Beta-lactam-induced immediate hypersensitivity reactions: A genome-wide association study of a deeply phenotyped cohort

Background: b -lactam antibiotics are associated with a variety of immune-mediated or hypersensitivity reactions, including immediate (type I) reactions mediated by antigen-speciﬁc IgE. Objective: We sought to identify genetic predisposing factors for immediate reactions to b -lactam antibiotics. Methods: Patients with a clinical history of immediate hypersensitivity reactions to either penicillins or cephalosporins, which were immunologically conﬁrmed, were recruited from allergy clinics. A genome-wide association study was conducted on 662 patients (the discovery cohort) with a diagnosis of immediate hypersensitivity and the main ﬁnding HLA-DRB1*10:01 increased the risk of immediate hypersensitivity at a genome-wide level (odds ratio, 2.96; P 5 4.1 3 10 2 9 ). No association with HLA-DRB1*10:01 was identiﬁed in 268 patients with delayed hypersensitivity reactions to b -lactams. Conclusions: HLA-DRB1*10:01 predisposed to immediate hypersensitivity reactions to penicillins. Further work to identify other predisposing HLA and non-HLA loci is required. (J Allergy Clin Immunol 2021;147:1830-7.)

HLA-DRB1*10:01 increased the risk of immediate hypersensitivity at a genome-wide level (odds ratio, 2.96; P 5 4.1 3 10 29 ). No association with HLA-DRB1*10:01 was identified in 268 patients with delayed hypersensitivity reactions to b-lactams. Conclusions: HLA-DRB1*10:01 predisposed to immediate hypersensitivity reactions to penicillins. Further work to identify other predisposing HLA and non-HLA loci is required. (J Allergy Clin Immunol 2021;147:1830-7.) Key words: Type I hypersensitivity, b-lactams, penicillins, cephalosporins, allergy, anaphylaxis, pharmacogenomics b-Lactam (BL) antibiotics cause a wide spectrum of hypersensitivity reactions (sometimes termed allergy). The self-reported incidence of BL allergy ranges from 1% to >10%, 1 but in clinic populations, most patients (;95%) are not found to be truly allergic with validated skin testing and oral challenge. Indeed, a high proportion are intolerant 2 as adverse effects such as diarrhea after the use of BLs are often mistakenly reported as allergy by patients.
True BL hypersensitivity reactions are classified according to the time of onset of the reaction following drug intake. 3 Immediate hypersensitivity reactions develop in minutes or hours after drug intake and are due to cross-linking of specific IgE molecules on the mast cell surface with release of vasoactive mediators such as histamine leading to vasodilation, increased vascular permeability, and smooth muscle contraction. 4 Clinically this is manifested as urticaria, angioedema, bronchospasm, and hypotension. Anaphylaxis is the most severe and feared form of immediate hypersensitivity. By contrast, delayed hypersensitivity reactions occurring >6 hours after dosing are typically T-cellmediated and have variable manifestations including maculopapular exanthem, drug reaction with eosinophilia and systemic symptoms, and Stevens-Johnson syndrome/toxic epidermal necrolysis. 3 Medicines are among the main cause of fatal anaphylaxis with a mortality rate higher than with other agents. 5 Penicillins and cephalosporins are still the 2 most common drug classes associated with anaphylaxis, 6 with penicillins having a higher incidence (1-5 per 100,000) 7 compared with cephalosporins. 1 Crossreactivity between penicillins, cephalosporins, and other BLs not sharing an R1 or R2 side chain is now thought to be <2%. 8,9 Potential clinical risk factors for immediate hypersensitivity reactions are family history, atopy, concomitant virus infections, and the route of administration. 10 Genetic predisposing factors have also been identified: 10 the most comprehensive was an analysis of 107,398 single nucleotide polymorphisms (SNPs) that identified that the HLA-DRA locus may protect against penicillin-induced immediate hypersensitivity reactions. 11 To further investigate the role of genetic factors in BL-induced immediate hypersensitivity reactions, we have undertaken a genome-wide association study of the largest deeply phenotyped patient cohort assembled so far.

