Coupling Genotyping and Computational Modeling in Prediction of Antiepileptic Drugs that cause Stevens Johnson Syndrome and Toxic Epidermal Necrolysis for Carrier of HLA-B*15:02

– Purpose. The importance of HLA-B*15:02 genotyping to avoid carbamazepine induced SJS/TEN and molecular modeling to predict the role of HLA-B*15:0 and AEDs induced SJS/TEN are investigated. Methods . DNA was extracted from eighty-six patients. The patients were genotyped by AS-PCR. Computational modeling of the HLA-B*15:02 followed by docking studies were performed to screen 26 AEDs that may induce ADR among HLA-B*15:02 carriers. Results. Odd ratio for CBZ induced SJS/TEN and HLA-B*15:02 was 609.0 (95% CI: 23-15873; p =0.0002). Molecular modeling studies showed that acetazolamide, ethosuxiamide, lamotrigine, oxcarbazepine, phenobarbital, phenytoin, primidone and sodium-valproate may induce ADR in HLA-B*15:02 carriers alike CBZ. Conclusion. We confirmed HLA-B*15:02 as a predictor of SJS/TEN and recommend pre-screening. Computational prediction of DIHR is useful in personalized medicine.


INTRODUCTION
Drug-induced hypersensitivity reactions (DIHR) such as Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN) have resulted in high mortality and lowered the quality of life of patients. The most commonly studied example is carbamazepine (CBZ) induced SJS/ TEN. The pathophysiological mechanisms underlying these two fatal cutaneous drug reactions, however, are not well understood.
Several hypotheses have been established explaining how CBZ induces SJS and TEN. Oxidative metabolites such as epoxide were thought to be the culprit of serious cutaneous reactions [1]. Many studies have suggested that CBZ might affect the antioxidant capacity of an individual but the mechanism by which CBZ affects the oxidantantioxidant balance of an individual and its association with adverse drug reaction (ADR) remained unclear [2][3][4]. However, an immune pathogenesis is most likely as drug-specific T cells had been identified in patients who experience SJS/TEN [5]. With reference to the hapten/prohapten hypothesis, the active chemical molecule forms a covalent bond with an endogenous peptide, which is intracellularly processed and presented by HLA [6]. The p-i concept (direct pharmacologic interaction of a drug with immune receptors) implies that the direct and reversible interaction of drugs with TCRs or peptide-loaded HLA initiates immune reactions [7]. Nevertheless, there is very few evidence that prove these hypotheses. Several theories postulate the drug-driven antigenic stimulation of immunologic receptors in drug induced hypersensitivity. Studies on HLA class I alpha (α) chain refolding assay showed the direct interaction of eight synthesized nanomer peptides with HLA class I α chains. _________________________________________ Out of the eight nanomer peptides, B27 (B*2705) peptide was derived from binding heat shock protein peptide HSP89 α (201-209) and the other seven were derived from the human tumor suppressor protein p53 sequence. Testing a total of 36 allospecific HLA class-I protein have shown that the majority of these peptides possess Arginine-Lysine (ARG-LYS) motif that firmly binds to α chains of B27 but binds weakly to A31 α chains. Whereas, the peptides of p53 (173-181) VRRCPHHER and (334-342) GRERFEMFR with the ARG-ARG motif showed a considerable binding affinity to A31 α chains but not with other α chains. This depicts both the specificity and the cross-peptide-binding of peptides to HLA allelic products of different HLA loci [8]. A previous study has shown that CBZ binds to HLA-B*15:02 that results in an alteration in the repertoire of the presented self-peptides [9].
Involvement of the major histocompatibility complex for CBZ induced cutaneous reaction has been known ; 100% of Han Chinese patients with carbamazepine-induced SJS/TEN were positive for human leukocyte antigen HLA-B*15:02 allele, compared with 8% in the healthy population [10,11]. These DIHRs are significantly correlated with HLA-B*15:02 polymorphism which is common in Thailand, India, Malaysia, Hong Kong, and in descendants of immigrants from Southeast Asia [12].
A meta-analysis performed by Tangamornsuksan et al. reported that in Thailand, Singapore and Malaysia, usage of carbamazepine represents up to one-third of all SJS and TEN cases [13].The firm association between human leukocyte antigen HLA-B*15:02 and carbamazepine-induced SJS in Asians is however not seen among the Europeans [14]. Similarly, some studies among the Japanese reported lack of association with HLA-B*15:02 [15,16]. Evidence from the published literature supporting this relationship was summarised, and recommendations for the use of carbamazepine based on HLA-B genotype was provided by The PharmGKB (http://www.pharmgkb.org) which is a publicly available Web-based knowledge base [17]. The guideline suggested that carbamazepine should not be recommended for carbamazepine-naive individuals who have at least one copy of the HLA-B*15:02 allele. If a patient has previously used carbamazepine for longer than three months without incidence of cutaneous adverse reactions, consider cautious use of carbamazepine if no alternative is available. Besides, , cautious use of alternative AEDs is recommended, but there is a lack of evidence linking serious cutaneous ADR induced by alternative AEDs with HLA-B*15:02 allele.
In this study, the association of HLA-B*15:02 and CBZ induced SJS/TEN is investigated using the in-house developed PCR-based detection method. Two different perspective groups of patients were genotyped; a retrospective group with patients prescribed CBZ and developed SJS/TEN and another prospective group in which patients were genotyped before CBZ therapy was started. The retrospective group allows the odd ratios to be calculated while the prospective group provide insights to the importance of pre-genotyping among patients. In addition, we have developed three dimensional structural models for HLA-B*15:02 using homology modeling approach and endogenous peptide interaction to the binding site of the above model was done through peptide docking studies. Subsequently, docking studies were used to screen the AEDs that may cause drug induced hypersensitivity similar to CBZ, which are associated with HLA-B*15:02 allele. Our study implies the hapten/prohapten hypothesis, the chemical active molecule forms a covalent bond with an endogenous peptide, which is intracellularly processed and presented by HLA to induce ADR. The data generated will add more confidence to the local practitioners on the impact of pharmacogenomics based clinical practises.

