Method and system for screening diet-Parkinson's disease (PD) related genes

Abstract

The invention provides a method and a system for screening diet-Parkinson's disease (PD) related genes. The method comprises the following steps: firstly, carrying out difference analysis on transcriptome data sets GSE8397 and GSE20295; a gene module is constructed through WGCNA; 8 types of diet phenotype target gene sets (high cholesterol / high glucose / high fat / high carbohydrate diet and egg / whole grain / mutton / fish intake), PD target gene sets (3907 targets) and DEGs are integrated, and candidate genes are screened in a cross mode; an analysis model is constructed by adopting a machine learning algorithm, a random forest algorithm is preferably selected, and 22 key genes are screened out by utilizing an SHAP value. A detection kit and an analysis system are developed based on the gene combination and are applied to PD mechanism research of A53T transgenic mouse models and human midbrain organs, and the molecular mechanism that diet intake affects PD pathology by adjusting dopaminergic neuron related genes is disclosed for the first time.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biomedical technology, and particularly relates to a diet-Parkinson's disease (PD) correlation gene screening method based on multi-omics analysis, a key gene combination and application thereof. BACKGROUND

[0002] Parkinson's disease (PD) is the second most common neurodegenerative disease, with more than 100 million patients worldwide. With the aggravation of population aging, the incidence of PD is showing a sustained upward trend. The typical pathological features of PD include selective loss of dopaminergic neurons in the substantia nigra of the midbrain and abnormal aggregation of alpha-synuclein, and the clinical manifestations include motor symptoms such as bradykinesia, resting tremor and muscle rigidity, and non-motor symptoms such as constipation, sleep disorders and depression. At present, the clinical treatment mainly includes dopamine replacement therapy (such as levodopa) and deep brain stimulation, but these methods can only temporarily relieve symptoms and cannot delay disease progression, and long-term use can cause serious side effects such as dyskinesia.

[0003] It is worth noting that in recent years, studies have found that intestinal flora dysbiosis participates in the pathological process of PD through the "microbiota-gut-brain axis". PD patients generally have reduced levels of short-chain fatty acids (SCFA) and damaged intestinal barrier integrity, leading to increased release of inflammatory factors (such as TNF-α, IFN-γ) and neuroinflammation. Studies have shown that the concentration of SCFA (especially propionic acid) in the feces of PD patients is 40% lower than that of healthy controls, and is significantly negatively correlated with the severity of motor symptoms (UPDRSIII score). These findings reveal the important value of the intestinal microenvironment as a new target for PD intervention, and also provide a scientific basis for dietary intervention.

[0004] Epidemiological studies have shown that different dietary patterns have a significant regulatory effect on the occurrence and development of PD, and its core mechanism involves the regulation of intestinal flora, the inhibition of inflammation and the maintenance of metabolic homeostasis. Studies have shown that people who consume ≥11 ultra-processed foods (sugar-containing beverages, processed meats, etc.) per day have a 2.5 times higher risk of developing three or more PD prodromal symptoms (such as rapid eye movement sleep disorder, constipation, and depression) than those who consume less. The mechanism is related to the direct damage of additives such as nitrite and artificial sweeteners in ultra-processed foods to neurons, or through triggering intestinal flora imbalance and systemic inflammation to accelerate α-synuclein pathological deposition. On the contrary, the Mediterranean diet and high-fiber diet show significant neuroprotective effects. Patients who strictly follow the Mediterranean diet variant have a 43% slower disease progression rate, and its core components such as walnut dietary fiber and deep-sea fish play a role by increasing SCFAs and inhibiting neuroinflammation. The team of Yunnan Agricultural University has further found that walnut dietary fiber not only prevents rotenone-induced PD symptoms in mice, but also treats individuals who have already developed the disease, and its mechanism is closely related to the regulation of the 'gut-brain axis'.

[0005] Although epidemiological studies have clearly shown that different dietary patterns have a significant regulatory effect on the occurrence and development of Parkinson's disease (PD), there are still obvious deficiencies in revealing its molecular mechanisms in existing technologies. Existing research mainly focuses on macro-epidemiological associations (such as increased intake of ultra-processed foods increasing the risk of PD prodromal symptoms) and the 'gut-brain axis' pathway mediated by intestinal flora (such as decreased SCFA levels and increased neuroinflammation), however, these studies have failed to explore the specific targets at the genetic level of dietary intervention. For example, the pathological process of PD involves the damage of dopaminergic neurons and the deposition of α-synuclein, but there is still a lack of systematic evidence on how diet directly regulates the expression network of related genes (such as neuronal transport, mitochondrial function, or inflammation pathways). This leads to the lack of precise targets for existing prevention strategies (such as dietary adjustment), making it difficult to develop targeted intervention measures.

[0006] Further, the existing screening methods of diet-PD associated genes have limitations. Traditional methods usually rely on single omics analysis (such as differentially expressed genes of transcriptome), but cannot integrate multi-dimensional data (such as diet phenotype target points and PD pathway genes) to identify core regulatory genes. For example, GWAS and gene databases (such as GeneCards) can provide candidate gene sets, but they do not combine functional enrichment and machine learning modeling to verify their causal relationship with PD. This causes the screened genes to be redundant and functionally ambiguous, making it difficult to distinguish between core driver genes (such as genes involved in neuronal signaling or metabolic homeostasis) and secondary regulatory factors. Notably, although recent studies have attempted to use weighted gene co-expression network analysis (WGCNA) to construct gene modules (such as PD-related modules), they have not cross-validated them with diet phenotype target points to reveal the direct regulatory effects of specific diets (such as high cholesterol or high sugar diets) on PD pathways. SUMMARY

[0007] The purpose of the present application is to provide a method for screening diet-Parkinson's disease (PD) associated genes to solve the problems in the prior art.

