Genetic and environmental determinants of human TCR repertoire diversity

The large sequence diversity of the TCR repertoire is a hallmark of the adaptive immune system, and varies markedly across individuals [1‐4]. This diversity, estimated to exceed 10⁶ sequences in humans [5‐7], is shaped by stochastic [8] and genetic [9] effects in conjunction with continuous immunological challenges throughout life [9]. In the thymus, VDJ recombination facilitates random rearrangement of the complementary determining region 3 (CDR3) within the TCR α and β loci, followed by random nucleotide insertion and deletions at junction sites [10]. The CDR3 regions of the TCR are primarily responsible for interacting with the peptide presented by MHC [11], with the potential diversity of CDR3β exceeding that of CDR3α [12]. Whether a particular TCR joins the periphery depends on its behavior during thymic selection, in which TCRs interact with both self peptide and MHC [13, 14]. TCRs that fail to bind to peptide-MHC complexes and those that bind too strongly are eliminated [15, 16]. Those TCRs that survive thymic selection are responsible for mounting productive immune responses through continuous interaction with self and foreign peptides bound to MHC molecules. TCR diversity can determine how efficiently one rejects pathogens such as viruses, and potentially cancer cells. Accordingly, considerable effort has been devoted to understanding how MHC genetic variation impacts the TCR repertoire.

MHC restriction is the cornerstone of T cell recognition [17], and prior reports have assessed the effect of the presence of specific MHC alleles on TCR V gene usage [18, 19] and repertoire sharing [9, 20]. These data, together with structural studies of the TCR-MHC interface [11, 21‐24], have provided key insights into how the TCR binds MHC and peptide. However, it remains unknown to what extent HLA polymorphism affects TCR repertoire diversity in humans.

We sought to address this question. Thus, we studied a cohort of 666 individuals, with annotated CMV serostatus, ethnicity, age, sex, high-resolution HLA class I and class II genotypes, and bulk TCRβ sequencing from PBMCs [9, 25] (Additional file 1: Table S1). 85% of the individuals were white, 52% were male, and 45% were female, with the remainder of unknown sex. We first quantified TCR repertoire diversity by applying two measures widely used in repertoire and ecological studies—the number of unique CDR3β amino acid sequences (a.k.a. richness), and Shannon entropy, a diversity measure that is weighted by the abundance of each CDR3 [26]. We found both measures to be highly correlated, and observed high variability in TCR repertoire diversity across the cohort (Fig. 1a; richness range 1055–415,509, Shannon entropy range 8.1–18.7; R = 0.78, P < 0.0001). Accordingly, we anticipated that CMV—a chronic infection prevalent in 30–90% of adults [27] and a model system for the study of public T cell responses [28]—would be a key determinant of the observed wide variation in repertoire diversity. Indeed, individuals with CMV (CMV+) exhibited a reduction in TCR repertoire diversity compared to those without (CMV-) (Fig. 1b-c). This reduction was most striking when using Shannon entropy (P < 0.0001, Wilcoxon test; Fig. 1b), consistent with prior work demonstrating that CMV alters the diversity, but not overall size of the CD8+ T cell response [29]. Altogether, these data suggest that CMV diminishes TCR repertoire diversity, and are in line with a recent study demonstrating dramatic reduction of the antibody repertoire after measles infection [30]—highlighting the need for widespread and continuous vaccination against infectious disease.

We next limited our analyses to individuals with complete HLA-I (HLA-A, B, & C) and II (HLA-DRB, DPB, DQB, DQA, & DPA) genotypes, and given the impact of CMV on the TCR repertoire described above, considered CMV+ and CMV- individuals separately (Additional file 2: Fig. S1). We used a linear model to test the association between HLA polymorphism—measured here as the number of different HLA-I alleles in each individual—and TCR repertoire diversity in CMV- individuals. Strikingly, we observed that TCR repertoire diversity was positively associated with the number of HLA-I alleles (P = 0.02; Fig. 1d). Furthermore, we observed that CMV- individuals fully heterozygous at HLA-I genes had higher TCR repertoire diversity than individuals who were homozygous at least in one HLA-I locus (P = 0.02; Fig. 1e). We found the same associations when considering Shannon entropy instead of richness (Additional file 2: Fig. S2a-b). Importantly, these results were independent of age, previously shown to be negatively correlated with TCR repertoire diversity [31‐33] and shown here to be independent of the number of HLA-I alleles. Interestingly, we found no association between HLA-II polymorphism and TCR repertoire diversity (Fig. 1f-g and Additional file 2: Fig. S2c-d). These data may suggest that heterozygosity at HLA-II may be disadvantageous given the strong associations between many HLA-II haplotypes and susceptibility to autoimmune disease [34]. Finally, we repeated these analyses in CMV+ individuals, and observed no association between HLA polymorphism and TCR repertoire diversity (Additional file 2: Fig. S3). Notably, we found no association between age and number of unique CDR3s in CMV+ individuals either (P = 0.41; Fig. 2a), whereas in CMV- negative individuals, we observed that the number of unique CDR3s diminished with age (P = 0.002; Fig. 2b). When considering Shannon entropy instead of richness, the effect of age was weaker in CMV+ individuals (P = 0.03; Additional file 2: Fig. S4a) than in CMV- individuals (P = 0.0005; Additional file 2: Fig. S4b). These results suggest a dominant role of chronic infection over host genetics and age in significantly altering the TCR repertoire. However, HLA diversity may affect antigen-specific TCRs rather than the whole repertoire as suggested by past studies [35], and should be the subject of future complementary analyses focused on the diversity of CMV-specific expanded clones.

