An exosomal characteristic protein composition for diagnosing HIV-related asymptomatic neurocognitive disorders and its application

By screening combinations of exosomal characteristic proteins such as DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A, and combining them with machine learning algorithms to construct a diagnostic model, the problems of insufficient subjectivity and specificity in the early diagnosis of ANI were solved, and a highly accurate non-invasive diagnosis was achieved.

CN122631899BActive Publication Date: 2026-11-10BEIJING YOUAN HOSPITAL CAPITAL MEDICAL UNIV
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202610680836.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-05-18
Publication Date
2026-11-10
Estimated Expiration
2046-05-18

AI Technical Summary

Technical Problem

Existing technologies make it difficult to achieve early and accurate diagnosis of HIV-related asymptomatic neurocognitive impairment (ANI). Traditional diagnostic methods are highly subjective, lack specificity, and are highly invasive. Single protein biomarkers are insufficient to comprehensively capture the molecular characteristics of the disease.

Method used

By screening six exosome characteristic protein combinations, including DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A, and combining them with DIA mass spectrometry technology and machine learning algorithms, a diagnostic model is constructed to provide a non-invasive and accurate diagnostic tool.

Benefits of technology

It achieves highly accurate diagnosis of ANI, with strong specificity, can identify 100% of ANI patients, and is non-invasive and convenient, suitable for large-scale screening and long-term follow-up.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122631899B_ABST
    Figure CN122631899B_ABST
Patent Text Reader

Abstract

This invention discloses an exosome characteristic protein composition for diagnosing HIV-related asymptomatic neurocognitive impairment (ANI), comprising the following characteristic proteins: DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A. Compared with non-ANI individuals, the relative abundance of DSCAM and IGF1 is increased in ANI individuals, while the relative abundance of CAMK2D, UPF1, ALDH2, and EIF5A is decreased. This exosome characteristic protein composition has the following advantages: (1) high diagnostic accuracy, comprehensively covering the core pathways of ANI pathological mechanisms, and accurately identifying ANI patients; (2) high specificity, clearly reflecting the specific expression patterns of exosome proteins in ANI patients, thereby effectively distinguishing ANI from NCI and healthy individuals; (3) non-invasive and convenient, based on peripheral plasma testing, improving subject compliance, and suitable for large-scale screening and long-term follow-up; (4) easy clinical translation: the kits prepared for the above exosome characteristic proteins are easy to operate, can be mass-produced and used, and can be implemented in hospitals and disease control centers at all levels, making them suitable for widespread application.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of medical diagnostic technology, specifically to an exosome characteristic protein composition for diagnosing HIV-related asymptomatic neurocognitive disorders and its application. Background Technology

[0002] Despite the widespread use of combination antiretroviral therapy (CART) which has significantly reduced HIV-related mortality, HIV-associated neurocognitive disorders (HAND) remain a core challenge in the clinical management of acquired immunodeficiency syndrome (AIDS). Numerous studies have confirmed that even in patients with effectively suppressed viral load, the prevalence of HAND remains as high as 30%-50%. According to the Frascati diagnostic criteria, HAND can be divided into three subtypes: asymptomatic neurocognitive impairment (ANI), mild neurocognitive disorder (MND), and HIV-associated dementia (HAD). ANI has the highest prevalence, reaching 23.5%. Although patients with acute myocardial infarction (ANI) perform normal daily activities, they are 2 to 6 times more likely to progress to symptomatic neurocognitive impairment than cognitively normal HIV-infected individuals. Therefore, achieving early and accurate diagnosis of ANI is crucial for improving patient prognosis.

[0003] Currently, the diagnosis of Handicap Disorder (HAND) primarily relies on neuropsychological assessments, but this method has significant limitations: it is highly subjective, and results are easily influenced by factors such as education level, cultural background, and language differences. Ardila et al. demonstrated through five dimensions that culture affects neuropsychological test results in terms of ability patterns, cultural values, familiarity with test content, language factors, and education level, leading to the omission or misdiagnosis of early cognitive impairment in ANI patients. Meanwhile, although cerebrospinal fluid testing can provide markers of neuronal damage (such as NFL and Aβ) and inflammatory markers (such as HMGB1), it is an invasive procedure, making clinical implementation difficult and resulting in low patient compliance. Therefore, developing non-invasive, objective, and highly specific peripheral blood biomarkers has become a core direction for HAND diagnostic research.

[0004] Exosomes, as extracellular vesicles with lipid bilayers, provide an ideal carrier for the non-invasive diagnosis of HAND (Hyperencephaly-Related Atrophic Disorder). They act as intercellular communication mediators, carrying proteins, RNA, lipids, and other substances reflecting cellular state, and can cross the blood-brain barrier, playing a crucial role in the pathophysiology of HIV. Studies by András et al. have found that HIV infection affects exosome release and amyloid-β (Aβ) levels in brain endothelial cells, suggesting that exosomes can be used to elucidate pathological processes such as blood-brain barrier dysfunction and neuroinflammatory responses. Neuron-derived exosomes (NDEs) from plasma, as a "liquid biopsy" tool, can non-invasively reflect the state of the central nervous system. Studies by Sun et al. have confirmed that the levels of proteins such as HMGB1, NF-L, and amyloid-β in NDEs from HIV-infected individuals are associated with cognitive impairment. Pulliam et al. further pointed out that NDE protein biomarkers can provide "real-time" information on neuronal health status, complementing traditional cerebrospinal fluid and neuroimaging indicators.

