Updates on the Immunopathology and Genomics of Severe Cutaneous Adverse Drug Reactions

Abstract

Severe cutaneous adverse reactions (SCAR) such as Stevens-Johnson syndrome, toxic epidermal necrolysis (SJS/TEN) and drug reaction with eosinophilia and systemic symptoms/drug-induced hypersensitivity syndrome (DRESS/DIHS) cause significant morbidity and mortality and impede new drug development. HLA class I associations with SJS/TEN and DRESS/DIHS have aided preventive efforts and provided insights into immunopathogenesis. In SJS/TEN, HLA-class I restricted oligoclonal CD8+ T-cell responses occur at the tissue level. However, specific HLA risk allele(s) and antigens driving this response have not been identified for most drugs. HLA risk alleles also have incomplete positive and negative predictive values, making truly comprehensive screening currently challenging. Although, there have been key paradigm shifts in drug hypersensitivity knowledge, there still remain many open and unanswered questions about SCAR immunopathogenesis as well as genetic and environmental risk. In addition to understanding the cellular and molecular basis of SCAR at a single-cell level, identification of the MHC-restricted drug reactive self or viral peptides driving the hypersensitivity reaction will also be critical to advancing pre-marketing strategies to predict risk at an individual and drug level. This will also enable identification of biological markers for earlier diagnosis and accurate prognosis as well as drug causality and targeted therapeutics.

INTRODUCTION

Severe cutaneous adverse reactions (SCAR) are life-threatening, T-cell mediated, delayed drug hypersensitivity reactions (DHR). In clinical practice, there are three main clinical reaction SCAR phenotypes (Figure 1). These include acute generalized exanthematous pustulosis (AGEP), drug reaction with eosinophilia and systemic symptoms (DRESS)/drug-induced hypersensitivity reaction (DIHS), and Stevens-Johnson syndrome and toxic epidermal necrolysis (SJS/TEN). While these reactions are distinguishable by clinical and immunopathological features, all are thought to involve human leukocyte antigen (HLA) presentation of drugs and/or metabolite altered peptides to T-cell receptors, resulting in a robust immune response.^{1, 2} SCAR can therefore differ in terms of the specific effector T-cells involved, and the resultant chemokines/cytokines produced, leading to cell homing to the skin and tropism for specific target tissues.

Figure 1.

Clinical presentation and immunopathology of SCAR. AGEP, DRESS/DIHS and SJS/TEN are the reaction phenotypes that comprise SCAR. These reactions vary in mortality, immunopathological features, and clinical characteristics. Adapted from Hama et al. 2022¹. AGEP: CD4+ T-cells secrete IL4, IL5, IL13, IFNγ, TNFα, IL8, IL17, and IL22 (Th17). IL8 drives neutrophil and T-cell recruitment to the epidermis to form sterile pustules. DRESS/DIHS: CD4+ and CD8+ T-cells, plasma DCs, and monocytes are enriched in the dermis. DCs produce CCL17 to recruit CCR4+ Th2 T-cells. Th2 cells and ILC2 produce IL5 to induce activation and migration of eosinophils that drive inflammation. TNFα, IFNγ (Th1), IL-4, IL-5, and IL-13 (Th2) are observed with HHV reactivation and Tregs in lesional skin. SJS/TEN: Cytotoxic CD8+ T-cells accumulate in blisters and release perforin, granzyme B, and granulysin to kill keratinocytes. N, neutrophil; APC, antigen-presenting cell; Th, T helper; DC, dendritic cell; IL, interleukin; TNFα, tumour necrosis factor alpha; IFNγ, interferon gamma; Treg, T regulatory cell; E, eosinophil; TRM, tissue-resident memory T-cell; CCR, chemokine receptor; CCL, C-C motif chemokine ligand; ILC, innate lymphoid cell; NKT, natural killer T-cell; NK, natural killer cell; GM-CSF, granulocyte macrophage colony-stimulating factor; HHV, Human Herpesvirus; CMV, Cytomegalovirus; EBV, Epstein-Barr virus.*5–10% of SJS/TEN-cases with more than 10% skin detachment do not have mucosal erosions.

created using Biorender.com

Over the last two decades, significant progress has been made in terms of understanding genetic risk and immunopathogenesis of SCAR. Strong associations with HLA-class I risk alleles have been described for many drugs causing SJS/TEN and DRESS/DIHS (but not yet for AGEP). Herein, we review the recent progress made in our understanding of the immunopathogenesis and genetic risk for SCAR that has led to greater insights into pathways for prediction and prevention, earlier diagnosis, prognosis, and more targeted and effective therapeutics. In the future, improved understanding of SCAR utilizing single-cell and high-throughput studies at a functional level will improve cost-effective strategies to avoid the morbidity and mortality associated with these reactions and improve the safety of drugs through prevention, early, and more precise drug-specific diagnosis and targeted therapies.