We next sought to assess the combined effect of age and HLA-I polymorphism on repertoire diversity. We developed three separate linear models in CMV- individuals—one with age alone, one with number of unique HLA-I alleles alone, and one with both age and number of unique HLA-I alleles. We selected the most strongly supported model based on the Akaike information criterion (AIC), i.e., the best-fit model yields the smallest AIC value [36]. We found that the best model that explained the observed TCR repertoire diversity in these individuals included both variables, (AIC = 4601.28; Fig. 2c; AIC = 575.11, Additional file 2: Fig. S4c). As expected, the combined effect of age and HLA-I polymorphism was not observed in CMV+ individuals (Additional file 2: Fig. S4d).

Recent studies have demonstrated that HLA-I evolutionary divergence (HED), a continuous and granular metric of HLA-I polymorphism, measures the breadth of the immunopeptidome bound by an individual’s MHC-I molecules [37, 38]. Therefore, we sought to investigate the association between mean HED, an aggregate measure of HED across the three classical HLA-I loci [37, 38], and TCR repertoire diversity. High mean HED—defined here as mean HED greater than the median—was associated with increased TCR repertoire diversity in CMV- individuals, and was independent of age (P = 0.03; Additional file 2: Fig. S5a). These results suggest that individuals with more divergent HLA genotypes, and correspondingly broader immunopeptidomes, have increased TCR repertoire diversity. Moreover, consistent with our earlier analysis, the best model of TCR repertoire diversity included both high mean HED and age (AIC = 577.6; Additional file 2: Fig. S5b). This effect was not observed in CMV+ individuals (Additional file 2: Fig. S5c-d). These data thus provide further evidence for the notion that TCR repertoire diversity is increased in individuals with greater HLA diversity.

To our knowledge, our study is the first to show empirically that HLA-I polymorphism increases TCR repertoire diversity in humans, and has several important implications. First, our results add an important dimension to the HLA heterozygote advantage hypothesis, which states that HLA-heterozygous individuals present a broader immunopeptidome for recognition by cytotoxic T cells [17, 35, 37‐41]. In particular, our data suggest that an additional potential consequence of enhanced peptide presentation in HLA-heterozygous individuals is a more diverse TCR repertoire, which could improve immune protection and evolutionary fitness.

An important consideration is the direction of the association between HLA polymorphism and TCR repertoire diversity, which has remained controversial. The TCR depletion hypothesis suggests that increasing individual MHC polymorphism—leading to more self peptides presented during negative selection—creates “holes” in the TCR repertoire, thereby decreasing its diversity in the periphery [42‐45]. However, there is no experimental evidence for the TCR depletion hypothesis in humans. Our results may suggest a dominant role for positive selection in influencing TCR repertoire diversity. This idea is in line with prior theoretical work suggesting that additional MHC variants enhance positive selection, and consequently the number of T cells that can survive negative selection [46]. Indeed, our study motivates empirical investigation of how each step of thymic selection affects the TCR repertoire in an MHC-dependent fashion, which remains unclear.

Finally, while our study demonstrates that HLA-I polymorphism, age, and chronic infection shape the TCR repertoire, a full account of the determinants of TCR repertoire diversity remains unknown. Of note, variation in TCR repertoire diversity across individuals may be driven in part by differences in T cell sampling. The sample size variation in our cohort spans an order of magnitude, and may be driven in part by uncontrolled factors in the sequencing process. As a possible control for sample size variation, we quantified TCR repertoire diversity in CMV- individuals using a normalized form of the Shannon entropy, (Methods). Using this slightly corrected measure, we still observed a positive association between HLA-I polymorphism and TCR repertoire diversity (Additional file 2: Fig. S6). Indeed, our analyses suggest that despite confounding variation in TCR repertoire sample sizes across individuals, biological factors such as age and HLA-I polymorphism independently affect TCR repertoire diversity. In addition, prior studies have suggested that the TCR repertoire differs by sex [42, 47]. We also detected a trend towards reduced TCR repertoire diversity in CMV- males in our cohort (P = 0.09, Additional file 2: Fig. S7). However, larger numbers may be required to clarify this association. Future work may investigate how the TCR repertoire is shaped by vaccination, or how HLA polymorphism and TCR repertoire diversity act in concert to influence overall mortality.