Practical implementation of genetics: new concepts in immunogenomics to predict, prevent and diagnose drug hypersensitivity

Abstract

Delayed drug hypersensitivities are CD8+ T-cell mediated reactions associated with up to 50% mortality. Human leukocyte antigen (HLA) alleles are known to predispose disease, specific to drug, reaction, and patient ethnicity, with pre-treatment screening recommended for a handful of the strongest associations to identify and prevent drug use in high-risk patients. However, an incomplete predictive value implicates other HLA-imposed risk factors, and low carriage of many identified HLA-risk alleles combined with the high cost of sequence-based typing has limited economic viability for similar recommendation of screening across drugs and healthcare systems. To mitigate, an expanding armoury of low-cost polymerase chain reaction-based screens is being developed, and HLA-imposed risk factors are being discovered. These include (1) polymorphic variants of metabolic and endoplasmic reticulum aminopeptidase enzymes, towards multi-allelic screening with increased predictivity, (2) regulation by immune checkpoint inhibitors, enabling de-tolerised animal models of human disease, and (3) immunodominant T-cell receptors (TCR) on clonally expanded CD8+ T-cells. For the latter, HLA-risk restricted TCR provides immunogenomic strategies and samples from a single patient to identify novel HLA-risk associations in underserved minority populations, tissue-relevant effector biomarkers towards earlier diagnosis and treatment, and HLA-TCR-presented immunogenic structures to aid future drug development.

INTRODUCTION TO IMMUNOGENOMICS AND DRUG HYPERSENSITIVITY

Immunogenomics incorporates sequencing and genetic engineering platforms to assess how variance of immune-related genes influences immune function and disease onset and severity. Despite the average human genome differing from reference consensus by 4.1–5 million single nucleotide polymorphisms (1), the human leukocyte antigen (HLA) complex has been a disease-spanning focus of inter-individual immune susceptibility as the most polymorphic loci in the human body. Indeed, there are now >30,000 HLA allelic variants described (2). With critical function in presentation of antigen to T-cells, different HLA alleles have been associated with altered susceptibility to T-cell mediated disease including delayed drug hypersensitivity reactions (DHR). CD8+ T-cell mediated DHR often target specific tissues. Severe manifestations include drug-induced liver injury (DILI) or cutaneous reactions including drug reaction with eosinophilia and systemic systems (DRESS) or Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN), associated with up to 9%, 10%, or 50% mortality, respectively (3–5). Treatment approaches remain limited without high-resolution approaches to dissect and characterise the immunopathogenic T-cells driving disease in immunologically heterogenous tissues, driving critical need to identify predictive risk factors for prevention. This was provided by seminal association in 2002 between HLA-B*57:01 and abacavir hypersensitivity syndrome (HSS), which developed in up to 8% of patients within the first 6 weeks of treatment, presenting as fever with rash and gastrointestinal and respiratory symptoms. Importantly, abacavir is a nucleoside reverse-transcriptase inhibitor used for treatment of HIV with favourable long-term toxicity, yet onset of HSS necessitated patient withdrawal from this efficacious treatment for replacement by less-efficacious regimens (6). The complete 100% negative predictive value (NPV) of this association with a 55% positive predictive value (PPV), demonstrated this allele was necessary albeit insufficient for disease to occur. This then led to the first double-blind randomised trial of its kind to demonstrate efficacy of abacavir pre-treatment HLA screening to improve drug safety by avoiding use in HLA-B*57:01 positive patients (7). Genome wide association studies (GWAS) have since identified other strong HLA associations for which a few are now similarly implementable in clinic specific to drug, reaction phenotype, and patient population with shared genetic ancestry. However, this has driven health disparities with risk discovery limited to only the most prevalent populations as minority populations lack the necessary large control cohorts. Moreover, most associations are of lower NPV and PPV, limiting viability of costly sequence-based HLA typing for clinical recommendation across national cohorts. However, low-cost single-allele screens are being developed, and HLA-imposed risk factors identified, which propose to increase PPV towards more predictive multi-allelic screening strategies. Moreover advanced immunogenic approaches are being implemented to predict genetic risk in a single patient, to characterise the single-cell effector signatures in tissue towards diagnostics and treatments and define HLA-presented moieties to inform structural risk for avoidance in future drug development. We provide an overview of the immunopathogenic models of DHR, providing background for current predisposing genetic risk factors, before elucidating the new immunogenomic concepts and technologies driving advanced prediction, prevention, and treatment strategies. Importantly, these are critically dependant on clinician-led curation of ethnicity-captured paired tissue samples and international biorepositories.

