To predict how frequently epitopes overlapping with TMHs are presented, a similar analysis strategy was applied as described in (7) for several alleles of both MHC-I and MHC-II, and for a human, viral and bacterial proteome. To summarize, for each proteome, all possible 9-mers (for MHC-I) or 14-mers (MHC-II) were derived. For each of these peptides, we determined if it overlapped with a predicted TMH and if it was predicted to bind to the most frequent alleles of each MHC allele.

For MHC-I, 9-mers were used, as this is the length most frequently presented in MHC-I and was used in our earlier study (7). For MHC-II, 14-mers were used, as this is the most frequently occurring epitope length (19). A human (UniProt ID UP000005640_9606), viral (SARS-CoV-2, UniProt ID UP000464024) and bacterial (Mycobacterium tuberculosis, UniProt ID UP000001584) reference proteome was used. TMHMM (9) was used to predict the topology of the proteins within these proteomes. To predict the affinity of an epitope to a certain HLA allele, EpitopePrediction (7) for MHC-I and MHCnuggets (20) for MHC-II was used. Both MCH-I and MHC-II alleles were selected to have a high prevalence in the population, where the alleles of MHC-I are the alleles representing the 13 supertypes with over 99.6% coverage of the population’s MHC-I repertoire as defined by (1, 21), and the 21 MHC-II alleles, have a phenotypic frequency of 14% or more in the human population (22).

We define a protein to be a binder if, for a certain MHC allele, any of its 9-mer or 14-mer peptides have an IC50 value in the lowest 2% of all peptides within a proteome ( Supplementary Tables S1 , S2 ), this differs from our previous study where we defined a binder as having an IC50 in the lowest 2% of the peptides within a protein. This revised definition precludes bias of proteins that give rise to no or only very few MHC epitopes. To verify that the slight change in method yields similar results, a side by side comparison is shown in the Supplementary Materials , Supplementary Figures S1 , S2 .

To obtain experimental confirmation that peptides stemming from TMHs are presented by MHC-I and MHC-II, we mined the IEDB (18) for confirmed human MHC-ligands. We queried the IEDB for all linear epitopes obtained from MHC ligand assays in healthy humans, carrying the MHC alleles as used in this study. From these epitopes, we kept those that were present exactly once in the human reference proteome with UniProt ID UP000005640_9606. We concluded that the epitope overlapped with a TMH if at least 1 amino acid was overlapping with a TMH, as predicted with TMHMM (9), from which we concluded if the epitope is overlapping with a TMH with at least 1 amino acid.

The full analysis can be found at (https://github.com/richelbilderbeek/bbbq_article_issue_157).

Figure 1A shows the predicted presentation of TMH-derived peptides in MHC-I, for a human, viral and bacterial proteome. Per MHC-I allele, it shows the percentage of binders that overlap with a TMH with at least one residue. The horizontal line shows the expected percentage of TMH-derived epitopes that would be presented, if TMH-derived epitopes would be presented just as likely as epitopes derived from soluble regions, when assuming equal incidence of soluble and TMH-derived epitope presentation. For 11 out of 13 MHC-I alleles, TMH-derived epitopes are predicted to be presented more often than the null expectation, for a human and bacterial proteome. For the viral proteome, 12 out of 13 MHC-I alleles present TMH-derived epitopes more often than expected by chance. The extent of the over-presentation between the different alleles is similar for the probed proteomes, which strengthens our previous conclusion (7) that the hydrophobicity of the MHC-binding groove is the main factor responsible for the predicted over-presentation of TMH-derived peptides.

We next wondered if the over-representation of TMH-derived peptides would also be present for MHC-II. Figure 1A shows the percentages of MHC-II epitopes predicted to be overlapping with TMHs for our human, viral and bacterial proteomes. We found that TMH-derived peptides are over-presented in all of the 21 MHC-II alleles, for a human, bacterial and viral proteome, except for HLA-DRB3*0101 in M. tuberculosis. See Supplementary Table S5 for the exact TMH and epitope counts.

