Construction method of alzheimer disease resistance related gene expression interaction network, marker screening method and application thereof

Abstract

This invention relates to the fields of bioinformatics and neurodegenerative diseases, and discloses a method for constructing an Alzheimer's disease resistance-related gene expression interaction network, a biomarker screening method, and their applications. The method includes: constructing disease model sample populations with multiple genetic backgrounds and corresponding normal control sample populations; acquiring cognitive function detection data, pathological burden detection data, and transcriptome sequencing data for each sample; calculating the average resistance index for each strain based on cognitive and pathological data; constructing a consensus co-expression module based on transcriptome data; constructing disease-related and control topological networks for each strain using characteristic genes of the consensus module as nodes; calculating the network topological reconnection index through edge set comparison; performing correlation analysis between this index and the resistance index to screen key consensus modules; extracting core genes and upstream transcription factors from the key modules to construct a resistance-related gene expression interaction network, providing a technical solution for Alzheimer's disease research and diagnosis.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and neurodegenerative disease research, specifically to methods for constructing gene expression interaction networks related to Alzheimer's disease resistance, methods for screening biomarkers, and their applications. Background Technology

[0002] Alzheimer's disease is a progressive neurodegenerative disease characterized by senile plaques formed by extracellular deposition of β-amyloid protein and neurofibrillary tangles formed by hyperphosphorylation of Tau protein. In recent years, clinical trials of drugs targeting single Aβ and Tau targets have repeatedly encountered setbacks (such as Aducanumab and Bapineuzumab), suggesting significant limitations in relying solely on "pathological clearance" strategies. Against this backdrop, research focus in the field has gradually shifted from "pathological clearance" to "cognitive resistance" mechanisms, that is, exploring the biological basis for how some individuals can maintain relatively normal cognitive function under the same or similar pathological burdens.

[0003] Current methods for identifying Alzheimer's disease-related genes often employ single-genetic-background experimental models, failing to distinguish between two core phenotypic differences: first, samples with impaired congenital pathological development and minimal lesion accumulation; and second, samples with sufficient pathological burden but exhibiting active cognitive resistance. The gene expression patterns of these two types of samples are fundamentally different; the former reflects differences in the pathogenesis process, while the latter directly relates to the resistance mechanism itself. If these two types of samples are analyzed together within a single genetic background, the molecular signals related to disease resistance will be masked or confused by differences in pathological burden, leading to systematic biases in biomarker selection. Furthermore, single-genetic-background models cannot reflect the genetic heterogeneity of AD in the population, making it difficult to translate research findings into clinical applications. Therefore, a method is urgently needed to analyze the gene expression and interaction networks related to cognitive resistance within a multi-genetic-background system to overcome the aforementioned shortcomings of existing technologies. Summary of the Invention

[0004] This invention provides a method for constructing an expression interaction network of Alzheimer's disease resistance-related genes, a biomarker screening method, and their applications, in order to construct a genetic diversity-Alzheimer's disease animal model evaluation system that can simulate the genetic heterogeneity of human populations, in order to solve the technical problem that existing single genetic background models cannot distinguish between "pathological obstruction" and "functional resistance under pathological burden".

[0005] In a first aspect, the present invention provides a method for constructing an expression interaction network of Alzheimer's disease resistance-related genes, the method comprising the following steps: S1. Construct a genetic diversity-Alzheimer's disease animal model system. The genetic diversity-Alzheimer's disease animal model system includes diseased model sample populations with multiple different genetic backgrounds (AD-CC) and normal control sample populations with the same background as the diseased model sample populations (Ntg-CC).

[0006] In one alternative implementation, the disease model sample population includes diseased mice; The diseased mice were selected from at least one of the following: AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING.

[0007] In one optional implementation, the disease model sample population was obtained by crossing six different paternal lines (C57BL / 6J, A / J, PWK / Phj, LOT_FC, STUCKY_HF, PIPING_BD) with the maternal line B6.Cg-Tg (APPswe, PSEN1dE9) double transgenic mice; the normal control sample population was the Ntg-CC population, obtained by crossing the same paternal line with the wild-type maternal line. Each subpopulation had a sample size of no less than 10 mice, for a total sample size of 319 mice.

[0008] S2. Determine the cognitive function test data, pathological burden test data, and transcriptome sequencing data of each sample in the normal control sample population and the disease model sample population.

[0009] In one alternative implementation, cognitive function testing data includes the percentage of rigidity time in a situational fear memory test; pathological burden testing data includes the percentage of Aβ (6E10 labeled) immunohistochemically positive area in the hippocampus; and transcriptome sequencing data are derived from high-throughput sequencing of total RNA extracted from hippocampal tissue.

[0010] S3. Based on cognitive function test data and pathological load test data, determine the average resistance index of each strain.

[0011] In one optional implementation, the resistance index is calculated as follows: using the C57BL / 6J strain as a reference genetic background, a linear regression model is established with Aβ load as the independent variable and situational fear memory level as the dependent variable; the residual of each individual's actual memory level relative to the model's predicted value is calculated; the residual is positively shifted to eliminate negative values ​​to obtain the resistance index, where a higher resistance index indicates a stronger cognitive maintenance ability of the individual under the same Aβ load.

[0012] S4. Generate a sample expression matrix based on transcriptome sequencing data.

[0013] In one alternative implementation, constructing the sample expression matrix includes quality control of the raw sequencing reads, alignment to a reference genome, gene-level quantification, and batch effect correction.

[0014] S5. Perform multiple random samplings on the sample representation matrix, and construct a co-representation network for each sample subset obtained from each random sampling.

[0015] In one alternative implementation, the random sampling is performed 2000 times, with 80% of all samples randomly selected each time to form a subset; the co-expression network is constructed based on the WGCNA method, using the β power of the Pearson correlation coefficient as the weighted adjacency matrix, and the soft threshold β=10.

[0016] S6. Based on the co-expression network, determine the consensus module; the consensus module includes multiple module feature genes.

[0017] In one optional implementation, determining the consensus co-expression module includes: counting the number of times each gene pair is co-sampled in a random subset and assigned to the same module, and constructing a consensus matrix; performing hierarchical clustering using 1 minus the consensus matrix as a distance metric, dividing the consensus module using a dynamic tree cutting algorithm, and merging adjacent modules whose feature gene correlation coefficient is greater than 0.85.

[0018] S7. Using the module feature genes as network nodes, construct the disease topology network corresponding to the disease model sample population and the control topology network corresponding to the normal control sample population for each strain.

[0019] In one alternative implementation, the edges of the topology network are valid connections after the Pearson correlation coefficient between the feature genes of the modules is filtered by a double threshold, where the double threshold includes an absolute value of the correlation coefficient |r|>0.6 and an FDR-corrected q value<0.1.

[0020] S8. Based on the diseased topology network and the control topology network, determine the network topology reconnection index of the corresponding strain.

[0021] In one alternative implementation, the network topology reconnection index = (number of lost edges + number of gained edges) / total number of edges in the superimposed network × 100%, where lost edges are edges that exist only in the control topology network but not in the diseased topology network, and gained edges are edges that exist only in the diseased topology network but not in the control topology network.

[0022] S9. Perform correlation analysis between the network topology reconnection index of each strain and the average resistance index of each strain, and select key modules that are significantly correlated with the resistance index from the consensus modules.

[0023] In one optional implementation, the correlation analysis is Pearson correlation analysis; the key module is the Top 5 consensus module, which identifies the network topology reconnection index and the resistance index with the largest absolute value of their correlation coefficients and which are statistically significant.

[0024] S10. Extract the core genes and upstream transcription factors that regulate the core genes from the key modules. Based on the regulatory relationship between the core genes and upstream transcription factors, construct an expression interaction network of genes related to Alzheimer's disease resistance.

[0025] In one alternative implementation, the core gene is obtained through protein-protein interaction network analysis and node centrality sorting; upstream transcription factors are obtained through transcription factor binding site enrichment analysis of the TRANSFAC database and / or JASPAR database.

[0026] In one alternative implementation, the core gene includes at least one of SSR1, BNIP3L, SEPT7, SELT, ZBTB6, CCNC, GMPS, NFIB, and TRAPPC13; the upstream transcription factor includes NR2F6 and / or SOX1.

[0027] In an alternative implementation, the method further includes a step of cross-species validation of core genes and / or upstream transcription factors in a population database.

[0028] In an optional implementation, step S9 includes: for each consensus module, determining the module reconnection score corresponding to the consensus module based on the neighbor set of the disease topology network and control topology network corresponding to the consensus module in different strains; For each consensus module, determine the correlation between the module reconnection score of the consensus module and the average resistance index of the corresponding strain, and determine the resistance correlation coefficient and robustness index corresponding to the consensus module. The comprehensive score of each consensus module is determined based on the module reconnection score, resistance correlation coefficient, and robustness index. Based on the ranking of comprehensive scores, key modules that are significantly correlated with the resistance index are selected.

[0029] In one optional implementation, step S3 includes: constructing a linear regression model based on cognitive function test data and pathological burden test data of the reference strain; Based on the linear regression model, as well as the cognitive function test data and pathological burden test data of each sample, the individual resistance index of each sample was determined. The average resistance index of each individual sample under the same strain is calculated by averaging the individual resistance indices.

[0030] In an optional implementation, step S6 includes: based on all co-expression networks, counting the number of times each gene pair is sampled in the same sampling subset, and constructing a gene co-occurrence indicator matrix; Count the number of times gene pairs with common sampling relationships are assigned to the same module, and construct a co-clustering matrix; Based on the gene co-occurrence indicator matrix and co-clustering matrix, all genes are clustered using average link hierarchical clustering, and a dynamic tree cutting algorithm is used for clustering to obtain multiple consensus modules.

[0031] In an optional implementation, step S7 includes: determining the correlation coefficient between the feature genes of each module; Gene pairs whose absolute correlation coefficient is greater than a preset threshold and meet the preset requirements after multiple tests are selected as valid edges. For each strain, using module feature genes as network nodes and correlation coefficients as edge weights, disease-related topology networks and control topology networks were constructed based on the selected effective edges.

[0032] In an optional implementation, step S8 includes: superimposing the diseased topological network of the same strain with the control topological network to obtain a superimposed network; Statistically compare lost edges in a topological network with gained edges in a diseased topological network; The network topology reconnection index is obtained by dividing the sum of the number of lost edges and the number of gained edges by the total number of edges in the superimposed network.

[0033] Secondly, this invention also provides a biomarker screening method based on the expression interaction network of Alzheimer's disease resistance-related genes, the biomarker screening method comprising the following steps: S11. Construct an expression interaction network of Alzheimer's disease resistance-related genes using the first approach; S12. Extract core genes and / or upstream transcription factors from the expression interaction network as candidate biomarkers; S13. Validate the association between candidate biomarkers and cognitive resistance index in an independent population cohort, and identify biomarkers that demonstrate significant validation as Alzheimer's disease resistance-related biomarkers.

[0034] In an optional implementation, step S12 includes: evaluating the gene characteristics of consensus modules in the Alzheimer's disease resistance-related gene expression interaction network, and determining the module membership degree and gene significance of characteristic genes in each module; the consensus module includes multiple module characteristic genes. Key genes are obtained by screening characteristic genes of a module based on intra-module gene connectivity, module membership, and gene significance. This screening includes: selecting genes whose module membership is greater than a first preset percentile of the membership of all genes within the corresponding consensus module and whose module membership reaches a first preset significance level; selecting genes whose intra-module gene connectivity is greater than a second preset percentile of the connectivity of all genes within the consensus module; selecting genes whose gene significance reaches a second preset significance level after verification and correction; and identifying genes that simultaneously meet all three screening criteria as key genes. Pre-defined sequences upstream and downstream of the transcription start sites of each key gene were extracted as promoter regions. Based on the position weight matrix of transcription factors in a preset transcription factor database, the promoter region is scanned according to a preset score threshold to obtain potential transcription factor binding sites; Using the promoter sequences of all genes within the same consensus module as the background set, enrichment tests were conducted based on the distribution of potential transcription factor binding sites to screen upstream transcription factors that meet the preset correction significance criteria and enrichment intensity requirements. Upstream transcription factors and / or key genes are used together as candidate biomarkers.

[0035] In one alternative implementation, validation of the population cohort includes: correlation analysis of candidate gene expression levels and cognitive resistance index based on the Allen Brain Aging / TBIStudy database, and subgroup analysis based on Aβ pathological burden stratification.

[0036] In one alternative implementation, the Alzheimer's disease resistance-related biomarkers include at least one of BNIP3L, SELT, GMPS, and NR2F6.

[0037] In an optional implementation, after step S13, the method further includes: S14, acquiring transcriptome data of brain tissue and clinicopathological sample data of the population, and constructing a population disease resistance index based on a preset regression model. S15. Screen key genes and / or upstream transcription factors for human homologous genes as candidate genes. S16. Conduct multi-dimensional population association analysis on candidate genes and verify the cross-species conservation of the regulatory relationship between key genes and upstream transcription factors; multi-dimensional population association analysis includes the association between candidate genes and population disease resistance index, the association between candidate genes and Alzheimer's disease pathology and neuroinflammatory traits, and the predictive evaluation of candidate genes on dementia diagnosis results. S17. Identify candidate genes and / or upstream transcription factors that have been validated through multidimensional population studies as human-related biomarkers for resistance to Alzheimer's disease.

[0038] The technical solution of this invention has the following advantages: First, a resistance index evaluation system based on pathological-cognitive regression residuals has been added to the animal model phenotypic evaluation steps.