[0005] More importantly, exosomes are not only a source of biomarkers but also directly involved in HIV-related neuropathological processes: exosomes produced by HIV-infected macrophages and astrocytes carry viral proteins such as Tat and Nef, which can induce neuronal apoptosis and synaptic damage. Simultaneously, exosomes regulate neuronal-glial cell interactions, blood-brain barrier integrity, and microglia activation; changes in their molecular composition can precisely reflect the pathological state of the central nervous system. However, current exosome research still has significant shortcomings: most studies focus on single protein biomarkers, lacking specific protein combinations for screening in the early stages of acute myeloid neuropathy (ANI); the association between exosomal proteins and ANI pathological mechanisms (such as neuroinflammation, synaptic damage, and cortical atrophy) has not been systematically explored; and the diagnostic specificity and sensitivity of existing biomarkers are limited, making it difficult to meet the needs of precise clinical diagnosis.

[0006] Furthermore, the pathological mechanisms of ANI caused by HIV infection are complex, involving multiple pathological processes such as persistent chronic neuroinflammation, synaptic dysfunction, and mitochondrial dysfunction. Single protein biomarkers are insufficient to comprehensively capture the molecular characteristics of the disease. Studies have shown that HIV infection can induce astrocyte activation through viral proteins, releasing glutamate and causing excitatory neurotoxicity. It can also promote Aβ deposition by interfering with proteasome and autophagy pathways. These pathological changes all leave traces in the protein composition of neuronal-derived exosomes. Therefore, screening for exosomal characteristic protein combinations directly related to the pathological mechanisms of ANI and constructing diagnostic models using machine learning algorithms is essential to overcome the limitations of single biomarkers and achieve early and accurate identification of ANI.

[0007] Although existing studies have suggested that exosomal proteins could serve as potential biomarkers for hand-and-mouth disease (HAND), no research has yet screened for highly specific exosomal protein combinations targeting acute myopathic lesions (ANI), and there is a lack of in-depth validation of the correlation between these proteins and the pathological progression of ANI. Therefore, this invention utilizes data-independent acquisition (DIA) mass spectrometry to screen for characteristic protein combinations closely related to ANI from plasma exosomes. Combined with machine learning algorithms, a diagnostic model is constructed to address the problems of high subjectivity, insufficient specificity, and high invasiveness in existing diagnostic methods. This provides a non-invasive and accurate early diagnostic tool for ANI, filling a gap in current HAND diagnostic research. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention provides an exosome-specific protein composition for diagnosing HIV-related asymptomatic neurocognitive disorders and its application.

[0009] In one aspect, the present invention provides an exosome characteristic protein composition for diagnosing HIV-related asymptomatic neurocognitive impairment (ANI), the exosome characteristic protein composition being composed of six characteristic proteins: DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A.

[0010] Furthermore, compared to individuals other than those with HIV-related asymptomatic neurocognitive impairment: The relative abundance of DSCAM and IGF1 increased, while the relative abundance of CAMK2D, UPF1, ALDH2 and EIF5A decreased.

[0011] Furthermore, taking the total abundance of the exosome characteristic protein composition as 100%, wherein The relative abundance of DSCAM was 18% or higher, IGF1 was 15% or higher, CAMK2D was 16% or lower, UPF1 was 17% or lower, ALDH2 was 14% or lower, and EIF5A was 19% or lower.

[0012] Furthermore, taking the total abundance of the exosome characteristic protein composition as 100%, wherein The relative abundance of DSCAM was 18–24%, IGF1 was 15–21%, CAMK2D was 10–16%, UPF1 was 11–17%, ALDH2 was 8–14%, and EIF5A was 13–19%.

[0013] Furthermore, in the exosome characteristic protein composition, the relative abundance ratio of the six characteristic proteins is DSCAM:IGF1:CAMK2D:UPF1:ALDH2:EIF5A = 21:18:13:14:11:16.

[0014] In another aspect, the present invention provides the use of the above-described exosome characteristic protein composition in the preparation of reagents or kits for diagnosing HIV-related asymptomatic neurocognitive disorders.

[0015] Furthermore, the reagent or kit is used for qualitative and quantitative detection of the characteristic protein.

[0016] A third aspect of the present invention provides a kit for diagnosing HIV-related asymptomatic neurocognitive impairment, comprising specific detection reagents for detecting the above-described exosome characteristic protein composition.

[0017] Furthermore, the specific detection reagent includes specific capture antibodies against each of the six characteristic proteins; Preferably, the kit further includes one or more of the following: enzyme-labeled detection antibody, gradient concentration standards, blocking solution, washing solution, colorimetric solution, and stop solution.

[0018] The fourth aspect of the present invention provides the application of the above-mentioned exosome characteristic protein composition in constructing a computer model for diagnosing HIV-related asymptomatic neurocognitive impairment; the computer model is preferably an XGBoost diagnostic model.

[0019] Furthermore, the computer model performs the following functions: The detection results of the six characteristic proteins of the subject were obtained, and the relative abundance of each characteristic protein in the total amount of the six characteristic proteins was calculated; The relative abundance of each characteristic protein of the subject is compared with the relative abundance in the exosome characteristic protein composition. When the values ​​match, the subject is determined to be an HIV-related asymptomatic neurocognitive disorder patient, and the determination result is output.

