Integrated Genomic Analysis of Alzheimer’s Disease and Related Dementias Identifies 16 New Loci Among 91 Risk-Associated Loci
Background: Distinguishing Alzheimer’s Disease Within Dementia Genetic Signals
A genome-wide association study (GWAS) meta-analysis combining 52 studies in populations of European ancestry identified 91 loci associated with the risk of Alzheimer’s disease and related dementias. Sixteen of these loci were newly identified in European-ancestry samples at the time of the analysis. Published in Nature Genetics on June 3, 2026, the study examined whether each signal was also linked to clinically diagnosed Alzheimer’s disease and postmortem brain pathology.
Earlier large-scale studies sometimes produced inconsistent results despite partially overlapping samples because they used different reference datasets for genotype imputation and different analytical methods. Including proxy cases classified by a family history of dementia or biobank data based on diagnostic codes can increase statistical power. However, this makes it harder to distinguish genetic signals for Alzheimer’s disease from those for other dementias.
The researchers combined data from 72,721 people with Alzheimer’s disease, 55,960 proxy cases, 614,267 controls, and 235,566 proxy controls. Proxy cases were participants who reported that a parent or sibling had dementia, rather than patients who had themselves received a diagnosis.
Key Findings: Associations Also Observed in Clinical Diagnoses and Postmortem Pathology
Analysis of approximately 20.05 million variants after quality control identified 91 loci meeting the genome-wide significance threshold. Eighteen additional candidate loci were listed separately as requiring external validation. A locus is a genomic region showing an association signal; this does not by itself establish that a particular gene in that region causes disease.
Of the 91 loci in the primary analysis, 75 remained significant when proxy cases were excluded. In an analysis focused on clinically diagnosed Alzheimer’s disease that excluded large biobanks, 56 remained significant. This does not mean those 56 loci were found only in patients with a clinical diagnosis. The researchers varied the sample composition to assess the robustness of the signals from the primary analysis.
Genes enriched for association signals were linked to high expression in microglia in data from several brain regions. Signals were also concentrated in pathways related to tau, amyloid, lipids, immunity, and endosomal and lysosomal processes. These findings suggest directions for further mechanistic research, but do not demonstrate restoration of intracellular transport function or inhibition of protein aggregation.
The researchers also developed a polygenic score (PGS) excluding the apolipoprotein E (APOE) region. The primary score included 115 variants. They assessed its associations with 11 neuropathological measures in autopsy data from 677 participants in ACT and 5,808 in ADC/NACC. In ADC/NACC, participants in the top 10% of scores had 2.05 times the odds of a higher Braak neurofibrillary tangle stage and 1.96 times the odds of moderate or greater neuritic amyloid plaque pathology compared with participants in the 40th–60th percentile group. These are comparisons between groups based on postmortem pathology; they cannot be interpreted as meaning that an individual’s future probability of developing the disease is doubled.
Implications and Outlook: The Gap Between Genetic Risk Maps and Clinical Prediction
These results add evidence that a score combining genetic signals beyond APOE is associated with Alzheimer’s disease-related pathology. However, the score’s explanatory power was limited. In ADC/NACC, Nagelkerke pseudo-R² values for Braak stage and neuritic plaque measures were 3.98% and 4.37%, respectively. These values do not represent diagnostic accuracy.
Larger samples with well-characterized clinical features and autopsy data are needed to distinguish the effects of individual loci on Alzheimer’s disease pathology from their effects on other neuropathologies. Whether findings centered on people of European ancestry apply to populations with other genetic backgrounds also requires further validation. The current results alone do not establish a test for early diagnosis or treatment selection. The study design and findings are available in the Nature Genetics paper.
Nature Genetics, Published online: 03 June 2026. DOI: 10.1038/s41588-026-02583-1
Summary: Resolving the sample-size constraints and missing heritability barriers that historically limited early-stage diagnostic screening within neurodegenerative pipelines, this updated multi-cohort genome-wide association study (GWAS) meta-analysis maps the genomic landscape of Alzheimer’s disease (AD). Analyzing large-scale datasets from cohorts of European ancestry, the computing platform identifies 91 distinct genetic risk loci, with 56 loci specifically localized within clinically diagnosed AD phenotypes. This extensive genetic discovery isolates molecular cascades governing microglial immunometabolism and endosomal trafficking kinetics, delivering an expanded, non-invasive computational baseline to refine multivariate polygenic risk score (PRS) models, optimize clinical patient stratification, and guide future target therapeutic decision-making.
The practical use of these findings is to select candidates for validation in follow-up research and improve the reproducibility of analyses. For example, researchers can compare the publicly available summary statistics from the primary and sensitivity analyses to examine whether candidate-locus signals change depending on the inclusion of proxy cases or biobank data. Teams studying microglia or intracellular transport pathways could use the results to guide candidate selection, then investigate their actual functions in functional experiments. The paper reports that summary statistics were made available through the GWAS Catalog and NIAGADS, and the score through the PGS Catalog. These resources do not guarantee diagnostic accuracy for individuals or accurate prediction of drug response. Any potential clinical use would require separate evaluation of predictive performance and utility in independent samples and diverse populations.