METHODS Cases
All subjects were recruited between 2009 and 2013 as part of International Consortium on Drug Hypersensitivity, involving 5 recruitment centers worldwide (Australia, France, Italy, Spain, and United Kingdom). The study was approved by ethics committees in all countries, and all patients gave written informed consent.
We recruited 662 patients (the discovery cohort) with a diagnosis of immediate hypersensitivity to BL antibiotics ( Table I). The diagnosis of immediate hypersensitivity was made in specialist allergy clinics, as per published criteria. 12 All patients required immunological assessment (total and specific IgE, skin testing including skin prick test and intradermal and/or oral provocation) as part of the inclusion criteria. Independent adjudication of all cases was undertaken by N.H.S. and P.S.F. For replication of any signals, we separately recruited another 98 patients with immediate hypersensitivity from a clinic in Spain, diagnosed according to the same criteria.
To determine specificity of any signals identified in patients with immediate hypersensitivity, we also evaluated 268 patients with delayed hypersensitivity reactions across multiple BLs. The diagnosis again was in accordance with published guidance, 12 and all cases were adjudicated by N.H.S. and P.S.F. We also included an additional 17 BL-induced delayed hypersensitivity reaction cases analyzed in Shen et al. 13

Controls
We used general population samples as study controls. This comprised 9217 European ancestry controls from multiple available sources enriching the group with Spanish, Italian, and north European origin samples because cases were mainly recruited from those countries. We used the Wellcome Trust Case Control Consortium (http://www.wtccc.org.uk); the Population Reference Sample (POPRES), 14 PGX4000119, 13 LAM30004, 13 and Spanish Bladder cancer cohort (phs000346.v1) 15 from the Database of Genotypes and Phenotypes (dbGaP); Hypergenes cohort (http://www.hypergenes.eu/); the National Spanish DNA Bank (http://www.bancoadn.org/); and Toscani in Italia ([TSI] HapMap data) to obtain ancestry control data. In addition, we also recruited a group of 137 penicillin-tolerant controls from Italy.

Genotyping
Genome-wide genotyping of DNA extracted from whole blood was performed at the Broad Institute (Boston, Mass) for 662 cases with BLinduced immediate hypersensitivity and 268 cases with delayed hypersensitivity reaction, and from 137 penicillin-tolerant controls from Italy. In 474 (354 BLinduced immediate and 120 BL-induced delayed) cases, the Illumina Infinium HumanCoreExome BeadChip (Illumina, Inc, San Diego, Calif) was used while for 439 (308 BL-induced immediate and 131 BL-induced delayed) cases, the Illumina HumanOmniExpress BeadChip was used. In this last batch, we also genotyped 137 Italian penicillin-tolerant controls. In addition, the BL-induced delayed case group also included 17 BL-delayed hypersensitivity cases previously genotyped by the Illumina 1M Duo chip, extracted from a larger Stevens-Johnson syndrome/toxic epidermal necrolysis study that included multiple drugs, as described by Shen et al. 13 Other control cohorts were publicly available (see Table E1 in this article's Online Repository at www.jacionline. org). For each of the genotyping cohort, standard quality control was conducted at both single marker and subject levels as previously described. 13 This was followed by SNP and HLA imputation and amino acid analysis (see the Methods section in this article's Online Repository at www.jacionline.org). UK) or iPLEX MassArray genotyping platform (Agena Biosciences, Hamburg, Germany). High-resolution genotyping of HLA-A, HLA-B, HLA-C, DRB1, DQA1, and DQB1 was performed by Histogenetics (Ossining, NY). Sequencing data files were analyzed using Histogenetics' proprietary analysis software (Histomatcher and HistoMagic) for HLA genotype calling. Allele assignments are based on IMGT/HLA Database (release version 2.21.0, dated April 2008; http://www.ebi.ac.uk/imgt/hla/).

