henhaobio
Sign in

NatureGenetic & Genomic MedicinePeer-reviewed journal article

Impact of Rare and Common Genetic Variation on Cell Type-Specific Gene Expression

Anna S. E. CuomoiD, Eleanor SpenceleyiD, Hope A. TanudisastroiD, Blake Bowen, Albert HenryiD, Hao Lawrence HuangiD, Angli XueiD, Wei ZhouiD, Matthew J. WellandiD, Samantha J. BryeniD, Oscar A DongiD, Arthur S. LeeiD, Jeffrey M. PulliniD, Kristof WingiD, Owen TangiD, Michael Patrick GrayiD, Michael FrankliniD, Michael Harper, Michael A. SilkiD, Katalina BobowikiD, Alexander StuckeyiD, John MarshalliD, Vivian Bakiris, Caitlin UreniD, Rachael A. McCloyiD, , Joseph E. PowelliD

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

Abstract

Understanding the genetic basis of gene expression can shed light on the regulatory mechanisms underlying complex traits and diseases. Single cell-resolved measures of RNA levels and single-cell expression quantitative trait loci (sc-eQTLs) have revealed genetic regulation that drives sub-tissue cell states and types across diverse human tissues. Here, we describe the first phase of TenK10K, the largest-to-date dataset of matched whole-genome sequencing (WGS) and single-cell RNA-sequencing (scRNA-seq). We leverage scRNA-seq data from over 5 million cells across 28 immune cell types, and matched WGS, from 1,925 individuals, which provides power to detect associations between rare and low-frequency genetic variants that have largely been uncharacterised in their impact on cell-specific gene expression. We map the effects of both common and rare variants in a cell type-specific manner using a recently introduced method that increases power by modelling single cells directly rather than relying on aggregated pseudobulk counts. We identify putative common regulatory variants for 83% of all 21,404 genes tested and cumulative rare variant signals for 47% of genes. We explore how genetic effects vary across cell type and state spectra, develop a framework to determine the degree to which sc-eQTLs are cell type-specific, and show that about half of the effects are observed only in one or a few cell types. By integrating our results with functional annotations and disease information, we also further characterise the likely molecular modes of action for many disease-variant associations. Finally, we explore the effects that genetic variants have on gene expression across continuous cell states and functions, and effects that vary cell state abundance directly.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 2 (11 Nov 2025) · CC BY 4.0

Single-cell eQTL mapping in 5.4 million PBMCs from 1,925 individuals with matched WGS finds common-variant eQTLs for 83% and rare-variant signals for 47% of 21,404 genes.

Most eQTL studies have used array genotypes, which capture mainly common variants, or bulk tissue, which hides regulation that acts only in specific cell types. Here population-scale whole-genome sequencing is paired with single-cell expression, so rare and low-frequency variant effects can be tested in each immune cell type, and a new framework scores how specific each eQTL is without requiring the gene to be expressed in every cell type.

Key results

  1. ShownModelling single cells per cell type found common-variant cis eQTLs for most tested genes. The eGene count matched the largest whole-blood bulk study despite a far smaller cohort and was nearly three times that of the previous largest blood sc-eQTL map.

    154,932 eQTLs and 17,674 eGenes (83% of 21,404 tested) at FDR<5%, vs 16,987 eGenes in eQTLGen (over 30,000 individuals vs 1,925) and 6,469 in OneK1K (982 individuals); 4,572 (25.9%) eGenes absent from eQTLGen; all-cell-type pseudobulk found 11,359; SuSie added 24,925 non-primary signals for 7,131 eGenes. · Single-cell eQTL mapping reveals common variant…; Common variant eQTLs

  2. ShownGene-level burden and SKAT tests on WGS variants with MAF<1% detected combined rare-variant effects for nearly half of tested genes, mostly through SKAT. Restricting tests to annotated or open-chromatin variants raised discovery with distance-to-TSS weights, but not when rarer variants were up-weighted.

