Carbamazepine Induces Focused T Cell Responses in Resolved Stevens-Johnson Syndrome and Toxic Epidermal Necrolysis Cases But Does Not Perturb the Immunopeptidome for T Cell Recognition

Antiseizure medications (ASMs) are frequently implicated in T cell-mediated drug hypersensitivity reactions and cause skin tropic pathologies that range in severity from mild rashes to life-threatening systemic syndromes. During the acute stages of the more severe manifestations of these reactions, drug responsive proinflammatory CD8+ T cells display classical features of Th1 cytokine production (e.g. IFNγ) and cytolysis (e.g. granzyme B, perforin). These T cells may be found locally at the site of pathology (e.g. blister cells/fluid), as well as systemically (e.g. blood, organs). What is less understood are the long-lived immunological effects of the memory T cell pool following T cell-mediated drug hypersensitivity reactions. In this study, we examine the ASM carbamazepine (CBZ) and the CBZ-reactive memory T cell pool in patients who have a history of either Stevens-Johnson syndrome (SJS) or toxic epidermal necrolysis (TEN) from 3-to-20 years following their initial adverse reaction. We show that in vitro drug restimulation of CBZ-reactive CD8+ T cells results in a proinflammatory profile and produces a mainly focused, yet private, T cell receptor (TCR) usage amongst human leukocyte antigen (HLA)-B*15:02-positive SJS or TEN patients. Additionally, we show that expression of these CBZ-reactive TCRs in a reporter cell line, lacking endogenous αβTCR, recapitulates the features of TCR activation reported for ASM-treated T cell lines/clones, providing a useful tool for further functional validations. Finally, we conduct a comprehensive evaluation of the HLA-B*15:02 immunopeptidome following ASM (or a metabolite) treatment of a HLA-B*15:02-positive B-lymphoblastoid cell line (C1R.B*15:02) and minor perturbation of the peptide repertoire. Collectively, this study shows that the CBZ-reactive T cells characterized require both the drug and HLA-B*15:02 for activation and that reactivation of memory T cells from blood results in a focused private TCR profile in patients with resolved disease.


INTRODUCTION
Antiseizure medications (ASMs) are routinely used to treat epilepsy and other neuropsychiatric conditions, such as neuropathic pain and bipolar affective disorder. Commonly prescribed ASMs, including carbamazepine (CBZ) and phenytoin (PHT), have been implicated in life-threatening drug hypersensitivity reactions (DHRs) that predominantly target the skin (1). These drug reactions typically occur in the first 2-3 months of drug administration and span a range of clinical pathologies, including mild rash [e.g. maculopapular exanthema (MPE)], systemic symptoms [e.g. drug reaction with eosinophilia and systemic symptoms (DRESS)], as well as more severe bullous reactions involving rapid development of blisters and lesions accompanied by skin detachment [e.g. Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN)] (2)(3)(4)(5). Whilst early drug withdrawal can ameliorate symptoms in milder cases, the more severe cutaneous reactions require specific clinical treatments and hospitalization for disease resolution.
Small molecule drugs, such as ASMs, can promote immune reactions via T cell activation. ASM-induced T cell activation is proposed to occur via non-covalent and labile interactions between the drug, or a metabolite of the parent drug, and the HLA/peptide complex and T cell receptor (TCR) (21). Additionally, whilst peptide occupancy of the HLA molecule is necessary, there is no requirement for de novo peptide/HLA complex formation (21) and the drug does not markedly alter the anchor residue preference of HLA-B*15:02 suggesting it is not binding within the primary anchor pockets during peptide loading (22). Contesting the latter, a recent report has suggested that the CBZ metabolite, carbamazepine-10,11epoxide (ECBZ), is capable of altering the HLA-B*15:02 immunopeptidome and anchor residue prevalence (23). To date there is no published structure of the interaction of CBZ and the peptide/HLA and/or TCR, although diverse in silico models have been generated ranging from our previous proposal of occupation of the antigen-binding cleft, to more surface exposed positions, or, with the additional inclusion of identified drug-reactive TCRs, the HLA-TCR interface (21,22,24).
In contrast to the antiretroviral drug abacavir, which drastically alters the immunopeptidome and facilitates a diverse and polyclonal T cell response (25)(26)(27), examination of CBZ-induced T cells revealed a much more focused TCR usage associated with SJS/TEN (28). A recent study utilized next generation sequencing to examine the blister cells of CBZinduced SJS/TEN patients identifying a public HLA-B*15:02restricted abTCR [complementarity determining region (CDR)3 sequence; TCRa VFDNTDKLI and TCRb ASSLAGELF] that also recognized structural analogs of CBZ (24). These TCR signatures contrast with an early report demonstrating the expansion of VA-22 and VB-11-ISGSY dominant clonotypes derived from the blister fluid of SJS/TEN patients recruited in Taiwan expressing HLA-B*15:02 (28). These TCRs were identified using traditional Sanger sequencing and were confounded by in vitro T cell co-culture with antigen presenting cells (APCs), which has been suggested to bias the outgrown T cell repertoire (29). Another study examining the ex vivo TCRb repertoire in peripheral blood mononuclear cells (PBMCs) isolated from CBZ-induced SJS (n=5) or TEN (n=1) patients showed that diversity was directly linked to disease severity, with the TEN patient having a significantly decreased TCRb repertoire compared to SJS patients (30).
What is less understood are the long-lived immunological features of the CBZ-responsive memory T cell pool in patients with resolved disease. Do these patients exhibit similar immunological profiles to those reported during active disease, such as effector functions including Th1 cytokine production (e.g. IFNg) (28) and cytolytic molecules (e.g. granzyme B, perforin) (31), TCR repertoire clonality, mode of drug recognition and cross-reactivity towards other ASMs? This study shows that the in vitro drug expanded TCR repertoire of resolved CBZ-induced SJS or TEN patients remains relatively clonal years after acute disease. These T cells respond to HLA-B15-positive APCs in the presence of CBZ, as well as related compounds, a finding that is recapitulated by expression of these cloned TCRs in a reporter cell line. Here, the responses required the continuous presence of soluble drug and HLA-B15 expression, without the need for de novo generation of the peptide/HLA complex. No marked drug-induced alteration in peptide anchor preference of HLA-B*15:02 was induced by any of CBZ, the metabolite ECBZ, or the structurally related ASM OXC. Together these data support the presence of a long-lived memory pool of CBZ responsive T cells in SJS or TEN patients, which are activated by structurally related ASMs in the absence of marked alteration of the immunopeptidome.