Statistical analysis
The effect of population structure was assessed through principal component analysis using the smartPCA program from the EIGENSTRAT package (version 3.0; Alkes Price, Harvard T.H. Chan School, Boston, Mass). 16 Single marker and haplotype association analyses and heterogeneity test analyses were carried out by PLINK 1.07. 17 The statistical association of each marker, HLA alleles and SNPs, was determined in a logistic regression framework with scores for the first 7 principal components as covariates under an additive model using PLINK. We used the same statistical test for subpopulation analyses, using the 2, 7, and 10 most significant principal components as covariates in Italian, Spanish, and North European populations, respectively. We set the genome-wide traditional significance P-value threshold to 5.0 3 10 28 to correct for multiple testing and MHC-wide significance threshold to 2.0 3 10 24 to correct for total number of predicted alleles. When we obtained genome-wide significant signals, we tested for independent effects from the neighboring variants by including the most associated variants as a covariate and then testing the significance of others in the region. All detailed analyses and Manhattan plots were performed with R (version 3.0.2; R Foundation, Vienna, Austria). Regional plots were drawn by Locus-Zoom. 18 Meta-analysis was performed using a fixed-effect model in the metafor package (http://www.metafor-project.org/doku.php/metafor).

Patient cohorts
The clinical characteristics of the patients are shown in Table I. Clinical manifestations in the discovery cohort included angioedema (35%), bronchospasm (24%), and urticaria (34%), while hypotension was reported in only 4% of cases. The length of reaction in patients with immediate hypersensitivity was 2 to 11 days, while it ranged from 21 to 26 days for patients with delayed hypersensitivity reactions. Patients were included if they had positive diagnostic assessment, as highlighted in Table I. Penicillins accounted for 75% of cases, with the most common culprit drug being amoxicillin accounting for 58% of cases in the discovery cohort.

Association with immediate reactions to BLs
We first conducted a genome-wide association study on 662 patients of European descent with immediate hypersensitivity jacionline.org). The frequency of rs71542416 in our control population was comparable with that reported in publicly available sources (Table II). HLA allele imputation using HLA genotype imputation with attribute bagging, or HIBAG, 19 showed the HLA-DRB1*10:01 (OR, 2.95; 95% CI, 1.99-4.36; P 5 6.0 3 10 28 ) and HLA-DQA1*01:05 (OR, 2.93; 95% CI, 1.92-4.45; P 5 5.4 3 10 27 ) alleles to be significantly associated with the immediate reactions, with consistent ORs (Table II) and were tagged by rs71542416 (r 2 5 0.76). Haplotype analysis identified that HLA-DRB1*10:01 was a risk factor even without the HLA-DQA1*01:05 allele (see Table E2 in this article's Online Repository at www.jacionline.org). HLA-DRB1*10:01 was seen in 3% of cases and <1% of controls. The frequency of the HLA alleles within the Italian penicillin-tolerant controls was 10 times less than in the Italian general population (0.1% vs 1%).
The HLA allele effect size was similar across the 3 major clusters (heterogeneity test, P 5.11) ( Table III). The positive predictions in cases were fully validated by direct HLA typing. An additional 67 cases with low-quality predictions in both the loci were also typed. Among them, we found only 1 positive carrier for HLA-DRB1*10:01. All cases were also genotyped for rs71542416-this showed a concordance of 99% between typed and imputed genotypes of rs71542416. HLA-DRB1*10:01 cooccurred with rs71542416 in 89% of the HLA-DRB1*10:01-positive patients, while 12% of all cases carried rs71542416 alone.
Including rs71542416 or the HLA alleles as covariates revealed a residual protective effect of the HLA-DRA locus, tagged by rs114632839, an intronic gene variant, in accordance with the findings of Gueant et al 11 (Table II and see Figs E3  loci with the minor alleles showing a lower intron excision ratio. Both effects were detected in whole blood and shared across other tissues (Fig E4, B).
A replication cohort of 98 patients with anaphylaxis induced by either amoxicillin or amoxicillin-clavulanate (Table I) was recruited separately from Spain. We identified 7 individuals who were positive for HLA-DRB1*10:01, as confirmed by HLA typing. Comparison using the 11 Spanish HLA-typed cohorts reported at allelefrequency.net provided a total of 3137 Spanish subjects (see Fig E5 in this article's Online Repository at www. jacionline.org) as ethnically matched population controls. This analysis replicated the association with an OR of 2.80 (95% CI, 1.17-6.71; Fisher exact test, P 5 .016) (Fig 1, B).
Meta-analysis of the discovery and replication cohorts showed that HLA-DRB1*10:01 increased the risk of immediate hypersensitivity at a genome-wide level (OR, 2.96; 95% CI, 1.99-4.37; P 5 4.1 3 10 29 ) (Fig 1, B). The sensitivity and specificity of the allele is 0.06 and 0.98, respectively, while the positive and negative predictive values are 17% and 94%, respectively.