CLINICAL APPROACHES TO SCAR

The diagnosis of specific SCAR, and differentiating from SCAR mimickers, is based on specific clinical symptoms and histopathological features of skin biopsy (Figure 1), as well as different timelines to onset of the reaction (Figure 2).³ Importantly, similar parameters may also help to dissect reaction severity, with DRESS/DIHS with shorter latency recently reported and associated with less severe disease and lower incidence of facial edema.^4–6 In addition, algorithms have been established aiding clinical diagnosis and prognosis of SCAR, including the European Registry of (Regi)SCAR consortium, diagnostic criteria for DRESS and the severity-of-illness score for TEN (SCORTEN). For SJS/TEN, the algorithm of drug causality assessment for epidermal necrolysis (ALDEN) is used to help identify the culprit drug.⁷ However, algorithms directed at small molecules associated with SCAR remain incomplete, with the predictivity of SCORTEN uncontrolled for patient-specific factors including comorbidities and standard of care. Several recent adaptations have been proposed to increase accuracy.⁸ For example, the ABCD-10⁹ model uses a 5-item scoring system based on age, bicarbonate level, cancer, dialysis, and 10% involved body surface area. However, while ABCD-10 accurately predicts mortality, performance was not significantly different from the SCORTEN scale.⁹

Figure 2. Clinical timeline of SCAR.

Timelines showing the time from initiation of a specific drug to typical onset of symptoms and signs of a specific SCAR (latency period). The latency period differs not only between different SCAR reactions but also between SCAR reactions and benign exanthems. Adapted from Hama et al. 2022¹. DHR, delayed drug hypersensitivity reactions; SCAR, severe cutaneous adverse reaction; AGEP, acute generalized exanthematous pustulosis; DRESS, drug reaction with eosinophilia and systemic symptoms; DIHS, drug-induced hypersensitivity syndrome; SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis; MDE, morbilliform drug eruption; FDE, fixed drug eruption; SSLR, serum sickness like reactions.

Critical to evaluation of SCAR patients is the risk-benefit assessment of the potential implicated drugs and advice on future drug use that will impact safety.¹⁰ Following disease resolution, patch and/or intradermal testing can be used as adjunctive evidence to help identify a culprit drug and help risk-stratify patients for future therapy. While these in vivo allergy testing approaches have high drug specificity, they lack sensitivity, with ranges of 58–64% for AGEP, 32–80% for DRESS/DIHS, and 9–24% for SJS/TEN.^11,12 This decreased negative predictive value (NPV) means that patients with SCAR cannot be re-challenged to drugs based solely on negative patch or intradermal testing results. Further, ex vivo and in vitro tools including lymphocyte transformation tests and ImmunoSpot assays may be performed on peripheral blood mononuclear cells (PBMC) to detect drug-induced proliferation or cytokine secretion from T cells, respectively. However, these are non-standard techniques restricted to specialized research centers. Like in vivo approaches, these assays differ in terms of sensitivity across phenotypes with better performance for DRESS/DIHS than SJS/TEN. In addition, specificity can be incomplete and false positives have been reported. In many cases, drug-specific T-cells are only detected in blood several weeks after resolution. The sensitivity of ex vivo/in vitro testing is particularly poor for SJS/TEN from peripheral blood samples. Awad, et al. demonstrated a higher sensitivity for ELISpot to detect drug-specific response in blister fluid compared to PBMC.¹³ These data highlight the need to understand what is happening at a tissue level in addition to exploring different biological markers and cytokine outputs when developing ex vivo and in vitro assays.

Novel innovative approaches that may augment clinical and pathological diagnosis include using artificial intelligence (AI) tools such as deep neural networks. Optimistically, these AI tools will improve diagnostic capability. In one study, machine-learning and pattern recognition algorithms trained using a catalogue of photos of early reaction timepoints were used to differentially diagnose SCAR.¹⁴

There is no diagnostic approach with 100% NPV, and therefore integrated approaches are necessary. Merging clinical, in vivo/ex vivo/in vitro testing, and genetic markers that may be more specific for a drug-SCAR combination will improve diagnosis and help risk-stratify patients, particularly those who develop SCAR while on multiple drugs.

Although re-challenge is the gold standard to base the performance of other methods of diagnosis, this is unacceptable for most SCAR cases. The notable exception where evidence has accumulated is the management of tuberculosis (TB) and TB/HIV co-infected patients in resource-poor setting where the use of first-line treatments are essential to prevent TB-related mortality. Under these circumstances, sequential re-challenges of first-line anti-TB agents with methylprednisolone rescue upon the first sign of a drug reaction, have helped define the sensitivity and specificity of other diagnostic approaches, and have highlighted the low sensitivity of ex vivo approaches for drugs other than rifampin and the need to adjust diagnostic thresholds.^15,16

