Decoding Immune Regulation: From Genetic Variation to Mechanism Through Single-Cell Genomics Authors: Jung Hee Koh, Chun Jimmie Ye Published: 2026 ABSTRACT AbstractImmune cell states are not fixed. Rather, they emerge from dynamic transcriptional programs shaped by genetic variation, cellular context, and gene regulatory networks (GRNs). Single-cell and multi-omic technologies now enable population-scale immune profiling across molecular layers, revealing that cell-to-cell transcriptional variability is a functional feature that diversifies immune responses. Distribution-aware and tensor-based analytical frameworks capture this variability beyond mean expression, resolving coordinated gene programs across cell types, individuals, and conditions. Integrating human genetics with single-cell genomics demonstrates that genetic effects on gene expression, splicing, and chromatin accessibility are highly dependent on cell type, activation state, and differentiation trajectory. These variant-level signals converge on GRNs in which key transcription factors orchestrate context-dependent immune programs. High-throughput perturbation screens enable scalable functional validation of these networks, linking genetic variation to cellular function. Together, these integrative approaches translate molecular discoveries into clinical applications, from patient stratification to therapeutic target prioritization. Emerging spatial multi-omics and in situ perturbation screens further resolve neighbor-dependent regulation within intact tissue niches, offering a path from variant to mechanism to clinical translation.Keywords: Genetics, Human; Single-cell gene expression analysis; Immunogenetic phenomena; Gene regulatory networks; Quantitative trait loci INTRODUCTION The immune system is an inherently complex network composed of dynamic and heterogeneous cell populations that continuously sense, integrate, and respond to external pathogens and internal danger signals (1). Precise regulation of these circuits is crucial, as the cumulative effect of genetic variants and environmental triggers can disrupt immune tolerance, leading to the development of autoimmune diseases (2). Genome-wide association studies (GWAS) have identified thousands of risk loci for autoimmune diseases (3). However, linking these genetic variants to specific cellular functions remains challenging because genetic effects are often context-dependent and appear only in particular cell types or activation states (4). Traditional bulk profiling averages signals across heterogeneous populations, thereby masking genetic effects that are specific to rare or transient states in which genetic effects may manifest (5). The advent of single-cell technologies has revolutionized our ability to map previously unrecognized cell states at unprecedented resolution (5). High-throughput single-cell profiling of transcriptomes, chromatin accessibility, and protein epitopes enables comprehensive analysis of dynamic immune processes (6). These approaches have begun to resolve detailed immune cell states, uncovering activation trajectories and regulatory programs (6,7). Despite these advances, translating single-cell and multi-omic datasets into biological insights is difficult due to the analytical complexity of computational workflows, the cost of data generation, and the challenge of distinguishing true biological variation from technical noise (8). In addition, computationally defined cell states often lack direct experimental validation, thereby limiting their mechanistic interpretation and clinical relevance (9). To address these challenges, this review describes how integrating human genetics with single-cell genomics reshapes our view of immune heterogeneity and uncovers context-dependent genetic regulation. We highlight distribution-aware and multidimensional frameworks that capture cell-to-cell variability beyond mean expression. We also summarize analytical approaches that link GWAS signals to candidate effector genes, including cell-type- and state-specific expression quantitative trait loci (eQTLs), dynamic eQTLs whose effects vary along pseudotime and stimulation gradients, and molecular phenotypes beyond expression, such as RNA splicing and chromatin accessibility. Finally, we discuss how gene regulatory network (GRN) inference helps interpret cell-state-specific regulatory programs and outline strategies for validating computational findings. Overall, this review provides an integrative perspective that clarifies a path for translating complex genomic data into clinically relevant applications in immune-mediated diseases. TECHNOLOGICAL ADVANCES IN SINGLE-CELL IMMUNE PROFILING Recent advances in single-cell genomics have transformed RNA sequencing technologies from low-throughput transcript capture to massively multiplexed and population-scale profiling (Fig. 1). The advent of droplet-based microfluidics and sample multiplexing, enabled by oligo-based cell hashing and genotype-based demultiplexing, has dramatically increased both cellular throughput and cohort scale, supporting the parallel profiling of millions of cells while reducing batch effects and costs (10). These advances increase power to detect rare immune cell subsets in large cohorts (11). Technological and analytical advances in decoding immune cell heterogeneity. Innovations in single-cell genomics enable scalable and multimodal immune profiling. Advanced computational frameworks, including distribution-aware methods and tensor decomposition, enable quantification of cell-to-cell transcriptional variability beyond mean expression and resolve coordinated gene programs across cell types, individuals and conditions. These approaches reveal that transcriptional variability is a functional feature of immune cells. sci-CAR, single-cell combinatorial indexing chromatin accessibility and RNA-seq; ASAP-seq, assay for transposase-accessible