Subjects and sample collection
Approval from the constitutional Human Research Ethics Committee and Medical Research and Ethics Committee (MREC) was obtained. Eighty six (86) patients attending clinics at the Department of Neurology, in a local hospital were recruited and written informed consents were obtained from all patients. This is a single masked (Single-blind) study where the patients were randomly selected by the medical officers attending patients at the Department of Neurology, based on the inclusion and exclusion criteria. The technician who performed the pharmacogenetic test did not know the clinical status of the patients involved in this study. Patients were recruited if they were 18-55 years old, mentally healthy, on CBZ treatment for bipolar, neurologic pain or epilepsy at the department of Neurology, understand the study protocol and able to provide informed consent. Patients who did not understand the study protocol, unable to follow simple instructions or were not willing to provide signed informed consent were excluded. Subjects in the study were informed of the details of the study before giving their consent. Blood (5 ml) was taken from the patient by authorized trained personnel and DNA was extracted from the blood using the salting out method as described earlier [10]. Relevant clinical data was collected by the medical officers and revealed after the test was obtained.
The patients were categorized into prospective and retrospective groups. Among the 86 patients, 40 were newly registered patients and were genotyped before the patients being prescribed CBZ (prospective).The other 46 patients who were already receiving CBZ at the time of entry into the study were randomly selected by the doctor for HLAB*15:02 screening and they comprised the retrospective study group. The event of SJS/TEN was analyzed in relation to HLA-B*15:02 genotype.

HLA genotyping
HLA-B*15:02 genotype was determined by polymerase chain reaction (PCR) using sequence specific primers. The PCR based detection method of HLA-B*15:02 allele was adopted and modified from Man et al [12]. For each PCR run, both positive and negative controls were included as quality assurance as well as to avoid misinterpretation of the result. The positive control (DNA fragment containing the four nucleotide change denoting HLA-B*15:02was inserted into a plasmid. DNA samples were then randomly selected and sent to a third party for direct sequencing to verify the accuracy of the method.