THE ADAPTIVE IMMUNE-RESPONSE, IMMUNOPATHOGENESIS OF SCAR AND MODELS OF DRUG-INDUCED T-CELL ACTIVATION

The 3.6Mb HLA locus is the most polymorphic in the human genome, with >35,200 alleles described as of October 2022; this staggering variation, and the fact that HLA is a critical driver of T-cell activation, has made it a research focus of inter-individual immune susceptibility.¹⁷ Classical HLA-class I and II alleles are co-dominantly inherited, resulting in expression of one set each of HLA-class I and II alleles from each parent (Figure 3A). These classical HLA have co-evolved under the presence of infectious diseases.¹⁸ Specifically, HLA-class II present peptides from extracellular antigens (helminthic parasites) to CD4+ T-cells, while HLA-class I present peptides from intracellular antigens (viruses) to CD8+ T-cells (Figure 3Bi). Stable interaction of HLA, peptide, and corresponding T-cell receptor (TCR) is known as signal 1, or the immunological synapse. This synapse is required for T-cell signaling, clonal expansion, and effector response (Figure 3Bii). Importantly, HLA polymorphism is concentrated in the peptide-binding domain, providing different alleles with specificity for a varying set of peptides under evolutionary pressure. Indeed, HLA diversity evolved under the pressure of human migration and geographical variance in infectious diseases.¹⁸ Thus, HLA alleles evolved to have differing prevalence in diverse global populations.¹⁹ This has resulted in differential susceptibility to experiments of nature such as infection, but also experiments of humankind such as severe drug-induced T-cell mediated SCAR.

Figure 3. Inheritance of HLA alleles and their role in SCAR.

(A) HLA alleles are co-dominantly inherited. (Bi) Antigen processing varies between HLA-Class I (left) and II (right). HLA-class I: Intracellular antigens undergo proteasomal processing, with peptides loaded into the ER for loading on HLA-class I and presentation to CD8+ T-cells. HLA-class II: Extracellular antigens are phagocytosed for antigen processing in endosomes, which fuse with lysosomes from the ER carrying HLA class II for loading and surface presentation to CD4+ T-cells. (Bii) HLA-Class I presentation results in cytotoxicity. Briefly, antigen engages the TCR to enable phosphorylation of CD3 by Lck for binding of ZAP70, Grb2, and PLCγ1. This activates calcium, RAS, MAPK, and PI3K signalling and transcription factors for cell activation and release of cytotoxic granules. HLA, human leukocyte antigen; APC, antigen-presenting cell; Tel, telomeric; Cen, centromeric; ER, endoplasmic reticulum; TAP, tapasin; TCR, T-cell receptor; Lck, lymphocyte-specific protein tyrosine kinase; PI3K, Phosphatidylinositol 3 kinase; P, phosphorylated; ZAP70, Zeta-chain-associated protein kinase 70; LAT, linker for activation of T cells; PLC, phospholipase; Grb, Growth factor receptor-bound protein; SOS, Son of sevenless protein; Ca2+, calcium; RAS, ‘Rat sarcoma virus’ protein; MAPK, mitogen-activated protein kinase.

Figure created using Biorender.com

Models of HLA-restricted drug-induced T-cell activation and tolerance

Several different models have been proposed for how low-molecular weight drugs activate T-cells. The hapten model was developed during work on chemical sensitization in 1935²⁰ and remains largely unchanged over 87 years. This model states that low-molecular weight drugs are too small for immune recognition, and must first covalently or irreversibly bind to self-protein to become large neo-antigenic structures. The drug (or ‘hapten’) modified by protein can then be recognized by antigen-presenting cells (APC) for uptake, processing, and HLA presentation to corresponding TCRs on passing T-cells (Figure 4Ai). This model is most commonly associated with beta-lactam antibiotics.²¹ However, some drugs that cause DHR are chemically inert, and in vitro models have suggested that the parent drug itself may directly interact non-covalently or reversibly with an immune receptor (HLA or TCR). The pharmacological interaction (p-i) model (Figure 4Aii) has been proposed for sulfamethoxazole and oxypurinol, the active metabolite of allopurinol.^22,23 More recently, in relation to HLA-B*57:01 and abacavir, another model whereby the small molecule drug interacts non-covalently in a dose-dependent manner with HLA and self-peptide has been described.^24,25 During homeostatic immunosurveillance, endogenously-processed self-peptides associated with HLA that have previously been seen and tolerated are associated with T-cell anergy. In the altered peptide repertoire model, drug antigen may non-covalently bind to HLA, altering the shape and specificity of the peptide-binding cleft and consequently the repertoire of self-peptides presented (Figure 4Aiii). In this model, it is these previously unseen self-peptide-HLA complexes that are observed as foreign, and elicit HLA-class I restricted, T-cell mediated drug hypersensitivity responses. Peptide binding and elution studies, and the crystal structure of abacavir bound to peptide and HLA-B*57:01, support this model of abacavir hypersensitivity. To date, this model has only been proposed for HLA-B*57:01 restricted abacavir hypersensitivity. Furthermore, an HLA-B*57:01 transgenic mouse model supports that CD4+ T-cells actively suppress maturation of dendritic cells and mediate tolerance to the altered peptide repertoire induced by abacavir.^26,27 However, as only 55% of HLA-B*57:01 carriers will develop hypersensitivity, additional mechanisms likely contribute to pathogenesis, including peptide polymorphisms, genetic variation in peptide processing through endoplasmic reticulum aminopeptidases (ERAP), or variation in another part of the adaptive immune response. ^27,64

Figure 4. HLA-restricted models of drug-induced T-cell activation.

