A multi-ethnic reference map of T cell receptor germline diversity reveals evidence of natural selection on alpha chain genes

Abstract

T cell receptors (TCR) orchestrate adaptive immunity, yet the complex, repetitive architecture of the TCR loci has impeded systematic characterization of human genetic variation in the genes encoding the TCR. Using public long-read sequencing data from the Human Pangenome Reference Consortium and All of Us consortia spanning 2719 donors, we build a near-complete map of common alleles in TCR V, D, and J genes, revealing amino acid variation at almost every position within V genes. We observe allele frequency differences between populations for many individual TCR genes. We present evidence of natural selection on TCR genes, including signals of balancing selection and positive selection in the alpha chain locus. We find TCR allelic polymorphism alters core functional properties of T cells, including thymic fate commitment and cell-surface receptor abundance. Collectively, these findings position inherited variation in TCR genes as a key axis of immunological diversity that may shape interindividual differences in immune responses.

In this study, population-scale long-read sequencing reveals extensive germline polymorphism in T cell receptor V and J genes. This variation carries signatures of natural selection and shapes single-cell T cell phenotypes, underscoring its functional importance.

Introduction

Few regions in the human genome have been as profoundly shaped by evolutionary forces as the HLA locus^1,2. Over millennia, pathogen-driven selective pressures have sculpted extraordinary levels of genetic polymorphism in HLA class I and class II genes. This variation underlies interindividual differences in immune responses, with the HLA locus harboring more GWAS associations than any other genomic region^3,4.

However, antigen presentation is only one step of the adaptive immune response. The activation of T cells depends on the recognition of antigen–HLA complexes by the T cell receptor (TCR). A TCR chain is assembled by recombining gene segments chosen from the set of multiple V (variable), D (diversity), and J (joining) genes available in the corresponding locus. While the complementarity-determining region (CDR) 3 region of a TCR chain is partially generated through somatic recombination, most of the TCR sequence is encoded by these germline V, D, and J gene segments. The V gene provides the framework regions that form the structural backbone of the TCR and encodes the CDR1 and CDR2 which contact antigen-HLA complexes, while the D and J genes both contribute to the CDR3 sequence⁵. Many T cell responses are dominated by clonotypes sharing the same germline-encoded V or J gene, with residues in these genes determining key contact residues and binding geometry^5–9. Thus, germline genetic variation in TCR genes could be an important target of natural selection and may shape interindividual differences in immune responses^10–12.

The loci encoding TCR chains, including the TRA/D locus (chr14), the TRB locus (chr7), and the TRG locus (chr7), have a complex genetic architecture marked by extensive gene duplications with high sequence homology among paralogs. For example, the TRB locus encodes 67 Vβ genes, several of which are identical or nearly identical in sequence¹³. Short-read technologies are stymied by multi-mapping and are incapable of accurately surveying polymorphism in the TCR loci and other highly repetitive genomic loci^11,14–21. Thus, the TCR loci remain amongst the least characterized regions of the human genome; the largest studies to date survey fewer than 50 individuals, leaving human genetic diversity in TCR genes poorly understood^10,12,22,23. Furthermore, the lack of a complete reference map capturing comprehensive genetic variation within the TCR loci has prevented the potential discovery of genetic signals for human diseases¹⁰. Recent advancements in long-read sequencing technologies present a unique opportunity to overcome these challenges and faithfully characterize human genetic polymorphism in these key immune loci^22,23.

Here, we leverage long-read sequencing data from 2719 donors to reveal extensive human genetic diversity in TCR V, D, and J genes. We discover this polymorphism has been shaped by natural selection and influences core functional phenotypes of T cells, establishing inherited variation in TCR genes as a meaningful axis of immunological diversity (Fig. 1a).

Fig. 1. Uncovering the allelic diversity of human TCR genes.

a Schematic of study design. (Created in BioRender. Mantena, S. (2026) https://BioRender.com/fcyibc2). b Number of common V (left), D (center), and J (right) gene alleles observed as a function of the number of individuals sequenced. c Total number of known and novel common alleles for TCR V, D, and J gene groups. d One-field allelic diversity of TCR V genes. Top: Number of common alleles observed per gene. Bottom: Frequency distribution of one-field alleles for each gene; each color denotes a distinct allele.

Results

Population-scale discovery of TCR genetic diversity

To discover allelic polymorphisms in TCR loci, we harnessed PacBio HiFi sequencing data and corresponding haplotype-resolved assemblies generated from 2492 individuals from the NIH All of Us CDRv8 release and 227 individuals from the Human Pangenome Reference Consortium release 2^24,25 (Supplementary Note 1). We focused on discovering V, D, and J gene alleles, excluding pseudogenes incapable of contributing to functional TCR chains. Following International Immunogenetics Information System (IMGT)²⁶ convention, we defined alleles at one-field resolution based on nucleotide differences in the core coding exons (IMGT -REGION). Because noncoding regulatory variation in leader sequences, splice sites, and recombination signal sequences can alter TCR gene usage and functionality,^27–30 we also defined TCR alleles at ‘two-field resolution’ using nucleotide differences across extended IMGT gene units encompassing these flanking features (IMGT L-V-GENE-UNIT for V genes, D-GENE-UNIT for D genes, and J-GENE-UNIT through J-DONOR-SPLICE for J genes) (Methods).

At one-field resolution, we observed 638 total V gene alleles, 6 total D gene alleles, and 160 total J gene alleles within our cohort (Methods; Supplementary Fig. 1a). Of these, 491 V gene alleles and 79 J gene alleles were not present in the IMGT/GENE-DB database and were thus classified as novel. These alleles were named based on the VDJbase/OGRDB interim convention, wherein the variants that distinguish the allele from the closest known allele are listed in the name. This represents a 334% increase in the number of V gene alleles and a 98% increase in the number of J gene alleles relative to the IMGT/GENE-DB database (Supplementary Fig. 1b). We specifically focused on characterizing common alleles, which we defined as being present at an allele frequency of at least 0.01 in at least one of the five supergroup populations (AFR, AMR, EAS, EUR, SAS). At one-field resolution, 326 V gene alleles, 6 D gene alleles, and 109 J gene alleles were common (Fig. 1b-c). These common V, D, and J gene alleles were directly supported by an average of 10,828, 47,371, and 43,573 reads, respectively (Supplementary Fig. 2 and Supplementary Note 1). Rarefaction curves indicate our analysis captures approximately 93% of common TCR V, D, and J gene alleles in each of the five major ancestry groups (Fig. 1b; Supplementary Fig. 3), enabling us to perform the first population-scale interrogation of the genetic architecture of the human TCR loci.