STATISTICAL ANALYSIS
The frequencies of HLA-B*15:02 was calculated. Statistical analysis (Fisher's Exact Test and Chisquare test) was performed using SPSS software version 20. P values ≤ 0.05 were considered statistically significant. The strength of association between HLA-B*15:02 with CBZ-induced SJS-TEN was estimated by calculating the odds ratio.
Computational studies Homology modeling of HLA-B*15:02 structure The protein sequence for HLA-B*15:02 (accession no: C5J3Z8) was retrieved from the universal protein resource (UniProt), consisting of 362 amino acids [18]. The template for homology modeling was identified by PDB-BLAST (http://www.ncbi.nlm.nih.gov/blast/) available at the National Center for Biotechnology Information (NCBI). The HLA-B*15:01 structure (PDB ID: 1XR8 at 2.30 Å resolution) [19] with a sequence identity of 98.19 % to HLA-B*15:02 was identified as a template. Based on the template identified, the homology structure for HLA-B*15:02 was modeled using MODELLER9v11 [20]. MODELLER can simultaneously incorporate structural information from one or more reference proteins. Structural topologies in the reference proteins were used to derive spatial restraints, which are used to build model protein structures by using conjugate gradient and simulated annealing optimization procedures. Among the models generated by MODELLER, the best model was selected based on the lowest generated discrete optimized protein energy (DOPE) score.
Further, the quality and stereochemistry of the models were evaluated using the program PROCHECK, where the final model was selected based on the stereochemical quality [21] and RAMPAGE Ramachandran plot (http://mordred.bioc.cam.ac.uk/~rapper/rampage.ph p). The main-chain conformations for all amino acid residues were checked for the favored or allowed regions of the Ramachandran plot.

Modeling of the peptide-loaded HLA-B*15:02 complex
In order to model the peptide pp5 (FLFDGSPTY) loaded antigen binding groove (B-pocket) of HLA-B*15:02, the peptide of human UbcH6 and Epstein -Barr virus EBNA-3 in complex with HLA-B*15:01 (PDB ID: 1XR8) was used as a reference [22]. Further, the modeled FLFDGSPTY peptide conformation in agreement with the co-crystallized peptide of HLA-B*15:01 was visually inspected. The FLFDGSPTY peptide loaded HLA-B*15:02 complex was further optimized using FlexPepDock, a high-resolution peptide docking protocol server [23,24]. The complex was optimized using Rosetta FlexPepDock as mentioned above and the best energy threshold complex model was selected in each case for further studies.
PatchDock tool [25] (utilizes PDB coordinate files as input with default parameters) was used to generate HLA-peptide loaded complex mentioned above, subsequently, the FireDock program was used to perform energy minimization of the best models to get stable HLA-peptide complex [26] as a validating process. Each candidate transformation was evaluated by scoring function considering both geometric fit and atomic desolvation energy [27]. The candidate solutions were clustered using Root Mean Square Deviation (RMSD), discarding the redundant solutions.

Docking studies on the peptide loaded complexes
The 3D structures of 26 AEDs reported by the Epilepsy Society of UK [28] were obtained from the NCBI PubChem database and their structures were optimized using Openeye software (OpenEye Scientific Software Inc., Santa Fe, NM, USA). Further, the AEDs were docked to peptide loaded complex of HLA-B*15:02 as a pre-screening process to select the possible AED candidates that may induce hypersensitivity similar to CBZ. The entire docking studies were performed using Gold (Genetic Optimization for Ligand Docking, version 5.1) [29], considering the antigen binding site (Bpocket) as the target site with a cut off of 12 Å radius with 100% genetic algorithm (GA), using ChemScore as the fitness scoring function. Results divergent by less than 0.75 Å in ligand-all atom root mean square deviations (RMSDs) were clustered together. Further, the binding mode of each AED candidate and their interaction patterns were analyzed by visual inspection using PyMOL [30]. During the analysis, it was prioritized that the binding mode of each AED candidate was compared with the binding mode of CBZ as a reference.

