Application of acetylcholine and 1-carboxyl-6-hydroxy-3, 4-dihydro-beta-carboline in sow urine in early pregnancy diagnosis

Abstract

The invention provides application of acetylcholine and 1-carboxyl-6-hydroxy-3, 4-dihydro-beta-carboline (the code in the attached drawing of the abstract is CHEMBL461889) in early pregnancy diagnosis, non-targeted metabonomics detection and analysis are carried out on the urine of pregnant and non-pregnant sows by virtue of a liquid chromatography-mass spectrometry technology; the two metabolic markers capable of being used for early pregnancy diagnosis of the sows are screened out by introducing two deep learning models for joint analysis, the AUC values of ROC curves of the two metabolic markers are 1, the accuracy is very high, and the pregnancy state of the sows can be accurately judged only 18 days after the sows are bred. The invention provides a novel effective sow early pregnancy diagnosis method for the pig breeding industry, and has important significance for reducing the breeding cost.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of biotechnology diagnosis, and particularly relates to application of metabolites in sow urine in early pregnancy detection. BACKGROUND

[0002] In the process of breeding sows, the average number of weaned piglets per sow per year (PSY) is a core economic indicator, and the empty gestation days of sows is an important factor affecting PSY. When sows are mated, various factors such as the quality of boar semen, the physiological state of sows, and the feeding environment may cause mating failure and thus extend the empty gestation time of sows and reduce PSY. How to accurately and efficiently identify mating failure sows and timely perform supplementary mating or culling is a core problem currently focused on by sow breeding farms. The mainstream pregnancy diagnosis method of current pig farms is ultrasonic diagnosis method, but this method can only have a relatively reliable pregnancy diagnosis effect at 23-25 days after mating, and the estrus cycle of empty gestation sows is averagely 21 days, so it is easy to miss the best supplementary mating opportunity due to late diagnosis. The remaining pregnancy diagnosis methods include hormone injection method, boar estrus test method, and platelet count method, but they are not used at present due to reasons such as great operation difficulty, high cost, and great stimulation to sows.

[0003] Early pregnancy assisted diagnosis through detection of markers in sow urine has practical applications, but there are few markers that can be used, the accuracy is limited, and more markers with high diagnosis efficiency need to be developed and applied. SUMMARY

[0004] To solve the above technical problems, the present application provides application of new urine metabolic markers in early pregnancy diagnosis of sows.

[0005] The present application provides metabolic markers in urine for early pregnancy diagnosis of sows, which are acetylcholine and 1-carboxyl-6-hydroxy-3,4-dihydro-beta-carboline, respectively.

[0006] Further, the present application also provides application of the above metabolic markers in early pregnancy diagnosis of sows.

[0007] Specifically, the application of the above metabolic markers in early pregnancy diagnosis of sows is to detect the expression level of the metabolic markers in sow urine, and to judge the pregnancy state of sows. Those skilled in the art should understand that the level of metabolites can be determined by using conventional methods known in the art, including but not limited to enzyme-linked immunoassay, colloidal gold method, mass spectrometry, etc.

[0008] The two metabolite markers, acetylcholine or 1-carboxyl-6-hydroxy-3,4-dihydro-beta-carboline, can also be used to develop various early pregnancy diagnosis products for sows, including but not limited to ELISA detection kits, colloidal gold test strips and other products for early pregnancy diagnosis of sows.

[0009] Preferably, the time of early pregnancy diagnosis is 18 days after insemination of sows.

[0010] Preferably, the large and large binary hybrid sows are the source of samples.