[0008] To solve the above technical problems, the present application adopts the following technical solutions:

[0009] A method for screening diet-Parkinson's disease associated genes, comprising the following steps:

[0010] (a) performing differential analysis on transcriptome datasets GSE8397 and GSE20295 to obtain differentially expressed genes DEGs, respectively;

[0011] (b) constructing gene modules by weighted gene co-expression network analysis WGCNA: selecting soft strength 19 to construct 3 modules for DEGs of GSE8397, and selecting soft strength 14 to construct 5 modules for DEGs of GSE20295;

[0012] (c) integrating diet phenotype target gene set, PD target gene set and DEGs of step (a) to cross-screen candidate genes; the diet phenotype includes high cholesterol diet, high sugar diet, high fat diet, high carbohydrate diet, egg intake, whole grain intake, mutton intake and fish intake;

[0013] (d) constructing a gene expression analysis model for candidate genes using an integrated machine learning algorithm, and screening key genes using SHAP values.

[0014] Preferably, the construction of the diet phenotype target gene set in step (c) comprises:

[0015] The database GWAS catalog, GeneCards, UKBB GWAS and KoGES were searched with the species limit set as "Homo sapiens";

[0016] The genes with scores higher than the median in GeneCards were selected;

[0017] P<10 -6 was used as the threshold for gene annotation.

[0018] Preferably, the machine learning algorithm in step (d) is a random forest algorithm, and the parameter combination is:

[0019] High cholesterol diet: mtry=13; high sugar diet: mtry=7 or mtry=19; high fat diet: mtry=28; high carbohydrate diet: mtry=8; egg intake: mtry=8; whole grain intake: mtry=7; lamb intake: mtry=5; fish intake: mtry=12.

[0020] Preferably, the key genes in step (d) are the top genes in the SHAP value ranking of each dietary phenotype, and the obtained genes are de-duplicated. In a specific embodiment of the present application, 22 genes are finally integrated: DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

[0021] Based on the above, the present application also provides a gene combination for studying the association between diet and Parkinson's disease, which comprises the 22 genes described above: DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

[0022] The present application also provides the use of one or a combination of DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA in the preparation of a reagent for detecting Parkinson's disease.

[0023] Further, the present application also provides a detection kit, which comprises specific primer pairs for detecting the expression levels of the above-mentioned 22 genes.

[0024] The present application also provides a diet-PD association analysis system, which comprises:

[0025] a data input unit for receiving the expression data of the above-mentioned 22 genes;

[0026] an algorithm processing unit for analyzing the association between gene expression and diet phenotype by using a random forest algorithm, wherein the parameters are configured as:

[0027] mtry = 13 for high cholesterol diet; mtry = 7 or 19 for high sugar diet; mtry = 28 for high fat diet;

[0028] mtry = 8 for high carbohydrate diet;

[0029] a visual output unit for generating SHAP swarm plots to show the gene contribution.

[0030] Further, the present application also provides the application of midbrain organoids in the research of diet-PD mechanism, and the expression levels of the above-mentioned 22 genes in the organoids are detected by using the above-mentioned kit.

[0031] the association between the gene expression changes and the diet intervention is analyzed, wherein the organoids are cultured by the following methods:

[0032] ①differentiating H9 hESC cells in a neural induction medium containing SHH and FGF8b;

[0033] ②transferring the cells into a maturation medium containing BDNF and GDNF on the 8th day and culturing to 90 days.

[0034] Further, the present application provides a method for constructing an animal model for Parkinson's disease research, which comprises:

[0035] feeding A53T transgenic mice with high cholesterol / high sugar / high fat / high carbohydrate feed;

[0036] detecting the expression of the above-mentioned 22 genes in the substantia nigra tissue by using the kit;

[0037] when FOXA1, GBE1, TH, KCND3, PBX1 and SNCA are significantly up-regulated, it is determined that the model construction is successful.

[0038] The present application integrates differential analysis (GSE8397: 445 DEGs; GSE20295: 947 DEGs), WGCNA modules and multi-source database target cross (such as 25 candidate genes for high cholesterol diet), which significantly improves the efficiency of key gene screening and avoids the redundancy problem of traditional methods.

[0039] Based on 14 algorithms and 195 combinations, Random Forest (RF) performed best in all 8 dietary phenotypes: the highest accuracy of the test set was 0.882 (high-fat diet), the AUC was 0.941, and the lowest accuracy of 0.765 was still better than the conventional method. The SHAP value interpretation framework clearly explains the direction of gene contribution (such as low expression of DLD in high-sugar diet promotes PD risk), breaking through the "black box" limitation.

[0040] The 22 key genes (such as SNCA, TH, FOXA1) screened by the present application cover the core pathology of PD: SNCA is significantly up-regulated in A53T mice (P<0.05), TH regulates dopamine synthesis; FOXA1 is abnormally expressed in organoids and mice, driving abnormal neuronal differentiation.

