Machine learning reveals a non-canonical mode of peptide binding to MHC class II molecules

Abstract

MHC class II molecules play a fundamental role in the cellular immune system: they load short peptide fragments derived from extracellular proteins and present them on the cell surface. It is currently thought that the peptide binds lying more or less flat in the MHC groove, with a fixed distance of nine amino acids between the first and last residue in contact with the MHCII. While confirming that the great majority of peptides bind to the MHC using this canonical mode, we report evidence for an alternative, less common mode of interaction. A fraction of observed ligands were shown to have an unconventional spacing of the anchor residues that directly interact with the MHC, which could only be accommodated to the canonical MHC motif either by imposing a more stretched out peptide backbone (an 8mer core) or by the peptide bulging out of the MHC groove (a 10mer core). We estimated that on average 2% of peptides bind with a core deletion, and 0·45% with a core insertion, but the frequency of such non-canonical cores was as high as 10% for certain MHCII molecules. A mutational analysis and experimental validation of a number of these anomalous ligands demonstrated that they could only fit to their MHC binding motif with a non-canonical binding core of length different from nine. This previously undescribed mode of peptide binding to MHCII molecules gives a more complete picture of peptide presentation by MHCII and allows us to model more accurately this event.

Keywords: MHC class II; deletions; insertions; machine learning; non-canonical binding.

Figure 1

Correlation coefficient (average over 37 molecules) of the method versus the number of burn‐in iterations used to prime the networks. Networks were trained in cross‐validation with a maximum insertion length of one amino acid and a maximum deletion length of one amino acid. NoGap corresponds to the method trained without insertions and deletions.

Figure 2

Correlation coefficient for methods trained with different deletion (d) and insertion (i) maximum lengths. The method with at most one deletion and one insertion (d = 1; i = 1) had significantly higher performance than the method without insertions/deletions (d = 0; i = 0) with higher Pearson's correlation coefficient (PCC) in 34/37 molecules (P = 10⁻⁷). It also outperformed d = 1; i = 0 on 34/37 molecules (P = 10⁻⁷) and d = 0; i = 1 on 33/37 (P = 10⁻⁶) molecules. Allowing longer deletions of up to two amino acids (d = 2; i = 1) does not significantly improve cross‐validated performance compared with d = 1; i = 1 (P = 0·32). All P‐values were calculated with two‐tailed binomial tests. **Highly significant; ns Not significant.

Figure 3

Frequency of predicted binders containing deletions (black) and insertions (white) in the binding core using a burn‐in of 100 iterations. While some molecules have over 10% of predicted binders with a non‐canonical binding core (e.g. HLA‐DRB3*01:01 and HLA‐DRB5*01:01), others have almost no predicted binding cores with insertions or deletions (e.g. the mouse H‐2 alleles). The figure was generated, for each allele, from the top 10% scoring peptides out of 100 000 natural random 15mers.

Figure 4

Per cent of peptide–MHCs that must be removed to ensure no overlap between cross‐validation partitions, depending on the common‐motif threshold N on the x‐axis. Approximately the same number of peptides has to be removed (~14–15%) for N = 6 to N = 10, reflecting a possible bias in the procedures used to assay binding affinity with a sliding window over antigens of interest.

Figure 5

Correlation coefficient (average over 37 molecules) as a function of the burn‐in rate for networks trained on the low redundancy data set. Networks were trained in cross‐validation with a maximum insertion length of one amino acid and a maximum deletion length of one amino acid. NoGap corresponds to the method trained without insertions and deletions.

Figure 6

Predicted 9mer and 10mer binding cores for two DRB1*03:01 and two DRB5*01:01 ligands. The non‐canonical spacing of the anchor residues can only be accommodated with a deletion in the binding core, which is depicted here as a protrusion of the peptide chain at the predicted position of the deletion. Reference sequence logos are from NNA lign.15 [Colour figure can be viewed at wileyonlinelibrary.com]