Mechanisms of drug-induced T-cell activation

There are three mechanisms by which small molecular weight drugs drive T-cell activation, with all dependent upon formation of a risk HLA-antigen complex with a corresponding T-cell receptor (TCR). The first described was the hapten concept whereby drugs must first bind self-protein to become large (>1000Da) non-self-neoepitopes capable of immune recognition (Figure 1,A, i). Antigen presenting cells (APC) then engulf and process the drug-hapten into peptides for HLA presentation as a mechanism best described for beta-lactam antibiotics (8). Next, as the antibiotic sulfamethoxazole can directly stimulate T-cells but is not protein-reactive, Pichler established the pharmacological interaction (PI) concept (9). Here, the drug antigen either binds directly to the HLA or TCR in the absence of APC processing (Figure 1, A, ii). The most recently described is the altered self-peptide repertoire model, which factors in the continuous presentation of tolerised self-peptides by HLA during steady state immunosurveillance (10, 11). In the context of HLA-B*57:01, abacavir binds in such way that the HLA antigen-binding cleft is changed, altering the repertoire of endogenously processed self-peptides that can be presented. These previously untolerised HLA-specific antigen complexes are identified as foreign, with a multitude of potential self-peptides driving polyclonal TCR activation opposed to the oligoclonal response observed with more structurally restricted hapten- or PI-induced activation (Figure 1, A, iii) (12). Importantly, functional assessment of how each drug activates T-cells can be performed in vitro, either using post-reaction patient-derived memory T-cells (13, 14), or through naïve T-cell priming platforms developed to assess immunological risk in healthy drug-unexposed donors (15–18). Intriguingly, use of such assays has raised two non-mutually exclusive models for the origins of drug-reactive T-cells; a more traditionally described route whereby drug antigens prime naïve T-cells (17, 19, 20), and the other centred around heterologous immunity whereby drug drives cross-reactive re-stimulation of memory T-cells primed to previously encountered antigen. For this model, although the initial priming antigen is undefined, abacavir-responsive T-cells are observed in all HLA-B*57:01 positive but drug-exposed donors implicating broad early-life exposure pathogen(s) such as herpes simplex virus (HSV). Yerly et al (21) reported reactivity of an HLA-B*57:01-restricted abacavir T-cell clone to an HSV1/2-dervied peptide sequence. This model is postulated as reason for the tissue-specific targeting of disease, as virally primed memory T-cells (Figure 1, B, i) migrate to and reside as CD103-expressing tissue resident memory T-cells (TRM) in the skin and mucous membranes until proposed cross reactive drug exposure (Figure 1, B, ii). However, this remains a developing hypothesis, and while the accumulation of activated TRM (CD45RO+CD69+CD103+) is recently reported in skin biopsies after intradermal drug challenge in patients with different DHR, similar characterisation during acute disease is now warranted to pinpoint the origins of response-driving populations (22–24).

Figure 1.

Models of drug-induced T-cell activation. (A)(i) Hapten, (ii) pharmacological interaction, and (iii) altered self-peptide repertoire models of how drugs activate T-cells. (B) Heterologous immunity. (i) Patient infected with pathogen which primes naïve T-cell to drive TRM development. (ii) Drug cross-reacts with viral-primed TRM. APC, antigen presenting cell; TCR, T-cell receptor.