[0039] Specifically, the technical solution of this application changes the existing technology's reliance on a single behavioral reading. In steps S2 and S3, it explicitly introduces quantitative detection of Aβ immunohistochemical plaque burden and establishes a pathological-cognitive regression model based on a single reference genetic background (preferably the C57BL / 6J strain). During execution, this application first uses the percentage of Aβ (6E10 marker) positive area in the hippocampus as the pathological level of Alzheimer's disease (independent variable), and the percentage of situational fear memory rigidity time, after baseline correction, as the memory level (dependent variable). A simple linear regression model is then established in the reference strain (C57BL / 6J). Based on this, the residual value of the actual memory level of each test mouse i (including other genetically diverse strains) relative to the predicted value of the regression model is calculated. To further facilitate comparison and interpretation, all calculated residual values ​​were positively shifted to eliminate negative values. The disease resistance index was ultimately defined as follows: ,in The absolute value of the minimum residual for all samples (i.e., the shift constant) is taken. This index can quantify the differences in individual cognitive maintenance ability under the same Aβ load level, overcoming the systematic defect in traditional methods that confuses two fundamentally different phenotypes: "less pathology equals better cognition" and "functional resistance under pathological load".

[0040] Second, a closed-loop analysis step has been added to the transcriptome data analysis process: "topological reconnection analysis of disease-induced network based on consensus-weighted gene co-expression network, identification of core regulatory factors, and cross-species homology verification".

[0041] Specifically, the technical solution of this application changes the existing technology's approach of stopping at a list of differentially expressed genes or a single overall network construction, and explicitly adds the following specific analysis sequences in steps S5 to S8: ①Construction and module identification of consensus-weighted gene co-expression network: RNA-seq transcriptome sequencing was performed on hippocampal tissues of mice from various strains in the high-resistance and low-resistance groups. The consensus WGCNA method was used to integrate expression data from multiple strains to construct co-expression modules, thereby improving the robustness and cross-strain universality of module identification.

[0042] ② Quantitative calculation of the disease-induced network topology reconnection ratio: Using non-transgenic cognitively intact control samples (Ntg-CC) and diseased samples (AD-CC) from each strain as inputs, and consensus module characteristic genes as nodes, specific module topology networks for each strain were independently constructed. Based on this, the proportion of changes in network connectivity structure before and after AD induction was quantitatively calculated through edge set comparison, defined as the "disease-induced network topology reconnection ratio". Furthermore, a cross-strain association analysis was conducted between this network topology reconnection ratio and the Alzheimer's disease resistance level of each strain to identify the dynamic changes in topology networks associated with the resistance phenotype.

[0043] ③ Core gene screening and upstream transcription factor prediction based on multidimensional indicators: Modules with a significant negative correlation to network topology reconnection strength and resistance level were selected as key modules. Genes within a module were ranked using a combination of three dimensions: module membership, gene significance, and intra-module connectivity, identifying core genes occupying pivotal positions in the co-expression network. Enrichment analysis of transcription factor binding sites was performed on the promoter regions of the core gene set to predict potential upstream regulatory transcription factors.

[0044] ④ Cross-species homology verification and clinical relevance assessment of population data: RNA-seq data and accompanying clinicopathological information from human autopsy hippocampal tissue were obtained from the Allen Brain Institute or other public databases. A population cognitive resistance index was constructed using the same residual calculation method as in the mouse model. The correlation between the expression levels of the identified core genes and predicted transcription factors in human hippocampal tissue and the cognitive resistance index was analyzed. Further stratified analysis by sex and Aβ pathological burden was performed to assess its value as an independent prognostic indicator. Simultaneously, a comparative conservation analysis of the promoter sequences of core genes in humans and mice was conducted to verify the cross-species conservation of regulatory mechanisms at the genomic sequence level.

[0045] The addition of this specific step sequence represents a technological leap from "static strain-specific differential expression analysis" to "cross-species comparison of dynamic reconnection of disease-induced networks." By introducing a quantitative indicator of the reconnection ratio of topological networks, this application can not only identify gene modules related to resistance but also capture the changing patterns of gene synergistic relationships under AD pathological perturbations. Simultaneously, by adding a promoter sequence conservation analysis step, the screening results are validated across species from both sequence evolution and transcriptional regulation dimensions, significantly improving the reliability and reproducibility of the screened molecular targets in clinical applications. Attached Figure Description

[0046] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the method for constructing an interaction network of gene expression related to Alzheimer's disease resistance.

[0048] Figure 2 This is a schematic diagram of the construction and experimental procedures for the F1 generation mouse Ntg-CC and AD-CC populations.

[0049] Figure 3 This represents the changes in fear acquisition and fear memory in AD-CC mice with different genetic backgrounds; (AB) the percentage of fixation time in different groups (Ntg-female, female Ntg mice; Ntg-male, male Ntg mice; AD-female, female AD mice; AD-male, male AD mice) of F1 generation mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) on the situational fear memory test (A) and the cueed fear memory test (B). C57, A / J, PWK, PIPING: One-way ANOVA, P < 0.05; LOT, STUCKY: One-way ANOVA, P > 0.05. Data are presented as mean ± standard error, significance level: P<0.05; P<0.01; P < 0.001. N = 319.

[0050] Figure 4 This section presents the pathological characteristics of Aβ in AD-CC mice from different genetic backgrounds (AB). Aβ load in the hippocampus of female and male AD-CC mice from different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) (A) and its quantitative analysis (B), scale bar 250 μm. Each group has n = 3 mice. Data are presented as mean ± standard error, significance level: P<0.05 P<0.01 P<0.001.

[0051] Figure 5 The different genetic backgrounds of AD-CC mice showed different levels of AD resistance; (A) Regression of corrected episodic fear memory level against hippocampal Aβ load in the C57 background (red) was y=-0.2989x+0.7725, P=0.0132. (B) Resistance index of female and male AD-CC mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING). Compared with C57-F, P<0.05 (t-test); #, compared with C57-M, P<0.05 (t-test).

[0052] Figure 6 This section shows microglia changes in the hippocampus of AD-CC mice under different genetic backgrounds. (A) Representative images of Aβ plaques and Iba1+ microglia distribution in different subregions of the hippocampus of Ntg and AD mice from F1 generation mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING), scale bar 50 μm. (B) Representative images of microglia aggregation (Iba1+) and activation (CD68+Iba1+) around Aβ plaques in the hippocampus of AD-CC mice from F1 generation mice with different genetic backgrounds, scale bar 100 μm. (C) Percentage of Iba1-positive area in the hippocampus of Ntg-CC and AD-CC mice from F1 generation mice with different genetic backgrounds. Significance level: P<0.05; P<0.01; P < 0.001. Each group had n = 6, half male and half female; △, male; ○, female. (D) Number of Iba1-positive microglia per diameter (μm) Aβ plaque in the hippocampus of AD-CC mice from F1 generation mice with different genetic backgrounds. Significance level: P < 0.05 P<0.01 P < 0.001. Each group had n = 6, half male and half female; △, male; ○, female. (E) Percentage of activated microglia (CD68 and Iba1 double positive) around Aβ plaques per diameter (μm) in the CA1 region of the hippocampus of AD-CC mice from F1 generation mice with different genetic backgrounds compared to microglia (Iba1 positive). Significance level: P<0.05; P<0.01; P < 0.001. Each group has n = 6, with half males and half females, △ for males and ○ for females.

[0053] Figure 7 This section describes the changes in the complexity of hippocampal dendritic branching in AD-CC mice under different genetic backgrounds. (A) Representative Golgi staining images of the hippocampus from different groups (Ntg-female, female Ntg mice; Ntg-male, male Ntg mice; AD-female, female AD mice; AD-male, male AD mice) in F1 generation mice from different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING), with a scale bar of 50 μm. (B) The total number of concentric circles of different radii crossing neuronal processes in the hippocampus of different groups of F1 generation mice from different genetic backgrounds. Significance level: P<0.05; P<0.01; P < 0.001. n = 3 per group. (C) Crossover distribution characteristics of hippocampal numbers in different groups of F1 generation mice with different genetic backgrounds at concentric circles with radii of 50, 75, 100, 125, 150, 175, and 200 μm from the cell body. n = 3 per group.

[0054] Figure 8 This section shows the changes in hippocampal CA1 synaptic density in AD-CC mice under different genetic backgrounds. (A) Representative images of synaptic density in the hippocampal CA1 region of different groups (Ntg-female, female Ntg mice; Ntg-male, male Ntg mice; AD-female, female AD mice; AD-male, male AD mice) from F1 generation mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) under transmission electron microscopy. Red arrows indicate synaptic structures. Scale bar is 2 μm. (B) Number of intact synapses per unit area in the hippocampal CA1 region of different groups of F1 generation mice with different genetic backgrounds. Significance level: P<0.05; P<0.01; P < 0.001. n = 3 per group.

[0055] Figure 9This section shows the changes in postsynaptic PSD thickness in the CA1 region of the hippocampus in AD-CC mice under different genetic backgrounds. (A) Representative images of synaptic morphology in the CA1 region of the hippocampus from different groups (Ntg-female, female Ntg mice; Ntg-male, male Ntg mice; AD-female, female AD mice; AD-male, male AD mice) in the F1 generation of mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) under transmission electron microscopy. The PSD region is shown within the white dashed box. The scale bar is 0.1 μm. (B) PSD thickness of the synaptic CA1 region of the hippocampus in different groups of F1 generation mice with different genetic backgrounds. Significance level: P<0.05; P<0.01; P < 0.001. n = 3 per group.

[0056] Figure 10 This is a heatmap showing the correlation between module feature expression values ​​and samples. Based on consensus-weighted gene co-expression network analysis, 72 co-expression modules were identified. The ME (Mean Expression Value) of each module was calculated, and a heatmap showing its correlation with all samples was generated. Rows represent gene co-expression modules, columns represent samples, and color intensity reflects the correlation strength between module ME and samples (orange indicates positive correlation, blue indicates negative correlation). Samples are arranged by strain, sex, and treatment status. Each group has n=3.

[0057] Figure 11 These are the topological characteristics of gene expression networks under different genetic backgrounds and basal states; (AD) Topological characteristics of hippocampal transcriptome gene expression networks in F1 generation mice (female and male) of Ntg mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING), including the number of edges (A), network density (B), average number of nodes (C), and modularity coefficient (D).

[0058] Figure 12This section describes the changes in disease-induced gene network reprogramming in AD-CC mice. (A) Overlay of gene co-expression network topologies in mice with different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) under basal (Ntg) and disease (AD) states. Nodes represent modules, edges represent co-expression relationships between modules (Pearson correlation |r|>0.6, FDR<0.1), and edge colors represent the regulatory architecture of each strain under disease and basal states (blue, lost edges; orange, gained edges; gray, shared edges). (B) Quantitative comparison of the number of gene expression network topology edges in basal and disease states under different genetic backgrounds (blue, Ntg; orange, AD). (C) Correlation analysis between network topology reconnection index and strain disease resistance level, r = -0.92, P = 0.0094. The network topology reconnection index is defined as (number of new edges + number of lost edges) / total number of edges.

[0059] Figure 13 This is a scatter plot showing the correlation between the average reconnection score of each module and the disease resistance index, based on a multi-dimensional comprehensive score. Each point represents a module, and the top 5 key modules selected by the multi-dimensional comprehensive score (honeydew1, darkorange2, navajowhite2, plum2, cyan) are marked in red.

[0060] Figure 14 This study presents a stratified correlation analysis of key reconnection modules with neuropathological and behavioral phenotypes. The stratified correlations between key reconnection modules (honeydew1, darkorange2, navajowhite2, plum2, cyan) and nine neuropathological and behavioral phenotypes under different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING) are analyzed. The bar direction represents the Pearson correlation coefficient (upward for positive correlation, downward for negative correlation), and the significance level is: P<0.05; P<0.01; (P<0.001). Nine neuropathological and behavioral phenotypes were included: behavioral tests (episodic fear memory, cue fear memory), pathological indicators (Aβ load), neuronal indicators (synaptic density, PSD thickness, total number of dendritic crossings), and immune indicators (percentage of Iba1 positive area, number of microglia around each plaque diameter, and number of activated microglia around each plaque diameter).

[0061] Figure 15The expression levels of Nr2f6 and Sox1 were positively correlated with the resistance index; the expression levels of Nr2f6 (right) and Sox1 (left) log2(FPKM + 1) (horizontal axis) were positively correlated with the disease resistance index (vertical axis) of each sample. Different colored dots represent different genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING). Sox1: r=0.366, P=0.0281; Nr2f6: r=0.565, P<0.0001. N = 36.

[0062] Figure 16 This section presents the distribution of the population's disease resistance score and its relationship with AD pathology and dementia status. (A) Histogram of the population's disease resistance index distribution, Shapiro-Wilk test W = 0.839, P < 0.001. (B) Box plot of the population's disease resistance index under different dementia statuses. The resistance score in the non-dementia group was significantly higher than that in the dementia group (non-dementia group: 1.06 ± 0.27; dementia group: -1.11 ± 0.31; Wilcoxon rank-sum test: W = 1720, P < 0.001). (C) Scatter plot of the correlation between Aβ pathological burden and the predicted proportion of dementia. The horizontal axis represents the Aβ pathological level (percentage of hippocampal Aβ-positive area measured by IHC in a fixed sample), and the vertical axis represents the predicted probability of dementia risk. The scatter color represents the patient's actual disease status (dementia vs. non-dementia).