[0041] The detection kit (containing 22 gene-specific primers) and analysis system (RF algorithm + SHAP visualization) of the present application can be directly used for clinical or model research.

[0042] The present application realizes the efficient screening of diet-PD associated genes through multi-omics integration and machine learning optimization; 22 core genes and their detection / analysis system provide a powerful tool for PD mechanism analysis, early warning and precise nutritional intervention. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1. The transcriptomic landscape of PD is shown; where, (A), DEGs of PD vs. Control in GSE8397, red represents up-regulation, and blue represents down-regulation; (B), the fitting index and average connectivity of soft threshold vs. scale-free topology model in GSE8397, 19 is chosen as the optimal soft threshold; (C), the association of gene modules with different traits in GSE8397, the blue module is the most specific for PD; (D), DEGs of PD vs. Control in GSE20295, red represents up-regulation, and blue represents down-regulation; (E), the fitting index and average connectivity of soft threshold vs. scale-free topology model in GSE20295, 19 is chosen as the optimal soft threshold; (F), the association of gene modules with different traits in GSE20295, the blue module is the most specific for PD; (G), GO and KEGG enrichment analysis of DEGs in GSE8397; (H), GO and KEGG enrichment analysis of DEGs in GSE20295. (I), GSEA analysis of DEGs in GSE8397, using C2 as the reference gene set; (J), GSEA analysis of DEGs in GSE8397, using C5 as the reference gene set. (K), GSEA analysis of DEGs in GSE20295, using C2 as the reference gene set; (L), GSEA analysis of DEGs in GSE20295, using C5 as the reference gene set.

[0044] Figure 2 . The expression levels of key genes for diet phenotype affecting PD in dopaminergic neurons; where, (A), the upset plot of the intersection of high-cholesterol diet target gene set, PD target gene set and DEGs of two datasets; (B), 120 kinds of predictive model combinations using 10-fold cross-validation and ranking AUC index; (C), the feature importance ranking of high-cholesterol diet affecting PD; (D), the SHAP bee plot of high-cholesterol diet affecting PD.

[0045] Figure 3 . The expression levels of key genes for diet phenotype affecting PD in dopaminergic neurons; where, (A), the upset plot of the intersection of high-cholesterol diet target gene set, PD target gene set and DEGs of two datasets; (B), 120 kinds of predictive model combinations using 10-fold cross-validation and ranking AUC index; (C), the feature importance ranking of high-cholesterol diet affecting PD; (D), the SHAP bee plot of high-cholesterol diet affecting PD.

[0046] Figure 4Expression levels of key genes for diet phenotypes affecting PD in dopaminergic neurons; where (A), upset plot of the high-fat diet target gene set, the PD target gene set, and the DEGs of the two datasets intersected; (B), 120 predictive model combinations using 10-fold cross-validation and the ranking AUC index; (C), feature importance ranking of high-fat diet affecting PD; (D), SHAP swarm plot of high-fat diet affecting PD.

[0047] Figure 5 Expression levels of key genes for diet phenotypes affecting PD in dopaminergic neurons; where (A), upset plot of the high-carbohydrate diet target gene set, the PD target gene set, and the DEGs of the two datasets intersected; (B), 120 predictive model combinations using 10-fold cross-validation and the ranking AUC index; (C), feature importance ranking of high-carbohydrate diet affecting PD; (D), SHAP swarm plot of high-carbohydrate diet affecting PD.

[0048] Figure 6 Expression levels of key genes for diet phenotypes affecting PD in dopaminergic neurons; where (A), upset plot of the egg intake target gene set, the PD target gene set, and the DEGs of the two datasets intersected; (B), 120 predictive model combinations using 10-fold cross-validation and the ranking AUC index; (C), feature importance ranking of egg intake affecting PD; (D), SHAP swarm plot of egg intake affecting PD.

[0049] Figure 7 Expression levels of key genes for diet phenotypes affecting PD in dopaminergic neurons; where (A), upset plot of the whole grain intake target gene set, the PD target gene set, and the DEGs of the two datasets intersected; (B), 120 predictive model combinations using 10-fold cross-validation and the ranking AUC index; (C), feature importance ranking of whole grain intake affecting PD; (D), SHAP swarm plot of whole grain intake affecting PD.

[0050] Figure 8 Expression levels of key genes for diet phenotypes affecting PD in dopaminergic neurons; where (A), upset plot of the lamb intake target gene set, the PD target gene set, and the DEGs of the two datasets intersected; (B), 120 predictive model combinations using 10-fold cross-validation and the ranking AUC index; (C), feature importance ranking of lamb intake affecting PD; (D), SHAP swarm plot of lamb intake affecting PD.

[0051] Figure 9Expression levels of key genes affecting PD in dopaminergic neurons for diet phenotypes; (A), upset plot of the intersection of fish intake target gene set, PD target gene set and DEGs of two datasets; (B), 120 kinds of prediction model combinations using 10-fold cross-validation and ranking AUC index; (C), feature importance ranking of fish intake affecting PD; (D), SHAP bee plot of fish intake affecting PD.

[0052] Figure 10 qPCR experiment results; (A) for human midbrain organoids; (B) for mouse substantia nigra. DETAILED DESCRIPTION

[0053] The following examples further illustrate the present application but should not be construed as limiting the application. Modifications or variations of the method, steps or conditions of the application can be made by those skilled in the art without departing from the spirit and scope of the application.