Human leukocyte antigen and current state of risk implementation

Human leukocyte antigen loci are divided into classes I-III (Figure 2, A, i). While class III are non-classical and represent complement, tumour necrosis factor, and heat shock proteins, HLA class I encompasses HLA-A, HLA-B, and HLA-C genes with protein expressed on all nucleated cells to present endogenous peptides to cytotoxic T cells (Figure 2, A, ii) for physiological role in clearance of virus-infected cells. In contrast, HLA class II includes HLA-DP, HLA-DQ, and HLA-DR genes for which protein is more selectively expressed on APC for presentation of exogenous peptides (Figure 2, A, ii) to helper T-cells in response to extracellular infections. Since discovery association between HLA-B*57:01 and abacavir hypersensitivity, several other strong HLA associations with complete NPV are described and clinically implemented for ethnic populations of high allelic carriage (Figure 2, B). The association of HLA-B*15:02 with hypersensitivity to the antiepileptic drug carbamazepine and pre-emptive screening was firstly conducted in a couple of countries, especially in Taiwan. Taiwan, Hong Kong, Thailand, and Singapore now endorse HLA-B*15:02 pre-prescription screening (25–27). Moreover, prospective screening of HLA-B*13:01 is proven to reduce incidence of dapsone-hypersensitivity in Chinese (28) and Thai (29) populations, and screening for HLA-B*58:01 associated with hypersensitivity to the anti-gout drug allopurinol has seen selective implementation in Southeast Asia (30, 31). However, routine HLA testing is not yet similarly advised by the European Medicines Agency or US Food and Drug Administration (32). A recent evaluation of a Canadian populace showed that HLA-B*58:01 screening was underutilised in countries outside East Asia, with less than half of allopurinol-exposed patients of high-risk East Asian-ancestry tested (33) inciting need for greater clinical awareness. Importantly, it is now clear that risk alleles are not only specific to a drug, but also the ethnic population and life-threatening reaction phenotype. This is exemplified by association between HLA-A*31:01 and carbamazepine-induced DRESS but not SJS/TEN in both European and Chinese populations (34), and associations between HLA-B*15:02 and HLA-B*57:01 with carbamazepine-induced SJS/TEN in Chinese and European populations, respectively (35, 36). New risk alleles continue to be discovered, including strong associations for high-utility antibiotics. These include HLA-A*32:01 and vancomycin-DRESS in Europeans in 2019 (37), and HLA-B*13:01 and −B*15:02 with co-trimoxazole-induced severe cutaneous adverse reactions (SCAR) in Asian populations in 2020 and 2021 (38, 39). While a diversity of alleles are now associated with diverse drugs as extensively reviewed (40, 41), it is intriguing that the same risk alleles are shared by structural diverse antigens. This is epitomised by association of B*35:01 with liver injury caused by polygonum multiflorum in 2019 (42), green tea in 2020 (43) and co-trimoxazole in 2021 (44). Shared binding is of interest to identify alternate risk alleles for the same drug across ethnicities, as demonstrated with similar risk of HLA-C*04:01-induced hypersensitivity across HLA-C*04 variants with a shared F pocket motif, including alleles prevalent in Hispanic (C*05:01) and African (C*18:01) populations (45). However, few risk alleles outside those discussed above have translated to clinic. This is due to both an incomplete NPV and considerable diversity in allelic carriage between different ethnic populations (Figure 2, C), implicating other undefined HLA-risk alleles, and a low PPV, implicating undefined non-HLA genetic risk factors. In recent years this has driven two strategies; development of low-cost single HLA-allele screening assays and genomic investigation of HLA-imposed risk factors to reduce the number of patients needed to test to prevent one case (NNT).

Figure 2.

Human leukocyte antigen risk and drug hypersensitivity. (A) Schematic of HLA (i) locus organisation and (ii) peptide antigen loading pathways. (B) Key HLA class I risk alleles with implemented, potential, or pre-emptive screening assays. (C) The allele frequency (%) of key HLA risk alleles in representative ethnic populations. (D) Map showing location of drug hypersensitivity focussed clinical repositories and consortia. DRESS, drug reaction with eosinophilia and systemic symptoms; ER, endoplasmic reticulum; NNT, number needed to treat; NPV, negative predictive value; PPV, positive predictive value; SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis; TAP, tapasin.

