MainGenome-wide association studies (GWASs) have revealed hundreds of thousands of association signals between genetic variants and human diseases and traits4, but the modes of action by which most of these variants affect phenotypes remain to be characterized, especially non-coding variants. The characterization of the impact of genetic variants on gene expression—known as expression quantitative trait locus (eQTL) mapping—can help linking variants to the putative genes they regulate1,5,6,7,8,9, thus identifying the genes and pathways involved in disease pathogenesis, which is a critical step in identifying opportunities for therapeutic intervention10.However, important limitations remain. First, eQTL mapping has traditionally been limited to common genetic variation (minor allele frequency (MAF)) greater than 1–5%) due to relatively small sample sizes and the widespread use of genotyping arrays, which do not accurately genotype most rare variants and complex structural variants11, compared with whole-genome sequencing (WGS) approaches. Second, eQTLs are extremely context specific, with many studies demonstrating that genetic variants frequently affect gene expression levels in ways detectable only in specific tissues3, sorted cell types6,10,12,13 or on stimulation14,15. Single-cell eQTL mapping has further demonstrated that the effects of many genetic variants on gene expression can only be detected within specific cell types and even transient cellular states3,16,17,18,19,20,21,22,23.Here we introduce phase 1 of the TenK10K (up to 10,000 cells for 10,000 individuals) study, encompassing 5.4 million cells from 1,925 individuals of European ancestry selected from a broader cohort of 2,103 individuals. In accompanying papers, we leverage the same dataset to investigate the effect of tandem repeat variation on gene expression24 and the putative causal links between gene expression measurements and human diseases and traits25, and an overlapping set of samples to explore the impact of genetic variation on chromatin accessibility26.Here we use an association method we recently developed, SAIGE-QTL27, to map common-variant and rare-variant single-cell eQTLs across immune cell types. We identify more than 150,000 common-variant effects and more than 30,000 rare-variant effects on gene expression, including signals for nearly 5,000 genes (roughly 25%) with no previously reported eQTLs in a large blood eQTL study5. We propose a framework for quantifying the extent of cell-type specificity or sharedness of eQTLs and show that about half of the effects are found in only one or a few cell types, and almost none are ubiquitous across all. We perform statistical colocalization between our eQTLs and 60 disease phenotypes and blood traits, identifying more than 60,000 colocalization events for more than 5,000 genes, providing potential molecular mechanisms in almost 900 genes associated with human phenotypic variation. We use a biologically driven deep learning method, scDeepID, to identify continuous cell states related to cell function and map more than 15,000 eQTLs that vary dynamically along those states. Finally, we identify 164 independent genetic variants that regulate the abundance of cell states.The TenK10K phase 1 cohortAfter quality control, a total of 5,438,679 peripheral blood mononuclear cells (PBMCs) from 1,925 donors were considered for analysis (Fig. 1, Supplementary Table 1 and Methods). We collected and sequenced PBMCs from all individuals, yielding an average of 2,825 cells per individual after quality control (Supplementary Figs. 1 and 2 and Methods).Fig. 1: Population-scale scRNA-seq and matched WGS from 1,925 individuals identified 28 immune cell types and 66 million variants.Full size imagea, Overview of the data used in this paper introducing phase 1 of the TenK10K cohort. High-quality matched WGS and scRNA-seq data were generated for 1,925 individuals of European ancestry (after quality control). After computational analysis and processing, we conducted both common and rare-variant eQTL analysis and identified cell-type-specific and state-specific effects of genetic regulation in immune cells. We also performed colocalization analysis with GWAS from 60 traits and diseases and mapped QTLs affecting csaQTL. b, Accompanying papers perform complementary analyses using the TenK10K phase 1 cohort. In brief, ref. 25 integrates sc-eQTLs from the present study and 100 GWAS studies to explain the causal role of gene expression in immune cell types across complex traits; ref. 24 uses the WGS data to generate a catalogue of more than 1.7 million polymorphic tandem repeats and dissect their effects on single-cell-resolved gene expression; ref. 26 combines common genetic variants called from WGS with scATAC-seq generated for a subset of roughly 1,000 individuals