For MHC-I, we previously showed that the over-presentation of TMH-derived peptides is caused by the hydrophobicity of the peptide binding grooves (7). Figures 1B , C show the extent of over-presentation of TMH-derived epitopes as a function of the hydrophobicity preference score for the different human MHC alleles. An assumed linear correlation explains 88% of the variability in MHC-I. For MHC-II, 62% of the variability is explained by hydrophobicity. This indicates that TMH-derived peptides are over-presented, because the peptide binding grooves of most MHC-I and -II alleles are relatively hydrophobic.

The Immune Epitope Database (IEDB) from the National Institutes of Health contains millions of linear epitope sequences obtained by MHC ligand assays. For the MHC alleles used in this study, we obtained 54,303 and 2,484 linear epitope sequences for the MHC-I and MHC-II alleles from human origin respectively. There are relatively few epitopes for MHC-II, as MHC-II has many more different alleles than MHC-I, whereas we selected only the human epitopes found for the 21 MHC-II alleles used in this study.

Figure 2A and Supplementary Figure S4 show there are similar levels of over-presentation of TMH-derived epitopes between (1) the percentage of TMH-derived epitopes that is reported in the IEDB database versus (2) the percentage of TMH-derived epitopes that is predicted to be presented in MHC-I alleles. For MHC-II alleles, there were too few epitopes per MHC allele to result in an informative figure.

In Figure 2B we grouped all the epitopes presented by MHC-I and MHC-II alleles by the percentage of TMH-derived epitopes, which are 22% and 10%, respectively.

These findings robustly confirm that epitopes derived from human TMHs are presented in both MHC-I and MHC-II, and support that they are over-presented. See the Supplementary Table S6 for the exact values.

We also mined the IEDB database for epitopes for any type of T cell response from the specified alleles. From the total reports, 36% and 7% concerned TMH-derived epitopes in MHC class I and II, respectively (see Supplementary Figure S5 ).

This data confirms that not only TMH derived epitopes are presented on MHC, but this also elicits T cell mediated immune responses.

We addressed the question whether there is an evolutionary advantage in presenting TMHs. We determined the conservation of TMHs by comparing the occurrences of SNPs located in TMHs or soluble protein regions for the genes coding for membrane proteins. We obtained 911 unique gene names associated with the phrase ‘membrane protein’, which are genes coding for both membrane-associated proteins (MAPs, which have no TMH) and transmembrane proteins (TMPs, which have at least one TMH). These genes are linked to 4,780 protein isoforms, of which 2,553 are predicted to be TMPs and 2,237 proteins are predicted to be MAPs. We obtained 37,630 unique variations, of which 9,621 are SNPs that resulted in a straightforward amino acids substitution, of which 6,062 were located in predicted TMPs. See supplementary Supplementary Table S3 and S4 for the detailed numbers and distributions of SNPs.

Per protein, we calculated two percentages: (1) the percentage of a protein sequence length bearing TMHs, and (2) the percentage of SNPs located within these predicted TMHs. Each percentage pair was plotted in Figure 3A . The proportion of SNPs found in TMHs varied from none (i.e., all SNPs were in soluble regions) to all (i.e., all SNPs were in TMHs). To determine if SNPs were randomly distributed over the protein, we performed a linear regression analysis, and added a 95% confidence interval on this regression. This linear fit nearly goes through the origin and has a slope below the line of equality, which shows that less SNPs are found in TMHs than expected by chance.

We determined the probability to find the observed amount of SNPs in TMHs by chance, i.e., when assuming SNPs occur just as likely in soluble domains as in TMHs. We used a binomial Poisson distribution, where the number of trails (n) equals the number of SNPs, which is 21,208. The probability of success for the ith TMP (p_i), is the percentage of residues within a TMH per TMP. These percentages are shown as a histogram in Figure 3B . The expected number of SNPs expected to be found in TMHs by chance equals Σp ≈ 4,141. As we observed 3,803 SNPs in TMHs, we calculated the probability of having that amount or less successes. We used the type I error cut-off value of α = 2.5%. The chance to find, within TMHs, this amount or less SNPs equals 6.8208·10^–11. We determined the relevance of this finding, by calculating how much less SNPs are found in TMHs, when compared to soluble regions, which is the ratio between the number of SNPs found in TMHs versus the number of SNPs as expected by chance. In effect, per 1000 SNPs found in soluble protein domains, one finds 918 SNPs in TMHs, as depicted (as percentages) in Figure 3C .

