Method for non-targeted screening analysis of veterinary drug residues in animal-derived product

Abstract

The invention provides a non-targeted screening analysis method for veterinary drug residues in an animal-derived product, and belongs to the technical field of food safety detection.The method comprises the following steps that 1, after the animal-derived product is pulped, pretreatment is conducted, and sample solutions with different concentrations are obtained; (2) detecting the sample solutions with different concentrations by using an HPLC-MS / MS (High Performance Liquid Chromatography-Mass Spectrometry / Mass Spectrometry) method to obtain original data of samples with different concentrations; 3) performing metabonomics analysis on the original data by using analysis software, and screening variable projection importance VIPgt; all variables of 1; and 4) utilizing a support vector machine recursive feature elimination method SVMRFE to assist metabonomics analysis, and finally determining feature variables representing known veterinary drugs. Compared with simple metabonomics analysis, the method provided by the invention has the advantages that the screening precision is higher, and veterinary drug residues in animal-derived products can be comprehensively, accurately and quickly screened.

Description

Technical Field

[0001] The invention belongs to the technical field of food safety detection, and in particular relates to a method for non-targeted screening and analysis of veterinary drug residues in animal-derived products. Background Art

[0002] The illegal addition of veterinary drugs to animal-derived foods is a global food safety issue of great concern and has attracted widespread attention from all over the world. With the rapid development of the modern chemical industry, more and more new veterinary drugs may enter and remain in animal-derived foods, ultimately affecting human life and health through the food chain. Given that the current methods for screening veterinary drugs in animal-derived foods are mostly to first establish a database containing a variety of veterinary drug residues, and then search for veterinary drug residues in food based on this database, this method has a good effect on the non-targeted screening of veterinary drugs in the database, but it has the obvious defect of being unable to identify veterinary drugs outside the database, which poses a huge hidden danger to food safety. The identification of unknown and unexpected veterinary drug residues in animal-derived foods by continuously establishing new databases is a passive screening method that consumes a lot of manpower and material costs. It can no longer meet the growing demand for veterinary drug residue detection. Therefore, it is necessary to develop an effective non-targeted screening method.

[0003] Taking advantage of the advantages of metabolomics in analyzing massive data with complex characteristics such as small sample size, many interferences, and high noise, it has been applied to non-targeted screening of contaminants in food and has achieved good results. This method is based on high-resolution gas or liquid chromatography tandem mass spectrometry data. After multivariate and univariate analysis, the process of searching for residual contaminants is transformed into the process of searching for corresponding variables in metabolomics, and characteristic variables indicating residual contaminants (i.e., "marker compounds") are obtained. Then, through network or self-built database retrieval, pollutant identification is completed, thereby achieving the purpose of non-targeted screening. This method overcomes the shortcomings of traditional methods that cannot identify residual veterinary drugs outside the database, has broad development prospects and application value, and provides strong technical guarantees for relevant departments to comprehensively supervise the safety of animal-derived food and crack down on illegal addition of veterinary drugs.

[0004] However, there is currently no comprehensive, accurate and rapid method for screening veterinary drug residues in animal-derived foods. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method for non-targeted screening and analysis of veterinary drug residues in animal-derived products. The method combines metabolomics analysis and support vector machine analysis to more comprehensively screen out characteristic variables, and has higher screening accuracy than simple metabolomics analysis.

[0006] The present invention provides a method for non-targeted screening and analysis of veterinary drug residues in animal-derived products, comprising the following steps:

[0007] 1) After beating the animal-derived product, an internal standard solution is added and pre-treated to obtain sample solutions of different concentrations;

[0008] 2) Using HPLC-MS / MS method to detect sample solutions of different concentrations to obtain raw data of samples of different concentrations;

[0009] 3) Performing metabolomics analysis on the raw data of samples of different concentrations obtained in step 2) using analysis software to form a data matrix with variable names as horizontal coordinates and sample names as vertical coordinates; sorting by numerical values, and screening all variables with variable projection importance VIP>1;

