Human leukocyte antigen (HLA) molecules play a vital role in the immunological T cell response. HLA-A and HLA-B code for major histocompatibility complex (MHC) class I, which mainly presents peptide fragments derived by proteasomal degradation of “self” proteins on the cell surface of the antigen presenting cell to cytolytic T cells (1–3). HLA encoding genes are highly polymorphic and over 8,500 unique HLA-A and HLA-B haplotypes have been identified to date (4, 5), each presenting different peptide fragments of mostly nine amino acids in length (6). This polymorphism provides an evolutionary advantage, because pathogens cannot easily develop resistance by mutating residues critical for binding to all HLA haplotypes. In this article, we reason that another way how resistance is prevented is by the presentation of stretches of amino acids which are critical for protein function: those located within transmembrane helices (TMHs) of integral membrane proteins.

Each HLA-A and HLA-B haplotype presents peptides with preferential requirements for the charge and hydrophobicity of each residue within the peptide (7). Based on these requirements for HLA binding, the haplotypes of HLA-A and HLA-B can be grouped in five and seven super types, respectively (7–10). The binding peptides (binders) for these super types can be predicted from protein sequences by bioinformatics tools (11, 12), whose high accuracy was empirically confirmed in several studies (13–16). For most HLA-A and HLA-B super types, the presence of multiple hydrophobic residues is favorable for peptide binding (7).

The occurrence of multiple hydrophobic residues in close proximity of each other is a hallmark of TMHs. Due to the hydrophobic nature of lipid membranes, a transmembrane motif of mostly 23 amino acids in length containing non-polar headgroups is required to span the membrane (17). These stretches are frequently flanked by bulky hydrophobic residues, such as tryptophan, that allow for anchoring of the protein at the water–lipid interface, and by positively charged lysine and arginine residues that allow for electrostatic interactions with negatively charged lipid headgroups (18–20). Similar to HLA binders, the presence of TMHs can be predicted with high accuracy from protein sequences (21). Approximately 25% of all human genes code for integral membrane proteins that are anchored by one or more TMHs to the plasma membrane or to the lipid membranes of organelles.

Since both HLA-A/B binders and TMHs contain multiple hydrophobic residues, it is likely that HLA-A and HLA-B haplotypes would preferentially present peptides derived from TMHs. This is supported by a comparison of the predicted binders with predicted TMHs from the human proteome (Figure 1A). Predicted TMHs contain more predicted HLA binders than expected from a random distribution of binders for all HLA-A and most HLA-B super types, and this was significant for all HLA super types (see Methods; P < 0.001; individual P values in Table S1 in Supplementary Material). Comparing the hydrophobicity (22) of all peptides in the human proteome reveals that peptides derived from TMHs are more hydrophobic than those originating from other protein regions (Figure 1B) and more than 90% of the 10% most hydrophobic peptides originate from TMHs (Figure 1C).

We performed two control analyses to determine to what extent the preferential HLA binding of TMH-derived peptides was related to their hydrophobicity. First, we scored the hydrophobic preference for each HLA super type (see Methods). Haplotypes for which this score is positive will bind peptides with above average hydrophobicity. We found a strong correlation (r = 0.88) between these hydrophobic preference scores and the percentages of predicted HLA-binding peptides derived from TMHs (Figure 1D). For a second control, we selected a set of peptides derived from non-transmembrane protein regions of the human proteome, but with a matching hydrophobicity distribution as of those derived from TMHs (Figure 2A). If the predicted preferential presentation of peptides locating to TMHs were entirely attributable to their large hydrophobicity, we would expect these control peptides to be equally well presented. Indeed, this was largely the case (Figure 2B), except for HLA-A02. For HLA-A02, peptides derived from TMHs were predicted to be presented with a lower probability than peptides from the control set. Possibly this may be due to the specific compositional requirements of TMHs resulting in less-favorable positioning of residues as in other hydrophobic domains critical for HLA-A02 binding.

To investigate this compositional bias further, we also determined the positions of the predicted binders relative to the positions of the predicted TMHs (Figure 3). These positions seem to correlate well with the requirements for peptide binding to the different HLA haplotypes. For instance, predicted binders of the HLA-A02 super type are mainly located within the hydrophobic core of TMHs, which is consistent with the requirement of multiple hydrophobic amino acids within the peptides (7). In contrast, binders for the HLA-B27 super type are enriched at the lipid–water interface of TMHs, correlating to its preference for charged amino acids within the epitope (7).

A survey of the literature confirmed that membrane-buried peptides are presented in MHC-I. Comparison of the predicted TMHs with the MHC-I epitopes identified from B lymphoblastoid cell lines by peptide elution coupled to mass spectrometry, revealed that approximately 1% of naturally presented MHC-I epitopes are predicted to be located within TMHs (24). The hydrophobicity of these detected epitopes is somewhat lower than of the predicted binders (compare Figure 4A with Figure 1B), and approximately 15% of the 10% most hydrophobic detected epitopes is derived from TMHs (Figure 4B). This percentage is lower than one would expect purely based on the predicted binding affinity (Figure 1C), which might be due to a lower abundancy of transmembrane proteins compared to soluble proteins. Alternatively or additionally, the lower percentage of peptides derived from TMHs might be caused by a lower detection efficiency, as hydrophobic peptides are often underrepresented in mass spectrometry due to their poor solubilization and ionization characteristics (25, 26). In any case, this survey of peptide elution studies (24) shows that MHC-I presents epitopes derived from TMHs.