(A) Models of HLA-restricted T-cell activation include the (i) hapten, (ii) pharmacological interaction (PI), and (iii) altered self-peptide repertoire models. Hapten: Drug-antigen binds self-protein before intracellular processing forms drug-modified peptides, which bind irreversibly and covalently to risk HLA. PI: Drug-antigen binds directly, reversibly, and non-covalently to risk HLA or TCR without a need for antigen uptake and processing. Altered self-peptide repertoire: Drug-antigen binds to risk HLA in such way that it alters the repertoire of self-peptides that may bind, which are subsequently seen as immunogenic. (B) Models for the HLA-restricted origins of drug-reactive T-cells in SCAR patient skin include (i) naïve priming and (ii) the heterologous immunity model. Naïve priming: Drug antigen primes the naïve T-cell which generates long-lived TRM, which are re-activated upon subsequent exposure to the same drug. Heterologous immunity: The patient is first infected with a virus, from which peptides are presented by the risk HLA to prime naïve T-cells. In later life, exposure of the same patient to a drug-antigen which shares structural cross-reactivity and HLA-restriction drives re-stimulation of the viral-primed TRM and SCAR. APC, antigen-presenting cell; HLA, human leukocyte antigen; TCR, T-cell receptor; TRM, tissue-resident memory T-cell.

Figure created using Biorender.com

Origins of pathogenic T-cells and heterologous immune models of SCAR

In vitro studies demonstrate that naïve T-cells from drug-inexperienced donors can be primed to different drugs, including vancomycin and sulfasalazine.^28,29 However, the latency period between first introduction of a drug and onset of symptoms differs widely, from hours to a few days in cases of SJS/TEN, and abacavir hypersensitivity, and a few days in AGEP, to weeks for DRESS/DIHS (Figure 2). In addition, evidence suggests that many T-cell mediated reactions induce life-long immunity, yet it is unclear what maintains an immune response to a drug-altered antigen in the absence of ongoing exposure. The propensity for some drug reactions, like abacavir hypersensitivity, SJS/TEN, and AGEP to have short latency periods while maintaining long-lasting memory T-cell responses suggests an ability to bypass the sensitization or priming phase (Figure 4Bi). It also suggests that cross-reactive memory T-cells could be important in those exposed to drug for the first time (Figure 4Bii).³⁰ Abacavir-responsive T-cells were detected in HLA-B*57:01 positive abacavir-naïve donors, suggesting the drug could re-stimulate an existing memory T-cell population primed to an earlier, but unknown and cross-reactive, viral antigen restricted by the same HLA-class I allele. This is known as the heterologous immunity model of drug hypersensitivity which is non-mutually exclusive to other models of T-cell activation (Figure 4Bii).³¹

Characterization of T-cells at the site of tissue damage

A public TCRαβ was found to be expressed on CD8+ T-cells in blister fluids of unrelated patients with carbamazepine (CBZ)-induced SJS/TEN from Europe and Asia.³² The functional relevance of this public TCR was supported in mice administered CBZ where the dominant TCR expressing T-cells were adoptively transferred with reproduction of a skin phenotype.³² To date, the public TCR clonotype defined for CBZ-SJS/TEN appears to be the exception, as other small molecules associated with SJS/TEN, such as oxypurinol, have been associated with dominant, private TCRαβ clonotypes expressed on CD8+ T-cells in SJS/TEN blister fluid. Studies at the site of tissue damage have increased interest in the role of skin resident T-cells (TRM) in SJS/TEN and DRESS/DIHS immunopathogenesis.^33,34 Recent literature highlights that CD8+ T-cells with either a TRM or effector phenotype (CD103+, CCR7-, CD57-) are prominent in SJS/TEN blister fluid.³⁵ In addition, TRM may persist in the skin after SCAR and may play a role in the local, delayed response seen at the site of a positive drug patch or intradermal test for DRESS/DIHS.³⁶ Single-cell studies examining the cellular and molecular signatures at the site of drug-induced tissue damage have generated further insights into the immunopathogenesis of SCAR and other T-cell mediated hypersensitivities and may pave a pathway to precision medicine approaches for prevention, earlier diagnosis and targeted treatment.³⁷