[0010] 4) Using the support vector machine recursive feature elimination method SVMRFE to analyze the data matrix obtained in step 3), arranging the square values ​​of variable weights according to numerical values ​​to form a variable weight table;

[0011] 5) determining the position corresponding to the last variable with variable projection importance VIP>1 obtained in step 3) in the variable weight table obtained in step 4), and the variables before this position in the variable weight table are confirmed as pre-selected characteristic variables capable of characterizing the marker compound;

[0012] 6) Comparing the preselected characteristic variables with the retention time and accurate molecular weight of known veterinary drugs, and finally determining the characteristic variables representing the known veterinary drugs.

[0013] Preferably, the internal standard solution is a methanol solution of enrofloxacin-d5.

[0014] Preferably, in step 3), all variables with a screening variable projection importance VIP>1 are overlapping parts of the sample comparison data of different concentrations.

[0015] Preferably, the preselected characteristic variable is the overlapping portion of comparative data of samples of different concentrations.

[0016] Preferably, the metabolomics analysis in step 3) is performed on the Workflow4Metabolomics analysis platform.

[0017] Preferably, the support vector machine recursive feature elimination method SVMRFE described in step 4) is run under the radial basis function RBF; the data matrix is ​​randomly divided into 5 equal subsets, and 5-fold cross validation is performed, 4 of which are used for model training and the other 1 is used for testing.

[0018] Preferably, the animal-derived product comprises fish product.

[0019] Compared with the prior art, the present invention has the following beneficial effects: the method for non-targeted screening and analysis of veterinary drug residues in animal-derived products provided by the present invention is mainly based on metabolomics analysis, assisted by support vector machine analysis, with higher screening accuracy, and can comprehensively, accurately and quickly screen veterinary drug residues in animal-derived products, thereby achieving the purpose of effectively ensuring that consumers can eat safe fish. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Total ion chromatogram of standard added fish sample on W4M platform;

[0021] Figure 2 This is the PCA score plot of the standard added fish meat sample;

[0022] Figure 3 This is the cluster analysis diagram of the standard added fish meat samples;

[0023] Figure 4 OPLS-DA score graph of standard added fish meat samples;

[0024] Figure 5 This is the S-plot of the standard added fish sample;

[0025] Figure 6 This is the permutation test diagram for the standard added fish meat sample;

[0026] Figure 7 is the total ion chromatogram of the carp sample containing ciprofloxacin;

[0027] Figure 8 This is the total ion current of the carp sample containing tetracycline. DETAILED DESCRIPTION

[0028] The technical solutions provided by the present invention are described in detail below in conjunction with the embodiments, but they should not be construed as limiting the protection scope of the present invention.

[0029] Example 1

[0030] 1 Materials and Methods

[0031] 1.1 Reagents and Materials

[0032] The fish samples (negative) used to establish the screening method were cod imported from Russia; the fish samples used to verify the screening method were cod imported from the United States and carp purchased from a local farmers' market in Dalian.

[0033] Methanol, acetonitrile and n-hexane (chromatographic grade, Merck, Germany); formic acid (chromatographic grade, Shanghai Anpu Laboratory Technology Co., Ltd.); C18 adsorbent (Sinopharm Chemical Reagent Co., Ltd.); filter membrane (0.22 μm, Agilent, USA); ultrapure water (milliq ultrapure water system, Merck, Germany); enrofloxacin-d5 internal standard solution (100 μg / mL, methanol phase, First Standard, USA).

[0034] The 102 veterinary drugs were purchased from Sigma, First Standard, Dr. Ehrenstorfer, and TRC, all with a purity greater than 98%. The detailed information of all veterinary drugs is shown in Table 1.

