Cross-Tissue WGCNA Reveals Conserved Brain-Blood Transcriptomic Signatures and Potential Biomarkers for Asymptomatic Alzheimer's Disease Diagnosis

Technical Summary: Cross-Tissue WGCNA Reveals Conserved Brain-Blood Transcriptomic Signatures for Asymptomatic Alzheimer's Disease

Problem Statement
Alzheimer's disease (AD) is characterized by a prolonged prodromal stage where molecular changes occur before overt symptoms, leading to late diagnosis and limited therapeutic efficacy. Current diagnostic methods, such as cerebrospinal fluid analysis and PET imaging, are invasive and symptom-dependent, rendering them unsuitable for asymptomatic detection. While peripheral blood offers a non-invasive alternative for early biomarker discovery, existing approaches face significant hurdles:

1. Signal Dilution: Systemic noise and physiological heterogeneity in blood obscure subtle, coordinated transcriptional alterations.
2. Methodological Limitations: Conventional differential expression analysis (DEG) focuses on individual genes with statistically significant changes, failing to capture the subtle, coordinated network shifts typical of the prodromal stage.
3. Lack of Cross-Tissue Validation: Many studies construct co-expression networks solely within peripheral blood, lacking the anatomical anchoring necessary to confirm a central nervous system (CNS) origin. Furthermore, many investigations are biased toward symptomatic cohorts rather than the critical asymptomatic phase.

Methodology
The authors developed a multi-tiered, cross-tissue bioinformatic pipeline to identify peripheral biomarkers for asymptomatic AD (AsymAD), utilizing public transcriptomic datasets from the Gene Expression Omnibus (GEO).

1. Data Acquisition and Preprocessing:

   • Brain Cohort (GSE118553): Postmortem human brain tissue (entorhinal and frontal cortices) from cognitively healthy controls, early-stage AsymAD, and late-stage symptomatic AD.
   • Blood Cohort (GSE282742): Peripheral blood samples from patients with progressive/stable mild cognitive impairment (MCI) and symptomatic AD.
   • Preprocessing: Data underwent log2 transformation, quality control (removing outliers based on sample connectivity), and batch effect correction using linear models with covariates (diagnosis, age, sex).

2. Network Construction (WGCNA):

   • An unbiased co-expression network was constructed in the brain tissue using the top 10,000 most variable genes (selected via Median Absolute Deviation) to avoid pre-filtering based on differential expression.
   • Soft-thresholding powers were optimized to achieve scale-free topology (R² > 0.8).
   • Hierarchical clustering identified distinct gene modules in the entorhinal cortex (early stage) and frontal cortex (late stage).

3. Cross-Tissue Preservation Analysis:

   • Brain-derived modules were projected onto the independent peripheral blood dataset to assess structural conservation.
   • Preservation was quantified using the Z-summary statistic, where scores >10 indicate strong preservation, 2–10 indicate moderate preservation, and <2 indicate no preservation.

4. Hub Gene Prioritization and Validation:

   • Hub Selection: Genes within preserved modules were prioritized using two metrics: Module Membership (MM) (structural connectivity) and Gene Significance (GS) (correlation with disease status).
   • Machine Learning: A Random Forest classifier was trained on the prioritized hub genes using an independent peripheral blood microarray dataset (GSE63060) to distinguish AsymAD patients from healthy controls.
   • Performance Metrics: Model accuracy and Area Under the Curve (AUC) were calculated.

Key Results

• Module Identification: In the early-stage entorhinal cortex, 10 co-expression modules were identified. In the late-stage frontal cortex, 7 modules were identified.
• Cross-Tissue Preservation: Only one module from the early-stage network, the Green module, demonstrated strong structural preservation in peripheral blood (Z-summary = 15.25). Moderate preservation was observed for yellow, blue, and magenta modules, but the majority of brain-specific modules (including the dominant turquoise and trait-correlated purple modules) lost topological integrity in blood.
• Biological Function: The preserved Green module (973 genes) is enriched for mitochondrial energy pathways, protein-folding machinery, proteostasis, and apoptotic signaling.
• Five-Gene Signature: From the Green module, five hub genes were prioritized based on high MM and GS: ATP6AP2, CKS1B, STAMBPL1, SUB1, and BET1.
  • ATP6AP2 served as the primary structural regulator (MM > 0.85).
  • CKS1B and BET1 showed the strongest direct association with the AsymAD phenotype (GS > 0.4).
• Diagnostic Performance: The Random Forest classifier trained on these five genes distinguished AsymAD patients from healthy controls in the independent validation set with:
  • Accuracy: 69.1%
  • AUC: 0.722
  • Feature Importance: SUB1 was the dominant predictive feature (score 100.00), followed by CKS1B (66.51). Notably, ATP6AP2, despite being the topological anchor, had a predictive weight of 0.00 in the classification model.

Significance and Claims
The paper claims to present a robust framework for early AD screening by overcoming the limitations of conventional DEG analysis and single-tissue studies. Its primary contributions include:

1. Anatomical Anchoring: By anchoring the network in the entorhinal cortex (the earliest site of AD pathology) and confirming preservation in blood, the study provides evidence that the identified signature originates from CNS neurodegeneration rather than peripheral noise.
2. Unbiased Network Approach: The methodology avoids pre-filtering for differentially expressed genes, allowing for the detection of subtle, coordinated transcriptional shifts that characterize the asymptomatic stage.
3. Non-Linear Dynamics: The study highlights a decoupling between global module stability and linear disease progression. While the Green module's global eigengene did not correlate linearly with AsymAD status, specific hub genes within it did. This supports the hypothesis that early neurodegeneration involves multidimensional biological collapse rather than linear protein accumulation.
4. Translational Potential: The identification of a five-gene panel (ATP6AP2, CKS1B, STAMBPL1, SUB1, BET1) that functions as a computationally predictive, blood-based biomarker offers a potential pathway for non-invasive early diagnosis.

Limitations and Future Directions
The authors modestly frame their findings as a "computational, transcript-level proof of concept." They acknowledge limitations including:

• Tissue Heterogeneity: The use of bulk RNA-seq prevents the attribution of signals to specific cell types (e.g., specific leukocyte subsets).
• Study Design: The cross-sectional nature of the public cohorts precludes establishing temporal causality.
• Clinical Translation: The findings require validation in large-scale, prospective clinical trials and translation into high-throughput molecular assays (e.g., RT-qPCR) before clinical utility can be established.

The paper concludes that while the 5-gene panel shows diagnostic potential, future research must prioritize longitudinal validation to track expression dynamics as patients transition from prodromal stages to clinical cognitive decline.