To validate alleles in independent data, we collected publicly available single-cell TCR sequencing data and found 99.4% of V gene and 100% of J gene alleles common in donors of European ancestry could be observed in human repertoires (Supplementary Note 2; Supplementary Fig. 1g-h). Furthermore, of those alleles that were not observed in the IMGT/GENE-DB database release used in this study, 127 V gene alleles and 19 J gene alleles were independently observed in a recent study that profiled the expressed TCR alleles of 45 donors¹².

Coding diversity in TCR genes

First, we sought to characterize the broad diversity of TCR gene classes. The 102 TCR V genes had an average of 3.20 common one-field alleles, with some genes harboring up to nine, while the 73 J genes had an average of 1.49 common one-field alleles (Fig. 1d). We quantified the variability of each gene across the population with the Shannon entropy of its one-field allele frequencies. TCR V genes were more entropic than J genes, and TRAV genes specifically exhibited greater entropy than TRBV genes (Supplementary Fig. 3d). When examining locus-wide variant calls, the TRA, TRB, and TRG loci exhibited SNP densities 1.67-, 1.46-, and 2.25-fold higher, respectively, than those of the chromosomes on which they reside, highlighting their unusually elevated polymorphism (Supplementary Fig. 3e). Donors of African ancestry exhibited the highest levels of diversity, consistent with previous observations³¹ in other genomic regions.

We constructed dendrograms of TRAV and TRBV genes; as expected, alleles of each gene clustered together (Fig. 2a-b). Next, we examined the distribution of nonsynonymous changes within TCR V genes. In our cohort, 41.4% of TRAV genes and 40.4% of TRBV genes carried nonsynonymous mutations in their CDRs, while 93.1% of TRAV genes and 97.9% of TRBV genes carried nonsynonymous mutations in their framework regions (Supplementary Fig. 3f-g). Because all TRAV genes encode the same component of the α chain, we aggregated functional alleles across TRAV genes and observed that nearly every position exhibited amino acid variation, a pattern that was also observed for TRBV genes (Fig. 2c-d). This remarkable coding diversity, reflecting both the divergence between genes and the pervasive allelic diversity within them, underscores the breadth of antigen-recognition potential encoded in the germline repertoire.

Fig. 2. Codon-level diversity in TCR V genes.

a Dendrogram of common functional TRAV alleles. Each leaf represents a distinct one-field allele, labeled by gene name. Novel alleles are in dark blue and marked with a black dot at the leaf tip. b Analogous to (a), but for common functional TRBV gene alleles. c Amino acid sequence logo for functional TRAV gene alleles. Sites with no amino acid present represent gapped positions. d Analogous to (c), but for TRBV gene alleles. e Variance in amino acid physicochemical characteristics across functional TRAV gene alleles. Atchley factors were used to numerically represent physicochemical characteristics, and variance was minimax normalized for each Atchley factor. Columns in grey represent gapped positions. f Analogous to (e), but for TRBV genes.

The CDRs, as expected, are a focal point of this variation, which is consistent with their established role in contacting peptide–HLA complexes (Fig. 2). Prior studies have demonstrated CDR mutations can alter recognition of key epitopes^32,33, and the extensive genetic polymorphism we observe within them likely modulates antigen recognition across diverse contexts.

The framework regions, which serve as the structural backbone of the TCR, are typically considered relatively invariant compared to the CDRs³⁴. However, the framework regions also displayed substantial sequence heterogeneity (Fig. 2). To determine whether framework diversity reflected changes among chemically similar amino acids or functionally distinct ones, we used Atchley factors³⁵ to numerically represent physicochemical characteristics. Several framework codons exhibited surprisingly high variance in these physicochemical traits (Fig. 2e-f). For example, among all sites, including those in the CDRs, IMGT position 91 in the framework 3 region of TRAV genes exhibited the greatest variance in amino acid hydrophobicity and IMGT position 69 in the framework 3 region of TRBV genes exhibited the greatest variance in amino acid charge.

To understand how this diversity is distributed across human populations, we performed a PCA of donor one-field TCR V and J genotypes, which revealed separation of donors by their ancestry (Fig. 3a). Several TCR alleles exhibited population differentiation, with their frequencies varying markedly across populations (Fig. 3b–d). For example, the TRBV18*01_a76g allele is present at a frequency of 0.43 in donors of African ancestry but is nearly absent in European and Asian populations (Fig. 3c). These differences in TCR allele frequencies across ancestries raise the possibility that evolutionary forces have shaped TCR gene polymorphism across human populations. Thus, we investigated signals of natural selection on TCR genes.

Fig. 3. TCR alleles vary in frequency across ancestries.

a Principal component analysis of donors by their one-field TCR V and J gene genotypes. Ancestry-specific density contours are shown in PC1–PC2 space. b Fst assessing the population differentiation of one-field alleles of V and J genes. c Frequencies of alleles of V genes with high population differentiation across ancestries. Allele frequencies are expressed in fractions, from 0 to 1. d Analogous to (c), but for J genes.

Natural selection on TCR genes

Natural selection offers a powerful lens to understand if inherited variation in TCR genes is biologically consequential; evidence of selection on TCR genes would imply that TCR polymorphisms have shaped immune responses to a degree substantial enough to alter organismal fitness. We hypothesized TCR genes may be subject to three distinct selective pressures: (1) Diversifying selection, wherein specific codons across a gene group (e.g., across TRAV genes) evolve under pressure to recognize diverse ligands, (2) Balancing selection, wherein multiple alleles of a specific TCR gene are maintained at intermediate frequency to recognize evolving pathogenic threats, and (3) Positive selection, wherein a particular allele of a specific TCR gene confers a fitness advantage and rises in frequency. Because TCR gene segments have undergone duplication and divergence throughout vertebrate history, diversifying selection across a TCR gene group may capture both ancient and more recent evolution. In contrast, balancing and positive selection may act on variation in individual TCR genes and specifically reflect evolution in the human lineage.

We first tested for diversifying selection; as T cells must recognize a broad array of peptide–MHC (pMHC) molecules, selection may have favored the accumulation of amino acid diversity at specific sites within TCR V genes. Selective pressures may not act uniformly across all V genes, so we employed the mixed effects model of evolution, which is sensitive to detecting episodic diversifying selection³⁶. We observed diversifying selection at two codons within the CDR1 of TRAV genes and four codons in the CDR1/2 of TRBV genes (Supplementary Fig. 4), consistent with prior reports of diversifying selection in the CDRs³⁷. In many crystal structures, the CDR1/2 regions of TCRs contact the helices of HLA molecules³⁸; diversifying selection may have acted on germline-encoded CDRs to ensure compatibility with the wide range of HLA alleles an individual can inherit. Additionally, as CDR1/2 often directly engage the peptide^5,8, antigen-driven pressures may have contributed to diversifying selection at these CDR codons. Unexpectedly, we also observed diversifying selection within seven codons in the framework regions of TRAV genes and nine codons in the framework regions of TRBV genes, suggesting that polymorphism within these structural regions of the TCR has been evolutionarily advantageous. To test whether allelic variation within individual genes contributed to these signals, we repeated the analysis using only the most common allele of each gene, finding that five fewer TRAV codons demonstrated evidence of diversifying selection (Supplementary Fig. 4).