[0035] Table 1 Basic information of 102 veterinary drugs

[0036]

[0037]

[0038]

[0039]

[0040] 1.2 Instruments

[0041] 8010ES meat grinder (Blender, USA); JJ200Y balance (sensitivity 0.01 g) (Changshu Shuangjie Testing Instrument Factory); MS-3 vortex mixer (IKA, Germany); 3-18KS high-speed refrigerated centrifuge (Sigma, USA); QYN100-2 nitrogen blowdown apparatus (Shanghai Bajiu Industrial Co., Ltd.); Q Exactive Plus quadrupole / electrostatic field orbital trap high performance liquid chromatography-tandem mass spectrometer (Thermo Fisher Scientific, USA); Accucore RP-MS chromatographic column (100 mm×2.1 mm, 2.6 μm) (Thermo Fisher Scientific, USA).

[0042] 1.3 Solution preparation

[0043] Methanol solutions containing 102 veterinary drugs (veterinary drug concentration of 100 μg / mL) were prepared respectively, 1 mL of each was mixed, and then diluted with methanol to a 1 μg / mL solution. The methanol solution of enrofloxacin-d5 (100 μg / mL) was diluted with methanol to obtain a 100 ng / mL mixed solution.

[0044] 1.4 Sample preparation and pretreatment process

[0045] (a) Cut the fish meat sample into small pieces and grind them into minced meat using a meat grinder;

[0046] (b) 2.0, 5.0 and 10.0 g of minced meat and 20, 50 and 100 μL of a mixed solution containing 102 veterinary drugs (1 μg / mL) were added to 50 mL polypropylene centrifuge tubes. 0.5 mL of enrofloxacin-d5 methanol solution (100 ng / mL) was then added to each centrifuge tube to calibrate the recovery of the sample pretreatment process;

[0047] (c) 10 mL of acetonitrile / water (4 / 1, v / v) was injected into the centrifuge tube and vortexed for 1 minute. Subsequently, the sample was centrifuged at 6000 r / min for 10 minutes;

[0048] (d) Transfer all the supernatant to a new centrifuge tube containing 500 mg of C18, and continue to add 10 mL of acetonitrile-saturated n-hexane solution. After vortexing for 1 minute, the sample is centrifuged at 6000 r / min for 10 minutes, and the lower layer solution is transferred to a new centrifuge tube;

[0049] (e) The solution was blown to near dryness with nitrogen, and the residue was dissolved with 1 mL of methanol / water (1:1, v / v), filtered through a 0.22 μm filter membrane, and 102 mixed sample solutions of veterinary drugs with theoretical mass concentrations of 20, 50, and 100 ng / mL were prepared. Each concentration experiment needed to be repeated 9 times to obtain 9 parallel samples for subsequent metabolomics analysis.

[0050] 1.5 Sample grouping and naming

[0051] The three concentration groups of samples were named 20ng / mL-1~20ng / mL-9, 50ng / mL-1~50ng / mL-9 and 100ng / mL-1~100ng / mL-9. 30μL of each sample was extracted and mixed to obtain a quality control (QC) sample. The QC sample was injected 3 times before and after the injection of each concentration group sample, for a total of 12 times, named QC-1, QC-2, ..., QC-12, and used to evaluate the stability of high performance liquid chromatography-tandem mass spectrometer (HPLC-MS / MS).

[0052] 1.6 Analysis methods

[0053] M+H adducts were obtained by HPLC-MS / MS in the positive ion mode with a heated electrospray ionization (ESI) source for metabolomics analysis of all veterinary drugs and internal standards. 10 μL of sample was injected each time and separated by an Accucore RP-MS column (100×2.1 mm, 2.6 μm). The mobile phases were 0.1% formic acid in water (eluent A) and 0.1% formic acid in acetonitrile (eluent B) at a flow rate of 0.25 mL / min. The gradient program was as follows: 5% B at 0 min; increased to 60% B from 0.5 to 5.0 min; continued to increase to 100% B from 5.0 to 6.0 min; maintained at 100% B from 6.0 to 11.0 min; decreased to 5% B from 11.0 to 11.5 min; maintained at 5% B from 11.5 to 15.0 min. The column temperature was stabilized at 40°C. The instrument parameters were set as follows: the capillary and heating temperatures were both 320 °C; the spray voltage and lens voltage were 3200 and 50 V, respectively; the sheath gas and auxiliary gas were both nitrogen, with flow rates of 40 and 10 arb, respectively; the scanning mode was full scan / data-dependent secondary scan; the primary mass spectrometer full scan resolution, automatic gain control target, maximum dwell time, and m / z scanning range were 70000, 1×10 6 , 100ms and 100~1000; the secondary mass spectrometer resolution, automatic gain control target and maximum dwell time are 17500, 2×10 5 and 50ms.