Study Cohort
Eight patients were recruited from Hong Kong, with six HLA-B*15:02-positive patients experiencing CBZ-induced SJS or TEN, one HLA-B*15:02-negative patient with CBZ-induced SJS and one CBZ-tolerant ( Table 1, Supplementary Table 1). As described in our previous studies (32, 33) the diagnosis of SJS or TEN was based on the criteria by Roujeau and Stern (2) defined by skin detachment in two or more mucosal sites, and was confirmed by dermatologists. Healthy HLA-B*15:02 + individual controls (AP numbers; n=7; Table 1) were also recruited from both Monash University and the Australian Bone Marrow Donor Registry. All study participants provided written consent, with ethics approval granted by the Joint

Cell Preparation and In Vitro Expansion of Drug-Induced T Cells
Blood samples were collected from study participants, with PBMCs isolated using Ficoll-Paque (GE Healthcare, Uppsala, Sweden) density gradient centrifugation and either used immediately or cryopreserved in fetal calf serum (FCS) containing 10% DMSO (Sigma-Aldrich, MO, USA) at -196°C until required. Before use for T cell stimulation, PBMCs were quickly thawed in 37°C and washed twice in RPMI 1640 (Gibco, Life Technologies, NY, USA) and resuspended in Complete Medium [RPMI 1640 supplemented with 2 mM MEM nonessential amino acid solution (Gibco), 100 mM HEPES (Gibco), 2 mM L-glutamine (Gibco), Penicillin/Streptomycin (Gibco), 50 mM 2-mercaptoethanol (Sigma-Aldrich) and 10% heat inactivated human blood group AB serum (Sigma-Aldrich)].
Drug-induced T cells were in vitro expanded following PBMC stimulation at a density of 5 million per 2 mL of Complete Medium in a 24-well plate with 25 mg/mL of either CBZ (Sigma) or the metabolite ECBZ (Sigma or SYNthesis med chem, Australia). On days 4 to 14, T cell cultures were supplemented with 50 U/mL of recombinant human IL-2 (Peprotech, NJ, USA) and subcultured as required to ensure optimal outgrowth.

Drug-Pulsed APCs and T Cell Stimulation
Functionality of the drug-induced T cells, including crossreactivity, was assessed using intracellular cytokine staining (ICS) for production of either IFNg and/or TNF (36 Figure  2A). All monoclonal antibodies were purchased from BD and titrated for optimal staining efficiency. A maximum of 50,000 lymphocytes were acquired on a BD LSRII flow cytometer utilizing BD FACSDIVA ™ software (FlowCore, Monash University) and analyzed using FlowJo software (version 10, BD). Representative gating strategy is shown in Supplementary Figure 2B.  Figure 2C). Single cells were sorted on a BD Influx flow cytometer (FlowCore, Monash University) directly into 96-well PCR plates (Bio-Rad, Hercules, CA, USA) based on CD8 + IFNgfor drug-unresponsive T cells (negative control) and CD8 + IFNg + for drug-induced T cells (Supplementary Figure 2C). Sorted plates were immediately stored at -80°C until required. Analysis of paired TCRa and b genes was carried out by multiplex nested RT-PCR and sequencing of a and b products as described previously (37). Both external and internal rounds of PCR included 40 TRAV and 27 TRBV forward primers, and a TRAC and TRBC reverse primer, as detailed elsewhere (37). Sequences were analyzed according to the ImMunoGeneTics/V-QUEry and STandardization web-based tool (38). All TCR nomenclature was according to Folch et al. (39). CDR3 amino acid sequences described within the text start from CDR3position 3, which is equivalent to amino acid position 107 of the TRAV and TRBV segments, and end at TRAJ-position 10 or TRBJ-position 6.

