Method for screening endogenous biomarkers in serum of depression patient

Abstract

The invention discloses a method for screening endogenous biomarkers in serum of a patient with depression, which comprises the following steps: collecting serum samples of the patient with depression and a healthy control, and screening the endogenous biomarkers in serum of the patient with depression based on ultra-high performance liquid chromatography-orbital ion trap high-resolution mass spectrometry by coupling non-targeting and targeting screening technical means. The method comprises the following steps: representing inter-group differences of endogenous small molecule metabolites in serum, identifying differential metabolites between a depression patient and a healthy control group, and then determining a biomarker closely related to depression by using a regression model. And constructing a depression diagnosis model by respectively taking the single biomarker and the specific value as input variables and taking a binary variable for diagnosing whether depression occurs as an output variable. According to the invention, high-sensitivity and full-coverage screening of serum metabolites is realized by means of coupling targeting-non-targeting metabonomics, multiple biomarkers closely associated with depression are identified, and basic data can be provided for analysis of the occurrence mechanism of depression and development of an early diagnosis method.

Description

Technical Field

[0001] The invention belongs to the field of biotechnology and medical diagnosis, and relates to a method for screening endogenous biomarkers in the serum of patients with depression. Background Art

[0002] Depression is a common mental illness and a major cause of disability. It impairs cognitive function and increases the risk of cardiovascular metabolic diseases and related brain diseases. The social burden caused by depression prompts the health system to invest a lot of resources every year to improve mental health, but often with little success.

[0003] Depression is a disease that is strongly affected by endogenous factors. Most patients will have varying degrees of disorder in the metabolism of energy, lipids, amino acids, and nucleic acids. The conclusions drawn from current studies on the pathogenesis of depression lack consistency. Identifying specific metabolic biomarkers associated with depression can help determine the pathogenesis of depression and provide reference value for further clinical intervention and management of depression in the future. Metabolomics, as an important omics technology, has been widely used in metabolic feature description, phenotype determination, and disease-specific biomarker identification. It systematically identifies and quantifies key metabolites of biochemical pathways affected by pathological or stressful conditions (such as exposure to environmental toxins) in biological samples.

[0004] Metabolomics can be divided into targeted and untargeted approaches. Targeted metabolomics focuses on accurate and selective quantification of known metabolites by isolating specific targets, which improves sensitivity but limits the coverage of metabolites to a small to moderate number (tens to hundreds) of known compounds. In contrast, untargeted metabolomics sacrifices some sensitivity for a broader detection capability, focusing on acquiring data on as many species as possible, annotating metabolites, and reviewing known and unknown metabolic changes. The data can be used for relative quantification between sample groups and provide hypotheses that can be further studied through targeted approaches. Combined analysis of untargeted and targeted metabolomics can improve the coverage and sensitivity of metabolite identification, providing a promising strategy for comprehensive characterization of trace and high-abundance metabolite markers.

[0005] At present, psychiatrists mostly use the Patient Health Questionnaire-9 (PHQ-9) or the Hamilton Rating Scale for Depression (HAMD) to screen for depression. Since these questionnaires are self-assessment questionnaires, heterogeneous response behaviors cannot be avoided. Although the diagnostic accuracy is usually not low for psychiatrists with clinical experience, thorough clinical interviews still require time and experience, and there is an urgent need to develop effective rapid diagnostic tools for depression. Summary of the invention

[0006] The purpose of the present invention is to provide a method for screening endogenous biomarkers in the serum of patients with depression. The screening method is as follows: serum samples of patients with depression and healthy controls are collected, and based on ultra-high performance liquid chromatography-orbital ion trap high-resolution mass spectrometry, non-targeted and targeted screening techniques are coupled to characterize the inter-group differences in endogenous small molecule metabolites in serum, identify the differential metabolites between patients with depression and healthy controls, and then use the LASSO regression model to determine biomarkers closely related to depression, including 11 metabolites such as thymidine ribonucleoside, N-acetylglycine, and phenylalanine. Based on a machine learning algorithm, a single biomarker and a biomarker ratio are used as input variables, and a binary variable "diagnosing whether depression occurs" is used as an output variable to construct a depression diagnosis model.

