Drug hapten‐specific T‐cell activation: Current status and unanswered questions

Drug haptens are formed from the irreversible, covalent binding of drugs to nucleophilic moieties on proteins, which can warrant adverse reactions in the body including severe delayed‐type, T‐cell mediated, drug hypersensitivity reactions (DHRs). While three main pathways exist for the activation of T‐cells in DHRs, namely the hapten model, the pharmacological interaction model and the altered peptide repertoire model, the exact antigenic determinants responsible have not yet been defined. In recent years, progress has been made using advanced mass spectrometry‐based proteomic methods to identify protein carriers and characterise the structure of drug‐haptenated proteins. Since genome‐wide association studies discovered a link between human leukocyte antigens (HLA) and an individual's susceptibility to DHRs, much effort has been made to define the drug‐associated HLA ligands driving T‐cell activation, including the elution of natural HLA peptides from HLA molecules and the generation of HLA‐binding peptides. In this review, we discuss our current methodology used to design and synthesise drug‐modified HLA ligands to investigate their immunogenicity using T‐cell models, and thus their implication in drug hypersensitivity.


Statement of significance
Understanding of the mechanisms underlying haptenspecific T-cell activation remains elusive, especially with respect to the antigenic determinant responsible for the reaction. Our article reviews the current status of drughaptenated protein characterisation and their implication in drug hypersensitivity. Current methodology and challenges in depicting the antigenic epitope have been addressed.

Delayed-type hypersensitivity reactions
This review focuses on the mechanisms underlying delayed-type, T-cell mediated DHRs. The T-cell subset and effector mechanism involved can differ in these reactions, allowing further subdivision into four categories as shown in Table 1. Type IVa is a Th1 response with monocytic inflammation whereas Type IVb is a Th2 response with eosinophilic inflammation. Cytotoxic CD8+ T-cells and helper CD4+ T-cells mediate the cell death in Type IVc and lastly, a T-cell response causing neutrophilic inflammation is evident in Type IVd [3].
Although different types of DHRs have been classified, it is important to note that there is likely an overlap between each type meaning many patients will show symptoms from each. Moreover, drug responses will exhibit interindividual variability [4].
DHRs have multiple clinical manifestations and can involve different organ systems such as the skin, liver and kidneys. With that said, the skin is the organ that is mostly commonly affected in DHRs [5]. It remains unknown why certain organs are targeted over others but it is possible the skin is more susceptible to drug reactions due to the large vascular network and abundance of dendritic cells and macrophages that survey the local environment for signs of stress and antigenic material [6].

Cutaneous reactions
Cutaneous reactions vary in appearance and severity. While more than half of cutaneous ADRs may present as a mild cutaneous reaction known as maculopapular rash [7], a proportion of cutaneous reactions are deemed more serious and are potentially fatal. Severe cutaneous adverse reactions include Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN) and drug reaction with eosinophilia and systemic symptoms (DRESS) [8].

Maculopapular exanthema
Maculopapular exanthema (MPE) is typically skin eruptions of both flat and raised lesions with a diameter of approximately 1-5 mm known as macules and papules, respectively, which often merge forming a patch or plaque. This condition distributes symmetrically and bilaterally affecting the face, neck, upper trunk and limbs, and may occur alongside a fever [9]. MPE studies have demonstrated the presence of CD4+ and CD8+ T-cells in drug-induced cutaneous eruptions, particularly in the dermo-epidermal junction and epidermis of the skin [10,11]. Several drugs were involved in the development of these eruptions including amoxicillin, carbamazepine (CBZ) and cefazoline. Moreover, sulphamethoxazole (SMX)-specific T-cells generated from a patient who developed a skin eruption following SMX therapy were capable of killing keratinocytes [12]. This combined evidence suggests immune involvement for the basal cell vacuolisation and keratinocyte death seen in drug-induced MPE [13]. Consistent throughout literature, CD4+ T-cells are the dominant effector cells in MPE [14][15][16], as they are present in higher numbers compared to CD8+ T-cells and clinical MPE symptoms can occur in the almost absence of CD8+ T-cells showing that CD4+ cells are adequate to trigger this reaction. Activation of CD8+ T-cells predominantly appears to be typical with more severe reactions such as bullous drug exanthems or those involving the liver [15].

