Home · Research · Epigenetic Architectures
Project · Alzheimer's Epigenetics

Epigenetic Architectures of Alzheimer's Disease

A synthesis of blood DNA-methylation evidence for Alzheimer's disease: distinct profiles, predictive signatures, cognitive resilience, and probe reliability — drawn together from a series of longitudinal studies.

CohortsFHS · ADNI · AIBL · AddNeuroMed
TissuePeripheral blood
PlatformIllumina EPIC
StudiesZhang et al. 2023 · 2024 · 2025
§01 · Distinct profiles

The epigenetic landscape of AD is not one landscape

Key finding · Zhang et al., 2023
DNA methylation signatures associated with CSF biomarkers (Aβ42, pTau181) are distinct between Alzheimer's disease (AD) patients and cognitively normal (CN) subjects. Very few CpG sites appear in both lists — the pre-clinical and symptomatic phases of AD involve fundamentally different epigenetic programs.

In the 2023 study, CSF biomarker–associated methylation was analysed separately in CN and AD subjects. The overlap of significant CpGs between the two groups was minimal, arguing against a single universal "AD methylation signature" that simply changes in magnitude over the disease course. Instead, the preclinical phase looks like a regulated, possibly compensatory response, while the symptomatic phase looks like the breakdown of that response.

Figure 1
Overlap of CSF-biomarker-associated CpG sites in CN vs. AD

Figure 1. Stacked composition of significant CSF-biomarker-associated CpGs across biomarker categories. Bars decompose into CN-specific, AD-specific, and shared CpGs. The shared fraction is small, consistent with stage-specific epigenetic regulation.

i.
Distinct biological processes
In CN, methylation changes likely reflect homeostatic or compensatory mechanisms. In AD, they reflect the failure of those mechanisms.
ii.
Implication for diagnostics
A universal AD panel may miss preclinical cases. Early-screening tools must target CN-specific signals — not the same CpGs that define late-stage disease.
iii.
Liquid biopsy still useful
Despite the distinctness, peripheral blood DNAm correlates strongly with central pTau181 — supporting blood-based assays, provided they are stage-specific.
§02 · Predictive signatures

A 44-CpG / 44-DMR signature for incident dementia

Using longitudinal data from the Framingham Heart Study and ADNI, the 2025 Alzheimer's & Dementia paper derived a blood methylation signature that predicts dementia onset years before clinical diagnosis. The signature consists of 44 individual CpG sites plus 44 differentially methylated regions (DMRs). DMRs (clusters of co-regulated CpGs) are preferred over single sites because they are more biologically functional and more statistically robust against site-level noise.

Pathway enrichment — systemic, not just neural

The 44-CpG signature is enriched in peripheral immune, metabolic, and vascular pathways as strongly as in CNS pathways. Three biological axes stand out:

  • Peripheral immune dysregulation. Enrichment of immune-gene methylation challenges the "immune-privileged" view of the brain. Leukocyte DNAm changes reflect "inflammaging" — a low-grade chronic inflammatory state that communicates with the CNS via cytokine signalling.
  • Metabolic dysfunction ("type-3 diabetes"). Dysregulation of glucose and lipid metabolism genes mirrors brain insulin resistance. Peripheral energy deficits disproportionately affect the brain, which consumes ~20% of systemic glucose.
  • DMR-level regulation. Coordinated regional methylation (DMRs) captures joint regulation of gene neighbourhoods that single CpGs miss.
Cross-cohort validation
Replicated in AIBL (Australia) and AddNeuroMed (Europe)
§03 · Cognitive resilience

Why some people with pathology remain cognitively intact

The 2025 Alzheimer's Research & Therapy paper studied the Healthy Brain Initiative and ADNI cohorts to ask why some individuals maintain cognition despite accumulated pathology. We operationalised a methylation-based Resilience Index (mRI) from a lifestyle-based resilience score, and identified a blood DNAm signature that mediates the association between lifestyle and cognitive trajectory, independent of known risk factors.

The widget below is a schematic illustration — adjust lifestyle inputs to see how the model weights three canonical resilience genes (APOC2, UBAP1, GSTCD) and an inferred heart–brain axis term.

Lifestyle inputs
Diet (MIND / Med)Mod.
Physical activityMod.
Cognitive activityMod.
Stress / mindfulnessAvg.
Methylation-based resilience
50/100
Epigenetic targets — modulation
APOC2
Lipid metabolism. Enhances triglyceride clearance; supports lipid homeostasis in brain tissue.
UBAP1
Protein clearance (ESCRT). Boosts lysosomal degradation of toxic aggregates.
GSTCD
Oxidative defence. Enhances antioxidant capacity; protects neurons from ROS.
Heart–brain axis
Vascular health. Cardiomyopathy-pathway enrichment supports a vascular contribution to cognitive preservation.

Note. The simulator is for illustration only; weights are schematic. Actual analyses estimate mediation on the methylation-based Resilience Index in HBI and ADNI participants, adjusting for age, sex, APOE, and standard risk factors.

§04 · Blood–brain bridge

Peripheral methylation as a proxy for central biomarkers

A central question for liquid-biopsy biomarkers is whether blood methylation meaningfully reflects central-nervous-system biology. Across the Zhang et al. studies we report that:

  • Several HOX-family and immune-regulatory loci show coordinated methylation in blood and brain tissue.
  • Blood methylation at these loci correlates with CSF pTau181 in matched participants, supporting a blood–brain epigenetic axis.
  • Concordance is loci-specific, not genome-wide — caution is warranted when extrapolating from whole-blood DNAm to brain epigenetics.

This concordance is what makes a peripheral assay worth pursuing, provided the signature is anchored on loci where blood–brain agreement is empirically demonstrated rather than assumed.

§05 · Methodological rigor

Probe reliability on the Illumina EPIC v1.0 BeadChip

The 2024 Epigenetics paper performs a systematic evaluation of probe reliability on the EPIC v1.0 BeadChip. Not every probe is equally trustworthy — reliability varies with beta-value range, probe chemistry, and intra-array characteristics. We recommend a filtering workflow to retain probes with reproducible measurement properties before downstream inference.

Tier Probe class Recommended action
I.High reliabilityReproducible ICC > 0.75 across technical replicates. Suitable for discovery and clinical use.
II.Moderate reliabilityICC 0.4–0.75; retain for discovery with sensitivity analyses.
III.Low reliabilityICC < 0.4. Exclude from candidate biomarker lists.
IV.Cross-reactive / ambiguousKnown problematic probes; exclude by default.

Table 1. Probe reliability tiering. Applied before downstream biomarker discovery, this filter reduces false-positive rates and improves reproducibility across cohorts.

§06 · Key references

Underlying studies

Distinct profiles · 2023

Zhang, W., et al. (2023). Distinct CSF biomarker-associated DNA methylation in Alzheimer's disease and cognitively normal subjects. Alzheimer's Research & Therapy.

Probe reliability · 2024

Zhang, W., et al. (2024). Critical evaluation of the reliability of DNA methylation probes on the Illumina MethylationEPIC v1.0 BeadChip. Epigenetics.

Predictive signature · 2025

Zhang, W., et al. (2025). Blood DNA methylation signature for incident dementia: evidence from longitudinal cohorts. Alzheimer's & Dementia.

Cognitive resilience · 2025

Zhang, W., et al. (2025). DNA methylation signature of a lifestyle-based resilience index for cognitive health. Alzheimer's Research & Therapy.

See the full list on Google Scholar.