ABSTRACT
Patients treated with etrolizumab, a monoclonal antibody drug for ulcerative colitis, often develop anti‐drug antibodies (ADA) and neutralising antibodies (NAb). To investigate genetic risk factors, we evaluated associations between HLA alleles and ADA/NAb development. HLA‐DQB1*06:03 demonstrated the most significant association with both ADA (OR = 7.3, p ≤ 7.6 × 10−13) and NAb (OR = 13.1, p ≤ 7.7 × 10−13). After controlling for HLA‐DQB1*06:03, HLA‐DQA1*03:03 emerged as the next most significant allele associated with ADA (OR = 2.8, p ≤ 1.0 × 10−3) and NAb (OR = 5.0, p ≤ 1.3 × 10−3). Notably, all patients carrying both alleles (n = 5) developed ADA, suggesting these two alleles are sufficient to induce ADA to etrolizumab.
Keywords: anti‐drug antibodies, etrolizumab, HLA, neutralising antibodies, ulcerative colitis
1. Introduction
Etrolizumab, a monoclonal antibody targeting the β7 subunit of α4β7 and αEβ7 integrin heterodimers, was developed for the treatment of ulcerative colitis [1, 2, 3] and evaluated in Phase III clinical trials. However, these trials were discontinued due to mixed efficacy results. A fraction of patients treated with etrolizumab developed undesirable immune responses, characterised by the formation of anti‐drug antibodies (ADA), a subset of which were neutralising antibodies (NAb). ADAs can alter pharmacokinetics and induce adverse reactions, thereby impacting the overall therapeutic efficacy and safety profiles. NAbs can neutralise the therapeutic effects of the drug, further complicating the therapeutic landscape.
Inherited genetic variation in HLA molecules is known to play a role in ADA development, given their function in presenting peptide antigens to T cells. Previous studies reported associations between specific HLA class II alleles and the risk of ADA development during treatment with therapeutic proteins such as interferon beta (IFNβ), anti‐PD‐L1 and anti‐tumour necrosis factor (anti‐TNF) antibodies [4, 5, 6]. For example, the HLA‐DQA1*05 allele was associated with ADA development (hazard ratio (HR) = 1.9, p ≤ 5.9 × 10−13) in Crohn's disease patients treated with infliximab or adalimumab [5]. Similarly, HLA‐DRB1*01:01 was associated with ADA (odds ratio (OR) = 2.0, p ≤ 3.4 × 10−5) and NAb development (OR = 2.3, p ≤ 2.8 × 10−7) in cancer patients treated with atezolizumab [6].
Based on evidence from monoclonal antibody treatments such as infliximab, adalimumab and atezolizumab, we investigated whether the variable presentation of etrolizumab peptides via HLA molecules contributes to ADA formation. Here, we present a genetic association study involving 615 patients treated with etrolizumab for ulcerative colitis across three different phase III clinical trials: Hickory [1], Laurel [2] and Gardenia [3]. The impact of etrolizumab‐associated ADAs and NAbs development on clinical outcomes is currently under investigation and will be reported separately. In this study, we focus on understanding the baseline genetic characteristics of patients that may predispose them to developing ADAs or NAbs.
2. Materials and Methods
2.1. Studies and Subjects
This research utilised samples from participants who consented to genetic analyses in Roche sponsored clinical trials: Hickory (NCT02100696), Laurel (NCT02165215) and Gardenia (NCT02136069). In total, we included 615 patients of European ancestry across these three studies. Separate analyses in non‐European cohorts could not be performed due to statistical power limitations. The largest non‐European subgroup within any study comprised 23 individuals (most subgroups contained 4–10). Patient clinical and demographic characteristics are detailed in Table 1 and Table S1.
TABLE 1.
Number of patients (n), ADA and NAb rates according to study and treatment arm.
| Study (Study ID) | Treatment | n | n ADA+ (%) | n NAB+ (%) |
|---|---|---|---|---|
| Hickory (GA28950) | Etrolizumab | 178 | 45 (25.3%) | 20 (11.2%) |
| Etrolizumab/Etrolizumab | 74 | 16 (21.6%) | 6 (8.1%) | |
| Etrolizumab/Placebo | 87 | 29 (33.3%) | 14 (16.1%) | |
| Laurel (GA29102) | Etrolizumab | 71 | 23 (32.4%) | 2 (2.8%) |
| Etrolizumab/Etrolizumab | 45 | 16 (35.6%) | 8 (17.8%) | |
| Etrolizumab/Placebo | 48 | 16 (33.3%) | 6 (12.5%) | |
| Gardenia (GA29103) | Etrolizumab/Placebo | 112 | 37 (33%) | 21 (18.8%) |
| Total | 615 | 182 (29.6%) | 77 (12.5%) |
Note: Patients in Etrolizumab arms were treated with etrolizumab. Patients in Etrolizumab/Etrolizumab arms were treated with Etrolizumab in both induction and maintenance phase. Patients in Etrolizumab/Placebo arms were treated with Etrolizumab in the induction phase, but not in the maintenance phase.