We split this analysis for TMPs containing only a single TMH (so-called single-membrane spanners) and TMPs containing multiple TMHs (multi-membrane spanners). We hypothesized that single-membrane spanners are less conserved than multi-membrane spanners, because multi-membrane spanners might have protein-protein interactions between their TMHs, for example to accommodate active sites, and thus might have additional structural constraints. From the split data, we did the same analysis as for the total TMPs. Figure 4A shows the percentages of TMHs for individual proteins as a function of the percentage of SNPs located in TMHs. For both single- and multi-spanners, a linear regression shows that less SNPs are found in TMHs, than expected by chance.

We also determined the probability to find the observed amount of SNPs by chance in single- and multi-spanners. For single-spanners, we found 452 SNPs in TMH, where ≈ 462 were expected by chance. The chance to observe this or a lower number by chance is 0.319. As this chance was higher than our α = 0.025, we consider this no significant effect. For the multi-spanners, we found 3,351 SNPs in TMH, where ≈ 3, 678 were expected by chance. The chance to observe this or a lower number by chance is 8.31584·10^–12, which means this number is significantly less as explained by variation. The TMHs of multi-spanners are thus significantly more conserved than soluble protein regions, whereas this is not the case for single-spanners.

Also, for single- and multi-spanners, we determined the relevance of this finding by calculating where and how much less SNPs are found in TMHs when compared to soluble regions, as depicted in Figures 4B, C . In effect, per 1,000 SNPs found in soluble protein domains, one finds 978 SNPs in TMHs of single-spanners and 911 SNPs in TMHs of multi-spanners.

Although our data show that TMH-derived epitopes are presented in all classical MHC-I and MHC-II alleles, the molecular mechanisms of how integral membrane proteins are processed for MHC presentation are largely unknown (7). Most prominently, the fundamental principles of how TMHs are extracted from their hydrophobic lipid environments into the aqueous vacuolar lumen, leading to subsequent proteolytic processing are unresolved.

A first possibility is that the extraction of TMPs from the membrane is mediated by the ER-associated degradation (ERAD) machinery. For MHC class I (MHC-I) antigen presentation of soluble proteins, the loading of the epitope primarily occurs at the endoplasmatic reticulum (ER). The chaperones tapasin (TAPBP), ERp57 (PDIA3), and calreticulin (CALR) (26) first assemble and stabilize the heavy and light chains of MHC-I. Later, this complex binds to the transporter associated with antigen processing (TAP) leading to the formation of the so-called peptide-loading complex (PLC). The PLC drives import of peptides into the ER and mediates their subsequent loading into the peptide-binding groove of MHC-I (27). Membrane proteins first will have to be extracted from the membrane before they become amenable to this MHC-I loading by the PLC. In the ER, this process can be orchestrated by the ERAD machinery, consisting of several chaperones that recognize TMPs, ubiquitinate them, and extract them from the ER membrane into the cytosol (retrotranslocation) for proteasomal degradation (28, 29). Similar to the peptides generated from soluble proteins, the TMP-derived peptides might then be re-imported by TAP into the ER for MHC-I loading. This ERAD-driven antigen retrotranslocation might be facilitated by lipid bodies (LBs) (30), since LBs can serve as cytosolic sites for ubiquitination of ER-derived cargo (31).

A second possibility is that TMPs are proteolytically processed by intramembrane proteases that cleave TMHs while they are still membrane embedded. Supporting this hypothesis is the well-established notion that peptides generated by signal peptide peptidases (SPPs), an important class of intramembrane proteases that cleave TMH-like signal sequences, are presented on a specialized class of MHC-I called HLA-E (32). The loading of peptides generated by SPP onto MHC-I does not depend on the proteosome and TAP, possibly because the peptides are directly released into the lumen of the ER (32). However, this mechanism cannot explain how most membrane proteins can be processed for antigen presentation, because SPPs only cleave TMH-like signal sequences at their C-termini, and N-terminal domains will hence not be removed. Nevertheless, the presentation of peptides with a high hydrophobicity index was shown to be independent of TAP as well (33), suggesting that the TMH peptides might perhaps be released directly in the ER lumen by other intramembrane proteases.