HLA analysis in patients with delayed hypersensitivity reactions
To determine whether the association with HLA-DRB1*10:01 was limited to patients with immediate hypersensitivity reactions, Minor allele frequency for external data obtained from allelefrequncy.net for HLA alleles or GnomAD for SNPs. ORs are of the logistic regression model, correcting for population stratification. P values are logistic regression P. *HLA haplotype was HLA-DRB1*10:01-HLA-DQA1*01:05.
The marker rs114632839 has merged with rs3135392.
Drug-specific associations with immediate hypersensitivity HLA-DRB1*10:01 was associated with penicillins as a class (OR, 3.07), but not with cephalosporins (Table IV). Among the penicillins, the strongest signals were for amoxicillin (OR, 3.48), amoxicillin clavulanic acid (OR, 2.85), and phenoxymethylpenicillin (OR, 6.66) (Table IV). When we combined amoxicillin and amoxicillin clavulanic acid cases (assuming that amoxicillin rather than clavulanic acid was the culprit), the OR was 3.1 (95% CI, 2.01-4.85; P 5 4.0 3 10 27 ). Additional drug-specific HLA allele associations that we identified will need confirmation (see Tables E5 and E6 in this article's Online Repository at www.jacionline.org).
In the drug-specific analysis, a genome-wide signal (rs71437970) on chromosome 13 upstream of SLC15A1 (see Fig  E6 in this article's Online Repository at www.jacionline.org) was   (Table E6). This association was shared across the European subpopulations and with amoxicillin-clavulanate cases (see Tables  E7 and E8 in this article's Online Repository at www.jacionline. org). However, we failed to replicate the association, with an allele frequency that was lower than that observed in Spanish controls (0.007 vs 0.04).

