遗传变异影响免疫细胞单细胞染色质可及性与疾病风险
Genetic variants shape single-cell chromatin accessibility and disease risk
研究团队对922名供体的350万个外周血单个核细胞核开展scATAC-seq,在28种免疫细胞中鉴定出243,225个染色质可及性数量性状位点(caQTL)。超过一半的caQTL具有细胞类型特异性,caQTL与eQTL整合得到7万余个共定位信号,并识别出28,834个候选顺式调控元件。
Main
Understanding how genetic variation influences complex traits and disease risk remains a major challenge in human genomics. Although genome-wide association studies (GWAS) have identified tens of thousands of risk loci, most lie in non-coding regions6, complicating interpretation. One common approach to bridge this gap is to map molecular QTLs (molQTLs), identifying genetic variants that affect intermediate molecular phenotypes such as gene expression or chromatin accessibility, and then test for colocalization with GWAS loci to infer potential regulatory mechanisms7. Most molQTL studies have focused on eQTLs derived from bulk RNA sequencing (RNA-seq), which have low sensitivity for cell-type-specific effects8,9. Recently, well-powered single-cell RNA-seq (scRNA-seq) studies have shown that genetic regulation of gene expression varies across cell types and states10,11,12,13,14. However, many GWAS signals still lack corresponding regulatory effects, a gap often termed the ‘missing regulation’ problem15,16,17.
caQTLs offer a complementary modality that directly reflects the physical state of regulatory elements and captures non-coding variant effects on enhancer and promoter activity. Recent studies have suggested that caQTLs colocalize with more GWAS loci than eQTLs15,16,18,19, particularly at distal elements, but current caQTL maps remain limited in scale or resolution20,21, or rely on genotypes imputed from the chromatin data22. Many caQTLs also lack corresponding eQTLs for reasons that remain unclear16,23, and dynamic, cell-state-dependent regulatory effects are often overlooked11,24.
Here we present a large-scale atlas of genetic effects on single-cell chromatin accessibility in peripheral blood mononuclear cells (PBMCs). We profiled over 4.6 million single-cell assay for transposase-accessible chromatin with sequencing (scATAC-seq) and 95,728 multiome (RNA + ATAC) nuclei across 1,042 individuals. After quality control, we analysed 3.5 million high-quality nuclei from 922 donors, with matched whole-genome sequencing data. We identified 243,225 caQTLs across 28 immune cell types, with more than half showing cell-type-specific effects. Integrating caQTL and eQTL data revealed over 70,000 colocalized signals, including 28,834 candidate cis-regulatory elements (cCREs) supported by Mendelian randomization. We further detected thousands of cell-state-dependent caQTLs and GWAS colocalizations missed by eQTLs alone, and found that combining caQTL and eQTL data improved gene regulatory networks (GRNs) inference over conventional multiome-based methods.
Together with accompanying articles exploring the broader effects of common and rare genetic variation5 and tandem repeat variants25 on gene expression, this work provides a single-cell view of genetic regulation of chromatin accessibility in human immune cells, filling gaps left by eQTL-only approaches to reveal how non-coding variation shapes immune function and disease risk.
The TenK10K multiome cohort
We generated the TenK10K phase I multiome dataset by profiling chromatin accessibility (scATAC-seq) in 4.6 million nuclei from 952 donors, as well as multiome (RNA + ATAC) from 90 donors (Methods and Supplementary Tables 1–6). We obtained the 952 donors from the TenK10K pilot cohort, known as OneK1K or the Tasmanian Ophthalmic Biobank10 (Methods). These data were complemented by 5.4 million PBMC scRNA-seq profiles from 1,925 donors5, including 932 that overlapped with matched scATAC-seq profiles (Methods and Extended Data Fig. 1a). Cell types were annotated via label transfer against a CITE-seq reference (Methods), yielding 28 immune cell types consistent with the scRNA-seq dataset (Extended Data Figs. 1b and 2 and Supplementary Tables 4–6). After quality controls, 3,472,552 scATAC-seq nuclei from 922 donors were retained for discovery analyses, with an additional 60 donors used for replication (Methods, Extended Data Fig. 1c and Supplementary Fig. 1). On average, each donor contributed 3,766 nuclei; CD14+ monocytes (CD14Mono) were the most abundant cell type, whereas proliferating CD8+ T cells (CD8Proliferating) were the rarest (Extended Data Fig. 1d).
Chromatin peaks were called independently in each cell type and merged into a unified list of 440,996 peaks (Methods and Extended Data Fig. 1e). Peak widths ranged from 201 to 5,545 bp (median of 499 bp). Most peaks were distal to the promoter, with only 4.0% within 200 bp of a transcription start site (TSS; Extended Data Fig. 1f). Compared with existing chromatin accessibility datasets, only 44,015 peaks (9.98%) lacked overlap with known peaks or cCREs (Methods and Supplementary Table 7).
To assess the biological specificity of these peaks, we clustered cell-type-enriched features and observed grouping by haematopoietic lineage (Extended Data Fig. 2c,d). Transcription factor motif enrichment analyses supported known lineage functions: for example, monocyte-specific peaks were enriched for CEBP family motifs (CEBPB and CEBPD), consistent with their established roles in monocyte differentiation and activation (Extended Data Fig. 2e). Together, these results confirm that chromatin accessibility profiles capture cell-type-specific regulatory signals across the immune landscape.
Mapping common variant single-cell caQTLs
We mapped caQTLs across 28 immune cell types in 922 donors, testing 440,996 peaks against common variants (minor allele frequency (MAF) ≥ 5%; approximately 9 million SNPs and 3.4 million indels) within 1 Mb of the peak centre (Methods). Using permutation-based correction (q < 0.05), we identified 243,225 significant caQTLs, corresponding to 100,483 unique peaks (caPeaks) with at least one association (Fig. 1a). caQTL discovery scaled with sample size per cell type, and wider peaks were more likely to be caPeaks (Fig. 1b and Supplementary Fig. 2). Top caQTL variants were typically proximal to the regulated peak (median distance of 7.7 kb; 88.1% within ±100 kb), although 2.7% lay more than 500 kb away (Supplementary Fig. 3). Expanding the testing window to ±1 Mb in CD14Mono identified 4,291 additional peaks, approximately 85% with more than tenfold increased statistical support (Supplementary Fig. 4), underscoring the importance of distal regulatory effects26.
a, Number of significant caQTLs across 28 immune cell types. ASDC, AXL+SIGLEC6+ dendritic cells; CD4CTL, CD4+ cytotoxic T cells; dnT, double-negative T cells; gdT, γδ T cells; HSPC, haematopoietic stem and progenitor cells; ILC, innate lymphoid cells; MAIT, mucosal-associated invariant T cells; pDC, plasmacytoid dendritic cells. b, Relationship between the number of caQTL signals and nuclei per donor across cell types. c, Cell-type specificity of CD4+ T cell caQTLs across five scenarios (pie charts). The bar plot (right) shows caPeaks by the number of cell types sharing the effect (scenarios 3–5). The brown-framed pie chart (CD4TCM caQTLs in CD14Mono) is compared further in d. d, Comparison of caQTL significance (−log10(q value)) between CD4TCM and CD14Mono. The yellow diamond highlights the caPeak shown in e. e, Violin and boxplot of the caQTL (chr. 13: 24670806–24672096) at its top variants in CD4TCM and CD14Mono, showing z-score-normalized pseudobulk values per genotype group (n = 922 individuals; left). Box plots show the median (centre line), the first and third quartiles (box bounds), and whiskers extending to the most extreme values within 1.5× the interquartile range of the box bounds; the same applies to all other boxplots. A locus plot of −log10(nominal P values) against genomic position, with the top variants circled in black, the peak as a blue vertical line and the gene track below (right). f, ENCODE v3 cCRE annotation across all peaks, top caQTL variants and caPeaks (left), and caPeaks stratified by peak width (right). pELS, proximal enhancer-like signature; pLS, promoter-like signature. g, Spearman’s correlation between observed and ChromBPNet-predicted effect sizes, grouped by top or fine-mapped variants and stratified by variant distance to the peak. Each dot is one cell type (n = 25 per box). h, Observed effect size (slope) versus predicted log2 fold change for fine-mapped variants within tested peaks (n = 14,213 variants with posterior inclusion probability (PIP) ≥ 0.5 within 1,000 bp of the peak). i–k, Rare variant 12:9650869:G:A. The dashed lines mark the variant position. Genomic context relative to CLEC2D (i). Observed and predicted scATAC-seq signal (±500 bp) in the cell type with the largest ChromBPNet-predicted allelic difference (j); the ENCODE SCREEN distal enhancer-like sequence (dELS) is a yellow bar. ChromBPNet contribution scores for the reference (Ref) and alternative (Alt) alleles (±50 bp); the highest-scoring transcription factor motif overlapping the variant (reference sequence) is indicated (k).
To quantify cell-type specificity, we adopted a five-category model5 comparing caQTL signals across cell-type pairs (Methods, Fig. 1c, Extended Data Fig. 3 and Supplementary Fig. 5), as existing methods, such as mashR27, are restricted to peaks detectable in all cell types, thus, only a subset of peaks can be tested (Supplementary Fig. 6). Categories ranged from cell-type-exclusive chromatin accessibility (scenario 1) to shared caQTLs with concordant or discordant allelic effects (scenarios 4 and 5), with intermediate scenarios capturing cases where the same peak was regulated by distinct variants in different cell types (scenario 3). For example, at one peak (chromosome 13 (chr. 13): 24670806–24672096), independent top variants were identified in CD4+ central memory T (CD4TCM) cells and classical monocytes (CD14Mono) (linkage disequilibrium r2 = 4.5 × 10−7) and conditional analysis in CD14Mono revealed a secondary signal in near-perfect linkage disequilibrium with the CD4TCM lead variant (linkage disequilibrium r2 = 0.94), indicating that cell-type-specific effects can arise from distinct causal variants at the same locus (Fig. 1d,e).
Functional annotation using ENCODE v3 showed that caPeaks were enriched for promoter and enhancer elements relative to background peaks (64.2% overlapping distal enhancers, 22.8% proximal enhancers; chi-squared P < 2.2 × 10−16), with broader peaks more likely to overlap promoters and narrower peaks mapping to CTCF sites (Fig. 1f). We also identified 11,585 caPeaks in gene deserts (more than 500 kb without annotated genes), enriched for distal enhancers (72.1% versus 64.2% genome-wide, P < 2.2 × 10–16), suggesting long-range regulatory roles. Fine-mapping using SuSiE28 identified 242,838 credible sets across 213,264 caQTLs, with 88.56% harbouring a single credible set (defined at posterior inclusion probability (PIP) > 0.9; Supplementary Fig. 7). Compared with the top variants, fine-mapped variants are on average slightly closer to the peaks (Supplementary Fig. 3). Integrating these fine-mapped results helps to explain cell-type-specific signals (scenario 3; Fig. 1e). For the aforementioned peak (chr. 13: 24670806–24672096), we observed one credible set in CD4TCM cells but two in CD14Mono cells(Supplementary Fig. 8): a shared set proximal to the caPeak present in both cell types, and a second, intergenic set detected only in CD14Mono cells.
We assessed replication in three ways (Methods and Extended Data Fig. 4). First, compared with two published PBMC caQTL studies20,21 (n = 13 and 48), we observed high replication (63–95%) and effect-direction concordance (78–90%) despite differences in cell annotation and association model, while detecting substantially more caPeaks (100,483 versus 10,884 combined). Second, internal replication using multiome data from 60 donors yielded replication rates of 82.5–88.1% in the two most abundant cell types, with lower rates in rarer cell types probably reflecting reduced nuclei per donor. Third, using within-individual allele-specific chromatin accessibility counts (mixQTL29) as an orthogonal validation, 94.0–100% of testable lead caQTLs were sign-concordant with allele-specific effects, with effect sizes strongly correlated between methods (Spearman ρ = 0.81–0.93).
Finally, we evaluated caQTL effect sizes using ChromBPNet30, a deep learning model trained on scATAC-seq data to predict regulatory variant effects (Methods). Predicted and observed effect sizes were strongly correlated for fine-mapped variants within peak regions (Spearman’s ρ = 0.75; Fig. 1g), with accuracy decreasing with distance (Fig. 1h and Supplementary Figs. 9 and 10). Fine-mapped variants 1–5 kb from peaks retained high accuracy (ρ = 0.7) compared with top variants (ρ = 0.4), suggesting that distal top variants are more likely to represent tagging single-nucleotide polymorphisms (SNPs) lacking direct regulatory function, whereas fine-mapped variants at similar distances remain enriched for functional effects. By contrast, novel peaks exhibited consistently lower ChromBPNet prediction accuracy and greater donor-level sparsity than known peaks (Supplementary Figs. 11 and 12). These findings highlight the ability of chromatin accessibility data to predict the regulatory impact of genetic variants, particularly those within functional regions.
Rare-variant effects on accessibility
Although several studies have identified common variants associated with chromatin accessibility, rare variant contributions remain largely unexplored. Using SAIGE-QTL31 (Methods and Supplementary Fig. 13), we tested rare variants (MAF < 5%) within ±100 kb of each peak centre and identified 216,159 rare caQTLs across 26 cell types (Extended Data Fig. 5a). To rule out linkage with nearby common variants, we conditioned on significant common-variant genotypes. A total of 62,866 rare caQTLs (38,161 unique peaks) remained significant, of which 24.1% (15,154 of 62,866) overlapped common-variant caPeaks in the same cell type (Extended Data Fig. 5a and Supplementary Table 8). The variants closer to the common-variant MAF threshold (1–5%) were more likely to lose significance after conditioning, consistent with stronger linkage disequilibrium with common variants (Extended Data Fig. 5b), and restricting to MAF < 1% left 8,450 significant rare caQTLs.
To functionally annotate rare caQTLs, we used ChromBPNet to predict allele-specific chromatin accessibility and test whether associations could be explained by transcription factor-binding motif disruption (Methods). One example, chr. 12:9650869:G:A (12:9650869:G:A), lies 14.1 kb upstream of CLEC2D within a distal enhancer (ENCODE cCRE; H3K27Ac-marked across cell lines) with no significant common caQTLs at this peak (Fig. 1i,k). The variant is associated with reduced CLEC2D expression across multiple cell types, most strongly in CD4+ T effector memory (CD4TEM) cells (Extended Data Fig. 5c). CLEC2D encodes LLT1, a ligand for the natural killer (NK) cell inhibitory receptor CD161 that suppresses NK cell cytotoxicity32,33 and stimulates IFNγ release34. ChromBPNet predicts that the reference allele (G) generates a strong accessibility signal at this peak (chr. 12: 9650649–9651305), whereas the alternative allele (A) markedly attenuates it. Further contribution score analysis indicates that this is driven by disruption of a canonical ELF1-binding motif (MA0473.4) centred on the variant (Fig. 1k and Supplementary Table 9). Together, this suggests that the rare G > A variant reduces accessibility at this distal element by disrupting ELF1 binding, plausibly contributing to decreased CLEC2D expression.