[0011] The present application firstly proves that acetylcholine and 1-carboxyl-6-hydroxy-3,4-dihydro-beta-carboline can be used as metabolite markers for early pregnancy diagnosis of sows by non-targeted metabolomics detection and analysis of sow urine combined with deep learning model, wherein 1-carboxyl-6-hydroxy-3,4-dihydro-beta-carboline is a new compound, and the two metabolite markers have high detection efficiency for pregnant and non-pregnant sows. The present application provides a new and effective detection method for early pregnancy detection of sows, and also provides a new direction for product development for early pregnancy diagnosis of sows. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 The correlation analysis graph for QC samples; Figure 2 The cluster heat map of pregnancy, non-pregnancy and quality control samples; Figure 3 The PCA analysis graph of pregnancy, non-pregnancy and quality control samples; Figure 4 The PLS-DA analysis graph of pregnancy, non-pregnancy and quality control samples; Figure 5 The volcano plot of differential metabolites in urine; Figure 6 The 1D-CNN model parameter graph; Figure 7 The weight heat map of each metabolite in the model in the 1D-CNN sample feature model of different samples; Figure 8 The global average feature weight of the 1D-CNN sample feature model (only test set); Figure 9 The Transformer model parameter graph; Figure 10 The importance of each metabolite in the Transformer model; Figure 11 The metabolites with a weight greater than 0.001 in the 1D-CNN model; Figure 12Metabolites with impact weight over 0.001 in the Transformer model; Figure 13 Weinberg chart of metabolites with significant expression difference, high random forest impact weight and high support vector machine impact weight; Figure 14 ROC curve chart of three metabolites; Figure 15 Box plot chart of expression difference of three metabolites in different groups of samples; Figure 16 ROC curve of diagnostic efficiency of three metabolites in independent samples; Figure 17 Box plot chart of expression of CHEMBL461889 in urine of pregnant and non-pregnant sows; Figure 18 ROC curve chart of CHEMBL461889 in urine of pregnant and non-pregnant sows; Figure 19 ROC curve chart of CHEMBL461889 in independent samples. DETAILED DESCRIPTION

[0013] Embodiments of the present application will be described in detail below with specific examples, which are only used to illustrate the present application and should not be regarded as limiting the scope of the present application, and it should be understood that they are more detailed descriptions of certain aspects, characteristics and embodiments of the present application. Unless otherwise specified, the instruments, reagents and materials used in the following examples are commercially available.

[0014] Example 1, preparation of urine samples 1. Treatment of test animals The experimental samples used in the present study were provided by Xinjiang Tiankang Animal Husbandry Technology Co., Ltd. They were 225-250 day-old, 135-150 kg body weight, good body condition, fed under the same conditions, and housed in the same conditions of long-grown crossbred gilts. All sows were in good health, without genetic diseases or deformities, the shed temperature was between 20-25°C, and the humidity was between 50-55%. The production strategy was all-in all-out, and the sows were fed with full-price feed. After the same estrus, the sows were batched and inseminated, and whether they were successfully pregnant was detected by B-ultrasound at 23-25 days of age after insemination.

[0015] 2. Sample collection Eighteen days after batch insemination, 14 sows were randomly selected, and urine was collected under the condition of no food and water. At the same time, 10 sows that failed to be inseminated or were not inseminated in the same batch were randomly collected. The collected samples were stored in liquid nitrogen, and then transferred to a-80°C refrigerator for storage.

[0016] 3. Sample pretreatment The samples were thawed at 4°C, vortexed, and 100 μL of each sample was taken and added to 300 μL of pre-cooled methanol. The mixture was placed at -20°C for 30 min, and then centrifuged at 16,000 g at 4°C for 20 min. The supernatant was collected for mass spectrometry analysis.

[0017] 4. Preparation of quality control samples All samples were mixed in equal amounts to obtain quality control samples (hereinafter referred to as QC samples), which were used to balance the chromatography-mass spectrometry system and the state of the measuring instrument, and to evaluate the stability of the system during the entire experimental process.