[0063] Figure 17 This is a validation of the expression of key genes and transcription factors in a population cohort; (A) Bubble plot showing the correlation between the expression levels of nine key genes and two transcription factors and the resistance index in different pathological load stratifications and genders. The bubble size represents the absolute value of the correlation coefficient (|r|), and the color indicates the direction of the correlation (red for positive correlation, blue for negative correlation); ·, P<0.1; P < 0.05. Analysis was performed separately in different sexes within the high pathological burden group (Aβ pathology above median) and the low pathological burden group. (B) Scatter plot of the correlation between the expression levels of the three genes BNIP3L, SELT, and GMPS in the high pathological burden male sample (n=22) and the resistance index. BNIP3L (r=-0.570, P=0.006), SELT (r=-0.476, P=0.025), and GMPS (r=-0.446, P=0.038) were all significantly negatively correlated with the resistance index. The scatter plot color represents the patient's actual pathological state (no dementia, dementia).

[0064] Figure 18This is a heatmap showing the association between key genes and transcription factors and AD-related traits in a population cohort; and a heatmap showing the association between nine key genes (SSR1, BNIP3L, SEPT7, SELT, ZBTB6, CCNC, GMPS, NFIB, TRAPPC13) and two transcription factors (NR2F6, SOX1) and 22 AD-related traits (including AD pathology, glial cells, neuroinflammatory factors, and Tau pathology) in different sexes. †, 0.05 <P<0.1; P<0.05; P < 0.01. Females (bottom), n = 37; Males (top), n = 46.

[0065] Figure 19 This study presents a logistic regression prognostic analysis of the predictive power of key gene and transcription factor expression on dementia status. The expression levels of nine key genes (SSR1, BNIP3L, SEPT7, SELT, ZBTB6, CCNC, GMPS, NFIB, TRAPPC13) and two transcription factors (NR2F6, SOX1) were independently predicted in all samples (A), as well as in female and male samples (B). Dotted lines represent OR values, i.e., 95% CI intervals. Gray indicates no significance; orange indicates P < 0.05.

[0066] Figure 20 This is a heatmap of co-expression correlations of transcription factors and key gene axes in a population cohort. Detailed Implementation

[0067] The following embodiments are provided to better understand the present invention, but the following embodiments do not constitute a limitation on the content and scope of protection of the present invention. Any product that is the same as or similar to the present invention, derived by any person under the guidance of the present invention or by combining the features of the present invention with other prior art, falls within the scope of protection of the present invention.

[0068] Unless otherwise specified, all experimental steps or conditions in the examples were performed according to conventional experimental procedures and conditions in the art. Reagents or instruments whose manufacturers are not specified are all commercially available products.

[0069] According to an embodiment of the present invention, an embodiment of a method for constructing an Alzheimer's disease resistance-related gene expression interaction network is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0070] This embodiment provides a method for constructing an Alzheimer's disease resistance-related gene expression interaction network, which can be used on servers, terminals, and mobile terminals, such as mobile phones and tablets. Figure 1 This is a flowchart of a method for constructing an Alzheimer's disease resistance-related gene expression interaction network according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: S1. Construct a genetic diversity-Alzheimer's disease animal model system. The genetic diversity-Alzheimer's disease animal model system includes diseased model sample groups with different genetic backgrounds, as well as normal control sample groups with the same background as the diseased model sample groups.

[0071] In one alternative implementation, the disease model sample population includes diseased mice; The diseased mice were selected from at least one of the following: AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING.

[0072] In one optional implementation, the disease model sample population was obtained by crossing six different paternal lines (C57BL / 6J, A / J, PWK / Phj, LOT_FC, STUCKY_HF, PIPING_BD) with the maternal line B6.Cg-Tg (APPswe, PSEN1dE9) double transgenic mice; the normal control sample population was the Ntg-CC population, obtained by crossing the same paternal line with the wild-type maternal line. Each subpopulation had a sample size of no less than 10 mice, for a total sample size of 319 mice.

[0073] In the specific implementation process, six paternal lines with different genetic backgrounds were selected and crossed with a maternal line carrying the AD pathogenic gene (B6.Cg-Tg(APPswePS1dE9) mice) to obtain an F1 generation AD mouse population with genetic diversity (e.g., Figure 2 ).

[0074] The paternal strains include: the traditional laboratory inbred mouse strains A / J and C57BL / 6J; the co-inbred mouse strains LOT_FC, STUCKY_HF, and PIPING_BD; and the wild-source inbred mouse strain PWK / Phj. Among these, A / J, C57BL / 6J, and PWK / Phj were bred by the Jackson Laboratory in the United States and introduced to the Institute of Laboratory Animal Science, Chinese Academy of Medical Sciences for breeding; LOT_FC, STUCKY_HF, and PIPING_BD were bred by the University of Western Australia and introduced to the Institute of Laboratory Animal Science, Chinese Academy of Medical Sciences for breeding. The LOT_FC, STUCKY_HF, and PIPING_BD strains are standard strains for genetically diverse mouse populations. These strains have been included in the National Animal Model Resource Bank for Human Diseases of the Institute of Laboratory Animal Science, Chinese Academy of Medical Sciences (https: / / www.namr.org.cn) and are publicly accessible. Those skilled in the art can legally obtain these resources from the database.

[0075] The maternal strain was the B6.Cg-Tg (APPswePS1dE9) mouse carrying double transgenes, bred by the Jackson Laboratory in the United States and introduced and bred by Beijing Huafukang Biotechnology Co., Ltd. Male mice from the aforementioned paternal strain were crossed with female mice from the maternal strain to obtain the F1 generation of genetically diverse AD mouse populations (i.e., the AD-CC population, including subpopulations such as AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING) and their littermates' control mice that do not carry the AD transgene (i.e., the Ntg-CC population, including subpopulations such as Ntg-C57, Ntg-A / J, Ntg-PWK, Ntg-LOT, Ntg-STUCKY, and Ntg-PIPING).

[0076] Male mice from the aforementioned paternal line were crossed with female mice from the maternal line to obtain F1 generation genetically diverse AD mouse populations (i.e., AD-CC populations, including subpopulations such as AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING) and their littermate control mouse populations that do not carry the AD transgene (i.e., Ntg-CC populations, including subpopulations such as Ntg-C57, Ntg-A / J, Ntg-PWK, Ntg-LOT, Ntg-STUCKY, and Ntg-PIPING).

[0077] Thus, a genetic diversity-Alzheimer's disease animal model system with genetic diversity was constructed to simulate the heterogeneous relationship between AD pathology and cognitive performance in human populations.

[0078] S2. Determine the cognitive function test data, pathological burden test data, and transcriptome sequencing data of each sample in the normal control sample population and the disease model sample population.

[0079] In one alternative implementation, cognitive function testing data includes the percentage of rigidity time in a situational fear memory test; pathological burden testing data includes the percentage of Aβ immunohistochemically positive area in the hippocampus; and transcriptome sequencing data is derived from high-throughput sequencing of total RNA extracted from hippocampal tissue.

[0080] In one alternative implementation, behavioral tests were performed on mice aged 8-10 months. The subjects were mice from all AD-CC populations (including subpopulations such as AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING) and Ntg-CC populations (including subpopulations such as Ntg-C57, Ntg-A / J, Ntg-PWK, Ntg-LOT, Ntg-STUCKY, and Ntg-PIPING). To ensure statistical power, the minimum number of animals in each experimental group (same strain, sex, and genotype) was 10; a total of 319 animals were used in this study.

[0081] The situational fear memory testing device includes a fear conditioning chamber that records the rigidity behavior of mice, with a programmable electric shock grid at the bottom of the chamber. The testing procedure is conducted over four days: Day 1 is the adaptation period, during which mice are placed in the device and allowed to explore freely for 10 minutes to adapt to the environment; Day 2 is the learning period, after which mice are placed in the device and their baseline rigidity is recorded for 3 minutes, followed by pairing training with 5 pairs of auditory stimuli (30 seconds, 70 dB, 5000 Hz) and electric shock stimuli (1 second, 0.5 mA), with a 30-second interval between each pair; Day 3 is the situational fear memory test, during which mice are returned to the original device for 5 minutes without any stimulation, and the percentage of rigidity time is recorded as an indicator of situational fear memory; Day 4 is the cueed fear memory test, after changing the shape, base material, and odor of the device, only auditory stimuli are given, and the mice's cueed memory of the auditory conditions (e.g., ...) is assessed. Figure 2 ).

[0082] The percentage of rigidity time in the situational fear memory test was calculated as follows: Percentage of rigidity time = (Rigidity time / Total test time 300 seconds) × 100%. A higher value indicates better preservation of hippocampal-dependent cognitive memory function in the mice. Analysis of the group behavioral test data from the AD-CC and Ntg-CC subgroups provides valid supporting evidence for the successful construction of the AD genetically heterogeneous mouse model in this patent.

[0083] Following the above method, conditioned fear memory behavioral tests were performed on female and male mice from the AD-CC and Ntg-CC subgroups. The test results are as follows: Figure 3 As shown.

[0084] In one alternative implementation, within 24 hours of the completion of behavioral testing, samples were collected and pathologically examined from all AD-CC groups (including subgroups such as AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING) and Ntg-CC groups (including subgroups such as Ntg-C57, Ntg-A / J, Ntg-PWK, Ntg-LOT, Ntg-STUCKY, and Ntg-PIPING). To ensure statistical power, a minimum of 3 animals were used in each experimental group (with the same strain, sex, and genotype), for a total of 72 animals.

[0085] The experimental procedure was as follows: Cardiac perfusion and brain harvesting: Mice were deeply anesthetized by intraperitoneal injection of tribromoethanol (0.2 ml / 10 g, 1.25%), and then perfused with pre-cooled PBS (0.01 M, pH 7.4) until the liver turned white. Subsequently, the liver was fixed by perfusion with 4% paraformaldehyde (PFA, dissolved in PBS). After perfusion, the whole brain was harvested and fixed in 4% paraformaldehyde for 24 hours (4℃).

[0086] Dehydration and sectioning: After fixation, the brain tissue was dehydrated with graded ethanol, cleared with xylene, and then embedded in paraffin. Coronal sections were prepared using a microtome, with a section thickness of 5 μm. Based on the Allen mouse brain atlas (Reference Atlas, version 1, 2008), brain slices were selected at four planar locations: -1.655 mm, -2.255 mm, -2.88 mm, and -3.455 mm posterior to the anterior fontanelle (with an allowable error of ±0.1 mm), and the slices were attached to glass slides.

[0087] Immunohistochemical staining: After dewaxing in xylene and rehydration with graded ethanol, the sections were subjected to antigen retrieval in citrate buffer. After blocking with goat serum, anti-Aβ primary antibody was added, and the sections were incubated at 4°C for 16 hours. After washing with PBS, the sections were incubated sequentially with biotin-labeled secondary antibody and horseradish peroxidase-labeled streptavidin according to the kit instructions. DAB staining was performed, followed by dehydration and clearing, and then mounting with neutral resin.

[0088] Image acquisition and data analysis: The stained sections were scanned in their entirety using a digital slide scanner. ImageJ was used to calculate the area of ​​Aβ immunopositive deposits in the hippocampus. The Aβ plaque burden percentage was calculated as follows: Aβ burden = (Aβ immunopositive area / hippocampal area) × 100%. A lower value indicates a milder degree of AD pathological damage.

[0089] By analyzing the Aβ pathological quantitative data of the AD-CC group (a total of 6 genetic background subgroups) and the Ntg-CC control group, the population distribution characteristics of Aβ deposition in AD mice under different genetic backgrounds can be obtained. The hippocampus of each subgroup of AD-CC mice clearly showed Aβ load, while the corresponding Ntg-CC strains showed no Aβ load, and identifiable differences existed between different subgroups. This serves as the core pathological support for the successful construction of the AD genetic heterogeneity mouse model in this patent. Typical results can be seen in the corresponding pathological images and quantitative statistical graphs in the attached figures (e.g., Figure 4 (As shown).

[0090] The purpose of this step is to construct an animal model system of Alzheimer's disease (AD-CC) with genetic diversity to simulate the heterogeneous relationship between AD pathology and cognitive performance in the human population.

[0091] S3. Based on cognitive function test data and pathological load test data, determine the average resistance index of each strain.

[0092] In one optional implementation, step S3 includes: Step b1: Based on the cognitive function test data and pathological burden test data of the reference strain, construct a linear regression model; Step b2: Based on the linear regression model and the cognitive function test data and pathological burden test data of each sample, determine the individual resistance index of each sample; Step b3: Calculate the average resistance index of all samples under the same strain to obtain the average resistance index corresponding to each strain.

[0093] In one optional implementation, the resistance index is calculated as follows: using the C57BL / 6J strain as a reference genetic background, a linear regression model is established with Aβ load as the independent variable and situational fear memory level as the dependent variable; the residual of each individual's actual memory level relative to the model's predicted value is calculated; the residual is positively shifted to eliminate negative values ​​to obtain the resistance index, where a higher resistance index indicates a stronger cognitive maintenance ability of the individual under the same Aβ load.

[0094] In one alternative implementation, the resistance index is calculated as follows: 1. Establishment of a reference regression model Using the C57BL / 6J strain as a reference genetic background, the percentage of Aβ (6E10 marker) positive area in the hippocampus of this strain of mice was used as the pathological level of Alzheimer's disease (independent variable x). The percentage of situational fear memory rigidity time measured in the above steps, after being adjusted to the corresponding baseline, was used as the memory level (dependent variable y). A simple linear regression model was established in the C57BL / 6J strain. This model characterizes the expected relationship between Aβ pathological burden and cognitive performance in a reference strain.