Two additional examples follow the same pattern. At chr. 19: 16289808 (C > T; allele count = 1), 34.8 kb upstream of KLF2 in an ENCODE cCRE distal enhancer, the T allele is associated with outlier under-expression of KLF2 across multiple cell types and is predicted by ChromBPNet to reduce accessibility at the corresponding peak (chr. 19: 16289593–16290014) via weakened CREB1 binding. CREB1 is an annotated upstream regulator of KLF2 in chromatin immunoprecipitation followed by sequencing compendia (ChEA)35, although not resolved to this specific element, consistent with but not direct evidence for altered CREB1-mediated enhancer activity in this single carrier (Supplementary Table 9 and Supplementary Fig. 14). At chr. 21: 44932815 (C > T), 735 bp upstream of the ITGB2 TSS, the T allele is associated with outlier under-expression of ITGB2 across multiple cell types (most strongly in NK cells) and is similarly predicted to reduce accessibility at the corresponding peak (Supplementary Fig. 15).
Together, these examples illustrate how rare variants can act through transcription factor motif disruption to shape chromatin accessibility and, in turn, cell-type-specific gene expression relevant to immune function, which is likely to be missed in analyses restricted to common variants.
Characterizing cell-type-resolved cCREs
To link chromatin peaks to genes, we performed colocalization between TenK10K caQTLs and eQTLs across approximately 3.7 million peak–gene–cell-type pairs within ±1-Mb windows using coloc36 (Methods), identifying 76,754 colocalization events (posterior probability of a shared causal variant (PPH4) > 0.8) involving 28,834 peaks and 11,204 genes (Fig. 2a and Supplementary Table 10). We define these peaks as cCREs, enabling peak–gene linkage without paired multiome data.
a, Cell-type-specific and shared cCREs identified by colocalizing caQTLs and eQTLs. The number of peak–gene pairs colocalized in or shared by multiple cell types (right). The number of colocalized peak–gene pairs (PPH4 > 0.8) per cell type, grouped by lineage in the same colour scheme (left). b, Combined violin and boxplots of the top variants for the caQTL (chr. 10: 45592479–45592785) and eQTL (MARCH8) in NK cells and cDC2, showing standardized pseudobulk chromatin accessibility and gene expression across the three genotype groups (n = 922 individuals). Slope estimates and P values from a two-sided Student’s t-test are annotated. c, Locus plot of the caQTL. The −log10(nominal P values) from a two-sided Student’s t-test against genomic position, with the two top variants circled in black and the peak as a blue vertical line (top). Gene track, with forward-strand genes in blue and reverse-strand genes in green (middle). Zoomed-in MARCH8 locus and chromatin peak map across five cell types (bottom). d, Zoomed-in gene and peak tracks. Each dash in the peak track is an open chromatin peak. The sky blue vertical line marks the caPeak (chr. 10: 45592479–45592785).
Colocalization events varied by cell type, with classical monocytes (CD14Mono) showing the most events (12,940), including 8,402 CD14Mono-specific pairs and 1,743 shared with non-classical monocytes (CD16Mono) (Fig. 2a). For example, the inflammatory bowel disease (IBD) risk gene IRGM showed a regulatory effect via promoter accessibility (chr. 5: 150846177–150847636) in CD8TEM cells but not CD8+ naive T cells (CD8Naive; Supplementary Fig. 16). This cell-state-dependent pattern is consistent with the known role of IRGM as an IBD risk locus37 and illustrates how integrating caQTL and eQTL data can resolve context-specific regulatory activity.
Although coloc infers whether two phenotypes share a causal variant, it does not estimate effect direction or magnitude. To address this, we applied summary-based Mendelian randomization (SMR)38, using chromatin accessibility as the exposure and gene expression as the outcome (Methods). Testing 3,667,924 peak–gene–cell-type combinations (matching those tested in coloc), we identified 81,550 significant associations at PSMR < 1.363 × 10−8 (Bonferroni correction threshold; Supplementary Table 11). SMR significance was strongly concordant with coloc support: colocalized pairs with significant SMR results had a much higher median PPH4 than non-significant pairs (0.794 versus 0.023), and HEIDI test results aligned with coloc’s PPH3 (posterior probability of independent causal variants), consistently flagging distinct causal variants (Extended Data Fig. 6 and Supplementary Fig. 17). Conversely, 29,517 of the 81,550 SMR-significant pairs showed low colocalization support (PPH4 < 0.01) but strong evidence for distinct causal variants (mean PPH3 = 0.98), of which 86.16% (25,433) also showed significant heterogeneity (PHEIDI < 0.05).
Of 76,730 colocalization events also tested by SMR, 39,045 (50.89%) were also SMR significant (15,980 cCREs, 6,205 genes), strengthening confidence in their causal regulatory role. A subset of 749 events (454 unique peak–gene pairs) achieved PPH4 = 1 (indicating a fully shared causal variant) and were frequently observed across multiple related cell types (Supplementary Fig. 18). We further identified 516 bidirectional cCREs, that is, peaks with opposite regulatory effects on different genes, in different cell types, or both (Methods and Supplementary Note 1). These directionality patterns illustrate the complex, context-dependent logic of genetic regulation across the immune system (for example, bivalent chromatin39). We highlight a locus where the chromatin peak chr. 10: 45592479–45592785 shows opposite effects on the gene MARCH8: negative in NK cells (bSMR = −0.087, PSMR = 2.0 × 10−22) and positive in conventional dendritic cell type 2 (cDC2; bSMR = 0.277, PSMR = 6.8 × 10−9; Fig. 2b–d). In NK cells, the top caQTL variant (10:45592613:T:C), an intronic variant of MARCH8 within the peak, shows an accessibility-increasing minor allele (Fig. 2b). In cDC2, the top variant is 10:45429520:C:A, an intronic variant of the upstream gene ALOX5. The two top variants are in high linkage disequilibrium (r2 = 0.73) and show concordant effects on chromatin accessibility but opposite effects on gene expression, indicating distinct regulatory mechanisms for the same cCRE across cell types. This peak–gene pair also showed strong coloc support in both cell types (PPH4 = 0.997 in NK cells, 0.946 in cDC2), consistent with a shared underlying causal variant (Fig. 2c and Supplementary Fig. 19). This example illustrates how combining caQTL and eQTL summary statistics can not only identify regulatory elements but also resolve their effect size and direction across cellular contexts.
Resolving GWAS loci missed by eQTLs
GWAS have identified tens of thousands of loci for complex traits, yet the regulatory mechanisms underlying most remain unresolved. We reasoned that some loci act through chromatin accessibility rather than steady-state expression, particularly where the causal variant lies far from the target gene. Colocalizing GWAS signals for 16 diseases and 44 blood traits with caQTLs and eQTLs separately (Methods), we found that caQTLs colocalized with 9.8% more loci on average (range 4.5–22.6%; PPH4 > 0.8), a pattern reproduced by SMR (Fig. 3a,b and Supplementary Table 12). Adding bulk eQTLs from GTEx and eQTLGen raised colocalization by 2.2–21.4%, and adding TenK10K eQTLs and caQTLs to eQTLGen nearly doubled it (Extended Data Fig. 7a,b), indicating that much of the cell-type-specific regulation underlying GWAS associations is invisible to bulk analysis. Consistent with this, 27.7% of colocalizing loci on average were detected in only a single cell type (for example, 14 of 72 caQTL colocalization events in IBD), with detection dependent on cell-type abundance (Extended Data Fig. 7c,d).
a, Proportion of GWAS loci colocalizing with caQTLs or eQTLs across 16 diseases and 44 blood traits using coloc (top) or SMR (bottom). Trait categories are colour coded for complex diseases, blood serum markers and blood cell counts. b, Distribution of GWAS loci by colocalization pattern. The stacked bars indicate the proportion of loci colocalizing with caQTLs only, eQTLs only, both or neither. SLE, systemic lupus erythematosus; T1DM, type 1 diabetes mellitus. c, Locus plot of ADCY7–BRD7–TENT4B showing colocalization signals detected only by multi-variant coloc, reflecting multiple underlying variants shared with the IBD GWAS. Each colocalized eQTL–GWAS credible set pair is a different colour; credible sets in black indicate no colocalizing pair. CS, credible set. d, Aggregated percentage of IBD GWAS loci with colocalization evidence across QTL datasets and different analytical settings. e, Fine-mapping with mvSuSiE in CD14Mono identifies a credible set (L1) affecting IBD risk, gene expression and chromatin accessibility, with PIP = 1 for the causal SNP. SNPs are ordered by sequential rather than actual chromosome position for visualization. f, Estimated minor allele effect sizes of the causal variant on chromatin accessibility, gene expression and IBD risk. The frame highlights the peak containing the L1 SNP.
To ask why caQTLs capture signals that eQTLs do not, we first tested whether they reach more distal elements than the ±100-kb cis-window commonly used for eQTL mapping. At the IBD-associated ETS2 locus, the lead GWAS variant (21:39094644:G:A) lies approximately 269 kb downstream of ETS2 within a gene-desert enhancer, outside a ±100-kb window and therefore untestable for eQTL colocalization. The same variant fell within a caPeak and colocalized strongly with caQTLs in CD14 monocytes (PPH4 = 0.995), CD16 monocytes and cDC2, with the risk allele reducing accessibility, consistent with previous reports at this locus40 (Extended Data Fig. 8a). Widening the eQTL window to ±1 Mb recovered the eQTL colocalization (PPH4 = 0.996, PSMR = 4.2 × 10−19, versus PPH4 ≈ 0 at ±100 kb), one of 35,382 gene–cell type–trait combinations affected by window choice. Window size therefore explains individual loci but not the aggregate gap: caQTLs retained a 15.9% (average across all 60 traits) colocalization advantage even against ±1 Mb eQTLs (versus 16.7% at ±100 kb; Supplementary Tables 12 and 13), and colocalizing and non-colocalizing caQTLs had comparable effect sizes (Supplementary Fig. 20), arguing against a simple power explanation.
We therefore tested two further explanations (Supplementary Note 2). Allelic heterogeneity accounted for a substantial share: 35.6% of caQTL-specific signals showed strong evidence of distinct causal variants (PPH3 > 0.8), and SuSiE-based colocalization41, which models multiple credible sets per locus, raised colocalizing features from 47.2% to 53.0% for IBD. At ADCY7, an IBD risk locus with six credible sets where single-variant coloc returned H3 ≈ 1, multi-variant coloc resolved four cell-type-specific associations, each matched to a different credible set: ADCY7 and TENT4B in CD4TCM cells, BRD7 and NOD2 in CD16 monocytes (Fig. 3c). Enhancer priming, tested using CD4 naive cells as an unstimulated proxy, explained only approximately 10% of caQTL-only signals (approximately 16% for promoters; Supplementary Fig. 21), leaving most of the gap attributable to mechanisms that we did not test directly, such as post-transcriptional buffering. The converse case, 5.2% of loci colocalizing with eQTLs but not caQTLs, is consistent with reduced caQTL power (Supplementary Fig. 22 and Supplementary Note 2).
Together, combining cell-type-resolved caQTLs and eQTLs with wider cis-windows and multi-causal-variant modelling raised the proportion of IBD loci with colocalization evidence from 16% to 65% relative to bulk eQTLs alone (Fig. 3d and Supplementary Table 14), indicating that much of the missing regulation reflects data resolution and modelling rather than absent regulatory effects.
Multiomic fine-mapping of disease loci
To better pinpoint causal variants acting on disease risk, we applied a multi-trait fine-mapping framework42 to summary statistics from GWAS, caQTL and eQTL analyses (Methods). We selected focal genes by colocalization with GWAS (PPH4 > 0.8), caPeaks by colocalization with eQTL (PPH4 > 0.8) and set focal regions to ±1 Mb of the gene body. We applied this to IBD, one of the best-powered autoimmune GWAS, across all immune cell types.
Across 21 immune cell types, we identified 1,085 credible set–phenotype pairs, of which 882 were single-variant credible sets, among the 916 matching focal gene–cell-type pairs (Supplementary Table 15 and Supplementary Fig. 23). For example, 11:1852991:C:T, 2 kb upstream of the IBD-associated gene LSP1, was fine-mapped as a causal variant for IBD risk. In CD14 monocytes, this variant was strongly associated with both LSP1 expression (PeQTL = 2.21 × 10−131) and accessibility at chr. 11: 1852420–1853362 (PcaQTL = 9.77 × 10−47; Fig. 3e), which colocalized completely (PPH4 = 1). The minor allele increased both accessibility and expression, consistent with a positive regulatory relationship and with the peak–gene SMR estimate (bSMR = 0.47, PSMR = 2.38 × 10−38). The framework also recovered the ETS2 lead eQTL (21:39094644:G:A) in a CD14Mono credible set (PIP = 0.812; Supplementary Fig. 24), matching previously prioritized functional results40 that GWAS-only fine-mapping could not resolve because of high linkage disequilibrium among candidate SNPs. By jointly modelling chromatin accessibility, gene expression and disease risk, this approach resolves the coordinated regulatory effects through which individual variants act, including at loci where high linkage disequilibrium compromises GWAS-only fine-mapping.
caQTLs vary along cell-state trajectory
To test whether genetic effects on chromatin accessibility vary with cell state, we inferred pseudotime across the CD4+ T cell compartment and tested caQTL-pseudotime interactions at single-cell resolution with CASTIE31,43 (Methods). We identified 209,189 significant interactions (false discovery rate < 0.05), of which 57,308 (27.3%) lacked a detectable main effect in the primary caQTL mapping. One such peak (chr. 16: 50326863–50327752) falls within the ADCY7 locus, where multi-variant colocalization resolved a CD4TCM-specific IBD association (Fig. 3c). Here the effect of SNP 16:50336792:G:C on accessibility varied along the trajectory in a genotype-dependent manner: G/G carriers showed a positive association between pseudotime and accessibility, whereas alternative genotypes showed the opposite trend (Fig. 4a–c). This caQTL was positively associated with ADCY7 expression in CD4TCM cells by SMR (bSMR = 0.218, PSMR = 2.88 × 10−101). ADCY7 encodes a membrane-bound adenylate cyclase that catalyses cAMP synthesis and has been shown to regulate the helper 1 T (TH1)/TH2 cytokine balance in primary human CD4+ T cells, with loss of function shifting cytokines towards a TH2 phenotype44. Using Reactome activity inferred by scDeepID (Methods and Code availability) as the interaction term, we identified four pathways, including G protein-mediated events, whose activity scores correlated positively with peak accessibility, with correlation strength differing across genotype groups (Fig. 4d,e). These observations are consistent with cell-state-dependent, allele-specific accessibility at this peak modulating ADCY7 expression along the CD4+ T cell differentiation trajectory. Together, these results highlight regulatory effects that emerge only within dynamic cellular contexts and help to explain additional layers of genetic regulation not captured by conventional main-effect analyses.
a, Cell-state-dependent caQTL (chr. 16: 50326863–50327752) in CD4TCM cells. Chromatin accessibility at this peak, stratified by the 16:50336792:G:C genotype and pseudotime group, shows decreased effect size in the later pseudotime group. The numbers above each violin indicate donors per group. CPM, counts per million. b, Inferred pseudotime within the CD4+ T cell population (n = 493,520), with normalized chromatin accessibility at chr. 16: 50326863–50327752 (SNP: 16:50336792:G:C) on the y axis and points coloured by genotype group. c, Locus plot integrating the ADCY7 eQTL, the caQTL for chr. 16: 50326863–50327752 in CD4TCM cells, and IBD GWAS signals. The first four panels show caQTL and eQTL associations with their fine-mapped results in CD4TCM cells, as indicated in the subtitles. The vertical brown line marks the caPeak position. Point colour denotes linkage disequilibrium with the lead variant (purple). The green square and purple triangle indicate two credible sets. The fifth panel shows the SMR association with IBD GWAS (PSMR). The y axes show the PIP from SuSiE in fine-mapping panels and −log10 nominal P values in QTL and GWAS panels. Gene annotations are shown below. d, UMAP of the CD4+ T cell population and pathway activity scores for G protein-mediated events. e, Correlation between peak accessibility and G protein-mediated event pathway activity score at the pseudobulk level. Pearson’s correlation coefficient is estimated for all donors and for the three genotype groups separately.