Immunopeptidome Analysis
The C1R.B*15:02 cell line was in vitro expanded in RF10, and treatment of cells with either CBZ (Sigma), ECBZ (Sigma or SYNthesis med chem, Australia) or OXC (SYNthesis med chem) for peptide elutions was performed at 25 mg/mL for 48 hours in roller bottles prior to harvesting. Cells (0.9-1.1 x 10 9 ) were pelleted, washed twice in PBS and snap-frozen in liquid nitrogen. Cells were lysed, and the HLA class I isolated by immunoaffinity purification using solid-phase bound pan HLA class I antibody W6/32 as described previously (42). Peptides were dissociated using 10% acetic acid and separated from the HLA heavy and light (beta-2 microglobulin; b 2 m) chains by Reversed Phase HPLC (RP-HPLC) as described (43), monitoring elution by 215nm absorbance and collecting 500 mL fractions. The retention times of CBZ, ECBZ and OXC were determined through subjecting each molecule to the same RP-HPLC protocol. Peptide containing fractions, avoiding regions containing b 2 m and heavy chain, were vacuum concentrated and concatenated to generate 13 pools, including 3 pools aligned with the retention times of CBZ (pool 10), ECBZ (pool 12) and OXC (pool 13). Pools were vacuum concentrated to remove residual acetonitrile (ACN) and reconstituted in 15 mL 2% ACN, 0.1% formic acid (FA), spiked with 250 fmol iRT peptides (44).
Pools were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) using a data dependent acquisition (DDA) strategy on a Q-Exactive Plus Hybrid Quadrupole Orbitrap (Thermo Fisher Scientific) utilizing a Dionex UltiMate 3000 RSLCnano system (Thermo Fisher Scientific). 5-6 µL of concentrated material was loaded onto an Acclaim PepMap 100 (100 µm x 2 cm, nanoViper, C18, 5 µm, 100Å; Thermo Scientific) in 2% ACN, 0.1% FA at a flow rate of 15 µL/min. Peptides were separated over an Acclaim PepMap RSLC (75 µm x 50 cm, nanoViper, C18, 2 µm, 100Å; Thermo Scientific) at 250 nL/min using a gradient of Buffer A (0.1% FA) and Buffer B (80% ACN, 0.1% FA) as follows: 2.5-7.5% B in 1 min, 7.5-32.5% B in 55 min, 32.5-40% B in 5 min, 40-99% B in 5 min, 99% B for 6 minutes, and 99-2.5% B in 1 min, prior to reequilibration at 2.5% B for 20 min. Data  To characterize the HLA-B*15:02 binding motif, peptide sequences identified at a 5% peptide FDR in control data sets representing the endogenous HLA class I and II of C1R cells were removed. So too were peptides assigned without a protein accession or mapping exclusively to proteins in the contaminant database. For peptide counts, motif and length distribution analysis, only 7-20mer peptides were considered and peptides sequences containing deamidations or cysteinylations were treated as distinct from the native sequence. Figures were generated using Prism 9.0, GraphPad software (San Diego, CA). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (45) partner repository with the dataset identifier PXD023545 and 10.6019/PXD023545.

Identification of Co-Purified Drugs/Metabolites
Drugs/metabolites were detected using a targeted LC-multiple reaction monitoring-MS (LC-MRM-MS) approach on a SCIEX QTRAP ® 6500 plus mass spectrometer, coupled to an Ekspert ™ nanoLC 415. For pools aligned with the RP-HPLC retention time of CBZ (pool 10), ECBZ (pool 12) and OXC (pool 13), residual sample post-peptide analysis was diluted by a factor of~2. 6 mL diluted sample was loaded onto a NanoLC Trap ChromXP C18 column (350 mm x 0.5 mm, 3 µm particle size, 120 Å pore size) in 2% ACN, 0.1% FA at a flow rate of 5 µL/min for 5 minutes, prior to separation over a ChromXP nanoLC C18 column (75 µm x 15 cm, 3 µm particle size, 120 Å pore size) at 300 nL/min using an increasing gradient of Buffer B ( (CE 25). CE values were optimized based on injection of solubilized drug/metabolite. Three transitions for each iRT peptide were also monitored. Drug peaks areas are the sum peak areas of the monitored transitions. Drug peak areas were normalized to the sum peak areas of iRT peptides B-G (GAGSSEPVTGLDAK, VEATFGVDESNAK, YILAGVENSK, TPVISGGPYEYR, TPVITGAPYEYR, DGLDAASYYAPVR). Figures were generated using Prism 9.0, GraphPad software.

Statistical Analysis
All data were reported as mean ± standard error of the mean (SEM), unless stated otherwise. Statistical significance was determined by nonparametric one-way ANOVA with post-hoc Tukey's multiple comparison test, Mann-Whitney test or multiple comparisons using Holm-Sidak method (Prism 8.0, GraphPad software) with *p<0.05, **p<0.005 and ****p<0.0001. Standard error of the difference between mean amino acid prevalence was calculated using the Welch t-test (Prism 9.0, GraphPad software).

Drug-Induced Recall Responses Are Restricted to CD8 + T Cells in Recovered SJS or TEN Patients
We examined the immune reactivity profiles of convalescent SJS or TEN patients (7 cases and one drug-tolerant case) to determine whether they maintain a memory pool of T cells specific to the culprit drug that can be reactivated following subsequent drug exposures ( Table 1). PBMCs treated with 25 mg/mL CBZ were expanded in vitro for 14 days. Outgrown CBZinduced T cell lines were then restimulated in a 6-hour ICS assay with either no drug (untreated), CBZ (25 mg/mL), or with HLA-B*15:02 + APCs C1R.B*15:02 or C1R.B*15:02+CBZ (25 mg/mL) to measure activation via production of Th1 cytokines (IFNg and TNF). Group data of the recovered patients showed that the CBZ-reactive T cell response, where activation was observed following CBZ and APC+CBZ stimulation, was primarily restricted to CD8 + T cells. However, the CD4 + T cells did show immune recognition of CBZ alone, likely via the mechanism of self-T cell presentation (i.e. T-T presentation). Here, 6 out of 7 patients expressing HLA-B*15:02 demonstrated non-specific CD4 + T cell responses towards the APCs most likely due to HLA class II mismatches expressed by the C1R.B*15:02 transfectant ( Figure 1A). Optimization experiments, using healthy donors, showed similar CD4 + T cell immune responses (i.e. production of Th1 cytokines IFNg and TNFa) to both the C1R parental and C1R.B*15:02 APC in either the presence or absence of CBZ (data not shown). Dissection of the individual CD8 + T cell responses profiled either by single or dual expression of IFNg and TNF demonstrated significantly higher responses were driven by IFNg production (p-value <0.05; Mann-Whitney test; Figure 1B). As expected, the drug-tolerant case did not respond to CBZ (red dot; Figure 1B).
Priming of Drug-Naïve HLA-B*15:02 + Individuals Required Multiple Rounds of Drug Exposure To demonstrate both the kinetics and magnitude of T cell activation following drug exposure we in vitro stimulated HLA-B*15:02 + drug-naïve PBMCs (n=7; Table 1) with either CBZ or its metabolite ECBZ at 25 mg/mL. On days 13 and 27, outgrown T cells were restimulated with drug (as per day 0) for continued in vitro expansion. Drug-induced T cell lines were tested prior to each restimulation and at day 40 (3 intervals total: day 13, 27 and 40) in a 6-hour ICS assay with either no drug (untreated), drug (CBZ or ECBZ; 25 mg/mL), or using C1R.B*15:02 or C1R.B*15:02+drug (CBZ or ECBZ; 25 mg/mL) as APC with activation quantitated by either IFNg and/or TNF production. For CBZ-induced T cells, we observed similar findings to resolved cases ( Figure 1) but with delayed kinetics as drug-specificity predominantly mediated by CD8 + T cells that was more pronounced by day 40 (Figure 2A). For ECBZinduced T cells, the greatest magnitude of immune recognition was 2.7-fold lower than the CBZ parent drug on day 40 for IFNg producing cells (C1R.B*15:02+drug; mean ± SEM: 14.28% ± 4.6 for CBZ and 5.16% ± 1.1 for ECBZ; p=0.0029 multiple comparisons using Holm-Sidak method) ( Figure 2B).