[0095] Following the above method, a linear regression equation was established using the Aβ load of C57BL / 6J reference mice (n=12, half male and half female) as the independent variable and the adjusted situational fear memory level as the dependent variable. Figure 5 As shown in Figure A, the regression equation is y = -0.2989x + 0.7725 (R² = 0.32, P = 0.0132), indicating that in the C57BL / 6J reference strain, hippocampal Aβ loading is significantly negatively correlated with cognitive memory level.

[0096] 2. Calculation of residuals Based on the regression model described above, the residual value of the actual memory level of each test mouse i (including the C57BL / 6J strain and other genetically diverse strains) relative to the model's predicted value was calculated. A positive residual indicates that the individual's cognitive performance is better than expected based on their Aβ pathological burden, i.e., they exhibit resistance to AD pathology; a negative residual indicates that their cognitive performance is lower than expected.

[0097] 3. Definition of Resistance Index To facilitate comparison and interpretation, all calculated residual values ​​are positively shifted to eliminate negative values. The disease resistance index is defined as:

[0098] in, Let be the residual value of the i-th mouse. The absolute value of the minimum residual for all samples is taken (i.e., the translation constant). The higher this index, the more significantly the memory performance of this individual mouse is superior to the expected level of the C57BL / 6J reference strain under the same Aβ load, that is, the stronger its resistance to Aβ pathology.

[0099] Calculate the resistance index for each subgroup of mice using the formula described above. For example... Figure 5 As shown in B, the resistance indices of each subpopulation, from highest to lowest, are: AD-STUCKY (female, male), AD-LOT (female, male), AD-PWK (male), AD-A / J (female, male), AD-PWK (female), AD-C57 (female, male), and AD-PIPING (male, female). This result indicates that AD-CC mice from different genetic backgrounds exhibit significantly different levels of AD resistance, with the STUCKY strain conferring the strongest resistance phenotype.

[0100] In an optional implementation, S3 further includes multidimensional auxiliary pathological phenotypic detection: To further validate the biological rationale for the resistance index and provide multidimensional phenotypic anchoring for subsequent network analysis, the following auxiliary tests were performed on the brain tissue of mice from each strain: 1. Detection of neuroinflammatory markers Immunofluorescence staining was used to detect microglia-related indicators in the hippocampus to assess the level of neuroinflammatory response in AD mice with different genetic backgrounds. Specific indicators included: (i) the percentage of Iba1-positive area, reflecting the overall level of microglia; (ii) the number of Iba1-positive microglia around each unit diameter Aβ plaque, reflecting the degree of aggregation of plaque-related microglia; and (iii) the proportion of Iba1- and CD68-positive microglia to Iba1-positive microglia around each unit diameter Aβ plaque, reflecting the phagocytic activation status of microglia.

[0101] Immunofluorescence staining and image analysis were performed on mice in each experimental group (n=3 mice / strain / sex) according to the following steps: (i) Preparation of frozen sections of brain tissue Mice were deeply anesthetized by intraperitoneal injection of tribromoethanol (0.2 ml / 10 g, 1.25%), and then perfused with pre-cooled PBS (0.01 M, pH 7.4) until the liver turned white. The liver was then fixed by perfusion with 4% paraformaldehyde (PFA, dissolved in PBS). Brain tissue was completely removed and post-fixed in 4% PFA for 48 hours (4°C). The fixed brain tissue was then sequentially transferred to 15% sucrose solution (prepared in PBS) and 30% sucrose solution, and dehydrated for 24 hours each (4°C) until the tissue completely settled. The dehydrated brain tissue was trimmed into blocks, and coronal sections including the hippocampus were taken, embedded in OCT embedding medium, and then flash-frozen at -80°C. Using a cryostat (Leica CM1950), continuous coronal sections of brain tissue were prepared at -20°C, with a section thickness of 40 μm. Brain slices containing the CA1, CA2, CA3, and DG subregions of the hippocampus were collected and stored in cryopreservation solution at -20°C for later use. After the sections were naturally dried at room temperature for 2 hours, subsequent staining could be performed or they could be stored at -20°C.

[0102] (ii) Immunofluorescence staining Remove frozen sections and allow them to thaw at room temperature for 15 minutes. Rinse three times with PBS for 5 minutes each time. Then permeabilize with PBS containing 0.5% Triton X-100 at room temperature for 20 minutes. Discard the permeabilization buffer, add goat serum blocking buffer (PBS containing 5% goat serum), and block at room temperature for 1 hour. Place the sections in a humidified chamber to prevent drying. Discard the blocking serum, add a primary antibody mixture (rabbit anti-Iba1, 1:500, Wako; rat anti-CD68, 1:200, Abcam), ensuring complete coverage of the tissue (approximately 200 μL per section), and incubate overnight at 4°C (12-16 hours). The next day, recover the primary antibody and rinse three times with PBST (PBS containing 0.05% Tween-20) for 10 minutes each time. Add a mixture of fluorescein-labeled secondary antibodies (Alexa Fluor 488-labeled goat anti-rabbit, 1:500, Invitrogen; Alexa Fluor 594-labeled goat anti-rat, 1:500, Invitrogen) and incubate at room temperature in the dark for 60 minutes. After incubation, rinse three times with PBST for 10 minutes each time in the dark. Add a fluorescence quenching mounting medium containing DAPI stain (Vector Labs), and gently mount with a coverslip, avoiding air bubbles. Acquire fluorescence images of the mounted sections within 48 hours.

[0103] (iii) Image acquisition and quantitative analysis Images of the hippocampus were acquired using a fluorescence microscope under a 20x objective lens. Quantitative analysis was performed using ImageJ software. Total microglia levels were expressed as the percentage of Iba1-positive area to the total area analyzed. Plaque-associated microglia quantification: Five to eight Aβ plaques with a diameter of 50-100 μm were selected from each brain slice, and the number of Iba1-positive cells within a 50 μm radius around the plaques was measured. The results were expressed as the number of cells per 100 μm diameter periplasmic region. Microglia activation status: Within the same periplasmic region, the percentage of Iba1 and CD68 double-positive cells was counted out of the total number of Iba1-positive cells.

[0104] Comparison of the percentage of Iba1-positive area in Ntg and AD mice across different genetic backgrounds revealed a significant genetic background dependence in the overall level of microglia induced by AD pathology. Figure 6 A, C). Using C57-AD as a control, the number of Iba1-positive cells per unit diameter plaque was significantly increased in the A / J, PWK, LOT, and STUCKY strains compared to C57, while there was no significant difference between PIPING and C57. Figure 6 B, D). It is noteworthy that the PIPING genetic background is characterized by high overall activation but low plaque targeting; while in the resistant genetic background (LOT, STUCKY), microglia exhibit a unique "high recruitment-low phagocytosis" activation pattern. Figure 6 B, E). No sex differences were observed in the Ntg and AD groups within any genetic background. Figure 6 Representative immunofluorescence images are shown in Figures C, D, and E. Figure 6 .

[0105] 2. Neuronal dendrite morphology detection Golgi staining was used to observe the morphology of mouse hippocampal neurons to assess the impact of Alzheimer's disease (AD) pathology on the dendritic morphology of hippocampal neurons under different strains. Quantitative morphological assessment of neurons in the CA1 region of the hippocampus was performed using Sholl analysis. Specifically, concentric circles with radii ranging from 50 μm to 200 μm and intervals of 25 μm were drawn with the neuronal cell body as the center, and the number of intersections between each concentric circle and the neuronal processes was counted. This number of intersections quantitatively reflects the complexity and spatial distribution characteristics of dendritic branching.

[0106] Golgi staining and morphological analysis were performed on mice in each experimental group (n=3 mice / strain / sex) according to the following steps: (i) Brain tissue preparation and Golgi staining Golgi staining was performed using the FD Rapid Golgi Stain kit. After animal sacrifice, brain tissue was rapidly removed, and the surface blood was quickly rinsed off with double-distilled water. The tissue was then placed in a 1:1 mixture of solution A and solution B prepared one day in advance and soaked for 24 hours. The solution was then replaced with a fresh mixture, and the tissue was stored at room temperature in the dark for 2 weeks. The tissue was then transferred to solution C and kept at room temperature in the dark for 72 hours, with the solution C being replaced after 24 hours. At -20°C, the tissue was cut into 100 μm sections using a cryostat and transferred to slides containing solution C. Excess liquid was blotted with filter paper, and the sections were air-dried at room temperature in the dark for 2 days. The sections were rinsed twice with double-distilled water for 4 minutes each time. The sections were then placed in a freshly prepared 1:1:2 mixture of solution D and solution E and double-distilled water and stained in the dark for 10 minutes. The sections were rinsed twice with double-distilled water for 4 minutes each time. The sections were dehydrated sequentially in 50% ethanol, 70% ethanol, and 95% ethanol, for 4 minutes at each concentration gradient; then dehydrated four times in anhydrous ethanol, 4 minutes each time. Clearing was performed three times in xylene, 4 minutes each time. The sections were then mounted with neutral resin mounting medium and allowed to dry at room temperature in the dark before being observed under a microscope.

[0107] (ii) Image acquisition and Sholl analysis Images of intact neurons in the CA1 region of the hippocampus were acquired using an optical microscope under a 20x objective lens. Three morphologically intact and clearly stained pyramidal neurons were selected from each mouse. Quantitative analysis was performed using ImageJ software with the Shol analysis plugin: concentric circles with radii ranging from 50 μm to 200 μm and intervals of 25 μm were drawn with the neuronal cell body as the center, and the number of intersections between each concentric circle and the neuronal process was counted.

[0108] The experimental groups were tested using the method described above, and the results are as follows: The total crossover number of neurons in each group under different genetic backgrounds was compared. The results showed that in the C57, A / J, and PIPING genetic backgrounds, the AD group had a significantly lower number of crossovers than the Ntg control group (consistent between males and females), while there were no significant differences among the groups in the PWK, LOT, and STUCKY genetic backgrounds. Figure 7 B). Further analysis showed that the decrease in crossover number was mainly concentrated in the concentric circles at a distance of 125 μm from the cell body ( Figure 7 C). The above results indicate that reduced dendritic branching only occurred in C57, A / J, and PIPING genetic backgrounds, and was mainly manifested as branch loss at a distance of 125 μm from the cell body, suggesting a genetically background-dependent difference in susceptibility of hippocampal neurons to AD pathology. Representative Golgi staining images are shown below. Figure 7 A.

[0109] 3. Synaptic ultrastructure detection Transmission electron microscopy was used to observe the ultrastructure of the CA1 region of the hippocampus in Ntg-CC control mice and AD-CC model mice from various genetic backgrounds. Statistical analysis was performed on synaptic density and postsynaptic compact band thickness per unit electron microscopic field to assess the damage patterns of synaptic structures in mice with different genetic backgrounds under AD pathological conditions and their association with resistance phenotypes.

[0110] The following steps were followed to prepare transmission electron microscopy samples and perform image analysis on mice in each experimental group (n=3 mice / strain / sex): (i) Sample preparation After animal sacrifice, fresh hippocampal tissue was harvested and immediately immersed in pre-cooled 2.5% glutaraldehyde fixative (0.1 M phosphate buffer, pH 7.2) at 4°C for 2 hours. After fixation, the sample was rinsed three times with pre-cooled 0.1 M phosphate buffer (pH 7.2) for 10 minutes each time. The sample was then transferred to 1% osmium tetroxide fixative and fixed at 4°C for 2 hours. After fixation, the sample was rinsed three times with double-distilled water for 10 minutes each time. The sample underwent graded alcohol dehydration: sequentially immersed in 50%, 70%, and 90% ethanol solutions for 10 minutes each, followed by two 15-minute immersions in 100% ethanol. Propylene oxide was used as a transition solvent for displacement treatment twice, for 15 minutes each time. The samples were then progressively infiltrated in a mixture of propylene oxide and EPON 812 resin: a 1:1 propylene oxide:resin mixture for 1 hour at room temperature; a 1:4 propylene oxide:resin mixture for 1 hour at room temperature; and finally, pure EPON 812 resin for 2 hours at room temperature. The fully infiltrated samples were then placed in embedding plates containing fresh pure EPON 812 resin and subjected to gradient temperature polymerization: 45°C for 5 hours, 60°C for 5 hours, and 80°C for 5 hours.

[0111] After trimming, the polymerized embedding blocks were first semi-thinly sectioned (approximately 1 μm thick), stained with azuril-methylene blue, and observed under an optical microscope to locate and define the target region. Once the target region was identified, ultrathin sections (90 nm thick) were prepared using an ultramicrotome. The sections were then transferred to a copper mesh and stained sequentially with uranyl acetate and lead citrate. The stained ultrathin sections were then observed and imaged using a JEM-1400 transmission electron microscope under accelerating voltage.

[0112] (ii) Image acquisition and quantitative analysis Four electron micrographs (20,000× magnification) of the neural tapetum region of the CA1 area in the hippocampus were collected from each mouse. Synapse counts were determined by clearly visible presynaptic membranes, synaptic clefts, and postsynaptic compact bands, and the results were expressed as the number of synapses per unit area. PSD thickness was measured using ImageJ software, with 10-15 synapses randomly selected from each image for measurement, and the average value was taken.