QTL-guided peak–gene and GRN inference
Dissecting how genetic variants act within regulatory networks requires robust inference of GRNs from multiome data. Existing methods45,46 rely largely on co-expression or co-accessibility and do not incorporate genetic variation, which can serve as a causal anchor for network inference. We therefore used caQTL and eQTL summary statistics as priors for peak–gene and GRN inference with the graph-based GLUE framework47, which models peak–gene relationships from multiome data and accepts prior knowledge as regulatory scores (Methods). In CD14Mono cells, eQTL guidance raised the median GLUE score for peak–gene links supported by regulatory evidence from 0.36 to 0.93, and recovered known enhancer–promoter interactions, including multiple peaks upstream of the COL1A2 locus48 (Extended Data Fig. 9a and Supplementary Fig. 25).
To generalize, we benchmarked guidance types on the 10X Genomics PBMC 10K multiome dataset (Methods). Relative to bulk GTEx v8 eQTL guidance used in the original implementation, cell-type-specific eQTLs markedly improved peak–gene inference (for example, a 0.30 increase in CD4 naive cells; Extended Data Fig. 9b). Guidance from more abundant cell types generally produced larger gains, consistent with greater eQTL discovery power, as regulatory T (Treg) cells (approximately 1.46% of the cohort) gave only modest improvement compared with CD4 naive cells (approximately 12.86%) or CD14 monocytes (approximately 21.74%; Supplementary Fig. 26). Cell-type matching nonetheless outweighed abundance. Using 48,248 scRNA-seq and 52,042 scATAC-seq Treg cells as a baseline, matched Treg eQTL guidance increased the mean GLUE score by 0.57, whereas unmatched CD14Mono guidance, despite approximately 20-fold greater abundance, increased it by only 0.04. Inference remained accurate even when RNA and ATAC compositions diverged (CD14Mono: 2.3% of RNA cells versus 22.8% of ATAC cells; mean GLUE score of 0.77). Across three validation sets (GWAS, eQTL and CRISPR), our approach performed comparably with state-of-the-art methods such as pgBoost49, with further gains from QTL guidance (Fig. 5a and Supplementary Fig. 27), and distal peaks 75–100 kb from the TSS benefited most, reflecting greater ambiguity in their regulatory targets (Fig. 5b). Using peak–gene pairs with eQTL guidance from both TenK10K RNA + ATAC improved the discrimination for CRISPR-validated pairs by 17.2% and 15.6% in CD14Mono and CD4TCM cells, respectively (Extended Data Fig. 9c,d).
a, Benchmarking of GLUE, GLUE with eQTL guidance, and pgBoost for peak–gene linkage inference. We stratified pairs by CRISPR-based regulatory support and quantified performance by area under the precision-recall curve (AUPRC). Peak–gene pairs without and with CRISPR support, respectively: GLUE n = 1,216 and 172; GLUE with eQTL guidance n = 2,293 and 310; pgBoost n = 4,992 and 577. b, Distribution of GLUE scores without (blue) and with (orange) prior regulatory guidance (n = 1,681, 1,497, 1,464, 1,372, 1,412 and 1,283 from left to right). The number is the same for two boxplots in the same distance group. c, Benchmarking of GLUE against CellOracle, SCENIC and SCENIC+ for GRN inference. We evaluated performance using three independent transcription factor (TF)-target validation datasets and quantified it with the early precision ratio (EPR). The error bars indicate the 95% confidence interval (1.96 × s.e.) calculated from five independent runs as computational replicates. d, Genome track plot for APOBEC3A showing SMR-based associations with chromatin accessibility and gene expression. The top three tracks show ATAC peaks linked to the APOBEC3A promoter in CD14Mono, using different sources of linking evidence. Specifically, we assigned peaks to the gene based on caQTL–eQTL SMR (P < 5 × 10−8, based on chi-squared test), GLUE (score > 0.8) and genomic proximity (±10 kb of the TSS).
Finally, we integrated caQTL–eQTL SMR results alongside GLUE-inferred networks to refine transcription factor–target assignment. GLUE draws on chromatin expression co-variation with prior graphs, whereas SMR uses shared genetic variants as instruments to infer causal peak–gene relationships, so the two capture complementary interactions (Methods, Fig. 5c, Extended Data Fig. 10 and Supplementary Fig. 28). Integration yielded 128 additional transcription factor–target gene pairs (a 22.4% increase) in CD14 monocytes, 66 of them at GWAS loci for the 16 disease traits (Extended Data Fig. 10 and Supplementary Table 16). At the APOBEC3A locus, genomic proximity alone yielded few reliable peak–gene links. Adding GLUE and SMR evidence enriched the inferred network and, after GRN pruning with SCENIC, predicted APOBEC3A as a target of IKZF1 and STAT1 (Fig. 5d), both of which have approved inhibitors. Consistent with this, target–indication pairs supported by SMR evidence were enriched across almost all stages of clinical development in the Open Targets Platform (v25.12), most strongly among approved indications (Methods and Supplementary Fig. 29).
Integrating genetic evidence with multiome profiles therefore yields more accurate, cell-type-specific regulatory networks, recovering interactions that co-variation in multiome data alone does not resolve.
Discussion
In this study, we have shown that genetic variants shape chromatin accessibility in a cell-type-specific manner to influence immune function and disease risk, capturing regulatory effects largely missed by transcriptomic approaches alone. Profiling 3.5 million nuclei from 922 individuals by scATAC-seq alongside matched eQTL and whole-genome sequencing data, we identified 243,225 caQTLs across 28 immune cell types, with 22.8% of peaks showing genetic associations, and linked 28,834 peaks to 11,204 genes as candidate cis-regulatory elements, over half of which were supported by Mendelian randomization.
These findings address a long-standing challenge in human genetics: the limited overlap between GWAS loci and eQTLs. Across 16 diseases and 44 blood traits, caQTLs colocalized with GWAS loci that eQTLs did not, revealing distal regulatory effects that standard eQTL analyses miss. Several mechanisms could explain this. Accessible chromatin does not always drive immediate transcription: a variant may alter accessibility at a poised or primed element without yet producing genotype-dependent expression differences. Transcription factor binding at cis-regulatory elements may be restricted to rare or developmental cell states, or require environmental triggers, and would then be detected as response eQTLs50. Alternatively, accessibility changes may be buffered post-transcriptionally through mRNA stability, splicing or translational control, which modulate transcript abundance independently of chromatin state51,52,53.
Several limitations remain. caQTLs are mapped to peaks, which are not fixed genomic features and vary with scATAC-seq platform and peak-calling strategy, complicating cross-study comparison. However, we observed high reproducibility across cohorts using different technologies and association models. caQTL power is also sensitive to the cis-window used in testing: narrow windows reduce computational burden but miss distal effects, and we found variants nearly 1 Mb away exerting strong effects on accessibility, probably mediated through 3D genome organization and enhancer–promoter looping54. Wider windows (±1 Mb) are therefore necessary, and methods that capture true long-range regulation without inflating technical noise remain needed.
Cell-type annotation in scATAC-seq also lacks a gold standard, particularly for rare populations; we found that a bridging multiome reference substantially outperformed direct label transfer from imputed gene activity. Our 28-cell-type annotation balances accuracy against caQTL power, but large-scale higher-resolution references would probably resolve more granular cell-type-specific effects.
Looking forwards, chromatin accessibility captures only one dimension of regulation. Incorporating histone modifications, DNA methylation and protein expression, ideally measured in the same cells, would give a fuller view of molecular regulation. Reliance on resting cells may also obscure effects that emerge only under stimulation, and extending caQTL mapping beyond blood to disease-relevant tissues will be essential for assessing generalizability. Finally, structural variants and tandem repeats25 may reveal regulatory mechanisms invisible to SNP-centric and indel-centric analyses.
Together, these results establish chromatin accessibility as a key layer through which non-coding variation acts on immune cell function and disease risk, and show that measuring it at cell-type resolution recovers regulatory mechanisms that expression data alone leaves unexplained.
Methods
Sample collection and library processing
To build a chromatin accessibility atlas for human PBMCs, we collected blood samples from 1,042 consenting individuals from the TenK10K phase I cohort5, which is a combination of 952 from the Tasmanian Ophthalmic Biobank (TOB) cohort, 64 from the BioHeart (BH) cohort55, and 26 from the Liquid Biopsy Biobank (LBIO) cohort. We obtained the 952 donors from a subset of the 982 donors of European ancestry in OneK1K10, the pilot cohort of TenK10K. We processed TOB donor blood samples in different batches than those used to process scRNA-seq samples in Cuomo et al.5. The BioHEART cohort was recruited between 29 March 2016 and 2 May 2023, and included participants who presented for clinically indicated cardiovascular imaging or with acute ST-elevation myocardial infarction. The LBIO was established to collect peripheral blood from participants with advanced cancer receiving intravenous therapy. Recruitment for LBIO commenced in December 2019 and is ongoing. No statistical methods were used to predetermine sample size. The study did not involve randomization or blinding.
Ethics oversight
This study was conducted in accordance with all relevant ethical regulations. It was overseen by the Human Research Ethics Committees at the Royal Children’s Hospital, Northern Sydney Local Health District, St. Vincent’s Hospital and the University of Tasmania. We obtained ethics approval for all three datasets: TOB (ethics protocol 2020/ETH02479), BioHEART study (ethics protocol 2019/ETH08376) and LBIO (2019/ETH13113). All participants provided informed consent.
Single-cell ATAC-seq
In total, we processed 952 individuals from the TOB cohort using the 10X Genomics Chromium Next GEM single-cell ATAC kit v2. This is the main dataset used for caQTL mapping. In brief, we thawed, washed, counted and pooled cryopreserved PBMC samples into 119 pools of 8 individuals, with equal numbers of live cells per donor. We isolated nuclei and transposed them before running each pool across two single-cell reactions, targeting either the same or different numbers of nuclei, ranging from 20,000 to 50,000. The specific number for each library can be found in Supplementary Table 1. We generated and sequenced 238 Illumina sequencing libraries at a read depth of approximately 30,000 reads per nucleus using S4 flow cell kits on the NovaSeq 6000.
Multiome sequencing
The BioHEART and LBIO samples were processed with the 10X Genomics Chromium Next GEM Single Cell Multiome ATAC + Gene Expression kit v1. The BioHEART multiome subcohort consisted of 64 individuals split into 4 pools of 16 individuals each, and the LBIO subcohort consisted of 26 individuals divided into 4 pools of 6–7 individuals each. As with the ATAC-only reaction, we thawed, washed, counted and pooled the cryopreserved PBMC samples in equal numbers of live cells per donor before isolating and transposing nuclei. We then ran each pooled cell suspension in a single reaction. We generated ATAC and gene expression libraries from the same pool of pre-amplified transposed DNA and cDNA. We sequenced ATAC and RNA libraries independently at approximately 30,000 reads per nucleus using S4 flow cell kits on the NovaSeq 6000.
Whole-genome sequencing and genotyping
We extracted DNA for whole-genome sequencing (WGS) from blood samples using the Qiagen DNeasy Blood and Tissue Kit, and measured DNA concentrations with Qubit. We prepared libraries using the KAPA PCR-free kit and sequenced them on an Illumina NovaSeq 6000 at 30× coverage. Cuomo et al.5 have provided more details on quality control and processing. We obtained genotype array data for all 952 TOB individuals from Yazar et al.10 and used them only for demultiplexing. Because 20 TOB individuals lack WGS data, we excluded them from caQTL mapping.
We subsetted common variants, including SNPs and indels (MAF ≥ 5%), from the WGS data for common-variant caQTL mapping. Variants with minor allele count < 20 or missing genotype rate > 15% were also excluded. We extracted rare variants (MAF < 5%) to perform rare-variant caQTL mapping. We used genotype array data generated with the Illumina Infinium Global Screening Array in the OneK1K study10 for demultiplexing.
Quality control and preprocessing
Read alignment (Cell Ranger ATAC)
Cell Ranger ATAC (v2.2.0) was used to align sequencing reads to the reference genome (GRCh38 v1.2.0); ‘the-force-cells’ parameter was set to 20,000, 40,000 or 50,000, depending on the targeted cell capture number for each sequencing library, resulting in 9,457,411 nuclei in total (Supplementary Table 1). Using an older version without the-force-cells function (v2.1.0), which relied on the automatic cell calling algorithm, resulted in 5,639,779 nuclei (Supplementary Table 2). We performed initial peak calling using the Cell Ranger built-in MACS2 method56. We used Cell Ranger ARC to process the multiome sequencing libraries (RNA + ATAC) for the BioHeart and LBIO cohorts, yielding a total of 105,887 nuclei.
Demultiplexing and doublet detection
To assign nuclei to each donor and identify doublets, we used Vireo implemented with Demuxafy (v2.1.0)57. Because we used the-force-cells parameter during Cell Ranger count, many low-quality nuclei were classified as unassigned during demultiplexing and removed (1,155,771). In the TOB cohort, we detected 3,644,297 doublets. The average doublet rate was 37.0% across all 238 libraries, compared with 32.7% if no force cell was used (Supplementary Table 3 and Supplementary Fig. 30). There are several big discrepancies in the quality control metrics between the two versions but such discrepancies are direct and expected consequences of the-force-cells setting used in the new pipeline and does not reflect any inconsistency in the underlying data. Critically, both pipelines converge on essentially the same high-quality nuclei after our standard quality control step, and we performed all downstream analyses on this quality control-filtered set. Two donors (10,003 nuclei) were not mapped to any donor genotype; thus, we retained 4,586,910 singlets from 950 donors after this step. For the multiome sequencing results from the BioHEART and LBIO cohorts, demultiplexing in the ATAC layer retained 41,248 singlets from 48 donors for BioHEART and 15,303 singlets from 12 donors for LBIO. We removed one BioHEART library and two LBIO libraries with poor sequencing quality and demultiplexing performance. In the gene expression layer, the data quality was also low, with only 11.9% of nuclei reliably assigned to a donor. Therefore, we used only the ATAC layer results for downstream analyses.