2.2. Genotyping and HLA Typing
For the patients in the Hickory and Laurel clinical trials, genomic DNA was extracted from blood samples using the DNA Blood400 kit (Chemagic) and eluted in 50 μL Elution Buffer (EB, Qiagen). DNA was sheared (Covaris LE220) and sequencing libraries were prepared using the TruSeq Nano DNA HT kit (Illumina Inc.). 150 bp paired‐end whole‐genome sequencing (WGS) data was generated to an average read depth of 30× using the HiSeq platform (Illumina X10, San Diego, CA, USA) and processed using the Burrows Wheeler Aligner (BWA)/Genome Analysis Toolkit (GATK) best practices pipeline. Short reads were mapped to hg38/GRCh38 (GCA_000001405.15), including alternate assemblies, using an alt‐aware version of BWA to generate BAM files. All sequencing data was checked for concordance with SNP fingerprint data collected before sequencing.
HLA‐HD [7] was used to infer HLA genotypes starting from BAM files generated as described above. HLA calls at 3‐field resolution were reduced to 2‐field resolution, reflecting the amino‐acid sequence of the HLA protein.
For the Gardenia study, samples were genotyped with the Infinium Global Screen Array (v2) by Illumina and genome‐wide imputation was performed using BEAGLE (v5.1) [8] and the 1000 Genomes reference panel (GRCh38 aligned version). HLA alleles were imputed using the HIBAG [9] software with ancestry‐specific reference panels.
Although different HLA typing methods were employed (HLA‐HD for the Hickory and Laurel; HIBAG for Gardenia), HLA allele frequencies were comparable across all studies (Figure S1). Genetic ancestry for each cohort was inferred from SNP genotyping data using the ADMIXTURE software [10].
2.3. ADA and NAb Bioanalytical Assays
Anti‐etrolizumab antibodies were detected and characterised using a tiered strategy. Initially, serum samples were screened using a validated bridging enzyme‐linked immunosorbent assay (ELISA) assay. Positive samples were then confirmed through competitive binding with etrolizumab (10 μg/mL). Once confirmed, these samples were further characterised for their neutralisation activity using the neutralising ADA (NAb) assay.
2.4. Statistical Analyses
For each clinical trial, we performed logistic regression to assess the association between ADA or NAb and HLA alleles with carrier frequencies between 2% and 98%, assuming a dominant inheritance model. This choice was based on statistical power considerations, as homozygosity for most alleles was rare within individual cohorts and often absent when considering ADA carrier status. A dominant model reduced potential bias in downstream meta‐analyses while capturing the fact that one copy of a risk allele is typically sufficient to increase risk for many HLA‐associated traits. Haplotype analyses were not pursued because phased data were unavailable across cohorts and haplotype resolution in the HLA region is limited, leading to reduced power and potential bias; therefore, we focused on single‐allele models for robust cross‐cohort comparability.
We included age, sex, and three principal components (PCs) as covariates to adjust for population stratification. We conducted a random‐effects meta‐analysis using the R package “meta” to calculate effect estimates, 95% confidence intervals and p values. We applied the Benjamini–Hochberg method to correct for multiple testing.
3. Results
ADA measures for each patient were obtained using a bridging enzyme‐linked immunosorbent assay (ELISA) with positives confirmed via competitive binding with etrolizumab. Neutralisation activities of confirmed positive samples were characterised with a NAb assay. Among 615 patients of European ancestry who provided consent for genetic analyses, 182 (29.6%) patients developed treatment emergent anti‐drug antibodies (ADAs) to etrolizumab across the three studies (Table 1, Table S1). Seventy‐seven of the 182 ADA‐positive patients also developed neutralising antibodies (see Supporting Information for details).