chromatin with select antigen profiling by sequencing; TEA-seq, transcription, epitope, and accessibility sequencing; DOGMA-seq, central dogma sequencing; scMNT-seq, single-cell multi-omics nucleus tomography sequencing; scDD, single-cell differential distribution; BASiCS, Bayesian analysis of single-cell sequencing data; IDEAS, individual digital evaluation of allele-specific expression. Parallel to transcriptomic advances, epigenetic landscape profiling has been adapted to single-cell technologies. For example, single-cell assay for transposase-accessible chromatin (scATAC) using sequencing maps accessible chromatin regions across different cell types and states, revealing their regulatory potential (12). To profile specific histone modifications, targeted methods such as single-cell cleavage under targets and tagmentation were developed to provide high sensitivity and resolution (13). Building upon this, multifactorial profiling approaches, including multiple target identification by tagmentation, and nanobody-tethered transposition followed by sequencing, simultaneously map multiple chromatin epitopes in the same cells, facilitating the assessment of cell-specific combinatorial chromatin states (14,15). Beyond single-modality profiling, joint measurement of RNA and chromatin accessibility links gene expression with cis-regulatory elements within the same cell, strengthening element-to-gene inference at scale (16). Similarly, cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) simultaneously measures surface proteins and gene expression, linking classical immunophenotypes with transcriptional states (17), while assay for transposase-accessible chromatin with select Ag profiling by sequencing pairs chromatin accessibility profiling with protein phenotypes, enabling the association of epigenetic landscapes with protein markers (18). Integrated trimodal assays further enable simultaneous quantification of three distinct molecular layers. Technologies such as transcription, epitopes, and chromatin accessibility by sequencing and DOGMA-seq, simultaneously measure the transcriptome, chromatin accessibility, and protein profiles in the same cell (18,19). Furthermore, single-cell nucleosome, methylation, and transcription sequencing extends single-cell epigenetic profiling to DNA methylation, chromatin accessibility, and transcription (20). These multimodal methods improve cell-type and cell-state resolution and enable direct tests of regulatory congruency and uncoupling across two or more molecular layers (Fig. 1). Single-cell genomics has continued to expand in both molecular resolution and spatial contexts. Long-read single-cell RNA sequencing (scRNA-seq) improves the reconstruction of full-length transcript isoforms, enabling the detection of alternative splicing variants that are often missed by short-read assays (21). Moreover, spatial transcriptomics restores the anatomical context lost in dissociated single-cell assays by mapping cell states in situ within tissue sections and microanatomical niches (22). Recent advances in spatial transcriptomics have addressed a major barrier to clinical translation by enabling in situ total RNA capture in formalin-fixed paraffin-embedded tissues using random primers at single-cell resolution (23). In immunology, these advances are particularly useful for testing how the local tissue microenvironment shapes immune cell activation states and cell-cell interactions and for localizing where the specific cell states identified by single-cell profiling are spatially enriched (24). Beyond observations, perturbation-based single-cell assays enable the functional validation of candidate genes, regulatory elements, and putative causal variants at scale (10). In pooled single-cell clustered regularly interspaced short palindromic repeats (CRISPR) screens, Perturb-seq uses each cell as an independent experimental unit and captures a guide barcode alongside the single-cell transcriptome, enabling researchers to link specific genetic perturbations to complex cellular phenotypes (25). Together, these technological and analytical advances establish a multimodal foundation for resolving immune cell heterogeneity at single-cell resolution. DECODING IMMUNE CELL HETEROGENEITY Functional implications of transcriptional variability In bulk RNA-seq, variability across samples is often dismissed as technical noise. However, single-cell profiling has revealed that cell-to-cell variability is a functional feature that diversifies immune responses (26). Even among genetically identical immune cells, stochastic transcription and epigenetic heterogeneity can create variable activation thresholds, allowing the immune system to respond flexibly while remaining tightly regulated to specific stimuli (26,27). This functional heterogeneity is evident in the innate immune system. Classic single-cell imaging and transcriptomic analyses have further demonstrated that key cytokine genes are expressed in only a fraction of LPS-stimulated dendritic cells (DCs) rather than uniformly across the population. This heterogeneity reflects the stochastic activation of an antiviral transcriptional program involving Irf7 and Stat2, propagated through a paracrine IFN-feedback loop, allowing a small number of first-responder cells to alert the remaining cells (28). Similarly, variability plays a key role in fine-tuning adaptive immune response activation. Eling et al. (29) quantitatively modeled mean–variance relationships in scRNA-seq data to identify genes whose mean-corrected residual dispersion changes upon activation of CD4+ T cells. They found that while biosynthetic and metabolic genes became more uniform as cells entered a coordinated activation program, key immunoregulatory and effector genes exhibited increased dispersion in their expression. This reflects a mechanism of controlled heterogeneity that diversifies