DRESS
DRESS is a serious, multi-organ condition characterised by a high fever, skin eruption, haematological abnormalities and mucosal/internal organ involvement. The majority of patients will experience a skin eruption covering over 50% of the body surface area. Eosinophilia is the most common haematological abnormality in patients as indicated by the name, followed by atypical lymphocytes, which confirms immune involvement. Two or more internal organs are affected in over half of cases. Overall, the liver appears to be the main internal target in DRESS followed by the kidneys and lungs [17]. Approximately 50 drugs have been shown to cause DRESS. CBZ appears to be the most frequently involved followed by allopurinol (ALP), an anti-gout drug [18].
ALP is known to cause DRESS notably with greater kidney involvement in comparison to CBZ [17], therefore those with underlying kidney problems may be at greater risk. Other risk drugs include dapsone (DDS) and sulphonamides [19].

SJS/TEN
SJS/TEN are life-threatening skin conditions characterised by epidermal necrosis resulting in the epidermis-dermis separation, mucosal erosions and lesions, and are often accompanied with a fever. SJS and TEN are classed according to a single disease severity spectrum.
SJS patients have 10% or less total skin detachment, whereas more severe detachment of 30% or more is considered as TEN [22]. Similar to DRESS, internal organs may be involved including the liver or, more commonly in SJS, oral cavity and eye involvement [21]. SJS/TEN is predominantly drug induced; common culprits include antibiotics, nonsteroidal anti-inflammatory drugs, ALP and antiepileptic drugs [23].

Idiosyncratic drug-induced liver injury
Drug-induced liver injury (DILI), also referred to as hepatotoxicity, is an ADR causing liver dysfunction or abnormal liver serology [25]. The major metabolic function of the liver makes it highly susceptible to direct drug-induced damage. In fact, adverse hepatic events are more likely to occur with drugs that are majorly metabolised in the liver [26].
DILI is generally classified as either dose dependent and predictable or idiosyncratic (IDILI) with a complex dose relationship [27]. The former refers to reactions such as acetaminophen (APAP) toxicity that has been studied extensively [28][29][30]. It is widely understood that APAP is oxidised to a reactive intermediate N-acetyl-p-benzoquinoneimine (NAPQI) by the microsomal P-450 system, which under normal doses is detoxified and eliminated by glutathione (GSH) conjugation. In APAP overdose cases, GSH is depleted, therefore the excess NAPQI reacts with cysteine residues in cellular proteins, leading to APAP-protein adducts. The binding of APAP to proteins in the mitochondria induces oxidative stress and subsequent cell necrosis [31]. On the other hand, IDILI accounts for approximately 10-15% cases of acute liver injury [32]. The pathogenesis of IDILI is a lot more obscure and due to the lack of preclinical in vivo evidence of hepatotoxicity in IDILI drugs, animal models have not been readily obtainable. It is possible that co-stimulatory innate signalling is augmented by mitochondrial injury causing oxidative stress and/or drug-induced inhibition of pumps that are responsible of exporting bile, resulting in the build-up of antigenic determinants at the site of tissue injury [33]. Moreover, drug-specific T-cells are observed in patients with IDILI caused by flucloxacillin and co-amoxiclav [34,35], providing evidence of immune involvement.

HOW DRUGS ACTIVATE T-CELLS
The activation of T-cells is predominately determined by two signals.

Pathways of T-cell activation
The

Hapten
The earliest proposed pathway is the hapten hypothesis discovered in 1930s [40], using chemical sensitisers and since described as the clas-sical explanation for DHRs. The term hapten describes low molecular weight compounds that are incapable of initiating an immune response alone. However, when covalently bound to circulating proteins, these low molecular weight compounds are able to form an antigenic determinant that can activate T-cells. A drug is considered a prohapten if its derived metabolite covalently binds to proteins and becomes antigenic, as seen with SMX and its nitroso metabolite (SMX-NO). It is assumed that these protein adducts are endogenously processed into drug-modified peptides and presented by MHC molecules on the surface of APCs as de novo antigens, which can stimulate antigenspecific T-cells [41].
Evidence β-Lactams are often used as model haptens in DHRs. Piperacillin and benzylpenicillin (BP)-human serum albumin (HSA)-specific T-cell clones have been identified, which do not respond in the presence of fixed APCs or proteasome inhibitors, demonstrating the requirement of antigen processing for T-cell activation [42]. Furthermore, the covalent binding of β-lactams is confirmed by the activation of piperacillin T-cell clones in the presence of drug-pulsed APCs [38] and the binding of BP to cellular proteins in dendritic cells [43]. The involvement of drug hapten protein binding in T-cell activation has been demonstrated directly through the detection of patient T-cells that are responsive towards drug-HSA adducts [42,43] and synthetic drug-haptenated peptides [44][45][46]. Haptenic metabolites such as SMX-NO can activate T-cells in a processing-dependent manner [47,48]; however, SMX-NO also forms covalent bonds with nucleophilic amino acid residues in peptides associated with MHC molecules expressed on the surface of APC, which complicates delineation of pathways of drug-specific T-cell activation. This concept also applies to the penicillins since BP-T-cell clones have been shown to be activated in a processing-independent manner [49]. Therefore, there are limitations to using antigen processingblocking assays to distinguish between the hapten and p-i pathways.