Next, we asked if polymorphisms in TCR genes may be maintained by balancing selection. Classical HLA genes, which present peptides to TCRs, are canonical examples of balancing selection^1,39. Shifting spatial and temporal pathogen pressures, together with heterozygote advantage, have maintained multiple HLA alleles at intermediate frequencies across human populations^1,40. To test for balancing selection in TCR genes, we applied BetaScan⁴¹ across the TCR loci, the HLA locus, and to the TRB orphon locus on chromosome 9, which encodes nonfunctional genes with TRBV-like sequence content and serves as a neutrally evolving control. A hallmark of balancing selection is the build-up of linked polymorphisms whose frequencies mirror that of the selected allele; BetaScan quantifies this clustering of frequency-matched variants to discover balancing selection^41,42. We found the strongest signal of balancing selection in a region of the TRA/D locus containing TRAJ37 (Fig. 4a). This signal was robust across all five ancestries, consistent with a pattern of long-term balancing selection (Supplementary Fig. 5). To corroborate this finding, we computed Tajima’s D, which detects an excess of intermediate-frequency alleles by contrasting mean pairwise diversity with the number of segregating sites⁴³. This region also exhibited markedly elevated Tajima’s D values, exceeding 2.9 in every population (Fig. 4b). Though pinpointing the causal variant driving this signature is difficult, TRAJ37 showed striking coding-allele variability, with its Shannon entropy ranking among the top 5% of all TRAJ genes in all ancestries, so we examined it further (Fig. 4c). Unlike the TCRβ chain’s CDR3, which is primarily shaped by junctional insertions, the TCRα chain’s CDR3 is more strongly determined by its J gene^44,45. Thus, we wondered if TRAJ37 may have undergone balancing selection because it contributed to the recognition of an immunodominant antigen.

Fig. 4. Balancing and positive selection in the TCR alpha locus.

a Distribution of BetaScan-computed β₁^* scores of balancing selection across the TCR genes, genes known to be subject to balancing selection (HLA class I and II), and genes under neutral evolution (TRB orphon genes). Bounds of the box plot represent the first and third quartiles, the center represents the median, and whiskers extend to the most extreme data points within 1.5 times the interquartile range. b Distribution of Tajima’s D across the TRAJ gene array; the vertical line indicates the window containing the TRAJ37 gene. c Distribution of the Shannon entropy of one-field allele frequencies across TRAJ genes. TRAJ37’s entropy is indicated with a vertical line. d iSAFE plot to detect variants subject to positive selection in individuals of East Asian ancestry. Color indicates linkage (r²) to iSAFE lead variant. e Population differentiation (Fst) between East Asian vs African ancestry (left) and East Asian ancestry vs European ancestry (right). Fst values above 0.3 are shaded darker. f XP-EHH statistic comparing the East Asian ancestry’s haplotype structure to that of African ancestry (left) and to that of European ancestry (right). XP-EHH values less than –2 are shaded darker, indicating East Asian donors exhibit longer haplotypes in this region.

To investigate this, we analyzed antigen specificity data from VDJdb⁴⁶. TRAJ37 was strongly enriched in TCRs binding the key GILGFVFTL epitope of the influenza matrix protein (Fisher’s exact p-value < 10⁻¹⁶), with nearly 70% of TRAJ37⁺ TCRs in VDJdb binding this peptide (Supplementary Fig. 5c). Prior studies have found TRAJ37 contributes to a dominant TCR clonotype that recognizes this epitope on a common HLA allele (A*02:01), and structural analyses show the CDR3α makes critical contacts with the peptide^47,48. If balancing selection acted on TRAJ37 to enable recognition of diverse influenza-derived epitopes as the pathogen evolved, alleles of TRAJ37 may be differentially capable of recognizing those epitopes. Indeed, by analyzing a single-cell dataset⁴⁹ of CD8⁺ T cells stained with an influenza GILGFVFTL–A0201 pMHC dextramer (Methods), we found TRAJ37*02⁺ cells were significantly more likely to recognize this pMHC than TRAJ37*01⁺ cells (p = 4.3 × 10⁻⁷; 13.1% vs 6.1%; Supplementary Fig. 5d). These two alleles differ by a nonsynonymous change in the CDR3α region and three nonsynonymous changes in the FW4 region. Thus, the retention of multiple TRAJ alleles encoding distinct antigen-contacting residues may have conferred a selective advantage in the face of recurrent exposure to a rapidly-evolving virus.

Additionally, we detected balancing selection in TCR V genes, with the strongest signal in a region containing TRAV14/DV4 (Supplementary Fig. 5). Despite these notable examples, variants within TCR loci generally did not show evidence of balancing selection, especially compared to signals observed in the classical HLA genes (Fig. 4a). This, in part, may be due to the differences in the genetic architecture of these loci; on a given haplotype there are only three HLA class I genes that present peptides to CD8⁺ T cells, while there are dozens of TCR V and J genes. This may buffer the pressure for balancing selection to act broadly on all TCR genes.

Finally, we asked if positive selection may have acted on specific variants in TCR genes. To investigate this, we used integrated Selection of Allele Favored by Evolution (iSAFE)⁵⁰ analysis to identify variants with evidence of positive selection. The strongest signal emerged from the TRAV gene array in donors of East Asian ancestry (Fig. 4d). This region exhibited elevated population differentiation, with Fst values exceeding 0.4, and the lead variant identified by iSAFE had a higher allele frequency in East Asian donors (0.67) relative to African donors (0.10) and European donors (0.19), consistent with selection on standing variation, often termed a ‘soft sweep’ (Fig. 4e; Supplementary Fig. 5e). To further investigate this signal, we applied haplotype-based statistical methods. Furcation analysis revealed long-range haplotypes carrying the iSAFE lead variant, and integrated haplotype score (iHS)⁵¹ analysis supported a signature of directional selection, with SNPs in this region exhibiting extreme |iHS | > 2 (Supplementary Fig. 5f). Additionally, cross-population extended haplotype homozygosity tests (XP-EHH)⁵² comparing the East Asian haplotype structure in this region to that of African and European ancestries corroborated the population-specific nature of this selective event (Fig. 4f).