SCAR ASSOCIATED WITH IMMUNOMODULATORY DRUGS

Antibodies for co-inhibitory programmed death (PD)-1 and cytotoxic lymphocyte antigen (CTLA)-4 receptors increase drug-specific responses after naïve T-cell priming.³⁸ Pharmacological blockade of these signals is now a routine and efficacious treatment for cancer, with immune checkpoint inhibitors (ICI) used to re-activate tumor antigen-specific T-cells where exhaustion due to prolonged TCR stimulation is mediated by up-regulation of co-inhibitory receptors. However, the same receptors are expressed on all antigen-stimulated T-cells, including those responding to infection. Thus, ICI are not specific to anti-tumor T-cells but are present on all T-cell clonotypes.³⁹ As a result, adverse events including SCAR associated with ICI and other immunomodulatory drugs are well-documented.^40–42 Additionally, a number of clinical reports now note increased DHR to dacarbazine, vemurafenib and sulfasalazine in patients previously-exposed to ICI, suggesting ongoing therapeutic dysregulation to low-molecular weight drugs potentially creating an immune environment permissive to T-cell mediated hypersensitivity.^43–44 Indeed, sulphapyridine-reactive T-cells were recently characterized from the blood of ICI- and sulfasalazine-treated hypersensitive patients.²⁹ It is currently unknown whether this occurs in an HLA-class I restricted permissive environment or, as suspected in many ICI reactions, an environment is created allowing this to occur in the absence of an HLA-class I risk allele.

IMMUNOGENOMICS OF SCAR

Discovery of strong HLA-class I associations and SCAR

The discovery of a strong association between HLA-B*57:01 and abacavir hypersensitivity syndrome over 20 years ago and the subsequent licensing trial and successful implementation into routine HIV clinical practice as a pre-treatment screening test has created a roadmap for clinical translation.^46,47 With a 100% NPV and 55% positive predictive value (PPV) and a prevalence of HLA-B*57:01 in 4–10% of the developed world, only 13 patients would need to be tested to prevent one case of clinically diagnosed abacavir hypersensitivity.⁴⁷ The development of cost-effective, quality-assured, single allele assays for HLA-B*57:01 therefore facilitated the ease of implementation into routine clinical practice worldwide and has resulted in the disappearance of abacavir hypersensitivity as a clinical entity.⁴⁸

Over the last two decades, there have been several other HLA associations described with T-cell-mediated reactions including SJS/TEN, DRESS/DIHS, and drug-induced liver injury (Suppl. Table 1). For AGEP, a clear HLA association with the common drugs implicated has not been investigated yet. For other SCAR, like SJS/TEN and DRESS/DIHS, HLA associations have been almost exclusively HLA-class I and have been largely specific to both the drug and the clinical phenotype of the reaction. However, there are still many examples where both SJS/TEN, DRESS/DIHS and even single organ involvement such as drug-induced liver injury (DILI) are associated with the same HLA allele. This includes HLA-B*58:01 and allopurinol SCAR and HLA-B*13:01 and dapsone SCAR (Suppl. Table 1).

Strong associations continue to be discovered using traditional DNA-sequence-based HLA typing. However, new tools and discovery platforms for HLA and other genetic associations, including RNA-seq, whole genome sequencing (WGS), whole exome sequencing (WES), and array-based typing with validated imputation of HLA, have emerged.⁴⁹ Although more costly, these new platforms provide more diverse genetic information with bioinformatic algorithms designed to detect HLA variants, other genes in linkage disequilibrium, and other epistatically linked genetic variants.^50,51 WES-imputed HLA was recently applied by the Taiwan (T)SCAR consortium to expand the study of HLA-B*13:01 risk with trimethoprim-sulfamethoxazole-induced SCAR across Asian populations following initial association in a Thai cohort in 2020.^52,53 Other recent updates of HLA-SCAR associations include a similarly expanded risk of HLA-B*13:01 for dapsone-SCAR across Asian populations, and novel associations of HLA-B*57:01 with CBZ-SJS/TEN and HLA-A*32:01 with vancomycin-DRESS in European populations.^54–58 These associations have been supported by in vitro and functional studies on patient-derived T-cells.^59–61 In silico modeling recently demonstrated the structural cross-reactivity of HLA-B*13:01 with both dapsone and sulfamethoxazole and γ-interferon ELISpot assays showed a small but detectable risk of cross-reactivity for HLA-A*32:01 carriers with vancomycin-DRESS and other glycopeptide antibiotics.^52,60

Challenges and Opportunities for implementation

The imperfect PPV of most HLA associations with T-cell mediated drug hypersensitivity reactions suggests that HLA is necessary, but not sufficient, to predict hypersensitivity risk. The number needed to test for a specific HLA allele to prevent one case of drug hypersensitivity is dependent not only on the PPV of the HLA risk allele, but also the carriage rate of the risk allele and prevalence of the phenotype in the population. The PPV varies between SCAR and hypersensitivity phenotypes and can be as high as 55% for HLA-B*57:01 in abacavir hypersensitivity, 20% for HLA-A*32:01 in vancomycin-DRESS and 3% for HLA-B*58:01 in allopurinol SCAR (Figure 5). The incomplete PPV of most HLA associations with DHR indicates that other genetic and ecological factors aside from HLA may be important. Previous studies identified genes epistatically linked to HLA, including ERAP, which is involved in the N-terminal trimming of peptides, as well as genes involved in drug metabolism, immunosuppression and co-morbidities.^62,63 In 2020, Pavlos et al. demonstrated a protective association between hypoactive allotypes of ERAP and abacavir tolerance in HLA-B*57:01 risk-predisposed patients.⁶⁴ Further, in 2022, Sun et al. reported that epigenetic regulation increased expression of transporter for antigen processing (TAP) 1 and 2 and antigen presentation in HLA-B*13:01-positive dapsone hypersensitive patients but not HLA-matched tolerant controls.⁶⁵

Figure 5. HLA class I risk is thought to be necessary but not sufficient for predicting many SCAR.