RESULTS
The clinical and demographic HLA-B*15:02 genotypes of the 86 patients (prospective and retrospective group) are shown in Table 1. HLA-B*15:02was observed in 24.4% of the patients. No significant difference was found in terms of sex, race and indication of CBZ when compared with patients who carry HLA-B*15:02 (p-values> 0.05).
Among the patients in the prospective group, 17.5% were positive for HLA-B*15:02. No significant differences were found in term of sex, race and indication of CBZ where the p-values from these groups are 0.273, 0.229, and 0.705 respectively.
In the retrospective group, 15 patients were of HLA-B*15:02 (32.6%) genotype. CBZ was withdrawn from 14of those patients due to SJS/TEN or rash and all were positive for HLA-B*15:02. One patient with positive HLA-B*15:02, however, did not develop SJS/TEN. No significant differences in the percentage frequency of HLA-B*15:02were found in the different ethnic groups or for different indications for CBZ (Table 1). However, nine patients who were detected as negative for HLA-B*15:02 developed mild rashes.
The occurrence of cADR among the patients in the retrospective group with respect to the genotype of HLA-B*15:02 is shown in Table 2. The strength of association, expressed as the odds ratio for the ADR in subjects of HLA-B*15:02 genotype relative to those who were not was 609.0 (95% CI: 23 -15873). p = 0.0002.

Homology modeling
The homology model for HLA-B*15:02 was constructed using HLA-B*15:01 (PDB ID: 1XR8) as a template, with a sequence identity of 98.19 %. The protein sequence of HLA-B*15:02/HLA-B*15:01 ( Fig. 1) was aligned using ClustalW. Further, the homology model generated by MODELLER was selected based on the lowest DOPE score. The DOPE score for HLA-B*15:02 was -28405.27.

Evaluation of model quality
The model with the lowest DOPE score was selected and subjected for quality evaluation. The quality of the predicted model was evaluated using RAMPAGE. Fig. 2a shows Ramachandran plot generated by RAMPAGE with most of the residues in the favorable zones i.e. the allowed regions for HLA-B*15:02 model. The Ramachandran plot analysis for HLA-B*15:02 shows that the residues in the most favored region are 98.2 %, and the residues in allowed region are 1.5 %.
Further model validation was carried out using PROCHECK based G-factors that indicates the quality of covalent and bond-angle distance. The Gfactors, indicating the quality of covalent and bond angle distance, were 0.01 for dihedrals, 0.41 for covalent, and overall 0.18 for HLA-B*15:02. The main chain bond length was 100 % while the main chain bond angle was 98 %within the limits for HLA-B*15:02. The overall main-chain and sidechain parameters as evaluated by PROCHECK were favorable for HLA-B*15:02. The modeled structure of HLA-B*15:02 is shown in (Fig. 2b).
Thus, the modeled structure was reliably employed for further studies.
HLA-B*15:02 belong to the same MHC class I that have domains such as α 1 -domain, α 2-domain, βpleated sheets and α 3-domain. The antigen or the peptide binding groove is located at the interface of α 1 and α 2 domain, beneath the β-pleated sheet. In HLA-B*15:01 the guanidine group of ARG62 forms salt bridge with the carboxylate of GLU63. While GLU63 mutation of to ASN63 in HLA-B*15:02 does not allow salt bridge formation. As a result, the HLA-B*15:02 antigen binding sites remain unaffected, and the mutation of this specific amino acid might be crucial for the specificity of the peptide binding or the antigen recognition at the B-pocket (Fig. 3a).

Modeling HLA-B*15:02 peptide-loaded complex
The HLA-B*15:02/FLFDGSPTY peptide complex was modeled based on the peptide (LEKARGSTY) complexed with the crystal structure of HLA-B*15:01 as a reference. The peptide loaded complex was optimized using FlexPepDock, a protein-peptide docking server, which generated top 10 protein-peptide complexes for each input complex. Among them, the best complex was selected entirely based on the best full-atom energy and subsequently compared to their corresponding reference peptide binding mode. The top full-atom total energy score of the best complex of HLA-B*15:02/FLFDGSPTY was -584.611. The resulting top full-atom total energy scored complex was considered for further studies. To cross validate the FlexPepDock results, PatchDock and FireDock studies were performed, and the ten best docked solutions were selected from refinement and rescoring analysis. For the best docking model of HLA-B*15:02/FLFDGSPTY binding energy was -108.04. FlexPepDock and PatchDock/FireDock docked binding modes were in agreement to each other for the complex (Fig. 3b).
All the 26 AEDs structures (Fig. 6) were docked using GOLD that predicts the binding of all the AEDs with the antigen binding groove of the peptide-loaded complex. The results of the docking studies were analyzed for the binding ability of AEDs that plausibly may induce ADR. Given the fact that both HLA-B*15:02 involve in CBZ induced ADR, molecular level interaction study reveals their feasibility to induce hypersensitivity upon CBZ binding, albeit they belong to different HLA subtypes. The docked ChemScore function for CBZ on B-pocket for HLA-B*15:02/FLFDGSPTY complex was 27.77. The ChemScore and CBZ binding mode of HLA-B*15:02/FLFDGSPTY complexes was in agreement upon analysis and, therefore, this complex was considered for further studies.