[0007] The purpose of the present invention is achieved through the following technical solutions: A method for screening endogenous biomarkers in the serum of patients with depression, comprising the following steps: (1) Serum collection: Serum was collected from healthy controls and patients with depression and mixed with isotope-labeled internal standards for further analysis; The depression patients include moderate depression patients (MOD) and major depression patients (MDD); The serum is further purified; The purification is to mix the serum with pre-cooled methanol, incubate for several hours, and centrifuge at below 4°C to remove proteins; (2) Non-targeted metabolomics analysis: Step (1) The serum sample is analyzed by ultra-high performance liquid chromatography and quadrupole / colorimetric high-resolution mass spectrometry to collect a number of fragment ion spectra; the fragment ion spectra are further screened and analyzed to screen out a number of compounds; The screening analysis includes one or more steps of peak extraction, integration, missing value filling, feature filtering and metabolite annotation; (3) Targeted metabolomics analysis: Step (1) Serum samples were subjected to ultra-high performance liquid chromatography tandem with quadrupole / orbitrap high-resolution mass spectrometry to construct a high-throughput detection method that can simultaneously detect multiple metabolites. The chromatographic peaks of each compound were extracted based on the precise molecular mass of the parent ion, and quantitative analysis was performed using the isotope internal standard method. (4) Screening for differential metabolites between patients with depression and healthy controls: The metabolites identified by non-targeted metabolomics analysis and targeted metabolomics analysis were combined for analysis, and the common differential metabolites between patients with depression and healthy controls were used as endogenous markers of depression; The combined analysis preferably uses a supervised multivariate statistical analysis method (such as partial least squares discriminant analysis (PLS-DA)), and evaluates whether the model is overfitting through a permutation test, and screens differential metabolites based on a rank sum test and a difference fold analysis.

[0008] Endogenous biomarkers in the serum of patients with depression are the following compounds: N-methyl-L-alanine, γ-aminobutyric acid, N-acetylneuraminic acid, N-acetylglycine, L-methionine, L-phenylalanine, thymidine ribonucleoside, xanthine, decanoylcarnitine, aspartate-phenylalanine, and uric acid.

[0009] The above-mentioned endogenous biomarkers in the serum of patients with depression can be used to construct a depression diagnostic model.

[0010] The concentration and concentration ratio of endogenous biomarkers in the serum of patients with depression can be used to construct a depression diagnosis model; The endogenous biomarkers in the serum of patients with depression are: N-methyl-L-alanine, gamma-aminobutyric acid, N-acetylneuraminic acid, N-acetylglycine, L-methionine, L-phenylalanine, and thymidine ribonucleoside.

[0011] The ratio of N-acetylglycine to thymidine ribonucleoside concentration can be used to construct a diagnostic model for depression.

[0012] A method for constructing a depression diagnosis model comprises the following steps: S1: Obtain multiple endogenous biomarkers in serum; S2: Establish training sample sets and validation sample sets based on various endogenous biomarkers; (the training samples in each training sample set belong to the same endogenous biomarker in serum) S3: training machine learning models respectively by using training sample sets corresponding to various endogenous biomarkers to obtain various primary diagnostic models; wherein, during the training process, the concentration and concentration ratio of the endogenous biomarkers are used as input features of the machine learning model, and the depression status of the serum subject collected by the endogenous biomarkers is used as a label; the depression status includes suffering from depression and not suffering from depression; The machine learning model may be a random forest model, a neural network model, or others; S4: For various primary diagnostic models, the accuracy is verified by validation sample sets corresponding to various endogenous biomarkers; S5: Select the primary diagnostic model with the highest accuracy as the final depression diagnostic model.

[0013] A device for constructing a depression diagnosis model, comprising: An acquisition module, used to obtain multiple endogenous biomarkers in serum; A sample construction module, used to establish a training sample set and a validation sample set based on various endogenous biomarkers; A training module, which trains machine learning models respectively through training sample sets corresponding to various endogenous biomarkers to obtain various primary diagnostic models; wherein, during the training process, the concentration and concentration ratio of the endogenous biomarkers are used as input features of the machine learning model, and the depression status of the serum subject collected by the endogenous biomarkers is used as a label; the depression status includes suffering from depression and not suffering from depression; The machine learning model may be a random forest model, a neural network model, or others; Verification module: For various primary diagnostic models, the accuracy is verified through the verification sample sets corresponding to various endogenous biomarkers; Selection module: Select the primary diagnostic model with the highest accuracy as the final depression diagnosis model.

[0014] Compared with the prior art, the present invention has the following advantages and effects: 1. The present invention achieves high-sensitivity and full-coverage screening of serum metabolites by coupling targeted-nontargeted metabolomics, and identifies a variety of biomarkers closely related to depression, which can provide basic data for the analysis of the pathogenesis of depression and the development of early diagnosis methods.