[0018] Example 2, screening of differential metabolic molecules 1. Non-targeted metabolomics detection using liquid chromatography-mass spectrometry (LC-MS) technology During the entire analysis process, the samples were placed in a 4°C autosampler. The samples were analyzed using a SHIMADZU-LC30 ultra-high performance liquid chromatography system (UHPLC) with an ACQUITY UPLC® HSS T3 (2.1 x 100 mm, 1.8 µm) (Waters, Milford, MA, USA) chromatographic column. The injection volume was 10 μL, the column temperature was 40°C, and the flow rate was 0.3 mL / min. The mobile phase A was 0.1% formic acid in water, and the mobile phase B was 0.1% formic acid in acetonitrile. The gradient elution program was as follows: 0-2 min, 0B; 2-6 min, B from 0% linearly to 48%; 6-10 min, B from 48% linearly to 100%; 10-12 min, B maintained at 100%; 12-12.1 min, B from 100% linearly to 0%; 12.1-15 min, B maintained at 0%.

[0019] Each sample was detected in positive (+) and negative (-) ion modes using electrospray ionization (ESI). After UPLC separation, the samples were analyzed by mass spectrometry using a QE Plus mass spectrometer (Thermo Scientific). The ionization conditions were as follows: Spray Voltage: 3.8kv (+) and 3.2kv (-); Capillary Temperature: 320 (±); Sheath Gas: 30 (±); Aux Gas: 5 (±); Probe Heater Temp: 350 (±); S-Lens RF Level: 50.

[0020] The mass spectrometry acquisition settings were as follows: Mass spectrometry acquisition time: 15 min. Precursor ion scan range: 75-1050 m / z, primary mass spectrometry resolution: 70,000 @ m / z 200, AGC target: 3e6, primary maximum IT: 100 ms. Secondary mass spectrometry analysis was performed using the following method: after each full scan, the secondary mass spectra of the 10 highest intensity precursor ions were acquired (MS2 scan). Secondary mass spectrometry resolution: 17,500 @ m / z 200, AGC target: 1e5, secondary maximum IT: 50 ms, MS2 activation type: HCD, isolation window: 2 m / z, normalized collision energy (Setpped): 20, 30, 40.

[0021] Four QC samples were prepared in this study, such as Figure 1 The correlation analysis graph of the QC samples shown includes six scatter plots in the lower left corner representing the expression level of each metabolic molecule in each pair of corresponding samples; histograms on the diagonal show the data distribution in each QC sample; and six graphs in the upper right corner represent the correlation coefficients between the metabolite expression levels of corresponding pairs of samples. As can be seen from the graph, the expression levels of metabolic molecules in the four QC samples are close to a linear relationship, the expression data of each metabolic molecule in different samples are close to a normal distribution, and the correlation coefficients between the four QC samples are all greater than 0.97, which is at a high level. This indicates that the detection data in the four samples have good consistency, confirming the reliability of the mass spectrometry results.

[0022] 2. Screening for differentially metabolized molecules After standardizing the expression levels of ion peaks in mass spectrometry (z-score), different models were used for analysis. First, dimensionality reduction methods were employed to describe the data characteristics of two types of samples: pregnant (represented by P in the chart) and non-pregnant (represented by NP in the chart). The dimensionality reduction analysis methods used were mainly principal component analysis (PCA) and partial least squares analysis (PLS-DA), and volcano plots were used to visualize the P-values ​​and fold changes.

[0023] exist Figure 2 The hierarchical clustering heatmap shown reveals significant metabolic differences in the urine of pregnant (P1~P14) and non-pregnant (NP1~NP10) sows. Meanwhile, the expression levels of metabolic molecules in QC samples (QC1~4) show high consistency, indicating that QC samples are highly stable and the LC-MS experiment has high reliability.

[0024] In the PCA dimensionality reduction results, a clear separation trend between pregnant and non-pregnant samples can be observed, with a significant classification boundary between the two classes.Figure 3 As shown, the QC samples are highly aggregated in the PCA plot, indicating that the instrument performance is stable and reliable during the experiment, and the experimental results are highly reliable; as shown in Figure 4 As shown in partial least squares discriminant analysis (PLS-DA), the separation trend of pregnant and non-pregnant samples can be seen, indicating that there are significant metabolic differences between the urine of pregnant and non-pregnant sows; and the QC samples are also highly concentrated in the partial least squares discriminant analysis, confirming the reliability of the detection results.