The p-i concept
Challenging the requirement for antigen processing and covalent bind- Evidence SMX-specific T-cell responses are observed in the presence of SMX with fixed APCs but not with drug-pulsed APCs, collectively indicating that SMX activates T-cells in a processing-independent manner, without binding covalently to a peptide or an immune receptor [52,53]. SMX responses are abrogated in the absence of APCs and exhibit MHC restriction, suggesting that SMX interacts with the MHC rather than directly activating the TCR. Similar observations of processingindependent activation of specific T-cells and labile binding have been observed with CBZ, ALP and its metabolite oxypurinol (OXY) [54,55].
Interestingly, unlike SMX, CBZ can activate T-cells in the absence of APCs suggesting a direct TCR interaction. Furthermore, the downregulation of the TCR and the influx of Ca 2+ which is indicative of TCR engagement have been observed rapidly in SMX-, lidocaine-and ALP/OXY-specific T-cells clones following drug exposure. This is inconsistent with the kinetics of antigen processing, which usually requires 4-6 h [55,56].
It is important to note that T-cell activation by a drug may not be restricted to a sole mechanism. This phenomenon is best described with SMX hypersensitivity since T-cells are activated by both p-i and hapten pathways as mentioned previously [57].

Altered peptide
The altered peptide hypothesis arose following the report of human leukocyte antigen (HLA)-B*57:01 restricted T-cells in abacavirinduced hypersensitivity [58], and is based on the structural elucidation that a drug can bind directly to the pockets of the MHC binding groove. This changes the conformation and chemistry of the binding groove, and subsequently, the repertoire of peptides that are presented. It is worth emphasising that the peptides presented are not drug-modified peptides as seen in the hapten pathway. Alternatively, they are self-peptides that would not usually be presented by that particular HLA allele, as a shift in amino acid residues occurs to accommodate the bound drug. Drug-HLA binding may occur in the endoplasmic reticulum prior to intracellular loading or on the cell surface after a peptide-HLA complex is formed [59]. The altered peptide repertoire is considered a third independent model of drug hypersensitivity by many [60,61]; however, this has been challenged in literature and rather viewed as a branch of the p-i model [62].

SMX-NO
SMX hypersensitivity can affect multiple organs including injury to the skin, liver or kidney. Reactive metabolites of SMX bind to protein thiols [47] therefore they are capable of forming the antigenic determinants that are likely implicated in SMX hypersensitivity. Defining the mechanism is more difficult for a prohapten since the metabolite is not readily available and an in vitro metabolising system does not exist. However, in some cases, the reactive metabolites have been successfully synthesised. SMX oxidation generates two oxidative metabolites, a hydroxylamine intermediate (SMX-NHOH) and the previously mentioned SMX-NO [78]. Early studies confirm SMX-NO protein adducts in human plasma [79] and cellular models [80][81][82].
While the presence of the drug modification was confirmed, the exact haptenic structure was not characterised. SMX-NO adducts have been detected with Ig and albumin in mouse serum models and cysteine capping revealed that SMX-NO was interacting with cysteine residues in mouse serum albumin (MSA) [83]. Later LC-MS/MS analysis has revealed that SMX-NO forms multiple adducts with proteins.
An N-hydroxysulphinaminde-HSA adduct has been detected with the modification on Cys34. SMX-NO also forms N-hydroxysulphonaminde adducts with human GSH S-transferase pi (GSTP), as well as sulphinamide and N-hydroxysulphinaminde adducts. Interestingly, GSTP was modified on Cys47 residue exclusively. Further protein adducts have been detected with GSH and synthetic peptides containing cysteine residues [84]. Recently, novel HSA adducts were discovered on lysine and tyrosine residues including a Schiff base adduct and arylazoalkane adduct, and unlike any previous studies, a SMX-lysine adduct was detected in patient sera [85].   [98]. The discovery of these associations has led to the development of in vitro assays to define the mechanisms underlying drug-HLA associations.