Within the haplotype tagged by the iSAFE signal (r² > 0.85), we found two candidate causal variants in annotated coding or regulatory regions of TCR genes: a splice acceptor variant in TRAV26-2 which renders the gene nonfunctional and a framework 3 region nonsynonymous variant in TRAV30, located three amino acids away from the CDR2 loop. Both of these variants are estimated to have arisen prior to out of Africa migration, over 9,200 generations ago⁵³. Due to the tight linkage between these candidates, unambiguously identifying the causal driver variant is a challenge, as it is for many other loci under positive selection. However, because mutations in framework regions have been shown to alter T cell specificity^54,55, we hypothesize that the TRAV30 framework variant may have enhanced germline-encoded recognition of a specific pathogen circulating in ancestral East Asian populations, leading to its positive selection.

Loss-of-function alleles

The TCR loci contain numerous gene segments encoding the same protein domain; this redundancy buffers against the impact of individual loss-of-function (LoF) mutations and may dampen negative selection, allowing nonfunctional alleles to persist at appreciable frequencies. Nonfunctional alleles cannot contribute to productive TCR chain rearrangements capable of antigen recognition at the cell surface. However, their expression may still be detected if they contribute to nonproductive chains at the mRNA level. Some TCR genes are nonfunctional in all humans, while others harbor LoF alleles that vary across individuals. To systematically understand this heterogeneity, we annotated the functionality of TCR alleles, focusing on genes with variable functionality across the population. 20% of TRAV and 22.9% of TRBV genes harbored a common nonfunctional allele, while only 5.9% of TRAJ genes and 7.7% of TRBJ genes did (Fig. 5a). Including rare alleles, 55.6% and 54.2% of TRAV and TRBV genes harbored a nonfunctional allele, while 17.6% and 30.8% of TRAJ and TRBJ genes did (Fig. 5a).

Fig. 5. Loss-of-function alleles across TCR V and J genes.

a Percentage of TCR genes in each gene group that have a common loss-of-function allele (left), and percentage of genes that have a common or rare loss-of-function allele (right). b Distribution of the number of distinct functional alleles of TRAV (left) and TRBV (right) genes across donors. c Analogous to (b), but for TRAJ and TRBJ genes. d Frequency of common nonfunctional alleles in disease-associated TCR genes.

Several V genes, including TRBV10-1, TRAV26-2, and TRBV16, carried a high burden of nonfunctional alleles, with LoF allele frequencies over 0.1 (Supplementary Fig. 6a). A subset of genes we discover to have common LoF alleles play key roles in disease contexts (Fig. 5d). For example, we identify a LoF allele in TRAV26-1 at a frequency of 0.13 in donors of admixed American ancestry. TRAV26-1 is the dominant Vα gene employed by pathogenic T cells that recognize the HLA-DQ2.5–gliadin-α2 peptide–MHC complex in celiac disease, with germline-encoded TRAV26-1 framework motifs facilitating binding to the gliadin autoantigen^56–58. In our cohort alone, several donors are homozygous for this LoF allele; they may be protected against developing celiac disease. Furthermore, we find a LoF allele of TRDV2 at 0.02 frequency in donors of African ancestry. TRDV2/TRGV9⁺ cells are the most abundant γδ subset in peripheral blood and play critical roles in anti-tumor and anti-pathogen responses; homozygosity for this LoF allele may impair these responses^59–61. We also identify a TRBV12-4 LoF allele at 0.03 frequency in donors of African ancestry; the bacterial superantigens SpeC and TSST-1 specifically activate TRBV12-3/12-4⁺ T cells, so carriers of this allele may be partially protected from superantigen-induced cytokine storm⁶².

Surprisingly, we also identified LoF alleles within TCR genes required for the development of unconventional T cell subsets that recognize non-HLA ligands. Specifically, TRAV10, TRBV4-1, and TRAJ33—required for CD1d-restricted iNKT cells, CD1b/c-restricted T cells, and the predominant population of MR1-restricted MAIT cells, respectively^63–66—harbored LoF alleles that were rare in donors of African and admixed American ancestry. The roles of these unconventional subsets remain incompletely understood; studying the phenotypes of individuals genetically unable to generate them offers a natural model for probing their in vivo function^67,68. Furthermore, the presence of these LoF alleles in presumably immunocompetent individuals suggests loss of these T cell subsets may be tolerated.

Next, we asked if individuals vary in the overall number of distinct functional TCR alleles they carry at one-field resolution. We discovered substantial interindividual variation; donors carried between 46 and 64 (95% empirical CI: 49–61) functional TRAV alleles and between 45 and 68 functional TRBV alleles (95% empirical CI: 47–66). Thus, some individuals have only 66% as many functional TRBV alleles as others (Fig. 5b-c). Donors of African ancestry carried, on average, 2.2 more functional TRAV and 5.6 more functional TRBV alleles than donors of European ancestry (Supplementary Fig. 6b). To identify the sources of this variability, we used a linear model to partition the variance into three components: LoF allele burden, functional heterozygosity (the number of genes at which individuals are heterozygous for two functional alleles), and genomic deletions. Functional heterozygosity accounted for nearly all the interindividual variance in functional allele counts (99% for TRAV and 92% for TRBV; Supplementary Fig. 6c). To mount an effective response, thymocytes must respond to a vast array of antigens. Heterozygosity at classical HLA genes, which expands the range of peptides presented to T cells, is associated with reduced cancer risk, improved immunotherapy outcomes, and better control of infections like HIV^69–72. Similar effects may extend to individuals carrying more functional TCR genes, and HLA and TCR heterozygosity may interact to alter immune responses in these disease contexts. Further studies are needed to elucidate these possibilities.

Influence of TCR allelic polymorphism on single-cell phenotypes

Having established that polymorphism in TCR genes has been shaped by selective forces, we wondered how this class of genetic variation may influence core functional properties of T cells. Although the TCR is central to T cell identity, the impact of inherited TCR variation on human T cell biology remains largely unexplored¹⁰. To investigate this, we leveraged the near-complete map we developed of common human TCR alleles to genotype TCR chains in published single-cell datasets. As donor-level diploid TCR genotypes are difficult to infer robustly from expressed repertoires, we performed all analyses at single-cell resolution: for each cell, we determined the V gene allele from the full-length TCR sequence (Methods). To ensure unambiguous allele calls, we restricted analyses to V genes whose common alleles differ within the germline-encoded V segment upstream of the CDR3 (Supplementary Fig. 7 and Supplementary Note 3). Then, we tested how allelic polymorphism influences cellular fate commitment and cell surface TCR abundance. Unlike most single cell association frameworks, our covariate of interest is germline encoded, so identified signals likely represent causal effects mediated by the TCR genotype.