The CBZ binding mode in HLA-B*15:02/FLFDGSPTY complex
The docked binding mode of CBZ within the Bpocket/antigen binding groove of HLA-B*15:02/FLFDGSPTY complex forms a stable interaction. The ketone group of 5-carboxamide of CBZ forms a hydrogen bond with the guanidine group of ARG62, the tricyclic ring of CBZ forms a hydrophobic interaction with the side chain ring of TYR59, indole ring of TRP167 and the aliphatic chain of LEU163. Interestingly, the tricyclic ring of CBZ forms T-shaped π-π stacking with PHE1 of FLFDGSPTY-nonameric peptide. Geometrically, the tri-cyclic ring of CBZ fits well in the B-pocket of HLA-B*15:02 (Fig. 4a).
Docking studies of 26 AEDs with HLA-B*15:02/FLFDGSPTY complex In the second phase of the study, all the 26 AEDs were docked to HLA-B*15:02/FLFDGSPTY complex to screen the AEDs that could bind in a similar fashion alike CBZ and to be possible putative candidate that that may induce SCAR among the carriers of HLA-B*15:02. Following the docking protocol, analyzes were done based on ChemScore and visually inspection in PyMOL. Among the 26 AED's eight were selected based on their interaction pattern. Aacetazolamide, ethosuxiamide, lamotrigine, oxcarbazepine, phenobarbital, phenytoin, primidone and sodiumvalproate were selected (Fig. 4b).
Out of the eight AEDs, ethosuxiamide, oxcarbazepine, phenobarbital, phenytoin, primidone, and sodium-valproate shows similar binding mode to that of CBZ, as their structural scaffolds are similar. While acetazolamideand lamotrigine despite having a different chemical scaffold to that of CBZ yet shows a stable interaction at binding B pocket alike CBZ.
In acetazolamide, the sulfamoyl group oxygen forms hydrogen bond with the guanidine NH of ARG62. While the NH forms hydrogen bond with carboxyl oxygen GLU58 and the acetamide group forms a hydrogen bond with the guanidine NH of ARG170. While LEU163 forms hydrophobic interaction with the methyl group of acetamide as shown in (Fig. 5a).
In the case of ethosuxiamide, the pyrrolidine oxygens form a hydrogen bond with the guanidine NH of ARG62 and with the guanidine NH of ARG170 respectively. The ethyl group of ethosuxiamide forms hydrophobic interaction with LEU163 and PHE1 phenyl group of the FLFDGSPTY peptide as shown in (Fig. 5b).
In the case of lamotrigine, dichlorophenylring forms hydrophobic interaction with PHE1 of FLFDGSPTY-nonameric peptide, while the amine attached at the ortho position of triazine forms hydrogen bond with carboxyl oxygen of Glu58. The triazine ring forms hydrophobic interaction with TYR59, indole ring of TRP167 and aliphatic chain of LEU163 (Fig.5c).Oxcarbazepine binding mode shows that the hydroxy group attached to the azepine ring forms hydrogen bond with the guanidine group of ARG170, while the carboxamide oxygen forms hydrogen bond with the guanidine NH of ARG62. Additionally the tricyclic ring of oxcarbazepine forms hydrophobic interaction with the side chain ring of TYR59, indole ring of TRP167 and aliphatic chain of LEU163. Interestingly the tricyclic ring of oxcarbazepine forms T-shaped π-π stacking with PHE1 of FLFDGSPTY-nonameric peptide. (Fig.5d). While in the case of phenobarbital, the guanidine NH of ARG62 and ARG170 forms hydrogen bond with oxygen attached to the phenylpyrimidine trione, and the TYR59, TRP167, LEU163 and PHE1 of FLFDGSPTY forms hydrophobic interaction with phenylpyrimidine trione ring (Fig.5e). Whereas, in the case of phenytoin, the guanidine NH of ARG62 and ARG170 forms hydrogen bond with oxygen attached to imidazolidine, and the TYR59, TRP167, LEU163 and FLFDGSPTY PHE1 of peptide forms hydrophobic interaction with diphenylimidazolidine as shown in (Fig.5f). In the case of primidone the diazinane-4, 6-dione oxygen forms hydrogen bond with the guanidine NH of ARG62, while the phenyl ring of benzodiazepine forms π-π stacking with the indole group of TRP167 and hydrophobic interaction with PHE1 of FLFDGSPTY peptide, while the ethyl group forms hydrophobic interaction with LEU163 side chain as shown in (Fig.5g). In case of sodium-valproate the pentanoate oxygen forms hydrogen bond with the guanidine NH of ARG170, where the propyl group forms hydrophobic interaction with TRP167 and the phenyl group of PHE1 of FLFDGSPTY peptide (Fig.5h).
The current computational study shows the identification of AEDs specific molecular interaction patterns to differentiate between inducer and non-inducer of SJS using CBZ interaction pattern as a reference point.