CBZ-Reactive T Cell Cross-Reactivity Towards Tricyclic Aromatics
A previous study examining active cases of HLA-B*15:02positive SJS (and one SJS/TEN) and CBZ-tolerant controls showed CBZ-induced CD8 + T cell cross-reactivity towards other tricyclic aromatic compounds (21). We wanted to determine whether T cell cross-reactivity can be achieved following reactivation of drug-induced T cells in our cohort of resolved HLA-B*15:02-positive SJS or TEN patients and a drug-tolerant control, where PBMCs were collected between 3to-20 years post-reaction (Table 1). Here, in vitro expanded CBZ-induced T cell lines (generated for Figure 1) were restimulated in a 6-hour ICS assay with media alone (untreated), CBZ (25 mg/mL) or C1R.B*15:02 APCs in the absence or presence of drug (CBZ, ECBZ, OXC, PHT; 25 mg/ mL) and T cell activation measured via IFNg production, with statistical significance only observed between untreated and C1R.B*15:02+CBZ (p=0.111, one-way ANOVA with post-hoc Tukey's multiple comparison test). We show that CD8 + T cells primed against the parent drug CBZ can cross-react towards other ASMs following restimulation with C1R.B*15:02+drug. As expected, restimulation in the presence of CBZ+APC generated the highest reactivity (mean ± SEM; 6.771 ± 2.220), followed by cross-reactivity towards OXC (3.014 ± 0.831) and ECBZ (2.943 ± 0.496) (Figures 3A, B). Whilst, minimal CD8 + T cell crossreactivity was observed towards the aromatic drug PHT (1.657 ± 0.561), reflecting the more chemically diverse structure of PHT compared to the other more closely related compounds to CBZ ( Figure 3C).

CD8 + T Cells That Respond to CBZ Exhibit a High Degree of TCR Clonality
To evaluate whether the TCR repertoire of activated CD8 + T cells was remodeled following CBZ culture, we analyzed in vitro expanded HLA-B*15:02-positive PBMCs from two drug-naïve donors (AP022, AP026), five CBZ-induced SJS cases (T00016, T00024, E10076, E10314, E10630) and one CBZ-induced TEN case (E10056) ( Table 1). CBZ-reactive T cell lines established from each individual were restimulated for 4-hours with A B FIGURE 1 | Restimulation of drug-reactive memory cells in recovered CBZ-induced SJS and TEN patients is restricted to CD8 + T cells. PBMCs were co-cultured in the presence of 25 mg/mL CBZ for 14 days. T cell subset phenotype (CD4 + or CD8 + ) and activation status (production of either IFNg and/or TNFa) were measured following a 6-hour restimulation with either no drug (untreated), CBZ (25 mg/mL), C1R.B*15:02 or C1R.B*15:02+CBZ (25 mg/mL). Data was acquired on LSRII flow cytometer (BD) and analyzed using FlowJo software (version 10, BD). All data is expressed as mean ± SEM. (A) Grouped data demonstrated a drug-reactive response in CD8 + T cells expressing the different combinations of proinflammatory cytokines, but this was not observed for CD4 + T cells (n=7, single data).   Table 2. We selected three CBZ-reactive CD8 + TCRs for further functional validation to understand antigen processing requirements for T cell stimulation ( Figure 4, Table 2).

Clonotypic TCRs Exhibit Cross-Reactivity Towards CBZ Metabolites and Related ASMs
Retroviral transduction of selected CBZ-reactive CD8 + TCRs into the SKW3 reporter cell line enabled us to conduct cellular investigations of TCR recognition to confirm drug specificity without confounding background T-T presentation (SKW3 reporter cell lines are HLA-B15-negative). Here, TCR activation was determined by cell surface upregulation of CD69, with the shift in geometric MFI being measured by flow cytometry ( Figure 5A). We next explored whether HLA    Figure 5B, middle panels). Notably, none of the TCRs examined displayed responses to any of the drugs in the presence of the parental C1R cell line, which lacks HLA-A and -B expression. An irrelevant TCR control, SKW3.LC13, was also examined to demonstrate TCR specific recognition of C1R.B*08:01/FLR and non-activation by drug in the presence of C1R.B*08:01 or C1R.B*15:02 APCs ( Figure 5B, lower panel).
To determine whether TCR recognition of the non-risk HLA allotype B*15:01 was drug dose dependent, the drug was titrated 4-fold from 25 mg/mL down to 3.125 mg/mL in the presence of APCs. For both SKW3.AP026 and SKW3.E10630, as anticipated TCR activation reduced in accordance with decreasing concentrations of drug. However, this effect was comparable in both the risk and non-risk HLA allotypes, B*15:02 and B*15:01, respectively. Whilst SKW3.E0056 showed remarkable specificity and sensitivity towards CBZ and decreasing responses towards the less immunogenic ECBZ, but only when presented by B*15:02-expressing APC ( Figure 5C).