effector functions within clonally expanding populations (29). Beyond activation, this transcriptional variability drives the lineage bifurcation critical for determining long-term cell fate. During viral infection, the differentiation of CD8+ T cells into terminal effector or memory precursor populations is initiated by a pronounced transcriptional divergence immediately following the first cell division (30). This early asymmetry is subsequently stabilized by epigenetic silencing of memory-associated determinants, effectively consolidating distinct functional lineages (30). Similarly, in germinal center B cells, single-cell profiling reveals dynamic transitions between the light and dark zones (31) and identifies transcriptionally distinct precursor subpopulations for memory B cells and plasma cells that emerge within the light zone (32). Distinguishing true transcriptional variability from clonal diversity is a unique challenge in adaptive immunity, where heterogeneous gene expression can reflect either dynamic cell transitions or stable differences between clonotypes (33). To disentangle cell state dynamics from clonal identity, paired scRNA-seq and single-cell V(D)J sequencing link transcriptomic states to TCR or B cell receptor (BCR) clonotypes (34). This paired profiling resolves clonal expansion across transcriptional states, suggesting that Ag specificity is associated with cell differentiation and functional states (35). For BCR repertoires, somatic hypermutation-based lineage reconstruction further connects clonal evolution to B cell differentiation states (36). Collectively, these studies demonstrate that immune regulation leverages transcriptional variability to fine-tune the activation thresholds, differentiation trajectories, and effector diversity. Therefore, capturing gene expression distributions across time and clonal lineages requires a shift toward distribution-aware computational frameworks. Distribution-aware analysis: from mean to variance While standard differential expression (DE) and pseudobulk approaches are robust for detecting shifts in average expression, they can miss distributional changes that underlie cell-to-cell variability (37,38). To address these limitations, recent single-cell methods test distributional differences beyond the mean and often separate technical noise from true biological variability. Bayesian analysis of single-cell sequencing data estimates residual overdispersion and identifies cell-to-cell variability changes between conditions, consistent with subset-driven or heterogeneous responses (39). Single-cell differential distribution considers each gene’s expression as a mixture of sub-distributions and classifies between-group differences into distributional patterns (38). Methods such as distinct and individual digital evaluation of allele-specific expression handle random donor effects, mean–variance dependence, and single-cell-specific confounders, such as donor-level structure and single-cell sparsity, linking distributional shifts to disease mechanisms (40,41). Moment-based DE analysis is a method-of-moments framework that tests differences in mean, variability, and gene–gene correlation at scale (42). This enables detection of immune regulatory effects, such as activation-associated transcriptional stochasticity, even when the average expression remains unchanged. It can also detect subtle donor-associated changes in gene regulation that are not apparent in classical pseudobulk analyses (42). This method has been applied to CRISPR-perturbed CD4+ T cells to map gene regulatory interactions underlying T cell activation, effectively reconstructing parts of the GRN from changes in variability and co-variation (42). Tensor decomposition: uncovering coordinated patterns across patients Although gene-wise variability metrics provide granular insights, they often fail to capture simultaneous, coordinated changes across multiple axes, such as cell subtypes, donor groups, and conditions. To resolve this multidimensional covariation structure, researchers have used tensor decomposition (43). Unlike standard matrix factorization, this approach models the data as a multidimensional array (e.g., patient × gene × cell type), identifying coordinated programs across specific cell lineages and patient groups (44). For instance, by applying Tucker tensor decomposition to a 3D tensor (patient × gene × cell type), Mitchel et al. (45) identified coordinated multicellular patterns linked to systemic lupus erythematosus (SLE) severity and clinical phenotypes, such as lupus nephritis. This demonstrates how tensor analysis can stratify heterogeneous patient populations based on molecular endotypes, complementing clinical classifications. However, conventional tensor methods assume aligned cell axes across conditions, which is inconsistent with single-cell data characterized by unmatched cell numbers and compositions across patients and conditions. To overcome this, advanced algorithms such as RISE (adapting PARAFAC2) address the practical challenges posed by an unaligned cell dimension across conditions (46). This approach leverages full datasets to highlight disease-associated inflammatory programs, including contributions from less abundant cell populations, helping connect high-dimensional single-cell variation to clinically relevant stratification. These distribution- and tensor-based frameworks provide quantitative summaries of coordinated programs across cells and patients. To understand the genetic basis of this transcriptional heterogeneity, we discuss how inherited variants regulate gene expression and chromatin states in a cell-type- and state-dependent manner (Fig. 2). Translating context-dependent genetic regulation into mechanistic insights and clinical applications. sc-eQTL mapping reveals that the effects of genetic variants on gene expression are highly dependent on cellular