IDENTIFICATION OF DRUG-ASSOCIATED HLA LIGANDS
The identification of peptide epitopes is crucial to the understanding of a cellular immune response and is significant in the development of peptide-based vaccines and cancer immunotherapy [99,100]. The

Naturally processed HLA ligands
The identification of natural HLA-peptides involves the immunoaffin- As mentioned previously HLA-B*57:01 peptides have been identified in the presence of abacavir using this HLA peptide elution method.
Similar studies have been attempted for CBZ. Initial studies reported that similar peptides are bound to HLA-B*15:02 in the presence and absence of CBZ and concluded that the p-i concept was most appropriate to explain this DHR [101]. This has now been challenged by the report of a shift in the peptide repertoire due to CBZ, favouring smaller residues at P4 and P6 and increase of hydrophobic residues at several positions [59]. Unlike abacavir, CBZ did not alter the anchor residues and the shifts were more subtle, nevertheless the same pathway of T-cell activation could apply. More recently, it has been dis-

Designer HLA ligands
Alternatively, reverse engineering has been used to design and synthesise drug-modified peptides with HLA-binding motifs for use in T-cell models. 'Designer' peptide studies have successfully identified immunogenic T-cell epitopes, initially reported by the Weltzien group using synthetic BP-modified HLA-DRB1*04:01 binding peptides. Such T-cell responses revealed that activation was dependent on the sitespecific modification of the peptide backbone [103]. Expanding on this, drug-haptenated candidate peptides have been generated from drugprotein conjugates including BP-haptenated peptides synthesised from BP-HSA conjugates [44] and an amoxicillin-haptenated peptide synthesised from an exosomal protein, SRY-box 30, which covalently bound to amoxicillin in hepatocytes [104]. Both peptides were designed around HLA-binding predictions and shown to stimulate naïve T-cells in healthy donors. More recently, amoxicillin-modified HLA-DRB1*15:01 binding peptides were incorporated into a DILIpatient T-cell model, which resulted in successful T-cell activation [46].

STRATEGIES FOR THE DESIGNER PEPTIDE MODEL
The prediction of drug hypersensitivity remains the principal challenge in the field. Since the exact antigenic determinant has not been defined as discussed, they cannot be used in T-cell assays as a predictive tool. While T-cell models for parent drugs exist, they do not define drug hypersensitivity at the peptide level thus the functional epitopes remain unspecified. Rather than focusing on the unknowns, our knowledge of HLA associations of DHRs and the pivotal role of peptides in T-cell activation can be used to generate and investigate potential antigens in the designer peptide T-cell model.

HLA ligand design rationale
Peptides are designed with the aim of binding to HLA risk alleles, react- P9 for MHC class II ligands [106,107]. Incorporating these anchor residues enhances the binding affinity of a peptide to the HLA, which can be predicted and scored using peptide binding databases such as is incorporated into the peptide sequence depending on the reactivity of the drug. Lysine residues are used for β-lactam binding due to the known reactivity of lysine with the β-lactam ring, whereas cysteine residues are used for the binding of SMX-NO. As mentioned previously, the position of a drug modification influences a T-cell response therefore its placement must be deliberate and TCR binding preferences conferred [46]. Structural immunogenicity studies have revealed that positions P4-6 of MHC I peptides are most important in the interaction between TCR and pMHC. Moreover, TCR contact sites have a preference for bulky, aromatic residues [109]. Based on this an MHC I peptide would be designed with a central lysine residue for optimal TCR contact. Chemically synthesised peptides have a free N-terminal that is more reactive than the ε-amino groups of lysine residues. Consequently, peptides are designed with an N-terminal protection group such as FMOC to ensure site-specific modification.
Lastly, the physical chemistry of a peptide is considered to assess its solubility for cellular assays as hydrophobic peptides are inclined to aggregate in solution, thus hydrophilic residues may be used as a replacement.

Synthesis and drug modification
In the last decade, the discovery of protein targets and peptide epitopes in drug development has massively increased due to advances in omics technologies [110], thus there has been immense demand for peptide synthesis in the pharmaceutical industry for the development of therapeutic peptides. The innovative work of Bruce Merrifield introduced solid-phase peptide synthesis (SPPS) and the automation of peptide synthesis [111], and while multiple approaches to chemical peptide synthesis exist such as solution-phase synthesis, this is the most commonly used method for peptide production today [112]. The principle of SPPS is the anchoring of an amino acid at the C-terminus to an insoluble polymer such as resin that acts as a solid 'platform' for the addi-

Translation to in vitro T-cell assays
Drug-modified peptides are used to explore the immunogenicity of a specific epitope as well as a predictive tool for HLA susceptibility.