[0020] Furthermore, the method for constructing the computer model is as follows: Input the relative abundance data of the six characteristic proteins in the exosome characteristic protein composition, and after Z-score normalization, stratify and sample in an 8:2 ratio to obtain the training set and validation set; Five-fold cross-validation was used to optimize the initial parameters using a grid search. The initial parameters included nrounds=100, max_depth=6, learning_rate=0.1, subsample=0.8, and colsample_bytree=0.8. The optimal parameters were finally determined to be nrounds=100, max_depth=6, learning_rate=0.07, subsample=0.8, and colsample_bytree=0.8. A model is constructed based on the optimal parameters, and performance verification is completed.

[0021] The beneficial effects of the present invention are as follows: The exosome characteristic protein composition for diagnosing HIV-related asymptomatic neurocognitive impairment (ANI) provided by the present invention has the following advantages: (1) High diagnostic accuracy: The six proteins selected in this application are combined in a specific ratio to fully cover the core pathways of ANI pathological mechanism; the computer model constructed based on the exosome characteristic protein composition has a validation set AUC of 0.944 and a sensitivity of 1.000, and can identify ANI patients 100%; (2) High specificity: The specific expression ratio of the six characteristic proteins can clearly reflect the specific expression pattern of exosomal proteins in ANI patients (DSCAM and IGF1 are upregulated, while CAMK2D, UPF1, ALDH2 and EIF5A are downregulated), thus effectively distinguishing ANI from NCI and healthy individuals. (3) Non-invasive and convenient: The exosome characteristic protein composition provided in this application can be detected based on peripheral plasma, avoiding invasive operations, thereby improving subject compliance and making it suitable for large-scale screening and long-term follow-up; (4) Easy to translate into clinical practice: The kits prepared for the above-mentioned exosome characteristic proteins are easy to operate, can be mass-produced and used, and can be implemented in hospitals and disease control centers at all levels, making them suitable for widespread application. Attached Figure Description

[0022] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative examples are not intended to limit the embodiments. The term "illustrative" as used herein means "serving as an example, embodiment, or illustration." Any embodiment illustrated herein as "illustrative" is not necessarily to be construed as superior to or better than other embodiments.

[0023] Figure 1 Electrophoresis images of exosomes from each group of subjects were obtained using Western blotting.

[0024] Figure 2The following are the transmission electron microscopy observation results of exosomes in each group of subjects: A is the morphology of exosomes in the ANI group; B is the morphology of exosomes in the NCI group; and C is the morphology of exosomes in the HCs group.

[0025] Figure 3 The particle size distribution maps of exosomes from each group of subjects were detected by a nanoparticle tracking analyzer. Among them, A is the exosome particle size distribution of the ANI group; B is the exosome particle size distribution of the NCI group; and C is the exosome particle size distribution of the HCs group.

[0026] Figure 4 This is a quality control chart for exosomal proteomics data from each group of subjects.

[0027] Figure 5 The differential protein screening diagrams for the ANI group and the HCs group are shown below. In A, the differential protein volcano plots of the two groups are shown, with red representing differentially expressed proteins that are upregulated, blue representing differentially expressed proteins that are downregulated, and gray representing proteins with no significant difference. In B, the heatmaps of the top 50 differentially expressed proteins of the two groups are shown, with blue representing low expression and red representing high expression.

[0028] Figure 6 Cluster diagrams showing the expression trends of exosomal proteins in each group of subjects are provided. The horizontal axis represents different groups, and the vertical axis represents the change in expression levels after normalization. The lines of each cluster represent a class of proteins with the same expression trend.

[0029] Figure 7 The intersection protein analysis plots for Differentially Expressed Proteins (DEPs) and Fuzzy C-means Clustering (Mfuzz) are shown, where: A is the statistical count of the number of intersections between the differentially expressed proteins and proteins in Cluster 6 and Cluster 7; B is the expression level heatmap of the intersection proteins.

[0030] Figure 8The following diagrams represent the screening of disease-specific proteins in the ANI group: A is the error curve of the Random Forest (RF) algorithm decision tree; B is the top 10 core proteins selected by the Random Forest algorithm; C is the top 10 core proteins selected by the eXtreme Gradient Boosting (XGBoost) algorithm; D is the regularized path diagram of the Least Absolute Shrinkage and Selection Operator (LASSO) regression algorithm; E is the result of screening disease-specific proteins by the Support Vector Machine (SVM) algorithm; and F is the intersection result of the core protein screening by the four algorithms: Random Forest, eXtreme Gradient Boosting, LASSO regression, and SVM.

[0031] Figure 9 Figure A shows the functional enrichment analysis of the intersecting proteins. Figure B is a Mānji bubble diagram of pathway enrichment from the Kyoto Encyclopedia of Genes and Genomes (KEGG). Figure B is a Mānji bubble diagram of enrichment from the Gene Ontology-Biological Process (GO_BP).

[0032] Figure 10 This figure shows the expression differences of six core characteristic proteins in three groups of subjects.

[0033] Figure 11 The graph shows the receiver operating characteristic (ROC) curves for the six core characteristic proteins.