context. Genetic regulatory effects can be specific to distinct cell lineages or activation states. Furthermore, dynamic eQTLs continuously shift in effect size and direction along differentiation trajectories or pseudotime. Additional QTL frameworks are summarized in Table 1. These variant-level signals are integrated into GRNs and functionally validated through targeted genetic experiments and perturbation-based single-cell screens, enabling causal inference and systematic state-resolved validation at scale. The mechanistic insights are translated into clinical applications, including patient stratification, causal target prioritization, and optimization of cell-based therapies. SNP, single nucleotide polymorphism. GENETIC REGULATION OF IMMUNE CELL STATES: FROM eQTL TO MOLECULAR PHENOTYPES Cell-type and state-specific genetic effects GWAS risk variants are predominantly enriched in non-coding regions with largely unexplored regulatory functions (47). eQTL mapping establishes links between genetic variants and gene expression levels (Table 1). Integrating eQTL with colocalization and transcriptome-wide association studies (TWAS) helps connect regulatory variations to disease phenotypes (48,49). Table 1Overview of molecular QTL frameworks and their biological interpretations in single-cell genomicsFrameworkData modalityKey modelBiological interpretationRepresentative referencessc-eQTLscRNA-seqDonor-level aggregation by cell typeGenetic variants associated with gene expression in cell type/state; effect sizes/directions can differ across cell types/statesvan der Wijst et al. (2018) (73)csaQTLscRNA-seqDonor-level association testing of cell-state abundanceVariants associated with shifts in the relative abundance of transcriptionally defined cell states (composition)Rumker et al. (2024) (55)vQTLscRNA-seqMean-variance modeling within a cell type/stateVariants associated with changes in cell-to-cell variability (noise/dispersion)Sarkar et al. (2019) (58)Dynamic QTLscRNA-seq (time-course)Trajectory-based modeling of gene-variant effecteQTLs where the effect size varies continuously according to continuous axes (i.e., time, stimulation dose, pseudotime/latent state score)Nathan et al. (2022) (61), Soskic et al. (2022) (64)co-eQTL/network QTLscRNA-seqCo-expression network analysisVariants that modulate the co-expression strength or pairwise correlation between genes, indicating rewiring of gene regulatory networksvan der Wijst et al. (2018) (73)sQTLscRNA-seq (5’ or full-length)Splicing ratio/PSI quantificationLoci associated with variation in transcript isoform ratios or splicing eventsTian et al. (2024) (68)caQTLscATAC-seq, multiome (RNA + ATAC)Association between genotype and chromatin accessibilityVariants altering chromatin accessibility, providing a mechanistic explanation for colocalized eQTLsBenaglio et al. (2023) (76)CrisprQTLPerturb-seqCRISPR interference with scRNA-seqEstablishes causal links between non-coding elements and target genesGasperini et al. (2019) (79)csaQTL, cell-state abundance quantitative trait locus; PSI, percent spliced-in; sQTL, splicing quantitative trait locus; ATAC, assay for transposase-accessible chromatin. Overview of molecular QTL frameworks and their biological interpretations in single-cell genomics csaQTL, cell-state abundance quantitative trait locus; PSI, percent spliced-in; sQTL, splicing quantitative trait locus; ATAC, assay for transposase-accessible chromatin. Before the advent of single-cell technologies, complementary strategies addressed cellular heterogeneity by profiling sorted immune cell subsets to map cell-type-specific eQTLs (4,50) and by using deconvolution to infer cell type and genotype interactions (51). Parallel studies have established context dependence by exposing cells to external stimuli, revealing that eQTL effect sizes can change dynamically with environmental cues (52). Single-cell sequencing has resolved cellular heterogeneity, enabling precise mapping of cell-type- and cell-state-specific genetic regulation. Single-cell expression quantitative trait locus (sc-eQTL) analyses have demonstrated that genetic effects vary significantly in magnitude and direction across immune cell lineages (53,54). For instance, Yazar et al. (53) reported that the rs4987360-G allele showed opposite allelic effects across lineages, decreasing SELL expression in monocytes while increasing it in a B cell subset. In the same study, most trans-acting eQTLs exhibited cell-type-specific effects (53). Similarly, an autoimmune risk locus at TNFRSF14 (peak SNP rs8725) was associated with reduced TNFRSF14 expression only in activated CD4+ T cell subsets, demonstrating activation state-dependent genetic regulation (54). Beyond allelic effects, non-genetic covariates, such as biological sex, can also shape activation dependent gene expression across CD4+ T cell subsets, helping to explain sex-biased autoimmune risk (54). In addition to per-cell regulatory effects on gene expression (eQTLs), genetic variants can manifest through changes in cell-state abundance (55). Recent single-cell frameworks, such as GeNA, have been used to identify five loci that influence the relative abundance of immune cell states (55). For instance, rs3003-T is associated with an increased abundance of NK cells that express tumor necrosis factor response programs. This locus colocalizes with psoriasis risk, linking genetically driven compositional changes to disease susceptibility (55). Variance quantitative trait loci (vQTLs) quantify how inherited variants modulate cell-to-cell variability within a given cell type or state (56). Genotype-dependent increases in variability may reflect reduced expression stability and an increased propensity to enter pathogenic states (56). Alternatively, vQTLs can uncover gene–environment interactions when genetic effects differ across unmeasured contexts (57). However, early single-cell vQTL