Challenges in drug-modified peptide synthesis
While drug-modified HLA peptides have been successfully generated for penicillin [44] amoxicillin [46], flucloxacillin and DDS-NO (unpublished data), challenges have arisen. Firstly, chemistry is conducted at a microscale level therefore the ideal method of using SPPS to generate drug-modified peptides by incorporating drug-modified-lysine monomers residues into the sequence production, as executed by the Pallardy group [44], is not cost-effective. Consequently, yielding sufficient peptide in batches for T-cell assays is extremely laborious. Difficulties significantly revolve around peptide stability and side reactions. Without solubility, peptides are not modifiable nor are they of any use for in vitro assays. For some HLA-binding peptides it is tolerable to replace hydrophobic residues for more hydrophilic residues to increase solubility. However, in some cases, this can reduce the binding affinity of the peptide to the MHC, therefore is counterproductive for a model based on HLA binding. While reconstituting the sample with solvents such as acetonitrile for HPLC analysis could overcome this problem, the sample must dissolve in the mobile phase. If the composition of the mobile phase in a gradient system is lower than the diluent needed to dissolve the sample, peak shape and retention time are affected. DMSO is often used to dissolve peptides with poor solubility; however, this can have considerable implications on cellular behaviour as it is toxic to cells at low concentrations and will have to be controlled in cellular models. Importantly, side reactions occur with both drugs and residues; β-lactams can modify the N-terminus of a peptide and although an N-terminal protection group can prevent this as discussed, it must be removed via an extra HPLC purification step that increases labour time and decreases the yield. Moreover, βlactams yield an array of degradation products therefore a more complex peak profile is formed making the drug-modified peptide peak less distinguishable. Drug dimer-or trimerisation also adds to the complexity of identifying drug-modified products that occur extensively with amoxicillin. Some residues are more prone to side reactions in peptide synthesis than others based on their reactivity. Cysteine residues are known to form disulphide bonds causing aggregation in solution. Usually in post-translational modification studies, this is overcome using cysteine capping; however, this is not feasible in some drug designer peptide synthesis where cysteine residues represent the main target F I G U R E 3 The process of peptide synthesis and characterisation. The generation of synthetic drug-modified peptides is described in four steps: (1) The FMOC-protected peptide is incubated with drug for covalent binding. A drug like amoxicillin is bound to a lysine residue in a peptide through an amide linkage. (2) The drug-peptide mixture is purified using a succession of HPLC experiments, and the suspected drug-modified peptide is fractionated. (3) The collected sample is analysed using LC-MS/MS to confirm successful drug modification. Peptides are identified using de novo sequencing thus drug-characteristic fragments are used to identify successful drug modification. The N-terminal FMOC-protection group is confirmed by a peak with m/z at 179. (4) The peptide is deprotected using reagents like piperidine that removes the N-terminal FMOC protection group. HPLC purification and LC-MS/MS confirmation steps are repeated to confirm the removal of FMOC and yield a final deprotected drug-modified peptide residue for drug modification (i.e. DDS modifications). Additionally, it is difficult to control oxidation status as cysteine can undergo mono-, di-and tri-oxidation. Tryptophan and methionine also form oxidation products. Challenges are not limited to experimental procedures but also data analysis. A limitation of current peptide identification soft-ware such as PEAKS or Protein Pilot to analyse drug-modified peptides is that they do not recognise the neutral loss of drugs, which is common for drugs like β-lactams, therefore modifications must be identified manually, which is timely and has higher error rates than computational methods.

CONCLUSION AND FUTURE WORK FOR THE FIELD
DHRs cannot be predicted therefore remain a significant burden for patients and healthcare providers and a challenge for the pharmaceutical industry. Over the last decade significant progress has been made to define the critical antigen responsible for T-cell activation in the development of a reaction, including the identification and characterisation of antigenic drug-protein conjugates and natural drug-modified HLAbinding peptide. Moreover, HLA association studies have confirmed the importance of the pMHC-TCR interaction in DHRs. Despite this, the exact peptide repertoire presented and the binding interactions between the TCR and MHC remain to be fully defined. The interaction between the TCR and pMHC is considered weak therefore it dissociates rapidly making it difficult to detect T-cell responses in vitro [113].
With that being said, epitope identification facilitates the potential use for tetramers in drug hypersensitivity. A panel of drug-modified HLAbinding peptides can be used to generate multimeric peptide-MHC complexes to enhance a monomeric response, as the avidity of multimerisation significantly prolongs this interaction. Tetramers can be used to screen the diversity of drug-specific T-cells, including their frequency and phenotype. This will identify immunodominant epitopes in the patient population and the extent of cross reactivity. The dominant epitope tetramers may then be used as a screening tool in healthy individuals. Furthermore, the molecular docking approach and X-ray crystallography studies will be used to model the interaction between the MHC and TCR and successfully define the critical epitope.