[0034] Figure 12 The ROC curves for the diagnostic model based on six core characteristic proteins are shown, where the dashed line represents the training set curve and the solid line represents the validation set curve. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] Furthermore, to better illustrate the present invention, numerous specific details are set forth in the following detailed embodiments. Those skilled in the art will understand that the present invention can still be practiced even without certain specific details. In some embodiments, materials, elements, methods, and means well known to those skilled in the art are not described in detail in order to highlight the spirit of the invention.

[0037] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0038] Unless otherwise specified, all test materials used in the following examples are available through conventional commercial channels.

[0039] Example This invention is based on a prospective study that analyzed plasma samples from 72 subjects (18 ANI patients, 31 HIV-infected individuals without neurocognitive impairment (NCI)) and 23 healthy controls (HCs). Differential biomarkers were screened through systematic statistical analysis, and the quantitative proportions of each biomarker were derived based on inter-group abundance differences, as detailed below: 1. Study subject selection and baseline data Study Design: This was a prospective study approved by the Ethics Committee of Beijing You'an Hospital, Capital Medical University (Ethics No.: LL-2023-070-K). All participants signed written informed consent forms, which complied with the requirements of the Declaration of Helsinki.

[0040] 1.1 Inclusion criteria (1) HIV-infected individuals: Chinese men aged 20-50 who are male homosexual and HIV-infected, right-handed, whose HIV infection is confirmed by immunoblotting or PCR, and whose viral load is <40 copies / ml; (2) HCs: 20-50 years old, right-handed, no complaints of cognitive decline, normal standardized cognitive function test results, no history of neurological diseases or positive imaging findings.

[0041] 1.2 Exclusion Criteria (1) The presence of neurological diseases such as brain tumors, infections, strokes, and epilepsy that can cause cognitive impairment; (2) Mental illnesses such as depression and anxiety (HAMD score ≥ 7); (3) History of drug abuse, alcoholism, or drug use; (4) Having claustrophobia or contraindications for MRI examination; (5) Severe visual, hearing and reading difficulties, unable to cooperate with the examination.

[0042] 1.3 Baseline Data The study ultimately included 31 patients with NCI, 18 patients with ANI, and 23 patients with HC. There were no significant differences in demographic data (age, years of education) among the three groups (p>0.05). Compared with NCI, the ANI group had significantly lower t-scores in all six cognitive domains (p<0.001). The diagnosis of ANI strictly followed the Frascati criteria: (1) performance in ≥2 cognitive domains below the mean of the demographically adjusted normative score by at least one standard deviation; (2) no decline in daily living function; (3) impairment that did not meet the criteria for delirium or dementia; and (4) no evidence that ANI was caused by other factors.

[0043] 2. Experimental Preparation 2.1 Main Reagents EVtrap exosome extraction kit (including magnetic beads, loading buffer, washing buffer, 200mM TEA elution buffer); CD9, CD81, TSG101, Calnexin antibodies; RIPA lysis buffer, protease inhibitor, BCA protein quantification kit; DIA mass spectrometry detection reagents; 8M urea denaturing buffer, dithiothreitol (DTT), iodoacetamide, mass spectrometry grade trypsin; DSCAM, IGF1, CAMK2D, UPF1, ALDH2, EIF5A specific ELISA detection reagents (including pre-coated plate, standards, enzyme-labeled secondary antibody, chromogenic solution, and stop solution).

[0044] 2.2 Main Instruments Exosome-related instruments: transmission electron microscope, nanoparticle tracking analyzer, centrifuge, magnetic separator; Protein detection instruments: mass spectrometer, microplate reader, electrophoresis apparatus, chemiluminescence imaging system; Data processing equipment: High-performance computer, with MaxQuant software and R software installed (including XGBoost, pROC, ggplot2, and Mfuzz packages).

[0045] 3. Experimental Methods 3.1 Sample Collection and Preprocessing 3.1.1 Plasma Sample Collection 10 mL of peripheral venous blood was collected from all subjects using EDTA anticoagulant tubes and processed within 2 hours after collection. The plasma was centrifuged at 1900 g for 10 minutes at 4°C, and the upper plasma layer was carefully aspirated (avoiding aspiration of the middle layer of leukocytes) and transferred to a 600 μL nuclease-free EP tube. The remaining plasma was transferred to a 5 mL EP tube, centrifuged at 3000 g for 15 minutes at 4°C for purification, and 3.5 mL was used for exosome extraction. The remaining samples were aliquoted and stored at -80°C.

[0046] 3.1.2 Neuropsychological testing Six cognitive dimensions were assessed using a combination of Chinese-language neuropsychological tests. Raw scores were converted to T-scores and corrected. ANI was diagnosed according to the Frascati criteria, and the clinical grouping of the three groups of subjects was clarified.

[0047] 3.2 Exosome Extraction and Characterization 3.2.1 Exosome Extraction Exosome extraction: Take 3.5 mL of purified plasma, add 1 mL of loading buffer and 25 μL of EVtrap magnetic beads, shake well and incubate for 1 hour at room temperature; perform magnetic separation using a magnetic separator, collect the supernatant for later use; add washing buffer to suspend the magnetic beads, transfer the reaction system to a deep well plate, perform fully automated extraction according to the instrument's default program, collect 400 μL of eluent, and freeze at -80℃.