[0054] Veterinary drug residue analysis was performed using the Trace Finder 3.3 software provided by the instrument. For the target parent ion, the response intensity threshold, signal-to-noise ratio, and mass error were 10,000, 5.0, and 5 ppm, respectively; for the secondary fragment ions, the minimum number of matching ions, response intensity threshold, and mass error were 1, 10,000, and 5 ppm, respectively. An enrofloxacin-d5 standard curve was prepared, and its parent ion peak area was used for quasi-quantitative analysis to calculate the recovery rate.

[0055] 1.7 Metabolomics Data Processing

[0056] The HPLC-MS / MS output files in .RAW format were converted into .mzXML format files by MSConvertGUI software and then uploaded to the Workflow4Metabolomics (W4M) platform (https: / / workflow4metabolomics.usegalaxy.fr / ) for metabolomics analysis. After a series of processing processes such as peak detection, peak alignment and retention time calibration, as well as peak intensity normalization, centering, scaling and data conversion, the chromatographic peaks were processed to form a data matrix with variable names (such as M111T222, M and T represent the extracted molecular weight and retention time of the target compound, respectively) and sample names (such as 100 ng / mL-1) as horizontal and vertical axes, respectively. Principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) in multivariate analysis were performed in SIMCA 14.1 software, while cluster analysis was performed in Heml 1.0.3.7 software. Whether the OPLS-DA model is overfitted is evaluated by permutation test. Variable importance in projection (VIP) is an important parameter for screening qualified variables as candidate marker compounds, and the value is generally required to be greater than 1. After completing the above work, qualified variables are screened out and arranged in order of VIP value to form a VIP table.

[0057] SVMRFE runs under the commonly used kernel function - radial basis function (RBF). The data matrix obtained by the above analysis is randomly divided into 5 equal subsets, and a 5-fold cross-validation is performed. Model training is performed on 4 of them, and the remaining 1 is used for testing. The average of the 5 accuracy rates is calculated. Repeat the above process 10 times, and the average accuracy is used as the final classification accuracy of the model. The selection of kernel parameters is also completed through 5-fold cross-validation. The classifier is trained on any one of the 4 subsets and tested on the fifth subset. With the help of the LIBSVM software package, a set of parameters that can provide the best cross-validation accuracy is used for further analysis. After calculation, the parameters with the highest cross-validation accuracy are selected as kernel parameters for actual classification. The obtained variable weight square value (w 2 ) are arranged according to numerical values ​​to form a variable weight table for further analysis.

[0058] The running code for calculating variable weights is shown below. The process is performed in MATLAB software (R2019b version).

[0059] data=load(data);

[0060] [data_r,data_c]=size(data);

[0061] indices=crossvalind('Kfold',data_r,5);

[0062] for i=1:5;

[0063] test=(indices==i);

[0064] train=~test;

[0065] test_data=data(test,1:data_c-1);

[0066] test_label=data(test,data_c);

[0067] train_data=data(train,1:data_c-1);

[0068] train_label=data(train,data_c);

[0069] end;

[0070] classifer=fitcsvm(train_data,train_label);

[0071] predict_label=predict(classifer,test_data);

[0072] accuracy_svm=length(find(predict_label==test_labels)) / length(test_label);

[0073] sum_accuracy_svm=sum_accuracy_svm+accuracy_svm;

[0074] mean_accuracy_svm=sum_accuracy_svm / 5;

[0075] disp(mean_accuracy_svm);

[0076] n=size(data,2);

[0077] s=1:n;

[0078] r=[];

[0079] iter=1;

[0080] while~isempty(s);