Recent studies, including our own, have shown the TCR sequence influences thymic lineage commitment^73–75. During positive selection, for example, the specific V genes a T cell expresses biases its likelihood of CD8 lineage commitment^73,74. Thus, we hypothesized that germline variation within individual V genes may modulate fate determination and employed an omnibus logistic regression strategy to a combined cohort of 522 donors to test this hypothesis (Methods). After controlling for potential donor- and age-mediated effects and randomly sampling one cell per clonotype, we found seven V genes exhibited significant allele‑dependent effects on CD8⁺ fate acquisition (Fig. 6a). We observed the strongest signal for TRAV14/DV4⁺ cells (omnibus p = 2.4 × 10⁻³¹); 34.6% of cells expressing TRAV14/DV4*01_c306t were CD8⁺, while 39.9% of cells expressing TRAV14/DV4*02_a117c were CD8⁺. These alleles differ by a nonsynonymous change in CDR1, a region known to extensively contact pMHC ligands during thymic selection³⁸.

Fig. 6. Influence of TCR allelic variants on T cell functional phenotypes.

a Influence of V gene alleles on CD8⁺ fate commitment. Left: Nominal p-value from a likelihood ratio test comparing a null mixed-effects model to a full model including an allele term (i.e., omnibus test of allelic effects). Top 15 V genes are shown. The vertical dotted line indicates the Bonferroni-corrected significance threshold across all tests; blue diamonds denote genes with p-values passing this threshold. Right: Proportion of cells adopting a CD8⁺ fate stratified by allele. Bar plot indicates mean with error bars representing the 95% confidence interval. b Impact of V gene alleles on TCR surface abundance. Left: Nominal p-values from an omnibus mixed-effects linear regression model. Right: Centered log ratio normalized staining of CD3 antibody and pan–TCRαβ antibody of T cells carrying different alleles of TRAV8-6. Bounds of the box plot represent the first and third quartiles, the center represents the median, and whiskers extend to the most extreme data points within 1.5 times the interquartile range (left: n = 2,587, p = 3.5 × 10⁻¹⁰. right: n = 2,587, p = 4.4 × 10⁻⁴. p-values from two-tailed t-test).

Beyond influencing antigen specificity, TCR sequence variation may also modulate receptor surface expression, a key determinant of signaling strength. Artificially-introduced framework mutations can alter surface expression⁷⁶, so we asked whether common inherited variation in V genes has a similar effect. To test this, we applied our mixed-effects modeling approach to TCR + CITE‑seq data from 122 donors⁷⁷, using normalized anti-CD3 antibody staining as a proxy for surface TCR abundance. We focused on CD4⁺ naive cells, the most abundant subset. We observed the strongest effect in TRAV8-6⁺ cells (omnibus p = 3.5 × 10⁻⁷). TRAV8-6*02⁺ CD4⁺ naive cells exhibit significantly increased surface staining of both the CD3 and pan-TCRαβ antibodies in the discovery cohort, an effect that was replicated in an independent cohort⁷⁸ profiled with a distinct anti-CD3 antibody (Fig. 6b; Supplementary Fig. 7d). The two common TRAV8-6 alleles, TRAV8-6*01 and TRAV8-6*02, differ by two residues at IMGT positions 66 and 67 in the framework 3 region. Increased TCR surface expression may enhance sensitivity to cognate pMHC ligands; further experimental work is needed to determine how germline variation influences this sensitivity⁷⁶.

Discussion

Our study highlights a previously under-characterized class of human genetic diversity within loci central to adaptive immunity and suggests that it has functional implications. Evolutionary analyses reveal that TCR genes bear signatures of three distinct evolutionary forces: diversifying, balancing, and positive selection. Selection occurs when genetic variants confer interindividual fitness advantages. The fact that natural selection is observed in genes encoding the TCR—a component of the adaptive immune system traditionally thought to function through somatically-generated stochastic diversity—suggests that inherited TCR variation critically shapes immune responses.

Though our approach was equally powered to detect signals in both the alpha and beta chain loci, the strongest evolutionary and functional associations consistently localized to the alpha chain locus. Surprisingly, individual TRAJ genes exhibited greater allelic diversity than TRBJ genes and harbored fewer LoF mutations, even though there are nearly four times as many TRAJ genes as TRBJ genes. This suggests that the two TCR chains may achieve complementary layers of diversity. While the CDR3β region derives much of its diversity from stochastic junctional processes, the CDR3α region is more strongly shaped by inherited V and J genes⁴⁵. Germline polymorphism in alpha chain genes provides a substrate for natural selection to act upon, enabling populations to retain an immunogenetic “memory” of prior pathogen exposures. This conceptually parallels a recent study finding the kappa and lambda light chains (the immunoglobulin analog of the alpha chain) harbor germline-encoded biases towards recognizing common public epitopes of infectious pathogens⁷⁹.

Additionally, we observed substantial diversity in the framework regions. Recent TCR engineering studies have revealed mutations in framework codons modulate CDR loop orientation and flexibility, altering antigen recognition^54,55,80. Moreover, artificial mutations in framework regions influence TCR heterodimer stability and surface expression⁷⁶. Thus, polymorphism in the framework regions likely serves as another source of functional diversity in the TCR repertoire; T cells with identical CDR sequences may differ in antigen sensitivity due to inherited framework variation. Furthermore, given their molecular dependency, it is possible that the HLA and TCR loci exhibit signals of co-evolution; much larger cohorts with high-quality genotyping of both loci may enable such analyses to become possible. In a preliminary, population-level comparison, the African and admixed American ancestral subgroups exhibited high TRAV/TRBV allelic diversity and the highest allelic diversity for a subset of the classical HLA genes (Supplementary Fig. 3), though this observation likely reflects shared demography.

In this study, it was not tractable to directly associate TCR variants with human disease given the small size of the long-read cohorts and the lack of detailed phenotypic information. Despite the central role of T cells in human immunity, GWAS signals to the TCR loci remain sparse, with the most notable example being an association between a variant in the TRA V gene array and narcolepsy^81–83. This gap may well be driven by technical limitations, as SNP arrays widely used in GWAS studies have poor coverage in these regions^10,14,16. Reduced coverage in these regions may result in disease-relevant causal variation remaining untagged, obscuring associations to these loci. Furthermore, HLA variants may act in concert with TCR risk alleles to shape disease susceptibility and progression; for example, a TCR risk allele may only modulate disease risk in individuals whose HLA genotype enables presentation of the relevant antigenic peptide(s), e.g., HLA-TCR epistasis.