SKW3.TCR Reporter Cells Are Activated via a Non-Covalent Drug Interaction
We explored whether the SKW3.TCR cells were activated by the same non-covalent drug interactions as observed for in vitro expanded CBZ-induced T cell lines. Here, we tested the TCR transduced reporter cells capacity for activation when both C1R.B*15:02 and drug were present throughout the assay (drug addition), as well as when C1R.B*15:02 was pulsed overnight with drug and then washed prior to co-incubation. Both SKW3.AP026 (drug-naïve healthy control) and SKW3.E10630 (SJS patient) cell lines showed a significant decrease in TCR activation for C1R.B*15:02 drug-pulsed, compared to C1R.B*15:02 drug addition, across all three drugs tested (CBZ, ECBZ, OXC; p<0.0001 ANOVA with Tukey correction) ( Figure 6A). These observations mirrored T cell line data ( Figures 1B, 2A) demonstrating a labile drug-HLA-B*15 interaction in keeping with non-covalent binding.

SKW3.TCR Reporter Cells Do Not Require Intracellular Antigen Processing for Activation
We measured whether active antigen processing in APCs for de novo generation of peptide/HLA complexes was required for SKW3.TCR activation. Following fixation of C1R.B*15:02 with 1% paraformaldehyde, TCR activation was significantly reduced, compared to non-fixed C1R.B*15:02, but a notable drug-specific response was still observed for all three SKW3.TCR cell lines ( Figure 6B, p<0.0001 ANOVA with Tukey correction). These data align with the nature of interaction between TCR/drug/HLA molecule not requiring active antigen processing pathways. Finally, we investigated whether the drug itself binds stably to the TCR. Here SKW3.E10056 and SKW3.E10630 cells were either untreated or pulsed with drug (and washed to remove unbound drug) prior to stimulation with C1R.B*15:02 in either the absence or presence of drug. Here, maximal TCR activation was only observed when TCR/drug/C1R.B*15:02 were present throughout the assay. Thus, potential TCR/drug interactions alone were assumed to lack the stability to be maintained without either maintenance of the drug in solution or stabilization by the HLA present on APC ( Figure 6C, p<0.0001 and p<0.005 ANOVA with Tukey correction).
Minimal Impact of CBZ/ECBZ/OXC on Peptide Binding Specificity of HLA-B*15:02 Previous analyses of the ligandome of HLA-B*15:02 from CBZ treated cells have observed co-purification of CBZ (22,47). Furthermore, we previously described a minor alteration of the immunopeptidome reflected in conservation of anchor residue biases (P2, P9) of co-purified peptides, and modulation of residues across non-anchor positions. This earlier observation led us to postulate drug binding at a central position in the antigen-binding cleft (22). In contrast the metabolite ECBZ has been suggested to alter peptide binding to soluble HLA-B*15:02, through interaction in the region of the F-pocket (23). Given the responses of CBZ-reactive T cells to CBZ, ECBZ and OXC we sought to determine whether we could define common changes in peptide binding induced by these three small molecules that might shed light on the interaction. As previously, we utilized the C1R.B*15:02 cell line due to its demonstrated ability to present these molecules in an immunogenic fashion (Figures 5, 6). Membrane-bound HLA molecules were extracted from 0.9-1.1 x 10 9 cells through non-denaturing lysis and immunoaffinity purification, prior to acid elution of peptides, fractionation by RP-HPLC and LC-MS/MS analysis. Experiments were performed in duplicate for each of untreated and CBZ treated (25 mg/mL) cells, whilst single experiments for ECBZ and OXC treatment (25 mg/mL) conditions were performed to determine if ECBZ and OXC recapitulated observations for CBZ treatment.
As CBZ, ECBZ and OXC were observed to have distinct chromatographic retention times during RP-HPLC, aligned with fraction pools 10 (CBZ), 12 (ECBZ) and 13 (OXC), these specific pools were also analyzed by multiple reaction monitoring (MRM) to determine whether drug had co-purified with the HLA-B*15:02 during immunoaffinity purification. As anticipated, clear signal for CBZ was observed in pool 10 for CBZ treated samples at 36 minutes, for ECBZ in pool 12 of ECBZ treated samples at 28 minutes, and for OXC in pool 13 for OXC treated samples at 30 minutes (Supplementary Figure 3). Figure 3B vs 3C). In the ECBZ sample, a second peak was also observed at 21 minutes (Supplementary Figure 3B). Evolution of this second, earlier, peak was also seen over increased time in the autosampler for ECBZ prepared as a comparator. Subsequent analysis suggested that ECBZ is unstable over time in the loading conditions used for LC-MS analysis (2% ACN, 0.1% FA), with an increase in signal for a mass consistent with the hydrolysis product (10,11dihydroxycarbamazepine, m/z = 271.1+) co-eluting with the earlier peak (data not shown). It is therefore hypothesized that the earlier peak is 10,11-dihydroxycarbamazepine, some of which undergoes dehydration during electrospray ionization to regenerate ECBZ.

