Single-cell network biology characterizes cell type gene regulation for drug repurposing and phenotype prediction in Alzheimer's disease

Abstract

Dysregulation of gene expression in Alzheimer's disease (AD) remains elusive, especially at the cell type level. Gene regulatory network, a key molecular mechanism linking transcription factors (TFs) and regulatory elements to govern gene expression, can change across cell types in the human brain and thus serve as a model for studying gene dysregulation in AD. However, AD-induced regulatory changes across brain cell types remains uncharted. To address this, we integrated single-cell multi-omics datasets to predict the gene regulatory networks of four major cell types, excitatory and inhibitory neurons, microglia and oligodendrocytes, in control and AD brains. Importantly, we analyzed and compared the structural and topological features of networks across cell types and examined changes in AD. Our analysis shows that hub TFs are largely common across cell types and AD-related changes are relatively more prominent in some cell types (e.g., microglia). The regulatory logics of enriched network motifs (e.g., feed-forward loops) further uncover cell type-specific TF-TF cooperativities in gene regulation. The cell type networks are also highly modular and several network modules with cell-type-specific expression changes in AD pathology are enriched with AD-risk genes. The further disease-module-drug association analysis suggests cell-type candidate drugs and their potential target genes. Finally, our network-based machine learning analysis systematically prioritized cell type risk genes likely involved in AD. Our strategy is validated using an independent dataset which showed that top ranked genes can predict clinical phenotypes (e.g., cognitive impairment) of AD with reasonable accuracy. Overall, this single-cell network biology analysis provides a comprehensive map linking genes, regulatory networks, cell types and drug targets and reveals cell-type gene dysregulation in AD.

Fig 1. An integrative network-biology framework for analyzing cell type gene regulatory mechanisms in Alzheimer’s disease (AD).

(A) Predicting neuronal and glial cell type GRNs from multi-omics data in AD and control by integrating snRNA-seq with chromatin interaction data and TF binding site information. (B) Analyses of cell type GRN characteristics include identification of hub genes, regulatory hierarchy, network motifs, regulatory logics, modules of co-regulated genes, and drug-module associations. (C) Machine learning based prioritization of novel AD genes using network interaction patterns and prediction of clinical phenotypes using machine learning.

Fig 2. Centrality analysis reveal hub gene changes of cell-type gene regulatory networks in AD.

(A) An upset plot showing overlaps between the top 10% genes with largest out-degree and (B) betweenness centralities. The filled dots in the center matrix indicate the comparison between the respective sets (along the x-axis), and the bars on the top show size of the intersection. Blue and red rows indicate control and AD, respectively. (C) Scatter plots showing normalized gene centralities distribution and (D) the distribution of in-degrees in microglial AD and control networks. Genes with large changes in in-degree between AD and control are labelled. (E) Visualization of the subnetwork of 9 TFs with high betweenness centrality in microglia. Grey circles around the periphery of the network indicate target genes. Symbols of the nine central TFs are shown and the rest hidden for clarity. Blue and red edges indicate interaction in the control and AD networks, respectively. (F) A dot plot showing enrichment of gene ontology biological processes (y-axis) among genes with the most extensive changes in the in-degree centrality across all cell types (x-axis). The dot size is set according to the FDR-corrected p values, as shown in the key.

Fig 3. Hierarchy analysis of cell type gene regulatory networks in AD.

(A) The distribution of TFs in three levels of hierarchy in AD, control and 1000 random GRNs across all four cell types. (B) The rewiring scores of TFs (x-axis) across all three levels of hierarchies (y-axis). The distributions of (C) number of promoters targeted by TFs (y-axis) and (D) the fold change values (log scale; y-axis) of target genes (AD versus healthy controls) of TFs at the three levels of hierarchy (x-axes). (E) Enrichment of gene ontology biological processes (y-axis) within targets of top, middle and bottom layers of the regulatory hierarchy across cell type networks (x-axis).

Fig 4. Network motifs and regulatory logic across cell types in AD.

(A) Barplots showing the enrichment (x-axis; Z-score estimates from random networks) of various three-node triplets (y-axis) in AD (red) and control (blue) conditions across all four cell types. (B) Genes that frequently occur in feed-forward loops in cell type AD GRNs are depicted and colored uniquely for each cell type. (C) Barplot showing the frequency (x-axis) of various logic gates (y-axis) active within the feed-forward loops in AD (red) and control (blue) conditions across all four cell type networks. (D) Logic gate diagram showing PPARG-NFYA-CREBP triplet’s AND logic in AD and OR logic in control networks of microglia.

Fig 5. Coregulated gene modules reveal cell type-specific drug-repurposed targets and gene functions in AD.

(A) Illustration depicting the concept of gene set cohesiveness in a network. The bar plot below shows the number of gene ontology biological process terms (y-axis) that gain (blue) or lose (red) cohesiveness between control and AD networks across all cell types (x-axis; see Methods). (B) A heatmap showing the enrichment of co-regulated modules of the microglia AD network within differentially expressed genes in various AD pathologies. The average fold-change of genes within each module was transformed to a Z-score to derive the enrichment score. Negative and positive Z-scores indicate down- and up-regulation, respectively, of co-regulated modules (x-axis) in AD pathologies (y-axis). The grids of the heatmap are colored accordingly, with red indicating down-regulation and blue indicating up-regulation of the module. (C) Visualization of genes in module 1 of the microglia AD coregulatory network. Each circle in the plot is a gene, with TFs depicted as triangles, known AD-genes in octagons, and other genes as ellipses. Nodes are colored according to fold change values in AD pathology (early versus no pathology) as shown in the key. (D) Proposed mechanism-of-action for treatment of AD by everolimus using drug-target network analysis with microglia M1. (E) Proposed mechanism-of-action for treatment of AD by Rifampcian using drug-target network analysis with microglia M4.

Fig 6. AD Gene prioritization and clinical phenotype prediction using network-based machine learning.

(A) Boxplots showing the distribution of balanced accuracies (y-axis; obtained from 10 independent runs of five-fold cross-validation) in predicting known AD genes using interaction patterns in cell type GRNs as features (x-axis). (B) Gene ontology biological process terms enriched within the top 20% predictions in the microglia AD machine learning (ML) model. The terms are depicted along the y-axis, and the FDR corrected p-values are shown along the x-axis. (C) Genes were sorted according to their probability of being associated with AD in the microglia ML model, and the top 5% of the sorted list was used as features to predict AD phenotypes in an independent dataset (ROSMAP). The boxplots show the distribution of balanced accuracies (y-axis) obtained from testing four AD phenotypes (see Methods) and a set of randomly selected samples (x-axis). (D) Average feature importance scores of TFs at the three hierarchy levels in the microglia AD network. (E) Visualization of the subnetwork connecting top 10% TFs with highest feature importance scores in microglia AD network. Each grey node depicts a TF with border color set along a red gradient according to the disease-gene association score given in the DisGeneNet database (based on preliminary evidence collected from independent studies). (F) Feed-forward loops observed within top-ranked TFs in the microglia ML model (red line) and the distribution in 1000 random networks (grey bars). (G) Regulatory logics observed within top-ranked TFs in the microglia. (H) Enrichment of top-ranked genes within coregulated genes modules in microglia (*Permutation p value < 0.001).