Beyond this study, mounting evidence from structural biology, receptor engineering, and repertoire sequencing indicates germline-encoded motifs play key roles in shaping antigen recognition^{5–8,54,55,58,76}. Although large-scale long-read sequencing of disease cohorts is currently infeasible, many cohorts, including population biobanks, are already being profiled with short-read whole-genome sequencing. The allelic reference developed in this work, coupled with specialized imputation approaches, may make future association studies linking TCR germline variation to human disease risk possible. Importantly, disease-relevant effects may be mediated not only by the presence of particular alleles in the genome, but also by how those alleles are deployed in the expressed repertoire; for example, through gene usage biases, clonal selection, and antigen-driven expansion that vary across individuals and immune contexts. More broadly, the overall number of functional TCR alleles an individual carries may also influence disease risk. As these loci are increasingly profiled, updating and maintaining the allelic references employed by the community to study TCR repertoires will be key⁸⁴.

We also discovered LoF alleles in key autoimmune-associated TCR genes. Determining whether these LoF alleles are protective through genetic association studies would directly inform therapeutic strategies that ablate autoreactive T cell subsets, such as recent efforts targeting TRBV9⁺ cells in ankylosing spondylitis⁸⁵. Furthermore, the commonality of such LoF alleles suggests that V and J genes are not haploinsufficient, supporting the safety of such interventions⁸⁶.

Moreover, our analyses revealed hundreds of rare TCR alleles present at allele frequencies below 0.01 in all ancestries. Rare V gene alleles, in fact, outnumbered common V gene alleles. Given growing evidence that germline residues can mediate autoantigen recognition, it is plausible that these alleles may contribute to rare syndromes of autoimmunity^57,58,87.

Our study faced several limitations. Limited TCR sequencing data are available for γδ cells, preventing detailed single-cell analysis of this population. Furthermore, our analyses focused on coding variation. Although noncoding regulatory variation in the TCR loci (e.g., promoter/enhancer and microRNA binding site variants) likely contributes to inter-individual differences in TCR gene expression and gene usage, detailed functional annotations of these sites are lacking. Future studies that jointly analyze matched TCR genotypes and TCR repertoire data will be essential to characterize regulatory control of the TCR and quantify the heritability of TCR repertoires.

Ultimately, the TCR repertoire possesses a remarkable ability to recognize a vast universe of antigens. While the somatic variation produced by V(D)J recombination is a powerful mechanism for generating diversity in the repertoire, it does not act alone; it is complemented by extensive germline polymorphism within the TCR loci. Our findings suggest this germline variation shapes interindividual differences in T cell responses, motivating further investigation.

Methods

TCR allele discovery

Sequencing data were acquired from the Human Pangenome Reference Consortium’s (HPRC) second release and the NIH All of Us CDRv8 release. All samples in the CDRv8 release with PacBio HiFi sequencing data were included, and samples flagged for QC issues or relatedness were removed from consideration. The ancestral supergroup of each donor was derived from the HPRC and All of Us ancestry calls.

TCR alleles were discovered from genome assemblies and their raw read support was quantified. First, TCR V, D, and J alleles on IMGT/GENE-DB²⁶ (downloaded on June 11, 2025) were aligned to their corresponding gene in the T2T-CHM13v2.0 reference using minimap2 (v2.28) to transfer IMGT gene features, including leaders, exons, introns, splice sites, and recombination signal sequences, onto the T2T-CHM13v2.0 coordinates. The T2T-CHM13v2.0 reference was chosen over the GRCh38 reference, as the latter reference lacks sequence content for several TRBV genes.

Allele discovery was performed using haplotype-resolved de novo assemblies that were previously produced by the HPRC and All of Us consortia. These assemblies were produced via hifiasm, as described by these consortia^24,88. To discover TCR alleles from these assemblies, we mapped assembly contigs to the TCR loci in the T2T reference genome using minimap2 --asmx10 and retained only contigs that (i) mapped with high confidence (MAPQ ≥ 60), (ii) were ≥ 100 kilobases in length, and (iii) were non-overlapping with other retained contigs. Haplotype-resolved contigs greater than 100 kilobases long span substantial flanking noncoding sequence, so alignments can be uniquely localized within the TCR loci, enabling unambiguous assignment of each allele to its genomic location. Continuous contig segments mapping to the coordinates of each gene unit (L-V-GENE-UNIT for V genes, D-GENE-UNIT for D genes, and J-GENE-UNIT through J-DONOR-SPLICE for J genes) were extracted as candidate allele sequences.

After candidate TCR alleles were identified from these contigs, we mapped PacBio HiFi reads back to these candidate alleles and determined the number of reads whose sequence exactly matched the full candidate allele sequence (end-to-end, with no mismatches or indels). An allele was called if at least eight reads exactly supported the sequence present in the assembly in at least one donor or at least four reads exactly supported the assembly sequence across at least three donors.

Alleles were named at one-field and two-field resolution. Following IMGT convention, one-field alleles were defined by nucleotide changes in their coding (V-REGION/D-REGION/J-REGION) sequences. Any allele sequence absent from the IMGT/GENE-DB reference was designated as novel and annotated with a suffix that lists the variants that distinguish the novel allele from the closest known allele. Furthermore, because germline variation outside the coding exon (such as in leader regions, splice sites, or recombinational signal sequences) can alter TCR gene functionality, we also defined alleles at a “two-field” resolution. IMGT does not assign two-field alleles, so we constructed two-field allele sequences by drawing upon the relevant IMGT-annotated features²⁶. For V genes, the two-field sequence was the IMGT-defined L-V-GENE-UNIT, extending from the L-PART1 through to the V-RS. For D genes, the two-field sequence was the IMGT-defined D-GENE-UNIT, extending from the 5’D-RS through the 3’D-RS. For J genes, we expanded the IMGT J GENE-UNIT to additionally include the annotated donor splice site immediately 3′ of the J-REGION. Thus, two-field J gene alleles extended from the J-RS through the DONOR-SPLICE. Under this two-field definition, a sequence was considered known if an identical two-field sequence was present among IMGT/GENE-DB reference entries for that gene (restricted to the same IMGT-annotated feature span), and novel otherwise.

Digger v0.7.0⁸⁹ was used to annotate the functionality of TCR alleles; it assesses the validity of recombination signal sequences, leader sequences, canonical motifs, and sequence translation. Additionally, genes with non-‘GT’ splice donor sites and non-‘AG’ splice acceptor sites were determined to be nonfunctional.