A third possibility is that peptide processing and MHC-loading occur in multivesicular bodies (MVBs) (32). TMPs can be routed from the plasma membrane and other organelles by vesicular trafficking to endosomes. Eventually, these TMPs can be sorted by the endosomal sorting complexes required for transport (ESCRT) pathway into luminal invaginations that pinch off from the limiting membrane and form intraluminal vesicles. This thus results in MVBs where the membrane proteins destined for degradation are located in intraluminal vesicles. Upon the fusion of MVBs with lysosomes, the entire intraluminal vesicles including the TMPs are degraded (34). Via this mechanism, TMPs might well be processed for antigen presentation, particularly since the loading of MHC-II molecules is well understood to occur in MVBs (35–37). However, such processing of membrane proteins in MVBs for antigen presentation poses a problem, because complexes of HLA-DR with its antigen-loading chaperon HLA-DM were only observed on intraluminal vesicles, but not on the limiting membranes of MVBs (37), indicating that epitope loading of MHC-II also occurs at intraluminal vesicles. This observation hence raises the question how the intraluminal vesicles carrying the TMPs destined for antigen presentation can be selectively degraded, while the intraluminal vesicles carrying the MHC-II remain intact. A second problem is that phagosomes carrying internalized microbes lack intraluminal vesicles, and it is hence unclear how TMPs from these microbes would be routed to MVBs for MHC-II loading (37).

Alternatively to the enzymatic degradation of lipids in MVBs by lipases (38, 39), they might be oxidatively degraded by reactions with radical oxygen species produced by the NADPH oxidase NOX2 (40). This oxidation can result in a destabilization and disruption of membranes (40) and might thereby lead to the extraction of TMPs. Due to the hydrophobic nature of TMHs, however, the extracted proteins will likely aggregate and it is unclear how these aggregates would be processed further for MHC loading.

In general, one might expect that evolutionary selection shapes an immune system where surveillance is directed towards protein regions essential for the survival, proliferation and/or virulence or pathogenic microbes, as these will be most conserved. In SARS-CoV-2, for example, there is preliminary evidence that the strongest selection pressure is directed upon residues that change its virulence (41). These regions, however, may only account for a small part of a pathogen’s proteome. Additionally, the structure and function of these essential regions might differ widely between different pathogenic proteins. Because of this scarcity and variance in targets, one can imagine that it will be mostly unfeasible to provide innate immune responses against such rare essential protein regions, as suggested in a study on influenza (42), where it was found that the selection pressure exerted by the immune system was either weak or absent.

Evolutionary selection of pathogens by a host’s immune system, however, is more likely to occur for protein patterns that are general, over patterns that are rare. While essential catalytic sites in a pathogenic proteome might be relatively rare, TMHs are common and thus might be a more feasible target for evolution to respond to. Indeed, we have found the signature of evolution when both factors, that is, TMHs and catalytic sites are likely to co-occur, which is in TMPs that span the membrane at least twice. In contrast to single-spanners, where we found no significant evolutionary conservation, the TMHs of multi-spanners are more evolutionary conserved than soluble protein regions. Likely, the TMHs in many multi-spanners need to interact which each other for correct protein structure and function and they might hence be more structurally constrained compared to the TMHs of single-spanners. Thus, we speculate that the human immune system is more attentive towards TMHs in multi-spanners, as these are evolutionarily more conserved.

There have been more efforts to assess the conservation of TMHs, using different methodologies. One such example is a study by Stevens and Arkin (43), in which aligned protein sequence data was used. Also this study found that TMHs are evolutionarily more conserved, as the mean amino acid substitution rate in TMHs is about ten percent lower, which is a similar value as we found. Another example is a study by Oberai, et al. (44) that estimated the conservation scores for TMHs and soluble regions based on alignments of evolutionary related proteins, and also found that TMHs are more conserved, with a conservation score that was 17% higher in TMHs. Note that the last study also found that mutations in human TMHs are likelier to cause a disease, in line with our conclusion that TMHs are more conserved.

Together, from this study, two important conclusions can be drawn. First, the MHC over-presentation of TMHs is likely a general feature and predicted to occur for most alleles of both MHC-I and -II and for humans as well as bacterial and viral pathogens. Second, TMHs are genuinely more evolutionary conserved than soluble protein motifs, at least in the human proteome.