analyses highlighted that vQTL signals could be confounded by mean effects, emphasizing the need for explicit mean–variance decoupling and adequate sample sizes (58). Dynamic genetic effects Beyond discrete cell-type- and cell-state-specificity, modeling genetic effects along continuous cellular states, such as differentiation or activation gradients, cytokine dose, and gene co-regulatory module activity, reveals dynamic quantitative trait loci (QTLs) whose effect sizes vary along these axes. Methods such as CellRegMap and Struct-LMM capture these genotype–context interactions using continuous or multivariate cellular state variables (59,60). These approaches have revealed a multilayered regulatory architecture in which the effect of a genetic variant depends not only on the cell type but also on its transcriptional and epigenetic contexts. For instance, approximately one-third of cis-eQTLs in memory T cells are state-dependent. Their effect sizes are modulated by continuous cytotoxicity and regulatory capacity scores rather than by discrete cell lineages, such as CD4 versus CD8 (61). In B cells, dynamic eQTLs reveal that genetic regulatory effects vary along the pseudotime trajectory from immature and naïve B cells to memory B cells (53). For instance, BLK encodes a Src-family tyrosine kinase that modulates downstream BCR signaling strength, and BLK variants are associated with the risk of SLE (62). Notably, BLK eQTLs show the greatest allelic effects in late B cell states (53). Similarly, dynamic eQTL effects vary across immune cell states, even within the T cell compartment, particularly for major histocompatibility complex class II genes. The HLA-DQA1 eQTL (rs3104371) showed the strongest effects in cytotoxic T cell states (γδ T cells, cytotoxic CD4+ T cells, and cytotoxic CD8+ T cells), supporting dynamic genetic regulation along cytotoxic activation programs (63). Expanding on activation dynamics, time-course sc-eQTL mapping during CD4+ T cell activation showed that immune disease risk variants frequently acted as dynamic eQTLs, with allelic effects emerging or disappearing across stimulation time points (64). A representative example is an eQTL for TYK2 detected in 16 h-stimulated effector memory T cells (TEM) that colocalized with Crohn’s disease GWAS association. Carriers of the Crohn’s disease-protective allele showed lower TYK2 expression in TEM cells at 16 h of activation (64). Finally, this approach has potential applications in personalized medicine. In PBMCs exposed to immune stimuli and glucocorticoids, the transcriptional pseudotime was measured to quantify the per-cell drug response. This analysis identified dynamic eQTLs in which genetic variants showed different regulatory effects across the initial, middle, and late stages of the response. This result helps explain the inter-individual variability in glucocorticoid sensitivity, distinguishing responders from non-responders based on their genetic regulation dynamics (65). Genetic effects on splicing and chromatin regulation While eQTLs provide a direct link between genetic variants and gene regulation, colocalization analyses across three major immune cell populations suggest that only ~25% of autoimmune risk loci share genetic effects with cis-eQTLs (66). Consequently, splicing quantitative trait loci (sQTLs), in which genetic variants affect alternative splicing to determine which proteins or isoforms are produced, have emerged as compelling targets for systematic mapping across disease-relevant cell types and stimuli (67). sQTLs can reshape immune function through isoform selection, often independently of the total expression level (68). For instance, colocalization analyses in stimulated macrophages across risk loci associated with 22 immune-mediated diseases showed that ~20% of risk loci colocalized only with sQTL signals (compared to 9% for eQTL only) (69). Notably, approximately half of the colocalized sQTL signals arise from low-usage splice junctions, which are often considered noise in conventional analyses (69). To capture this complexity, recent workflows combine short-read based discovery with long-read RNA sequencing, which can provide end-to-end isoform structures and validate low-usage splice junctions, thus improving isoform detection and quantification for downstream genetic analyses (70,71). In addition to changing the levels or isoforms of single genes, genetic variants can modulate gene–gene co-regulation. Co-expression quantitative trait loci (co-eQTLs) influence the correlation strength between gene pairs or within gene regulatory modules (72). Single-cell sequencing enables the construction of personalized co-expression networks, linking edge strength to donor genotypes (73). For example, in PBMCs, the top co-eQTL, rs7297175, markedly altered the co-expression of RPS26 and RPL21 in CD4+ T cells. This co-eQTL is in strong linkage disequilibrium with a known type 1 diabetes risk variant, suggesting that disease susceptibility may arise from regulatory network disruption rather than a shift in mean expression (73). Furthermore, in pathogen-challenged monocytes, co-regulatory wiring is context-dependent and shaped by cell type, exposure time, and genetic background (74). Extending QTL mapping to additional molecular layers helps trace the propagation of genetic effects through regulatory mechanisms. Mapping chromatin accessibility quantitative trait loci (caQTLs) at single-cell resolution has shown that genetic effects are often highly context-dependent (75,76). For example, the G allele of rs1867687 increases chromatin accessibility in B cells and monocytes, whereas the A allele has the opposite effect in T and NK cells. This divergent effect is masked in bulk assays (76). Mu et al. (77) built a unified PBMC scATAC map and used topic-defined continuous cell states to identify dynamic caQTLs whose effects vary along cellular trajectories, such