Major histocompatibility complex (MHC) class II molecules are heterodimeric glycoproteins consisting of α and β chains, with approximate molecular mass of 33 kDa and 28 kDa respectively. MHC class II molecules are specialized peptide receptors that play a critical role in initiating and regulating immune responses by binding peptide fragments that are 10–30 amino acids long [1] and present them on the surface of antigen-presenting cells for recognition by CD4+ T cells. The class II region encodes genes for the human leukocyte antigen (HLA) or histocompatibility molecules class II structural genes DP, DQ and DR [2, 3]. While specific DP, DQ or DR alleles at the HLA class II locus have been shown to correlate with particular autoimmune diseases, a variety of confounding factors including strong linkage disequilibrium between the different HLA alleles, especially DR and DQ, complicates the exact identification of MHC susceptibility alleles.
Pemphigus Vulgaris (PV) is a potentially life-threatening form of autoimmune blistering skin disorder due to loss of integrity of normal intercellular attachments within the epidermis and mucosal epithelium. The disease is characterized by the presence of pathogenic autoantibodies directed mainly against a 130-kDa transmembrane glycoprotein, desmoglein-3 (Dsg3) [4], within the desmosomes of the spinous layer of the skin. Strong association of PV to the major histocompatibility complex class II serotypes DR4 and DR6 have been reported in the literature [5, 6, 7] with over 95% of PV patients possessing one or both of these alleles [7]. Direct nucleotide sequence analysis of DR4 and DR6 subtypes revealed that susceptibility to PV is strongly linked to DRB1*0402 and DQB1*0503 molecular subtypes, respectively [7, 8].
The use of computational techniques has been instrumental in advancing epitope-based vaccine research, with much work focusing on predicting the binding specificities of peptides to MHC molecules. Sequence-based predictive systems, based on identifying patterns in peptides with experimentally determined binding strength, are widely used to facilitate the identification of binding peptides to MHC class II molecules. Southwood et al. [9] developed a scoring matrix for DRB1*0401 based on a polynomial technique. Mallios [10, 11] reported the results of an iterative stepwise discriminant analysis meta-algorithm to identify binders from non-binders for DRB1*0101. Brusic et al. [12] applied a genetic algorithm to discriminate binders from non-binders for DRB1*0401. Noguchi et
al. [13, 14] utilized both fuzzy neural network and hidden Markov model to predict potential binders to DRB1*0401 and DRB1*0101. Hammer et al. [15] employed a peptide side chain scanning technique for screening peptides that interact with DRB1*0401. Nielsen et al. [16] used a Gibbs sampling method for discriminating DRB1*0401 specific binders from non-binders. Karpenko et al. [17] made use of an ant colony system to search for DRB1*0401 binding and non-binding peptides. Doytchinova and Flower [18] employed an additive method for predicting the binding affinity of peptides bound to DRB1*0401, DRB1*0101 and DRB1*0701 based on the sum of the contributions of the amino acids at each position of the bound peptide and various interactions between them. However, despite recent advances in sequence-based predictive techniques, computational models for the majority of PV implicated alleles have been lacking, mainly due to the paucity of sufficient peptides as training data, and are unsuitable for predicting peptide binding to PV implicated alleles. Also, most computational methods focus on predicting just peptide binders and non-binders, whereas our aim is to distinguish between different modes of binding conferred by susceptible and protective alleles.
An alternative approach to predicting peptide/MHC (pMHC) complexes without the need of a large training dataset is to use information derived from three-dimensional structures. Logean and Rognan [19] utilized a combinatorial built-up algorithm to construct the three-dimensional structure of pMHC complexes. Altuvia et al. [20] reported the use of a computational threading approach to rank potentially binding peptides to MHC class I molecules. Lim et al. [21] employed molecular dynamic simulations to examine the structures of A*0201 in complex with 9-mer peptides. Michielin et al. [22, 23] applied homology modeling to select peptides that bind to A*0201.
In addition to predicting the binding specificities of peptides to MHC molecules, three-dimensional models have also been used for structural classification of alleles into HLA "supertypes" based on structural features derived from the binding sites. Recently, Doytchinova et al. [24, 25] employed hierarchical clustering and principal component analysis to classify alleles based on structural features into eight HLA class I and twelve HLA class II supertypes. The structural classification of alleles facilitates the identification of allelic subgroups that may share similar binding specificities and shed light into their possible role in cellular immunity against pathogens.
In the present study, we have attempted to understand the functional correlation between MHC class II alleles and PV, from a structural interaction view point. Molecular modeling of ten PV associated and non-associated MHC class II receptors (DR4: DRB1*0401, *0402, *0404, *0406, DR6 (also classified now as DR14): DRB1*1401, *1404, *1405, DQ2: DQB1*0201, *0202 and DQ5: DQB1*0503) were performed to explore the structural organization of the binding groove of these alleles. Nine previously identified epitopes, Dsg3 96–112, Dsg3 191–205, Dsg3 206–220, Dsg3 252–266, Dsg3 342–356, Dsg3 380–394, Dsg3 763–777, Dsg3 810–824 and Dsg3 963–977 (numbered in accordance with Swiss-Prot [26] accession number P32926), capable of stimulating patient derived T cells, were selected. The binding of these peptides to the DR and DQ structural models were studied by our efficient computational docking protocol [27]. In the light shed by these atomic models, the binding specificities of each allele to the various Dsg3 peptides are discussed. The results obtained in the study are able to discriminate between PV associated and non-associated alleles, consistent with the experimental results obtained by Veldman et al. [28] and Sinha et al. [unpublished results for Dsg3 342–356, 810–824 and 963–977]. Insights into structural features behind the immune response provided by protective alleles for PV have also been obtained by our structural immunoinformatics approach.