Positive predictive values differ based on drug and specific hypersensitivity phenotype. 100 patients with reaction to each drug are shown, with the percentage of those that will express the defined risk allele coloured purple. The remainder of patients, coloured grey, would be tolerant to the drug despite the presence of the risk allele. For abacavir (left), 55% of those carrying HLA-B*57:01 develop abacavir hypersensitivity. For vancomycin (middle), 20% of those carrying HLA-A*32:01 develop vancomycin-DRESS/DIHS when exposed for greater than 2 weeks. For allopurinol (right), 3% of those carrying HLA-B*58:01 develop allopurinol-DRESS/DIHS or allopurinol-SJS/TEN. HLA, human leukocyte antigen; DHR, delayed drug hypersensitivity reaction; SCAR, severe cutaneous adverse reaction; AGEP, acute generalized exanthematous pustulosis; DRESS, drug reaction with eosinophilia and systemic symptoms; DIHS, drug-induced hypersensitivity syndrome; SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis.

In addition to the PPV gap in HLA screening, there is also a NPV gap where carriage of a single HLA allele does not identify individuals at risk of developing disease. Abacavir hypersensitivity appears to be a notable exception to this rule where HLA-B*57:01 is the only currently identified risk allele for abacavir hypersensitivity (100% NPV)..⁶⁶ Aside from the now widespread implementation of HLA-B*57:01 screening across different races and ethnicities in the developed world, a study in African Americans showed that HLA-B*57:01 was the only allele found in abacavir patch test positives with clinical histories consistent with abacavir hypersensitivity.⁶⁷

The influence of multiple alleles is now well-documented for predisposition to CBZ-SJS/TEN, initially associated with HLA-B*15:02 in a Taiwanese cohort.⁶⁸ While pre-prescription screening is now implemented across Asian populations, risk of CBZ-SJS/TEN has recently been associated with HLA-B*57:01 in Europeans.^57,68 HLA-A*31:01 is more highly associated with CBZ-DRESS and morbilliform eruption than CBZ-SJS/TEN.⁶⁹ The US Food and Drug Administration had previously approved labeling recommending HLA-B*15:02 screening before treatment with carbamazepine therapy in those of Southeast Asian ancestry. It is important to remember, however, that those carrying an HLA risk allele, regardless of self-identified race and ethnicity, are thought to be at equal risk for the SCAR in question. This has been raised as a recent area of concern as race- and ethnicity-based screening is unreliable and opens up the potential for structural racism.⁷⁰ Fang, et al. mined public data to show that US population screening would identify more than double the number of HLA-B*15:02 carriers at risk of CBZ-SJS/TEN than selective screening of Asian patients alone.⁷¹ Goodman and Brett recently demonstrated this point through comparison of HLA-B*58:01 frequencies, which had previously been associated with allopurinol-SCAR in South Asian populations. They highlighted that HLA-B*58:01 carriage varies by a similar degree between diverse ethnic populations and populations residing in the cities of Switzerland (a country with a high degree of racial homogeneity) suggesting again that screening for HLA risk alleles should not be race-based.⁷²

Biological and structural characteristics of HLA alleles are important to explain potential and shared risk amongst HLA alleles that share critical anchor residues in the peptide-binding domain. Interestingly, for abacavir, the HLA-B*57:01 restriction of abacavir-specific CD8+ T-cell responses can be mapped to the structure of the F pocket of the MHC-1 antigen binding cleft where even a single amino acid change within the binding pocket is enough to eliminate risk of an in vitro abacavir hypersensitivity response, and individuals carrying the B17 serotype alleles HLA-B*57:02, B*57:03 and B*58:01 also have been demonstrated to clinically tolerate abacavir.⁷³ This contrasts with the association between both HLA-B*57:01 and HLA-B*57:03 for flucloxacillin-DILI, which appears to be less restrictive across alleles.⁷⁴ For recently-characterized associations between HLA-B*57:01 and CBZ-SJS/TEN in Europeans, surface plasmon resonance was utilized to show that CBZ interacts with both HLA-B*15:02 and HLA-B*57:01, but not HLA-B*15:01 or HLA-B*51:01, suggesting shared and selective structural restriction.⁵⁷ Importantly, this aligns with early functional data demonstrating that for diverse alleles of the HLA-B75 serotype, (B*15:02, B*15:08, B*15:11, and B*15:21, but not B*15:01 or B*15:03), sharing critical anchor residues in the peptide-binding domain may also activate CBZ-responsive T-cells.⁷⁵