Quality controls
Nuclei that failed the following quality control thresholds were excluded from analyses: number of peak counts of more than 5,000 and less than 100,000, percentage of reads overlapping with peaks of more than 40, blacklist ratio of less than 0.01, nucleosome signal of less than 4 and TSS enrichment score of more than 4. We excluded seven sequencing libraries with a low number of nuclei after quality control (less than 300), resulting in the loss of one pool (eight donors excluded). After matching with WGS data, 20 donors could not be matched and were removed. We removed seven libraries (S0268_2, S0301_1, S0302_1, S0303_1, S0304_1, S0304_2 and S0331_1) after all quality controls. The final dataset included 3,472,552 nuclei from 922 TOB donors across 231 libraries (Supplementary Fig. 1). On average, each donor contributed 3,766 nuclei from scATAC-seq and 2,825 cells from matching scRNA-seq data.
Cell-type annotation by label transferring
To obtain fine-grained cell-type resolution for PBMCs, we adopted a two-step strategy to annotate the scATAC-seq data. We first curated a multiome bridging dataset combining eight multiome datasets (99,507 nuclei): six 10X public datasets, the BioHeart dataset and the LBIO dataset. We merged the bridging dataset into the same peak space using the ‘Reduce’ method, which merges all intersecting peaks. We then annotated it by label transfer from a CITE-seq reference (120,678 cells) provided by Hao et al.58, and removed nuclei with a maximum prediction score of less than 0.3. In total, 95,348 nuclei (31 cell types excluding 3 non-PBMC types) were retained for the next step (Supplementary Figs. 31 and 32). After obtaining cell-type labels for the bridging multiome dataset, we built each ATAC library with anchors to the ATAC layer in the multiome bridging dataset to obtain cell-type labels. We described the number of nuclei per cell type for each library in the TOB cohort (Supplementary Table 4) and for each donor in the TOB, BioHeart and LBIO cohorts (Supplementary Tables 5 and 6). We also visualized cell-type annotations for each donor in boxplots to check extreme outliers (Supplementary Fig. 33) and in a principal component analysis (PCA) plot annotated by how many cell types fell outside the third quartile + 1.5 × interquartile range threshold (Supplementary Table 17 and Supplementary Fig. 34). The two technical replicates for the same pool also showed reasonably high concordance (Supplementary Fig. 35).
The annotation strategy was designed to maximize both annotation accuracy and caQTL discovery power given the constraints of our study design. With approximately 2,825 nuclei per donor after quality control, accurate annotation of rare cell subtypes with an abundance below approximately 0.1% is not statistically reliable, as many such types would be represented by fewer than three nuclei per donor on average. This per-donor yield is partly a consequence of the pooled library design (eight donors per pool), which places a practical ceiling on nuclei recovery. Furthermore, caQTL discovery power scales with the number of nuclei per donor per cell type (Fig. 2b), so many finer-grained subtypes lack sufficient representation to support meaningful caQTL mapping. To evaluate whether a cut-off of cell-type annotation for scATAC-seq data would materially change our results, we re-ran the caQTL analysis on chromosome 1 for four cell types spanning a range of abundances (CD14Mono, NK, B naive and Treg cells) across four candidate thresholds (0.3, 0.5, 0.7 and 0.9). We compared the caQTL effect estimates against those from the original, uncut analysis. caQTL slope estimates are highly concordant across thresholds (Pearson R ≥ 0.88 in all conditions; Supplementary Table 18 and Supplementary Fig. 36), indicating that our downstream results are robust to this choice. Deviation from the identity line is noticeable only at the most stringent threshold (0.9). It largely reflects severe data loss in low-abundance populations (for Treg cells, the 0.9 cut-off leaves only approximately 4 nuclei per donor, which is not analytically viable). This analysis supports our decision to use the highest-scoring assignment without an additional cut-off. Together, these constraints motivate using 28 cell types as a level of granularity that balances annotation fidelity with statistical power.
Verifying cell-type annotation with imputed scRNA-seq data
After obtaining cell-type labels, we checked canonical gene markers for each cell type. We used the gene activity score generated in the previous step for each library. We then visualized the gene activity scores of the gene markers used in the Seurat58,59/Azimuth reference (see Code availability) for CD14Mono and cDC1 and confirmed that the canonical markers had enriched scores in the corresponding cell populations (Supplementary Fig. 32). We also visualized the gene activity scores of the markers used for all 26 cell types in the main scATAC-seq data (Supplementary Fig. 37).
Recalling peaks within each cell type
For each cell type, we recalled peaks from the raw fragment files using the MACS3 tool56 integrated into SnapATAC2 (ref. 60). The total number of peaks called in each cell type ranged from 427 (CD8+ proliferating) to 269,371 (CD16Mono), with a median of 170,292 (Fig. 1e). The peaks were then merged using the Reduce function from Signac61 to define a consensus list of peaks. In total, the merged object contains 440,996 unique peaks, with minimum, median and maximum widths of 201 bp, 499 bp and 5,545 bp. We matched the peak regions against two comprehensive datasets: ENCODE (v4)62 with 2,348,854 peaks, and a whole-body scATAC-seq atlas63 with 1,154,611 peaks, as well as the 10X 10k PBMCs ATAC v2 Chromium Controller dataset (164,487 peaks and 10,246 nuclei). After matching, 44,015 peaks (9.98%) did not overlap with any of these three datasets (Supplementary Table 7). The novel peaks are probably due to the increase in sample size, so we have more power to call peaks with low accessibility, as supported by the observation that the donor-level sparsity of novel peaks is consistently higher than that of known peaks (Supplementary Fig. 12). ChromBPNet also showed lower prediction accuracy for novel peaks than known peaks (Supplementary Fig. 11).
We annotated the peak list based on functional categories curated by overlapping the pre-annotated regions. In total, we mapped 437,561 peaks to 65 functional categories. One peak could match multiple categories, and when we calculated the proportion of each category, we used all matched pairs to compute the proportion for each category. For the ENCODE v3 annotation64, there were five main categories: promoter-like signature (pLS), proximal enhancer-like signature (pELS), distal enhancer-like signature (dELS), DNase–H3K4me3 and CTCF only.
Dimensional reduction and visualization
We used EpiScanpy65 to obtain a dimension reduction space for the scATAC-seq data. EpiScanpy is a toolkit for analysing single-cell epigenomic data, including scATAC-seq. We combined the quality-controlled scATAC-seq data into a single AnnData object without removing the 20 donors with unmatched WGS data (3,546,117 nuclei with 440,996 peaks), then binarized the data and selected the top 50,000 peaks based on the built-in feature variability score in EpiScanpy. We then further normalized the data such that the total count of each cell equals the median of total counts for all the cells before normalization. We then performed PCA with Scanpy66 to reduce dimensionality. The cell clustering pattern was as expected (Supplementary Fig. 38). We integrated the scATAC-seq data with scRNA-seq in TenK10K by projecting them into a co-embedding space (Extended Data Fig. 2a). One cluster showed limited overlap between the two layers. This is due to differences in the proportion of the monocyte population between the two datasets because the blood samples were processed in different batches. A more detailed explanation is provided in supplementary note 1 of the TenK10K flagship paper5.
The caQTL mapping
Overall, after the nucleus-level and sample-level quality controls, a total of 3,472,552 nuclei of scATAC-seq from 922 individuals, 5,438,679 cells of scRNA-seq from 1,925 individuals (including the 922 individuals with scATAC-seq) and 95,728 nuclei of multiome data from 60 individuals (independent from the other two cohorts) were used for subsequent analyses resulting in 9 million single cells or nuclei (Extended Data Fig. 1b).
Cell-type-specific caQTL mapping
To map genetic variants to the variation of chromatin accessibility across individuals, we ran an association test between genetic variants (SNPs + indels) and chromatin accessibility levels in each of the 28 cell types. We aggregated peak counts (by summing all nuclei for one donor) into a pseudobulk matrix for each cell type. We did this separately for each repeat. In each repeat, we corrected the GC content bias using the ‘withinLaneNormalization’ function in EDAseq67. We then removed peaks with more than 95% zero counts across donors. We converted the count data to counts per million (CPM) and then performed z-score normalization for all the peaks. We merged the normalized matrices from the two repeats by averaging the z-scores for overlapping peaks. For the unique peaks, we kept the original z-scores. Finally, we re-standardized each peak across all donors. We used the standardized matrix to generate chromatin accessibility principal components. We conducted sensitivity analyses to optimize association-mapping parameters, including normalization methods, window size and number of principal components (Supplementary Figs. 39–41). Although a smaller window size, such as 10 kb, yielded more significant associations, it would miss many associations outside the 10-kb window, and the number of tested variants overlapping with other QTL datasets would be limited. To generate consistent SNP–peak pairs between caQTL discovery and downstream colocalization analysis, we used ±1 Mb as the default window size.
The association model fits multiple covariates, including sex, age, genotype principal components and chromatin accessibility principal components. The final number of donors per cell type ranges from 120 to 922. We did not test the two rarest cell types, CD8+ proliferating (3 donors) and innate lymphoid cells (19 donors), because of the limited donor size. We determined the number of principal components fitted in the model using the Elbow method in the R package PCAForQTL68. Each peak was tested against all the SNPs located within the ±1-Mb window around the peak centre, and the number of SNPs tested for each peak ranged from 293 to 30,942, with a median of 6,067. For each peak, TensorQTL69 ran 10,000 permutations to correct P values using a β-approximation strategy. We then applied peak-level correction for each chromosome and called a caQTL if the q < 0.05, and we identified 243,225 caQTLs in total. We used graphics processing unit (GPU) acceleration to speed up the association tests, which substantially reduced computational time given the large number of peaks.
Because we measured scATAC-seq twice for each pool, we also ran the caQTL association for each repeated library separately and then compared the caQTL signals between the two runs. The significance level, quantified by −log10(q value), and effect sizes are largely consistent (Pearson’s correlation > 0.90) across cell types except for those rare cell types with limited power (Supplementary Fig. 42).
Defining the cell-type specificity of caQTL effects
To quantify cell-type specificity of caQTLs, we assessed different models to explain the specificity using the approach described in Cuomo et al.5 (Fig. 1c, Extended Data Fig. 3 and Supplementary Fig. 5). We defined five possible scenarios that can occur for a given caQTL identified at q < 5% in one cell type (cell type A) when compared pairwise with another cell type (cell type B):
-
Scenario 1 (specific caPeak): the caPeak associated with the caQTL is not sufficiently accessible (less than 1% of nuclei) in cell type B.
-
Scenario 2 (specific caQTL): the caPeak is sufficiently accessible in cell type B, but no caQTL is identified.
-
Scenario 3 (distinct caQTL): different caQTL variants in low linkage disequilibrium (r2 < 0.5) are identified for the same caPeak in cell types A and B.
-
Scenario 4 (shared caQTL, concordant effect): the same caQTL variant or variants in high linkage disequilibrium (r2 ≥ 0.5) are identified for the same caPeak in both cell types, with a concordant direction of allelic effect (that is, the same directional effect of the minor allele on chromatin accessibility).
-
Scenario 5 (shared caQTL, discordant effect): the same caQTL variant or variants in high linkage disequilibrium (r2 ≥ 0.5) are identified for the same caPeak in both cell types, but with a discordant direction of allelic effect.
Functional annotation of caQTLs
There are 243,225 caQTLs, spanning 100,483 peaks and 158,251 unique top variants. Similar to the annotation of chromatin peaks, we overlapped the caPeaks with the five cCRE categories in ENCODE v3, as well as the top variant for each peak (Fig. 1f). The annotations of caPeaks are further divided into five subgroups based on the peak width.
Fine-mapping of causal variants for caQTLs using SuSiE
We performed fine-mapping analysis for all caPeaks using SuSiE28. For each peak, we considered all variants located within a 1-Mb cis-window and calculated the corresponding variant correlation matrix using PLINK (v1.9)70. We obtained z-scores from TensorQTL summary statistics and defined the sample size N as the number of donors containing the corresponding cell type. These summary statistics were supplied to the ‘susie_rss’ function with L = 10. The analysis yielded credible sets for each peak and marginal PIPs for individual variants.
Replication of caQTLs in published and in-house datasets
To replicate our identified caQTL signals, we collected caQTL results from two published studies20,21. Because those published studies used different cell-type annotations, we counted a caQTL as replicated if it reached significance in the closest cell type. In other words, if at least one of our cell types contained a significant caQTL that overlapped with theirs, we counted it as being replicated. We also quantified replication of the direction of caQTL effects for Benaglio et al.21. As peak definitions differ across studies, we counted a replication if the peak regions overlapped between datasets, which might inflate concordance estimates. Conversely, fewer than 10% of our caQTLs replicate in previous studies, as expected given the limited statistical power of those earlier cohorts. We also conducted caQTL mapping on a combined multiome dataset comprising 60 donors from two cohorts (BioHeart and LBIO) after quality control. We used the same protocol for cell-type annotation, pseudobulking and caQTL association as in the main caQTL analysis for the TOB cohort.
Per-donor scATAC reads (n = 890 donors, because only the first repeated library was used) were realigned with STAR (v2.7.11b) in WASP mode71, retaining reads tagged vW:i:1 to correct for allelic mapping bias. We obtained heterozygous sites from OneK1K phased, imputed SNP-array genotypes (MAF ≥ 0.05, imputation R2 ≥ 0.3, biallelic SNPs only). Reference and alternate snATAC counts at heterozygous sites were then quantified with GATK ASEReadCounter (v4.6.2) on per-(donor, cell type) pseudobulk binary alignment map (BAM) files. For each peak, we resolved haplotype counts for each individual at heterozygous sites based on phased genotypes and then summed them across the peak interval. To run this analysis, we developed an optimized Python re-implementation of the mixQTL ASC method29 and ran it using per-peak haplotype counts within a 1-Mb cis-window flanking each peak, with default mixQTL parameters. To validate lead caQTLs, we calculated sign concordance and Spearman and Pearson correlations using TensorQTL and mixQTL effect size estimates for peak-internal lead caQTLs (that is, caQTLs whose lead variant fell within their associated peak) that were also ASC measurable. We considered lead caQTLs as ASC measurable if they passed the mixQTL default thresholds (asc_cut-off = 5, n_cut-off = 15; that is, peaks that had at least 15 donors with at least 5 SNP-intersecting reads for each haplotype within the peak). We excluded cell types with fewer than 30 such peak-internal ASC-measurable lead caQTLs from headline correlation and concordance analyses to ensure stable estimates.