[0015] 2. In the rapid diagnosis model for depression constructed by the present invention, the diagnostic accuracy of the model constructed with a single biomarker and a biomarker ratio is 82% to 99%, among which the N-acetylglycine / thymidine ribonucleoside ratio performs better than other models in diagnosing depression, with an accuracy of 99%. Therefore, the N-acetylglycine / thymidine ribonucleoside ratio can be used as a new biomarker for rapid diagnosis of depression, and the model constructed based on the biomarker has excellent performance in diagnosing depression. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The PLS-DA showed the differences in metabolic profiles between the healthy control group and the depression patient group.

[0017] Figure 2 The metabolites that changed significantly between the healthy control group and the depression patient group were analyzed based on the volcano plot.

[0018] Figure 3 are the chemical classes of metabolites that are differentially expressed between the healthy control group and the depression patient group.

[0019] Figure 4 It is based on KEGG pathway enrichment analysis of metabolic pathways closely related to depression.

[0020] Figure 5 The biomarkers for diagnosing depression were determined based on the LASSO regression model.

[0021] Figure 6 It is the diagnostic performance of the random forest model constructed based on biomarkers in the test set and validation set.

[0022] Figure 7 is the importance ranking of biomarkers in the random forest model.

[0023] Figure 8 , Fig. 9 , Fig.10 is the trend of concentration levels of biomarkers in patients with depression and healthy controls.

[0024] Fig.11 , Fig.12 is the diagnostic performance of the random forest model constructed based on the targeted biomarkers and their concentration ratios. DETAILED DESCRIPTION

[0025] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0026] Example The method for screening endogenous biomarkers in the serum of patients with depression comprises the following steps: (1) Cohort design and sample collection: The research subjects included patients with depression and healthy controls. The validation sample set data came from 15 patients with mild depression from April 2023 to March 2024. To reduce bias, healthy controls with inflammatory symptoms or dyslipidemia and patients with other mental illnesses such as anxiety were excluded, and subjects with missing serum samples were also excluded. The final cohort consisted of 161 healthy controls (HC), 199 patients with moderate depression (MOD), and 109 patients with major depression (MDD). Serum samples were stored in 1.5 ml centrifuge tubes at -80°C.

[0027] (2) Serum sample cleanup: Serum samples (150 μL) thawed at -80 °C were mixed with 450 μL of ice-cold methanol, vortexed and incubated at -80 °C for 2 hours. The mixture was centrifuged at 14,000 × g for 40 minutes at 4 °C to remove proteins, and then 300 μL of the supernatant was mixed with 50 μL of isotope-labeled internal standard and transferred to an amber glass bottle for analysis.

[0028] (3) Non-targeted metabolomics analysis: Ultra-performance liquid chromatography (UPLC; Vanquish UPLC system, Thermo Fisher Scientific, Germany) and quadrupole / colorimetric high-resolution mass spectrometry (Exploris 240, Thermo Fisher Scientific, Germany) were used for analysis. The chromatographic column was Hypersil GOLD AQ C18 (150 × 2.1 mm, 1.9 µm, Thermo Fisher Scientific, Germany). The mobile phases included 0.1% formic acid (A) and 100% methanol (B) for positive ions and 0.05% acetic acid in water (A) and 100% methanol (B) for negative ions. The column temperature was maintained at 35°C and the autosampler temperature was maintained at 10°C. The mobile phase gradient was as follows: 0-1 min: 2% B; 1-12 min: 2-98% B; 12-17 min: 98% B; 17.01-20 min: 2% B. The mass spectrometry detection was performed by electrospray ionization technology, with the spray voltage of 3.2 kV and -3.0 kV in positive and negative ion modes, respectively. The parameters of H-ESI were as follows: spray current: static; ion transfer tube temperature: 320 °C; evaporator temperature: 350 °C, gas mode: static, sheath gas: 45, auxiliary gas: 8 and scan gas: 1. Data acquisition included full scan mass spectrometry, target mass exclusion, vertex detection and ddMS2. In full scan mass spectrometry mode, the orbital resolution was 120,000, and the scan range was 70~1000 m / z. In dd-MS2 mode, the resolution was 15,000, and the first three high intensity ions were fragmented using step normalized collision energies of 30%, 50% and 80%, respectively. Finally, several fragment ion spectra were collected.