as the activation continuum of CD8+ effector memory T cells observed in coronavirus disease 2019 (COVID-19) patients. Notably, immune-mediated disease GWAS loci colocalized more often with caQTLs than with eQTLs. However, most GWAS-colocalizing caQTLs showed no corresponding gene expression effect in the same cell type, which limits target–gene assignment from chromatin data alone (77). Recent large-scale single-nucleus multiome studies have allowed joint mapping of caQTLs and eQTLs in the same donors, facilitating the inference of chromatin-to-expression regulatory relationships (78). While these integrated maps further refine GWAS interpretation, target–gene assignment often remains unresolved. To bridge the gap between regulatory elements and their downstream targets, CRISPR-based perturbations with expression readouts can directly map noncoding cis-regulatory elements to their target genes. Gasperini et al. (79) introduced crisprQTL mapping, which couples multiplexed clustered regularly interspaced short palindromic repeats interference (CRISPRi) with scRNA-seq to associate perturbed elements with transcript-level changes. They identified 664 cis-regulatory enhancer–gene pairs, and one-third of enhancer–gene pairs skipped at least one proximal transcription start site of another expressed gene, highlighting the limitations of nearest-gene assignment (79). In primary CD4+ T cells, CRISPRi perturbation of an enhancer upstream of PHF19 implicated TRAF1 as the putative effector gene at an rheumatoid arthritis (RA) GWAS locus, rather than the nearest gene (PHF19) (80). Collectively, extending QTL mapping across splicing, co-expression, and chromatin layers has clarified where and when genetic variants act within dynamic immune contexts, while CRISPRi-based element-to-gene mapping helps nominate target genes for noncoding signals. However, most QTL frameworks remain associative and do not fully specify regulatory programs that propagate through gene networks to shape immune cell states. FROM VARIANTS TO REGULATORY NETWORKS GRN inference and validation As observed through co-eQTLs (72), heterogeneity does not occur in isolation, but is shaped by complex GRNs in which regulators are connected to target genes by interaction edges (81). These networks determine immune cell development by linking extrinsic cues to intracellular signaling cascades and various regulators, including transcription factors (TFs), chromatin modifiers, microRNAs, and long noncoding RNAs (82). Genetic variants often converge on key regulatory nodes within these networks, propagating their effects through cis- and trans-regulatory cascades to reshape immune cell states (72). Bulk transcriptome studies first mapped co-regulated gene modules and canonical regulatory circuits, particularly in myeloid cells stimulated by Toll-like receptors (83). Advances in single-cell and multi-omic technologies have enabled the reconstruction of developmental lineages and context-dependent programs (84). To unravel the regulatory program changes underlying these states, tools such as SCENIC define cell states by estimating TF-regulon activity (85), whereas CellOracle integrates chromatin accessibility with network inference for in silico perturbation (86). Pando combines multi-omic data with sequence-based predictions of TF-binding sites to infer a global GRN during organoid development (87). These computational frameworks enable the prioritization of candidate regulators, which can subsequently be validated through targeted genetic studies to establish causal regulatory interactions. Complementing these computational approaches, experimental studies have established causal roles for specific TFs in shaping immune cell identity and plasticity. For instance, under chronic T cell stimulation, TOX and NR4A emerge as key TFs that reorganize chromatin to lock in the exhaustion program, paradoxically supporting T cell persistence (88). Targeted genetic perturbations causally link these regulatory networks to exhaustion, suggesting TOX and NR4A modulation as potential therapeutic strategies for immunotherapy (89). In Tregs, Foxp3 acts not as a single master regulator but as an amplifier of a pre-existing Treg identity. Depending on combinatorial interactions with other TFs, Foxp3 can either activate Treg programs or suppress conventional T cell programs (90). Within this regulatory framework, Smarcc1 has been identified as a controller of activated Treg-specific chromatin accessibility (90). The immune system often repurposes the same GRN modules across lineages. For instance, innate lymphoid cells use a similar set of master TFs as helper T cells, despite the lack of T cell Ag receptors (91). In tissue-resident macrophages, identity is enforced by niche-specific signaling. Microglia adopt a unique identity via SALL1 and SMAD4, which drive microglia-specific functions of the TGFβ–SMAD signaling axis (92), whereas alveolar macrophages differentiate under lung-derived GM-CSF, which induces peroxisome proliferator-activated receptor γ for their development and maintenance (93). The principle of tissue adaptation extends to the adaptive immune system. Treg tissue adaptation follows a continuum in which cells first engage a shared core non-lymphoid tissue program enriched for TNFRSF–NF-κB signaling, subsequently acquiring distinct skin- or intestine-specific signatures (94). This illustrates that cell-to-cell variability reflects a structured process of tissue adaptation, with gene regulatory programs tuned by environmental cues (94). Scalable functional validation through perturbation-based single-cell assays While the combination of GRN inference and conventional genetic experiments provides causal support for selected regulators, pooled perturbation screens with single-cell readouts enable systematic state-resolved validation at scale (25,95). These approaches, broadly referred to