Despite identical masses and monitored transitions, OXC and ECBZ are distinguished by distinct retention times and transition hierarchies (Supplementary
Analysis of the eluted peptides identified more than 9900 peptides of 7-20 residues per data set ( Figure 7A) as described in the Materials and Methods. Over 24,000 peptides were identified in total, with more than 7,000 peptides identified in all treatments ( Figure 7B). Peptides were predominantly 9 amino acids in length, which were heavily biased towards Leu (~30%)>Val (~16%)>Gln (~14%) at position 2 of the peptide, and Tyr (~48%)>Phe (~24%)>Met (~15%)>Leu (~10%) at position 9 ( Figure 7C). Proline was observed at P4 and P5 in approximately 12% and 10% of peptides respectively, whilst Ser, Thr and Val were often seen in positions 5-8 of the peptide ( Figure 7C). Only minor differences in amino acid prevalence were observed between individual treatments and the untreated control, and no common perturbation of residue preference was noted for the three molecules when considering the global immunopeptidome as might be expected from their crossreactivity (Supplementary Figure 4). Due to previous reports of increased prevalence of deamidated peptides bound to HLA-B*15:02 after treatment with ECBZ, particularly at P4 (23), we considered deamidated Asn and Gln separately from their native form in motif analysis. However, deamidated peptides were observed to be rare in our analysis (Supplementary Figure 5). We previously observed minor perturbations of non-anchor residues when considering peptides uniquely identified in HLA-I purifications of CBZ treated C1R.B*15:02 as compared to untreated cells, in analyses based on a smaller dataset (<2000 peptides/condition) (22). The current analysis incorporates more in the presence of drug throughout the assay (drug addition) or drug-pulsed overnight and then washed. Cells were then stained and data was acquired by flow cytometry (triplicate data, mean ± SEM). A significant decrease was observed when SKW3.TCR cells were stimulated with drug-pulsed C1R.B*15:02 (****p<0.0001; ANOVA with Tukey correction). (B) SKW3.TCR cells (AP026, E10056, E10630) were stimulated with either untreated or fixed C1R.B*15:02 for 17-20 hours. Cells were then stained and data was acquired by flow cytometry (triplicate data, mean ± SEM). Whilst a significant reduction in TCR activation was shown, a drugreactive response was still observed (****p<0.0001; ANOVA with Tukey correction). (C) SKW3.E10056 and SKW3.E10630 cells were either untreated or pulsed with drug prior to stimulation with C1R.B*15:02 ± drug. TCR/drug interactions alone were unable to induce CD69 upregulation (****p<0.0001 and **p<0.005; ANOVA with Tukey correction). than 9900 peptides (>6000 nonameric peptides) per experiment, and more than 18,000 peptides (>11,000 nonamers) per untreated and CBZ treated conditions when replicates are combined. We therefore again determined the peptides that were common (identified in at least 1 replicate of each condition) and unique (identified in at least 1 replicate of the condition, but never in the compared condition). We identified more than 14,000 peptides (>9000 nonamers) common to untreated and CBZ treatment conditions, with 4246 (2126 nonamers) and 3871 (2129 nonamers) peptides identified in untreated only (untreated unique) and CBZ treatment only (CBZ unique), respectively. Focusing on the nonamer peptides, as anticipated we saw similar anchor preferences to the whole data set analysis. As observed previously, there were some variations across the backbone of the peptide ( Supplementary  Figures 6), but few were recapitulated in similar analyses of ECBZ and OXC compared to untreated cells ( Supplementary  Figures 7, 8), and overall only 2738 (1211 nonamers) were only identified in the untreated condition (when compared to all drug treatment conditions) ( Figure 7B). It is therefore difficult to discern a compelling common drug induced motif across the three treatments, although a slight reduction in acidic residues at P1 and tyrosine at P3 were commonly observed ( Supplementary  Figures 6-8), as noted previously for CBZ (22). A subtle increase in Pro at P2 was also observed in peptides uniquely identified in drug treatments, although Leu remained the dominant anchor residue.