[0048] 3.2.2 Triple Characterization Validation Western Blot validation: Exosomal proteins were extracted and quantified using the BCA method; after 15% SDS-PAGE electrophoresis, the membrane was transferred and blocked with 1% BSA for 1 hour. Then, primary antibodies against CD9, CD81, TSG101, and Calnexin (calcified protein, negative control) were added, and the membrane was incubated overnight at 4°C. After incubation with secondary antibody, chemiluminescence detection was performed. Results are as follows: Figure 1 As shown, the specific markers CD9, CD81, and TSG101 were highly expressed in all groups, while the negative control calcinin was not detected, indicating that the obtained exosomes were of high purity and met the experimental requirements.

[0049] Transmission electron microscopy (TEM) observation: 10 μL of each exosome sample was added to a copper grid, negatively stained with uranyl acetate, dried at room temperature, and then imaged at 100 kV. Results are as follows: Figure 2 As shown, all three groups of exosomes exhibited typical cup-shaped or vesicle-shaped structures, with intact morphology and no damage, and their particle size was concentrated in the range of 50-150 nm.

[0050] Nanoparticle tracking analysis: Sample dilution was analyzed using a nanoparticle tracking analyzer, and the results are as follows. Figure 3 As shown, the main peak values ​​of the three groups of exosomes are concentrated in the range of 80-120 nm, which is consistent with the typical particle size characteristics of exosomes.

[0051] The above tests confirm that the extraction method used in this application is stable and reliable, and the purity and integrity of the samples can meet the needs of subsequent experiments, thus providing a basic support for the scientific nature of the feature protein screening.

[0052] 3.3 Proteomics Detection and Data Quality Control 3.3.1 Protein Extraction Add 6 volumes of RIPA lysis buffer (containing protease inhibitor) to 200 μL of exosome elution buffer, lyse on ice for 30 minutes, centrifuge at 12000g for 20 minutes, collect the supernatant, and determine the protein concentration (≥1 μg / μL) using the BCA method.

[0053] 3.3.2 DIA Mass Spectrometry Detection Protein digestion: Denatured with 8M urea, reduced disulfide bonds with DTT, and alkylated with iodoacetamide; add mass spectrometry grade trypsin (enzyme / protein = 1 / 50) and incubate at 37°C for 16 hours; Mass spectrometry analysis: Separation was performed using nanoliter liquid chromatography, followed by gradient elution for 120 minutes. Mass spectrometry was then performed with the following parameters: mass range 350-1800 m / z, resolution 70,000 (200 m / z), and DIA acquisition in 21 windows (25 Da each) to obtain the relative peak area data of 6 proteins. Ratio calculation: Calculate the average relative peak area of ​​the 6 proteins in the ANI group, and deduce the relative abundance ratio and range accordingly (as described in the technical solution). Data quality control: Calculate the RSD of protein expression in the three groups of samples, and the results are as follows. Figure 4 As shown, the median RSD of the HCs group was approximately 14.9%, the ANI group was approximately 14.7%, and the NCI group was approximately 15.0%, all of which were within the biologically acceptable range. This indicates that the experimental data had good reproducibility and reliability, which can ensure the accuracy of subsequent differential protein screening and characteristic protein identification.

[0054] 3.4 Feature Protein Screening 3.4.1 Differential Protein Screening Between-group difference analysis: Differentially expressed proteins were screened using t-tests, and the results are as follows: Figure 5 As shown, with |FC|>1.2 and p<0.05 as thresholds, the ANI group and HCs group screened out 491 differentially expressed proteins (including 416 upregulated differentially expressed proteins and 75 downregulated differentially expressed proteins). The top 10 upregulated and downregulated differentially expressed proteins were marked in the volcano plot, and the top 50 differentially expressed proteins were shown in the heatmap. The NCI and HCs groups screened out 701 differentially expressed proteins.

[0055] Mfuzz clustering: Two ANI-specific modules, Cluster6 and Cluster7, were selected using fuzzy C-means clustering. Figure 6 The differentially expressed proteins were then intersected with the proteins in the two feature modules to obtain 35 candidate proteins that are highly correlated with the occurrence and development of ANI. These 35 candidate proteins were used as the input dataset for the subsequent screening by four machine learning algorithms. Figure 7 ).

[0056] Functional enrichment: GO / KEGG analysis showed that the candidate proteins can participate in key pathways such as synapse formation and axonal guidance, thereby affecting neurocognitive function.

[0057] 3.4.2 Machine Learning Validation Four algorithms were used for cross-selection: core proteins were screened from candidate proteins using four algorithms: RF, XGBoost, LASSO, and SVM. The intersection of these algorithms was combined with the effects of the CAMK2D pathway to finally identify six core characteristic proteins: DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A.

[0058] To avoid bias from a single algorithm, four classic feature selection algorithms—Random Forest (RF), Extreme Gradient Boosting (XGBoost), Minimum Absolute Shrinkage and Regression by Selection Operator (LASSO), and Support Vector Machine (SVM)—were used. Based on the expression matrices of the aforementioned 35 candidate proteins, independent selection was conducted using the ANI phenotype as the classification label. The implementation details and selection results of each algorithm are as follows: (1) Random Forest (RF) Algorithm Selection: Random forest is based on the Bagging ensemble learning idea. It forms a strong classification model by constructing multiple decision trees. The mean decrease in Gini coefficient is used to evaluate the contribution of each feature to the classification. The higher the value, the stronger the protein's ability to distinguish between ANI and non-ANI samples. Using the expression level of 35 candidate proteins as features and ANI / non-ANI as the classification outcome, a random forest classification model is constructed. The core parameters such as the number of decision trees and the minimum number of samples for node splitting are optimized through 5-fold cross-validation to prevent model overfitting. Finally, the top 10 proteins contributing to ANI classification are selected based on the feature importance score output by the model. They are: EIF5A, DSCAM, SPRR2G, IMMT, ALDH2, IGF1, S100A11, UPF1, PPP4C, EPB41 ( Figure 8 (A and B in the text).

