HLA-E, H2-Qa1 and RT-BM1 occupy similar positions within the MHC

Early comparative maps of the human and mouse MHC placed the HLA-E locus in a much more telomeric position than the H2-Qa1 locus [for example, see Figure 3 of [9]]. This contributed to the general perception that these two loci were unlikely to occupy conserved syntenic positions, and were therefore unlikely to be orthologues (we use this term to indicate that sequences derived from the same ancestral locus). If the positions of invariant genes within the MHC region are taken as a reference point [29], however, the CD94L loci HLA-E, H2-Qa1 (also called H2-T23) and RT-BM1 are all found in the very same genetic location: the portion of the MHC class I 'island' between GNL1 and RPP21 [30, 31]. On the grounds of genetic mapping, therefore, there is no reason to believe that the primate and rodent loci for CD94L could not be bona fide orthologues.

Certain residues are CD94L-specific, and others are homogenised within species

To investigate the evolutionary relationship between the CD94L and class Ia loci, we searched the nucleotide database to identify species in which sequences were available covering the whole extracellular portions (domains α1, α2 and α3) of both CD94L and class Ia molecules. This search yielded four rodent species [mouse (Mus musculus), rat (Rattus norvegicus), Chinese hamster (Cricetulus griseus), deer mouse (Peromyscus maniculatis)] and four primate species [human (Homo sapiens), chimpanzee (Pan troglodytes), Rhesus macaque (Macaca mulatta), and cotton-top tamarin (Saguinus oedipus)]. CD94L and/or class Ia sequences are available for many other rodent and primate species, but either as partial sequences, or only for either a CD94L or class Ia molecule(s). Attempts to identify potential CD94L molecules in species outside of the primate and rodent genera by homology searches in sequence databases yielded no promising candidates (not shown).

For each of the species listed above, in addition to the CD94L sequence, we picked one representative of each of the characterised class Ia loci. For the hamster and deer mouse, because the definite locus information was lacking, we picked sequences that were as different from one another as possible (two for hamster, and three for deer mouse). To extend our investigation beyond the CD94L-type class Ib molecules, we also included the rat RT1-M3 and mouse H2-M3 class Ib sequences. RT1-M3 and H2-M3 are clear orthologues, and are both specialised in the presentation of bacterial N-formylated peptides [32].

Altogether, we selected 29 sequences, and aligned the protein sequences of their extracellular domains (Fig. 2). All these sequences are closely homologous to one another, with no gaps or deletions in the alignment, and there is perfect consensus between all the class I molecules at 87 positions (32.2%).

At 24 positions along the sequence (indicated by a blue number above the alignment), some residues have clearly been conserved among the CD94L molecules (in blue letters) within either the rodent or the primate orders (for example, see positions 116 and 140 for rodent CD94L, and 124 and125 for primate CD94L). At six of these positions (68, 69, 70, 143, 147, 155), the same residues are found in either seven or all eight of the primate and rodent CD94L sequences.

Given the previously reported observation that HLA-E, H2-Qa1 and RT-BM1 are each more closely related to their respective class Ia counterparts, we also inspected the alignment to get an idea of the localisation of the species-specific residues (Because the number of class Ia sequences included in this alignment is very small compared to all those known in the different species, the list of residues we identified in this manner should, however, only be considered as indicative rather than exhaustive and/or definitive). Such positions, where all the shown class Ia and Ib sequences from at least one species are identical to one another but diverge from the general consensus, are indicated by a red number above the alignment (for example, position 52 is a valine in all three macaque sequences, and position 54 is an arginine in all three hamster sequences). Fourteen such positions occur within the first 180 residues, which corresponds to the α1/α2 codomain that constitutes the PBR. There is very little room for doubt that all four primate CD94L genes descend from a common ancestral gene, and similarly for all four rodent CD94L genes. Both in the primate and in the rodent ancestral species, there must, therefore, already have been a co-existence of the loci for ancestral CD94L molecules and for ancestral class Ia molecules. The most likely explanation for residues found in the class Ia and CD94L molecules of only one species is the occurrence of gene conversion between the various MHC class I genes in that species, including those for class Ia and CD94L molecules, resulting in localised homogenisations, which would ultimately lead to concerted evolution.

