henhaobio
Sign in

Nature GeneticsGenetic & Genomic MedicinePeer-reviewed journal article

Genome-wide meta-analysis, fine-mapping, and integrative prioritization identify new Alzheimer's disease risk genes

Jeremy SchwartzentruberiD, Sarah CooperiD, Jimmy Z. LiuiD, Inigo Barrio‐HernandeziD, Erica BelloiD, Natsuhiko KumasakaiD, Adam M. H. Young, Robin J.M. FrankliniD, Toby JohnsoniD, Karol EstradaiD, Daniel J. GaffneyiD, Pedro BeltrãoiD, Andrew Roger BassettiD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Genome-wide association studies (GWAS) have discovered numerous genomic loci associated with Alzheimers disease (AD), yet the causal genes and variants remain incompletely identified. We performed an updated genome-wide AD meta-analysis, which identified 37 risk loci, including novel associations near genes CCDC6, TSPAN14, NCK2, and SPRED2. Using three SNP-level fine-mapping methods, we identified 21 SNPs with greater than 50% probability each of being causally involved in AD risk, and others strongly suggested by functional annotation. We followed this with colocalisation analyses across 109 gene expression quantitative trait loci (eQTL) datasets, and prioritization of genes using protein interaction networks and tissue-specific expression. Combining this information into a quantitative score, we find that evidence converges on likely causal genes, including the above four genes, and those at previously discovered AD loci including BIN1, APH1B, PTK2B, PILRA, and CASS4.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (27 Jan 2020) · CC BY 4.0

Meta-analysing UK Biobank family-history GWAS with diagnosed-AD GWAS yields 37 loci, four novel (NCK2, TSPAN14, SPRED2, CCDC6); fine-mapping, colocalisation and network scoring prioritize causal genes.

Genome-wide association studies (GWAS) have mapped many Alzheimer's disease (AD) loci, but the causal variants and genes are clear at only a few, and eQTL integration usually covers only a handful of datasets assumed to be relevant. This work combines a denser diagnosed-AD GWAS (11.5 vs. 7.1 million variants in the one used by earlier meta-analyses) with three fine-mapping methods, colocalisation across 109 eQTL datasets and a quantitative gene score.

Key results

  1. ShownCombining a UK Biobank proxy-case GWAS (family history of AD/dementia) with the Kunkle et al. stage 1 diagnosed-AD GWAS gives 34 genome-wide significant loci, four of them new. Three suggestive loci were also carried forward.

    GWAX: 53,042 affected individuals vs 355,900 controls; Kunkle: 21,982 cases, 41,944 controls; 10,687,126 overlapping variants. 34 loci at p < 5×10^−8 (22 in Kunkle, 8 in Jansen or Marioni); novel near NCK2, TSPAN14, SPRED2, CCDC6; suggestive (p < 5×10^−7) IKZF1, TSPOAP1, TMEM163. GWAX-only PRL and FAM135B hits had p > 0.1 in Kunkle. · Meta-analysis discovers 37 loci associated with…; GWAS on family history of AD; AD meta-analysis

  2. ShownThree fine-mapping approaches (single-causal Bayesian, FINEMAP and annotation-aware PAINTOR) were averaged per SNP, using only FINEMAP and PAINTOR at single-signal loci. This gives 21 variants with >50% mean causal probability, including high-confidence variants at newly reported loci.

    21 variants >50% and 79 more at 10 - 50%. NCK2 rs143080277 >99% (MAF 0.4%); APH1B missense rs117618017 90%; APP-ADAMTS1 rs2830489 72%; SCIMP rs61182333 61%; SPRED2 rs268120 56%; ABCA7 missense rs4147918 55%. PAINTOR enrichment: missense 19.2-fold (1% of SNPs), blood/immune DNase 6.4-fold (16% of SNPs, best likelihood), SpliceAI 9.3-fold. · Fine-mapping identifies credibly causal variants; Annotation-based fine-mapping

  3. ShownEach AD signal, including conditionally independent ones, was colocalised with 109 eQTL datasets, implicating 80 genes at 27 of 36 non-APOE loci. A simple maximum-probability score recovered candidates supported by other evidence slightly better than a microglia/brain/immune-weighted score.

    391 colocalisations with ≥80% shared-causal-variant probability (80 genes, 27 loci), e.g. PTK2B, BIN1, PILRA, CD33, TREM2, FCER1G, TSPAN14, APH1B, ACE. The top 80 genes by maximum coloc score recovered 20 of 40 genes prioritized by other evidence vs 18 for the weighted score (16 shared). · Colocalisation between AD risk loci and gene…; Figure legends

  4. ShownPersonalized PageRank over a combined STRING/IntAct/BioGRID network, seeded with candidate genes from all other loci, ranks AD candidates highly. It also highlights genes below the significance threshold such as LILRB2, ABCA1, SREBF1 and AGRN.