Mapping rare caQTLs using SAIGE-QTL
We assessed the effects of rare variants on single-cell chromatin accessibility using the set-based test implemented in SAIGE-QTL. All SNPs and indels with MAF < 5% and missingness < 15% were included. For each peak, we tested the combined effects of variants within a cis-region defined as ±100 kb around the peak centre. We used a different window size here because it maximized the number of caQTLs in a sensitivity analysis of chromosome 1 and CD4 naive cells (Supplementary Fig. 13). Following the SAIGE-QTL framework, we performed both burden tests and sequence kernel association test (SKAT). Peak-level P values were combined using the Cauchy method, similar to the SAIGE-QTL paper31. Consistent with the procedure applied to common variants above, we controlled for multiple testing using the q value method and considered results significant at false discovery rate < 5%.
To rule out scenarios in which rare signals are induced by common variants in linkage disequilibrium with the rare variants tested, we further performed a conditional analysis that conditions on nearby significant common caQTLs. For each peak, we conditioned on the common variant with the lowest P value within a 100-kb cis-window centred on the peak, as identified from the TensorQTL results. Following stepwise conditional analysis, 53,195 significant rare caQTLs remained significant in the set-based test.
Regulatory effect prediction using ChromBPNet
Processing of single-cell ATAC-seq data
We first normalized the sequencing depth and aggregated the 238 scATAC-seq libraries. Then, we created pseudobulk fragment files by pooling all fragments from each predicted cell type. Next, we randomly split the fragments into two pseudo-replicates and called peaks with the ENCODE ATAC-seq pipeline (v2.2.3; https://github.com/ENCODE-DCC/atac-seq-pipeline) with default parameters. We used the overlap peak set for all downstream analyses.
ChromBPNet bias model training
To select the initial hyperparameters for the bias model, we partitioned the data into train, validation (chromosomes 8 and 20) and test (chromosomes 1, 3 and 6). Next, we trained bias models in three cell types with a range of available cells, including CD14 monocytes (n = 767,874), NK cells (n = 432,705) and B naive cells (n = 152,441), with a bias threshold factor from 0.1 to 1 at 0.1 intervals. We picked the best bias model of each cell type based on three criteria: (1) the Pearson’s R between predicted and observed non-peak counts was high; (2) the Pearson’s R between predicted and observed peak counts was low, but greater than −0.3; and (3) the model only captured bias motifs (such as TN5 or DNase-I) and did not capture any real transcription factor motif in transcription factor-MODISCO-derived motifs.
Next, we trained the ChromBPNet bias-factorized model30 using the best-performing bias models from each of the three cell types. We then selected the final bias model based on two criteria: the highest Pearson’s R between predicted and observed peak counts, and the lowest Pearson’s R between predicted and observed non-peak counts, while ensuring the model’s Tn5 bias motif was corrected.
Finally, we trained the final bias model (NK cell with bias threshold factor of 1) with fivefold cross-validation. We followed the chromosome partition scheme used in the ChromBPNet method30, where each chromosome is used as a test set exactly once across the folds.
ChromBPNet bias-factorized model training
We trained ChromBPNet models in 25 of 28 cell types, except CD8+ proliferating (4 cells), innate lymphoid cells (33 cells) and CD4+ proliferating (145 cells), using fivefold cross-validation with the final bias model selected in the previous step.
ChromBPNet variant effect prediction
For all top SNPs associated with a peak, we generated a 2,114-bp (ChromBPNet input length) region centred on the variant. Then, we computed cell-type-specific variant scores as the log2 fold change in predicted total counts of the alternative versus reference allele using cell-type-specific ChromBPNet bias-factorized models. We also performed the same prediction using fine-mapped SNPs with high PIP scores in each peak, and we compared prediction accuracy across different PIP thresholds (Supplementary Fig. 10).
Transcription factor motif matching
To assess motif matches of the caQTL variants, we used the JASPAR 2026 vertebrate database72. For each variant, we computed contribution scores and performed motif discovery using Tangermeme (v0.5.1). We then assigned the best-matching transcription factor based on the most significant motif P value near the variant position.
Colocalized caQTL and eQTL to generate peak–gene links
To identify peak–gene links, we performed colocalization analysis using coloc36 by integrating eQTL and caQTL summary statistics. We tested all significant caPeaks against genes within ±1 Mb of the peak centre. To keep a consistent variant list across colocalization analyses, we excluded indels from the summary statistics and focused only on SNP variants. The coloc results indicate the posterior probability of five different models: PPH0: no association with either trait; PPH1: association with trait 1, but not with trait 2; PPH2: association with trait 2, but not with trait 1; PPH3: association with trait 1 and trait 2, two independent SNPs; and PPH4: association with trait 1 and trait 2, one shared SNP. In our case, trait 1 is the chromatin accessibility and trait 2 is gene expression.
Although a colocalization event suggests a peak–gene association, it does not necessarily inform the directionality or effect size. To provide causal inference support for peak–gene pairs that we identified through coloc analysis, we further adopted the Mendelian randomization framework to infer causal association of the caPeaks on nearby genes using OmicsSMR73. We first curated the caQTL and eQTL summary statistics into the MatrixEQTL format, and accordingly generated the BESD files as inputs for the SMR analysis. To select a significant peak for testing, we applied a nominal P value threshold < 5 × 10−8. This peak-selection threshold is much more stringent than q < 0.05, so OmicsSMR tests far fewer peak–gene pairs than coloc. For example, in CD14Mono, coloc tested 740,024 pairs, and OmicsSMR tested 191,423 pairs. After accounting for overlap, we kept 190,494 pairs, so the OmicsSMR pairs are a subset (99.51%) of coloc pairs. When comparing the two methods, we focused only on the overlapping pairs.
Colocalization of caQTL, eQTL and GWAS
To further understand the genetic regulatory mechanisms of GWAS associations, we conducted colocalization between caQTLs and GWAS associations from 16 complex diseases using coloc, as well as 25 blood serum markers and 19 blood count traits, following the same traits tested in Margoliash et al.74. The coloc results between eQTL and GWAS were acquired from Cuomo et al.5. For further comparison, we also obtained bulk eQTL summary statistics from eQTLGen9 (n = 31,684) and GTEx v10 whole-body tissues75 (n = 70–816) and ran the same colocalization analysis. Similar to coloc analyses, we also ran SMR analysis between eQTL and GWAS. In the SMR analysis between eQTL and GWAS, we tested genes with a top eQTL nominal P < 5 × 10−8. We performed the same SMR analysis between caQTL and GWAS.
To assess how much GWAS signals colocalize with eQTLs and caQTLs, we used a two-step approach. First, we performed clumping analysis to select independent loci for each GWAS, using a significance threshold of P < 5 × 10−8, a linkage disequilibrium r2 threshold < 0.01, and a window size ± 1 Mb. Next, we examined the raw eQTL and caQTL summary statistics, along with the outputs from coloc and SMR analyses. For each GWAS locus, we compiled a list of genes and peaks with significant QTL associations (nominal P < 5 × 10−8). A locus was considered colocalized with eQTLs or caQTLs if at least one gene or peak from the lists met the following criteria: for coloc, PPH4 > 0.8; for SMR, PSMR < 5 × 10−8. We then aggregated results across all cell types to calculate the overall proportion of GWAS loci colocalizing (Fig. 3a,b).
Conventional coloc function assumes a single causal variant when assessing colocalization between two phenotypes. However, in practice, both phenotypes may harbour multiple causal variants. To relax this assumption, we integrated SuSiE, testing all pairs of credible sets across the two phenotypes, as implemented in the coloc.susie method41. Specifically, we performed SuSiE fine-mapping separately for each phenotype and then evaluated each pair of credible sets using conventional coloc. If any pair showed evidence of colocalization (PPH4 > 0.8), we considered the two phenotypes to colocalize.
Multi-trait fine-mapping for identifying putative causal variants
We applied mvSuSiE42, a variable selection method for multivariate linear regression, to identify the putative causal variants that simultaneously affect disease risk, gene expression and chromatin accessibility. Specifically, we selected genes that colocalized with GWAS (PPH4 > 0.8) as the focal genes. For each focal gene, we then selected caPeaks colocalizing with eQTL (PPH4 > 0.8) within ±1 Mb of the gene body. We then extracted summary statistics from the eQTL, caQTL mapping and GWAS and constructed a z-score matrix, with rows representing different SNPs and columns representing different phenotypes. We also set up canonical covariance matrices for the remaining phenotypes with mashr27, as well as the linkage disequilibrium matrix for SNPs in the focal region with PLINK70. We then fit an mvSuSiE model using the z-score matrix, covariance matrices and the linkage disequilibrium matrix as input.
The key output of a fitted mvSuSiE model includes a list of credible sets that are intended to capture at least one causal SNP; a PIP score quantifying the probability that a SNP is causal; and posterior estimates of SNP effects on each phenotype. We used the local false sign rate in mvSuSiE as a phenotype-level significance measurement to define the phenotypes affected by the credible sets (cut-off = 0.01). Collectively, we identified 1,085 credible sets that affect at least one layer of phenotypes (Supplementary Table 15), and we focused on credible sets that simultaneously affect disease risk, gene expression and chromatin accessibility.
Genetic regulatory interaction with cell state
Inferring pseudotime as a proxy for cell state using Palantir
To model cell-state progression, we performed pseudotime analysis using Palantir76 on 1,176,697 cells aggregated across all pools and donors (N = 942) within the CD4 lineage (CD4 naive, CD4TCM, CD4TEM, CD4+ proliferating, CD4CTL and Treg cells). In brief, Palantir infers pseudotime by projecting cells into diffusion space, constructing a neighbourhood graph and estimating relative progression from a user-defined starting population, while also modelling lineage probabilities towards terminal states through a Markov chain framework. We first performed PCA using 75 principal components, followed by Harmony77 batch correction using the sequencing pool as the batch variable. We then generated a UMAP embedding from the batch-corrected 75 principal components using a 30-nearest-neighbour graph. Next, we used these embeddings to compute 100 diffusion components, before running Palantir with k-nearest neighbour = 50 and num_waypoints = 8,000.
Testing the cell-state-dependent caQTL effect using CASTIE
Using cell-specific pseudotime estimates from Palantir, we applied CASTIE43 within the SAIGE-QTL framework31 (v0.2.5.1), an extension of SAIGE-QTL designed to efficiently model heterogeneous genetic effects across varying cellular contexts directly from single-cell data, without pseudobulk aggregation. To reduce computational cost, we tested 493,520 cells from a subset of 386 donors in this analysis. First, we fitted a null Poisson mixed model for each gene across all cells. We included the inferred pseudotime as the dynamic (environmental) covariate, whereas sex, age, genotype principal components and chromatin-accessibility principal components were incorporated as static covariates. Next, we performed association tests for each genetic marker with MAF > 0.01 and minor allele count > 5, processing 10,000 markers per chunk. Finally, we used the ACAT test78 to aggregate marker-level results into gene-level P values.
To visualize interaction patterns, we stratified the cell population into three pseudotime bins (early, mid and late), corresponding to pseudotime ranges of 0–0.33, 0.33–0.66 and 0.66–1.00, respectively.
Quantification of cell function scores in CD4+ cells
We computed cell function scores in CD4+ cells as cell states using scDeepID, a flexible multi-task transformer that leverages biological pathway databases as prior knowledge to train the model. First, we trained a scDeepID model with 2,390,948 CD4+ cells (including CD4+ naive, CD4TCM, CD4TEM, CD4CTL, Treg and CD4+ proliferating cells) from TenK10K phase I single-cell RNA-seq data. The training used expression features of 4,000 highly variable genes selected with the Scanpy function highly_variable_genes (n_top_genes = 4000, min_mean = 0.0123, max_mean = 3 and min_disp = 0.5), with the human Reactome database as a guide, filtered to the top 200 most informative pathways based on the presence of pathway genes in the highly variable gene list. We then extracted the attention-based latent space for UMAP representation and extracted cell function scores, followed by transforming each cell function score to approximate a normal distribution with zero mean and unit variance.
GRN inference using multiome data
Inferring cell-type-specific GRN with GLUE
We performed GRN inference using GLUE47 in a cell-type-specific manner. Specifically, we aggregated cells from each major cell type across both scRNA-seq and scATAC-seq datasets and subsequently generated integrated, cell-type-specific objects. Following the GLUE framework recommendations, we adopted a standard Scanpy-based preprocessing pipeline. For scRNA-seq data, we normalized each cell to total counts across all genes, log-transformed the normalized counts and selected 6,000 highly variable genes per major cell type. We clipped each gene expression to max_value = 10, then scaled to unit variance across cells. We then applied PCA (n_pcs = 100), followed by Harmony in Scanpy, to account for technical batch effects between sequencing libraries. For scATAC-seq, we adopted the latent semantic indexing method implemented in GLUE for dimensionality reduction. We used the derived principal components to construct a nearest-neighbour graph and run UMAP in Scanpy with default parameters.
Next, we constructed the guidance graph required by GLUE by incorporating both genomic proximity and prior regulatory knowledge derived from eQTL data. For genomic distance, we connected an ATAC peak to a gene with a positive edge if it lies within a 150-kb window around the promoter of the gene. To incorporate regulatory evidence from eQTL data, we added an edge for each peak–gene pair where the peak overlaps a genomic locus associated with gene expression (eQTL). To ensure numerical stability, we incorporated self-loops into the graph. We then applied GLUE to integrate the multi-omics data using the constructed guidance graph to learn low-dimensional cell embeddings for each omics layer. We then performed cis-regulatory inference based on these embeddings. We computed the resulting GLUE regulatory score as the cosine similarity between gene and peak embeddings, with higher scores indicating stronger putative regulatory interactions between peak–gene pairs. The peak–gene pairs were annotated as ‘with support’ if at least one putative regulatory signal (such as an eQTL) colocalized with the pair. From the inferred peak–gene regulatory interactions, we utilized SCENIC45 to build the GRN for each major cell type.
To incorporate caQTL results to enhance GRN inference, we used causal inference (SMR) results linking eQTL and caQTL to identify high-confidence peak–gene pairs with strong regulatory potential (restricted to caQTL–eQTL associations with SMR P < 5 × 10−8). We used these interactions to enrich the cis-regulatory relationships inferred by the GLUE model, which then served as input for SCENIC-based GRN construction in CD14Mono. Integrating these curated regulatory links helps to identify transcription factor–target gene relationships that might otherwise be missed.
Comparison of inferred cis-regulatory scores between paired and unpaired multiome data
To evaluate the benefits of incorporating eQTL data and to compare differences in inferred cis-regulatory scores across datasets, we used the 10k human PBMC dataset processed with Cell Ranger ARC 2.0.0 from 10X Genomics. In parallel, we generated a randomly sampled subset of the TOB multiome cohort with a comparable size (10,000 cells) to the 10X dataset. Both datasets underwent the same preprocessing procedure, as described in the previous section. We incorporated GTEx and cell-type-specific eQTLs into GLUE model training.
Comparison of QTL-derived and multiome-derived peak–gene links
To compare peak–gene links derived from QTL colocalization with those inferred from multiome data, we performed complementary analyses in CD14 monocytes. We compared caQTL–eQTL colocalization results (PPH4) with peak–gene links inferred from multiome data using GLUE. We assessed concordance between the two approaches using Spearman correlation and further evaluated it with a binned analysis of GLUE scores across increasing levels of colocalization evidence. In addition, we repeated analyses after restricting to peak–gene pairs with colocalization evidence (PPH4 > 0.8).