Critical need 1. development of economically viable screening assays for clinical implementation

DNA sequence-based HLA typing is the ‘gold standard’, providing discrimination of most HLA class I alleles through focussed amplification of exons 2 and 3 at resolution capable of identifying novel allelic variants. However, with advances of next-generation sequencing (NGS), bioinformatic-algorithm informed computational tools have been developed to match NGS fragments to different HLA allele variants with high accuracy. These include HLA-HD, OptiType, HLAforest, HLAreporter, and PHLAT; capable of HLA prediction and imputation using either whole exome sequencing (WES) or whole genome sequencing (WGS) (46). As the cost of NGS is decreasing, prediction and imputation of HLA may serve as an alternative future method for HLA genotyping (47). Nonetheless, sequencing methods remain time-consuming, dependent upon specialist expertise and equipment, and costly, with HLA-A*31:01 typing reported to be 300 GBP (48). This is particularly true for alleles with incomplete NPV and PPV and with expression restricted to particular populations, complicating implementation in countries with a diverse multi-ethnic populace. However, the converse strategy of only testing patients from high-carriage populations will leave other patients at risk. For example, non-selective screening in US populations identified more than twice the ‘at-risk’ patients for HLA-B*15:02-restricted carbamazepine-hypersensitivity than screening of Asian patients alone (49). To facilitate broader risk screening, multigene panel tests have been developed encompassing high-risk alleles to provide extended risk genotype information. One such panel costing 50GBP and inclusive of HLA-A*31:01, HLA-B*15:02, HLA-B*57:01, HLA-B*58:01, HLA-B (158T), and HLA-DQB1 (126Q) is economically viable across alleles with 0.0069 increased quality-adjusted life year gain and total cost savings of $491 USD, but not for individual drugs including carbamazepine with HLA-B*15:02, and allopurinol with HLA-B*58:01 (50). Importantly, for these well-defined highly predictive alleles, real-time polymerase chain reaction (RT-PCR)-based assays have been developed including HLA-B*58:01 (51) or HLA-B*57:01/HLA-B*58:01 (52), and HLA-B*15:02 (53) or HLA-B*15:02/HLA-A*31:01 (54). Reverse transcription-PCR is both rapid and cheap, requiring only initial sequence specific design for highly sensitive target-allele amplification within a few hours, and is a highly established technique suited to most on-site clinical labs. Moreover, multiple genes can be targeted through multiplexed assays enabling simultaneous assessment of risk and house-keeping control genes to prevent false negatives (55). Thus, expansion of the single-allele screening tool armamentarium is required to provide similar strategies across drugs and populations and drive clinical validation studies. For example, a single-allele screening assay would provide cost-effective validation of recently described associations between life-threatening green tea-, polygonum multiflorum-, and co-trimoxazole-induced liver injury and HLA-B*35:01 (41–43). A clear example demonstrating how this need can be met is the recent development of an HLA-A*32:01 RT-PCR screen for risk of DRESS associated with vancomycin (36). Although not yet translating guidelines, the 100% sensitivity and specificity for HLA-A*32:01 detection with as little as 10ng of DNA (56) and short-turnaround may be appropriate to enable immediate treatment and post-prescription screening, with therapy altered in those patients subsequently identified as positive.

Critical need 2. non-HLA risk factors toward multi-allelic screening or predictive models of disease

While HLA screening provides an immediate avenue for DHR prevention, the incomplete PPV of all associations to date indicates other patient-specific risk factors that align with CD8+ driven immunopathogenic disease. These now span genetic and environmental risk, proposing advanced gene-engineered models of disease or multi-allelic screening.