[0081] ifmod(iter,10)==0;

[0082] str=['===',num2str(iter),'==='];

[0083] disp(str);

[0084] disp('processing.....');

[0085] end;

[0086] X=data(:,s);

[0087] v=5;

[0088] [bestCVaccuracy,bestc,bestg]=SVMcgForClass(label,X,-8,8,-8,8,v,0.8,0.8,4.5);

[0089] cmd=['-c',num2str(bestc),'-g',num2str(bestg)];

[0090] model=svmtrain(label,X,cmd);

[0091] model=svmtrain(label,X);

[0092] w=model.SVs'*model.sv_coef;

[0093] c=w.^2;

[0094] [c_minvalue,f]=min(c);

[0095] r=[s(f),r];

[0096] ind=[1:f-1,f+1:length(s)];

[0097] s=s(ind);

[0098] iter=iter+1;

[0099] end

[0100] When the variable with VIP>1 is arranged in the weight table in a continuous position, it indicates that the VIP method and the weight sorting method are completely consistent in finding effective variables; but when the arrangement position is not continuous, it means that some variables with VIP<1 are also included in the weight table. Generally, SVMRFE has a better prediction effect than OPLS-DA, that is, the weight sorting method can obtain effective variables more accurately than the VIP method. Therefore, the present invention determines the position corresponding to the last variable with variable projection importance VIP>1 obtained by metabolomics analysis in the variable weight table, and the variables before this position in the variable weight table are confirmed as pre-selected characteristic variables that can characterize the marker compound (wherein the part of the variables with VIP<1 in the weight table is retained to prevent the loss of some effective variables during the VIP calculation process). Compare the qualified variables in the 20 and 100 ng / mL mass concentration groups (theoretically, they should have significantly low and significantly high peak intensities, respectively), screen out overlapping variables, and perform paired t-tests (completed in SPSS Statistics V17.0 software) and concentration multiple change analysis and other univariate analyses to further confirm the validity of the variables. By comparing the retention times and accurate molecular weights (the absolute value of the error was less than 5 ppm) of 102 veterinary drugs, as shown in Table 1, the characteristic variables representing the marker compounds were finally determined.

[0101] 2 Results and analysis

[0102] 2.1 Data Preprocessing

[0103] The total ion currents of the three concentration groups of samples on the W4M platform are shown in Figure 1 The fish matrix is ​​relatively complex, and the pretreatment process may cause significant differences in the recovery rate of each pollutant in different samples, resulting in a lack of obvious correlation between the peak intensity and concentration level in the total ion flow diagram, such as Figure 1As shown in Table 2. In order to eliminate the influence of recovery on peak intensity, enrofloxacin-d5 (parent ion m / z 365.21094; fragment ions m / z 348.19690, 322.21803 and 246.11012; retention time 3.87min) was used as the internal standard for recovery and added to the sample for pretreatment experiments. The enrofloxacin-d5 standard curve (5, 10, 25, 50 and 100 ng / mL) was prepared using blank fish sample extracts to calculate the recovery of the internal standard in the sample. As shown in Table 2, the recoveries of enrofloxacin-d5 in the 20, 50 and 100 ng / mL groups were 81.3% to 97.8%, 81.3% to 96.4% and 83.7% to 97.4%, respectively. The recovery of the internal standard compound in each sample was multiplied by a calibration factor to obtain its 100% recovery. The peak intensity of the internal standard compound and all other variable peak intensities were also multiplied by the calibration factor to obtain the calibrated values. After deleting the variables with relative standard deviation of peak intensity greater than 30% in the QC group and the three concentration groups, a 867×39 data matrix was established for the next analysis.