as Perturb-seq, couple pooled CRISPR perturbations with scRNA-seq by capturing guide barcodes or single-guide RNA sequences to assign perturbations to individual cells. This framework allows researchers to analyze how perturbing key TFs reshapes gene programs across heterogeneous activation states (25). For instance, Perturb-seq in stimulated myeloid cells revealed opposing effects of lineage-determining TFs (Cebpb vs. Irf8) on monocyte/macrophage versus DC programs. It also identified distinct roles of Rela versus Stat1/2 in shaping inflammatory versus antiviral response modules. These results suggest lineage-biased wiring of stimulus responses, where monocytes favor inflammatory modules, and DCs favor antiviral modules (95). In primary human T cells, genome-wide pooled CRISPR knockout screens identified regulators of TCR signaling following stimulation (96). Notably, the loss of RASA2 enhanced T cell functional responses, nominating RASA2 as a potential target to improve anti-tumor immunity (96). Multimodal extensions such as expanded CRISPR-compatible CITE-seq (ECCITE-seq) connect perturbation-driven transcriptional changes to surface protein phenotypes, enabling the dissection of how regulatory perturbations propagate from signaling to cell-surface and effector outputs (97). In an ECCITE-seq screen of immune checkpoint regulation, simultaneous measurement of PD-L1 (CD274) mRNA and surface protein levels under IFN-γ stimulation allowed stratification of regulators according to whether they primarily affect PD-L1 transcription or protein stability (98). Notably, multiple perturbations shifted PD-L1 protein abundance without appreciable changes in PD-L2 or CD86, illustrating how multimodal single cell screens can distinguish transcriptional from protein-level regulation for checkpoint phenotypes (98). Genome-scale Perturb-seq targeting all expressed genes with CRISPRi systematically maps the transcriptional consequences of perturbations, assigning genes to functional modules and uncovering new regulators of ribosome biogenesis, transcription, and mitochondrial respiration (99). It also dissects composite phenotypes such as aneuploidy and chromosomal instability, while revealing stress-specific regulation of the mitochondrial genome (99). FROM CELL STATE TO PATIENT PHENOTYPE: TRANSLATING MOLECULAR SIGNATURES INTO CLINICAL INSIGHT Stratifying patients and predicting outcomes Clinical translation of single-cell signatures often begins with patient stratification. Perez et al. (11) demonstrated that in SLE, type 1 IFN–stimulated gene expression defines a distinct IFN-high patient subset, primarily driven by classical monocytes. This molecular stratification has clinical relevance because higher baseline IFN signatures correlate with greater disease activity and flare risk (100). Recently, trials for IFN-blocking biologics have utilized these signatures as predictive biomarkers to identify subgroups likely to show greater clinical benefit (101). Similar cell state-to-phenotype links define RA heterogeneity. Anti-citrullinated protein Ab (ACPA)-positive RA is characterized by robust type II IFN response signatures and adaptive immune cell crosstalk, whereas a proinflammatory innate immune signature with a distinct type I IFN-stimulated gene signature predominates in ACPA-negative RA (102,103). Beyond these subtypes, cell–cell communication analysis reveals that disease activity also drives heterogeneity through distinct signaling programs. Specifically, type II IFN pathway activity is increased in RA (104). The dominant senders of type II IFN signaling are naïve CD4+ T cells and NK cells in patients with moderate to high disease activity, whereas classical and IL-1β monocytes dominate in those with low disease activity (104). Single-cell signatures can predict outcomes in patients with drug-free remission. An increase in circulating CD45RO+PD1hi T cells and CD27+CD86+CD21− B cells predicts a flare characterized by T cell-specific clonal expansion and accumulation of functionally impaired Treg cells (105). Mapping variants and prioritizing targets Integrating single-cell genomics with GWAS helps shift the field from enrichment-based interpretations to causal inference and therapeutic target prioritization. Colocalization frameworks have identified putative causal genes, such as CTLA4 (associated with RA and type 1 diabetes), and mapped their effects on specific T cell subsets (53). CTLA4 haploinsufficiency causes immune dysregulation in humans, supporting its central role in immune tolerance (106). In addition, the clinical efficacy of CTLA4-Ig in RA provides independent therapeutic validation of CTLA4 as a target (107). Building on locus-level colocalization, single-cell genetics enables transcriptome-wide target prioritization by combining immune cell sc-eQTLs with Mendelian randomization (MR) and colocalization. For instance, Ying et al. (108) incorporated pathway analysis, protein–protein interaction data, and clinical-trial evidence into a tier system to prioritize immune-mediated targets for COVID-19. Using this framework, NCR3 was prioritized as a drug candidate with immune cell specificity across COVID-19 severity (108). A similar sc-eQTL-based MR and Bayesian colocalization workflow was applied in SLE, where candidate causal genes with eQTLs were further evaluated using phenome-wide association studies to flag potential off-target effects and cross-reference with existing compounds to nominate potentially lower-risk therapeutic targets (109). Furthermore, integrating sc-eQTL summary statistics with epigenomic annotations and 3D chromatin interaction information enables cell-type-level TWAS to prioritize candidate risk genes with cell-type-specific effects and nominate drug-repurposing candidates (110). In parallel, a joint model of bulk-eQTLs and sc-eQTLs improves power and stabilizes cell-type-specific sc-eQTL effect