Conceptually, the QTL-based framework evaluates whether a shared genetic variant influences both chromatin accessibility and gene expression, whereas GLUE infers peak–gene links from co-variation patterns across cells. These approaches capture distinct aspects of regulatory relationships and are therefore expected to yield partially overlapping but non-identical peak–gene link sets. As a technical constraint, we limited GLUE analysis to peak–gene pairs within 150 kb and to highly variable genes.
Benchmarking peak–gene link methods and GRN inference methods
To better understand the performance of the GLUE framework in peak–gene links and GRN inference, we conducted two sets of benchmarking with existing tools. First, we benchmarked the peak–gene links from GLUE47 and pgBoost49 on the same three validation sets (GWAS79,80, eQTL75,81 and CRISPR82,83,84,85,86,87,88,89,90) used in pgBoost. pgBoost is a recent method that infers variant-to-gene regulatory strengths, which has demonstrated superior performance than other approaches. We modified the framework to evaluate the peak–gene links. For a given evaluation set and given method, we ranked candidate peak–gene links by linking score. At each unique linking score c assigned to any candidate link, we computed the enrichment of the set of ‘predicted’ links (linking score ≥ c) for the set of ‘true’ evaluation links as \(\frac{\frac{|{\rm{p}}{\rm{r}}{\rm{e}}{\rm{d}}{\rm{i}}{\rm{c}}{\rm{t}}{\rm{e}}{\rm{d}}\,{\rm{l}}{\rm{i}}{\rm{n}}{\rm{k}}{\rm{s}}\cap {\rm{t}}{\rm{r}}{\rm{u}}{\rm{e}}\,{\rm{l}}{\rm{i}}{\rm{n}}{\rm{k}}{\rm{s}}|}{|{\rm{t}}{\rm{r}}{\rm{u}}{\rm{e}}\,{\rm{l}}{\rm{i}}{\rm{n}}{\rm{k}}{\rm{s}}|}}{\frac{|{\rm{p}}{\rm{r}}{\rm{e}}{\rm{d}}{\rm{i}}{\rm{c}}{\rm{t}}{\rm{e}}{\rm{d}}\,{\rm{l}}{\rm{i}}{\rm{n}}{\rm{k}}{\rm{s}}|}{|{\rm{c}}{\rm{a}}{\rm{n}}{\rm{d}}{\rm{i}}{\rm{d}}{\rm{a}}{\rm{t}}{\rm{e}}\,{\rm{l}}{\rm{i}}{\rm{n}}{\rm{k}}{\rm{s}}|}}\). We defined ‘candidate links’ as all peak–gene pairs involving a gene scored by all methods and with peak-TSS distance of less than 1 Mb and more than 1 kb (for CRISPR analyses, we further restricted to peak–gene pairs experimentally tested by CRISPR). For a given evaluation set, we computed R ≤ 1, the maximum recall achieved by all methods. For each method, we then computed average enrichment across recall values (or average enrichment) as \(\frac{({\sum }_{i=1}^{{n}_{R}}{\mathrm{enrichment}}_{i}\times {(1-\mathrm{recall}}_{i}))}{{n}_{R}}\), where \({n}_{R}\) denotes the number of ‘true’ evaluation links corresponding to recall R. We obtained standard errors on average enrichment for each method, and on the difference in average enrichment between two methods, by bootstrapping genes using 1,000 bootstrap iterations.
We designated pgBoost as a reference to evaluate whether GLUE can achieve comparable results. Using the PBMC 10k multiome dataset from 10X Genomics, we extracted cells from four major cell types (CD14Mono, CD4TCM, CD4 naive and CD8 naive), which encompass more than 1,000 cells. We applied both GLUE (with and without TenK10K eQTL prior guidance) and pgBoost separately to the four cell types. By default, GLUE uses the top 6,000 highly variable genes for inferring peak–gene links. For benchmarking analysis, we further added the 1,949 genes from CRISPR validation datasets and resulted in 7,233 genes. This explains the difference in the number of peak–gene pairs between benchmarking using 10X multiome data and discovery analysis using TenK10K. Validation followed the same sets used in the pgBoost study, and we assessed method performance by evaluating whether predicted peak–gene pairs from each method scored higher for pairs present in the validation sets (Fig. 5a and Extended Data Fig. 9).
Second, we benchmarked the GRN inference results against CellOracle91 and SCENIC+46, two state-of-the-art methods that can integrate single-cell multiome data to infer GRNs. We applied GLUE (with and without incorporating TenK10K eQTL and caQTL for enhancement) and the two comparison methods to the 10X Genomics PBMC dataset, deriving a list of transcription factor–target gene pairs for each approach. For quantitative evaluation, we curated three validation sets from different types of evidence: gene expression changes following transcription factor knockout92, chromatin immunoprecipitation followed by sequencing data93,94,95 and functional protein–protein interactions96. Following established benchmarking practice97, we used EPR as the evaluation metric. Specifically, early precision is defined as the proportion of true-positive edges among the top-k-predicted edges. We computed EPR as the ratio of this early precision to the expected early precision of a random predictor, accounting for differences in network size and density. Error bars indicate 95% confidence intervals (1.96 × s.e.) computed from five independent runs using different random seeds.
Drug enrichment analysis
To assess whether single-cell caQTL data can support identifying drug targets, we performed an enrichment analysis of target–indication pairs using the Open Targets Platform98 (OTP; v25.12), following an approach analogous to Henry et al.99. We first used peak-to-trait SMR results to shortlist peaks with putative causal effects on selected disease traits, then integrated these with peak–gene SMR results to nominate candidate effector genes for each trait. We then intersected effector gene–trait pairs with target–indication pairs from OTP and extracted the highest clinical development phase achieved for drugs with mechanisms of action matching the relevant target–indication pair. We stratified pairs by maximum clinical phase reached (phases I, II, III, I+, II+, III+ and approved), and for each stratum constructed a 2 × 2 contingency table enumerating target–indication pairs supported or not supported by our caQTL-derived effector genes.
We note two important limitations. First, the number of target–indication pairs supported by both caQTL evidence and clinical validation (true positives) is small, so interpret the estimated odds ratios with caution. Second, and more fundamentally, the enrichment framework assumes that the absence of a caQTL-derived or eQTL-derived effector gene nomination for a given target–indication pair constitutes a negative result. However, this reflects incomplete data coverage rather than a true biological negative. Many nominally unsupported pairs may yet prove relevant as QTL mapping power improves or as additional traits are integrated. The enrichment results should therefore be interpreted as a lower-bound estimate of the concordance between QTL-nominated targets and established drug mechanisms.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
Individual-level data (raw sequencing outputs and processed count matrices from scATAC-seq and multiome) generated in this study are available through the European Genome–phenome Archive (EGA; EGAS50000001653). WGS and scRNA-seq data generated by the TenK10K flagship project and used in this study are available through the EGA; the study IDs are EGAS50000001653 (TenK10K phase I: single-cell data) and EGAS50000001654 (TenK10K phase I: WGS data). All key summary results, including peak annotations, caQTL summary statistics (common and rare variants), fine-mapping results (SuSiE and mvSuSiE), ChromBPNet model weights and predicted scores, cell-state-dependent caQTL summary statistics, colocalization results (coloc and SMR), and GLUE peak–gene and transcription factor–gene links, are available on the Hugging Face platform (https://huggingface.co/datasets/anglixue/TenK10K_multiome). Fine-mapped GTEx eQTL data are available on Finucane Lab (https://www.finucanelab.org/data). GTEx eQTL summary statistics are available on GTEx Portal (https://www.gtexportal.org/home/downloads/adult-gtex/qtl). The CRISPR datasets are available on GitHub (https://github.com/EngreitzLab/CRISPR_comparison/blob/main/resources/crispr_data/EPCrisprBenchmark_combined_data.training_K562.GRCh38.tsv.gz). The JASPAR 2026 vertebrate database is available (https://jaspar.elixir.no). The corresponding authors can provide any additional information needed to reanalyse the data reported in this paper on request. Azimuth reference is available on HuBMAP (https://azimuth.hubmapconsortium.org/references/#Human%20-%20PBMC).
References
Aguet, F. et al. Molecular quantitative trait loci. Nat. Rev. Methods Primers 3, 4 (2023).
Kanai, M. et al. Population-scale immune multiome atlas reveals regulatory disease mechanisms. Nature https://doi.org/10.1038/s41586-026-11078-2 (2026).
Yin, J. et al. Chinese Immune Multi-Omics Atlas. Science 391, eadt3130 (2026).
Cuomo, A. S. E., Nathan, A., Raychaudhuri, S., MacArthur, D. G. & Powell, J. E. Single-cell genomics meets human genetics. Nat. Rev. Genet. 24, 535–549 (2023).
Cuomo, A. S. E. et al. Rare and common genetic variants shape single-cell immune gene expression. Nature https://doi.org/10.1038/s41586-026-10721-2 (2026).
Watanabe, K. et al. A global overview of pleiotropy and genetic architecture in complex traits. Nat. Genet. 51, 1339–1348 (2019).
Neumeyer, S., Hemani, G. & Zeggini, E. Strengthening causal inference for complex disease using molecular quantitative trait loci. Trends Mol. Med. 26, 232–241 (2020).
Lloyd-Jones, L. R. et al. The genetic architecture of gene expression in peripheral blood. Am. J. Hum. Genet. 100, 228–237 (2017).
Võsa, U. et al. Large-scale cis- and trans-eQTL analyses identify thousands of genetic loci and polygenic scores that regulate blood gene expression. Nat. Genet. 53, 1300–1310 (2021).
Yazar, S. et al. Single-cell eQTL mapping identifies cell type-specific genetic control of autoimmune disease. Science 376, eabf3041 (2022).
Nathan, A. et al. Single-cell eQTL models reveal dynamic T cell state dependence of disease loci. Nature 606, 120–128 (2022).
Perez, R. K. et al. Single-cell RNA-seq reveals cell type-specific molecular and genetic associations to lupus. Science 376, eabf1970 (2022).
Natri, H. M. et al. Cell-type-specific and disease-associated expression quantitative trait loci in the human lung. Nat. Genet. 56, 595–604 (2024).
Haglund, A. et al. Cell state-dependent allelic effects and contextual Mendelian randomization analysis for human brain phenotypes. Nat. Genet. 57, 358–368 (2025).
Connally, N. J. et al. The missing link between genetic association and regulatory function. eLife 11, e74970 (2022).
Jeong, R. & Bulyk, M. L. Chromatin accessibility variation provides insights into missing regulation underlying immune-mediated diseases. eLife 13, RP98289 (2024).
Lee, S. & Mostafavi, H. Missing regulatory effects on complex traits: contribution of distal variants. Cell Genom. 5, 100809 (2025).
Chun, S. et al. Limited statistical evidence for shared genetic effects of eQTLs and autoimmune-disease-associated loci in three major immune-cell types. Nat. Genet. 49, 600–605 (2017).
Arthur, T. D. et al. Multiomic QTL mapping reveals phenotypic complexity of GWAS loci and prioritizes putative causal variants. Cell Genom. 5, 100775 (2025).
Mu, Z. et al. Impact of disease-associated chromatin accessibility QTLs across immune cell types and contexts. Cell Genom. 6, 101061 (2026).
Benaglio, P. et al. Mapping genetic effects on cell type-specific chromatin accessibility and annotating complex immune trait variants using single nucleus ATAC-seq in peripheral blood. PLoS Genet. 19, e1010759 (2023).
Wenz, B. M. et al. Genotype inference from aggregated chromatin accessibility data reveals genetic regulatory mechanisms. Genome Biol. 26, 81 (2025).
Alasoo, K. et al. Shared genetic effects on chromatin and gene expression indicate a role for enhancer priming in immune response. Nat. Genet. 50, 424–431 (2018).
Valencia, C. et al. Modeling heterogeneity in single-cell perturbation states enhances detection of response eQTLs. Nat. Genet. 57, 2882–2890 (2025).
Tanudisastro, H. A. et al. Tandem repeat variation shapes immune cell type-specific gene expression. Nature https://doi.org/10.1038/s41586-026-10438-2 (2026).
Heintzman, N. D. & Ren, B. Finding distal regulatory elements in the human genome. Curr. Opin. Genet. Dev. 19, 541–549 (2009).
Urbut, S. M., Wang, G., Carbonetto, P. & Stephens, M. Flexible statistical methods for estimating and testing effects in genomic studies with multiple conditions. Nat. Genet. 51, 187–195 (2019).
Wang, G., Sarkar, A., Carbonetto, P. & Stephens, M. A simple new approach to variable selection in regression, with application to genetic fine mapping. J. R. Stat. Soc. Series B Stat. Methodol. 82, 1273–1300 (2020).
Liang, Y., Aguet, F., Barbeira, A. N., Ardlie, K. & Im, H. K. A scalable unified framework of total and allele-specific counts for cis-QTL, fine-mapping, and prediction. Nat. Commun. 12, 1424 (2021).
Pampari, A. et al. ChromBPNet: bias factorized, base-resolution deep learning models of chromatin accessibility reveal cis-regulatory sequence syntax, transcription factor footprints and regulatory variants. Preprint at bioRxiv https://doi.org/10.1101/2024.12.25.630221 (2025).
Zhou, W. et al. Efficient and accurate mixed model association tool for single-cell eQTL analysis. Nat. Genet. https://doi.org/10.1038/s41588-026-02798-2 (2026).
Aldemir, H. et al. Cutting edge: lectin-like transcript 1 is a ligand for the CD161 receptor. J. Immunol. 175, 7791–7795 (2005).
Rosen, D. B. et al. Cutting edge: lectin-like transcript-1 is a ligand for the inhibitory human NKR-P1A receptor. J. Immunol. 175, 7796–7799 (2005).
Mathew, P. A. et al. The LLT1 receptor induces IFN-γ production by human natural killer cells. Mol. Immunol. 40, 1157–1163 (2004).
Lachmann, A. et al. ChEA: transcription factor regulation inferred from integrating genome-wide ChIP-X experiments. Bioinformatics 26, 2438–2444 (2010).
Giambartolomei, C. et al. Bayesian test for colocalisation between pairs of genetic association studies using summary statistics. PLoS Genet. 10, e1004383 (2014).
Parkes, M. et al. Sequence variants in the autophagy gene IRGM and multiple other replicating loci contribute to Crohn’s disease susceptibility. Nat. Genet. 39, 830–832 (2007).
Zhu, Z. et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat. Genet. 48, 481–487 (2016).
Yagi, M. et al. Bivalent chromatin instructs lineage specification during hematopoiesis. Cell https://doi.org/10.1016/j.cell.2025.05.011 (2025).
Stankey, C. T. et al. A disease-associated gene desert directs macrophage inflammation through ETS2. Nature 630, 447–456 (2024).
Wallace, C. A more accurate method for colocalisation analysis allowing for multiple causal variants. PLoS Genet. 17, e1009440 (2021).