To orthogonally validate alleles using data from expressed TCR repertoires, FASTQ files from three recent single-cell studies that profiled full-length TCR transcripts with the 10x Genomics 5′ V(D)J kit were reprocessed (Supplementary data 1). These single-cell studies included Edahiro et al. (234 donors, ~0.8 M T cells), COMBAT (122 donors, ~0.4 M T cells), and Terekhova et al. (166 donors, ~1 M T cells), all of which sequenced T cells from PBMC samples. In addition, all sequencing runs matching the query “(((5’) AND TCR) AND VDJ) AND “Homo sapiens”[orgn:__txid9606] NOT (3’) NOT (3 prime)” were downloaded from the NCBI Sequence Read Archive on December 22, 2024. Those found to contain valid TCR sequence content were processed with CellRanger V(D)J 8.0. A BLAST database containing the coding regions of germline TCR alleles observed in our cohort was constructed, and IgBlast v1.22.0⁹⁰ was used to genotype all TCR chains present in the cellranger ‘filtered contigs’ outputs across the datasets. A novel allele was deemed validated in expression data if it was found in at least five cells with a 100% sequence identity match against the expressed chain (as determined by IgBlast) and no ties. Additionally, allele sequences were obtained from a recent study that profiled expressed TCR alleles from 45 donors using RNA sequencing¹². The corresponding GitHub repository for this manuscript specifies which alleles were independently observed in that study and overlap with those identified here.

When estimating the allele frequencies of particular TCR alleles, we only considered chromosomes in which the corresponding TCR gene was discovered. Allele frequency was therefore computed as:

$$AF\left(alleley\right)=\frac{\mathrm{\#}haplotype-resolvedcontigscarryinggenegthatcontainalleley}{\mathrm{\#}haplotype-resolvedcontigscarryinggeneg}$$ 1

This approach ensures that allele frequencies are insensitive to structural variation/hemizygosity, as haplotypes in which a gene is absent do not enter the denominator. In other words, the allele frequencies computed are conditioned upon gene presence. Thus, estimates of allele frequencies derive only from chromosomes harboring that specific TCR gene. Allele frequencies are expressed as fractions throughout the manuscript, from 0 to 1.

Per-donor genotypes derived from the haplotype-resolved assemblies and PacBio HiFi reads were used only to estimate allele frequencies of the discovered TCR alleles. These per-donor allele calls were not used for the subsequent population-genetic or evolutionary analyses in the manuscript. These analyses instead use the independent long-read variant callset released by the All of Us consortium, which was generated by aligning PacBio HiFi reads to the T2T reference genome and calling variants, as is standard in the field.

To estimate the theoretical number of common circulating alleles, a Michaelis-Menten curve, which has been extensively used to estimate species richness⁹¹, was fit to the rarefaction curve of each ancestry; the estimated asymptote was taken to be the theoretical number of common alleles circulating in the population.

For the analysis of per-donor functional allele counts, we included only HPRC samples whose genome assemblies fully spanned both the TRA/D and TRB loci. The number of distinct functional one-field alleles was computed for every donor; each functional allele was counted once, regardless of zygosity. To partition the variance in interindividual functional allele counts, an ordinary least squares linear regression model was fit and a Shapley decomposition of model R² was performed. In addition, graphs showing the frequency of specific nonfunctional alleles represent the frequency of these alleles computed within the HPRC cohort.

To compute the Shannon entropy of each TCR gene, we employed the following formula: $-{\sum }_{x\in X}p\left(x\right){\log }_{2}p\left(x\right)$, where $X$ represents the set of alleles for that gene, $x$ represents a specific allele of that gene, and $p\left(x\right)$ represents the frequency of that allele. To compute the Shannon entropy of each classical HLA gene, we employed the same formula and used four-digit HLA allele frequencies from the 1000 G HLA reference⁹². To compute the global Fst for each V and J gene across the five ancestral populations, we used the Weir–Cockerham estimator as implemented in the R package hierfstat v0.5-11.

Evolutionary genetics analyses

For evolutionary genetics analyses, we used a joint variant callset from the All of Us CDRv8 release, as sample preparation, sequencing, and bioinformatic processing were done in a uniform manner across this cohort. Per-sample variant calling was previously performed with DeepVariant against the T2T-CHM13v2.0 reference, and samples flagged for QC issues or relatedness were removed from consideration⁸⁸. Per-sample variant calls for chromosomes 7, 9, and 14 were jointly called with GLNexus v1.4.⁹³ and underwent quality control by filtering out multiallelic sites and variants with minor allele frequency <0.01, missingness > 0.1, or QUAL < 30. SHAPEIT4⁹⁴ was used to phase the resulting joint callset.

To determine codon-level diversifying selection, a codon alignment consistent with the IMGT positioning scheme was used. The mixed effects model of evolution (MEME) method³⁶ was run as implemented on Hyphy v2.5⁹⁵. The R seqlogo package v3.2.1 was employed to visualize the amino acid alignments, and the R ape package v5.8.1 was used to plot dendrograms.

For analysis of balancing selection, BetaScan⁴¹ was run with default parameters to compute the β₁^* statistic across all loci considered. To determine which ligands TRAJ37 may bind to, antigen specificity data was downloaded from VDJdb⁴⁶ on April 20, 2025. VDJdb entries from 10x dCODE dextramer experiments were removed. A Fisher’s exact test was used to quantify the enrichment of TRAJ37 in specific antigen binding groups.

To identify candidates of positive selection, iSAFE v1.0.4⁵⁰ was applied with the following parameters: MaxFreq = 0.8 and MaxRank = 300. Furthermore, selscan⁹⁶ v1.2.0a was employed to calculate and normalize XP-EHH across chromosome 14, and the R rehh package v3.2.2⁹⁷ was used to calculate and normalize iHS across the polarized chromosome 14. The R popgenome package v2.7.5⁹⁸ was used to compute the site frequency spectrum-based statistic Tajima’s D using a 5000 bp window size. Pairwise Fst between ancestries was computed for each SNP using the Weir and Cockerham method implemented in vcftools v0.1.14. Ancestral alleles were obtained from ENSEMBL’s release 115 and were lifted-over from the hg38 to the T2T-CHM13v2.0 genome build.

Single-cell analyses

To investigate the impact of germline variation on T cell phenotypes, we utilized a single-cell association framework. Rather than inferring donor-level diploid genotypes, which can be biased by limited sampling depth and allelic dropout, we performed all analyses at single-cell resolution. This approach treats each cell as an independent observation of a V-gene allele and its associated phenotype, utilizing mixed-effects models with a random donor intercept.

(1) V-gene allele assignment. We considered high-quality, full-length productive TCR sequences captured by the 10x Genomics 5′ V(D)J assay and assembled by cellranger vdj. Allele assignment was performed using IgBLAST, aligning sequences against the reference of common TCR alleles. To ensure high-confidence allelic assignments, we restricted our analysis to the 95% of TRAV/TRBV genes whose common alleles are distinguished by substitutions within the germline-encoded V segment (IMGT positions 1–104). These genes are listed in Supplementary data 2. By excluding V genes where allelic differences reside solely within the CDR3-contributing region (IMGT position ≥ 105), we avoided ambiguities introduced by somatic V(D)J recombination. Only chains achieving IgBLAST-determined 100% sequence identity to a reference V-segment allele were retained.