[0029] (4) Non-targeted metabolomics screening analysis: The fragment ion spectra were further screened and analyzed. The peak extraction (m / z and retention time), integration, missing value filling, feature filtering and metabolite annotation were all performed using CompoundDiscovery 3.3. The screened compounds were identified and annotated by searching in databases such as mzCloud, Mass Lists, ChemSpider, and mzVault. To evaluate the data quality, 10 μL aliquots of supernatant were extracted from each sample to prepare quality control (QC) samples and blank samples during the preparation process to eliminate interference caused by material and reagent contamination. The stability of the UPLC-MS system was evaluated by analyzing quality control samples at the beginning and end of each batch and after every 10 serum samples. The maximum relative standard deviation before quality control correction was 30%, and the maximum relative standard deviation after correction was 25%. At least 50% of the available hits were required for each compound. In addition, since data collected at different times may have batch differences, the random forest systematic error elimination method was used to solve the batch variability and longitudinal drift problems and eliminate unnecessary systematic variations in large sample sets.

[0030] (5) Targeted metabolomics analysis: A high-throughput detection method that can simultaneously detect 283 metabolites, including 103 amino acids, 83 carbohydrates, 72 nucleotides, and 25 energy metabolism metabolites, was constructed using ultra-high performance liquid chromatography coupled to a quadrupole / orbitrap high-resolution mass spectrometer. The chromatographic conditions and mass spectrometry methods were the same as those for non-targeted screening. The chromatographic peaks of each compound were extracted based on the precise molecular mass of the parent ion, and quantitative analysis was performed using the isotope internal standard method.

[0031] (6) Screening of differential metabolites between depression patients and healthy controls: Based on the non-targeted metabolomics screening workflow, 1,310 metabolic features were identified in the positive ion mode and 1,243 metabolic features were identified in the negative ion mode. In addition, the quantification of 84 metabolites was completed using a targeted quantitative method. Partial least squares discriminant analysis (PLS-DA) showed that the quality control samples formed a tight cluster, indicating that the metabolomics data had good stability. In addition, there were significant differences between MOD and MDD compared with HC, while there were no significant differences between MOD and MDD, which may be due to the similarity of the disease progression process. The permutation test confirmed that the PLS-DA model had strong explanatory power and high predictive ability for both independent and dependent variables ( Figure 1 ).

[0032] Based on fold change >1.2 or <0.83 and pAll metabolic features were combined for analysis with a criterion of p<0.05, resulting in 1,773 differential features between HC and MOD, and 1,658 differential features between HC and MDD ( Figure 2 ). The quality scores given to the differential metabolic features discovered by non-targeted metabolomics were then screened according to the mzCloud database, and the metabolic features with scores above 85 and obtained by targeted metabolomics were further annotated by matching the HMDB database. Finally, 76 differential metabolites were found between HC and MOD, 80 differential metabolites were found between HC and MDD, and 70 common differential metabolites were found between MOD / MDD patients and HC patients, indicating that these metabolites may be associated with the development of depression. These metabolites belong to 6 chemical categories, with the top two being organic acids and their derivatives (68.6%) and organic heterocyclic compounds (11.4%) ( Figure 3 ). Metabolic pathway enrichment analysis performed by the online tool MetaboAnalyst 6.0 found 27 dysregulated metabolic pathways that may be associated with depression. Key pathways include histidine metabolism, valine, leucine, and isoleucine biosynthesis, β-alanine metabolism, alanine, aspartate, and glutamate metabolism, arginine biosynthesis, pantothenate and CoA biosynthesis, and phenylalanine metabolism. Among the seven pathways with significant enrichment, histidine metabolism was the most prominent, containing four related differential metabolites ( Figure 4 ).