A 2020 study using allele-specific PCR and a retrospective review of medical records reported significant association between CBZ-SJS/TEN and HLA-B*15:21 in Filipino populations.⁷⁶ Moreover, in 2022, Wong, et al. confirmed association between HLA-B*15:11 and CBZ-SCAR in previously uncharacterized Chinese populations.⁷⁷ Similar peptide-binding specificity is now proposed to share risk of HLA-C*04:01-restricted nevirapine hypersensitivity with HLA-C*18:01 and HLA-C*05:01 alleles prevalent in Hispanic and African populations, respectively, and a shared E45-L116 motif is known to predispose methazolamide-induced SJS/TEN in Han Chinese expressing HLA-B*59:01 and HLA-B*55:02.⁷⁸

While peptide-binding algorithms have been designed to inform similar HLA sequence specificities, characterizing the actual HLA-presented structures will be critical for understanding complex structural restriction for shared HLA risk and shared structural risks to avoid during future drug development. Importantly, advanced liquid chromatography–mass-spectrometric (LC-MS/MS) methods have recently been established to directly elute and characterize peptides presented by risk HLAs in vitro. Applying these methods to investigate HLA-B*57:01 in the presence of abacavir was critical for showing perturbation of the immunopeptidome, development the altered self-peptide model and recently (2020) to identify six naturally-processed flucloxacillin-modified peptides from the HLA-B*57:01-presented immunopeptidome.^24,25,79,80 Importantly, the same peptides were confirmed by a separate group in 2021, who utilized an HLA-B*57:01 transgenic mouse to confirm peptide immunogenicity.

Traditional HLA typing utilizing high-resolution sequencing for preventative screening is not cost-effective due to the high cost of sequence-based typing across pre-prescription cohorts, which for a single allele are >110 USD/patient in a national healthcare setting.⁸¹ To enable implementation, alternative and low-cost single-allele screening assays have been developed. These include assays for well-defined risk alleles including HLA-B*15:02, HLA-B*58:01, and recently (2019) for HLA-A*32:01.^82,83,84 Importantly, compared to sequence-based typing, which can have a turnaround of a week or longer and require specialist equipment, operators, and transplant immunologists for interpretation, these PCR-based assays use as little as 10ng of DNA and have same-day turnaround in most standard on-site clinical labs at reagent cost (<10USD).⁸⁴ The use of these assays is best exemplified by the rollout and quality assurance program of HLA-B*57:01 single allele screening prior to abacavir, which is available through many commercial laboratories.⁴⁸ These assays show not only the feasibility of pre-prescription screening, but also testing to aid in preemptive management risk stratification. An example of this is an HLA-A*32:01 single allele assay to risk-stratify patients going on more than 2 weeks of vancomycin treatment for DRESS.⁵⁸ Given that DRESS has a latency of 2–8 weeks, it would be possible to risk-stratify based on a single allele assay done within the first 2 weeks of treatment.⁸⁴ Panel tests have also been developed providing similar allele-focused typing, but of several high-risk alleles, to enable both screening for the current drug and extended risk-typing for future treatment. Design of such panels resulted in total cost savings of 491USD compared to standard typing procedures.⁸⁵ Importantly, the development of these assays for other alleles will not only serve to enable cost-effective clinical prevention strategies, but also diagnostics, and provide lower cost assays for clinical trial and risk validation studies where an HLA risk allele is proposed. However, the problem of low PPV and incomplete NPV for most drug-HLA associations to-date remains, and patients should continue to be counseled about risks even after negative genetic screening.