DISCUSSION
A study report in 2010 indicated that carbamazepine, allopurinol and cotrimoxazole were the three main drugs which cause SJS/TEN in Malaysia. These accounted for 24.0%, 18.8% and 12.5%, of all ADR cases reported from 2001 to 2008 in Malaysia [31]. Therefore, there is a continuous interest among different groups of researchers in Malaysia to determine causative agents associated with these incidences. This is in line with the strategies to improve the quality of patients care and in realizing personalized medicine in Malaysia. Subsequent to data published by Chung et al [11] and Man et al [12], the association of HLA-B*1502 genetic marker with CBZ induced SJS/TEN was reported by Chang et al [32]. Whereby 21 subjects with CBZ-induced TEN/SJS were studied. However, HLA-B*15:02 was present in 75.0% (12/16)  Similarly high frequency of HLA-B*15:02 allele in CBZ induced SJS-TEN (94.1%) was reported in Thailand [33,34]. In India, eight individuals who fulfilled the diagnostic criteria of SJS induced by carbamazepine were identified and HLA-B molecular typing was performed. Out of eight patients, six patients were found to have the HLA-B*15:02 allele [35].A meta-analysis conducted by Tangamornsuksan et al [13] revealed a high risk of developing SJS and TEN among carbamazepine users with HLA-B*15:02. For Han-Chinese, Thais, Koreans, and Malaysians, the risks of SJS and TEN among carbamazepine users carrying the HLA-B*15:02 allele are approximately 115-fold, 60-fold, 25-fold, and 220-fold higher, respectively.   [32] reported the presence of HLA-B*15:02 in 15.7% (47/300) of normal controls. We reported the presence of the allele in 17.5% of patients who requires CBZ therapy. A study that recruited 56 Caucasian patients who developed serious hypersensitivity reaction which include SJS/TEN found that none of the patients carry HLA-B*15:02 [36]. Another study identified twelve cases of CBZ-induced SJS/TEN (retrieved from RegisSCAR). And only 4 of the patients were positive for HLA-B*15:02 allele. However, those four positive HLA-B*15:02 allele were from Asian ancestry [14]. Therefore, HLA-B*15:02 is not a marker of SJS/TEN among the Caucasian and other alleles but could be responsible for the occurrence of CBZ related hypersensitive reactions in these populations. The search for genetic markers therefore continues. The HLA-A*31:01 allele which has a prevalence of 2 to 5% in the Northern European populations was later found to be significantly associated with the CBZ related hypersensitivity syndrome (P = 3.5×10−8). Genotyping confirmed the variant as a risk factor for the hypersensitivity syndrome (odds ratio, 12 [36]. However, HLA-B*15:02 is rare in the Japanese population and is not found in patients with SJS induced by CBZ [15]. Further research by Ikeda et al. suggested that HLA-B*59:01 is one of the candidate markers for CBZ-induced SJS in Japanese [16].
Interestingly, there were 9 patients who developed mild rashes among the patients, but they were negative for HLA-B*15:02. Similar to a study conducted earlier by another research group in Malaysia, Then et al. found 6 out of 27 patients who developed skin rashes after CBZ administration were not positive for the HLA-B*15:02 allele [37]. Therefore, further research is required to determine alleles that caused skin rashes other than HLA-B*15:02.
