Treatment- and population-specific genetic risk factors for anti-drug antibodies against interferon-beta: a GWAS

BACKGROUND: Upon treatment with biopharmaceuticals, the immune system may produce anti-drug antibodies (ADA) that inhibit the therapy. Up to 40% of multiple sclerosis patients treated with interferon β (IFNβ) develop ADA, for which a genetic predisposition exists. Here, we present a genome-wide association study on ADA and predict the occurrence of antibodies in multiple sclerosis patients treated with different interferon β preparations.

METHODS: We analyzed a large sample of 2757 genotyped and imputed patients from two cohorts (Sweden and Germany), split between a discovery and a replication dataset. Binding ADA (bADA) levels were measured by capture-ELISA, neutralizing ADA (nADA) titers using a bioassay. Genome-wide association analyses were conducted stratified by cohort and treatment preparation, followed by fixed-effects meta-analysis.

RESULTS: Binding ADA levels and nADA titers were correlated and showed a significant heritability (47% and 50%, respectively). The risk factors differed strongly by treatment preparation: The top-associated and replicated variants for nADA presence were the HLA-associated variants rs77278603 in IFNβ-1a s.c.- (odds ratio (OR) = 3.55 (95% confidence interval = 2.81-4.48), p = 2.1 × 10-26) and rs28366299 in IFNβ-1b s.c.-treated patients (OR = 3.56 (2.69-4.72), p = 6.6 × 10-19). The rs77278603-correlated HLA haplotype DR15-DQ6 conferred risk specifically for IFNβ-1a s.c. (OR = 2.88 (2.29-3.61), p = 7.4 × 10-20) while DR3-DQ2 was protective (OR = 0.37 (0.27-0.52), p = 3.7 × 10-09). The haplotype DR4-DQ3 was the major risk haplotype for IFNβ-1b s.c. (OR = 7.35 (4.33-12.47), p = 1.5 × 10-13). These haplotypes exhibit large population-specific frequency differences. The best prediction models were achieved for ADA in IFNβ-1a s.c.-treated patients. Here, the prediction in the Swedish cohort showed AUC = 0.91 (0.85-0.95), sensitivity = 0.78, and specificity = 0.90; patients with the top 30% of genetic risk had, compared to patients in the bottom 30%, an OR = 73.9 (11.8-463.6, p = 4.4 × 10-6) of developing nADA. In the German cohort, the AUC of the same model was 0.83 (0.71-0.92), sensitivity = 0.80, specificity = 0.76, with an OR = 13.8 (3.0-63.3, p = 7.5 × 10-4).

CONCLUSIONS: We identified several HLA-associated genetic risk factors for ADA against interferon β, which were specific for treatment preparations and population backgrounds. Genetic prediction models could robustly identify patients at risk for developing ADA and might be used for personalized therapy recommendations and stratified ADA screening in clinical practice. These analyses serve as a roadmap for genetic characterizations of ADA against other biopharmaceutical compounds.