Modeling HLA associations with EBV-positive and -negative Hodgkin lymphoma suggests distinct mechanisms in disease pathogenesis

Abstract

HLA genotyping and genome wide association studies provide strong evidence for associations between Human Leukocyte Antigen (HLA) alleles and classical Hodgkin lymphoma (cHL). Analysis of these associations is complicated by the extensive linkage disequilibrium within the major histocompatibility region and recent data suggesting that associations with EBV-positive and EBV-negative cHL are largely distinct. To distinguish independent and therefore potentially causal associations from associations confounded by linkage disequilibrium, we applied a variable selection regression modeling procedure to directly typed HLA class I and II genes and selected SNPs from EBV-stratified patient subgroups. In final models, HLA-A*01:01 and B*37:01 were associated with an increased risk of EBV-positive cHL whereas DRB1*15:01 and DPB1*01:01 were associated with decreased risk. Effects were independent of a prior history of infectious mononucleosis. For EBV-negative cHL the class II SNP rs6903608 remained the strongest predictor of disease risk after adjusting for the effects of common HLA alleles. Associations with "all cHL" and differences by case EBV status reflected the subgroup analysis. In conclusion, this study extends previous findings by identifying novel HLA associations with EBV-stratified subgroups of cHL, highlighting those alleles likely to be biologically relevant and strengthening evidence implicating genetic variation associated with the SNP rs6903608.

Keywords: EBV; HLA; Hodgkin lymphoma.

Figure 1

Forest plot showing final models for EBV‐positive and EBV‐negative cHL. Panel A. EBV‐positive Hodgkin lymphoma. Panel B. EBV‐negative Hodgkin lymphoma. The set of allele effects that best predicted EBV‐positive and EBV‐negative Hodgkin lymphoma cases was selected from 44 HLA alleles and three selected MHC SNPs. Effects with a posterior probability of association (PPA) ≥ 50%, estimated in a Bayesian variable selection model, were selected. The ORs and 95% CIs presented here were estimated after refitting in a Firth logistic regression model. For EBV‐positive Hodgkin lymphoma, HLA‐A*01:01 and A*02:01 were included as adjustment variables and were not subject to variable selection; analyses were adjusted for sex and age group. Symbol size reflects allele frequency.