Our results indicate that the HLA-B*15:02 allele is strongly associated with increased risk of developing SJS and TEN in patients using carbamazepine and further justified the clinical practice of pre-genotyping in Malaysian. This severe adverse event could be prevented if genetic information were known a priori. Physicians and policymakers should consider our findings to support the implementation of genetic testing before initiation of carbamazepine therapy. In addition to the clinical importance, the availability of simple and rapid genotyping kit is necessary to implement pharmacogenotyping of HLAB*15:02 in clinics. Currently, the method adopted in this study allows identification of patients at risk within 4 hours and using the method optimized and validated in our study. We were able to detect patients with HLA-B*15:02 correctly. The genotype was reconfirmed with direct DNA sequencing technique by a third party. The prospective part of the study further shows that pharmacogenotyping is important as HLA-B*15:02 is relatively common in the Malaysian population. Clinicians are recommended to refer to the clinical pharmacogenetics implementation consortium guidelines for the use of carbamazepine based on HLA-B genotype [38].
In the present computational study, 26 AEDs have been virtually screened to identify AEDs that may cause ADR similar to that of CBZ, specific to HLA-B*15:02. The computational interaction study clearly discriminates between the inducers from non-inducers of ADR among HLA-B*15:02 carriers. Our study also reveals that the preloaded endogenous peptide specificity is crucial for the AEDs induction and stabilization in order to cause ADR. This study also suggests that apart from hydrogen bonding, hydrophobic interaction is also crucial for stabilization of the complex to induce ADR. This hypothetical molecular interaction study attempts to reveal, how some AEDs act antigenic in some individuals to induce drug mediated hypersensitivity. The current computational modeling study also demonstrates that the presence of specific endogenous peptide in theHLA-B*15:02 is essential to stabilize the complex, which in turn would be bound by the AEDs to induce hypersensitivity. Although the stability of the complex is important, the conformational orientation of the peptide also plays a crucial role in determining the drug interaction with the peptide mediated complex. Our modeling result also shows that the residue ARG62 at the B pocket is the key residue for interaction with the AEDs.
Our computational studies show that acetazolamide, ethosuxiamide, lamotrigine, oxcarbazepine, phenobarbital, phenytoin, primidone and sodium-valproate may cause ADR similar to that of CBZ in Southeast Asians carrying HLA-B*15:02 allele. AEDs like clobazam, clonazepam, Eslicarbazepine acetate, gabapentin, lacosamide, levetiracetam, nitrazepam, perampanel, piracetam, pregabalin, retigabine, rufinamide, stiripentol, tiagabine, topiramate, vigabatrin, and zonisamide did not show stable binding mode. The key chemical moiety attached to the core scaffold of the above AEDs results in unfavorable hydrogen bonding and hydrophobic interaction with the peptide loaded complex that affects their binding affinity. Henceforth, the study theoretically shows that these AEDs as non-binders which could be proposed as safe AEDs among individuals who develop drug induced ADR after clinical confirmation. However, additional clinical studies may be required to confirm the AEDs inducing ADR among HLA-B*15:02 allele carriers.

CONCLUSION
Screening patients for the HLA-B*15:02 allele before the initiation of carbamazepine treatment and withholding carbamazepine from HLA-B*15:02positive patients is important clinically to reduce the incidence of carbamazepine-induced SJS/TEN among the Malaysians. Pharmacogenotyping based practice of HLA-B*15:02 should be adopted by clinician as part of routine patient care. A new perspective for screening HLA-associated drug hypersensitivity using molecular docking technique was proposed in the present study. This approach will help to select safe AEDs for individuals with HLA-B*15:02 allele. This approach is particularly important if HLA polymorphism is widely present among the population.