(2) Per-cell quality control. To ensure each observation represented a high-confidence, independent T cell, we applied further filtering. First, within each donor, we discarded any V-gene allele supported by fewer than three cells to mitigate the impact of sequencing or assembly errors. Second, we retained only T cells containing exactly one productive TRA and one productive TRB chain to remove potential doublets. Finally, we grouped T cells into clones (defined by their V allele, J allele, and CDR3 nucleotide sequence) and randomly sampled one representative cell per clonotype, as cells from the same clone share identical TCR sequence features and similar transcriptional fates, so do not represent independent observations.

The above steps are schematized in Supplementary Fig. 7a, and the number of chains and cells remaining at each step in each of the datasets employed in the analyses is presented in Supplementary Fig. 7b. Further validation of the per-cell allele calls, including comparisons against known population allele frequencies and consistency checks for diploidy, is provided in Supplementary Note 3.

To test for associations between TCR V gene alleles and single-cell phenotypes, we employed mixed-effects regression models implemented in R lme4 package v1.1.23. Specifically, an omnibus regression framework was employed to evaluate the influence of TCR allelic polymorphism on cellular phenotypes. The models were formulated as follows:

Null Reduced Model:

$${outcome}_i=\theta +{\epsilon }_i$$ 2

Full Model:

$${outcome}_i=\theta +{\sum }_{q=2}^R{\beta }_q\times q_{present}+{\epsilon }_i$$ 3

Each observation i represents a T cell from a donor d. The models were fit once per TCR gene, where $\theta$ is the intercept. In the full model, ${\beta }_q$ is a fixed effect representing the effect of allele $q$ on the outcome, $q_{present}$ is an indicator variable with a value of 1 if that allele was observed in the cell, and $R$ is the number of alleles observed for that gene. A likelihood ratio test was used to compare the fit of the null reduced model to that of the full model. For each cellular phenotype, additional covariates were included in both the reduced and full models, as described below:

To test for V gene allele influences on fate commitment, an omnibus mixed-effects logistic regression approach of the above form was run on a combined cohort of 522 donors from three recently-published 5’ single-cell datasets with TCR-sequencing information^77,78,99. An additional fixed-effect for donor age, a fixed effect for donor sex, a fixed effect with three levels representing the dataset the cell was observed in, a fixed effect for donor COVID-19 status, and a random effect intercept for each donor was included. The outcome variable was set to 1 if the cell was in a CD8⁺ transcriptional cluster, and 0 if the cell was in CD4⁺ transcriptional cluster based on the cell annotations provided by the authors. All non-CD4⁺/CD8⁺ T cells were excluded. This model was run once per TRAV and TRBV gene; V genes with at least two alleles expressed in at least 250 clonotypes were considered.

To test for allelic influences on antigen binding, an omnibus fixed effects logistic regression approach of the above form was applied to dataset of CD8⁺ T cells from four donors profiled using 10x genomics dextramer staining technology⁴⁹. The two alleles of the TRAJ37 gene differ by several nucleotides in the CDR3 and FW4 regions and thus could be unambiguously genotyped. Dextramer binding calls were computed using our previously-described approach, wherein a negative binomial regression model was employed to remove background staining and strict thresholds were set to distinguish binders from non-binders; this strategy regresses out TCR expression and surface staining to isolate effects due to variation in TCR sequence content rather than differences in TCR expression⁷³. The outcome variable was a binary indicator representing whether the cell bound to the influenza GILGFVFTL dextramer. An additional fixed effect was included for each donor.

To test for V gene influences on cell surface TCR abundance, an omnibus mixed-effects linear regression model approach with the above form was fit to a cohort of donors profiled with CITE-seq⁷⁷. A fixed effect for donor age, a fixed effect for donor sex, a fixed effect for donor COVID-19 status, and a random effect intercept for each donor was included in the model. CITE-seq staining counts were centered log-ratio normalized (CLR) across all antibody-derived tags for each cell. The outcome variable was the CLR-normalized value of the CD3 antibody. All cells in the CD4⁺ naive compartment were considered, the model was run once per V gene, across all V genes with at least two alleles present in at least 250 clonotypes. The observed TRAV8-6 signal was replicated in a distinct cohort⁷⁸ profiled with a different anti-CD3 antibody.

For all the above models, Bonferroni correction was applied to correct for multiple testing.

Author contributions

S.M. and S.R. conceptualized the project. S.M. performed analyses with guidance from S.R. and A.A. R.E. and Y.O. curated and supported access to and analysis of single-cell data. All authors discussed the results and contributed to the final manuscript.

Peer review

Peer review information

Nature Communications thanks Benny Chain and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work is supported in part by funding from the NIH, including R01AR063759, P01AI148102, U01HG012009, R56HG013083, and T32GM144273. Y.O. was supported by JSPS KAKENHI (25H01057), and AMED (JP24km0405217, JP24ek0109594, JP24ek0410113, JP24kk0305022, JP223fa627001, JP223fa627002, JP223fa627010, JP223fa627011, JP22zf0127008, JP24tm0524002, JP24wm0625504, JP24gm1810011), JST Moonshot R&D (JPMJMS2021, JPMJMS2024), Ono Pharmaceutical Foundation for Oncology, Immunology, and Neurology, CiDER and CAMaD, Osaka University, RIKEN TRIP initiative (AGIS). R.E. was supported by JSPS KAKENHI (25K18835), AMED (JP25tm0424233, JP25gm7110002, JP24gm1810012, JP24ek0210207), and a Japan Foundation for Applied Enzymology grant.

Data availability

All sequencing data analyzed in this study were previously deposited in publicly accessible databases. The genetic data analyzed were provided by the All of Us CDRv8 (https://support.researchallofus.org/hc/en-us/articles/30294451486356-Curated-Data-Repository-CDR-version-8-Release-Notes) and Human Pangenome Reference Consortium release 2 (https://data.humanpangenome.org/) datasets, while the single-cell TCR repertoire expression data was acquired from the published studies listed in Supplementary data 1. All sequences of common validated TCR alleles are available at https://github.com/SreekarMantena/tcrdiversity. The All of Us consortium has restrictions concerning the public release of participant data; all summary statistics and publicly released data of this study comply with these restrictions.

Code availability

The code used to perform analyses in this study is available at https://github.com/SreekarMantena/tcrdiversity-analysis.

Competing interests

S.R. is a founder for Mestag, Inc, and serves as a scientific advisor for BMS, Janssen, Merck, Pfizer, Nimbus, and Bonito. The remaining authors declare no competing interest for this study.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73730-9.