Zou, Y., Carbonetto, P., Xie, D., Wang, G. & Stephens, M. Fast and flexible joint fine-mapping of multiple traits via the Sum of Single Effects model. Nat. Genet. 58, 454–462 (2026).
Liu, Y. C. et al. Scalable context-dependent single-cell eQTL mapping reveals disease-relevant regulatory variation beyond static models. Preprint at medRxiv https://doi.org/10.64898/2026.08.13.26360300 (2026).
Cardinale, C. J. et al. The ulcerative colitis risk gene adenylyl cyclase 7 restrains the T-helper 2 phenotype and class II antigen presentation. J. Crohns Colitis 19, jjaf030 (2025).
Aibar, S. et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017).
Bravo González-Blas, C. et al. SCENIC+: single-cell multiomic inference of enhancers and gene regulatory networks. Nat. Methods 20, 1355–1367 (2023).
Cao, Z.-J. & Gao, G. Multi-omics single-cell data integration and regulatory inference with graph-linked embedding. Nat. Biotechnol. 40, 1458–1466 (2022).
Kumasaka, N., Knights, A. J. & Gaffney, D. J. Fine-mapping cellular QTLs with RASQUAL and ATAC-seq. Nat. Genet. 48, 206–213 (2016).
Dorans, E., Jagadeesh, K., Dey, K. & Price, A. L. Linking regulatory variants to target genes by integrating single-cell multiome methods and genomic distance. Nat. Genet. 57, 1649–1658 (2025).
Oelen, R. et al. Single-cell RNA-sequencing of peripheral blood mononuclear cells reveals widespread, context-specific gene expression regulation upon pathogenic exposure. Nat. Commun. 13, 3267 (2022).
McManus, C. J., May, G. E., Spealman, P. & Shteyman, A. Ribosome profiling reveals post-transcriptional buffering of divergent gene expression in yeast. Genome Res. 24, 422–430 (2014).
Cusanovich, D. A., Pavlovic, B., Pritchard, J. K. & Gilad, Y. The functional consequences of variation in transcription factor binding. PLoS Genet. 10, e1004226 (2014).
Kiani, K., Sanford, E. M., Goyal, Y. & Raj, A. Changes in chromatin accessibility are not concordant with transcriptional changes for single-factor perturbations. Mol. Syst. Biol. 18, e10979 (2022).
Cuartero, S., Stik, G. & Stadhouders, R. Three-dimensional genome organization in immune cell fate and function. Nat. Rev. Immunol. 23, 206–221 (2023).
Kott, K. A. et al. Biobanking for discovery of novel cardiovascular biomarkers using imaging-quantified disease burden: protocol for the longitudinal, prospective, BioHEART-CT cohort study. BMJ Open 9, e028649 (2019).
Zhang, Y. et al. Model-based analysis of ChIP-seq (MACS). Genome Biol. 9, R137 (2008).
Neavin, D. et al. Demuxafy: improvement in droplet assignment by integrating multiple single-cell demultiplexing and doublet detection methods. Genome Biol. 25, 94 (2024).
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587.e29 (2021).
Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol. 42, 293–304 (2024).
Zhang, K., Zemke, N. R., Armand, E. J. & Ren, B. A fast, scalable and versatile tool for analysis of single-cell omics data. Nat. Methods 21, 217–227 (2024).
Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat. Methods 18, 1333–1341 (2021).
Moore, J. E. et al. An expanded registry of candidate cis-regulatory elements. Nature https://doi.org/10.1038/s41586-025-09909-9 (2026).
Zhang, K. et al. A single-cell atlas of chromatin accessibility in the human genome. Cell 184, 5985–6001.e19 (2021).
ENCODE Project Consortium. Expanded encyclopaedias of DNA elements in the human and mouse genomes. Nature 583, 699–710 (2020).
Danese, A. et al. EpiScanpy: integrated single-cell epigenomic analysis. Nat. Commun. 12, 5228 (2021).
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Risso, D., Schwartz, K., Sherlock, G. & Dudoit, S. GC-content normalization for RNA-seq data. BMC Bioinformatics 12, 480 (2011).
Zhou, H. J., Li, L., Li, Y., Li, W. & Li, J. J. PCA outperforms popular hidden variable inference methods for molecular QTL mapping. Genome Biol. 23, 210 (2022).
Taylor-Weiner, A. et al. Scaling computational genomics to millions of individuals with GPUs. Genome Biol. 20, 228 (2019).
Purcell, S. et al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet. 81, 559–575 (2007).
Asiimwe, R. & Alexander, D. STAR + WASP reduces reference bias in the allele-specific mapping of RNA-seq reads. Preprint at bioRxiv https://doi.org/10.1101/2024.01.21.576391 (2024).
Ovek Baydar, D. et al. JASPAR 2026: expansion of transcription factor binding profiles and integration of deep learning models. Nucleic Acids Res. 54, D184–D193 (2026).
Wu, Y. et al. Integrative analysis of omics summary data reveals putative mechanisms underlying complex traits. Nat. Commun. 9, 918 (2018).
Margoliash, J. et al. Polymorphic short tandem repeats make widespread contributions to blood and serum traits. Cell Genom. 3, 100458 (2023).
GTEx Consortium.The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 369, 1318–1330 (2020).
Setty, M. et al. Characterization of cell fate probabilities in single-cell data with Palantir. Nat. Biotechnol. 37, 451–460 (2019).
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat. Methods 16, 1289–1296 (2019).
Liu, Y. et al. ACAT: a fast and powerful P value combination method for rare-variant analysis in sequencing studies. Am. J. Hum. Genet. 104, 410–421 (2019).
Kanai, M. et al. Insights from complex trait fine-mapping across diverse populations. Preprint at medRxiv https://doi.org/10.1101/2021.09.03.21262975 (2021).
Weeks, E. M. et al. Leveraging polygenic enrichments of gene features to predict genes underlying complex traits and diseases. Nat. Genet. 55, 1267–1276 (2023).
Wang, Q. S. et al. Leveraging supervised learning for functionally informed fine-mapping of cis-eQTLs identifies an additional 20,913 putative causal eQTLs. Nat. Commun. 12, 3394 (2021).
Thakore, P. I. et al. Highly specific epigenome editing by CRISPR-Cas9 repressors for silencing of distal regulatory elements. Nat. Methods 12, 1143–1149 (2015).
Fulco, C. P. et al. Systematic mapping of functional enhancer-promoter connections with CRISPR interference. Science 354, 769–773 (2016).
Ulirsch, J. C. et al. Systematic functional dissection of common genetic variation affecting red blood cell traits. Cell 165, 1530–1545 (2016).
Wakabayashi, A. et al. Insight into GATA1 transcriptional activity through interrogation of cis elements disrupted in human erythroid disorders. Proc. Natl Acad. Sci. USA 113, 4434–4439 (2016).
Huang, J. et al. Dissecting super-enhancer hierarchy based on chromatin interactions. Nat. Commun. 9, 943 (2018).
Fulco, C. P. et al. Activity-by-contact model of enhancer-promoter regulation from thousands of CRISPR perturbations. Nat. Genet. 51, 1664–1669 (2019).
Gasperini, M. et al. A genome-wide framework for mapping gene regulation via cellular genetic screens. Cell 176, 377–390.e19 (2019).
Schraivogel, D. et al. Targeted Perturb-seq enables genome-scale genetic screens in single cells. Nat. Methods 17, 629–635 (2020).
Klann, T. S. et al. CRISPR-Cas9 epigenome editing enables high-throughput screening for functional regulatory elements in the human genome. Nat. Biotechnol. 35, 561–568 (2017).
Kamimoto, K. et al. Dissecting cell identity via network inference and in silico gene perturbation. Nature 614, 742–751 (2023).
Feng, C. et al. KnockTF 2.0: a comprehensive gene expression profile database with knockdown/knockout of transcription (co-)factors in multiple species. Nucleic Acids Res. 52, D183–D193 (2024).
Liu, Z.-P., Wu, C., Miao, H. & Wu, H. RegNetwork: an integrated database of transcriptional and post-transcriptional regulatory networks in human and mouse. Database 2015, bav095 (2015).
Han, H. et al. TRRUST v2: an expanded reference database of human and mouse transcriptional regulatory interactions. Nucleic Acids Res. 46, D380–D386 (2018).
Garcia-Alonso, L., Holland, C. H., Ibrahim, M. M., Turei, D. & Saez-Rodriguez, J. Benchmark and integration of resources for the estimation of human transcription factor activities. Genome Res. 29, 1363–1375 (2019).
Szklarczyk, D. et al. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res. 47, D607–D613 (2019).
Pratapa, A., Jalihal, A. P., Law, J. N., Bharadwaj, A. & Murali, T. M. Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat. Methods 17, 147–154 (2020).
Koscielny, G. et al. Open Targets: a platform for therapeutic target identification and validation. Nucleic Acids Res. 45, D985–D994 (2017).
Henry, A. et al. Single-cell genetics identifies cell type-specific causal mechanisms in complex traits and diseases. Preprint at medRxiv https://doi.org/10.1101/2025.08.28.25334614 (2025).
Schug, J. et al. Promoter features related to tissue specificity as measured by Shannon entropy. Genome Biol. 6, R33 (2005).
Acknowledgements
We thank K. Zhang for guidance on running the SnapATAC2 software; B. Ren for providing the quality-controlled scATAC-seq objects from ref. 63; Z. Cao for guidance on running GLUE; T. Stuart for guidance on running Signac; R. Satija for guidance on running Seurat and using the Azimuth reference; and M. Geaghan and E. Urng for bioinformatics support.
Funding
A.X. is supported by an NHMRC Investigator grant (2033018). J.E.P. is supported by an NHMRC Investigator grant (2034556) and a Fok Family Fellowship. D.G.M. is supported by an NHMRC Investigator grant (2009982). G.A.F. and the BioHEART study have been supported by an NHMRC Investigator grant, the NSW Health Office of Health and Medical Research, and the NSW Health Statewide Biobank scheme. The contents of this article are solely the responsibility of the authors and do not reflect the views of the NHMRC. This research was undertaken with the assistance of resources from the National Computational Infrastructure (NCI Australia), an NCRIS-enabled capability supported by the Australian Government, through the National Computational Merit Allocation Scheme (NCMAS; grant number NCMAS-2026-63). This research was also supported by a Ramaciotti Health Investment grant (2024HIG05).
Author information
Author notes
These authors contributed equally: Jianan Fan, Oscar A. Dong, Hao Lawrence Huang, Ling Chen, Peter C. Allen, Eleanor Spenceley
Authors and Affiliations
Translational Genomics Program, Garvan Institute of Medical Research, Sydney, New South Wales, Australia
Angli Xue, Jianan Fan, Oscar A. Dong, Hao Lawrence Huang, Peter C. Allen, Eleanor Spenceley, Eszter Sagi-Zsigmond, Blake Bowen, Albert Henry, Anna S. E. Cuomo, Zhen Qiao, Drew R. Neavin, Anne Senabouth, Caitlin Bartie, Rachael A. McCloy, Venessa Chin & Joseph E. Powell
Faculty of Medicine and Health, University of New South Wales, Sydney, New South Wales, Australia
Angli Xue, Jianan Fan, Hao Lawrence Huang, Peter C. Allen, Eleanor Spenceley, Blake Bowen, Albert Henry, Anna S. E. Cuomo, Hope A. Tanudisastro, Zhen Qiao, Anne Senabouth, Venessa Chin, Daniel G. MacArthur & Joseph E. Powell
Faculty of Science, University of Sydney, Sydney, New South Wales, Australia
Oscar A. Dong
UNSW Cellular Genomics Futures Institute, University of New South Wales, Sydney, New South Wales, Australia
Hao Lawrence Huang, Anna S. E. Cuomo & Joseph E. Powell
Illumina Artificial Intelligence Laboratory, Illumina, San Diego, CA, USA
Ling Chen, Eyal Ben-David & Kyle Kai-How Farh
Centre for Population Genomics, Garvan Institute of Medical Research, Sydney, New South Wales, Australia
Anna S. E. Cuomo, Hope A. Tanudisastro, Katrina M. de Lange & Daniel G. MacArthur
Centre for Population Genomics, Murdoch Children’s Research Institute, Melbourne, Victoria, Australia
Anna S. E. Cuomo, Hope A. Tanudisastro, Katrina M. de Lange & Daniel G. MacArthur
Faculty of Medicine and Health, University of Sydney, Sydney, New South Wales, Australia
Hope A. Tanudisastro
Genomics and Inherited Disease Program, Garvan Institute of Medical Research, Sydney, New South Wales, Australia
Zhen Qiao
Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, MA, USA
Elizabeth Dorans & Alkes L. Price
Harvard Medical School, Boston, MA, USA
Elizabeth Dorans
Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, MA, USA
Elizabeth Dorans, Linfeng Hu, Yijia Christiana Liu, Wei Zhou & Alkes L. Price
Stanley Center for Psychiatric Research, Broad Institute of MIT and Harvard, Cambridge, MA, USA
Linfeng Hu, Yijia Christiana Liu & Wei Zhou
Institute for Molecular Medicine Finland, University of Helsinki, Helsinki, Finland
Linfeng Hu, Yijia Christiana Liu & Wei Zhou
Massachusetts General Hospital, Boston, MA, USA
Arthur S. Lee
The Broad Institute of MIT and Harvard, Cambridge, MA, USA
Arthur S. Lee
The Kinghorn Cancer Centre, St Vincent’s Hospital, Sydney, New South Wales, Australia
Venessa Chin
Psychiatric and Neurodevelopmental Genetics Unit, Center for Genomic Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA
Wei Zhou
Novo Nordisk Foundation Center for Genomic Mechanisms of Disease, Broad Institute of MIT and Harvard, Cambridge, MA, USA
Wei Zhou
Department of Biostatistics, Harvard T.H. Chan School of Public Health, Boston, MA, USA
Alkes L. Price
Charles Perkins Centre, University of Sydney, Sydney, New South Wales, Australia
Gemma A. Figtree
Kolling Institute of Medical Research, Royal North Shore Hospital, Sydney, New South Wales, Australia
Gemma A. Figtree
Menzies Institute for Medical Research, University of Tasmania, Hobart, Tasmania, Australia
Alex W. Hewitt
Department of Ophthalmology, Royal Hobart Hospital, Hobart, Tasmania, Australia
Alex W. Hewitt
Authors
- Angli Xue
- Jianan Fan
- Oscar A. Dong
- Hao Lawrence Huang
- Ling Chen
- Peter C. Allen
- Eleanor Spenceley
- Eszter Sagi-Zsigmond
- Blake Bowen
- Albert Henry
- Anna S. E. Cuomo
- Hope A. Tanudisastro
- Zhen Qiao
- Elizabeth Dorans
- Eyal Ben-David
- Kyle Kai-How Farh
- Linfeng Hu
- Yijia Christiana Liu
- Drew R. Neavin
- Arthur S. Lee
- Anne Senabouth
- Caitlin Bartie
- Rachael A. McCloy
- Venessa Chin
- Wei Zhou
- Alkes L. Price
- Katrina M. de Lange
- Gemma A. Figtree
- Alex W. Hewitt
- Daniel G. MacArthur
- Joseph E. Powell
Contributions
A.X. and J.E.P. designed the study. A.X. designed and performed the single-cell processing and caQTL mapping pipelines and led the writing of the manuscript. A.X. and J.F. designed and performed the causal inference and GRN inference analyses. A.X. and O.A.D. designed and performed the caQTL and colocalization analyses. A.X., J.F. and O.A.D. designed and performed the replication analysis. A.X., B.B. and O.A.D. designed and performed the allele-specific analysis. H.L.H. designed and performed the dimensional reduction and fine-mapping analyses. E.S., E.S.-Z. and C.B. designed and performed the single-cell ATAC and multiome-seq experiments. A.X. and B.B. processed the single-cell ATAC-seq data. B.B. ran the 10X bioinformatics pipelines for read alignment. A.S.E.C. contributed the scRNA datasets and eQTL results. A.X. performed the quality controls and cell-type annotation of the scATAC-seq data, with help from J.F. and O.A.D. A.S.E.C., A.H. and J.E.P. designed and A.H. performed the cell-type specificity analysis. A.X. and O.A.D. designed and performed the rare caQTL analysis, with help from W.Z. H.A.T. first established the colocalization pipeline and conducted functional annotations. P.C.A., A.X., J.F. and O.A.D. designed and performed the cell-state-specific caQTL mapping. H.L.H. and A.X. designed and performed the cell function analysis with scDeepID. A.X. and J.F. designed and performed the GRN analysis. A.X., J.F., H.L.H., P.C.A., H.A.T., D.G.M. and J.E.P. contributed to downstream interpretation. J.F., H.L.H., B.B. and Z.Q. provided support with the caQTL mapping pipeline. B.B. optimized the mixQTL implementation. Z.Q. provided assistance with the pseudotime estimation and trajectory inference. L.C., E.B.-D. and K.K.-H.F. designed and performed the variant annotation using ChromBPNet. L.C. performed the identification of rare promoter–enhancer variants associated with outlier gene expression, with input from A.X. E.D. provided the curated validation datasets and conducted benchmarking analysis for pgBoost and GLUE, with help from J.F. and A.X. J.F. performed the benchmarking analysis for GRN inference methods with input from A.X. A.H. performed the drug enrichment analysis. L.H. improved the SAIGE-QTL software and computational efficiency. Y.C.L. developed the CASTIE method for interaction analysis. D.R.N. assisted with the scATAC-seq sample demultiplexing. A.S.L. contributed a curated list of regulatory annotations used to annotate the results. H.A.T. annotated caQTLs. A.S. contributed to the quality control of genotype array data. R.A.M. and V.C. designed the LBIO study. W.Z. helped implement SAIGE-QTL, designed the rare-variant mapping and interpreted the results. A.L.P. provided curated validation datasets and constructive guidance on benchmarking design and interpreting the results. K.M.d.L., D.G.M. and J.E.P. led the processing of the WGS data. G.A.F. designed the BioHEART study. A.W.H. designed the TOB study. A.X. drafted the manuscript. All authors contributed to editing the manuscript. A.X. and J.E.P. conceived the project. A.X. led the analytical team. J.E.P. supervised all aspects of the work.
