Several autoimmune disease models have been shown to be dependent on posttranslational modifications of relevant Ags (e.g. on acetylation of a myelin basic protein peptide (MBP-Ac1-11) [40] or on the isoaspartyl-modification of a self Ag in a mouse model of systemic lupus erythematosus [41] or of peptides in DR4-transgenic mice expressing human type II collagen [42].
Surprisingly, the role of the most common PTM, glycosylation, is poorly understood not only in the development of autoimmune diseases, but also in basic immune processes like T cell recognition.
Instead of using glycosylation-related annotations of SwissProt/UniProtKB or one of the PTM- specific databases (dbPTM or O-GLCBASE 6.0 [43, 44]), in our work, we chose to use Artificial Neural Network-based glycosylation prediction tools. By this approach, we could avoid errors of comparison resulting from publication bias, the lack of negative results and the diverse methods used to characterize glycosylation of molecules. Beside the numerous advantages, the authors are also aware of disadvantages of in silico technologies. Prediction methods use mainly stochastic algorithms, and a known portion of all provided answers is incorrect. This error, however, can be taken into account precisely, and be corrected during statistical analysis, while it is very hard to assess the systemic bias caused by errors in database entries.
In our study, we compared the glycosylation of experimentally verified T cell epitopes of the IEDB to that of the human proteome. According to our present knowledge, the most common types of glycosylation, N- and O-glycosylation, occur only in the rER. Nevertheless, earlier studies on glycosylation of the proteome have not yet considered the presence or absence of a (SigP) [45]. Our data are in accordance with the results of former studies regarding the frequency of glycosylation. Earlier, based on the analysis of a smaller pool of 6000 human proteins, it has been demonstrated that N-glycosylation preferentially occurs on proteins associated with transport and binding functions [45]. In another article, the clustering phenomenon of O-glycosylation was also reported for 221 glycoproteins [15]. Furthermore, N-glycosylation was reported to occur preferentially in the central and N-terminal regions of proteins, whereas O-glycosylation showed preference for both the N- and C-termini [46].
In this study, we found very low glycosylation of the experimentally verified T cell epitopes in comparison with the glycosylation of the entire human proteome. This is logical given the necessity of a TCR to interact simultaneously with the alpha helices of the MHC molecule and the presented peptide. If this peptide is glycosylated, the trimolecular complex formation might be prevented because of steric reasons. Such interference of epitope glycosylation with TCR binding has been indeed, reported already [1, 2, 3]. However, on the other hand, there are also data on recognition of glycopeptide epitopes by T cells [2, 47, 48, 49, 50, 51].
Epitope glycosylation may not only interfere with TCR binding but also with binding of a peptide to MHC. Many of the T cell epitopes in the IEDB may represent immunodominant epitopes, those few peptides that are found in the majority of peptide-MHC complexes during an immune response after natural intracellular processing of an Ag [52, 53]. Whether glycopeptides suffer any disadvantage because of their glycan moiety when competing for MHC binding, appears to be an interesting question with practical consequences.
The fact that we found only a minimal proportion of MHC II-restricted peptides to be glycosylated in case of most MHC alleles, supports the notion that the glycan moiety might indeed negatively affect glycopeptide presentation. On the other hand, the relatively high proportion of glycosylated peptides predicted to bind to certain HLA alleles, might be an artifact caused by the highest sensitivity, thus, a bit lower specificity of the prediction. Higher permissivity of these alleles towards probable N- but not O-glycosylation of peptides, however, might have relevance to pathological immune responses, but this difference still needs to be experimentally confirmed.
T cell epitope glycosylation may also have implications to autoimmunity. It may be hypothesized that within the thymus, glycosylation of peptides interferes with their presentation by protecting proteolytic cleavage sites. This would be a mechanism parallel to that already reported in case of O-glycosylation [3]. Not only the fact of glycosylation, but also the size and complexity of the N-glycan moiety might play a significant role in the presentation of N-glycosylated peptides [4]. In the absence of presentation, certain glycosylated parts of proteins may indeed, escape thymic central tolerance induction. In the periphery, however, glycosidase enzymes (of microbial or inflammatory cell origin) may attack the glycan moieties, and the naked peptides may become target neoepitopes of an autoimmune recognition. In line with this concept, our earlier data showed that elevated glycosidase activities were indeed, predictors of rheumatoid arthritis [54]. Proteins with higher number of N-glycosylation sites thus, are more probably recognized by autoimmune T cells than other proteins. In this work, we observed that the N-glycosylation of autoantigens was significantly higher than that of length-matched randomly generated sequences or normal proteins.
Molecular mimicry has been shown in several human autoimmune diseases [55] including multiple sclerosis [56], Sydenham's chorea [57] rheumatic heart disease [58], autoimmune thyroiditis [59], Chagas heart disease [60], systemic lupus erythematosus, etc. [61]. Since overlapping human-bacterial sequences may trigger autoimmune processes, their immune recognition might be substantial regarding autoimmunity. We found a reduced glycosylation of the linear human-bacterial exact sequence matches. As both T cell epitopes and bacterial-human exact sequence matches are much shorter than full length proteins, the reduced glycosylation observed in our work, could have been attributed to the difference in peptide length. Indeed, we found that glycosylation probability increases in parallel with the peptide length. However, we have shown in this study that the reduced glycosylation of neither the T cell epitopes, nor the human-bacterial exact sequence matches were the function of peptide length only.
According to most in vitro results, low glycosylation rates of human T cell epitopes may be a prerequisite for an effective immune recognition. Given the necessity of both hypoglycosylation and the highest specificity for an epitope sequence; the length of experimentally verified human T cell epitopes occupies a niche optimized for these two features (Figure 6). At the relevant peptide lengths for MHC-II-restricted T cell epitopes, only a relatively low number of bacterial-human exact sequence matches can be found, and simultaneously the probability of peptide glycosylation is also minimal.