    Network of 18,055 genes, 540,421 edges. Candidates enriched for high network scores (Wilcoxon p = 5×10^−14); top 500 network genes had lower nearby (10 kb) SNP p values (p = 7×10^−7). Nearest gene was top within 200 kb at all four novel loci. Within-locus network ranks correlated with the other four evidence ranks (Spearman rho=0.47; p < 2.2×10^−16). · Network evidence prioritizes genes within and…; Network analysis; Integrative gene prioritization from five lines…

  5. SuggestedSumming distance, microglia/brain expression specificity, maximum colocalisation, network and coding fine-mapping scores recovers most established AD genes and nominates APH1B, FCER1G, ZYX and MS4A6A. Annotation-based fine-mapping separately points to candidate regulatory SNPs at TSPAN14 and CASS4.

    With distance removed, the top gene was the nearest gene at 21 of 37 loci and within 100 kb at 8 more. MS4A6A prioritized though 143 kb from the peak. TSPAN14 risk allele rs1870138-G alters an invariant TAL1 motif position; CASS4 5' UTR rs17462136 (7% probability, GERP 3.46, microglial ATAC peak) creates a TEAD1 motif. · Integrative gene prioritization from five lines…; Discussion; Fine-mapping identifies credibly causal variants; Gene prioritization

Methods, data and reproducibility

BOLT-LMM GWAX of UK Biobank family history of AD, rescaled to log odds and combined with Kunkle et al. stage 1 by inverse-variance fixed-effects meta-analysis. Then GCTA conditioning, WTCCC, FINEMAP and PAINTOR fine-mapping, coloc against 109 eQTL datasets, personalized PageRank network scoring and a summed five-component gene priority score.

  • UK Biobank European-ancestry GWAX: AD cases plus proxy cases (first-degree relative with AD/dementia) vs controls, one of each relative pair kept · human · n = 53,042 affected (3,046 cases, 52,791 proxies); 355,900 controls
  • Kunkle et al. stage 1 GWAS meta-analysis summary statistics for diagnosed late-onset AD · human · n = 21,982 cases; 41,944 controls
  • Primary microglia eQTL study from brain surgery donors, QTLs called with RASQUAL; also used for microglial expression levels · human · n = 93 donors
  • Brain cortex eQTL meta-analysis · human · n = 1433 samples
  • GTEx final release (v8) tissue eQTLs and median TPM expression · human (GTEx) · n = 49 tissues
  • eQTL Catalogue uniformly reprocessed datasets, including brain and sorted or stimulated blood immune cells · human · n = 57 datasets
  • Primary microglia ATAC-seq (Gosselin et al.), realigned and jointly peak-called for functional annotation · human (aligned to GRCh37) · n = 15 datasets
Methods & tools
BOLT-LMM, GCTA COJO conditional analysis, FINEMAP, PAINTOR v3.1, WTCCC single-causal-variant Bayesian fine-mapping (Wakefield approximate Bayes factors), coloc R package, RASQUAL, DeepSEA and SpliceAI, Ensembl VEP, Roadmap Epigenomics DNase and chromatin-state annotations, STRING v10.5, IntAct and BioGRID with igraph personalized PageRank, gProfiler
Software & environment
PAINTOR v3.1, KING v2.0, bwa 0.7.15, FeatureCounts 1.5.3, STRING v10.5, GTEx v8; analyses in GRCh37 coordinates (CrossMap used for GRCh38 datasets and display positions).
Reproducibility gaps
Summary statistics are promised to the NHGRI-EBI GWAS Catalog only upon publication. No container or environment file is described, and versions for several tools (GCTA, FINEMAP, coloc, BOLT-LMM) are not stated.
Limitations (6)
  • APOE was left out of conditional analysis and fine-mapping because its very strong signal would need a near-perfect LD reference match to avoid spurious results. The CLNK locus was dropped from PAINTOR because the program failed on it. · stated by the authors
  • Colocalisation can miss shared effects when a locus has several causal variants, LD differs between studies or power is low. Some likely genes (SPRED2, RIN3, PILRA) colocalised in only one or a few datasets, so a missing colocalisation is weak evidence. · stated by the authors
  • Low-frequency variants were filtered out (below 0.1% for GCTA, below 0.2% for FINEMAP) for two reasons: LD estimated from few allele copies is unreliable, and FINEMAP otherwise picked implausible pairs of weak rare variants. · stated by the authors
  • For APH1B, the authors cannot tell whether risk comes from the T27I protein change or from altered expression. The same allele also raises APH1B expression and is predicted to create a YY1 binding motif. · stated by the authors
  • The network seed genes and the 40-gene benchmark for colocalisation scores were partly drawn from the literature and the study's own evidence streams. These enrichment and score comparisons are therefore not fully independent validations. · noted in this summary
  • The proposed regulatory mechanisms (TAL1 motif loss at rs1870138, TEAD1 motif gain at rs17462136) rest on motif, chromatin and eQTL annotations alone. No reporter, allele-specific or genome-editing experiments were done. · noted in this summary

Read this if you need fine-mapped variant probabilities, eQTL colocalisation results or a ranked candidate-gene list for Alzheimer's disease GWAS loci, or a template for multi-evidence GWAS gene prioritization.

Citations
531
FWCI
56.96
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISJansen '18Marioni '18Kunkle '18Uffelmann '25Brabec '23Rajabli '23
78% semantic matchGenetics
Iris E. Jansen, Jeanne E. Savage, Kyoko Watanabe+48 more·bioRxiv·Feb 22, 2018· 30 citations