Background: HLA loci contribute the most GWAS signals for complex traits. However, the HLA region is typically excluded from post-GWAS analyses due to complex genetic diversity and distinctive immunobiology. Mechanistically, HLA risk alleles are thought to restrict the hypervariable CDR3 region of T-cell receptors (TCRs) recognising pathogenic antigens. To strengthen functional interpretation of HLA associations, we developed ImmunoXcan, a TCR-based analysis pipeline leveraging HLA variants as instruments.
Method: ImmunoXcan adapts principles from transcriptome-wide association studies and Bayesian fine-mapping to jointly model TCR and non-mediated genetic effects, adjusting for linkage disequilibrium and horizontal pleiotropy despite multicollinearity. Using CDR3β immunosequencing of healthy blood donors (n=628), we trained 21,575 sparse TCR expression models from HLA protein variation (average cross-validated accuracy R2/h2=0.61) and predicted trans-regulatory TCR features into HLA mapping data for 52 immune-mediated phenotypes in UK Biobank Serology Study (n=7,741), 23andMe (n=219,579), and VaccGene clinical trials (n=2,499). We mapped credible causal TCRs to pathogenic antigens using tetramer-sorted T-cell data, k-mer matching, and paratope hotspot clustering.
Results: ImmunoXcan identified multiple 'positive control' TCR-trait associations (posterior probability of association >0.90), supported by functional evidence of target antigen specificity and HLA-colocalisation evidence of shared causal variants. Noteworthy examples include: (i) TCRs specific to uncharacterised Mycobacterium tuberculosis (MTB) Rv1518/MT1568 antigens and memory T-cell response to MTB (HLA-DRβ1 colocalisation PPH4>0.90); (ii) CDR3β-S%GGET_ILV+ clonotypes reactive to myelin basic protein and multiple sclerosis risk (HLA-DQβ1 PPH4>0.70); (iii) Epstein-Barr virus (EBV) EBNA3A-specific TCRs restricted to HLA-B*08 risk alleles and EBV seropositivity (HLA-B PPH4>0.50); (iv) hepatitis B virus (HBV) core protein-specific TCRs and antibody response to HBV vaccination (HLA-DQα1~DQβ1 PPH4>0.99). Probabilistic annotation enrichment using mass spectrometry immunopeptidomics data revealed shared HLA peptide-binding motifs underlying candidate causal axes.
Conclusion: ImmunoXcan provides a framework for resolving constitutional HLA-TCR interactions in immune-mediated trait susceptibility, with implications for designing preventive immunotherapeutics.