DISCUSSION
We have identified an association between the SNP rs71542416 and immediate hypersensitivity reactions to penicillins. The SNP does not affect gene expression in GTEx but is in linkage disequilibrium with HLA-DRB1*10:01 and HLA-DQA1*01:05. Haplotype analysis identified that HLA-DRB1*10:01 was a risk factor even without the HLA-DQA1*01:05 allele suggesting that HLA-DRB1*10:01 may be the predominant driver of the association. However, 12% of cases carried rs71542416 but were negative for HLA-DRB1*10:01, suggesting that the SNP may be a tag for other rare HLA alleles, which is consistent with the hypothesis of Heap et al 21 who showed an association between HLA-DQA1-HLA-DRB1 variants and thiopurineinduced pancreatitis.
The association with HLA-DRB1*10:01 and rs71542416 was most pronounced in the Spanish cohort (Table III), but given that the ORs were of similar magnitude in all populations studied, there was overlap in the CIs, and the prevalence of the SNP and HLA allele, our findings can be generalized across the European subethnicities studied (Table III). However, further studies will be needed in both European and non-European populations to determine the global relevance of this association. Additionally, the association was limited to immediate reactions and was not observed with the delayed hypersensitivity reactions, highlighting the specific nature of the association. Evaluation of drug-specificity showed associations with amoxicillin, amoxicillin-clavulanate, and phenoxymethylpenicillin. However, given the limited sample size with the other penicillins, we cannot exclude the possibility of an association with all penicillins (Table IV). Similarly, we did not find an association with cephalosporins, but this may also be because of a lower sample size.
The clear strength of our study is that all patients were deeply phenotyped: there was a clear clinical history with a temporal relationship to drug intake, and the diagnosis was confirmed immunologically by skin testing and/or oral provocation. Such deep phenotyping is important because many patients claim to be allergic to penicillin, but very few are; of those claiming to be allergic, <1 in 20 have an acute reaction to an oral challenge (the gold standard clinical test to confirm an IgE-mediated reaction). 22 Our data add to the increasing evidence of HLA in predisposing to different clinical phenotypes of drug hypersensitivity reactions. 23 The most well-known of these associations is HLA-B*57:01 and abacavir hypersensitivity, 24 which has been implemented into clinical practice and has resulted in a significant reduction in abacavir hypersensitivity. 25 It is important to note that most of the HLA associations identified to date have been with delayed hypersensitivity reactions. 23 However, more recent studies have identified HLA alleles as predisposing factors for immediate reactions. For instance, HLA-DRB1*07:01 is a risk factor for the development of anti-asparaginase antibodies and immediate reactions. 26 Our data, which show that HLA-DRB1*10:01 predisposes to immediate hypersensitivity, are also consistent with the pathogenesis of immediate reactions where the interaction between B cells and CD4 1 /T H 2-positive cells, through HLA Class II alleles, is central to the immunoglobulin switching that leads to the generation of specific IgE antibodies. Different HLA alleles have been associated with other types of immune-mediated reactions caused by BLs. For example, HLA-B*57:01 predisposes to flucloxacillin-induced cholestatic hepatitis, 27 while liver injury caused by amoxicillin-clavulanate is associated with the Class II HLA haplotype HLA-DRB1*1501-DQB1*0602. 28 Mechanistic studies undertaken in our laboratory have shown that drugspecific, HLA-restricted T cells can be isolated from patients with a past history of liver injury due to flucloxacillin 29 and amoxicillin-clavulanate. 30 It will be valuable to conduct similar studies in patients with a history of penicillin-induced immediate reactions to understand the mechanistic basis of the association with HLA-DRB1*10:01. Another potentially interesting finding in this study was the association between SLC15A1 gene variants and amoxicillininduced immediate reactions. SLC15A1 encodes the human peptide transporter 1, which is known to transport amoxicillin. 31 Therefore, it is plausible that variation in the activity of human peptide transporter 1 could result in altered amoxicillin pharmacokinetics and thereby increase risk of a type I reaction. However, we were not able to replicate this finding, and further work (including functional studies) to understand whether this gene is important in predisposing to immediate reactions will be required.
What are the clinical implications of this finding? Given the rarity of penicillin-induced anaphylaxis, the low population prevalence and sensitivity of HLA-DRB1*10:01, and the very wide usage of penicillins, the prospective use of this allele in screening patients before penicillin prescription would not be practical or feasible in terms of both the high numbers needed to test to prevent 1 case and patients unnecessarily excluded from therapy. However, this association of immediate penicillin hypersensitivity with HLA-DRB1*10:01 may provide much novel insights into the mechanisms of immediate reactions associated with penicillins, including the mechanisms of sensitization and natural loss or waning of penicillin, which is known to occur over time. Moreover, the higher negative predictive value of the allele (94%) may be of use in risk stratifying patients where penicillin cannot be excluded as an etiological agent in the setting of an immediate reaction.
Our study has limitations. First, the overall sample size is small compared with that used in complex diseases, but it is larger than that used in many pharmacogenomic studies. Our efforts to identify deeply phenotyped patients in this study was a result of an extensive international collaboration highlighting the difficulties in achieving large sample sizes in pharmacogenomic studies. Furthermore, we were unable to perform permutation testing to validate the replication P value for HLA-DRB1*10:01. Second, because we used population controls, we could not adjust for self-reported ethnicity, but this is unlikely to have had a major impact as we accounted for this through an analysis of population stratification ( Fig E1). Third, matching cases and controls for age, sex, and other comorbidities was not possible because of the use of population controls, and because sex could not be determined due to the absence of X chromosome SNP data. Whether this impacts on the association with the genetic signals identified by us will require further study.
In summary, we have for the first time reported an association of HLA-DRB1*10:01 carriage in deeply immunologically phenotyped European ancestry individuals with penicillin-induced immediate type I hypersensitivity reactions. However, we cannot exclude the possibility of other HLA alleles or HLA haplotypes also being important in conferring susceptibility in some patients, and therefore further work in both European and non-European patients is required to identify other HLA alleles, and also whether HLA-DRB1*10:01 is universally important. It is also interesting to note that we also identified that rs114632839, which is a proxy for the HLA-DRA locus, protected against the development of immediate hypersensitivity reactions to BLs, consistent with a previous study. 11 rs114632839 is an expression QTL and splicing QTL for several HLA loci, suggesting that predisposition to immediate hypersensitivity to penicillins is likely to be complex and mediated by a combination of susceptibility and protective HLA and non-HLA alleles. Clearly we have reported associations, and proof of causality will require a full understanding of the immunopathogenesis of initial sensitization to penicillin, and in particular, the mechanism of antigen presentation (including the relative importance of the BL ring vs the side chains) and interaction with CD4 1 T cells that ultimately leads to IgE-switching and the generation of hapten-specific IgE antibodies.
We offer special thanks to Arthur Holden for his help and effort in guiding this collaborative work and to the Broad genotyping facility for their contribution to the genome-wide association study genotyping. We also acknowledge the contribution of all our clinical collaborators and the study participants.