[0059] (2) XGBoost Algorithm Selection: XGBoost is an ensemble learning algorithm based on the gradient boosting framework. It iteratively trains weak classifiers and accumulates and optimizes the loss function. The feature importance is calculated by the sum of the gains of the splits in all decision trees. It has excellent fitting ability and generalization in small sample classification tasks. The XGBoost model is constructed using the same dataset and classification labels as the RF algorithm. The key parameters such as tree depth, learning rate, and number of iterations are optimized by 5-fold cross-validation. Finally, the top 10 proteins are selected according to the feature importance scores output by the model, which are: S100A11, IMMT, DSCAM, IGF1, EIF5A, CCDC18, SPRR2G, ALDH2, UPF1, EPB41 ( Figure 8 (C in the middle).

[0060] (3) LASSO regression algorithm screening: LASSO regression is a linear regression method with L1 regularization. It can compress the coefficients of features that do not contribute to classification to 0 through the penalty term, realize the automatic screening of features, and effectively avoid the model overfitting problem under high-dimensional data. It is a classic method for disease biomarker screening. With ANI phenotype as the dependent variable and the standardized expression level of 35 candidate proteins as the independent variable, a LASSO-Logistic classification model was constructed. The optimal regularization parameter λ was selected through 10-fold cross-validation. Finally, the model retained 16 proteins with non-zero coefficients, namely: EPB41, DSCAM, IMMT, ECM2, HSPA4L, IGF1, ALDH2, S100A11, SEC13, EIF5A, CCDC18, BROX, H2AC21, CEMIP, UPF1, EML4 ( Figure 8 (D in the middle).

[0061] (4) SVM Algorithm Screening: Support Vector Machine (SVM) classifies samples by constructing an optimal classification hyperplane in a high-dimensional feature space. It has good adaptability to small-sample, high-dimensional biological omics data, and the contribution of each protein to the classification can be quantified by the feature weight coefficient. A linear kernel function was used to construct the SVM classification model. The core parameters such as the penalty coefficient C were optimized by 5-fold cross-validation. Based on the feature weight coefficients output by the model, nine proteins that made significant contributions to ANI classification and whose weight coefficients were not zero were screened out, namely: DSCAM, CEMIP, ALDH2, EIF5A, IGKV113, SEC13, IGF1, UPF1, ECM2 ( Figure 8 (E in the text).

[0062] 3.4.3 Multi-algorithm cross-validation and core protein intersection screening To ensure that the selected feature proteins maintain stable ANI classification performance across different algorithm models, the intersection of the protein sets selected by the four algorithms was taken, ultimately yielding five proteins that were identified as core features in all four algorithms: DSCAM, IGF1, UPF1, ALDH2, and EIF5A. Figure 8 (F in the original text). These five proteins were identified as key features for ANI classification in four independent machine learning algorithms, demonstrating their strong disease discrimination ability and result stability.

[0063] 3.4.4 Supplementation of CAMK2D and Final Identification of Core Feature Proteins Based on the above five overlapping proteins, and combining pathway function analysis, expression difference verification, and diagnostic efficacy assessment, the CAMK2D protein was added, ultimately forming a set of six core characteristic proteins, as follows: (1) Pathway function level: In the KEGG enrichment analysis of 35 candidate proteins, CAMK2D was the only protein that was involved in the core pathological pathways of ANI, including glioma, glucagon signaling pathway, HIF-1 signaling pathway, necrosis and apoptosis. Figure 9 ).

[0064] (2) Expression differences and diagnostic efficacy: The intergroup comparison results of this study showed that CAMK2D expression was significantly downregulated in the ANI group, with a highly statistically significant difference compared with the healthy control group (p<0.01), and also showed a significant expression difference compared with the NCI group (p<0.05); ROC curve analysis showed that the AUC value of CAMK2D alone distinguishing ANI from healthy controls reached 0.720, demonstrating good diagnostic efficacy. Figure 10 ).

[0065] In summary, by combining the five core proteins obtained through cross-screening using four machine learning algorithms, and CAMK2D which was validated through pathway function, expression differences, and diagnostic efficacy, six core characteristic proteins of ANI were finally identified: DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A.

[0066] Expression differential validation: The expression levels of six core characteristic proteins were detected, and the results are as follows. Figure 10As shown in the box plot, the expression levels of IGF1 and DSCAM in the ANI group were significantly higher than those in the NCI and HCs groups (ANI vs HCs: p < 0.001; ANI vs NCI: DSCAM p < 0.001, IGF1 p < 0.05); the expression levels of CAMK2D, UPF1, ALDH2, and EIF5A in the ANI group were significantly lower than those in the HCs group (p < 0.05 ~ p < 0.001), and also lower than those in the NCI group (p < 0.05). The statistical significance (p value) between the groups is clearly marked in the figure. Therefore, compared to the NCI and HCs groups, the expression levels of IGF1 and DSCAM were significantly upregulated in the ANI group, while the other four were significantly downregulated, and the differences between the groups were statistically significant.