estimates, strengthening target discovery (111). For example, in Crohn’s disease, cell-type-specific TWAS prioritized IRF3 as a novel target gene with significant effects in NK cells (111). Moving from discrete cell types to continuous cell states, TWAS methods applied to pseudotime trajectories show that gene–trait associations can vary along differentiation (112). These dynamic associations are highly specific to disease context and cell lineage. For instance, the HLA region showed the strongest dynamic association with RA during the naïve-to-memory transition of CD4+ T cells, whereas PYCARD showed the strongest dynamic association with SLE during B cell differentiation (112). However, pleiotropy remains a challenge for causal inference, as a single variant can influence multiple biological pathways. To address this, integrating frameworks combining MR with context-specific single-cell DE can further refine target prioritization (113). For example, analyzing CD4+ T cell activation time points revealed that disease-associated genetic effects vary dynamically over time, which helps to disentangle pleiotropic signals and prioritize therapeutic targets (113). Finally, beyond germline GWAS, VEXAS syndrome illustrates how somatic variations can be linked to distinct immune cell states. Single-cell transcriptomics of bone marrow cells have shown that inflammatory activation and myeloid bias occur early at the hematopoietic stem and progenitor cell stages and are enriched in UBA1-mutant cells (114). Genotyping of transcriptomes indicates that disease-associated transcriptional signatures are not confined to mutant cells but are also present in wild-type cells within the same patient (115). In addition, single-cell genotype-to-phenotype somatic mapping suggests that UBA1-mutant progenitors depend on the PERK arm of the unfolded protein response, proposing PERK pathway inhibition as a potential therapeutic strategy (116). Optimizing therapeutic T cell engineering Beyond target discovery, perturbation-based screens can directly inform therapeutic cell engineering strategies. Pooled CRISPR screening revealed that targeted chromosome loss occurs in Cas9-edited human T cells (117). Notably, non-activated T cells maintain higher mean p53 expression during Cas9 editing than activated T cells. This elevated p53 activates the DNA damage response, which selectively eliminates cells with chromosome loss (117). Based on this finding, performing genome editing in non-activated T cells was proposed as a practical workflow to minimize genotoxicity in clinical settings (117). Together, these approaches illustrate the next generation of precision medicine, in which targets are selected based on cell-type-specific causal evidence rather than generalized associations. CONCLUSIONS AND FUTURE PERSPECTIVES The integration of single-cell genomics with human genetics has fundamentally redefined our understanding of immune regulation. We have moved from viewing immune cells as static lineages defined by surface markers to recognizing them as dynamic states shaped by context-dependent regulatory programs. Genetic variation not only shifts gene expression levels but also modulates cellular transitions, rewires GRNs, and ultimately drives disease susceptibility through highly cell-type- and state-specific mechanisms. Scalable single-cell multiomics and perturbation screens support the functional validation of immune regulatory programs, translating molecular discoveries into clinical applications (Fig. 2). Despite this progress, most scalable single-cell and perturbation assays rely on tissue dissociation, losing the local spatial context, which obscures cell density effects and neighbor-dependent regulation. Therefore, integrating spatial multi-omics and in situ perturbation represents a key next step (118). Emerging spatially resolved CRISPR screens now enable pooled perturbations while preserving cellular neighborhoods, disentangling intracellular pathways and intercellular networks within intact tissue niches (119). Ultimately, the convergence of population-scale genetics, multimodal profiling, and spatial perturbation will provide a unified map from variant-to-function-to-phenotype, offering a more direct path to pathogenesis and precision therapeutic interventions. ACKNOWLEDGEMENTS This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (RS-2023-00246642) awarded to Jung Hee Koh. Figures were created using BioRender. Koh, J. (2026). BioRender.com/hdh9xd2. Abbreviations anti-citrullinated protein Ab B cell receptor chromatin accessibility quantitative trait locus cellular indexing of transcriptomes and epitopes by sequencing co-expression quantitative trait locus coronavirus disease 2019 clustered regularly interspaced short palindromic repeats clustered regularly interspaced short palindromic repeats interference dendritic cell differential expression expanded CRISPR-compatible CITE-seq expression quantitative trait locus gene regulatory network genome-wide association studies Mendelian randomization quantitative trait locus rheumatoid arthritis single-cell assay for transposase-accessible chromatin single-cell expression quantitative trait locus single-cell RNA sequencing systemic lupus erythematosus splicing quantitative trait locus effector memory T cells transcription factor transcriptome-wide association studies variance quantitative trait locus Footnotes Conflict of Interest: The authors declare no potential conflicts of interest. Author Contributions: Conceptualization: Koh JH, Ye CJ.Data curation: Koh JH.Writing - original draft: Koh JH.Writing - review & editing: Koh JH, Ye CJ. Conceptualization: Koh JH, Ye CJ. Conceptualization: Koh JH, Ye CJ. Data curation: Koh JH. Data curation: Koh JH. Writing - original draft: Koh JH. Writing - original draft: Koh JH. Writing - review & editing: Koh JH, Ye CJ. Writing - review & editing: Koh JH, Ye CJ. References