[0113] The results showed that synaptic density was significantly lower in female AD mice with C57 and A / J backgrounds compared to controls, and significantly lower in male AD mice with PIPING backgrounds; while there was no significant difference between AD and Ntg in PWK, LOT, and STUCKY backgrounds. Figure 8 PSD thickness was significantly increased in both male and female AD mice with C57 and PIPING backgrounds compared to controls, while no significant differences were observed in other backgrounds. Figure 9The above results indicate that the decreased synaptic density and increased PSD caused by AD are genetically and sex-dependent.

[0114] S4. Generate a sample expression matrix based on transcriptome sequencing data.

[0115] In one alternative implementation, constructing the sample expression matrix includes quality control of the raw sequencing reads, alignment to a reference genome, gene-level quantification, and batch effect correction.

[0116] In one optional implementation, within 24 hours of the completion of behavioral testing, tissue samples and transcriptome sequencing were performed on mice from all AD-CC populations (including subpopulations such as AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING) and Ntg-CC populations (including subpopulations such as Ntg-C57, Ntg-A / J, Ntg-PWK, Ntg-LOT, Ntg-STUCKY, and Ntg-PIPING). To ensure statistical power, a minimum of 3 animals were used in each experimental group (with the same strain, sex, and genotype), for a total of 72 animals. The purpose of this step was to obtain high-throughput transcriptome data from the hippocampus of Ntg-CC control mice and AD-CC model mice from each strain, providing a data foundation for the subsequent construction of gene co-expression networks.

[0117] 1. RNA extraction and quality testing Hippocampal tissue, flash-frozen in liquid nitrogen, was transferred to RNase-free centrifuge tubes containing TRIzol reagent for homogenization. Total RNA was extracted using the TRIzol-chloroform method, precipitated with isopropanol, washed with 75% ethanol, and then dissolved in RNase-free water. Further purification was performed using a purification kit, with residual genomic DNA removed by on-column DNase I digestion. Quality control standards for RNA samples included: (i) agarose gel electrophoresis analysis of RNA integrity and DNA contamination; (ii) spectrophotometric determination of RNA purity (OD260 / 280 and OD260 / 230 ratios); and (iii) precise determination of RNA integrity values ​​using an Agilent 2100 Bioanalyzer.

[0118] 2. Library construction and high-throughput sequencing mRNA was enriched from total RNA using Oligo(dT) magnetic beads, and fragmented to approximately 300 bp using ion fragmentation. Using the fragmented RNA as a template, first-strand cDNA was synthesized using random primers and reverse transcriptase, followed by second-strand cDNA synthesis. After end repair, A-tailing, and adapter ligation, the enriched library fragments were amplified by PCR, resulting in a final library size of approximately 450 bp. The libraries were quality-checked and effective concentrations were determined using an Agilent 2100 Bioanalyzer. After passing quality checks, the libraries were mixed according to their effective concentrations and target data volume requirements, and high-throughput sequencing was performed on the Illumina platform.

[0119] 3. Sequencing data preprocessing The raw sequencing data underwent quality filtering, specifically including: removing sequencing adapter sequences; eliminating non-AGCT bases at the 5' end of reads; trimming low-quality bases with sequencing quality scores below Q20 at the ends; filtering reads containing ≥10% N bases; and discarding short read fragments shorter than 25 bp. The filtered valid reads were aligned to the mouse reference genome (GRCm39) using HISAT2 software, and the read counts aligned to each gene were used as the raw gene expression levels using HTSeq software. The raw counts were standardized using FPKM (Fragments Per Kilobase Per Million fragments) to ensure comparability of expression levels between different samples and genes. Batch effect correction was applied to the standardized expression matrix to obtain the final expression matrix for subsequent network analysis.

[0120] S5. Perform multiple random samplings on the sample representation matrix, and construct a co-representation network for each sample subset obtained from each random sampling.

[0121] In one alternative implementation, the random sampling is performed 2000 times, with 80% of all samples randomly selected each time to form a subset; the co-expression network is constructed based on the WGCNA method, using the β power of the Pearson correlation coefficient as the weighted adjacency matrix, and the soft threshold β=10.

[0122] S6. Based on the co-expression network, determine the consensus module; the consensus module includes multiple module feature genes.

[0123] In one optional implementation, step S6 includes: Step c1: Based on all co-expression networks, count the number of times each gene pair is sampled in the same sampling subset, and construct a gene co-occurrence indicator matrix; Step c2: Count the number of times gene pairs with common sampling relationships are assigned to the same module, and construct a co-clustering matrix; Step c3: Based on the gene co-occurrence indicator matrix and co-clustering matrix, perform average link hierarchical clustering on all genes, and use dynamic tree cutting algorithm to cluster and group them to obtain multiple consensus modules.

[0124] In one optional implementation, determining the consensus co-expression module includes: counting the number of times each gene pair is co-sampled in a random subset and assigned to the same module, and constructing a consensus matrix; performing hierarchical clustering using 1 minus the consensus matrix as a distance metric, dividing the consensus module using a dynamic tree cutting algorithm, and merging adjacent modules whose feature gene correlation coefficient is greater than 0.85.

[0125] In one alternative implementation, the construction of the consensus-weighted gene co-expression network includes: This step is based on a publicly available consensus clustering analysis workflow (referencing the GitHub project czawora / WGCNA_consensus_clustering; for methodological principles, see the literature doi.org / 10.1023 / A:1023949509487). This method is applied for the first time to transcriptome data (72 cases in total) from a genetically diverse AD mouse model (AD-CC population) and its control samples (Ntg-CC) to construct a stable gene co-expression network. The specific steps are as follows: First, based on the batch-corrected standardized full-sample expression matrix, 2000 random subsamplings with replacement were performed, with each sampling randomly selecting 80% of all samples to form a subsample set. For each subsample set, a weighted co-expression network was independently constructed using the WGCNA R package: the β power of the Pearson correlation coefficient (screened by network topology analysis, with a soft threshold β=10) was used as the weighted adjacency matrix, which was then transformed into a topological overlap matrix. Average link hierarchical clustering was performed based on the distance between 1 and the topological overlap matrix, and co-expression modules were identified using a dynamic tree cutting algorithm, with the minimum module size set to 30 genes.

[0126] Secondly, based on the results of 2000 sub-network constructions, the number of times each gene pair was simultaneously sampled in the same sub-sample set was counted to construct a gene-gene co-occurrence indicator matrix; simultaneously, the number of times each gene pair was assigned to the same module under the condition of co-sampling was counted to construct a co-clustering matrix. The co-clustering matrix was divided element-wise by the co-occurrence indicator matrix to obtain the consensus matrix, whose value range is [0, 1], reflecting the stability of the co-module assignment of each gene pair in multiple random sub-samplings.

[0127] Finally, using 1 minus the consensus matrix as the distance metric between genes, all genes were subjected to average link hierarchical clustering, and the final consensus modules were divided using a dynamic tree slicing algorithm. Adjacent modules with a correlation coefficient greater than 0.85 were merged to reduce module redundancy. The module's characteristic genes—i.e., the first principal component of the gene expression matrix within the module—were used as representative values ​​of the overall expression level of each consensus module. The constructed consensus network and module characteristic genes will provide input data for subsequent steps in the quantitative analysis of disease-induced network reconnection and the testing of heterogeneity between strains.

[0128] Based on batch-effects-corrected genome-wide expression matrices, a total of 72 co-expression modules were identified (module gene count range: 30–215). The module characteristic gene (ME) of each module served as a representative value of the overall expression level of the module. To systematically evaluate the relationship between the ME of each module and each sample, a correlation heatmap between the ME of each module and all samples was generated. Figure 10 The heatmap results visually reveal the transcriptomic heterogeneity patterns among samples. The aforementioned consensus module and ME will provide input data for subsequent steps, including quantitative analysis of disease-induced network reconnection and testing for heterogeneity among strains.

[0129] S7. Using the module feature genes as network nodes, construct the disease topology network corresponding to the disease model sample population and the control topology network corresponding to the normal control sample population for each strain.

[0130] In one optional implementation, step S7 includes: Step d1: Determine the correlation coefficients between the feature genes of each module; Step d2: Select gene pairs whose absolute correlation coefficient is greater than a preset threshold and which meet the preset requirements after multiple tests and corrections as valid edges; Step d3: For each strain, using the module feature genes as network nodes and the correlation coefficient as edge weights, construct disease topology networks and control topology networks based on the selected effective edges.

[0131] In one alternative implementation, the edges of the topology network are valid connections after the Pearson correlation coefficient between the feature genes of the modules is filtered by a double threshold, where the double threshold includes an absolute value of the correlation coefficient |r|>0.6 and an FDR-corrected q value<0.1.

[0132] S8. Based on the diseased topology network and the control topology network, determine the network topology reconnection index of the corresponding strain.

[0133] In one alternative implementation, the network topology reconnection index = (number of lost edges + number of gained edges) / total number of edges in the superimposed network × 100%, where lost edges are edges that exist only in the control topology network but not in the diseased topology network, and gained edges are edges that exist only in the diseased topology network but not in the control topology network.

[0134] In one optional implementation, step S8 includes: Step e1: Overlay the diseased topological network and the control topological network of the same strain to obtain the overlay network; Step e2: Count the lost edges in the comparison topology network and the gained edges in the diseased topology network; Step e3: Divide the sum of the number of lost edges and the number of gained edges by the total number of edges in the superimposed network to obtain the network topology reconnection index.

[0135] In one optional implementation, to quantify the co-expression network architecture and disease-induced remodeling degree of various genetic backgrounds under basal conditions, this step independently constructs weighted co-expression topology networks based on Ntg-CC control samples (36 cases) and AD-CC model samples (36 cases) from each strain. Specifically, the R package igraph is used for network construction and analysis, and the main process is as follows.

[0136] First, the module feature genes calculated in the above steps are used as network nodes, and the Pearson correlation coefficient between module feature genes is used as the basis for edge weights. To preserve statistically significant co-expression relationships, a dual threshold screening strategy is adopted: the absolute value of the correlation coefficient |r| is required to be greater than 0.6, and the q-value after FDR multiple hypothesis testing correction is less than 0.1. Only gene pairs that simultaneously meet the above two conditions are retained as valid edges in the topological network.

[0137] Secondly, based on the filtered edge set, weighted undirected topological networks for each strain in the control state (Ntg-CC) and disease state (AD-CC) are constructed using the igraph package. For each network, the following topological characteristic indices are calculated using the igraph package: Number of sides ( Figure 11 A): The total number of edges contained in the network; Network density ( Figure 11 B): The ratio of the actual number of edges to the maximum possible number of edges; Average node degree ( Figure 11 C): The average degree of all nodes (i.e., the number of edges connected to that node); Modular coefficient ( Figure 11 D): Calculate and evaluate the degree to which the network is divided into modules based on algorithms such as cluster_louvain.

[0138] Subsequently, the control topology network and the AD topology network of the same strain were superimposed and compared. Based on the presence of edges in the two states, they were divided into the following three categories: Common edges: Edges that exist in both the control network and the AD network; Lost edges: Edges that exist only in the control network but disappear in the AD network; Gained edges: Edges that exist only in the AD network but not in the control network.

[0139] The network topology reconnection index is defined as the percentage of the sum of lost and gained edges to the total number of edges in the superimposed network (i.e., the size of the union of the edge sets of the two networks), used to quantify the degree of disease-induced network reprogramming. The calculation formula is as follows: Network topology reconnection index = (Number of lost edges + Number of gained edges) / Total number of edges in the superimposed network × 100% Finally, to examine the cross-strain association between network robustness and disease resistance, Pearson correlation analysis was performed on the calculated network reconnection index of each strain and the average resistance index of that strain (see step S3). The results of the correlation analysis will be used to assess whether there is a systemic association between the degree of network topology reconnection and cognitive function resistance related to Alzheimer's disease.

[0140] Based on the above method, all samples (Ntg and AD) were included, and control and disease networks were constructed for each strain. The network topology characteristics (number of edges, network density, average node degree, modularity coefficient) for each strain are shown below. Figure 11 A, B, C, D. The network topology reconnection index for each strain is calculated by comparing edge sets. Figure 12 A and B were compared with the average disease resistance index of the strains using Pearson correlation analysis. The results showed that the network topology reconnection index was significantly negatively correlated with the disease resistance index. Figure 12 C), meaning that genetic backgrounds with higher resistance indices exhibit smaller overall structural changes in their networks during disease states, and their regulatory architecture maintains higher stability under the impact of AD pathology; while strains with lower resistance indices show a greater range of edge set recombination. These results indicate a system-level negative correlation between the degree of network topology reconnection and AD cognitive resistance.

[0141] S9. Perform correlation analysis between the network topology reconnection index of each strain and the average resistance index of each strain, and select key modules that are significantly correlated with the resistance index from the consensus modules.

[0142] In one optional implementation, step S9 includes: Step a1: For each consensus module, determine the module reconnection score based on the neighbor set in the disease topology network and control topology network corresponding to the consensus module in different strains.

[0143] Step a2: For each consensus module, determine the correlation between the module reconnection score of the consensus module and the average resistance index of the corresponding strain, and determine the resistance correlation coefficient and robustness index corresponding to the consensus module. Step a3: Determine the comprehensive score of each consensus module based on the module reconnection score, resistance correlation coefficient, and robustness index; Step a4: Based on the ranking of the comprehensive scores, key modules that are significantly related to the resistance index are selected.