GENETIC VARIATION IN THE HLA-PRESENTED IMMUNOPEPTIDOME

The only non-HLA risk factor recommended for preventative DHR screening remains cytochrome (CYP)2C9, an enzyme involved in metabolism of phenytoin (57). In particular, the *3 variant of CYP2C9 has a 90% reduced functional capacity, limiting clearance of the immunogenic parent drug. Although individuals with this profile are not prohibited from treatment, the 2020 Clinical Pharmacogenetics Implementation Consortium guideline is for a 25% or 50% reduction of starting dose in HLA-B risk-restricted patients defined as intermediate or poor CYP2C9 metabolisers, respectively (58). Importantly, for other drugs it may be that multiple enzymes play a role, with polymorphic variants of both N-acetyl transferase (NAT)2 (59, 60) and Glutathione S-transferase (GSTM1) (61) associated with but not replicated in high-risk patients for onset of sulfamethoxazole hypersensitivity. There remains limited replicated metabolising enzyme associations with DHR, implicating only a minor role in predisposing disease; a conclusion supported by similar quantification of drug-protein conjugates piperacillin-allergic and -tolerant patients (62). However, the structure of the HLA-presented immunogenic antigen remains undefined with proposed influence of self- and/or drug-derived peptides. Critically, recent GWAS indicate genetic impact of intracellular peptide processing pathways to diversify the HLA-loaded immunopeptidome. Briefly, in a sub-Saharan African cohort, Carr et al (63) reported minor association between HLA-C*04:01-restricted nevirapine-rash and a specific endoplasmic reticulum aminopeptidase (ERAP) allotype. ERAP is an enzyme epistatically associated with HLA and involved in processing peptides to the correct length for loading onto HLA class I molecules. Hypoactive variants are previously described in the context of viral infection, which undertrim peptides to form longer less immunogenic moieties and impair HLA-restricted clearance of hepatitis C infection (64). In support, a recent targeted analysis utilised sequence-based ERAP typing to report that risk HLA-B*57:01 expressing patients with hypoactive ERAP1 allotypes were more likely to tolerate abacavir (65). Although by itself HLA typing for abacavir hypersensitivity is highly successful, the initial finding with nevirapine suggests importance of such variation across drugs. Thus, studies are now warranted to improve PPV and NNT using ERAP, but also other highly polymorphic HLA-regulating loci such as killer-cell like immunoglobulin receptors (KIR) linked to HLA-imposed predisposition to other immune-mediated disease (66). Discovery of these multi-allelic associations will remain dependant on large allergic and tolerant patient cohorts but will improve risk prediction for currently defined but lowly predictive HLA risk alleles towards cost-effective and implementable screening strategies.

DYNAMIC REGULATION PROPOSES INTRA-INDIVIDUAL SUSCEPTIBILITY

Whereas stable interaction of HLA-antigen-TCR is the dependant signal for T-cell activation, a second signal provided by co-activating and co-inhibitory signalling pathways regulates the T-cell activation threshold, with the summative outcome skewing towards activation (allergy) or anergy (tolerance), respectively. Furthermore, blocking of co-inhibitory checkpoints including programmed death (PD)-1 and cytotoxic lymphocyte antigen (CTLA)4 has been shown to increase the drug-specific response of T-cells after naïve priming (15, 67), most recently with sulfasalazine (SLZ) (14). Further, blockade of these same pathways using immune checkpoint inhibitors (ICI) to bolster anti-tumour T-cell response in cancer patients has been associated with increased rates of hypersensitivity with subsequently administered drugs. Studies have reported liver injury in patients treated with ipilimumab and dacarbazine, severe rash with vemurafenib after treatment with nivolumab, and a higher proportion of patients developing SLZ-hypersensitivity following treatment with pembrolizumab (68, 69). Importantly, immune-mediated adverse effects during immune checkpoint therapy is well-documented, but often attributed to enhanced reactivity to self-antigen and autoimmunity, yet these studies demonstrate similarly heightened immunogenic risk with drugs (70, 71). Moreover, polymorphic variants of these regulatory receptors are associated with predisposition to T-cell mediated autoimmune disease (72, 73), but are not yet similarly reported for risk of DHR. Nonetheless, these studies raise the likelihood that these regulatory elements drive both genetic and therapeutic inter- and intra-individual DHR risk. Future multi-omic investigations will elucidate environmental risk pressures on cell regulation and drive increased clinical awareness of potentially increased drug immunogenicity during and after ICI treatment (74, 75).