We applied logistic regression on each study to test for association between ADA status and HLA alleles (2‐field resolution). We limited analyses to HLA alleles with carrier frequencies between 2% and 98% (minimising the multiple testing burden that would result from adding lots of alleles with limited power). Age, sex and three genetic principal components were included as covariates. A meta‐analysis, using a random‐effects model implemented in the R package “meta”, identified statistically significant associations (FDR ≤ 0.05 after Benjamini–Hochberg correction for multiple testing) between multiple HLA class II alleles and ADA occurrence (Figure 1, Table S2). Of these alleles, HLA‐DQB1*06:03 demonstrated the most significant association (OR = 7.3, 95% CI = [4.2, 12.6], p ≤ 7.6 × 10−13, FDR ≤ 5.0 × 10−11), with each clinical trial showing an odds ratio greater than 6. HLA‐DQB1*06:03 is part of a common haplotype that encompasses the three top‐associated alleles (HLA‐DQB1*06:03, HLA‐DRB1*13:01, HLA‐DQA1*01:03, Table S2), which are therefore not statistically independent. A conditional analysis, including HLA‐DQB1*06:03 as an additional covariate, showed no residual signal at HLA‐DRB1*13:01 and HLA‐DQA1*01:03, and uncovered an independent association of a different HLA class II allele, HLA‐DQA1*03:03 (OR = 2.8, 95% CI = [1.5, 5.3], p ≤ 1.0 × 10−3, FDR ≤ 0.04, Figure 1, Table S3).
FIGURE 1.
Forest plot showing the effect estimates of the two independent HLA alleles: HLA‐DQB1*06:03 (column 2–3) and HLA‐DQA1*03:03 (column 4–5). The effects of HLA‐DQA1*03:03 were estimated after conditioning on the first independent allele HLA‐DQB1*06:03. Bars represent 95% confidence intervals (CI) for odds ratios.
We then conducted a subgroup analysis on patients who developed NAbs, which directly inhibit the biological activity of the drug by binding to its target binding domain. We compared NAb‐positive patients to ADA‐negative patients, excluding ADA‐positive but NAb‐negative patients. The analysis revealed that both independent alleles identified in the previous analysis were also significant in this subgroup: HLA‐DQB1*06:03 (OR = 13.1, 95% CI = [6.5, 26.5], p ≤ 7.7 × 10−13, FDR ≤ 4.9 × 10−11) and HLA‐DQA1*03:03 (OR = 5.0, 95% CI = [1.9, 13.3], p ≤ 1.3 × 10−3, FDR ≤ 0.04, Figure 1). This finding suggests that individuals with these alleles have a predisposition to developing immune responses against etrolizumab, affecting both the binding and neutralising capacities of the drug.
Case frequencies, stratified by allele dosage, indicate that both DQB1*06:03 and DQA1*03:03 are independently associated with ADA and NAb case frequencies (Figure 2A). Notably, the risk of developing ADA or NAb increases with the number of unique independent alleles carried by each individual in each study (Figure 2B). In particular, all 5 patients carrying both independently associated HLA alleles (HLA‐DQB1*06:03 and HLA‐DQA1*03:03) were positive for ADA, with 3 of them also positive for Nab. Though numbers are small, these findings may indicate that the presence of both alleles is sufficient to induce ADA formation.
FIGURE 2.
(A) Heatmaps showing case frequency of ADA (left) or NAb (right) across all cohorts, stratified by allele counts for DQB1*06:03 (x‐axis: 0, 1, 2 copies) and DQA1*03:03 (y‐axis: 0, 1, 2 copies). Circle area and colour encode the case frequency; the value inside each circle is the frequency, and the values below each circle show #case/#total counts for that cell; cells with zero total have no circle and display NA. (B) The risk of developing ADA or NAb (y‐axis) increases with the number of unique independent alleles carried by each individual (x‐axis). A value of 0, 1 and 2 on the x‐axis represent individuals carrying none, either and both of the two independent alleles, respectively. The presence of an allele, either homozygous or heterozygous, is counted as 1 toward the unique allele count. The size of the point represents the number of allele carriers.
4. Discussion
In the context of a limited understanding of patient factors driving anti‐drug antibody (ADA) risk, our study has identified two HLA class II alleles (HLA‐DQB1*06:03 and HLA‐DQA1*03:03) associated with an increased risk of ADA formation against etrolizumab. Our findings unveil one of the strongest reported associations to date between an HLA allele (HLA‐DQB1*06:03) and both ADA (p ≤ 7.6 × 10−13) and NAb (p ≤ 7.7 × 10−13) status for antibody therapeutics. Similarly, Sazonovs et al. [5] reported a strong association (p ≤ 5.9 × 10−13) between HLA‐DQA1*05 and ADA development in Crohn's disease patients with infliximab or adalimumab therapy. These findings suggest that while HLA plays a crucial role in ADA development, different alleles are involved in distinct patient populations. Further research on the generalisability and specificity of genetic signals across different antibody therapies could provide valuable insights for advancing drug development, such as informing clinical trial recruitment strategies and enhancing personalised treatment strategies, including the prospect of HLA typing prior to therapy administration.