[0067] The above verification can directly verify the intergroup differentiation ability of the six core characteristic proteins, confirm that they have a specific expression pattern in ANI patients, and thus confirm the diagnostic value of the characteristic protein combination.

[0068] 3.5 Core Protein ROC Analysis The diagnostic efficacy of each of the six proteins was analyzed using the "pROC" package in R software, and ROC curves were plotted. Figure 11 The paper presents the ROC curves and areas under the curve (AUC) of six characteristic proteins used individually for ANI diagnosis. The AUCs are: EIF5A = 0.857, DSCAM = 0.843, IGF1 = 0.795, ALDH2 = 0.773, UPF1 = 0.766, and CAMK2D = 0.720. Quantitative analysis confirms that each protein possesses independent diagnostic value, and it is expected that combined use can further improve diagnostic accuracy.

[0069] 4. Diagnostic Model Construction and Validation 4.1 Data Processing The relative abundance data of six proteins from 72 subjects were collected and standardized using the Z-score standardization formula (Z=(X-μ) / σ, where X is the relative abundance value of a single sample, μ is the mean of the relative abundance of the protein in the 72 samples, and σ is the standard deviation of the relative abundance of the protein in the 72 samples) to eliminate dimensional differences. Dataset partitioning: Stratified random sampling was used to divide the 72 samples into a training set (58 cases, including 14 ANI cases, 25 NCI cases, and 19 healthy controls) and a validation set (14 cases, including 4 ANI cases, 6 NCI cases, and 4 healthy controls) in an 8:2 ratio. This ensured that the sample proportions of the three groups were consistent with the total sample and avoided sample bias. Outlier handling: Outliers (those exceeding 1.5 times the interquartile range) are identified using box plots and filled using the K-Nearest Neighbors (KNN) algorithm (K=5) to ensure data integrity. 4.2 Model Construction Algorithm selection: The XGBoost algorithm is adopted. This algorithm integrates multiple decision trees through the gradient boosting framework, has strong anti-overfitting ability, and is suitable for multi-feature integrated diagnosis. Parameter initialization: Set the initial values ​​of the core parameters to nrounds=100 (number of iterations), max_depth=6 (decision tree depth), learning_rate=0.1 (learning rate), subsample=0.8 (sample sampling ratio), colsample_bytree=0.8 (feature sampling ratio), and reg_lambda=1 (L2 regularization coefficient). Parameter optimization: Five-fold cross-validation is used. The training set is randomly divided into 5 subsets. Each subset is selected as the cross-validation set, and the remaining 4 subsets are used as training subsets. This process is repeated 5 times to complete full validation. The cross-validation set AUC value is used as the objective function. The parameter range for grid search is: nrounds=80~120, learning_rate=0.05~0.15, max_depth=4~8. Finally, the optimal parameter combination is selected as: nrounds=100, max_depth=6, learning_rate=0.07, subsample=0.8, colsample_bytree=0.8, reg_lambda=1. Model training: Based on the optimal parameter combination, the XGBoost model is trained using the training set data to generate the ANI diagnostic model.

[0070] 4.3 Performance Evaluation Evaluation metrics: Calculate the AUC (area under the curve), accuracy, sensitivity, specificity, precision, and F1 score for the training and validation sets to comprehensively evaluate model performance; Performance results: Figure 12 The ROC curves of the diagnostic model with six characteristic protein combinations are shown. The dashed line represents the training set curve (AUC=0.963, accuracy=0.881, sensitivity=0.632, specificity=1.000, precision=1.000, F1 score=0.774), and the solid line represents the validation set curve (AUC=0.944, accuracy=0.923, sensitivity=1.000, specificity=0.889, precision=0.800, F1 score=0.889). Results validation: The model performance was significantly better than the single protein model (the highest single protein AUC = 0.857), confirming that the combination of characteristic proteins has excellent diagnostic efficacy and outstanding core advantages in the diagnosis of ANI.

[0071] 5. Diagnostic kit validation Composition and Configuration: Includes an exosome extraction module (EVtrap magnetic beads, loading buffer, washing buffer, 200mM TEA elution buffer), six protein-specific detection reagents (specific capture antibodies for DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A, enzyme-labeled detection antibodies, gradient concentration standards (0.1ng / mL, 1ng / mL, 10ng / mL, 50ng / mL, and 100ng / mL), and blocking buffer (1%). BSA, washing buffer (TBST), chromogenic buffer (TMB), stop solution (2 mol / L H2SO4), and analysis software with a built-in XGBoost model were used. A mixed capture antibody was prepared according to the formula DSCAM:IGF1:CAMK2D:UPF1:ALDH2:EIF5A = 21:18:13:14:11:16, with the following concentrations: DSCAM capture antibody 2.1 μg / mL, IGF1 capture antibody 1.8 μg / mL, CAMK2D capture antibody 1.3 μg / mL, UPF1 capture antibody 1.4 μg / mL, ALDH2 capture antibody 1.1 μg / mL, and EIF5A capture antibody 1.6 μg / mL, for simultaneous detection of the relative expression levels of six proteins. Detection procedure: Collect 5mL of peripheral venous blood → centrifuge and purify plasma → extract exosomes → detect the relative expression levels of 6 proteins → software calculates the relative abundance ratio → output "ANI positive / negative" result within 10 minutes (≥60 points is positive, original document confidence threshold); the software can automatically calculate the relative abundance ratio of 6 proteins and output the diagnostic result by matching the proportion range of claim 1 or 2. Performance validation: Based on the original document sample data, the kit was validated to have an intra-batch CV of <5%, an inter-batch CV of <8%, and a consistency of ≥92% with the mass spectrometry detection results, meeting the requirements for clinical use.