Genetic variation in the HLA-risk restricted T-cell receptor repertoire

There is theoretical 2×10¹⁹ diversity of TCRαβ (76), albeit with a more restricted individual human repertoire estimated at 2×10⁷ (77). This variety is driven during thymic development and somatic rearrangement of the variable (V) and joining (J) or variable (V), joining (J), and constant (C) domains of respective α and β chain loci that form the HLA-antigen-interacting heterodimer TCR protein. Further diversity in the antigen-binding complimentary determining region (CDR)3 is driven by random nucleotide additions and deletions (78). Application of next-generation sequencing with TCR-targeted primers has enabled fine characterisation of exact TRBV, TRRBJ, and beta CDR3 sequence(s). This was utilised by Chung to show preferential private TCRβ chain usage in blister from HLA-B*58:01-restricted allopurinol SJS/TEN patients (79). In 2021, Villani et al (23) similarly showed that CD8+ T-cell effectors of patients with diverse drug-induced TEN had highly restricted TCR usage, with the top 5 clonotypes accounting for up to 90% of all detectable TRBV, indicating relevance of immunodominant TCR across drugs. However, the beta chain incompletely defines antigen specificity, which is additionally influenced by the corresponding alpha variable chain (80, 81). Single-cell (sc)TCR sequencing now captures both the alpha and beta sequence. This was utilised by Pan et al (82) for blisters of patients with HLA-B*15:02-restricted carbamazepine-induced SJS/TEN to find a TCRα CDR3 “VFDNTDKLI”/ TCRβ CDR3 “ASSLAGELF” clonotype shared and dominantly expanded across patients. Critically, this clonotype was not dominantly expanded in peripheral blood, highlighting need to assess affected tissue, and not similarly observed in risk HLA-expressing drug-tolerant patients, indicating disease dependence upon expression of both a risk HLA and TCR. Thus, although novel allelic associations within as yet undescribed risk-associated genes will likely remain discovered by GWAS, the field now recognises the use of single-cell multimodal omics (83) to detect dominant TCR. These data can be subsequently used as strategy to uncover novel HLA-risk associations in minority patient populations, novel biomarkers of disease for diagnosis and treatment, and for the immunogenic risk HLA-TCR-presented structures driving disease, as discussed subsequently.

Critical need 3: a precision, single-patient-informed risk prediction strategy using ethnicity-recorded biorepositories

One major advantage of single-cell sequencing is provision of the full sequence of a TCRαβ, supplying opportunity for synthetic expression to test HLA- and antigen-restriction in vitro. Briefly, Pan et al (82) linked TCRα/TCRβ expression constructs of the interest expanded TCR clonotype to a mouse constant domain and transfected into a murine 5KC hybridoma. TCR-transfected cells were then co-cultured with HLA class I-deficient lymphoblastoid C1R APC or those expressing HLA-B*15:02 and drug for 48 hours, before assessment of activation via cytokine secretion. They confirmed that the clonotype of interest was HLA-B*15:02- and carbamazepine-restricted. Although HLA-B*15:02 is already a well-established risk for carbamazepine hypersensitivity, this study provides important insight for how scTCR-seq applied to patient samples of minority ethnicities could both propose and validate HLA risk in a single patient. Upon observing previously undefined HLA allele-restricted drug-induced response, this personalised immunogenomic approach could translate to inform risk more generally across the ethnicity-matched population. Appropriate ethnicity-matched samples can be selected from the growing number of diverse international registries for DHR pairing ethnicity-recorded patient metadata with clinical samples (DNA, RNA, blood, tissue) outlined in Figure 2, D. The benefit of these clinical repositories is epitomised by identification of the first HLA risk allele relevant to DHR-prevention in diverse minority Australian Indigenous populations (84). Briefly, Samogyi identified three indigenous Australian patients with phenytoin-induced DRESS who shared expression of HLA-B*56:02. This allele has up 19% expression in these populations but 0% in Europeans (84), highlighting diverse pharmacogenetic variation and a pathway to determine risk in understudied populations to mitigate knowledge and health disparities. While this study proposed HLA-B*56:02-restricted risk, single-cell sequencing on acute tissue will identify and enable synthetic expression of the dominant TCR for testing against autologous HLA-expressing APC. Such approach will confirm and validate this risk allele for extension across these minority populations.

