Gene Expression Biomarkers Across Alzheimer's Disease Progression

At a glance

Overview

This project applies a differential expression and pathway enrichment pipeline, built from scratch in R, to hippocampal gene expression data from Alzheimer's disease patients at different clinical stages. We approached this as staff bioinformaticians embedded in a neurobiology lab, tasked with identifying candidate genes and biological pathways that track with AD severity. Our specific focus was to compare moderate stage patients against incipient (early) stage patients, and severe stage patients against moderate stage patients — the two transitions that describe how the disease actually progresses once it has taken hold, rather than how it differs from a healthy baseline.

An estimated 6.9 million Americans age 65 and older were living with Alzheimer's in 2024, a number projected to reach 12.7 million by 2050 without a medical breakthrough. Despite that scale, AD still has no confirmed cause and no disease-modifying treatment in routine use. The dataset was built around a specific, practical question: as gene expression data becomes cheaper and more available from post-mortem and biopsy brain tissue, what can it tell us about which genes and pathways change as AD gets worse, and could any of them serve as biomarkers for staging the disease or as targets for treatment?

The analysis used the Blalock et al. (2011) hippocampal gene expression dataset (GEO series GSE28146), which profiled brain tissue from cognitively normal controls and AD patients at three clinical stages: incipient, moderate, and severe. Our two comparisons drew on small but balanced clinical groups: 8 moderate vs. 7 incipient, and 7 severe vs. 8 moderate. Given the small per-group sample sizes typical of post-mortem brain tissue studies, we filtered on a raw p-value threshold (p <= 0.01) rather than an FDR-corrected threshold, treating this as a hypothesis-generating, exploratory analysis rather than a confirmatory one.

Phase 1 — Differential expression: using a custom Welch two-sample t-test function (fnTTest), we ran a gene-by-gene comparison for each transition and filtered to genes with p <= 0.01, producing shortlists of 157 genes (moderate vs. incipient) and 195 genes (severe vs. moderate). The strongest signals in the moderate-versus-incipient transition included TNKS, EXOC7, and NDUFA10 (a mitochondrial complex I subunit); the strongest signals in the severe-versus-moderate transition included AKIP1, TNPO3, and CCNL1.

Phase 2 — Functional enrichment: each gene shortlist was submitted to EnrichR against roughly 30 curated databases (KEGG, GO Biological Process, Reactome, WikiPathways, BioPlanet, MSigDB Hallmark, and others). Because the input gene lists were modest in size (under 200 genes each) and the analysis was intentionally exploratory, we treated nominal p-values (unadjusted, generally p < 0.01 to p < 0.05) as a screening signal for which pathways were worth investigating in the literature, exactly as intended for this exploratory approach, rather than as strictly significant findings after multiple testing correction.

Phase 3 — Literature-grounded biological interpretation: from the candidate pathways, the team selected four with a clear, literature-supported connection to AD biology. The estrogen receptor pathway was significant in both transitions (p < 0.001 in each), with JUN and PPARA downregulated in the earlier transition and ACOX1 and CYP1B1 downregulated in the later one — consistent with the documented protective effect of estrogen against beta-amyloid release and the overrepresentation of postmenopausal women among AD diagnoses. The insulin resistance / insulin signaling pathway was significant in the moderate-versus-incipient transition (p = 0.01) and represented across 17 related pathway hits in the severe-versus-moderate transition, tracking with research describing AD as, in part, 'diabetes of the brain.' The neuroinflammation signal sharpened as disease severity increased — 2 differentially expressed genes in the earlier transition (p = 0.0044) versus a stronger 9-gene signal in the later one (p = 0.00044) — lining up with the established role of chronically activated microglia and astrocytes in driving neuronal damage. Finally, the mitochondrial protein import pathway (Reactome R-HSA-1268020) was significant in both transitions, with genes involved in importing proteins into mitochondria implicated at each stage, mechanistically linked to energy deficits, oxidative stress, and eventual neuron loss.

Across both transitions, four biologically coherent, literature-supported themes emerge: a progressive loss of estrogen-mediated neuroprotection, worsening insulin signaling dysfunction that intensifies from incipient to severe stages, an inflammatory signature (centered on NF-kB and microglial activation pathways) that grows stronger as disease severity increases, and a persistent mitochondrial protein import deficit that compounds energy failure in neurons over time. Several individual genes — JUN, MTOR, PPARA, and the NF-kB-associated gene set in particular — appear in more than one pathway or more than one transition, providing a reasonable basis for prioritizing them for follow-up rather than treating every nominally significant hit as equally interesting. This project shows the same rigorous statistics and pathway enrichment pipeline applied to a second disease and dataset, and, more importantly, it shows what happens after the enrichment output comes back: turning a spreadsheet of pathway names and p-values into a set of falsifiable, literature-grounded hypotheses about which genes and mechanisms might matter for Alzheimer's staging, prognosis, or treatment.

Highlights

• Custom Welch t-test differential expression pipeline in R (fnTTest) with per-gene screening across ~43,000 probes
• 157-gene shortlist (moderate vs. incipient) and 195-gene shortlist (severe vs. moderate) filtered at p <= 0.01
• Multi-database EnrichR enrichment across KEGG, Reactome, WikiPathways, GO BP, BioPlanet, and MSigDB Hallmark
• Four literature-grounded pathways: estrogen receptor signaling, insulin resistance, neuroinflammation, and mitochondrial protein import
• Cross-transition gene prioritization: JUN, MTOR, PPARA, and NF-kB-associated genes recurring across multiple pathways and stages
• Team presentation and written summary translating statistical signal into testable biological narrative for a mixed technical and clinical audience