DISCUSSION
Adaptive immune recognition of a foreign entity by T cells relies on two signals; (1) interaction between an immunogenic antigen/ HLA complex and the TCR and (2) co-stimulation provided by engagement of B7 (i.e. CD80/CD86) on APCs and CD28 expressed on T cells. The ensuing response drives expansion and proliferation of T cells containing an armory of effector functions, including pro-inflammatory cytokine production and release of cytotoxic molecules, that eliminate pathogen-infected or neoplastic cells. Once the elimination process is completed, attrition of effector T cells leads to the establishment of a memory T cell pool to combat secondary encounters with the same immunogenic antigen/HLA complex which lacks the necessity for signal 2 to allow rapid deployment. Whilst T cell-mediated immune responses are critical in controlling initial and subsequent challenges, responses against innocuous stimuli sometimes occur and are associated with autoimmunity and allergy, including DHRs. The impacts of ASM-induced T cell-mediated DHRs have been well reported [comprehensively reviewed (6,48)]. Key insights have been drawn from investigation of three main cohorts; ASM-induced DHRs [spanning mild (MPE), moderate (DRESS) and severe (SJS and/or TEN) cases], drug tolerant patients and drug-naïve healthy donors. Comparison between these cohorts has provided critical information regarding the culprit ASM (e.g. CBZ, OXC, PHT, lamotrigine) as well as the association of HLA allotype (e.g. B*15:02, A*31:01) and ethnicity (e.g. Han Chinese ancestry) on DHR outcomes. Importantly, immunopathological examinations during the active phase of the hypersensitivity reaction has demonstrated a dominant role for CD8 + T cells and the involvement of key cytolytic (e.g. granulysin, perforin, granzyme) and proinflammatory cytokines (e.g. IFNg, macrophage inflammatory protein-1a) (21,28,31,(49)(50)(51)(52).
In this study, we examined long-lived T cell reactivity profiles of resolved CBZ-induced SJS (and one TEN) patients, ranging from 3 to 20 years post-active disease. In particular, we were interested in determining whether CBZ-reactive memory T cells activated upon re-exposure to drug showed similar profiles to those reported in active disease, considering effector functions, TCR repertoire clonality, mode of drug recognition and crossreactivity towards other ASMs.
CBZ-reactive T cells derived from SJS or TEN patient PBMCs were expanded for 2 weeks in vitro following a single round of CBZ exposure. The outgrown bulk T cell culture was then restimulated with drug either in the absence or presence of an HLA-B*15:02-expressing APC, and immunophenotypic properties of CBZ-reactive effector T cells measured by flow cytometry. We showed that these restimulated T cells were primarily restricted to the CD8 + T cell lineage and were capable of Th1 pro-inflammatory cytokine production (IFNg > TNF), aligning with reported studies of SJS/TEN patients during active disease (21,28). For drug-naïve healthy donors, 2-3 rounds of drug exposure (CBZ or ECBZ) were required for in vitro priming and expansion of drug-reactive T cells. Whilst, CBZ-and ECBZ-reactive CD8 + T cells were observed by day 40 in all individuals, the ECBZ metabolite had lower levels of immune responsiveness (with IFNg > TNF) compared to the parent drug, CBZ.
A common complication of prescribed ASMs is that individuals who have experienced a DHR are at higher risk of developing the same, if not, a more severe form of DHR with an alternate ASM. Indeed, structural similarities exhibited by aromatic ASMs (i.e. CBZ, OXC, PHT, lamotrigine) provide a framework for cross-reactive T cell recognition that perpetuates secondary reactions in between 13-80% of DHR patients (53)(54)(55)(56). This observation is supported by cellular investigations in SJS/TEN patients with CBZ-induced T cells (generated following 4-5 rounds of in vitro culture) exhibiting cross-reactivity towards other aromatic compounds such as ECBZ, OXC, lamotrigine and eslicarbazepine (21). Whilst, non-aromatic ASMs (i.e. valproic acid and gabapentin) are not generally associated with DHRs (57). In this study, we demonstrated that CBZ-reactive T cells (generated following a single in vitro drug exposure) cross-react with HLA-B*15:02 + APCs in the presence of aromatic ASMs ECBZ and OXC. The majority of patients showed higher levels of cross-reactivity towards the ECBZ metabolite, albeit one individual (E10630) demonstrated greater reactivity towards OXC. These data suggest an immune hierarchy of T cell crossreactivity towards ASMs is likely driven by the common tricyclic ring structure between CBZ, ECBZ and OXC, but modulated by different side groups. These observations demonstrate that resolved SJS and TEN patients, many years post-DHR, can still vigorously respond to alternate aromatic ASMs, and that T cell immune monitoring should be undertaken when considering different treatment options.
We next examined the HLA-B*15:02-restricted TCR repertoire of CBZ-reactive CD8 + T cells in two drug-naïve individuals and six SJS or TEN patients. In most cases, a single major TCR clonotype was observed, which supports previous findings of focused TCR usage in SJS/TEN patients (28). However, most of these TCR repertoires showcased private specificities, with only two individuals (AP026 and E10630) sharing the same TRBV27 usage, but lacking common CDR3 regions. These findings diverge from previous reports of a more public TCR usage within the HLA-B*15:02-restricted CBZinduced TCR repertoire in SJS and/or TEN patients; that exhibited dominant Va and Vb clonotypes (28) and a public abTCR CDR3 sequence (24). Differences in TCR repertoire may also be attributed to either the site of origin for CBZ-reactive T cells [i.e. blister fluid/cells vs PBMCs [this study)] and/or collection of biospecimens during active or resolved (this study) disease states. It is also important to acknowledge reported biases associated with the in vitro expansion of T cell lines (29,58), with the likelihood of different TCR clonotypes being expressed by drug-naïve and drug-experienced individuals, as well as their localization at different sites of pathology (i.e. blister fluid vs PBMCs). In particular, the study by Pan et al. (24) could not recapitulate TCR clonotype findings reported by Ko et al. (28) in the PBMC and blister fluid of SJS/TEN patients. Although, we have previously shown in anti-viral TCRs that a bias was not induced between ex vivo PBMCs and day 14 in vitro T cell cultures (59).
Limited patient biospecimen availability makes it difficult to conduct a multitude of cellular assays, even with in vitro expansion of T cell lines and clones. Here we take advantage of a known reporter system utilizing abTCR deficient SKW3 cells (41,60) to examine drug-induced HLA-restricted T cell promiscuity and the mode of drug presentation. Single major TCR clonotypes, for one drug-naïve individual (AP026) and two patients (E10056, E10630), were cloned into the SKW3 reporter cell line. Firstly, we investigated whether these TCRs demonstrated HLA-B15 specificity towards SJS or TEN risk (B75 subgroup) and nonrisk (B62 subgroup) allotypes (21). Interestingly, we observed highly restricted TCR recognition for E10056 towards B75 subgroup members B*15:02 and B*15:21, whilst both AP026 and E10630 showed high degree of cross-reactivity towards risk and non-risk HLA-B*15 allotypes. This observation was apparent for all HLA-B15 APCs in the presence of CBZ, ECBZ or OXC in a dose dependent fashion. Furthermore, no responses were observed in the presence of HLA-B-negative C1R cells. These observations suggest that non-risk HLA-B15 molecules may also support the ligation of certain CBZ-reactive TCRs, implying the association of SJS or TEN with HLA-B*15:02, but not HLA-B*15:01, is not as simple as the ability of the drug to uniquely interact with one HLA allomorph but not another.