Critical need 4: single-cell sequencing-informed tissue-relevant effector biomarkers for diagnostics and targeted treatments

There remains no consensus treatment guideline during acute management of SJS/TEN due to a lack of clinical trials to support therapeutic intervention. In recent years, cyclosporine has demonstrated some observational success to reduce mortality (85), and use of etanercept improves time to re-epithelisation by reducing concentrations of the major cytotoxic mediator, granulysin (86), yet the effect on mortality was not statistically significant. This highlights the need to discover novel biomarkers to develop the first targeted treatments. However, this has been limited by a lack of appropriate methods for high-resolution and unbiased characterisation of immunologically heterogenous populations within diseased tissue and the signature of clonally expanded effector T-cells. Compared with bulk-sequencing, single-cell methods provide this resolution (Figure 3, A). The advent of droplet-based microfluidics devices have enabled a fully integrated multi-focal single-cell pipeline for RNA-TCR-’cellular indexing of transcriptome and epitopes sequencing (87), for unbiased transcriptome, functional TCR, and surface protein expression of each cell (Figure 3, B). Analysis of these ‘big-data’ sets requires initial bioinformatics for unsupervised clustering of cells by similar transcriptomic signature, with subsequent alignment to reference databases providing initial inference of cell type. However, these are becoming increasingly standardised practices with frequent release of new tools to aid users in downstream analyses (88–91). To date, single-cell application to investigate DHR remains limited, but includes comparative scRNA-TCR-seq study of abacavir positive patch test with drug-unexposed skin (11), and to identify novel treatment strategies in a patient with standard therapy-refractory DRESS. In the latter 2020 study, Kim et al (92) compared the transcriptomic profile of single-cells in skin and blood samples from this patient to identify upregulation of the JAK-STAT pathway. This led the clinical team who had exhausted all classical treatment options to administer the JAK1 and 3-inhibitor Tofacitinib for effective disease control. This is the first demonstrated example of single-cell sequencing to identify an unconventional but effective treatment for SCAR during active uncontrolled disease. Most recently, Hertzman et al (93) compared punch biopsy from unaffected and positive delayed intradermal test skin to understand the T-cell immunopathogenesis of rare corticosteroid-induced systemic contact dermatitis. These initial studies show the clinical utility of single-cell sequencing for immunogenomic elucidation of cellular immunopathogenesis at unparalleled resolution. Additionally integrated assays provide future scope to use the dominantly expanded TCR to define drug-reactive effector signatures, all cell and gene interactions within the broader cell atlas, and lineage tracking to inform dynamic response (Figure 3, C). These pipelines are now paired with transposase-accessible chromatin with sequencing (94, 95) to assess epigenetic modifications, and spatial transcriptomic methods for localisation of interest signatures in histological tissue sections (Figure 3, D) (96). Analyses of the as yet undefined single-cell signatures of clonally expanded effector populations will now be required across drug-induced reactions to provide novel reaction-specific biomarkers towards diagnostics and targeted treatments.

Figure 3.

Single-cell sequencing to explore immunopathogenesis. Schematic of (A) the resolution provided by single-cell compared to bulk sequencing methods, (B) single-cell RNA (scRNA)-TCR-CITE-sequencing methodology, (C) utility of single-cell data to explore dominant TCR, identify the clonally expanded effector population, differential gene expression signatures, and network and lineage analysis, and (D) spatial transcriptomics. ADT, antibody-derived tag; cDNA, complementary DNA; mRNA, messenger ribonucleic acid; UMI, unique molecular identifier.

Critical need 5: strategies to define and screen HLA-risk presented immunogenic structures

Patient T-cells can be stimulated through in vitro co-culture with culprit drugs or their synthetically stable metabolites, most recently tolvaptan (12), dapsone (13), vancomycin (97), and clozapine (98). Moreover, testing HLA-restricted patient T-cells against comparative drugs or structural analogues can elucidate cross-reactivities to ensure safe ongoing treatment (97,99,100). However, the actual HLA-presented immunogenic structures driving T-cell activation remain undefined with potential influence of diverse drug-derived- or self-peptides and diverse TCR. Understanding the antigenic structure is now possible through immunogenomic interpretation of risk HLA and TCR, with several approaches recently implemented to understand these structures. Initial mass spectrometry-focussed investigations eluted peptides from HLA-B*57:01 with and without abacavir treatment, demonstrating an altered self-peptide repertoire as a new model of T-cell activation (9, 10). Abacavir not only modifies the peptide repertoire expressed at the cell surface, but alters the intracellular tapasin-mediated loading of peptide complexes (101). With correlation between immunogenicity and stable HLA-peptide complex formation, methods have been recently developed to generate immunopeptidome thermostability curves capable of training artificial network models to predict the most immunogenic peptides (102). Human leukocyte antigenA elution methods have recently been applied to similarly characterise naturally processed flucloxacillin-modified peptides (103, 104).Because as HLA presentation alone does not drive T-cell activation, these position 4-restricted flucloxacillin-modified peptides were then used to immunise risk-associated HLA-B*57:01-expressing transgenic mice. The authors then demonstrated T-cell immunogenicity by monitoring IFN-γ secretion from subsequently extracted and antigen re-exposed splenocytes (104). However, the protein origins of such hapten complexes remain unknown, and the in vitro assays described are limited by the restricted protein repertoire of in vitro co-cultures or focus on high-abundance model protein. To navigate this issue the field now requires strategies for high-throughput peptidome library screening approaches incorporating both risk HLA-expressing APC and patient-blister inferred immunodominant TCR to identify the full array of antigenic sequences driving T-cell activation (105).