Imputation procedure
Prephasing and imputation was performed in batches by dividing the cases and controls according to the genotyping platform. For each batch, we first prephased the genotype scaffold using SHAPEIT (version v2.r727). E1 Imputation was then undertaken using IMPUTE2 (version 3) E2 using the reference panel from the 1000 Genomes Project (release v3). E1 We used the ''all ancestries'' reference panel to improve the quality of the imputation especially of lower frequency variants. E2 For downstream analysis, we used best-guess genotypes retaining imputed genotypes with posterior probability >0.9. In the combined dataset, we then retained SNPs with: (1) no significant difference in missingness between cases and controls (P > .0001); (2) no significant deviation from Hardy-Weinberg equilibrium (P > .0001); (3) < 5% missing genotypes in all batches; (4) imputation information score >0.8 in all batches; and (5) MAF in the European descent haplotypes from the 1000 Genomes Project reference panel of at least 0.01. Finally, we compared SNP allele frequencies in controls between genotyping platforms. Specifically, we tested for association of genotyping platforms with SNPs in same ethnicity controls in a logistic regression framework, under an additive genetic model, with adjustment for principal components from smartPCA to account for population structure. SNPs with nominal evidence of association (P < .0001) were excluded. Sex chromosomes and mitochondria were not imputed.
For each batch, HLA alleles were inferred using HIBAG E3 using the reference predictor panels specific for the genotyping chip and Caucasian ancestry. Amino acid changes were inferred using SNP2HLA utilizing the reference data collected by the Type 1 Diabetes Genetics Consortium. E4        OR and P value as defined in Table II.