    34,376 rare-variant signals for 10,005 eGenes (47% of genes tested) at FDR<5%; SKAT only 29,260 (85%), both 4,975 (14.5%), burden only 141 (0.5%); annotation-restricted tests gave 1.2 times as many eGenes on average, more in 23 of 26 cell types (SKAT, dTSS weights), but with Beta(1,25) weights all-cis tests gave 1.09 times as many; cells vs eGenes correlation=0.98. · Mapping the effects of rare variation on…; Rare variant effects on cell type-specific gene…; Rare variant eQTLs

  3. ShownA pairwise five-scenario framework, which does not need a gene to be expressed in all cell types, showed that about half of common-variant eGenes had significant eQTLs in at most a quarter of cell types. Almost no eGene had an eQTL in every cell type.

    Of 17,674 eGenes: 419 (2.4%) expressed in one cell type; 1,733 (9.8%) with an eQTL in only one of 28; 2,530 (about 14%) distinct eQTLs; 10,547 (about 60%) shared concordant; 2,445 (about 14%) discordant; 8,506 significant in at most 7 of 28; only 35 (0.2%) in all 28; rare-variant eGenes: 2,560 (26%) cell type-specific vs 7,445 (74%) shared. · Quantifying the cell type specificity of immune…; Cell type specificity analysis

  4. ShownColocalisation with GWAS for 60 diseases and blood traits linked thousands of genes to traits. Most events were cell type-specific and point to candidate mechanisms such as OSM in NK cells for IBD and IGHG4 in memory B cells for Alzheimer's disease.

    PP H4≥0.8 for 5,287 genes in 60,458 gene-cell type-trait combinations; 7,521 of 16,993 (44%) gene-trait pairs seen in one of 28 cell types vs 12% of all eGenes; 893 (17%) colocalising genes not eQTLGen eGenes; OSM-IBD in NK PP4=0.99 (other cell types <0.06); IGHG4-AD in Bmemory PP4=0.99; GWAS loci explained 16.1% vs 6.6% with pseudobulk. · Colocalisation analysis with disease phenotypes…; Statistical colocalisation

  5. SuggestedInteraction models on NK-cell function scores from scDeepID found eQTL effects that change along continuous cell states. Separately, GeNA neighbourhood tests found variants that shift cell-state abundance, including an LYZ missense variant in monocytes.

    15,535 cell-state eQTLs (7,709 of 16,759 gene-SNP pairs, 46%) across five GOBP functions in 850,833 NK cells; 164 independent csaQTLs (p<5×10^−8) in 8 major cell types, most in monocytes, dendritic and NK cells; four of five OneK1K csaQTLs replicated; LYZ 12:69350234:C:A p=1.2×10^−17, colocalising with monocyte count GWAS (PP4>0.99). · Identification of cell state eQTLs; Cell state abundance QTLs; Cell state eQTL mapping; csaQTL mapping

Methods, data and reproducibility

Cis eQTLs (+/-100kb) were mapped separately in 28 cell types with SAIGE-QTL Poisson mixed models on single-cell counts (single-variant tests for MAF>=1%; burden/SKAT set tests for MAF<1%). Results were then fine-mapped (SuSie), scored for cell type specificity, colocalised with GWAS, and tested for NK cell-state interactions (scDeepID, lme4) and csaQTLs (GeNA).