[0104] Table 2 Recovery rate and relative standard deviation of enrofloxacin-d5 in fish samples (%, n = 9)

[0105]

[0106] 2.2 Multivariate analysis

[0107] 2.2.1 PCA Results

[0108] like Figure 2 As shown, the confidence level of all fish samples is within 95%. The 12 samples of the QC group clustered at the origin, indicating that the data obtained in this study are of high quality and can be used for further analysis. The 9 parallel samples in the same concentration group were clustered clearly, among which the clustering degree of the 20ng / L concentration group samples was significantly higher than that of the 50 and 100ng / L concentration group samples, indicating that the 20ng / L concentration group had relatively better intra-group similarity. The three concentration groups were clearly separated on the first principal component axis ( Figure 2 ), indicating that there are variables with significant differences among the three concentration groups, laying a foundation for finding marker compounds among the concentration groups.

[0109] 2.2.2 Cluster analysis results

[0110] Cluster analysis can directly describe the similarity of samples, that is, the samples with the highest similarity are clustered together first, and then clustered step by step until all samples are clustered. Figure 3 As can be seen from the figure, samples in the same concentration group are clearly separated from samples in other concentration groups, showing good intra-group similarity. The inter-group separation can be clearly attributed to the difference in the peak intensity of the variables. The cluster analysis results also show that there are indeed variables with significant differences.

[0111] 2.2.3 OPLS-DA results

[0112] OPLS-DA is a two-class model, and class 1 and class 2 are located on both sides of the y-axis in the OPLS-DA score graph. Figure 4 As shown in the figure, category 1 and category 2 are clearly separated on the first principal component axis, indicating that there are significant differences between the two categories. 2 Y and Q 2 are two key parameters for evaluating the interpretability and predictability of the OPLS-DA model along the y-axis. When both parameter values ​​are close to 1, it indicates that the OPLS-DA model has good reliability and predictability. Figure 4 In, R 2 Y and Q 2 The values ​​are all greater than 0.99, indicating that the OPLS-DA model established in the present invention has good robustness.

[0113] Each point in the S-plot represents a variable ( Figure 5 ), the variables that are most likely to represent marker compounds should be located near the two ends of the "S-shaped" graph. These variables contribute more to distinguishing between Class 1 and Class 2 in the OPLS-DA model and have higher confidence. Figure 5 (a) The right end of the "S-shaped" graph looks for the marker compound in the low concentration group (20 ng / mL), while Figure 5 (b) The left end of the "S-shaped" graph was used to find the marker compound of the high concentration group (100 ng / mL). Whether the OPLS-DA model is overfitting was tested by 200 iterations of permutation test to obtain the index Q 2 If the P value is less than 0.05, the corresponding permutation test graph Q 2 The intercept of the regression line is less than 0.05, indicating that the OPLS-DA model is not overfitting. Figure 6 As shown, all OPLS-DA models of the present invention meet the above conditions, so all OPLS-DA models do not have overfitting.

[0114] Calculate the VIP value of the variable, can be used to evaluate the importance of the variable. Usually VIP>1 is used as the key threshold for selecting the characteristic variable that can represent the marker compound. According to the above principle, the present invention screens out 136 valid variables from the analysis of 20ng / mL concentration group vs 50ng / mL concentration group (hereinafter abbreviated as 20vs 50), and screens 165 valid variables from the analysis of 100ng / mL concentration group vs 50ng / mL concentration group (hereinafter abbreviated as 100vs 50). In the effective variables screened in the two groups, a total of 92 variables overlap, and the 92 variables are pre-selected characteristic variables. These 92 pre-selected characteristic variables are compared with the retention time and accurate molecular weight of known veterinary drugs, and finally determine 88 characteristic variables that can represent veterinary drugs.

[0115] 2.2.4 Support Vector Machine Results

[0116] After running the SVM code, the classification accuracy of the 5-fold cross validation was calculated to be 100%, indicating that the model has good generalization ability. In the 20vs 50 analysis, the best kernel function parameter (g) and the best penalty factor (c) obtained under the condition of 100% accuracy of the 5-fold cross validation were both 0.0035, while g and c were both 0.0036 in the 100vs 50 analysis. The 136 variables with VIP>1 obtained in the 20vs 50 analysis were included in the weight table of 224 variables, while in the 100vs 50 analysis, a total of 165 variables with VIP>1 were included in the weight table of 247 variables. A total of 169 variables overlapped in the two weight tables (224vs247). These 169 variables were pre-selected feature variables. After univariate analysis of these variables, these 169 pre-selected feature variables were compared with the retention time and accurate molecular weight of known veterinary drugs, and finally 102 feature variables that can represent veterinary drugs were determined.