[0144] In one optional implementation, the correlation analysis is Pearson correlation analysis; the key module is the Top 5 consensus module, which identifies the network topology reconnection index and the resistance index with the largest absolute value of their correlation coefficients and which are statistically significant.

[0145] In one alternative implementation, for each consensus module, the degree of disease-induced connection remodeling is quantified in each genetic context. Specifically, based on the strain control network (Ntg-CC) and disease network (AD-CC) constructed in the above steps, the neighbor set of each module node in both states is extracted. For module m in strain b: Compare the state neighbor set ( ): refers to the set of nodes in the control network that are connected by an edge to any node in module m (satisfying the threshold conditions of |r|>0.6 and FDR<0.1 in the above steps).

[0146] Disease state neighbor set ( ): refers to the set of nodes in the disease network that are connected to any node in module m by an edge.

[0147] Based on the two sets mentioned above, the neighbor Jaccard similarity is first calculated:

[0148] The value of this indicator ranges from [0, 1]. The larger the value, the more stable the neighbor composition of the module is before and after the disease.

[0149] Simultaneously calculate the degree change weight ( This is used to quantify the degree of change in the number of node connections within a module.

[0150] in, and These represent the average node degree (i.e., the average number of edges connected to nodes within this module) of all nodes in the control network and the disease network, respectively. The value ranges from [0, 1]. The larger the value, the more drastic the change in the overall connectivity of the module.

[0151] Combining the above two indicators, we obtain the single-line reconnection score ( ):

[0152] The value of this indicator ranges from [0, 1]. The larger the value, the more severe the disease-induced network reconnection in strain b. If a module is completely absent in the control network or the disease network (i.e., the nodes in the module have no edge connections in a certain state), the reconnection score is directly recorded as the maximum value of 1.

[0153] Finally, for all strains that successfully constructed control and disease networks, the cross-strain mean was calculated as the final reconnection score for that module:

[0154] In an optional implementation, the reconnection score of each strain is calculated using the average resistance index of AD samples from each genetic background (see step S3) as the phenotypic variable. The cross-strain Pearson correlation coefficient between the average resistance index of the strain and the strain's average resistance index ( This coefficient reflects the direction and strength of the association between the degree of module reconnection and disease resistance: a positive value indicates that the more intense the module reconnection, the stronger the resistance (i.e., the module is related to the protective mechanism); a negative value indicates that module reconnection is negatively associated with resistance.

[0155] In one alternative implementation, to assess the cross-strain robustness of the aforementioned association (correlation between reconnection score and resistance index), leave-one-out cross-validation is employed: each time a genetic background is excluded, the correlation coefficient is recalculated based on the remaining strains. For a dataset containing 6 strains, a total of 6 rounds of estimates were obtained.

[0156] Based on these six rounds of estimates, the following robustness indicators were calculated: mean ( ): The average correlation coefficient estimated by the leave-one-out method; Standard deviation ( ): Evaluate the degree of dispersion of the estimated values; coefficient of variation ( : A standardized indicator that measures the degree to which a correlation depends on a specific strain. The smaller the value, the less dependent the correlation is on the individual contribution of a particular strain, meaning it is more robust.

[0157] In one alternative implementation, the three dimensions mentioned above are integrated to construct a comprehensive scoring system: Step 1: Standardization Average reconnection score for each module ( ) and resistance correlation coefficient ( Perform min-max standardization separately, normalizing to the [0, 1] interval:

[0158]

[0159] Step 2: Calculate the overall score The composite score is defined as the product of the two:

[0160] For modules with negative resistance correlation coefficients (i.e., negatively correlated with resistance), their overall score is multiplied by a penalty coefficient of 0.1 to reduce their priority.

[0161] Step 3: Apply robust weights Robustness weights are defined as a function of the coefficient of variation based on the leave-one-out method: The design logic for this weight is: The larger the correlation (the more dependent the correlation is on a specific strain), the smaller the weight. When the weight is 0 (fully robust), the weight is 1. As the weight approaches infinity, the weight approaches 0.

[0162] The final score for the module is calculated as follows:

[0163] Step 4: Filter candidate modules All consensus modules are sorted from highest to lowest based on their final scores, and the top 5 modules are selected as candidate key modules for further in-depth analysis.

[0164] Using the above screening strategy, the top 5 candidate modules with the most severe reconnection and positive correlation with disease resistance were identified from all modules (Table 1). Figure 13 ), which is the critical reconnection module.

[0165] Table 1 Feature values ​​of the multi-dimensional comprehensive score screening Top 5 module

[0166] In an optional implementation, module-phenotypic association analysis and inter-strain heterogeneity testing are also included: This step aims to evaluate the association between candidate key modules (the Top 5 modules selected in the above steps) and Alzheimer's disease-related phenotypes, and to examine whether these associations differ significantly across different genetic backgrounds (i.e., genetic background dependent).

[0167] 1. Intra-strain module-phenotypic association analysis Using the module characteristic genes calculated in step S6 as independent variables and phenotypic characteristics (including the percentage of rigidity time in situational fear memory, Aβ plaque burden, etc.) as dependent variables, only AD-CC model samples were included. Pearson correlation coefficients were calculated stratified according to six genetic backgrounds (C57, A / J, PWK, LOT, STUCKY, PIPING). The p-values ​​of all module-phenotype combinations within each strain were corrected using the Benjamini-Hochberg method, with FDR < 0.05 as the significance threshold.

[0168] 2. Heterogeneity testing between strains For each significant module-phenotype combination, pairwise Fisher Z-transform heterogeneity tests were performed between strains. The correlation coefficients of each strain were then calculated. Transform into Calculate the ratio of the difference in Z-values ​​between the two strains to the standard error. .like (correspond If the module-phenotype association is found to be significantly heterogeneous between the two strains, then it is considered that the module-phenotype association is significantly heterogeneous between the two strains.

[0169] 3. Determination of genetic background-dependent associations For each module-phenotype combination, if at least one of the 15 strain pairs exhibits significant heterogeneity ( If the association is genetically dependent, it will be marked as a genetic background-dependent association and included in subsequent in-depth analysis.

[0170] Following the above method, stratified correlation analysis was performed on 5 candidate modules (honeydew1, darkorange2, navajowhite2, plum2, cyan) and 9 phenotypes (episodic fear memory, cue fear memory, Aβ load, synaptic density, PSD thickness, total dendritic crossovers, percentage of Iba1 positive area, number of microglia around each plaque diameter, and number of activated microglia around each plaque diameter). The results showed that the module-phenotype associations exhibited a significant genetic background-dependent pattern. Figure 14Among the five candidate modules, the plum2 module stands out as the most valuable candidate due to its broadest phenotypic association spectrum, most concentrated cross-contextual heterogeneity, and a positive correlation pattern with cognitive function and synaptic structure indicators consistent with high-resistance backgrounds. Its diametrically opposed phenotypic association directions between resistant and susceptible backgrounds suggest that the functional state of this module may be a key node determining the trajectory of disease phenotypes. Based on this, future research will focus on the plum2 module, conducting in-depth analysis of its core regulatory genes and upstream transcription factor networks to elucidate the specific molecular mechanisms by which genetic background mediates disease resistance.

[0171] S10. Extract the core genes and upstream transcription factors that regulate the core genes from the key modules. Based on the regulatory relationship between the core genes and upstream transcription factors, construct an expression interaction network of genes related to Alzheimer's disease resistance.

[0172] In one alternative implementation, the core gene is obtained through protein-protein interaction network analysis and node centrality sorting; upstream transcription factors are obtained through transcription factor binding site enrichment analysis of the TRANSFAC database and / or JASPAR database.

[0173] In one alternative implementation, the core gene includes at least one of SSR1, BNIP3L, SEPT7, SELT, ZBTB6, CCNC, GMPS, NFIB, and TRAPPC13; the upstream transcription factor includes NR2F6 and / or SOX1.

[0174] In an alternative implementation, the method further includes a step of cross-species validation of core genes and / or upstream transcription factors in a population database.

[0175] This embodiment also provides a biomarker screening method based on the expression interaction network of Alzheimer's disease resistance-related genes, which can be used on servers, terminals, and mobile terminals, such as mobile phones and tablets. The method includes the following steps: S11. Construct an Alzheimer's disease resistance-related gene expression interaction network by performing the construction method in the previous embodiment; S12. Extract key genes and upstream transcription factors that regulate key genes from the expression interaction network, and use key genes and / or upstream transcription factors as candidate biomarkers. S13. Validate the association between candidate biomarkers and cognitive resistance in a population cohort, and identify biomarkers that demonstrate significant validation as biomarkers related to Alzheimer's disease resistance.

[0176] In one optional implementation, step S12 includes: Gene feature assessment was performed on the consensus module in the gene expression interaction network related to Alzheimer's disease resistance to determine the module membership degree and gene significance of the characteristic genes in each module; the consensus module includes multiple module characteristic genes; Key genes are obtained by screening characteristic genes of a module based on intra-module gene connectivity, module membership, and gene significance. This screening includes: selecting genes whose module membership is greater than a first preset percentile of the membership of all genes within the corresponding consensus module and whose module membership reaches a first preset significance level; selecting genes whose intra-module gene connectivity is greater than a second preset percentile of the connectivity of all genes within the consensus module; selecting genes whose gene significance reaches a second preset significance level after verification and correction; and identifying genes that simultaneously meet all three screening criteria as key genes. Pre-defined sequences upstream and downstream of the transcription start sites of each key gene were extracted as promoter regions. Based on the position weight matrix of transcription factors in a preset transcription factor database, the promoter region is scanned according to a preset score threshold to obtain potential transcription factor binding sites; Using the promoter sequences of all genes within the same consensus module as the background set, enrichment tests were conducted based on the distribution of potential transcription factor binding sites to screen upstream transcription factors that meet the preset correction significance criteria and enrichment intensity requirements. Upstream transcription factors and / or key genes are used together as candidate biomarkers.

[0177] In an optional implementation, step S12 includes an overall feature evaluation of the key modules: The overall characteristics of the 84 genes within the key module (plum2 module) obtained from the above screening steps were evaluated, and the following two indicators were calculated: Module membership (kME value): The Pearson correlation coefficient between the expression profile of each gene in a module and the characteristic genes of that module (the first principal component of the expression matrix within the module), reflecting the connection importance of genes within the module.

[0178] Gene significance (GS value): The Pearson correlation coefficient between the expression level of each gene in the module and the average resistance index of the strain (see step S3), reflecting the strength of the association between the gene and the disease resistance phenotype.

[0179] For each key module, the proportion of genes that simultaneously meet the following two conditions is counted: (i) kME value > 0.8; (ii) GS value reaches a significant level after FDR correction (FDR < 0.05). If this proportion exceeds a preset threshold (e.g., 80%), the module is determined to be a highly compact functional synergistic unit, and a more stringent multidimensional screening strategy than traditional standards is required to identify the core regulatory hub.

[0180] In an optional implementation, step S12 further includes: In a highly compact modular context, this application employs a multidimensional screening strategy based on percentile thresholds to identify key genes. Specific screening conditions include: (i) Module membership condition: The kME value is greater than the 90th percentile of the kME values ​​of all genes in the module, and the kME value is statistically significant (P value is less than the preset significance level, which is set to 0.05 below). (ii) Intra-module connectivity condition: The connectivity index of a gene within a module is greater than the 90th percentile of the connectivity values ​​of all genes within that module; (iii) Phenotypic correlation conditions: The gene expression level and resistance index are statistically significant after multiple test correction (FDR is less than the preset threshold, which is set to 0.05 below).

[0181] Genes that simultaneously meet all the above conditions are identified as key genes for this module. Using a uniform percentile threshold ensures consistency of standards across all screening dimensions, effectively avoiding screening bias caused by differences in thresholds.

[0182] Based on the above criteria, nine key genes were identified in the plum2 module: Ssr1, Bnip3l, Septin7, Selenot, Zbtb6, Ccnc, Gmps, Trappc13, and Nfib (Table 2). The kME values ​​of these nine genes all ranged from 0.970 to 0.980, and their intra-module connectivity ranged from 72.7 to 73.5, all falling within the highest quantile of gene distribution within the module. This indicates that they occupy a core position in the module's topology and may act as key drivers regulating the co-expression pattern of the entire module and its association with disease resistance levels.

[0183] Table 2. Eigenvalues ​​of key genes in the plum2 module

[0184] In an optional implementation, step S12 further includes: predicting upstream transcription factors for the key gene set identified in the above steps. Specifically, the method involves extracting the region from 2000 bp upstream to 200 bp downstream of the transcription start site of each key gene as the promoter region. Using the position weight matrix of known transcription factors in the JASPAR database (2022 version), potential transcription factor binding sites in the promoter region are scanned using an 80% relative contour score threshold. Using the promoter sequences of all genes within this key module as the background set, Fisher's exact test is used to identify significantly enriched transcription factors in the promoter regions of key genes. The screening criteria are an FDR-corrected P-value less than 0.05 and an enrichment intensity (Odds ratio, OR) greater than 1. Based on the prediction results, a regulatory network between transcription factors and key genes can be constructed, providing candidate targets for subsequent cross-species validation.

[0185] Following the above method, two transcription factors with an odds ratio (OR) exceeding 6.6 were screened (Table 3). Sox1 showed the most significant enrichment, with its binding site present in the promoters of 48 Hub gene transcripts, compared to only 365 in background genes. Considering the far fewer Hub genes compared to background genes, this enrichment is extremely significant. Nr2f6 also showed remarkable enrichment. Both transcription factors were positively correlated with the resistance index (Sox1: r = 0.366; Nr2f6: r = 0.565), suggesting that their high activity is associated with disease resistance. Figure 15 ).