Secondly, we examined whether ASMs stimulated SKW3.TCR reporter cells using the same mechanisms required for T cell lines. As observed by others (21,49,52), we found it was essential to maintain soluble drug continuously within the assay for TCR activation, consistent with a non-covalent and labile interaction between the drug and peptide/HLA and/or TCR complex. Additionally, we showed that fixation of APCs did not abrogate drug-induced TCR activation, which aligns with the interaction occurring with pre-existing peptide/HLA complexes, hence not requiring additional antigen processing. Finally, we showed that TCR/drug interactions were not maintained on washing of the reporter cells at physiological pH suggesting any interactions lack sufficient stability to be maintained in the absence of the peptide/HLA complex. Whilst it has been suggested that binding can occur directly with a public CBZspecific abTCR in the absence of the peptide/HLA complex (24), we only observed reporter cell activation in the presence of drug treated HLA-B*15-positive APCs. Hence, the presence of peptide/HLA, and TCR are necessary for maximal functional coordination of the drug interaction. Collectively, these data validate the use of SKW3.TCR reporter cells as an immunological tool for DHR cellular investigations and confirm that non-covalent and labile drug/TCR/peptide/HLA interactions in the absence of new peptide/HLA generation underpin CBZ-induced DHRs.
In line with the inability of drug pulsed C1R.B*15:02 to stimulate reporter cell responses, modulation of the immunopeptidome by CBZ and its derivatives was subtle. This is distinct from the interaction of abacavir and HLA-B*57:01, where stable, non-covalent binding within the antigen binding cleft causes a perturbation of self-peptide presentation, initiated predominantly in the ER, and alteration of the PΩ anchor preference of abacavir occupied HLA-B*57:01 molecules (22,26,27). We previously observed subtle modulation of nonanchor residues of peptides bound to HLA-B*15:02 through comparison of a smaller dataset of HLA-B*15:02 ligands (<2000 peptides per condition). With the current larger dataset (>9900 peptides per condition) and comparison to cross-reactive molecules ECBZ and OXC not all changes were recapitulated, although a subtle reduction in acidic amino acids at P1 and tyrosine at P3 were consistent for peptides unique to drug treatments (22). This is not to say there are not HLA/peptide/ drug complexes formed but the mechanism for their formation clearly follows a different trajectory to that of HLA-B*57:01/ abacavir/peptide complexes. Interestingly, our data contrasted a recent report that ECBZ alters peptide presentation by soluble HLA-B*15:02 (23). Whether this is due to differences between soluble and membrane-bound HLA such as reduced interaction with the peptide loading complex, or the workflow used (e.g. the necessity for use of detergents to extract membrane-bound HLA, peptide elution conditions), which might impact complex stability, remains to be clarified.
As described previously, CBZ and ECBZ were detected in eluates from affinity purified HLA class I molecules of drug treated cells (22,23,47), an observation that was extended to OXC. This residual drug is maintained despite washing of the source cells, as well as the immunoaffinity purified HLA class I molecules, at physiological pH, a process shown to abrogate APC immunogenicity in functional experiments. Given the rapid titration of responses to CBZ, ECBZ and OXC, it is possible that whilst washing reduces the number of peptide/HLA/drug complexes below the threshold for activation, some level of binding is maintained. Regardless, failure to identify a distinct peptide motif in drug treated cells and the inability of either the peptide/HLA or TCR alone to maintain sufficient drug interaction for immunogenicity on removal of soluble drug, collectively suggest that CBZ and related compounds form a tripartite interaction with peptide/HLA/TCR. Furthermore, differences in recognition hierarchies of CBZ, ECBZ and OXC by the three TCRs investigated suggest the epoxide (ECBZ) and ketone (OXC) more prominently impact TCR ligation as opposed to interaction with the HLA, thus may be oriented towards the TCR.
In conclusion, this study confirms the presence of long-lived immunological effects in resolved CBZ-induced SJS and TEN patients, which are characterized by a highly clonal drug-reactive CD8 + TCR repertoire. Furthermore, using a combination of the definition of the functional requirements for TCR engagement using transduced reporter cell lines and detailed study of the impact of CBZ, ECBZ and OXC on HLA-B*15:02 peptide presentation, our work supports the hypothesis that peptide/ HLA and TCR are required for drug interactions able to elicit TCR activation.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ebi.ac. uk/pride/archive/, PXD023545.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Joint Chinese University of Hong Kong-New Territories East Cluster Clinical Research Ethics Committee   Supplementary Figure 6 | Amino acid prevalence for 9mer peptides unique to CBZ treatment when compared to control conditions. Combined lists of peptides identified in at least 1 replicate of the control condition (untreated) and CBZ treatment were generated and compared to generate 3 categories; untreated uniqueidentified in at least one replicate of the untreated control, but not in either replicate of CBZ treatment, commonidentified in at least one replicate of the untreated control and at least one replicate of the CBZ treatment, and CBZ unique -identified in neither replicate of the untreated control, but in at least one replicate of CBZ treatment. Bar graphs show the percentage of 9mer peptides possessing a given amino acid at each position of the peptide per category. n is the number of 9mer peptides in the category. Arrows indicate residues discussed in the text. Figure 7 | Amino acid prevalence for 9mer peptides unique to ECBZ treatment when compared to control conditions. A combined list of peptides identified in at least 1 replicate of the control condition (untreated) was generated and compared to the ECBZ treatment data set to generate 3 categories; untreated uniqueidentified in at least one replicate of the untreated control, but not in ECBZ treatment, commonidentified in at least one replicate of the untreated control and in the ECBZ treatment, and ECBZ uniqueidentified in neither replicate of the untreated control, but in ECBZ treatment. Bar graphs show the percentage of 9mer peptides possessing a given amino acid at each position of the peptide per category. n is the number of 9mer peptides in the category. Arrows indicate residues discussed in the text.

Supplementary
Supplementary Figure 8 | Amino acid prevalence for 9mer peptides unique to OXC treatment when compared to control conditions. A combined list of peptides identified in at least 1 replicate of the control condition (untreated) was generated and compared to the OXC treatment data set to generate 3 categories; untreated uniqueidentified in at least one replicate of the untreated control, but not in OXC treatment, commonidentified in at least one replicate of the untreated control and in the OXC treatment, and OXC uniqueidentified in neither replicate of the untreated control, but in OXC treatment. Bar graphs show the percentage of 9mer peptides possessing a given amino acid at each position of the peptide per category. n is the number of 9mer peptides in the category. Arrows indicate residues discussed in the text.