[0117] 2.3 Univariate analysis

[0118] In the univariate analysis of metabolomics, paired t-test and concentration fold change are the two most commonly used analysis methods. The former focuses on whether there is a significant difference in the peak intensity of the variable between any two concentration groups, while the latter focuses on the change in the ratio of the peak intensity of the variable between any two concentration groups. Variables that simultaneously meet the paired t-test significance level (P) less than 0.05 and the concentration fold change (FC) greater than 2 are usually selected as feature variables. The P values ​​(i.e., P values) of the 169 overlapping variables in the weight table between the concentration groups of 20 and 50, 50 and 100, and 20 and 100 ng / mL were calculated. 20vs50 , P 50vs100 and P 20vs100), and the results showed that they were all less than 0.05 (Table 3), indicating that there were significant differences between the 169 variables. In addition, the concentration fold changes of the 169 variables in the 50 and 20, and 100 and 20 ng / mL groups (FC 50vs20 and FC 100vs20 ) were all greater than 2 (Table 4), further confirming that the 169 variables obtained by screening met the qualification of characteristic variables.

[0119] Table 3 Paired t-test results of 169 variables

[0120]

[0121]

[0122]

[0123] Table 4 FC results of 169 variables

[0124]

[0125]

[0126]

[0127] As shown in Table 5, the weight ranking method was able to identify characteristic variables representing all marker compounds (a total of 102) among 169 variables, while the VIP method could only screen out 88 characteristic variables, indicating that the weight ranking method performed better than the VIP method in selecting characteristic variables representing marker compounds. Although 14 veterinary drugs, including thiabendazole, triclabendazole, erythromycin, roxithromycin, flumequine, nalidixic acid, oxolinic acid, nabumetone, tolmetin, carprofen, propranolol, penbutolol, prednisone and fluorometholone, were included in the weight table (Table 5), at least one group of VIP values ​​was less than 1 in the 20 vs 50 and 100 vs 50 analyses, so they were not included in the VIP screening table. The metabolomics information of the 67 variables that failed to represent marker compounds in the weight table was relatively complex, and further analysis was needed to determine whether they were derived from fish matrix or foreign impurities.

[0128] The present invention adopts the same method to analyze the blank fish meat sample, and the result shows that no variables corresponding to the 102 veterinary drugs are found, thereby eliminating the interference of the 102 veterinary drugs inherent in the fish meat matrix on the search for marker compounds.

[0129] Table 5 102 marker compounds screened from fish sample group

[0130]

[0131]

[0132]

[0133]

[0134]

[0135] Note: a The VIP values ​​come from 20vs50 and 100vs50 respectively; b The squared weights (×100) from 20vs50 and 100vs50, respectively; c Molecular weight error (ppm) = (W4M platform extracted molecular weight - LC-MS / MS extracted molecular weight) × 10 6 / LC-MS / MS extracted molecular weight.

[0136] 2.4 Detection limit

[0137] The limits of detection (LODs) of 102 veterinary drugs were calculated according to the method proposed by the U.S. Environmental Protection Agency. The specific steps are as follows: (a) 2.0 g of blank fish meat sample was subjected to the pretreatment process described above to obtain 1 mL of extract; (b) a mixed solution of 102 veterinary drugs (1 μg / mL, 20 μL) was added to the extract, and the solution concentration was 20 ng / mL; (c) the above process was repeated to obtain 7 parallel samples; (d) metabolomics analysis was performed to obtain the variable peak intensity of 102 veterinary drugs; (e) for each veterinary drug, the 20 ng / mL mass concentration level corresponds to the average of the 7 parallel peak intensities, so the mass concentration (ng / mL) of each veterinary drug can be calculated by the formula—respective peak intensity × 20 / average peak intensity. As shown in Table 5, the LODs of 102 agricultural and veterinary drugs were between 0.3 and 3.9 μg / kg.