[0186] Table 3. Transcription factor enrichment results and eigenvalues

[0187] In one alternative implementation, after step S13, the method further includes: S14. Obtain transcriptome and clinicopathological sample data of brain tissue from the population, and construct the population disease resistance index based on the preset regression model. S15. Screen key genes and / or upstream transcription factors for human homologous genes as candidate genes. S16. Conduct multi-dimensional population association analysis on candidate genes and verify the cross-species conservation of the regulatory relationship between key genes and upstream transcription factors; multi-dimensional population association analysis includes the association between candidate genes and population disease resistance index, the association between candidate genes and Alzheimer's disease pathology and neuroinflammatory traits, and the predictive evaluation of candidate genes on dementia diagnosis results. S17. Identify candidate genes and / or upstream transcription factors that have been validated through multidimensional population studies as human-related biomarkers for resistance to Alzheimer's disease.

[0188] In one alternative implementation, population data acquisition and disease resistance index construction include: 1. Data Source Population validation data were obtained from the Aging, Dementia and Traumatic Brain Injury Study (http: / / aging.brain-map.org / ) database published by the Allen Institute for Brain Science, while gene expression data were simultaneously hosted in the NCBI Gene Expression Database (GEO accession number: GSE104687). This study screened 107 donors for samples with complete hippocampal RNA-seq data, dementia diagnostic records, and Aβ pathological quantitative data, ultimately including 83 individuals (46 males, 37 females; 40 in the dementia group, 43 in the non-dementia group). Transcriptome data covered four brain regions: the temporal cortex, parietal cortex, white matter, and hippocampus, and were publicly released in FPKM format; hippocampal data were used for subsequent analysis in this study.

[0189] 2. Construction of Cognitive Resistance Score This study uses a logistic regression residual-based method to calculate the population disease resistance index. Its core logic is consistent with the construction idea of ​​the mouse model disease resistance index in step S3: under the premise of controlling the degree of neuropathological damage, the deviation between the individual's actual cognitive state and the model's predicted cognitive state is quantified to reflect the individual's cognitive reserve capacity under the same pathological load.

[0190] The specific steps are as follows: Using clinical dementia status (act_demented, 0 = non-dementia, 1 = dementia) as the dependent variable, and hippocampal Aβ pathological percentage (ihc_a_beta_percent), age, and sex as independent variables, a binary logistic regression model is constructed: logit(P(dementia)) = β0 + β1 (Aβ pathological percentage) + β2 (age) + β3 (gender) After model fitting, the Pearson residuals for each sample were extracted and used as the individual's disease resistance score. Positive residuals indicate that the individual's actual cognitive state is better than predicted by the pathological burden, suggesting strong disease resistance; negative residuals indicate that the degree of cognitive impairment exceeds the expected level of the pathological burden, suggesting lower disease resistance. The pathological indicator selected was the percentage of hippocampal Aβ deposition area detected by immunohistochemistry (IHC) to achieve methodological alignment with the pathological indicator (Aβ deposition) in the mouse model. Figure 16 ).

[0191] In one alternative implementation, multi-dimensional validation of candidate genes includes: The following analyses were performed on the human homologs of the key genes and transcription factors identified in the above steps: (i) correlation analysis with the population disease resistance index, and assessment of association robustness in subgroups stratified by sex and pathological load; (ii) association analysis with multimodal pathological traits such as Aβ, Tau, and neuroinflammation; and (iii) logistic regression prognostic assessment with dementia diagnosis as the dependent variable.

[0192] 1. Extraction and preprocessing of candidate gene expression data Based on the screening results of the above steps, 11 candidate genes were selected as population validation subjects: 9 human homologous core genes (SSR1, BNIP3L, SEPT7, SELT, ZBTB6, CCNC, GMPS, NFIB, TRAPPC13) and 2 upstream transcription factors (NR2F6, SOX1). The hippocampal FPKM expression values ​​of the above genes were directly extracted from the normalized expression matrix of the Allen Brain Aging / TBI Study database and matched and merged with the clinical and pathological information of 83 samples.

[0193] To reduce noise interference from low-expression genes and approximate a normal data distribution, the FPKM values ​​of all genes were standardized using log2 transformation before analysis. After transformation, descriptive statistics (mean, standard deviation, distribution test) were performed on the expression levels of each gene, and the expression distribution of each gene between the dementia group and the non-dementia group was visually examined to confirm data quality. All subsequent correlation analyses, regression analyses, and co-expression network analyses were based on the standardized expression values.

[0194] Based on the research hypothesis, all analyses were performed separately in the full sample and in sex-stratified subgroups (males: n = 46; females: n = 37) to capture potential patterns of sex dimorphism regulation.

[0195] 2. Correlation analysis between candidate genes and cognitive resistance index To assess the strength and direction of the association between the expression levels of 11 candidate genes and the population disease resistance index, this study employed Spearman rank correlation analysis as the primary statistical method. Spearman correlation does not assume normality of data distribution and is robust to outliers, making it suitable for the analysis context of non-normal distribution of population cognitive resistance scores in this study. Correlation analyses were performed independently in the full sample, male subgroup, and female subgroup to examine the sex specificity of the gene-population disease resistance score association.

[0196] To further eliminate the confounding effect of differences in AD pathological burden on the association between the gene and population disease resistance index, this study introduced a pathological burden stratification strategy in the correlation analysis: using the median percentage of Aβ immunohistochemically positive area within each group as the cutoff point, the entire sample and each sex subgroup were divided into high pathological burden subgroups and low pathological burden subgroups, and the above correlation analysis was repeated within each stratified subgroup. This strategy resulted in six analytical subgroups (high / low pathology in the entire sample, high / low pathology in males, and high / low pathology in females), systematically evaluating the robustness of the gene-population disease resistance index association under different pathological burden backgrounds.

[0197] In addition, to test whether there is a significant sex-related interaction effect in the association between genes and population disease resistance index, this study constructed a linear regression model for each gene that included gene expression level, sex, and the interaction term between the two (gene expression level × sex). The regression coefficient of the interaction term and its significance (P value) were used as the basis for testing the sex-related moderating effect.

[0198] All correlation analysis results are reported as Spearman correlation coefficient (r) and two-tailed p-values. The statistical significance threshold was set at P < 0.05, and marginal significance trends (0.05 ≤ P < 0.10) were noted for reference. Given that this study was an exploratory analysis and the sample size of each subgroup was limited, no correction was made for multiple comparisons. The interpretation of results was mainly based on the consistency of effect size direction with cross-analysis strategy.

[0199] The specific implementation results are as follows: Using the median percentage of Aβ pathological events as the cutoff point, 83 participants were divided into a high pathological burden group (n=42) and a low pathological burden group (n=41). Within the high pathological burden group, gene expression differences between individuals with and without dementia were analyzed. In the full sample analysis, the expression direction of 5 out of 9 core genes was consistent with expectations (higher in the non-dementia group than in the dementia group), but none reached statistical significance (Table 4). Figure 17 A). After sex stratification, the direction of 8 genes (88.9%) in women with high pathological burden was consistent with expectations (mean r = 0.172); in men with high pathological burden, 3 genes (BNIP3L, SELT, GMPS) showed significant negative correlations (P < 0.05, |r| > 0.44), consistent with the direction in the mouse model (Table 5). Gene-population disease resistance score correlation analysis further validated the above findings: BNIP3L (r = -0.570, P = 0.006), SELT (r = -0.476, P = 0.025), and GMPS (r = -0.446, P = 0.038) all showed significant negative correlations in men with high pathological burden, while the correlation disappeared in men with low pathological burden (|r| < 0.11) (Table 5). Figure 17B). In summary, BNIP3L, SELT, and GMPS showed robust negative correlations across analysis strategies in men with high pathological burden, making them the most well-supported candidate genes in this cohort.

[0200] Population validation of upstream transcription factors NR2F6 and SOX1 showed that their effect sizes were generally weaker than those of core genes. In full-sample correlation analysis, the two transcription factors showed the expected positive correlation in women (NR2F6: r=0.134; SOX1: r=0.067) and a negative correlation in men, neither of which reached statistical significance. After stratification for high pathological burden, the positive correlation was maintained in women (mean r=0.103), while the correlation approached zero in men (mean r=0.015). These results indicate that the evidence for conservation at the transcription factor level is insufficient in this dataset and requires further validation in larger cohorts.

[0201] Table 4. Differences in the expression of core genes and transcription factors between the dementia group and the non-dementia group within the high pathological burden group.

[0202] Note: The expected direction for the Hub gene is "no dementia > dementia" (low expression in resistant individuals); the expected direction for the TF gene is "no dementia > dementia" (high expression in resistant individuals). This indicates that the transcription factor is expected to be positively correlated (high expression without dementia), and is marked "No" here. "This means that the direction of expression is as expected, but the statistics are not significant."

[0203] Table 5. Correlation analysis of key genes and transcription factors with population disease resistance index by pathological burden and sex grouping.

[0204] Note:†, P <0.1; , P <0.05; , P <0.01. A negative correlation (r<0) indicates that high expression corresponds to low resistance, which is the expected direction for key genes; a positive correlation (r>0) is the expected direction for transcription factors.

[0205] 3. Association analysis between candidate genes and AD pathology and neuroinflammatory traits To systematically evaluate the potential functional roles of 11 candidate genes in AD pathological progression and neuroinflammatory regulation, this study integrated multimodal pathological quantitative data from the Allen Brain database into 22 analytical traits, covering four dimensions: Aβ pathological characteristics (4): percentage of hippocampal Aβ deposition area, percentage of Aβ40 area, percentage of Aβ42 area, and Aβ42 / 40 ratio as detected by IHC method; Tau pathological characteristics (5): percentage of hippocampal phosphorylated Tau protein (AT8) area and percentage of total Tau protein area (Tau2) as detected by IHC method; hippocampal phosphorylated Tau protein content, total Tau protein content, and ptau / totaltau ratio as detected by ELISA. Glial cell markers (2): hippocampal GFAP (astrocytic marker) area percentage, Iba1 (microglia marker) area percentage; Neuroinflammatory cytokines (11): CCL2, CCL3, CCL5, IL-6, IL-7, IL-10, TNF-α, BDNF, IFN-γ, IL-1β, VEGF (derived from Luminex protein quantification data associated with the database).

[0206] The association between the expression levels of each candidate gene and the 22 pathological traits was analyzed using Spearman rank correlation analysis, performed in the full sample and in sex-stratified subgroups (male and female) to capture potential sex-specific pathological association patterns. The correlation coefficient (r) and two-tailed p-value were used as the primary reporting indicators, with a significance threshold of p < 0.05. Marginal significance (0.05 ≤ p < 0.10) was indicated by "†".

[0207] Given the large number of traits analyzed (11 genes × 22 traits × 3 subgroups), the interpretation of the results focused on significant cross-sex associations and sex-specific associations with large effect sizes (|r|>0.3), and the screening and reporting were based on the dual criteria of biological significance and statistical significance.

[0208] The specific implementation results are as follows: In the female group, key genes were generally negatively correlated with Aβ and Tau core pathological markers. Figure 18 Among them, ZBTB6 showed the most significant pathological association, with its expression being significantly negatively correlated with Tau deposition, Aβ deposition, and Aβ42 levels, spanning the two core pathological axes of Aβ and Tau. Furthermore, several key genes exhibited complex association patterns with chemokines: CCNC was simultaneously positively correlated with CCL2 and negatively correlated with IL-6 and CCL5, suggesting that it may play a sophisticated immune regulatory role in women.

[0209] In the male group, the association patterns of key genes showed significant differences. Figure 18GMPS was negatively correlated with multiple inflammatory markers (CCL5, CCL3, IL-10, BDNF), and NFIB was also negatively correlated with IL-10 and BDNF. In contrast to inflammatory regulation, ZBTB6, SELT, and TRAPPC13 were positively correlated with GFAP in males. The association of SELT with GFAP was completely reversed between sexes (negative in females, positive in males), providing the clearest example of sex bimorphism.

[0210] Transcription factors NR2F6 and SOX1 exhibited consistent cross-sex associations in Aβ metabolism: both were positively correlated with the Aβ42 / Aβ40 ratio in both sexes. Furthermore, NR2F6 was negatively correlated with several pro-inflammatory chemokines (CCL2, IL-7, TNF-α) in males, suggesting that it may play a dual regulatory role in Aβ metabolism and neuroinflammation in men.

[0211] Based on the above results, the following systematic patterns can be summarized: (1) High expression of key genes in women is associated with the reduction of Aβ / Tau pathology and the increase of chemokine levels; (2) High expression of the same group of genes in men is associated with the reduction of inflammatory factors and the enhancement of glial cell activation, and some genes (GMPS, CCNC, SELT) show opposite regulatory directions between the sexes; (3) NR2F6 shows a cross-sex consistent association in Aβ metabolism, and achieves dual regulation of Aβ metabolism and neuroinflammation in men, making it the primary priority target for subsequent functional verification.

[0212] 4. Logistic regression prognostic assessment of candidate genes To assess the independent predictive value of the expression levels of each candidate gene for dementia diagnosis, this study constructed a series of binary logistic regression models with dementia diagnosis (act_demented: 1 = dementia, 0 = non-dementia) as the dependent variable.