[0054] Unless otherwise specifically indicated, the techniques used in the examples are conventional techniques known to those skilled in the art.

[0055] A method for screening diet-Parkinson's disease related genes, comprising the following steps:

[0056] (a) performing differential analysis on transcriptome datasets GSE8397 and GSE20295 to obtain differentially expressed genes DEGs, respectively;

[0057] (b) constructing gene modules by WGCNA: selecting soft threshold 19 to construct 3 modules for DEGs of GSE8397, and selecting soft intensity 14 to construct 5 modules for DEGs of GSE20295;

[0058] (c) integrating diet phenotype target gene set, PD target gene set and DEGs of step (a) to cross-screen candidate genes; the diet phenotypes include high cholesterol diet, high sugar diet, high fat diet, high carbohydrate diet, egg intake, whole grain intake, mutton intake and fish intake;

[0059] (d) constructing gene expression analysis model for candidate genes using integrated machine learning algorithm, and screening key genes using SHAP value.

[0060] Step (a) provides a molecular target basis for subsequent multi-omics integration by identifying significant DEGs in the disease state. In the embodiments of the present application, the data sets: GSE8397 (29 cases of PD vs 12 cases of control) and GSE20295 (11 cases of PD vs 23 cases of control) are taken from the GEO database, platform GPL96. Preprocessing process: RMA (Robust Multi-array Average) background correction and quantile (Quantile Normalization) normalization are performed using the R package affy. Probe annotation: GPL96 platform maps gene symbols (Homo sapiens). Tool: R package limma (v3.50.0). Threshold: |log2FC|>0.5 and adj.P<0.05 (Benjamini-Hochberg correction). Results: GSE8397: 445 DEGs (142↑, 303↓); GSE20295: 947 DEGs (351↑, 596↓).

[0061] Step (b) builds gene function modules based on differentially expressed genes (DEGs) through WGCNA. Specifically, hierarchical clustering is used to detect and remove outliers, and a suitable soft power value is selected to construct a weighted adjacency matrix. Then the matrix is converted into a topological overlap matrix, and gene modules are identified through dynamic tree cutting and given unique color labels. The module eigengenes (ME) are calculated to represent the expression pattern of each module, and the Pearson correlation coefficient is used to evaluate the correlation between each module and the Parkinson's disease state, identifying gene clusters that are highly co-regulated with Parkinson's disease (PD) and providing a topological correlation basis for multi-omics integration. In the embodiments of the present application, the soft threshold of data set GSE8397 is selected as β=19, and the scale-free topological fitting index R 2 >0.85 (satisfies the scale-free network requirement), and the average connection degree tends to be flat (avoiding excessive connection); the soft threshold of data set GSE20295 is selected as β=14, and the connection degree distribution is stable (R 2 >0.80), and the network connectivity is optimal when β=14.

[0062] Step (c) integrates the diet phenotype target gene set, the PD pathology target gene set, and the differentially expressed genes (DEGs) to precisely lock the core candidate genes that mediate the diet-PD association using a three-way cross-screening strategy, significantly improving targeting and reducing redundancy.

[0063] In step (c), the construction of the diet phenotype target gene set includes:

[0064] The database GWAS catalog, GeneCards, UKBB GWAS and KoGES were searched with the species limited to "Homosapiens"; in GeneCards, genes with a score higher than the median (for example, the median of the high-sugar diet score = 5.0) were selected; in UKBB GWAS and KoGES, P<10 -6 The genes were annotated with the threshold value.

[0065] The PD target gene set includes: DrugBank+GeneCards, with the keyword "Parkinson disease" as the screening keyword; KEGG pathway: 23 PD-related pathways (such as hsa05012), and 3907 genes were obtained after merging and deduplication.

[0066] The candidate genes = (diet target ∩ PD target) ∩ DEGs.

[0067] In step (d), an integrated machine learning algorithm is used to construct an analysis model for the candidate genes, the gene contribution degree is quantified through the SHAP interpretability framework, 22 key genes mediating the diet-PD association are screened out, and the transformation from association analysis to mechanism analysis is realized. In the embodiments of the present application, 14 algorithms and 195 parameter combinations (including neural networks, SVM, XGBoost, etc.) are tested, and the result is that the random forest (RF) has the best performance in all diet phenotypes. In this example, the key parameter mtry (the number of random genes at the split node) is determined by grid search. GSE8397 is used as the training set (10-fold cross-validation), and GSE20295 is used as the independent test set. The main indicators are AUC (>0.88) and accuracy (>0.76), and the secondary indicators are sensitivity and specificity.

[0068] The machine learning algorithm in step (d) is a random forest algorithm, and the parameter combination is:

[0069] High cholesterol diet: mtry = 13; high-sugar diet: mtry = 7 or mtry = 19; high-fat diet: mtry = 28; high-carbohydrate diet: mtry = 8; egg intake: mtry = 8; whole grain intake: mtry = 7; lamb intake: mtry = 5; fish intake: mtry = 12.

[0070] Preferably, the key genes in step (d) are the top-ranked genes of each dietary phenotype SHAP value, and the obtained genes are de-duplicated. In a specific embodiment of the present application, 22 genes are finally integrated: DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

[0071] 22 genes cover the PD core pathological mechanisms, including:

[0072] Dopamine synthesis: DDC, TH;

[0073] Alpha-synuclein: SNCA;

[0074] Neuroinflammation: NFKBIA, HSPB1;

[0075] Metabolic regulation: HMGCR, GBE1.

