The application of weighted gene co-expression network analysis in identifying key modules and hub genes associated with disease status in Alzheimer's disease

Abstract

Background: Alzheimer's disease (AD) is the most common neurodegenerative condition that affects more than 15 million individuals globally. However, a predictive molecular biomarker to distinguish the different stages of AD patients is still lacking.

Methods: A weighted gene co-expression network analysis (WGCNA) was employed to systematically identify the co-expressed gene modules and hub genes connected with AD development based on a microarray dataset (GSE1297) from the Gene Expression Omnibus (GEO) database. An independent validation cohort, GSE28146, was utilized to assess the diagnostic efficiency for distinguishing the different stages of AD. Quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR) and western blotting analysis were applied to examine the mRNA and protein level of GRIK1, respectively, in AD mice established with the expression of mutant amyloid precursor protein and wild type mice. The morphology of neurons was investigated using phalloidin staining.

Results: We identified 16 co-expressed genes modules, with the pink module showing significant association with all three disease statuses [neurofibrillary tangle (NFT), BRAAK, and mini-mental state examination (MMSE)]. Enrichment analysis specified that these modules were enriched in phosphatidylinositol 3-kinase (PI3K) signaling and ion transmembrane transport. The validation cohort GSE28146 confirmed that six hub genes in the pink module could distinguish severe and non-severe AD patients [area under the curve (AUC) >0.7]. These hub genes might act as a biomarker and help to differentiate diverse pathological stages for AD patients. Finally, one of the hubs, GRIK1, was validated by an animal AD model. The mRNA and protein level of GRIK1 in the AD hippocampus was increased compared with the control group (NC) hippocampus. Phalloidin staining showed that dendritic length of the GRIK1 pCDNA3.1 group was shorter than that of the NC group.

Conclusions: In summary, we systematically recognized co-expressed gene modules and genes related to AD stages, which gave insight into the fundamental mechanisms of AD progression and revealed some probable targets for the treatment of AD.

Keywords: Alzheimer’s disease (AD); gene modules; pathological stage; weighted gene co-expression network analysis (WGCNA).

Figure 1

Clustering of samples and detection of soft-thresholding power. (A) Sample clustering was conducted to detect outliers. All samples are situated in the clusters and passed the cutoff thresholds; (B,C) analysis of the scale-free fit index (B) and analysis of the mean connectivity (C) for numerous soft-thresholding powers.

Figure 2

Dendrogram of all genes grouped depending on a dissimilarity measure (1-TOM). TOM, topological overlap matrix.

Figure 3

Identification of modules associated with the clinical traits of AD. (A) A heatmap of co-expressed genes. Different colors in the X and Y axis signify different modules. The intensity of yellow represents the degree of connectivity of different modules; (B) correlation among modules and traits; (C) top: dendrogram of ME acquired by WGCNA; bottom: heatmap plot of the adjacencies of modules. Red represents high adjacency whereas blue represents low adjacency. AD, Alzheimer’s disease; ME, module eigengene; NFT, neurofibrillary tangle; MMSE, mini-mental state examination; PMI, postmortem interval; WGCNA, weighted gene co-expression network analysis.

Figure 4

Clinical relevance and functional analysis of genes in module Mepink. (A,B,C) A scatter plot of gene significance for NFT (A), BRAAK (B), and MMSE (C) vs. the module membership in the pink module; (D) the genes in the pink module and the red signify the hub nodes; (E) statistics of GO term enrichment for genes in different modules for WGCNA. Percentage is the ratio of the number of genes of a specific module in a certain pathway to the number of total genes. ME, module eigengene; NFT, neurofibrillary tangle; MMSE, mini-mental state examination; GO, gene ontology; WGCNA, weighted gene co-expression network analysis.

Figure 5

Efficacy evaluation of hub genes. (A) Hierarchical clustering of AD samples from the GSE28146 validation set using the hub genes in the pink module; (B) ROC curve for the six hub genes. AD, Alzheimer’s disease; ROC, receiver operating characteristic; AUC, area under the curve.

Figure 6

The expression of GRIK1 in the AD cortex and AD hippocampus. (A) The expression of GRIK1 in the AD cortex determined by qRT‐PCR. **, P<0.01. NC: The control group, n=10; the AD group, n=10; (B) Western blot showing the expression of GRIK1 in the AD cortex and control group; (C) relative expression of GRIK1 in (B). ***, P<0.005. n=3; (D) the expression of GRIK1 in the AD hippocampus determined by qRT‐PCR. *, P<0.05, NC: the control group, n=10; the AD group, n=10; (E) Western blot showing the expression of GRIK1 in the AD hippocampus and control group; (F) relative expression of GRIK1 in (E). **, P<0.01. n=3; (G) phalloidin staining showing the morphology of the primary neurons; (H) relative dendritic length in (G). **, P<0.01. n=3. The experiments were independently repeated 3 times. AD, Alzheimer’s disease; qRT‐PCR, quantitative real-time reverse transcription polymerase chain reaction.

Figure S1

Statistics of GO term enrichment for genes in blue, salmon, and lightcyan modules. GO, gene ontology.

Figure S2

ROC curve for the hub genes in blue (A), salmon (B), and lightcyan (C) modules. ROC, receiver operating characteristic; AUC, area under the curve.