[0025] After denoising and peak extraction of the ion peaks detected in the sow urine samples using chromatography-mass spectrometry (LC-MS) in the present application, a total of 1527 metabolic molecules in sow urine were identified, as shown in Figure 5 As shown in the volcano plot of differential metabolites, 553 metabolic molecules with significant expression differences and fold change (FC) values greater than 1.5 or less than 0.6667 between pregnant and non-pregnant sows were screened, of which 466 metabolic molecules were significantly up-regulated, i.e. red dots in the figure with log2(FC)>0.58, and 87 metabolic molecules were significantly down-regulated, i.e. blue dots in the figure with log2(FC)<-0.58.

[0026] 3. Deep learning model screening of differential metabolites After obtaining the mass spectrometry data, the metabolic molecules were sorted according to the retention time RT(min), and the samples were randomly divided into training set and test set according to the ratio of 8:2, then the training set was used to train the one-dimensional convolutional neural network model (One-dimensional convolutional neural network, 1D-CNN) and the Transformer model respectively, after the training was completed, the test set was used to detect the accuracy of the model, and the importance of each feature in the model was calculated by using Grad-CAM++ algorithm in 1D-CNN and input gradient significance in Transformer respectively.

[0027] 1) 1D-CNN model In the 1D-CNN, two convolutional layers were set, the first convolutional layer was 1*3*32, and the second convolutional layer was 1*3*64. After each convolutional layer, a max pooling layer was set. After the last pooling layer, the data of each channel was flattened, and then entered the full connection layer for calculation. After training, in order to obtain the weight of different metabolic molecules in different samples, the Grad_CAM++ algorithm was introduced in this model. The target class score of the prediction result was used to perform high-order derivation on each channel in the last convolutional layer, and weights were assigned to each feature channel and position. The feature map was weighted according to the weight, and then the "negative contribution" features were filtered out through the ReLU function. Finally, the weight heat map was obtained. The hyperparameters of the 1D-CNN model are shown in Table 1.

[0028] Table 1 1D-CNN hyperparameters

[0029] As Figure 6 In the 1D-CNN model, the AUC values of the ROC curves of the training set and the test set were both 1.00. In the confusion matrix, the model achieved perfect segmentation of pregnant and non-pregnant samples in the training set and the test set. In the prediction probability box plot, the model had good reliability in predicting pregnant and non-pregnant samples. The learning curve showed that when the number of training times reached 10, the model loss in the training set and the test set rapidly decreased, and the accuracy rapidly increased.

[0030] In Table 2, it can be seen that the 1D-CNN model performed well in predicting pregnant and non-pregnant sow samples. In the training set and the test set, the precision, recall, and F1 score for different categories of samples were all 1.00, indicating that the model could well fit the training set data and could perfectly predict the test set data.

[0031] Table 2 1D-CNN model performance evaluation results

[0032] In Figure 7 The metabolic molecule weight heat map is shown. It can be seen that in different samples, the weights of each metabolic molecule in the model have large differences, and there is a relatively obvious feature importance preference between pregnant and non-pregnant samples. In the heat map distribution, it can be seen that some specific metabolic molecules show high importance in multiple samples, indicating that they may be the core metabolic markers for distinguishing samples.

[0033] To calculate the global weight of features in the model, a certain number of samples are selected to calculate the average of feature weights, and the training set samples are the source of model parameters, which may contain noise due to model overfitting, leading to feature weight bias to the specific mode of training data. Therefore, only the average of feature importance of each sample in the test set is selected to obtain the global average weight of each metabolite. The results are shown in Figure 8

[0034] 2) Transformer model The hyperparameter settings of the Transformer model are shown in Table 3.