[0076] The related gene function verification shows that, for example, in the A53T transgenic mouse test after high-fat diet feeding, FOXA1 and SNCA are significantly up-regulated (FC>2.0, P<0.01); in the human midbrain organoid test, TH is down-regulated by 40% under the intervention of steroids (P=0.003).

[0077] Example 1 Construction of the method for screening diet-Parkinson's disease (PD) associated genes

[0078] I. Materials and methods

[0079] 1.1 Collection of dietary phenotype targets

[0080] The present application collects target information of 8 dietary phenotypes including 4 dietary structures and 4 food intake amounts by using databases (GWAS catalog, GeneCards, UKBB GWAS and KoGES), specifically: high cholesterol diet (2174), high sugar diet (2464), high fat diet (3557), high carbohydrate diet (2241); egg intake (1002), whole grain intake (589), lamb intake (499) and fish intake (1382). The process uses corresponding keywords for retrieval, and the species is limited to "Homo sapiens". In GeneCard, the "score" threshold is set to the median, and genes with "score" higher than the median are selected; in UKBB GWAS and KoGES, the present application uses annovar for gene annotation, and selects genes with P<10E-6 as the threshold to ensure that the obtained genes are highly related to the phenotype. After converting the gene alias to the standard gene symbol using the "alias2Symbol" function in the R package "limma", merging and deduplicating, the target gene set of the dietary phenotype is compiled.

[0081] 1.2 Construction of PD target

[0082] The present application obtains the target information of PD through databases (DrugBank, GeneCards, GWAS catalog, OMIM, Human Phenotype Ontology) and 23 PD-related pathways including "hsa05012: Parkinson disease". A total of 1928 PD targets are obtained by searching these databases using the "Parkinson disease" keyword, and a total of 2415 targets are obtained by counting the genes covered by the 23 PD-related pathways. After converting the gene alias to the standard gene symbol using the "alias2Symbol" function in the R package "limma", merging and deduplicating, the PD target gene set is obtained, which contains a total of 3907 PD targets.

[0083] 1.3 Sample collection and data preprocessing

[0084] The present application downloaded two human PD datasets (GSE8397 and GSE20295) from GEO database (https: / / www.ncbi.nlm.nih.gov / geo / ), both of which were converted to corresponding gene symbols by referring to GPL96 platform annotation information. The present application used R package "affy" to read probe data and perform background correction and normalization, converting original probe-level signal intensity to gene-level expression. The GSE8397 dataset contains 39 SN samples (29 PD samples and 12 Control samples), and the GSE20295 dataset contains 34 SN samples (11 PD samples and 23 Control samples).

[0085] 1.4 Differential analysis and weighted gene co-expression network analysis

[0086] The present application used R package "limma" to compare the expression profile data of PD samples and normal samples in the two datasets respectively to determine their DEGs, with a threshold of |log2FoldChange|>0.5 and P<0.05. R package "WGCNA" was used to perform WGCNA on the DEGs of the two datasets. Hierarchical clustering was performed to eliminate outliers, and appropriate soft power was selected to construct a weighted adjacency matrix, which was converted into a topological overlap matrix (TOM) characterized by different colors and module characteristic genes (MEs). Finally, the Pearson correlation coefficient was calculated to evaluate the correlation between each ME and the PD sample, and the module genes were checked.

[0087] 1.5 Functional enrichment analysis

[0088] The present application used R package "enrichplot" to perform gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis on the DEGs of the two datasets. C2:curated gene sets and C5:ontology genesets were used for GSEA enrichment analysis, with P value less than 0.05 considered significant

[0089] 1.6 Functional enrichment analysis comprehensive machine learning to construct PD prediction model and feature selection

[0090] In order to develop a diagnostic model with high prediction accuracy, the present application followed the following steps:

[0091] (1) The present application uses GSE8397 as a model training set and GSE20295 as a test set. The present application adopts 14 kinds of comprehensive machine learning algorithms, including neural network Neural Network (MLP), logistic regression Logistic Regression, linear discriminant analysis Linear discriminant analysis, quadratic discriminant analysis Quadratic discriminant analysis, KNN nearest neighbor k-Nearest Neighbor, decision tree classification Decision tree, random forest classification Random forest, XGBoost classification, ridge regression ridge regression, elastic network regression Elastic Net regression, support vector machine Support vector machine, gradient boosting machine Grandient Boosting Machine, stepwise logistic regression model stepwise+Logistic Regression, and naive Bayesian Naive Bayesian algorithm. 195 combinations of the 14 algorithms are applied to the training set, and 10-fold cross-validation is used for feature selection and model construction.

[0092] (2) Calculate the prediction accuracy of the training set and the validation set, calculate the average area under the receiver operating characteristic curve (AUC), rank all models according to the AUC index, take the top 120 models for visualization, and take the first ranked model as the most effective algorithm combination.

[0093] (3) Use SHAP as an explanation model for prediction and feature selection, use the R package "shapviz" to calculate the shap value of each candidate gene, and sort and explain the contribution of each feature to the model (the lower the importance, the less stable the feature ranking).