[0072] 6. Application Examples Sample processing: Collect 5 mL of peripheral venous blood from the subject, anticoagulate with EDTA, and centrifuge at 4℃ and 1900g for 10 minutes to obtain plasma; Exosome extraction: According to the kit instructions, add loading buffer and magnetic beads and incubate for 1 hour. After washing and elution, obtain exosomes and lyse them with RIPA lysis buffer. Protein detection: The expression levels of six proteins were detected by ELISA. The capture antibody mixture was prepared according to the ratio of DSCAM:IGF1:CAMK2D:UPF1:ALDH2:EIF5A=21:18:13:14:11:16. The absorbance was read at 562nm using an ELISA reader, and the concentration was calculated according to the standard curve. Results analysis: The concentration data was entered into the analysis software, and the "ANI positive / negative" result and confidence score (≥60 points for positive) were output within 10 minutes. Report generation: The software automatically generates standardized reports, including protein test results, diagnostic conclusions, and clinical recommendations, and supports PDF export.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A composition of exosome characteristic proteins for diagnosing HIV-related asymptomatic neurocognitive disorders, characterized in that, The exosome characteristic protein composition consists of six characteristic proteins: DSCAM, IGF1, CAMK2D, UPF1, ALDH2, and EIF5A.

2. The exosome-specific protein composition according to claim 1, characterized in that, Compared to individuals other than those with HIV-related asymptomatic neurocognitive impairment: The relative abundance of DSCAM and IGF1 increased, while the relative abundance of CAMK2D, UPF1, ALDH2 and EIF5A decreased.

3. The exosome-specific protein composition according to claim 2, characterized in that, Based on the total abundance of the exosome characteristic protein composition being 100%, wherein The relative abundance of DSCAM was 18% or higher, IGF1 was 15% or higher, CAMK2D was 16% or lower, UPF1 was 17% or lower, ALDH2 was 14% or lower, and EIF5A was 19% or lower.

4. The exosome-specific protein composition according to claim 3, characterized in that, Based on the total abundance of the exosome characteristic protein composition being 100%, wherein The relative abundance of DSCAM was 18–24%, IGF1 was 15–21%, CAMK2D was 10–16%, UPF1 was 11–17%, ALDH2 was 8–14%, and EIF5A was 13–19%.

5. The exosome-specific protein composition according to claim 4, characterized in that, In the exosome characteristic protein composition, the relative abundance ratio of the six characteristic proteins is DSCAM:IGF1:CAMK2D:UPF1:ALDH2:EIF5A = 21:18:13:14:11:

16.

6. The use of the exosome characteristic protein composition according to any one of claims 1 to 5 in the preparation of reagents or kits for diagnosing HIV-related asymptomatic neurocognitive impairment; The reagent or kit is used for qualitative and quantitative detection of the characteristic protein.

7. A kit for diagnosing HIV-related asymptomatic neurocognitive impairment, characterized in that, Includes specific detection reagents for detecting the exosome characteristic protein composition according to any one of claims 1 to 5.

8. The reagent kit according to claim 7, characterized in that, The specific detection reagent includes specific capture antibodies against each of the six characteristic proteins.

9. The reagent kit according to claim 8, characterized in that, The kit also includes one or more of the following: enzyme-labeled detection antibodies, gradient concentration standards, blocking solution, washing solution, chromogenic solution, and stop solution.

10. The use of the exosome-characteristic protein composition according to any one of claims 2 to 5 in constructing a computer model for diagnosing HIV-related asymptomatic neurocognitive impairment; said computer model is an XGBoost diagnostic model.

11. The application according to claim 10, characterized in that, The computer model performs the following functions: The detection results of the six characteristic proteins of the subject were obtained, and the relative abundance of each characteristic protein in the total amount of the six characteristic proteins was calculated; The relative abundance of each characteristic protein of the subject is compared with the relative abundance in the exosome characteristic protein composition. When it matches any one of claims 2 to 5, the subject is determined to be a patient with HIV-related asymptomatic neurocognitive impairment, and the determination result is output.

12. The application according to claim 11, characterized in that, The computer model is constructed as follows: Input the relative abundance data of the six characteristic proteins in the exosome characteristic protein composition, and after Z-score normalization, stratify and sample in an 8:2 ratio to obtain the training set and validation set; Five-fold cross-validation was used to optimize the initial parameters using a grid search. The initial parameters included nrounds=100, max_depth=6, learning_rate=0.1, subsample=0.8, and colsample_bytree=0.

8. The optimal parameters were finally determined to be nrounds=100, max_depth=6, learning_rate=0.07, subsample=0.8, and colsample_bytree=0.

8. A model is constructed based on the optimal parameters, and performance verification is completed.

Citation Information

Patent Citations

  • Compositions and methods for modulation of extracellular vesicle release and treatment of neurological disorders

    US20210260021A1