  • TenK10K phase 1 PBMC scRNA-seq (10x 3' GEX HT v3.1), demultiplexed by donor and annotated into 28 immune cell types · human · n = 5,438,679 cells from 1,925 individuals
  • Matched 30x short-read WGS (Illumina NovaSeq 6000, GRCh38, DRAGMAP and GATK HaplotypeCaller) with SNP and indel calls · human · n = 1,925 individuals; 66,658,242 SNP and indel sites
  • Donor cohorts from Australia: Tasmanian Ophthalmic Biobank (TOB) and BioHEART (cardiac presentations); only the European-ancestry subset was kept · human · n = 2,339 sampled; 1,925 retained (950 TOB, 975 BioHEART)
  • scATAC-seq (Xue et al. 2025): a TOB subset annotated example loci; cell type-matched peaks for 26 cell types annotated rare variants for set tests · human · n = 30,829 nuclei from 8 individuals (17,099 after QC) for example loci
  • Public GWAS summary statistics: 16 diseases and 44 UK Biobank blood count/serum traits, plus 26 Vuckovic et al. blood traits for csaQTL colocalisation · human · n = 16 diseases; 44 and 26 blood traits
  • Comparison eQTL maps: eQTLGen whole blood, OneK1K sc-eQTLs, and sorted-cell studies (BLUEPRINT, CEDAR, Kasela et al, Schmiedel et al) · human · n = eQTLGen over 30,000; OneK1K 982 individuals
  • Public integrated NK-cell scRNA-seq reference with NK subtypes (Rebuffet et al.), used to train scDeepID for label transfer
  • Released summary statistics (100kb and 1Mb cis windows), fine-mapping and colocalisation results · human · 10.5281/zenodo.17474113
Methods & tools
SAIGE-QTL (Poisson mixed model; single-variant, burden and SKAT tests), SuSie (susieR v0.14.2) fine-mapping, coloc v5.2.3 (coloc.abf), scDeepID (multi-task transformer that scores cell functions), GeNA v1.0.0 with CNA v0.1.6 for csaQTLs, lme4 genotype x cell-state interaction models, CellRanger v7.2.0 (STAR); Demuxafy v2.1.0 (vireo, scds, scDblFinder), scPred, scHPL with Azimuth reference; CellTypist v1.6.2, DRAGMAP v1.3.1, GATK4 HaplotypeCaller (DRAGEN mode), hail, PromoterAI, SpliceAI, AlphaGenome, Pangolin, Scanpy v1.8.2, harmonypy v0.0.10, fgsea v1.32.2, metafor v4.0.0, MACS3 v3.0.3 via SnapATAC2 v2.8.0; Cell Ranger ATAC v2.1.0
Data & accessions
gs://gcp-public-data--broad-references/hg38/v0/dragen_reference/https://gymreklab.com/science/2023/09/08/Margoliash-et-al-paper.html
Software & environment
CellRanger v7.2.0; Scanpy v1.8.2; Demuxafy 2.1.0; CellTypist 1.6.2; DRAGMAP v1.3.1; GATK v4.2.6.1; somalier v0.2.15; susieR 0.14.2; coloc 5.2.3; metafor 4.0.0; GeNA 1.0.0; CNA 0.1.6; harmonypy 0.0.10; fgsea 1.32.2; SnapATAC2 2.8.0
Reproducibility gaps
Individual-level genotypes and scRNA-seq counts are going to EGA but the upload is described as underway, so no EGA accession is given. Full SAIGE-QTL parameters are given only in the GitHub code.
Limitations (6)
  • The specificity threshold is empirical and conservative, and no independent dataset of similar size was available to check for winner's curse. The LD r 0.5 cutoff separating shared from distinct eQTLs is heuristic, and pairs in lower LD may inflate discordant (scenario 5) counts. · stated by the authors
  • Rare-variant tests lacked the power to test annotation classes separately. The matched scATAC-seq was too underpowered for an analysis of cell type-specific open chromatin alone, so it was pooled with annotations that are not cell type-specific, which may dilute the signal. · stated by the authors
  • Cell-state eQTLs used a simple linear mixed model on NK eQTLs already detected, so effects present only in specific states may be missed. csaQTLs overlapped little with cell count GWAS (low power, broad GWAS categories), and variants with MAF below 0.05 were dropped for poor calibration. · stated by the authors
  • SuSie was run on Z scores from the Poisson SAIGE-QTL model, which violates SuSie's linear assumption. Association testing kept only European-ancestry donors, and the authors call for more diverse ancestries, disease states and tissues. · stated by the authors
  • SAIGE-QTL Type I error was checked by shuffling individual IDs only in the most abundant cell type (CD4TCM). Calibration is not shown directly for rare cell types with far fewer cells and donors, such as ILC. · noted in this summary
  • Several numbers conflict: the colocalisation gain over pseudobulk is 5.2-fold in Results but 2.6 times in the Discussion, the count of colocalising genes is given as both 5,287 and 5,288, and the Methods list more disease GWAS than the stated 16. · noted in this summary

Read this if you build or use eQTL maps resolved by immune cell type, want WGS-based estimates of rare-variant gene regulation, or need cell type-specific colocalisation to prioritise GWAS genes.

Citations
28
FWCI
86.39
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
THISXue '24Li '22Henry '25Wang '22Kaptijn '25Study '18
74% semantic matchGenomics
Angli Xue, Seyhan Yazar, José Alquicira-Hernández+7 more·bioRxiv·May 5, 2024· 1 citations