[0138] 2.5 Verification of the practicality of the method

[0139] The method of the present invention was applied to 5 samples of cod imported from the United States and 15 samples of carp purchased from the market to verify the practicality of the method. The results showed that 1 carp sample contained ciprofloxacin at a content of 18.2 μg / kg (e.g. Figure 7 As shown in Figure 2), another carp sample contained tetracycline at a content of 14.3 μg / kg (as shown in Figure 2). Figure 8As shown), no veterinary drugs were detected in the other 18 samples. The test results verified the practicality of the support vector machine-assisted metabolomics method proposed in the present invention in the screening of non-targeted veterinary drug pollutants. Studies have pointed out that antibiotics in the environment may accumulate in sediments due to migration, and sediments provide abundant food for carp, resulting in higher levels of veterinary drug residues in carp tissues. It is reported that the concentration of veterinary drugs accumulated in carp is higher than that of fish living in other water levels, so the risk of veterinary drug contamination for carp is also relatively high, which is the main reason why the present invention chooses carp as the verification object.

[0140] In summary, the embodiment of the present invention uses edible fish as the research object, adopts support vector machine-assisted metabolomics analysis strategy, and establishes a non-targeted screening method for veterinary drugs. The method provided by the present invention can effectively identify all 102 added veterinary drugs. Only 88 veterinary drugs can be screened using the metabolomics method alone, indicating that the support vector machine helps to improve the screening accuracy of the metabolomics method. The support vector machine-assisted metabolomics method is significantly affected by the level of veterinary drug residues in fish, and has a good screening effect for veterinary drugs with a concentration higher than 10 μg / kg. The effective combination of non-targeted and targeted screening may be an effective way to solve the food safety problem caused by veterinary drug residue contamination in fish food.

[0141] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for non-targeted screening and analysis of veterinary drug residues in animal-derived products, characterized in that: The following steps are involved: 1) After beating the animal-derived product, an internal standard solution is added and pre-treated to obtain sample solutions of different concentrations; 2) Using HPLC-MS / MS method to detect sample solutions of different concentrations to obtain raw data of samples of different concentrations; 3) Performing metabolomics analysis on the raw data of samples of different concentrations obtained in step 2) using analysis software to form a data matrix with variable names as horizontal coordinates and sample names as vertical coordinates; sorting by numerical values, and screening all variables with variable projection importance VIP>1; 4) Using the support vector machine recursive feature elimination method SVMRFE to analyze the data matrix obtained in step 3), arranging the square values ​​of variable weights according to numerical values ​​to form a variable weight table; 5) determining the position corresponding to the last variable with variable projection importance VIP>1 obtained in step 3) in the variable weight table obtained in step 4), and the variables before this position in the variable weight table are confirmed as pre-selected characteristic variables capable of characterizing the marker compound; 6) Comparing the preselected characteristic variables with the retention time and accurate molecular weight of known veterinary drugs, and finally determining the characteristic variables representing the known veterinary drugs.

2. The method according to claim 1, characterized in that The internal standard solution is a methanol solution of enrofloxacin-d5.

3. The method according to claim 1, characterized in that In step 3), all variables with a screening variable projection importance VIP>1 are the overlapping parts of the sample comparison data of different concentrations.

4. The method according to claim 3, characterized in that The preselected characteristic variable is the overlapping part of the comparison data of samples with different concentrations.

5. The method according to claim 1, characterized in that Step 3) The metabolomics analysis was performed on the Workflow4Metabolomics analysis platform.

6. The method according to claim 1, characterized in that The support vector machine recursive feature elimination method SVMRFE described in step 4) is run under the radial basis function RBF; the data matrix is ​​randomly divided into 5 equal subsets, and a 5-fold cross validation is performed, 4 of which are used for model training and the other 1 is used for testing.

7. The method according to claim 1, characterized in that The animal-derived products include fish products.

Images