[0213] Age (at death), sex, and Aβ pathological burden (ihc_a_beta_percent) were included as covariates in the base model to control for confounding effects. Based on this, the log2 transformed expression values ​​of 11 candidate genes were individually introduced as additional predictive variables into the model, forming 11 single-gene extended models. The contribution of each gene to the model fit was evaluated by comparing the difference in Akaike Information Criterion (AIC) (ΔAIC) between each extended model and the base model: ΔAIC > 2 was considered a substantial improvement, and ΔAIC between 1 and 2 was considered a marginal improvement trend.

[0214] The regression coefficients, odds ratios (OR), and 95% confidence intervals of each candidate gene were reported as the main effect size indicators. An OR < 1 indicates that high expression of the gene is associated with a reduced risk of dementia, which is consistent with the direction of resistance.

[0215] The logistic regression analysis described above was performed in the full sample and in sex-stratified subgroups (males and females) to examine the sex specificity of the prognostic predictive effects of each gene. Given the limited sample size in each subgroup (male n = 46, female n = 37), the model estimation exhibited some instability. The results were interpreted primarily based on cross-analysis consistency between the OR direction and the improvement in AIC, supplemented by p-values ​​as a reference.

[0216] The specific implementation results are as follows: In the full sample analysis, none of the 11 candidate genes reached statistical significance. The AIC values ​​of all genes increased after being added to the model (ΔAIC<0), suggesting that the independent predictive power of single genes is limited. Figure 19 A).

[0217] After gender stratification, in the female subgroup, CCNC (OR=0.47, P=0.076) and GMPS (OR=0.52, P=0.094) were marginally significantly associated with a reduced risk of dementia, and their AIC values ​​decreased after being added to the model, suggesting that they had a marginal improvement effect on model fit, consistent with the finding that key genes in women are negatively associated with the pathological burden of AD. Figure 19 B). In the male subgroup, BNIP3L (OR=2.31, P=0.087) and GMPS (OR=1.85, P=0.142) showed an OR>1 trend, the opposite direction to that in the female subgroup, further confirming the functional directional differences between the sexes. Figure 19 B).

[0218] The above results indicate that, limited by the current sample size, the statistical power of single-gene logistic regression is insufficient to reliably capture independent prognostic effects. The marginally significant signals of CCNC and GMPS in women provide preliminary evidence of prognostic relevance, which requires further validation in larger independent cohorts.

[0219] In one alternative implementation, cross-species conservation verification of the regulatory mechanism includes: To verify the conservation of the transcription factor-key gene regulatory axis identified in the mouse model at the human transcriptome level, this study constructed pairwise Spearman correlation matrices (11 × 11) for the entire sample and sex-stratified subgroups (male and female) based on the log2 transformed expression values ​​of 11 candidate genes from 83 samples. The calculation of the correlation matrices and significance testing (two-tailed P-value) were performed in the R environment. The results were visualized as a correlation coefficient heatmap, with rows and columns arranged by gene category to visually represent the direction and intensity of co-expression between transcription factors and Hub genes.

[0220] The implementation results are as follows: NR2F6 and SEPT7 showed a significant negative correlation in the entire sample, in both males and females, making it the most robust transcription factor-key gene repressive link in this study. Figure 20 NR2F6 showed a consistent negative correlation with SSR1 and TRAPPC13 in both sexes, and a significant negative correlation with BNIP3L in males, showing a marginally significant trend across the entire sample. The negative correlation between these four key genes (SEPT7, SSR1, TRAPPC13, and BNIP3L) and NR2F6 is entirely consistent with the prediction of "high expression of transcription factors → low expression of key genes" in the mouse model, providing direct evidence for the cross-species conservation of the NR2F6-key gene repressive axis. Notably, NR2F6 showed a strong positive correlation with NFIB in females, contradicting the repressive hypothesis, suggesting that this regulatory relationship may have a different pattern in females.

[0221] The significant negative correlation between SOX1 and key genes was almost entirely concentrated in the female subgroup. Figure 20 In women, SOX1 showed significant negative correlations with BNIP3L, CCNC, and TRAPPC13, and a trend toward negative correlations with SEPT7, SELT, and ZBTB6; while in men, none of these associations were significant, except for a positive correlation between SOX1 and GMPS. This pattern suggests that the inhibitory regulatory relationships of SOX1 on multiple key genes are preserved in women, but essentially disappear in men.

[0222] In summary, among the 18 pairs of transcription factor-key gene combinations, 5 pairs showed significant negative correlations across the entire sample, 4 pairs in males and 7 pairs in females. The overall proportion of negative correlations is highly consistent with the prediction of "transcription factor repression of key genes." These results validate the population-related correlations of the plum2 module regulatory axis at the transcriptome level, providing a new perspective for understanding the molecular basis of AD resistance achieved by the sexes through different regulatory network topologies.

[0223] In an optional implementation, the genes and / or transcription factors validated in the above steps are identified as candidate biomarkers related to Alzheimer's disease resistance. Based on the aforementioned multi-dimensional population validation results, the following molecules are identified as candidate biomarkers related to AD resistance: The core gene layer (3 genes): BNIP3L, SELT, and GMPS. These three genes were significantly negatively correlated with cognitive resistance index in men with high pathological burden (r = -0.570, -0.476, and -0.446, respectively, P < 0.05), and the correlation disappeared in men with low pathological burden, exhibiting a pathological burden-dependent resistance-related expression pattern. Among them, GMPS showed a marginally significant association with a reduced risk of dementia in women (OR = 0.52, P = 0.094), and CCNC (OR = 0.47, P = 0.076) was included as a female candidate supplement.

[0224] Transcription factor layer (1): NR2F6. This transcription factor exhibits a consistent positive correlation across sexes in Aβ metabolism (related to the Aβ42 / Aβ40 ratio), and in males, it is simultaneously negatively correlated with multiple pro-inflammatory chemokines (CCL2, IL-7, TNF-α), achieving dual regulation of Aβ metabolism and neuroinflammation. Its inhibitory co-expression linkages with downstream SEPT7, SSR1, TRAPPC13, and BNIP3L have been validated in both sexes, making it a primary priority target for subsequent functional validation. Evidence of female-specific inhibitory regulation of SOX1 (significantly negatively correlated with BNIP3L, CCNC, and TRAPPC13) also supports its potential as a female candidate biomarker.

[0225] In summary, four molecules—BNIP3L, SELT, GMPS, and NR2F6—were identified as candidate biomarkers for AD resistance, validated across multiple population dimensions. CCNC and SOX1 were identified as supplementary biomarkers with gender-specific validation evidence. These biomarkers can be used, but are not limited to: (1) developing gene testing kits or microarrays for assessing individual AD resistance; (2) serving as candidate molecular targets for screening drugs that promote cognitive resistance to AD; (3) constructing predictive models for AD resistance; and (4) serving as independent indicators for AD prognosis assessment.

[0226] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for constructing an expression interaction network of genes related to Alzheimer's disease resistance, characterized in that, The construction method includes the following steps: S1. Construct a genetic diversity-Alzheimer's disease animal model system, which includes multiple disease model sample groups with different genetic backgrounds, and normal control sample groups with the same background as the disease model sample groups; S2. Determine the cognitive function test data, pathological burden test data, and transcriptome sequencing data of each sample in the normal control sample group and the disease model sample group; S3. Based on the cognitive function test data and the pathological burden test data, determine the average resistance index of each strain; S4. Based on the transcriptome sequencing data, generate a sample expression matrix; S5. Perform multiple random samplings on the sample expression matrix, and construct a co-expression network for each sample subset obtained from each random sampling. S6. Based on the co-expression network, determine the consensus module; the consensus module includes multiple module feature genes; S7. Using the module feature genes as network nodes, construct the disease topology network corresponding to the disease model sample population and the control topology network corresponding to the normal control sample population for each strain. S8. Based on the diseased topology network and the control topology network, determine the network topology reconnection index of the corresponding strain; S9. Perform correlation analysis between the network topology reconnection index of each strain and the average resistance index of each strain, and select key modules that are significantly related to the resistance index from the consensus modules. S10. Extract the core genes and upstream transcription factors that regulate the core genes from the key modules, and construct an Alzheimer's disease resistance-related gene expression interaction network based on the regulatory relationship between the core genes and the upstream transcription factors.

2. The construction method according to claim 1, characterized in that, Step S9 includes: For each consensus module, the module reconnection score is determined based on the neighbor set of the consensus module in the disease topology network and the control topology network corresponding to different strains. For each consensus module, determine the correlation between the module reconnection score of the consensus module and the average resistance index of the corresponding strain, and determine the resistance correlation coefficient and robustness index corresponding to the consensus module. Based on the module reconnection score, the resistance correlation coefficient, and the robustness index, the comprehensive score of each consensus module is determined. Based on the ranking of the comprehensive scores, the key modules that are significantly related to the resistance index are selected.

3. The construction method according to claim 1, characterized in that, Step S3 includes: A linear regression model was constructed based on the cognitive function test data and pathological burden test data of the reference strain. Based on the linear regression model, and the cognitive function test data and pathological burden test data of each sample, the individual resistance index of each sample is determined. The average resistance index of each individual sample under the same strain is calculated by averaging the individual resistance indexes.

4. The construction method according to claim 1, characterized in that, Step S6 includes: Based on all co-expression networks, the number of times each gene pair was sampled in the same sampling subset was counted, and a gene co-occurrence indicator matrix was constructed. Count the number of times gene pairs with common sampling relationships are assigned to the same module, and construct a co-clustering matrix; Based on the gene co-occurrence indicator matrix and the co-clustering matrix, all genes are subjected to average link hierarchical clustering, and a dynamic tree cutting algorithm is used for clustering to obtain multiple consensus modules.

5. The construction method according to claim 1, characterized in that, Step S7 includes: Determine the correlation coefficients between the characteristic genes of each module; Gene pairs whose absolute correlation coefficient is greater than a preset threshold and meet the preset requirements after multiple tests are selected as valid edges. For each strain, the diseased topology network and the control topology network are constructed based on the selected effective edges, using the module feature genes as network nodes and the correlation coefficient as edge weights.

6. The construction method according to claim 1, characterized in that, Step S8 includes: The diseased topological network and the control topological network of the same strain are superimposed to obtain a superimposed network; The missing edges in the control topology network and the acquired edges in the diseased topology network are statistically analyzed. The network topology reconnection index is obtained by dividing the sum of the number of lost edges and the number of gained edges by the total number of edges in the superimposed network.

7. The construction method according to any one of claims 1-6, characterized in that, The disease model sample population includes diseased mice; The diseased mice were selected from at least one of the following: AD-C57, AD-A / J, AD-PWK, AD-LOT, AD-STUCKY, and AD-PIPING.

8. A method for screening biomarkers based on the expression interaction network of Alzheimer's disease resistance-related genes, characterized in that, The marker screening method includes the following steps: S11. Construct an Alzheimer's disease resistance-related gene expression interaction network by performing the method described in any one of claims 1-6; S12. Extract key genes and upstream transcription factors that regulate the key genes from the expression interaction network, and use the key genes and / or the upstream transcription factors as candidate biomarkers. S13. Verify the association between the candidate biomarkers and cognitive resistance in a population cohort, and identify the biomarkers that demonstrate significant verification as Alzheimer's disease resistance-related biomarkers.

9. The marker screening method according to claim 8, characterized in that, Step S12 includes: Gene feature assessment was performed on the consensus module in the Alzheimer's disease resistance-related gene expression interaction network to determine the module membership degree and gene significance of the characteristic genes of each module; the consensus module includes multiple module characteristic genes; The key genes are obtained by screening the feature genes of the module based on the gene connectivity within the module, the module membership degree, and the gene significance; wherein, the screening includes: screening genes whose module membership degree is greater than a first preset percentile of the membership degree of all gene modules within the corresponding consensus module, and whose module membership degree reaches a first preset significance level; screening genes whose gene connectivity within the module is greater than a second preset percentile of the connectivity of all genes within the consensus module; screening genes whose gene significance reaches a second preset significance level after verification and correction; and determining the genes that simultaneously meet the three screening conditions as the key genes; Pre-defined sequences upstream and downstream of the transcription start sites of each key gene are extracted as promoter regions. Based on the position weight matrix of transcription factors in a preset transcription factor database, the promoter region is scanned according to a preset score threshold to obtain potential transcription factor binding sites; Using the promoter sequences of all genes within the same consensus module as the background set, enrichment tests are conducted based on the distribution of potential transcription factor binding sites to screen upstream transcription factors that meet the preset correction significance criteria and enrichment intensity requirements. The upstream transcription factor and / or the key gene are used together as the candidate biomarker.

10. The marker screening method according to claim 8, characterized in that, After step S13, the following is also included: S14. Obtain transcriptome and clinicopathological sample data of brain tissue from the population, and construct the population disease resistance index based on the preset regression model. S15. Screen the human homologous genes corresponding to the key genes and / or the upstream transcription factors as candidate genes. S16. Perform multi-dimensional population association analysis on the candidate genes and verify the cross-species conservation of the regulatory relationship between the key genes and the upstream transcription factors; the multi-dimensional population association analysis includes the association between the candidate genes and the disease resistance index of the population, the association between the candidate genes and the pathology and neuroinflammatory traits of Alzheimer's disease, and the predictive evaluation of the candidate genes for dementia diagnosis results. S17. Identify candidate genes and / or upstream transcription factors that have been validated through multidimensional population studies as human-related biomarkers for resistance to Alzheimer's disease.

Images