[0035] Table 3 Transformer hyperparameters

[0036] As Figure 9 , the AUC values of the ROC curves of the training set and the test set in the Transformer model are both 1.00; in the confusion matrix, the pregnant and non-pregnant samples in the training set and the test set can be perfectly segmented and predicted; the prediction probability box plot shows that the reliability of the model in predicting pregnant and non-pregnant samples is excellent; the learning curve shows that after the start of training, the model loss and accuracy of the training set and the test set change rapidly to the best value.

[0037] In Table 4, it can be seen that the Transformer model has good fitting and prediction performance on the non-targeted metabolomics results of pregnant and non-pregnant sow urine samples. The precision, recall and F1 score of different categories of samples in the training set and the test set are all 1.00, indicating that the model has excellent fitting effect on the training set data, and can also perfectly predict the test set data.

[0038] Table 4 Transformer model performance evaluation results

[0039] ​In the Transformer model, the metabolite sequences in the samples are first preprocessed into a two-dimensional tensor, then mapped to a third-order tensor, and two stacked encoder layers are added. Multi-head attention is used in the encoders to mine feature associations, and layer normalization and Dropout are performed in both encoder layers. The output of the last encoder layer is then pooled and compressed to obtain global features, which are fed into a fully connected layer to output class scores for binary classification. While multi-head attention can capture the importance of features in the model, this approach focuses more on interpreting feature correlations and is easily affected by data distribution. Therefore, this experiment uses the gradient significance algorithm to calculate the feature weights in the model, focusing more on the impact of features on the model output, making the selected high-importance metabolites more interpretable. The results are as follows: Figure 10 As shown.

[0040] Example 3: Determining Metabolic Markers and Their Diagnostic Performance This invention screened 553 metabolic molecules whose levels differed significantly between pregnant and non-pregnant samples, and which were significantly upregulated or downregulated; 126 metabolic molecules, calculated by Grad-CAM++, had a global importance greater than 0.001 in the 1D-CNN model, such as... Figure 11 As shown; in the Transformer model, 67 metabolic molecules with weights exceeding 0.001 were calculated using the attention mechanism, such as... Figure 12 As shown. Venn diagrams of metabolic molecules obtained by the three screening methods ( Figure 13 Three common metabolic molecules were screened out in the sample: Axitinib (coded NEG6871 in the figure), Muscaflavin (coded NEG2516 in the figure), and Acetylcholine (coded POS2031 in the figure).

[0041] ROC curves and the significance of expression differences of the three screened metabolic molecules in urine samples from pregnant and non-pregnant sows were calculated. In this experiment, the expression level of the metabolic molecules is the relative expression level, which is the area value of the metabolic molecules detected by liquid chromatography-mass spectrometry. Figure 14 In the ROC curves, all three metabolites showed high AUC values, with Acetylcholine having the highest AUC value of 1.00, followed by Axitinib at 0.90, and Muscaflavin at the lowest at 0.64. However, the expression levels of these three metabolites in the urine of pregnant and non-pregnant sows showed significant differences. Figure 15 It can be seen that the expression levels of Axitinib and Acetylcholine differed significantly in different groups of samples, and the expression levels of Muscaflavin also differed significantly, with all differences being statistically significant (P<0.05), which is consistent with the ROC curve.

[0042] As shown in Table 5, in the ROC curves of the three screened metabolites, the sensitivity, specificity and Youden index of Acetylcholine all reached 1.00, indicating that this metabolite can well distinguish between pregnant and non-pregnant samples; the sensitivity of Muscaflavin was 1.00, but the specificity and Youden index were only 0.45, and this metabolite was highly expressed in non-pregnant sow urine samples compared with pregnant samples, indicating that the model has strong judgment ability for non-pregnant samples and weak judgment ability for pregnant samples; the sensitivity, specificity and Youden index of Axitinib were all high, indicating that this metabolite has certain judgment ability for pregnant and non-pregnant samples.