[0094] (4) Take the top 10 features in the importance score of each dietary phenotype as the key genes of the dietary phenotype affecting PD for subsequent analysis.

[0095] 1.7 Mouse culture

[0096] Wild-type C57BL / 6J mice, A53T transgenic mice were purchased from Hangzhou Ziyuan Animal Experimental Animal Technology Co., Ltd., with a qualified certificate number of (20241029Abbb0105000546). The present application uses 6-month-old A53T transgenic homozygous male mice and littermate female C57BL / 6J mice for experiments. All mice were raised under specific pathogen-free conditions, with a 14-hour light / 10-hour dark cycle, and free access to food and water. All animal care and handling were performed in accordance with the Helsinki Declaration and were approved. The mice were randomly divided into different groups using a random number table. During the animal experiment, the researchers were blinded to the grouping. The experimental protocol was reviewed and approved by the Experimental Animal Center of Henan University of Chinese Medicine. The IACUC approval number is 202410034.

[0097] 1.8 Midbrain organoid culture

[0098] H9 hESCs were cultured in a 5% C02incubator at 37°C using standard procedures. The H9 hESC cell line was cultured in the absence of feeder layer on Matrigel (Corning) with mTeSR (STEMCELL). hESCs were dissociated into single cells with Accutase (Sigma) and 10,000 cells were seeded in each well of a low cell attachment 96-well culture plate with round bottom conical wells (Sumitomo Bakelite) to form uniform embryoid bodies in neural induction medium. The neural induction medium on day 0 was DMEM / F12: Neurobasal (1 : 1 ; Gibco), N-2 (1 %; Gibco), Gluta Max (1 %; Gibco), MEM NEM (1 %; Gibco), B-27 without vitamin A (2%; Gibco), SB431542 (10 mM; MCE), CHIR (0.8 mM; MCE), hapar (1 pg / ml; MCE), and Y27632 (10 mM; Miltenyi Biotec). Y27632 was supplemented for 2 days only after cell passaging, and the neural induction medium on day 2 was the same as day 0 but without Y27632. The medium was changed on day 4, which was the same as day 2 but with the addition of SHH (100 ng / ml; MCE) and FGF8b (100 ng / ml; MCE). The mesencephalic organoids were embedded in 30 pl of Matrigel droplet on day 7 and after 30 min of fixation in a 37°C incubator, they were transferred to neural maturation medium containing Neurobasal, 1 % N2 supplement, 2% B27(without VA) supplement, 1 % GlutaMAX, 1 % MEM NEM, 100 ng / ml SHH, 100 ng / ml FGF8b, and 200 ng / ml Laminin (MEMD millipore) and cultured in ultra-low attachment six-well plates (Servicebio). The embedded mesencephalic organoids were cultured under fixed conditions for 1 day and then transferred to an orbital shaker on day 8 and the medium was changed to Neurobasal, 1 % N2 supplement, 2% B27(without VA) supplement, 1 % GlutaMAX, 1 % MEM NEM, 10 ng / ml BDNF, 10 ng / ml GDNF, 100 mM VC (MCE), and 100 mM dCAMP (MCE) for 3 days. The medium was changed on day 15, which was the same as day 8 but without B27 without vitamin A and with the addition of B27 with vitamin A, and this was repeated for 3 days until maturation at 90 days.

[0099] 1.9 qPCR experiments

[0100] RT-PCR and RT-qPCR were used to analyze the expression levels of the 22 target genes mRNA in A53T transgenic mice (Control is the healthy control group, A53T is the disease group, 3 samples in each group) and midbrain organoids (A17 is the healthy control group, A16 is the disease group, 3 samples in each group) using Actin as the internal reference. The total RNA in the substantia nigra and organoids was extracted using Vazyme Cell / Tissue Total RNA Isolation Kit V2 (RC112), and the extracted total RNA was reverse transcribed into cDNA using HiScript IV All-in-One Ultra RT SuperMix for qPCR (R433) kit. The primers of the 22 target genes were designed through the NCBI website and then quantitative PCR was performed (Table 1 and Table 2). The quantitative reaction was performed by ChamQ Blue Universal SYBR qPCR Master Mix kit according to the actual and instructions, and the relative quantitative analysis was performed by ΔΔCt method. Each gene was measured 3 times in duplicate wells.

[0101] Table 1 Design of mouse-derived primers

[0102]

[0103]

[0104]

[0105] Table 2 Design of human-derived primers

[0106]

[0107]

[0108] II. Results

[0109] 2.1 Differential analysis and WGCNA

[0110] The present application performed differential analysis and WGCNA on two transcriptome datasets (GSE8397 and GSE20295) respectively. A total of 445 DEGs were obtained from GSE8397, including 142 up-regulated genes and 303 down-regulated genes ( Figure 1 A). Figure 1 B) WGCNA was performed on these DEGs by selecting soft strength 19, and a total of 3 modules ( Figure 1 C) were obtained. The highest correlation with PD was the blue module (0.77), which covered 147 genes.

[0111] GSE20295 got 947 DEGs, including 351 up-regulated genes and 596 down-regulated genes Figure 1 D) WGCNA was performed on these DEGs with soft strength 14 Figure 1 E) 5 modules were got Figure 1 F) The highest correlation with PD was brown module (0.60), which covered 206 genes.