SCAR AND FUTURE PROSPECTS FOR PREDICTION, PREVENTION, EARLIER DIAGNOSIS, AND TARGETED TREATMENTS

Future precision medicine approaches will improve prediction, prevention, and earlier diagnosis (Figure 6). Identifying implicated drugs, and avoiding cross-reactive drugs, will improve patient options and drug safety. Definition of cellular and molecular signatures at the site of tissue damage will guide targeted treatments. This will be essential to reduce SCAR-associated morbidity and mortality. As HLA-class I appears necessary but not sufficient for the development of many SCAR, increasing knowledge of other genetic and ecological factors will be helpful to risk-stratify patients. For SJS/TEN, granulysin has been previously identified as the major cytolytic peptide mediating keratinocyte cell death (Figure 1C), with levels notably increased early in peripheral blood and blister fluid compared to other delayed hypersensitivity phenotypes, leading to the proposal of point-of-care testing for the earlier detection of SJS/TEN.^70,86,87. However, the pro-inflammatory cytokine TNFα is also found in SJS/TEN blister fluid, with recent clinical trials demonstrating the capacity of the anti-TNF biologic etanercept to lower levels of granulysin and mortality.^88,89 Moreover, recent research also implicates other potential mechanisms of keratinocyte-specific cell death, which warrant investigation as alternate treatment pathways, which may be combined with the recently reported optimal and standardized Delphi-based consensus for supportive care.⁹⁰ First, Zhang et al. found that miR-375–3p was upregulated in plasma exosomes during SJS/TEN, which upon internalization by primary keratinocytes led to down-regulation of a key regulator of keratinocyte apoptosis, the X-linked inhibitor of apoptosis protein (XIAP), leading to induction of mitochondria-dependent apoptosis.⁹¹ Second, Kinoshita reported that skin-infiltrating T-cells also produce lipocalin-2, which could trigger neutrophil NETosis, formation of neutrophil extracellular traps (NETs), and secretion of antimicrobial peptide LL37. Importantly, LL37 was able to induce expression of formyl peptide receptor 1 (FPR1) on keratinocytes, making them susceptible to necroptosis⁹² (Figure 1C). Other studies have also recently identified upregulation of serum IL33 and Galectin-7 in SJS/TEN patients as potential future diagnostic biomarkers in peripheral circulation.^93,94For DRESS/DIHS, serum thymus and activation-regulated chemokine (sTARC) and soluble and CD4+ T-cell surface-bound OX40 may also be elevated early and associated with disease activity.^95,96and novel treatment strategies should employ knowledge of molecular and cellular signatures at the site of tissue damage followed prospectively through disease progression and under different treatments. For example, Kim, et al. used 3’ 10X-single-cell RNA sequencing and identified up-regulation of expression of several genes in the JAK-STAT pathway in CD4 and CD8 T-cells from patient skin during a case of refractory DRESS associated with trimethoprim-sulfamethoxazole. Based on this, the patient was treated with the JAK 1,3 inhibitor, tofacitinib, with complete symptom resolution.³⁷ Single-cell approaches identifying the immunodominant TCR at the site of tissue damage will help define specific antigens and determine HLA restriction at an “n=1 level”, further improving preventive, diagnostic and treatment efforts.³⁷

Figure 6. Precision medicine approaches to SCAR.

Approaches target prediction and prevention through risk assessment and genetic screening. Multifaceted approaches will facilitate earlier identification of disease. In vivo, ex vivo, in vitro and genetic studies combined with clinical factors will help to identify the culprit drug and improve future drug safety. Studies looking at disease at the site of tissue damage will inform immunopathogenesis, prognosis, oligoclonal T-cell receptor and identify targeted treatments. Functional studies will be important for defining molecular interactions as well as epitope discovery that along with the oligoclonal TCR (s) identified at the site of tissue damage will help define immunodominance and. HLA specificity of a given drug hypersensitivity reaction without the need for large scale epidemiological studies. HLA, human leukocyte antigen.

Figure created using Biorender.com

Abbreviations:

AGEP

Acute generalized exanthematous pustulosis

AI

artificial intelligence

ALDEN

algorithm of drug causality assessment for epidermal necrolysis

APC

antigen presenting cell

CBZ

carbamazepine

CCL

C-C motif chemokine ligand

CCR

chemokine receptor

CMV

cytomegalovirus

CTLA

cytotoxic lymphocyte antigen

DC

dendritic cell

DHR

delayed drug hypersensitivity reactions

DIHS

drug-induced hypersensitivity reaction

DILI

drug-induced liver injury

DNA

deoxyribonucleic acid

DRESS/DIHS

drug reaction with eosinophilia and systemic symptoms/drug- induced hypersensitivity syndrome

EBV

Epstein-Barr virus

PMBC

peripheral blood mononuclear cells

ELISpot

enzyme-linked immunospot

ERAP

endoplasmic reticulum aminopeptidase

FDE

fixed drug eruption

FPR1

formyl peptide receptor 1

Grb

growth factor receptor-bound protein

GM-CSF

granulocyte macrophage colony-stimulating factor

HHV

human herpesvirus

HIV

human immunodeficiency virus

HLA

human leukocyte antigen

ICI

immune checkpoint inhibitors

ILC

innate lymphoid cell

IL

interleukin

LAT

linker for activation of T-cells

Lck

lymphocyte-specific protein tyrosine kinase

MAPK

mitogen-activated protein kinase

MDE

morbilliform drug eruption

MHC

major histocompatibility complex

NETs

neutrophil extracellular traps

NK

natural killer cell

NKT

natural killer T-cell

NPV

negative predictive value

PCR

polymerase chain reaction

PD

programmed death

PI3K

phosphatidylinositol 3 kinase

PLC

phospholipase

PPV

positive predictive value

RAS

rat sarcoma virus’ protein

SCAR

severe cutaneous adverse reactions

SCORTEN

severity-of-illness score for TEN

SJS/TEN

Stevens-Johnson syndrome/toxic epidermal necrolysis

SOS

son of sevenless protein

SSLR

serum sickness like reaction

sTARC

serum thymus and activation-regulated chemokine

TAP

transporter for antigen processing

TB

tuberculosis

TCR

T-cell receptor

TRM

resident T cells

TNF

tumor necrosis factor

WES

whole exome sequencing

WGS

whole genome sequencing

XIAP

X-linked inhibitor of apoptosis protein

ZAP70

zeta-chain-associated protein kinase 70