A consensus meta-analysis of genome-wide association studies, published in Nature Genetics on 3 June 2026, reports 91 genome-wide significant risk loci for Alzheimer's disease and related dementias (ADRD), 16 of them new in European-ancestry samples. The analysis pooled 72,721 clinically diagnosed Alzheimer's cases, 55,960 proxy ADRD cases, 614,267 controls and 235,566 proxy controls from 52 studies, and tested 20,045,120 variants after quality control. Of the 91 loci, 56 were associated with risk in clinically diagnosed Alzheimer's cases specifically. A stepwise conditional analysis also turned up 25 independent secondary signals inside 16 already known regions, indicating that some of the familiar loci carry more than one distinct signal.

The point of the exercise was reconciliation rather than novelty alone. Previous large scans by the International Genomics of Alzheimer's Project, the European Alzheimer and Dementia Biobank and the Psychiatric Genomics Consortium Alzheimer's working group overlapped in samples but differed in imputation panels and analytic choices, and some results disagreed. The three groups, with the UK Biobank and FinnGen, agreed on one pipeline and ran everything through it. The work was coordinated by Celine Bellenguez and Jean-Charles Lambert at the Institut Pasteur de Lille, with contributing authors across Europe and the United States. The newly reported regions include EIF4G3, PTPRC, MGAT5, PPP2R3A, ADGRL3, FAM193B, TMEM184A, DOCK4, IPMK, UBFD1, VMAC, VAV1, LRRC25, CEP89, LILRB1/LILRB4 and SRC; the authors note ADGRL3 had already been reported in a recent trans-ethnic scan.

The limits are stated by the authors themselves. A large share of cases are proxy cases, meaning people who reported a parent or sibling with dementia rather than a confirmed Alzheimer's diagnosis, which adds statistical power but blurs Alzheimer's against other dementias. A further 18 loci, 15 of them new, are listed as requiring external validation, including five where significance fell away after conditional analysis. All samples were of European ancestry. A polygenic score built from the ADRD loci excluding APOE tracked mainly with Alzheimer's pathology rather than non-Alzheimer's pathology at autopsy: people in the top decile of the score had roughly twice the odds of Braak neurofibrillary tangle stage above 5 and of moderate to severe neuritic plaque pathology at death compared with those in the median group. The variance the score explained remained low, with Nagelkerke pseudo R-squared of 3.98 percent for Braak stage and 4.37 percent for CERAD score in the ADC/NACC dataset.

The practical distance from this map to the clinic is still large. A locus is a stretch of the genome where common variation tracks with risk across a population, not a named causal gene and not a validated drug target, and the association signals here converge on familiar biology: microglia and immunity, amyloid and tau, lipids, and endosome and lysosome pathways. Nothing in the paper is a test, a treatment or a prediction for any individual. The findings were first posted as a medRxiv preprint on 23 October 2025 and appeared peer reviewed in Nature Genetics 58(6), pages 1214 to 1225.