[0112] 2.2 Functional enrichment analysis

[0113] Functional enrichment analysis was performed on DEGs from two transcriptome datasets. GO and KEGG results showed that DEGs from GSE8397 were significantly enriched in “vesicle-mediated transport in synapse”, “neuronal cell body”, “transporter regulator activity” and “Neurodegenerative disease” and other pathways Figure 1 G) DEGs from GSE20295 were significantly enriched in “energy derivation by oxidation of organic compounds”, “transport vesicle”, “transport vesicle” and “Neurodegenerative disease” and other pathways Figure 1 H) GSEA results showed that the differential genes of the two datasets were enriched in “BLALOCK_ALZHEIMERS_DISEASE_UP”, “KIM_ALL_DISORDERS_CALB1_CORR_UP” “GOBP_CELL_CELL_SIGNALING”, “GOBP_SYNAPTIC_SIGNALING” and “GOCC_PRESYNAPSE” and other pathways Figure 1 I-L) The results showed that the functional enrichment of the two datasets was consistent.

[0114] 2.3 Construction of PD prediction model based on integrated machine learning

[0115] By crossing the diet phenotype target gene set, the PD target gene set and the DEGs of the two datasets, the candidate genes affected by diet phenotype on PD were obtained, including high cholesterol diet (25, such as Figure 2A) high sugar diet (25), high fat diet (37), high carbohydrate diet (27), egg intake (10), whole grain intake (9), lamb intake (6) and fish intake (15).

[0116] To determine the effect of diet phenotype targets on PD, the present application used 195 combinations of 14 machine learning algorithms for feature selection and model development, and visualized the accuracy of the top 120 models. In the training set, the present application used 10-fold cross-validation to evaluate each algorithm combination and calculate the average AUC value. It is worth noting that the random forest model under different parameters showed excellent performance in 8 diet phenotypes. In the high cholesterol diet, RF(mtry=13, 50% p) achieved an accuracy of 1.000 in the training set, an accuracy of 0.794 in the test set, and an average AUC of 0.897. The performance of XGBoost(cutoff: 0.75) was the same as RF in the high cholesterol diet (Best model: RF(mtry=13, 50% p), training set: 1.000, test set: 0.794, AUC: 0.897). Figure 2 B); in the high sugar diet, RF(mtry=7, 25% p) and RF(mtry=19, 75% p) showed consistent prediction performance, with an accuracy of 1.000 in the training set and an accuracy of 0.824 in the test set, and an average AUC of 0.912 (Figure S1B); in the high fat diet, RF(mtry=28, 75% p) achieved an accuracy of 1.000 in the training set and an accuracy of 0.882 in the test set, and an average AUC of 0.94; in the high carbohydrate diet, RF(mtry=8, 25% p) achieved an accuracy of 1.000 in the training set and an accuracy of 0.882 in the test set, and an average AUC of 0.941. In the 4 food intake, RF still showed excellent performance, with the lowest accuracy of 0.765 in the test set and an average AUC of 0.882; the specific cases were egg intake (Best model: RF(mtry=8, 75% p), training set: 1.000, test set: 0.765, AUC: 0.882), whole grain intake (Best model: RF(mtry=7, 75% p), training set: 1.000, test set: 0.794, AUC: 0.897), lamb intake (Best model: RF(mtry=5, 75% p), training set: 1.000, test set: 0.765, AUC: 0.882) and fish intake (Best model: RF(mtry=12, 75% p), training set: 1.000, test set: 0.765, AUC: 0.882).

[0117] 2.4 SHAP value explains the importance of diet intake targets

[0118] Feature selection methods can be used to reduce the number of useless features, thereby reducing the complexity of the final model. Since the feature attribution method of RF model is inconsistent, it may hinder the meaningful comparison of feature attribution values, while SHAP values can better explain the impact and importance of individual features. Therefore, the present application uses SHAP values to explain the impact and importance of individual features, ranks each feature according to the SHAP value, and uses the top 10 features with the largest contribution as the key genes of diet phenotype affecting PD (e.g. Figure 2 C), After integration, a total of 22 key genes (DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA) were obtained.

[0119] The swarm plot shows the summary of SHAP, which is color-coded according to the value of the corresponding variable on the y-axis, and color-coded using their associated Shapley values on the x-axis. The present application obtains the contribution magnitude and directionality of diet intake target points to PD prediction according to the value of the variable, while maintaining the importance ranking. For example, in high-sugar diet, samples with low expression of DLD and KCND3 have positive SHAP values, and samples with high expression of DLD and KCND3 have negative SHAP values; This indicates that samples with high expression of DLD and KCND3 will have a lower probability of being predicted as PD. This is also true in other diet intakes.

