While MHC class II molecules and their ligands are well characterised, the impact of peptide length on MHC affinity is not well understood. While several anecdotal observations have purported to show that peptide length may influence affinity we have for the first time systematically quantified this relationship using in silico analysis. Using novel computational analyses, we exploit the wealth of manually-curated quantitative binding data available within the AntiJen database [14].
When attempting to derive a relationship between sequence length and observed MHC affinity, one is limited by several factors. One such factor is the presence of different binding determinants in most sequences that could significantly bias any simple length-affinity correlation. Thus, if one examines the affinity of two sequences of different length and low sequence similarity, one cannot ascertain if an altered affinity is due to differences in the binding determinant or to changes in sequence length. To isolate such a relationship, one must normalise all factors (such as MHC allele, experimental conditions, and peptide sequence) except for sequence length before analysing the affect on affinity. To that end, we identified examples of affinity measurements, for a single sequence and an elongated equivalent, which were performed under the same experimental conditions. The benefits of this approach were twofold. Firstly, inter-laboratory variation in affinity measurements would be excluded. Secondly, bias from variations in the binding determinant would also be reduced if not eliminated, thus removing bias from intra- but not inter-sequence comparisons. Bias from inter-sequence comparisons is unavoidable. We optimised for the required task in order to avoid representing bias as effect.
The data presented in Figure 2 allowed us to examine the limits of the effect with respect to final peptide length. By examining the log fold-change in affinity for MHC class II as the length of the peptide increases it can clearly be seen that the impact on affinity lessens. The regression plot also shows that at approximately 19 amino acids the effect of peptide elongation on affinity becomes zero. Elongating peptides beyond this point would – in the majority of cases – result in a diminished affinity. This seems logical: if no such ceiling existed for the positive effect of peptide length then sequence length alone would be the key determinant of peptide-MHC affinity. This discovery is also strengthened by the findings of Lippolis et al. who eluted and sequenced peptides from HLA-DR*0401 and found that the most abundant species within a family of nested peptides were between 14 and 21 residues in length [15]. Our findings do not however suggest that a single optimal length would exist for all peptides for all MHC class II alleles, and indeed the relative heterogeneity of peptides binding to MHC class II molecules is well reported [15, 16, 17]. However, this approximation to an optimal length is consistent for a large number of peptide sequences and a large number of MHC class II alleles.
Many explanations for this length ceiling present themselves. It may be that as peptides become very long they are no longer able to form favourable interactions with the MHC or TCR. Alternatively, when MHC class II binding peptides attain a certain length they may form secondary or tertiary structures outside the binding site that are less compatible with MHC class II binding. An exception to this rule may be the 33-amino acid peptide initially described by Shan et al [18]. It binds to HLA-DQ2 and stimulates disease associated T cells. Despite its unusual length, this peptide retains its MHC class II binding abilities as its proline rich nature helps it to form a type II polyproline helix. This is the conformation normally seen in MHC class II binding grooves [3].
Our results clearly imply a general correlation between increasing MHC binding affinity and increasing peptide length. It is not peculiar to a restricted subset of peptides. Sercarz and Maverakis posited that longer peptides have a greater binding affinity for MHC class II [19]; this assertion was based on observations from a few articles, e.g. the study by Srinivasan et al [13]. However, to the best of our knowledge, ours is the first study to test this hypothesis systematically for several MHC class II alleles. The results of our study are compelling: across 1279 identified peptide elongation events for peptides binding to 19 distinctive MHC class II alleles, affinity predominately increased. It is important to note, however, that the examined peptides are more representative of the HLA-DR species than either HLA-DP or HLA-DQ, and this may result from an apparent publication bias in favour of HLA-DR which will be reflected in the AntiJen database.
Our findings have clear implications for experimental immunology and vaccinology, as well as computational immunology; and these implications are many. Increased affinity in vivo may result in increased reactivity to recently degraded peptides. Presumably, recently degraded peptides may be less completely digested and – as a result of increased length – have an increased affinity for MHC class II relative to shorter more fully degraded peptides. Such an effect would, in theory, increase the likelihood of T cells recognising more recently endocytosed peptides rather than a background of self-peptides. Additionally, the ability of exopeptidases to digest termini of peptides within the MHC class II binding cleft [6] may effectively reduce their MHC class II affinity. Thus, peptides susceptible to digestion by exopeptidases may prove to be less immunogenic in vitro. The field of subunit vaccine design may be able to augment vaccine efficacy by increasing peptide length or eliminating potential proteolytic cleavage sites, which may result in an inability of normal proteolytic enzymes to digest peptides to shorter equivalents with reduced immunogenicity.