The carrier frequencies of the two alleles reported here are close to 10% in people of European ancestry (13.3% for HLA‐DQB1*06:03 and 8.9% for HLA‐DQA1*03:03 in our dataset, Table S4), which suggests that a considerable subset of the patient population is predisposed to ADA formation to etrolizumab. For HLA‐DQB1*06:03, reported frequencies include 7.9% in > 3.4 million German donors and 15.0% in 23,595 Polish donors [11]. For HLA‐DQA1*03:03, reported frequencies range from ~0.4%–1.1% in Southern and Central European cohorts (e.g., Spain Murcia, n = 173; Italy Central, n = 380) to ~13.5% in a Dutch cohort (n = 155) and 12.9% in a U.S. Caucasian cohort (n = 307) [11]. These values align with the carrier frequencies observed in our study. Furthermore, the positive predictive value of these alleles for predicting ADA development is high, with 68.3% of HLA‐DQB1*06:03 carriers and 47.2% of HLA‐DQA1*03:03 carriers developing ADA. This prevalence and high predictive value underscore the importance of identifying and potentially eliminating immunogenic epitopes within recombinant large molecule therapeutics.
In addition, two other alleles, HLA‐DRB1*13:01 and HLA‐DQA1*01:03, which are in linkage disequilibrium with HLA‐DQB1*06:03, expectedly showed strong associations with ADA development (OR: 6.2 and 3.7, respectively; p ≤ 2.9 × 10−9 and 1.4 × 10−7, respectively). The haplotype containing these alleles likely contributes to ADA risk; however, further studies are required to identify the causal allele(s). Specifically, a haplotype‐based analysis utilising phased genotype data could be useful to pinpoint causal alleles in such haplotypes.
A limitation of our analysis was the generally small number of ADA and NAb cases among the DQB1*06:03 and DQA1*03:03 carriers in each individual study (Table S4). However, at least five cases were present in every group (except for NAb cases among DQA1*03:03 carriers in the Laurel study), and this within‐study limitation was mitigated by the overall meta‐analysis. Nevertheless, a larger combined cohort would be beneficial for future research and the possible identification of additional risk alleles. A minor limitation is the need for statistical imputation of HLA alleles from SNP genotyping data for one of the studies, as the 1000 Genomes reference panel does not fully capture global HLA variation [12]; however, since our analyses focused on common alleles in participants of European ancestry and the results are consistent with the other two studies based on inference from WGS, this limitation is unlikely to have materially impacted our results.
It is important to recognise that HLA alleles only partially explain the variance in ADA formation. Consequently, future research should continue to explore other potential risk factors and molecular mechanisms to better understand and mitigate the impact of ADA on therapeutic efficacy and patient outcomes, potentially leading to more personalised and effective treatment strategies.
Ethics Statement
The research was performed with samples from subjects who had given consent for genetic research in Roche sponsored clinical trials: Hickory (NCT02100696), Laurel (NCT02165215) and Gardenia (NCT02136069). In each of these trials, the Ethics Committees (EC) and Institutional Review Boards (IRB) who approved the trials also approved the Informed Consent Forms (ICF) used to obtain consent for genetic research from the study participants. No EC/IRB was additionally consulted to approve the specific genetic research reported here but an internal team of consent experts made sure the genetic research is covered by the ICFs signed by the study participants. Complete lists of ECs and IRBs are available in the Supporting Information (Tables S5–S7).
Conflicts of Interest
All authors are/were employees of Genentech/Roche. Christian Hammer is currently a full‐time employee of Altos Labs.
Supporting information
Data S1: tan70557‐sup‐0001‐supinfo.pdf.
Acknowledgements
Artificial intelligence (GPT‐4, Gemini‐2.5) was used to edit texts.
Saha A., McCarthy M. I., Fischer S. K., Sperinde G., and Hammer C., “ HLA Class II Alleles Are Associated With Anti‐Drug Antibodies in Ulcerative Colitis Patients Treated With Etrolizumab,” HLA 107, no. 2 (2026): e70557, 10.1111/tan.70557.
Data Availability Statement
HLA association summary statistics are available in the Supporting Information. Qualified researchers may request access to individual patient data used in this study through Roche data sharing platforms in accordance with the Global Policy on Sharing of Clinical Study Information: http://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_to_data_sharing.htm.