Corresponding authors
Correspondence to Angli Xue or Joseph E. Powell.
Ethics declarations
Competing interests
L.C., E.B.D. and K.K.-H.F. are employed at Illumina. D.G.M. is a paid advisor to Insitro and GSK, and receives research funding from Google and Microsoft, unrelated to the work described in this article. G.A.F. reports grants from the National Health and Medical Research Council (Australia), Abbott Diagnostics, Sanofi, Janssen Pharmaceuticals and NSW Health; an honorarium from CSL, CPC Clinical Research, Sanofi, Boehringer-Ingelheim, Heart Foundation and Abbott; serves as Board Director for the Australian Cardiovascular Alliance (past President), Executive Committee Member for CPC Clinical Research, Founding Director and CMO for Prokardia and Kardiomics, and Executive Committee member for the CAD Frontiers A2D2 Consortium; serves as CMO for the non-profit CAD Frontiers, with industry partners including Novartis, Amgen, Siemens Healthineers, ELUCID, Foresite Labs, HeartFlow, Canon, Cleerly, Caristo, Genentech, Artyra, Bitterroot Bio, Novo Nordisk and Allelica; and has the following patents: ‘patent biomarkers and oxidative stress’ awarded USA May 2017 (US9638699B2) issued to Northern Sydney Local Health District, ‘use of P2X7R antagonists in cardiovascular disease’ PCT/AU2018/050905 licensed to Prokardia, ‘methods for treatment and prevention of vascular disease’ PCT/AU2015/000548 issued to The University of Sydney/Northern Sydney Local Health District, ‘methods for predicting coronary artery disease’ AU202290266 issued to The University of Sydney, and ‘novel P2X7 receptor antagonists’ PCT/AU2022/051400 (23 November 2022), International application no.: WO/2023/092175 (01 June 2023), issued to The University of Sydney. The other authors declare no competing interests.
Peer review
Peer review information
Nature thanks Caleb Lareau, Stephen Montgomery and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data figures and tables
Extended Data Fig. 1 Population-scale single-cell chromatin accessibility and multi-omics profiling across 922 donors identifies 28 immune cell types and 12 million genetic variants.
(a) Overview of the TenK10K cohort, sequencing strategy, and analysis workflow. Schematic created in BioRender; Xue, A. https://biorender.com/896774q (2026). (b) UMAP projection of scATAC-seq profiles annotated via label transfer from scRNA-seq, showing 28 immune cell types. (c) Distribution of the number of nuclei per library and donor. (d) Nuclei count per cell type after quality control. Top: total number of nuclei per cell type; Bottom: violin plots showing per-donor distributions. (e) Number of high-quality chromatin accessibility peaks identified per cell type. (f) Peak width distribution across all peaks. Inset: pie chart showing the proportion of peaks by distance to the nearest transcription start site (TSS).
Extended Data Fig. 2 Additional data description for TenK10K scATAC-seq.
(a) Integrated UMAP of TenK10K scRNA-seq and scATAC-seq data (n = 8,911,231 cells/nuclei). The non-overlapping clusters are driven by differences in CD14 monocyte proportions between the scATAC and scRNA datasets. A more detailed explanation can be found in the Methods section. (b) Cell type composition across all donors (n = 922). (c) Color scale reflects log2(RPKM + 1) chromatin accessibility. The x-axis represents each of the 8 major cell type groups. The y-axis represents the cell type-restricted peaks (n = 47,809) inferred from the Shannon entropy-based method100 as implemented in ref. 63. (d) Heatmap of cell type-specific marker peaks (n = 73,763) across 26 immune cell types (two rare types excluded). Color scale reflects log2(RPKM + 1) chromatin accessibility. (e) Transcription factor motif enrichment in marker peaks across 26 immune cell types (two rare types excluded). Color scale reflects −log10(p-value) of enrichment using a one-sided hypergeometric test. The dendrogram tree of the heatmap indicates the subclusters of transcription factors and cell types based on hierarchical clustering.
Extended Data Fig. 3 Schematic locus plot for cell-type specificity of caQTLs into five different scenarios.
This combined plot demonstrates the five cell type-specific caQTL scenarios. Each panel contains the caQTL boxplot in CD4TCM and CD14Mono and one locus plot showing the caQTL results in two cell types. The boxplot and scatter in grey indicate the CD4TCM results, and orange indicates the CD14Mono results. The vertical blue shade indicates the genomic interval of the peak tested. All p-values are obtained from a two-sided Student’s t-test. (a) Scenario 1 - specific caPeak: chr1:14871143-14871388. The peak is not open in CD4TCM, so no association test was performed. (b) Scenario 2 - specific caQTL: chr1:248,568,988-248,569,740. (c) Scenario 3 - distinct caQTL: chr13:24,670,806-24,672,096. (d) Scenario 4 - shared caQTL, concordant effect: chr19:50,607,800-50,608,746. (e) Scenario 5 - shared caQTL, discordant effect: chr4:40,246,472-40,247,824.
Extended Data Fig. 4 Replication and allele-specific validation of caQTLs.
(a) Replication of caQTLs in two published PBMC datasets and one in-house replication cohort, showing the degree of overlap and effect direction concordance. (b) Allele-specific validation of caQTLs across cell types. Validation outcomes for tensorQTL-significant lead caQTLs (q < 0.05): ASC concordant (peak-internal, sign-matching mixQTL ASC β); ASC discordant(peak-internal, opposite-sign mixQTL ASC β); Not ASC-measurable (peak-internal, fails the mixQTL default thresholds asc_cutoff = 5 and n_cutoff = 15, i.e. fewer than 15 donors contributing ≥ 5 reads to each haplotype within the peak); Not peak-internal (lead variant lies outside its peak). The left panel shows fractions; the right panel shows absolute counts. Cell types are ordered by total caQTL count. (c) Per-cell-type tensorQTL slope versus mixQTL ASC β (natural-log allelic fold change) at peak-internal ASC-measurable lead caQTLs, for the 12 cell types with ≥ 30 such caQTLs. Insets show n(number of peak-internal ASC-measurable lead caQTLs), Pearson r, Spearman ρ, and sign-concordance fraction. Red dashed line: y = x; blue line: linear regression fit.
Extended Data Fig. 5 Rare caQTLs stratified by MAF bins and highlighted example for CLEC2D.
(a) Number of significant rare variant caQTLs across 28 immune cell types. The number above the stacked bar represents the number of rare caQTLs before adjusting for the nearby top common variant. The grey bar indicates the caQTLs that are not significant after conditioning. The dark blue bar indicates those that remain significant after conditioning. The light blue bar indicates those that are significant only after conditioning. (b) Number of rare caPeaks stratified by the MAF of the top variant. Stacked bars show rare caPeaks that remain significant (red) or become non-significant (pink) after conditioning on the top nearby common variant. Left y-axis: count of rare caPeaks; right y-axis: proportion remaining significant after conditioning. (c) Red lollipop plot showing target CLEC2D gene expression z-scores, ordered from the most to least affected. The z-scores represent the difference in expression between carriers of the alternative allele versus the reference allele. Blue lollipop plot indicates the corresponding difference in ChromBPNet-predicted chromatin accessibility of peak chr12:9,650,649-9,651,305.
Extended Data Fig. 6 Comparison between coloc and SMR results for peak–gene pairs.
(a) posterior probability of shared (PP.H4) or independent (PP.H3) causal variants. (b) concordance between coloc posterior probability (PP.H4) and SMR p-values from a two-sided Chi-Squared test, showing strong agreement between the two approaches.
Extended Data Fig. 7 GWAS colocalization pattern using TenK10K eQTL and caQTL.
(a) GWAS colocalization pattern using TenK10K eQTL and caQTL as baseline. The stacked bar plot indicates the percentage of GWAS loci showing colocalization with eQTL or caQTL datasets. The baseline is TenK10K sc- eQTL and caQTL. The “Shared” group indicates those loci colocalizing with both TenK10K sc- eQTL and caQTL. Additional datasets are included in the order of bulk eQTLs from the eQTLGen consortium and GTEx whole body tissues. (b) GWAS and xQTL colocalization pattern using eQTLGen as the baseline. The stacked bar plot indicates the percentage of GWAS loci showing colocalization with eQTL or caQTL datasets. The baseline is the bulk eQTL from the eQTLGen consortium. Additional datasets are included in the order of TenK10K sc- eQTL, caQTL, and bulk eQTL from GTEx whole body tissues. (c) Cell type-specific colocalization signals for IBD GWAS. The stacked bar plot depicts the % of eQTL and caQTL colocalization signals with IBDG GWAS in each cell type. (d) The percentage of caQTL-specific colocalization per cell type is dependent on the cell type abundance level. The dot size represents the total number of nuclei for the cell type. The color of the dot indicates the cell type and the color scheme is consistent with Fig. 1b.
Extended Data Fig. 8 Locus plots of two highlighted genes.
(a) Locus plot at the ETS2 locus illustrating colocalization results using eQTL (±100 kb and ±1 Mb window) and caQTL (±1 Mb window) in CD14Mono. Dot color represents linkage disequilibrium (estimated by LD r2) with the lead GWAS SNP (21:39094644:G:A). Gene annotations (bottom) show genomic context and transcriptional orientation (blue: forward strand; green: reverse strand). (b) Locus plot at the GSDMB locus illustrating a GWAS signal that colocalizes with an eQTL, but not with a caQTL, in NKCD56bright and NKProliferating cells. The p-values are obtained from a two-sided Student’s t-test.
Extended Data Fig. 9 Cell type-specific eQTL guidance improves GLUE peak–gene inference relative to bulk and unmatched eQTLs.
(a) Comparison of GLUE-inferred regulatory scores with and without cell type-specific eQTL guidance. Peak-gene pairs are stratified by the presence or absence of prior regulatory evidence. (b) GLUE scores for peak-gene pairs inferred using different eQTL sources. The first comparison contrasts GTEx bulk blood eQTLs with cell type-specific CD4Naive eQTLs in the 10× Genomics PBMC multiome dataset. The second compares two cell type-specific eQTL sets (CD14Mono and Treg) using Treg cells from the TenK10K multiome dataset. (c) Comparison of GLUE and GLUE with eQTL guidance for peak-gene linkage inference in CD14Mono. We stratify pairs by CRISPR-based regulatory support and quantify performance by AUPRC (Area Under the Precision-Recall Curve). Peak-gene pairs without and with CRISPR support, respectively: GLUE 274 and 123; GLUE with eQTL guidance 274 and 123. (d) The same comparison for CD4TCM population. The number of peak-gene pairs without and with CRISPR support, respectively: GLUE 132 and 62; GLUE with eQTL guidance 143 and 57.
Extended Data Fig. 10 GRN inference with prior QTL information.
(a) Expansion of GRNs through integration of caQTL and SMR-based peak–gene links, identifying 128 additional TF-target gene pairs in CD14Mono. Highlighted regulators include IKZF1 and STAT1, which modulate APOBEC3A via caQTL-mediated mechanisms and represent potential drug repurposing targets. (b) Genome-browser tracks for the CD55 locus. The SMR, GLUE, and distance tracks show ATAC peaks linked to the CD55 promoter in CD14Mono, using different sources of linking evidence. Specifically, peaks were assigned to CD55 based on caQTL-eQTL SMR (p-value < 5 × 10−8, two-sided Chi-Squared test), GLUE (score > 0.8), and genomic proximity (±10 kb of the TSS). Curved lines connect CD55 to inferred peaks under each method.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Xue, A., Fan, J., Dong, O.A. et al. Genetic variants shape single-cell chromatin accessibility and disease risk. Nature (2026). https://doi.org/10.1038/s41586-026-11173-4
Received: 12 August 2025
Accepted: 16 September 2026
Published: 28 October 2026
Version of record: 28 October 2026
DOI: https://doi.org/10.1038/s41586-026-11173-4
来源:Nature 综合主刊 · nature.com