In fact, many known MHC-I-binding epitopes locate within transmembrane helices. For HLA-A02, which is the most prevalent HLA-A haplotype in the Caucasian population (27), more than 10% of the epitopes in the Immune Epitope Database (28) correspond to peptide fragments overlapping with TMHs. These include many well-studied and clinically important epitopes that were found in in vivo studies. For example, multiple HLA-A02-binding epitopes are derived from the TMH of the viral coat protein gp41 from the human immunodeficiency virus (29). The same holds true for cancer recognition, and for instance an HLA-A02-binding epitope derived from the TMH of melanoma antigen recognized by T-cells 1 (Mart1; also known as Melan-A; residues 26–35) (30) is strikingly immunodominant, with few other MHC-I epitopes described (31).

The presentation of epitopes derived from TMHs could offer a clear evolutionary advantage. TMHs are well-conserved as there are strict compositional requirements for spanning lipid membranes (17). In addition to a hydrophobic core, the charge distribution of adjacent residues is a critical determinant for the orientation of a TMH within the membrane (32). The length of TMHs is important for the localization of transmembrane proteins to the correct organellar membrane (33). Because the precise composition of TMHs is essential for protein conformation and localization, pathogens might be unable to evade antigen presentation by mutating these regions. We believe this could constitute a novel mechanism contributing to the fidelity of antigen presentation.

It is mechanistically unclear how TMHs can be processed for MHC-I antigen presentation. For presentation of self-coded antigens in MHC-I, membrane proteins need to be first extracted from the membrane and subsequently targeted to the proteasome for the generation of antigenic epitopes (Figure 5). To date, this extraction has only been described for transmembrane proteins in the endoplasmic reticulum (ER) by a process that involves the ER-associated degradation pathway. This incompletely understood process is predominantly being studied in yeast (34, 35) and it is unclear if and to what extent it contributes to antigen presentation in mammals. However, self-presentation of an epitope derived from the TMH of Mart1 is dramatically enhanced with a mutant form that is selectively retained at the ER (36), suggesting that extraction of proteins from the ER membrane may well be of major importance for antigen presentation. Further evidence that TMH-derived epitopes are generated at the ER comes from the finding that a peptide fragment from the TMH of Mart1 is trimmed by the ER-associated proteases (ERAP) to release the MHC-I epitope (37).

The processing of TMHs for MHC-I presentation may well be mechanistically reminiscent to HLA-E presentation. HLA-E is a paralog of MHC-I specialized in the presentation of epitopes derived from the N-terminal signal peptides of other MHC-I paralogs, including HLA-A and HLA-B, for recognition by natural killer cells (38). These signal peptides resemble TMHs and are extracted from the ER membrane following cleavage by the signal peptide peptidase (39), and subsequently processed by the proteasome for HLA-E presentation (40). In this respect, it is interesting that downregulation of ERAP1 leads to upregulation of HLA-E, suggesting a functional link between HLA-E presentation and conventional antigen presentation (41).

It is also unclear whether and via which mechanisms transmembrane proteins locating at other organelles (i.e., outside the ER) are processed for MHC-I presentation. These transmembrane proteins are degraded in lysosomes via the formation of multi-vesicular bodies (Figure 5) (42). If and how these mechanisms contribute to the release of epitopes from TMHs for MHC-I presentation is yet unknown.

How integral membrane proteins can be processed for antigen cross-presentation is unclear as well. Cross-presentation is the process by which antigen presenting cells take up, process, and eventually present extracellular antigen on MHC-I to cytolytic T lymphocytes. There are currently two mechanistic pathways of antigen cross-presentation described (43), but for both pathways it is not easy to envision how they could result in cross-presentation of transmembrane proteins (Figure 5). In the first cytosolic pathway, ingested proteins translocate from endosomes into the cytosol where they become accessible for proteasomal degradation. The proteasome-derived peptides can then be imported into the ER or back into endosomes for subsequent loading onto MHC-I. The processing of transmembrane proteins would require either (i) the extraction of the TMHs from the lipid membrane within the lumen of endosomes or (ii) the translocation of large membrane fragments incorporating the antigen. In the second vacuolar pathway, proteins are degraded within endosomes by lysosomal proteases and subsequently loaded onto MHC-I. For processing of transmembrane proteins via this pathway, the TMHs need to be also extracted from the membrane or (iii) cleaved within the lipid membrane. All these processes have not been described to date.

Thus, from emerging bioinformatics information, the concept arises that most HLA-A and HLA-B haplotypes are biased to present epitopes derived from TMHs. Our present knowledge of antigen presentation is exclusively based on that of soluble antigens, and we currently cannot explain how antigen presenting cells can self- and cross-present integral membrane proteins. Both these processes can be expected to require unique, yet unknown, mechanisms and this might well relate to the disease-associations of specific HLA haplotypes. For instance, members of the HLA-B27 super type are associated with spondyloarthropathies (psoriasis, inflammatory bowel disease, reactive arthritis, ankylosing spondylitis) (44). As we explained in this Opinion article, this super type preferentially presents epitopes that are immediately adjacent to transmembrane helices and perhaps medullary thymic epithelial cells cannot efficiently present these peptide fragments. This would result in incomplete negative selection and a population of self-reactive cytolytic T cells, which might contribute to the pathology of these inflammatory diseases. Evidence for such an incomplete self-tolerance of membrane-buried epitopes comes from the finding that a large proportion of healthy HLA-A02 restricted individuals possess naive cytolytic T cells specific for the immunodominant epitope from the TMH of Mart1 (45). Resolving how TMHs are processed for antigen self-presentation and cross-presentation should allow for a better understanding of immunology, and it might ultimately allow for new insights into the mechanisms of the disease-associations of HLA haplotypes.