Scientists may have underestimated how much genes contribute to Alzheimer's risk

By Vijay Kumar MalesuReviewed by Susha Cheriyedath, M.Sc.Nov 2 2026

Drawing on genetic data from nearly 2.8 million people, researchers investigated how inherited risk relates to changes in the brain’s immune cells and neurons, and whether existing methods underestimate the genetic contribution to Alzheimer’s disease.

Study: Alzheimer’s disease genetics converges on microglial and neuronal dysregulation. Image Credit: Joyisjoyful / Shutterstock

In a recent study published in the journal Nature, a large international team of researchers identified genetic risk loci and implicated biological pathways, cellular processes, and potential therapeutic targets for Alzheimer’s disease (AD) across diverse ancestral populations.

Background

AD is the most prevalent type of dementia in the world, while dementia from all causes is projected to affect 153 million people worldwide by 2050. Twin studies estimate that 60%–80% of AD susceptibility is heritable, but earlier genome-wide association studies (GWAS) had identified about 80 risk loci, and previous estimates based on single-nucleotide polymorphisms (SNPs) suggested that common genetic variants accounted for only about 5% of disease liability. Current disease-modifying treatments primarily target amyloid-beta aggregates, but have shown limited effectiveness in relieving symptoms. Further research could clarify genetic risk, cellular changes and biological pathways and identify mechanisms that could inform new treatments and improve risk prediction.

About the study

The investigators combined de-identified genetic summary data from 30 cohorts, including 109,479 individuals with AD, 74,141 proxy cases, 2,131,799 controls, and 499,708 proxy controls. Proxy cases were participants whose mother or father was reported to have AD or dementia; participants with AD themselves were excluded from this analysis. The effective sample size was 416,493, with participants grouped by European, African, East Asian, admixed American, and South Asian ancestry. European ancestry accounted for approximately 91% of the effective sample size.

Participating cohorts checked variant and participant call rates, allele frequencies, and genotype consistency. AD cases were defined using clinical diagnoses or International Classification of Diseases, 10th Revision (ICD-10) codes, including unspecified dementia codes in some cohorts. GWAS analyses were conducted by ancestry group while controlling for age, sex, and other principal components and covariates of the study. Combined results were analyzed using inverse-variance weighted fixed-effect meta-analysis.

The researchers prioritized genes using fine-mapping, gene-based and gene-set analyses, quantitative trait locus (QTL) colocalization, and Mendelian randomization. Drug–gene interactions were assessed using the Drug–Gene Interaction Database. The researchers used SBayesRC to estimate SNP-based heritability and construct polygenic scores (PGSs), comparing heritability estimates with those from linkage disequilibrium score regression (LDSC). Prediction was evaluated in independent clinical and biobank cohorts. Simulations using United Kingdom (UK) Biobank data examined differences between heritability methods.

Study results

The analyses identified 118 significant loci in the multi-ancestry analysis and nine additional loci in the European-ancestry-only analysis, giving 127 loci overall; 48 were newly identified. Ancestry-specific results were less precise in groups with smaller samples. The investigators prioritized genes at the associated loci and identified 17 genes among the novel loci that interacted with drugs not approved for AD. Five potential drug targets were highlighted: RRM2B, SYK, KEAP1, CACNA1S, and AXL. SYK and AXL are involved in immune responses to amyloid-beta. Earlier experimental research reported promising findings for RRM2B, SYK, and KEAP1, while interventions involving CACNA1S and AXL did not improve outcomes in the cited human and mouse studies. These targets were not tested as treatments in the present study. The researchers also prioritized QPCT at a previously known locus, connecting it to an amyloid-beta modification targeted by donanemab.

Gene-set analyses identified 72 significant sets, which clustered around lipid binding, immune activation, amyloid catabolism, and cell adhesion. Additional pathways included EphrinB reverse signaling, glutathione metabolism, RAS signaling, sphingolipid metabolism, and synapse endocytosis.

Cell-type analyses showed that genes preferentially expressed in microglia and genes upregulated in microglia from AD cases were enriched for AD genetic associations. Genes underexpressed in three neuronal populations, namely small neuropeptide (Sncg)-expressing neurons, somatostatin (Sst)-expressing inhibitory interneurons, and layer 6 intratelencephalic projection (L6 IT Car3) neurons, also showed enrichment. After adjustment for average differential expression across cell types, the microglial association remained significant, whereas neuronal associations became only nominally significant. Inflammatory microglial states showed the strongest association, but no individual microglial state reached statistical significance. Whether the neuronal expression differences contribute to AD or follow neurodegeneration remains uncertain.

Gene-prioritization analyses suggested that increased astrocytic expression of KIF21B and reduced oligodendrocytic expression of MAF may increase AD risk. Several genes associated with microglial expression were related to phagocytosis, including INPP5D, SYK, BLNK, HAVCR2, PICALM, AXL, FCER1G, and PLCG2. ZKSCAN1, APH1, B, and SPRED2 were among neuronal genes associated with AD. The study also identified genetic overlap involving selected loci and other neurodegenerative conditions, including Parkinson’s disease and amyotrophic lateral sclerosis (ALS). At the UBE2V1–SPATA2 locus, colocalization suggested a shared causal variant in AD and ALS; overlap at other loci did not always involve the same causal variants.

Using SBayesRC, the estimated SNP-based heritability on the liability scale was 16% in European ancestry, compared with 6% using LDSC, assuming an AD population prevalence of 5%. Estimates were 13% in East Asian and 10% in African ancestry, but differences between ancestry groups were not statistically established, and the African-ancestry estimate was not significantly different from zero. A multi-ancestry PGS explained an average of 12.8% of liability variance in European cohorts beyond a baseline model accounting for genetic ancestry and reached 16.9% in one clinical cohort. In European participants aged 65 years or older in UK Biobank, observed AD prevalence was seven times higher in the top PGS decile than in the bottom decile. Prediction was weaker in non-European groups, supporting the need for larger and more diverse datasets. Using proxy cases increased statistical power and yielded 32 additional significant loci, but parental reports could affect diagnostic accuracy. These findings do not establish the clinical utility of PGS-based screening.

Conclusion

The findings expand the known genetic architecture of AD by identifying 127 risk loci, including 48 new loci, and suggest that common genetic variants contribute more to AD susceptibility than earlier estimates indicated. The analyses support a central role for microglia in AD and suggest possible neuronal dysregulation, alongside pathways related to immunity, lipids, and amyloid processing. PGSs improved risk prediction, particularly in European-ancestry cohorts, but require further clinical evaluation. The limited representation of non-European populations reduced the precision of ancestry-specific estimates, making larger, more diverse genetic studies necessary for equitable and reliable genetic risk assessment.