References
- 1. Peyrin‐Biroulet L., Hart A., Bossuyt P., et al., “Etrolizumab as Induction and Maintenance Therapy for Ulcerative Colitis in Patients Previously Treated With Tumour Necrosis Factor Inhibitors (HICKORY): A Phase 3, Randomised, Controlled Trial,” Lancet Gastroenterology & Hepatology 7, no. 2 (2022): 128–140, 10.1016/s2468-1253(21)00298-3. [DOI] [PubMed] [Google Scholar]
- 2. Vermeire S., Lakatos P. L., Ritter T., et al., “Etrolizumab for Maintenance Therapy in Patients With Moderately to Severely Active Ulcerative Colitis (LAUREL): A Randomised, Placebo‐Controlled, Double‐Blind, Phase 3 Study,” Lancet Gastroenterology & Hepatology 7, no. 1 (2022): 28–37, 10.1016/s2468-1253(21)00295-8. [DOI] [PubMed] [Google Scholar]
- 3. Danese S., Colombel J. F., Lukas M., et al., “Etrolizumab Versus Infliximab for the Treatment of Moderately to Severely Active Ulcerative Colitis (GARDENIA): A Randomised, Double‐Blind, Double‐Dummy, Phase 3 Study,” Lancet Gastroenterology & Hepatology 7, no. 2 (2022): 118–127, 10.1016/s2468-1253(21)00294-6. [DOI] [PubMed] [Google Scholar]
- 4. Link J., Ryner M. L., Fink K., et al., “Human Leukocyte Antigen Genes and Interferon Beta Preparations Influence Risk of Developing Neutralizing Anti‐Drug Antibodies in Multiple Sclerosis,” PLoS One 9, no. 3 (2014): e90479, 10.1371/journal.pone.0090479. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Sazonovs A., Kennedy N. A., Moutsianas L., et al., “HLA‐DQA1*05 Carriage Associated With Development of Anti‐Drug Antibodies to Infliximab and Adalimumab in Patients With Crohn's Disease,” Gastroenterology 158, no. 1 (2019): 189–199, 10.1053/j.gastro.2019.09.041. [DOI] [PubMed] [Google Scholar]
- 6. Hammer C., Ruppel J., Kamen L., Hunkapiller J., Mellman I., and Quarmby V., “Allelic Variation in HLA‐DRB1 Is Associated With Development of Antidrug Antibodies in Cancer Patients Treated With Atezolizumab That Are Neutralizing In Vitro,” Clinical and Translational Science 15, no. 6 (2022): 1393–1399, 10.1111/cts.13264. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Kawaguchi S., Higasa K., Shimizu M., Yamada R., and Matsuda F., “HLA‐HD: An Accurate HLA Typing Algorithm for Next‐Generation Sequencing Data,” Human Mutation 38, no. 7 (2017): 788–797, 10.1002/humu.23230. [DOI] [PubMed] [Google Scholar]
- 8. Browning B. L., Zhou Y., and Browning S. R., “A One‐Penny Imputed Genome From Next‐Generation Reference Panels,” American Journal of Human Genetics 103, no. 3 (2018): 338–348, 10.1016/j.ajhg.2018.07.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Zheng X., Shen J., Cox C., et al., “HIBAG—HLA Genotype Imputation With Attribute Bagging,” Pharmacogenomics Journal 14, no. 2 (2014): 192–200, 10.1038/tpj.2013.18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Alexander D. H., Novembre J., and Lange K., “Fast Model‐Based Estimation of Ancestry in Unrelated Individuals,” Genome Research 19, no. 9 (2009): 1655–1664, 10.1101/gr.094052.109. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Gonzalez‐Galarza F. F., McCabe A., dos Santos E. J. M., et al., “Allele Frequency Net Database (AFND) 2020 Update: Gold‐Standard Data Classification, Open Access Genotype Data and New Query Tools,” Nucleic Acids Research 48, no. D1 (2019): D783–D788, 10.1093/nar/gkz1029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Abi‐Rached L., Gouret P., Yeh J. H., et al., “Immune Diversity Sheds Light on Missing Variation in Worldwide Genetic Diversity Panels,” PLoS One 13, no. 10 (2018): e0206512, 10.1371/journal.pone.0206512. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1: tan70557‐sup‐0001‐supinfo.pdf.
Data Availability Statement
HLA association summary statistics are available in the Supporting Information. Qualified researchers may request access to individual patient data used in this study through Roche data sharing platforms in accordance with the Global Policy on Sharing of Clinical Study Information: http://www.roche.com/research_and_development/who_we_are_how_we_work/clinical_trials/our_commitment_to_data_sharing.htm.