[0033] (7) Construction of a rapid diagnostic model for depression: First, based on 70 differential metabolites obtained by screening serum samples of depression patients and healthy controls through non-targeted combined with targeted metabolomics technology, a LASSO regression model was established to determine endogenous biomarkers for diagnosing depression. The λ parameter was adjusted to reduce overfitting and minimize the deviation of the binomial distribution, and ten-fold cross-validation was used to test the robustness of the model. As a result, a variable combination of 11 endogenous biomarkers (Table 1) was identified as the best discriminant model that minimized the model deviation ( Figure 5). By analyzing the chemical structures of these 11 endogenous biomarkers, it was found that N-methyl-L-alanine and γ-aminobutyric acid are isomers with similar retention times and insufficient chromatographic separation, so they were merged into one variable. Subsequently, a training sample set and a validation sample set were established based on the 11 endogenous biomarkers as a whole. The total data of the validation sample set came from 15 patients with mild depression and 30% of random samples in the original cohort. The random forest machine learning model was trained with the training sample set to obtain a primary diagnostic model. Among them, the overall concentration of endogenous biomarkers during the training process was used as the input feature of the random forest model, and the depression status of the serum subjects collected by the endogenous biomarkers was used as a label; the depression status includes depression and non-depression; after adjusting and optimizing parameters such as ntree and mtry, the ROC analysis results of the primary diagnostic model constructed based on the overall biomarkers showed that the diagnostic accuracy was 100% ( Figure 6 ). Importance ranking based on mean drop accuracy and mean drop Gini score in the random forest model showed that L-aspartyl-L-phenylalanine, uric acid, and L-phenylalanine had high importance ( Figure 7 In addition, the concentration levels of most biomarkers in patients with mild depression were significantly different from those in healthy controls ( Figure 8 , Fig. 9 , Fig.10 ).

[0034] Table 1: 11 endogenous biomarkers

[0035] In addition, among the 11 endogenous biomarkers, 7 of them are metabolites that are accurately quantified by targeted metabolomics, namely N-methyl-L-alanine, γ-aminobutyric acid, N-acetylneuraminic acid, N-acetylglycine, L-methionine, L-phenylalanine, and thymidine ribonucleoside. In order to establish a more convenient and universal diagnostic standard, the above-mentioned method of training a random forest machine learning model was used to take the concentrations and concentration ratios of these 7 endogenous biomarkers as input features of the machine learning model, and the depression status of the collected serum subjects was used as a label to establish a random forest model to evaluate the potential of endogenous biomarkers for diagnosing depression. ROC analysis was performed on the various primary diagnostic models obtained, and the accuracy rate was 82%~99% ( Fig.11 , Fig.12). Among them, the accuracy of the preliminary diagnosis model trained by N-acetylglycine / thymidine ribonucleoside is the highest, which is 99%. The present invention finds that the ratio of N-acetylglycine / thymidine ribonucleoside = 2.434 can effectively preliminarily diagnose healthy samples and disease samples. In the external validation cohort, only one healthy sample was misclassified as a disease sample, and one disease sample was misclassified as a healthy sample.

[0036] Therefore, the N-acetylglycine / thymidine ribonucleoside ratio can be used as a new biomarker for rapid diagnosis of depression, and the random forest model constructed based on the biomarker has excellent performance in diagnosing depression.

[0037] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for screening endogenous biomarkers in the serum of patients with depression, characterized in that The following steps are involved: (1) Serum collection: Serum was collected from healthy controls and patients with depression and mixed with isotope-labeled internal standards for further analysis; (2) Non-targeted metabolomics analysis: Step (1) The serum sample is analyzed by ultra-high performance liquid chromatography and quadrupole / colorimetric high-resolution mass spectrometry to collect a number of fragment ion spectra; the fragment ion spectra are further screened and analyzed to screen out a number of compounds; (3) Targeted metabolomics analysis: Step (1) Serum samples were subjected to ultra-high performance liquid chromatography tandem with quadrupole / orbitrap high-resolution mass spectrometry to construct a high-throughput detection method that can simultaneously detect multiple metabolites. The chromatographic peaks of each compound were extracted based on the precise molecular mass of the parent ion, and quantitative analysis was performed using the isotope internal standard method. (4) Screening for differential metabolites between patients with depression and healthy controls: The metabolites identified by non-targeted metabolomics analysis and targeted metabolomics analysis were combined for analysis, and the common differential metabolites between patients with depression and healthy controls were used as endogenous markers of depression; The combined analysis is a supervised multivariate statistical analysis method, and the overfitting of the model is evaluated by permutation test, and the differential metabolites are screened based on rank sum test and difference fold analysis.

2. The method according to claim 1, characterized in that: The supervised multivariate statistical analysis method is partial least squares discriminant analysis.

3. The method according to claim 1, characterized in that: The screening analysis described in step (2) includes one or more steps of peak extraction, integration, missing value filling, feature filtering and metabolite annotation.

4. The method according to claim 1, characterized in that: The depression patients described in step (1) include moderate depression patients and severe depression patients.

5. The method according to claim 1, characterized in that: The serum described in step (1) is further purified.

6. The method according to claim 5, characterized in that: For the purification, the serum was mixed with pre-cooled methanol, incubated for several hours, and centrifuged at below 4°C to remove proteins.

Images