[0043] Table 5 ROC curve analysis of the diagnostic efficiency of three types of metabolites

[0044] Example Four, Diagnostic Efficiency of Target Metabolites in Independent Samples In combination with Table 6 and Figure 16 It can be seen that the diagnostic efficiency of the three metabolites in independent samples is different, among which Axitinib and Acetylcholine are both higher in pregnant samples than in non-pregnant, while Muscaflavin is higher in non-pregnant samples than in pregnant, and Axitinib has high specificity, while Muscaflavin has high sensitivity, indicating that the two metabolites have high reliability in the diagnosis of barren sows in independent samples. The sensitivity and specificity of Acetylcholine are both 1.00, indicating that its recognition accuracy for barren and pregnant sows is ideal.

[0045] Therefore, Acetylcholine, which is confirmed to have ideal recognition accuracy for barren and pregnant sows, is a metabolic marker for early pregnancy diagnosis in sow urine. In practical application, when the expression of Acetylcholine detected is higher than the optimal threshold (i.e. the diagnostic threshold), it is diagnosed as pregnant, and lower than the optimal threshold, it is diagnosed as non-pregnant. The mass spectrum characteristics of Acetylcholine are: mass-to-charge ratio is 146.1176, retention time is 2.219 min, and molecular structural formula is: .

[0046] Table 6 Diagnostic efficiency of three types of metabolites in independent samples

[0047] In this experiment, a new metabolite CHEMBL461889 was also noticed, which is a β-carboline derivative, and its molecular structural formula is: , according to its structure named 1-carboxy-6-hydroxy-3,4-dihydro-beta- carboline, its mass spectrum characteristics are: mass to charge ratio is 231.07629, retention time is 5.494 min. As shown in Figure 17 , the expression amount of the metabolite in pregnant and non-pregnant urine samples is extremely significantly different and has statistical significance (P<0.05), the AUC value in the ROC curve is 1.00 (as shown in Figure 18 ), the weight in the Transformer model is 0.001043, but the weight in the 1D-CNN model is only 0.0009914, so it is not parallel with the three metabolites mentioned above.

[0048] But in the ROC curve of this metabolite, as shown in Table 7, the sensitivity, specificity and Youden index are all 1.00, indicating that this metabolite has strong judgment ability for pregnant and non-pregnant samples, and has the potential to be a biomarker for sow pregnancy diagnosis. In practical application, when the expression amount of 1-carboxy-6-hydroxy-3,4-dihydro-beta-carboline is higher than its optimal threshold (i.e. diagnosis threshold), it is diagnosed as pregnancy, and lower than the optimal threshold, it is diagnosed as non-pregnancy.

[0049] Table 7 ROC curve analysis of CHEMBL461889 diagnostic efficiency

[0050] Combined with Table 8 and Figure 19 , it can be seen that the sensitivity, specificity, Youden index and AUC value in the ROC curve of CHEMBL461889 in the independent samples are all 1.00, indicating that it still has excellent pregnancy diagnosis accuracy and reliability in the independent samples, and has good development potential in sow pregnancy diagnosis.

[0051] Table 8 Analysis of CHEMBL461889 diagnostic efficiency in independent samples

[0052] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A metabolic marker in urine for early pregnancy diagnosis in sows, characterized in that, The metabolic markers are acetylcholine or 1-carboxy-6-hydroxy-3,4-dihydro-β-carboline.

2. The application of the metabolic marker of claim 1 in the early pregnancy diagnosis of sows.

3. The application as described in claim 2, characterized in that, The expression levels of the aforementioned metabolic markers in sow urine were detected and used to determine the pregnancy status of sows.

4. The application of the metabolic biomarker as described in claim 1 in the development of diagnostic products for early pregnancy in sows.

5. The metabolic biomarker as described in claim 1, or any application as described in claims 2-4, characterized in that, The early pregnancy diagnosis is performed 18 days after mating of the sow.

6. The metabolic biomarker as described in claim 1, or any application as described in claims 2-4, characterized in that, The sow in question is a Large White crossbred sow.

Images