Figures and data
Overview of trans-omic analysis for metabolic reactions in Alzheimer’s disease
The illustration shows how public multi-omic data from the DLPFC of cognitively normal subjects (CT) and patients with Alzheimer’s disease (AD) from the ROSMAP study were used to generate a metabolic trans-omic network (Steps 1-3) and to perform detailed analysis of energy production pathways (Step 4). Red indicates differentially expressed molecules with increased expression or abundance, regulatory events that activate metabolic reactions, or increased activity of a metabolic reaction. Blue indicates differentially expressed molecules with decreased expression or abundance, regulatory events that inhibit metabolic reactions, or reduced activity of a metabolic reaction. The sample size for each analysis: n = 87 (CT group) and n = 156 (AD group) for transcriptomics, n = 74 (CT group) and n = 71 (AD group) for proteomics, and n = 61 (CT group) and n = 144 (AD group) for metabolomics.
Construction of a trans-omic network for differentially regulated metabolic reactions
(A) The number of allosteric regulators among DEMs and the number of allosteric regulations mediated by them on metabolic enzymes identified with BRENDA. Some reactions are regulated by more than one DEM and some DEMs regulate multiple enzymes, thus the number of regulations exceeds the number of enzymes identified with BRENDA and the number of DEMs. (B) The number of substrates or products among DEMs and the number of regulations by them on metabolic enzymes identified with KEGG. Some reactions involve multiple substrates or yield multiple products and some substrates and products occur in multiple reactions, thus the number of regulations exceeds the number of enzymes identified with KEGG and the number of DEMs. (C) The trans-omic network for differentially regulated metabolic reactions (left). Nodes and edges indicate differentially expressed molecules and differential regulations involved in metabolic reactions, respectively. Schema of regulatory input into the metabolic reactions (right). TFs control metabolic enzyme mRNA abundance through gene regulation. Metabolite acts as substrates, products, and allosteric regulators in metabolic reactions. See Fig. S8 and Methods section for additional details. (D) Top 20 molecules with high degree centrality on the metabolic network. The metabolic network consisted of the nodes and edges from the “Enzyme Protein”, “Metabolic Reaction”, and “Metabolite” layers of the trans-omic network. Those shown in red text were increased in AD; those in blue were decreased in AD.
The metabolic dysregulations of energy metabolic pathways of AD
The trans-omic changes in the regulations of glycolysis (A), TCA cycle (B), oxidative phosphorylation (C), and ketone body metabolism (D). The information for each pathway were obtained from “glycolysis/gluconeogenesis” (hsa00010), “TCA cycle” (hsa00020), “oxidative phosphorylation” (hsa00190), and “butanoate metabolism” (hsa00650) in the KEGG database. Red upward triangles, increased molecules in AD; blue downward triangles, decreased molecules in AD. Red arrow, regulations that activate metabolic reactions; blue arrow, regulations that inhibit reactions. Unfilled circles with dashed outlines represent molecules that were not differentially expressed or were not measured. Unfilled diamond nodes with dashed outlines represent metabolic reactions. See Fig. S8 and Methods section for additional details.
Models of potentially dysregulated metabolic processes in AD
Diagrams of dysregulated metabolic pathways associated with the progression of AD. Metabolic processes affected by AD are highlighted in yellow boxes. Red, increased (circles) or activating (arrows) in AD; blue, decreased (circles and GLUT3) or inhibiting (arrows) in AD.
Information for the subjects from ROSMAP study.
(A) Information of the samples from ROSMAP Study used in this study, including sample size, sex, ApoE genotype, Braak Stage, CERAD score, age at death, and postmortem interval (PMI). (B) Venn diagrams showing the overlap in data available for each subject across the three omic levels: transcriptome, proteome, and metabolome.
Comparison of transcriptome, proteome, and metabolome data from the ROSMAP Study with data from other studies.
(A) Correlation between two transcriptomic datasets obtained from the ROSMAP Study and the MSSB Study for each gene. In the MSSB Study, RNA sequencing data were collected from brain tissue (FP; frontal pole) of individuals with AD (45 AD patients) and control subjects (45 CT subjects). For correlation analysis, the log2 fold-change values in 16,348 genes were calculated by dividing the average expression of the AD group by the average expression of the CT group. (B) Venn diagrams showing the overlap of upregulated DEGs between the ROSMAP study and those identified in a large-scale meta-analysis across seven brain regions [81]. (C) Venn diagrams showing the overlap of downregulated DEGs between the ROSMAP study and those identified in a large-scale meta-analysis across seven brain region [81]. (D) Correlation between two proteomic datasets obtained from the ROSMAP study and AMP-AD_DiverseCohorts study for each protein. TMT-MS data from each study were from brain tissue (DLPFC) of individuals with AD (637 AD patients) and control subjects (224 CT subjects). Correlation analysis was performed on the log2 fold-change values (average abundance of AD divided by average abundance of CT) for 9,180 proteins. Specifically, red nodes represent the subset of enzyme proteins which are involved in energy metabolism shown in Fig.3. (E) Rank-rank hypergeometric overlap (RRHO) heatmap comparing proteomic changes in AD compared to CT between the ROSMAP study (x-axis) and the AMP-AD_DiverseCohorts study (y-axis). Proteins were ranked based on -log10(q-value) multiplied by the sign of log2 fold-change in each dataset. Color intensity represents -log10(p-value) of the overlap at each rank threshold. Enrichment in bottom-left and top-right quadrants indicates concordant changes between the two datasets. (F) Differentially expressed metabolites (DEMs) identified in previous studies [71,72,73]. Note that each study used different definitions for AD and CT groups, had variations in measurement methods and brain regions analyzed. These studies had smaller sample sizes than our study. (G) RRHO heatmaps comparing metabolite changes in AD compared to CT between the ROSMAP study (DLPFC) (x-axis) and independent metabolomic datasets from multiple cohorts and brain regions [82,83] (y-axis). The Knight ADRC/DIAN dataset [83] consisted of 79 postmortem brain samples from cerebellar cortex (CER) of AD, 24 samples from CER of CT, 63 samples from temporal cortex (TCX) of AD, and 20 samples from TCX of CT and quantified 627 metabolites. The Mayo Clinic cohort datasets [82] consisted of 305 postmortem brain samples from parietal cortex (PC) of AD, 26 samples from PC of CT and quantified 685 metabolites. Metabolites were ranked based on -log10(p-value or q-value) multiplied by the sign of log2 fold-change in each dataset. Enrichment in bottom-left and top-right quadrants indicates concordant changes between the two datasets.