Conclusion

From a current clinical perspective (Figure 4, A), risk HLA typing is implementable for select high-risk drugs in high-risk populations to predict and prevent onset of DHR. However, the cost of sequence-based typing and a low PPV continues to hinder clinical implementation for most alleles and drugs, with the only screening-recommended non-HLA risk factor being low function CYP2C9 variants before treatment with phenytoin. To enable further implementation for HLA allele screening, an extended armamentarium of low-cost PCR-based screens are needed with advantage of having a short turnaround time suited to equipment and expertise available in most on-site clinical laboratories (Figure 4, B, need 1). Moreover, recent GWAS studies have highlighted HLA-imposed risk factors including ERAP, which now must be similarly assessed across drugs for potential to increase PPV and reduce NNT towards cost effective multi-allelic screening (Figure 4, B, need 2). However, one of the most significant discoveries of the last few years are immunodominant risk HLA- and drug-antigen-restricted TCR. These provide novel immunogenomic avenues to answer key remaining critical questions, but with discovery dependent upon clinician-led curation of disease-relevant biorepositories pairing clinical metadata, patient ethnicity, HLA-type, and relevant tissue samples from acute disease. Detailed clinical reporting to capture HLA-type and patient ethnicity, and submission of tissue-relevant acute clinical samples to these global biorepositories must now be considered part of the practical guide to treatment. This in turn will provide advanced immunogenomic data-driven diagnosis and treatment strategies for the patient at hand and enable future application to deliver population-personalised prediction and prevention strategies described. In their most accessible application, these biorepositories enable patient clustering to identify shared alleles as demonstrated by Samogyi et al (84) to propose risk of HLA-B*56:02 for development of Phenytoin-DRESS in Indigenous Australian populations. In their most complex application, state-of-the-art immunogenomics approaches combine single-cell sequencing and in vitro co-culture platforms to first identify and functionally validate HLA-restricted drug-TCR-risk in tissue from a single patient, with extended relevance across the ethnicity-matched population (Figure 4, B, need 3). Thereafter, the tissue-relevant signatures of clonally expanded effector T-cell populations provide biomarkers for diagnosis and targeted treatments (Figure 4, B, need 4), and the HLA-TCR-presented immunogenic drug- and/or self-derived moiety pathways for future structural risk prevention (Figure 4, B, need 5).

Figure 4.

Immunogenomics approaches. (A) Current. Large cohorts for HLA risk discovery by GWAS. (B) New. (Need 1) Low-cost PCR-based HLA screens. (Need 2) Multi-allelic screening. (Need 3) TCR- HLA-engineered co-culture systems to propose and confirm HLA risk. (Need 4) Single-cell sequencing identification of clonally-expanded effector population signatures. (Need 5) Mass spec risk HLA-elution to identify risk HLA-presented immunogenic antigen(s). ERAP, endoplasmic reticulum aminopeptidase.

Abbreviations used:

ADT

Antibody-derived tag

APC

Antigen presenting cell

cDNA

Complementary DNA

DRESS

Drug reaction with eosinophilia and systemic symptoms

ER

Endoplasmic reticulum

mRNA

Messenger ribonucleic acid

scRNA

Single-cell RNA

SJS/TEN

Stevens-Johnson Syndrome/Toxic epidermal necrolysis

TAP

Tapasin

TRM

Tissue -resident memory T cell

UMI

Unique molecular identifier