to investigate the genetic regulation of chromatin accessibility at cellular resolution. c, Schematic of the 28 immune cell types identified, organized by main cell type. d, The number of cells available for analysis by cell type for the 28 immune cell types is summarized in d. Top, bar plot showing total cell numbers per cell type on a log10 scale; bottom, violin plots showing the distribution across individuals on a log10 scale. ASDC, AXL+SIGLEC6+ dendritic cells; cDC1/2, conventional type 1/2 dendritic cells; CD4CTL, CD4+ cytotoxic T cells; dnT, double-negative T cells; gdT, γδ T cells; ILC, innate lymphoid cells; MAIT, mucosal-associated invariant T cells; pDC, plasmacytoid dendritic cells. Panels created in BioRender: a, Cuomo, A. https://biorender.com/0bkvg8x (2026); c, Cuomo, A. https://biorender.com/fyk570b (2026).Each cell was assigned to their donor of origin by means of demultiplexing and to an immune cell type using a combined approach of reference-based classification and hierarchical clustering as recommended by the single-cell eQTLGen Consortium28 (Methods). Each cell was assigned to one of 28 different cell types, spanning the key expected immune lineages (Fig. 1c, Supplementary Fig. 3 and Supplementary Table 2).In addition, we generated WGS data for all 1,925 individuals, which was processed and quality-controlled using established pipelines (Methods). We called 66,658,242 sites across single nucleotide polymorphisms (SNPs) and insertions and deletions (indels), of which 53,784,018 (80.6%) were rare (MAF 1%) variant sites within ±100 kilobases (kb) of the gene body on single-cell expression profiles (testing all sufficiently expressed genes; Methods). We used SAIGE-QTL27 to map eQTL effects using single-cell profiles modelled as Poisson counts, rather than relying on ‘pseudobulk’ aggregation. We mapped eQTLs separately in each of the 28 cell types and confirmed the type I error was controlled using permutations (Fig. 2a). At a false discovery rate (FDR) of less than 5%, we identified 154,932 eQTLs in at least one cell type from 17,674 unique eGenes (83% of all genes tested; Fig. 2b and Supplementary Fig. 5). The number of eQTLs identified varied substantially across cell types, largely due to the number of cells available for each cell type and the resulting number of individuals tested affecting statistical power, as previously shown in refs. 16,27 (Fig. 2b, Extended Data Fig. 1 and Supplementary Table 2).Fig. 2: Common and rare eQTL mapping in 28 immune cell types.Full size imagea, Quantile–quantile plot for all common (MAF = 1%, top) and rare (MAF 0.9, using SuSiE2; Methods) signals (credible sets) were categorized as either primary (including the top most significant variant) or non-primary. Non-primary signals are on average found further away from the gene’s TSS, less likely to be found in promoter regions (c) and more likely in enhancer regions compared with primary signals (d). In c, violin plots represent distribution over n = 17,674 independent signals, and the horizontal lines in the box plots therein represent the first quartile, median and third quartile. e, Proportion of eQTLs identified in each cell type by means of fine mapping, coloured by the number of independent signals (‘only one’ or ‘two or more’).We performed fine mapping using SuSiE29 (Methods), showing an extra 24,925 (a 16% increase) non-primary (secondary, tertiary and so on)30,31 eQTL signals for 7,131 eGenes (40.4%). We used 95% credible sets to quantify the number of independent signals for each eGene, and explored differences between primary effects (variants from the credible set containing the top variant for each eGene) and non-primary effects (all other credible sets). We find that primary effects contain larger credible sets, and genes in the higher-proportion cell types have substantially more non-primary effects (Fig. 2e and Supplementary Fig. 6). As shown by others31, non-primary signals were found further from the transcription start site (TSS) and more likely in enhancer as opposed to promoter regions (Fig. 2c,d).We compared the eGenes we identified with those from bulk blood eQTL studies. First, we considered the eQTLGen consortium study5: the largest eQTL study so far in whole blood. We identified a comparable number of eGenes compared to eQTLGen (17,674 and 16,987, respectively), despite the 15-fold smaller sample size (1,925 compared with more than 30,000 individuals; Extended Data Fig. 2a). More than one-quarter of our eGenes were not identified in eQTLGen (n = 4,572, 25.9%); those genes were slightly more lowly expressed, and their expression was more cell-type-specific (Extended Data Fig. 2b,c) compared with eGenes found by both studies; TenK10K-unique eGenes were also more likely to be eGenes in fewer cell types overall, in rare cell types (n