Cell type proportions estimation using a previous single-cell dataset.
Cell type proportions were estimated using a publicly available single-nucleus transcriptomic dataset from DLPFC tissue of 465 subjects in the ROSMAP cohort [84]. For each omic dataset, overlapping subjects were identified and cell type proportions were calculated for the matched samples. Box plots show the proportions of astrocytes, excitatory neurons, inhibitory neurons, microglia, oligodendrocytes, and OPCs in the AD group (blue) and the control group (orange) for the transcriptomic dataset (A; n = 91 AD, n = 50 CT), the proteomic dataset (B; n = 21 AD, n = 33 CT), and the metabolomic dataset (C; n = 58 AD, n = 25 CT). Statistical comparisons were performed using two-tailed t-tests with Bonferroni correction (significance threshold: p-value < 8.3 × 10-3). n.s., not significant.
Differential and Enrichment analysis for differentially expressed molecules of AD.
(A) Volcano plot for differentially expressed genes in AD. Increased mRNAs are shown in red. Decreased mRNAs are shown in blue. (B) Volcano plot for differentially expressed proteins in AD. Increased proteins are shown in red. Decreased proteins are shown in blue. Differentially expressed enzyme proteins are shown in yellow. (C) Volcano plot for differentially expressed metabolites in AD. Increased metabolites are shown in red. Decreased metabolites are shown in blue. (D) Enrichment analysis for DEGs, DEPs, and DEMs in AD performed with Metascape (for DEGs and DEPs) and MetaboAnalyst (for DEMs). Only significantly enriched terms (q-value < 0.05) are shown.
Construction of protein-protein interaction networks in DLPFC of AD.
(A) The protein-protein interaction (PPI) network for increased DEPs. Red nodes and gray edges indicated DEPs and their interactions, respectively. The nodes whose degree are over 18 their degree centralities are labeled for reasons of legibility (Supplementary Data 2). (B) The PPI network for decreased DEPs. Blue nodes and gray edges indicated DEPs and their interactions, respectively. The nodes whose degree are over 75 their degree centralities are labeled for reasons of legibility (Supplementary Data 2). (C) Top 30 DEPs with high degree centrality on the increased PPI network. (D) Top 30 DEPs with high degree centrality on the decreased PPI network.
Top 20 differentially expressed genes, proteins, and metabolites.
(A) The top 20 differentially expressed genes (DEGs) ranked and plotted according to log2 fold change (Log2FC) in mean expression between AD and CT. Red, DEGs with increased expression in AD (left); blue, DEGs with decreased expression in AD (right). (B) The top 20 differentially expressed proteins (DEPs) ranked and plotted according to log2 fold change (Log2FC) in mean expression between AD and CT. Red, DEPs with increased expression in AD (left); blue, DEPs with decreased expression in AD (right). (C) The top 20 differentially expressed metabolites (DEMs) ranked and plotted according to log2 fold change (Log2FC) in mean expression between AD and CT. Red, DEMs with increased expression in AD (left); blue, DEMs with decreased expression in AD (right).
Identification of differential regulations connecting differentially expressed molecules across different omic layers.
(A) The change in the encoding gene for the corresponding DEP grouped according to the increased DEPs and decreased DEPs in AD. (B) Correlation of the log2 fold changes (AD group vs. CT group) of measured molecules at the transcriptome level (x-axis) and proteome level (y-axis). Each point on a scatter plot represents a molecule measured as both mRNA and protein. DEPs are shown in blue or red. Blue represents those DEPs with encoding DEGs that change in the opposite direction; Red represents the rest of the DEPs. (C) Log2 fold change (AD group vs. CT group) of each differentially expressed transcription factor (DETF). The log2 fold change (Log2FC) of gene expression (transcriptome) or protein expression (proteome) of the indicated DETFs, which were identified from the 19 TFs that were inferred as potential regulators of DEGs with significantly altered expression in AD.
Contents of the trans-omic network for differentially regulated metabolic reactions between CT and AD.
(A) The definition of differentially expressed molecules in each layer (left). Differential regulations between adjacent layers and the databases used to identify them (right). (B) Classification of the differential regulations (activating or inhibiting in AD group). The methodology of constructing differential regulatory trans-omic network of metabolic reactions is based on our previous study [74,75]. Briefly, differentially expressed TFs (DETFs), genes (enzyme mRNAs), proteins (metabolic enzymes), metabolic reactions, and metabolites were integrated to construct a five-layer network for AD. Adjacent layers were linked by known regulatory relationships. Edge colors indicate the direction and functional sign of changes: red/blue denote activating/inhibiting regulation for TF-mRNA, mRNA-protein, and protein-reaction edges. For reaction-metabolite links, edge colors were assigned based on the metabolite change and its role (allosteric activator/inhibitor or substrate/product).
A list of transcription factors inferred to be involved in the regulation of DEGs in DLPFC of AD.