[0120] 2.5 qPCR experimental results of mouse substantia nigra and human midbrain organoids

[0121] To evaluate the molecular characteristics associated with diet-induced PD pathology, the present application performed qPCR detection on 22 key genes, and the samples were derived from human midbrain organoids and mouse substantia nigra tissue. In the midbrain organoids ( Figure 10 A), the comparison between A16 and A17 groups showed that AUH, CAPN3, HPRT1, ITPR1, L1CAM, NEFL, TAC1 and TH were significantly up-regulated in A16 organoids (P<0.05), while FOXA1, IRS1, KCND3, PBX1 and SNCA were significantly up-regulated in A16 organoids (P<0.05). In contrast, in the substantia nigra of A53T transgenic mice ( Figure 10B) More extensive transcriptional dysregulation was observed. In particular, FOXA1, GBE1, TH, KCND3 and PBX1 were significantly upregulated in A53T mice (P < 0.001), suggesting that the phenotype and electrophysiological properties of dopaminergic neurons might be altered. In addition, AUH, CAPN3, DDC, ELOVL4, HMGCR, HPRT1, NEFL and NFKBIA also showed varying degrees of difference (P < 0.05 to P < 0.001), involving mitochondrial function, inflammatory pathways and cytoskeleton regulation. Notably, SNCA was significantly upregulated in A53T mice (P < 0.05), consistent with the overexpression of mutant alpha-synuclein. In summary, these results reveal that human midbrain organoids and mouse models exhibit both overlapping and divergent transcriptional responses under PD pathological conditions, and that the organoid system to some extent reflects the molecular pathological processes related to Parkinson's disease, but at the same time highlights the species- and model-specific regulatory mechanisms. In summary, these results reveal that human midbrain organoids and mouse models exhibit both overlapping and divergent transcriptional responses under PD pathological conditions, and that the organoid system to some extent reflects the molecular pathological processes related to Parkinson's disease, but at the same time highlights the species- and model-specific regulatory mechanisms, indicating that they can be identified as markers.

[0122] A detection kit

[0123] The example provides a detection kit, which includes specific primer pairs for detecting the expression level of one or more of DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

[0124] Specific primers are shown in Table 2.

[0125] Further, a fluorescent probe or dye, a DNA polymerase, dNTPs, a reaction buffer, a positive control (a plasmid containing the gene), and nuclease-free water can be further included.

Claims

1. A method of screening for a diet-Parkinson's disease association gene, characterized by Comprising the following steps: (a) performing differential analysis on transcriptome datasets GSE8397 and GSE20295 to obtain differentially expressed genes (DEGs) respectively; (b) constructing gene modules by weighted gene co-expression network analysis (WGCNA): selecting soft strength 19 to construct 3 modules for DEGs of GSE8397, and selecting soft strength 14 to construct 5 modules for DEGs of GSE20295; (c) integrating diet phenotype target gene set, PD target gene set and DEGs of step (a) to cross-screen candidate genes; the diet phenotypes include high cholesterol diet, high sugar diet, high fat diet, high carbohydrate diet, egg intake, whole grain intake, mutton intake and fish intake; (d) constructing a gene expression analysis model for candidate genes using an integrated machine learning algorithm, and screening key genes using SHAP values.

2. The method of claim 1, wherein: The construction of the diet phenotype target gene set in step (c) comprises: Using databases GWAS catalog, GeneCards, UKBB GWAS and KoGES to search with species limited to "Homo sapiens"; Selecting genes with scores higher than the median in GeneCards; UKBB GWAS and KoGES were performed with P < 10 -6 Gene annotation was performed with a threshold of P < 0.

05.

3. The method of claim 1, wherein: The machine learning algorithm in step (d) is a random forest algorithm, and the parameter combination is: High cholesterol diet: mtry=13; High sugar diet: mtry=7 or mtry=19; High fat diet: mtry=28; High carbohydrate diet: mtry=8; Egg intake: mtry=8; Whole grain intake: mtry=7; Mutton intake: mtry=5; Fish intake: mtry=12.

4. The method of claim 1, wherein: The key genes in step (d) are the top genes in the SHAP value ranking of each diet phenotype, and the obtained genes are de-duplicated.

5. The method of claim 4, wherein: Finally, 22 genes are obtained: DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

6. One of the DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA genes or a combination thereof in the preparation of a reagent for detecting Parkinson's disease.

7. A test kit characterized in that Specific primer pairs for detecting the expression level of one or more of DDC, DLD, ELOVL4, FOXA1, GBE1, HK1, HMGCR, HPRT1, HSPB1, ITPR1, KCND3, PBX1, SNCA, TAC1, TGIF1, TH, AUH, CAPN3, IRS1, L1CAM, NEFL and NFKBIA.

8. A diet-PD association analysis system, characterized by Comprising: Data input unit: receiving the expression data of the 22 genes in claim 6; Algorithm processing unit: using random forest algorithm to analyze the correlation between gene expression and dietary phenotype, with the following parameter configurations: High cholesterol diet: mtry=13; High sugar diet: mtry=7 or 19; High fat diet: mtry=28; High carbohydrate diet: mtry=8; Visual output unit: generating SHAP swarm plot to show gene contribution.

9. Use of midbrain organoids in the study of diet-PD mechanism, characterized by: Using the kit in claim 7 to detect the expression level of the 22 genes in claim 6 in the organoids; Analyzing the correlation between gene expression changes and dietary intervention, wherein the organoids are cultured by the following methods: ① Differentiating H9 hESC cells in a neural induction medium containing SHH and FGF8b; ② Transferring to a maturation medium containing BDNF and GDNF on the 8th day and culturing for 90 days.

10. A method for constructing an animal model for Parkinson's disease research, characterized by Comprising: Feeding A53T transgenic mice with high cholesterol / high sugar / high fat / high carbohydrate feed; Using the kit in claim 7 to detect the expression of the 22 genes in claim 6 in the substantia nigra tissue; When FOXA1, GBE1, TH, KCND3, PBX1 and SNCA are significantly up-regulated, it is determined that the model is successfully constructed.

Images