HLA-DRB1 and HLA-DQA1 associated with immunogenicity to adalimumab therapy in patients with rheumatoid arthritis

Advanced targeted therapies including tumour necrosis factor inhibitors (TNFis) have transformed the clinical management of rheumatoid arthritis (RA). However, monoclonal antibody (MAb)derived TNFis are associated with development of immunogenicity resulting in low circulating drug levels (online supplemental figure S5). A genetic predictor of immunogenicity would have clinical utility by providing a pretreatment biomarker that could be used to inform therapy selection. Previous genetic studies of TNFi immunogenicity have focused on alleles within the HLA locus on chromosome 6. Patients were followed for 12 months with serum samples collected at 3 months, 6 months and 12 months following commencement on adalimumab (TNFi) therapy. Neutralising antidrug antibodies (ADAs) were detected using a drugsensitive/drugtolerant radioimmunoassay (Sanquin, NL). The presence of ADAs was determined by radioimmunoassay. A positive ADA titre was defined as >12 arbitrary units/mL. If a patient developed ADA at any time in the study, they were classed as ADA positive. Genotyping was carried out using the Illumina array, and HLA alleles were imputed using SNP2HLA and the T1DGC reference panel following standard data quality control (full details in online supplemental S1). Drug immunogenicity rates were determined using KaplanMeier analysis, and Cox proportional hazards regression, which was used to adjust genetic models for biological sex, age, concurrent conventional synthetic diseasemodifying antirheumatic drug (csDMARD) use, disease duration and first withinsample principal component from the genetic dataset. In total 445 patients were studied, of whom 96 (21.6%) became ADA positive during treatment. A total of 377 (85.3%) patients received cotherapy with csDMARDs of which 302 (81.4%) patients received methotrexate (MTX, online supplemental table S1). Disease duration modestly increased the rate of immunogenicity for every year since RA diagnosis (HR=1.02, p=0.01, table 1). Compared with TNFi monotherapy, combination therapy with csDMARD reduced the rate of ADA development by more than twofold (HR=0.379, p=1.27e−07). Importantly, a statistically significant difference in the rate of immunogenicity was observed when MTX cotherapy was compared with cotherapy with alternative csDMARDs; MTX conferring higher protection from immunogenicity (HR=0.425, p=1.27e−05). However, nonMTX csDMARD use also trended towards a reduced rate of immunogenicity (HR=0.66; 95% CI 0.429 to 1.012, p=0.056). Following quality control of the genetic data, 166 HLA alleles were available for analysis in 435 patients with nonmissing covariate data. The most statistically significant association with immunogenicity was observed for HLADQA1*03 (HR 0.6; 95% CI 0.474 to 0.775, p=6.4e−05) and HLADRB1*04 (HR 0.6; 95% CI 0.476 to 0.775, p=6.3e−05) (4digit and aminoacid results are reported in online supplemental material S1). In the KaplanMeier analysis, carriage of HLADQA1*03 and HLADRB1*04 alleles under an additive model was associated with reduced rate of immunogenicity (figure 1A–C). The two HLA alleles were in LD (R: 0.94), suggesting a single protective effect. In carriers of at least one copy of HLADQA1*03 or HLADRB1*04, MTX was observed to provide stronger protection against ADA development compared with other csDMARDs (HR 0.44; 95% CI 0.24 to 0.78, p=5.7e−03, figure 1B–D). We also investigated HLA alleles that have previously been reported on in RA and Crohn’s disease and provide support for alleles at HLADQA1*05, HLADRB1*11 and HLADRB1*03 (online supplemental figure S4). In conclusion, in the largest study of its type in RA to date, carriage of HLADQA1*03 and HLADRB1*04 reduced the rate of drug immunogenicity to adalimumab. The strongest protection from immunogenicity was conferred by csDMARD cotherapy, particularly in combination with MTX. Our results suggest that the use of alternative csDMARDs should be encouraged for patients treated with MAb TNFi who are MTX intolerant. Larger studies are now needed to determine if genetic testing could optimise Letter


Genotype sample processing
Genotyping was carried out using the Illumina Infinium HumanCoreExome 24 BeadChip kit (Illumina, San Diego, California, USA).250 ng of DNA was used, according to the manufacturer's guidance.Genotype calling was carried out using GenomeStudio software (Illumina, San Diego, California, USA).Standard QC was conducted on each individual array using PLINK v1.9 [2]: SNPs and samples were excluded if there was >2% missing data, and SNPs with MAF < 0.01 and Hardy Weinberg Equilibrium (HWE) p < 1 × 10 −4 were also excluded.Population stratification adjustment was done using HapMap 3 reference panel [3], that includes individuals of European descent, to determine genetic ancestry of each individual, followed by Principal Component Analysis (PCA) analysis.Only individuals of European descent were kept in the dataset.HLA information (types and amino acid) were imputed using SNP2HLA using T1DGC reference panel; imputation refers process of assigning SNP that were not genotyped in the array using a reference panel, the SNPs would then be assigned amino acids, and subsequently allele types [4].

Cox Regression Model
Cox proportional hazards regression model was used to determine immunogenicity rate association to HLA alleles using an additive genetic model.The final genetic model was adjusted for biological sex, age, concurrent conventional synthetic disease modifying anti-rheumatic drug (csDMARD) use, disease duration, and first within-sample principal component from the genetic dataset.After accounting for all available data that includes the above covariates, there were only 435 samples left.Smoking information which could be informative was excluded as there was high number of missing relative to the entire cohort (43%); inclusion of this variable would greatly reduce the power of the study.Another potentially informative variable to be excluded was BMI, this was also due to high number of missing data (18%).However, a final model that included BMI was built, the inclusion of BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)      The region marked in green are the alleles with the strongest association found in this study.The region marked in red are associations found in other studies but with effect sizes (and standard error) measured from the current study.For alleles found in other study, all alleles except for HLA-DRB1*07, were in the same direction, i.e. those found to be risk, were also found to be risk here.+ From Sazonovs et al [6], ++ From Liu et al [7] and +++ From Billiet et al [8].
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) for final Cox regression model for HLA-DRB1, DQA1.Given that DQA1 and DQB1 form a heterodimer, the results for tested alleles at the DQB1 locus are also presented.However, the results for DQB1*02 or DQB1*03 did not meet the p-values the threshold for multiple testing (p=3E-04).+ The results for HLA-DQA1*03 and HLA-DQA*0301 were identical because there is only one 4-digit allele for HLA-DQA1*03.

Figure S5 :
Figure S5: Violin plot for non-trough drug levels for ADA negative and positive samples.Mann-Whitney test indicated statistically significant difference with p: 5.7e-33.

Table S1 :
BMI information did not alter the results for either HLA-DRB1 or HLA-DQA1 in comparison to the same subset of patients where BMI was excluded from the model.For statistical testing the p-value threshold for significance was set to 3E-04.This value is derived from dividing p=0.05 by the number of alleles tested (n=166).Patient characteristics summary for this study.tableonepackagewas used to generate patient characteristics table[5].

Table S2 :
Statistical output BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)