Remarkably, within the α3 domain, which separates the PBR from the transmembrane domain (last 90 residues of the alignment), 22 positions have evolved in a species-related manner, and no single position is found among either the rodent or primate CD94L groups despite the fact that separate loci for class Ia and CD94L must have pre-existed the species' radiation within each genus.

From these observations, we conclude that, at least for the α3 domain, the CD94L have all undergone intra-species concerted evolution with their respective class Ia molecules. This phenomenon is very evident among rodent sequences, but it can also be observed within the primate ones (e.g. at positions 182, 194 and 228). The identification of 14 positions along the α1/α2 codomain that have undergone similar intra-species homogenisation further suggests that this type of phenomenon is likely to be true also of sequences outside exon 4.

Different topological distributions of the CD94L- and species-specific residues

We analysed the locations on the MHC class I molecule of those residues that have been conserved among CD94L molecules and those that evolve in a species related manner (Fig. 3A and 3B). The residues that appear to undergo intra-species concerted evolution are all located outside the PBR, whereas most of the residues that are conserved amongst CD94L molecules are situated at positions within the PBR. Three noticeable exceptions are positions 1, 90 and 91.

The observation that the residues of the α1/α2 codomain that undergo concerted evolution mostly tend to occupy positions on peripheral loops suggests that these positions may correspond to sites of interaction of the class I molecules with other proteins.

Remarkably, most of the α3 domain residues that have undergone concerted evolution occur furthest away from the PBR (Fig. 3B), and a handful of others lie at the CD8 interaction site (area indicated by a blue dotted oval) [33]. According to the recent report by Mitra et al[34], MHC class I molecules are not 'standing up' on the plasma membrane as is generally presumed, but adopt a supine position, as shown here. In this conformation, the portion of the α3 domain where residues have a strong tendency to undergo concerted evolution would be eminently accessible to interactions with one or several molecular partners.

Different portions of MHC class I molecules undergo different patterns of evolution

Using the protein alignment in Figure 2, we carried out successive phylogenetic comparisons focusing on various portions of the proteins. The tree shown in Figure 4A is a graphical representation of the widely accepted divergence pattern of the eight species in this study, together with the divergence times estimated according to the fossil record. We first compared the entire extracellular portions (270 amino acids; α1+α2+α3 domains) of the 29 sequences together with the pig class Ia sequence SLAI (used as an 'outlier') (Fig. 4B). Our findings confirm the previously reported observation that class Ia molecules (red) and class Ib molecules (blue and green) co-segregate within an order [11, 12].

Among the class Ia molecules the only orthologous relationships suggested by this analysis are between the human and chimpanzee B loci (HLA-B07 and Patr B), and human, chimpanzee and macaque A loci (HLA-A2, Patr A2 and Mamu A01), with no locus orthology suggested within rodents. Among the class Ib molecules, however, the independent groupings of CD94L sequences for primates and for rodents establishes beyond reasonable doubt that, within each group, all four loci must derive from a common ancestral CD94L locus. The fact that the comparison of primate sequences strongly suggests that the four CD94L are orthologues, whereas this is much less clear for the corresponding class Ia sequences confirms previous reports that primate CD94L molecules are much more evolutionarily conserved than are class Ia molecules [10, 35, 36], and extends this observation of high conservation to rodent CD94L and to the M3 class Ib molecules found in mouse and rat. If HLA-E and H2-Qa1 arose by convergent evolution, this must not only have taken place earlier than the intra-order species divergence, roughly 50 million years ago, and after the primate-rodent split 100 million years ago.

Given the observations collected from the sequence alignment in Figure 2, which suggested that the sequence of the α3 domain may evolve differently from that of the α1/α2 codomain, we next generated phylogenetic trees to compare separately the evolution of these domains (Fig. 5). Comparison of the α3 domains reveals a strong tendency of the class Ia sequences (red) and class Ib (blue and green) to group by species. This observation confirms our earlier conclusion from direct inspection of the sequence alignment;i.e. that concerted evolution within species does clearly take place between class Ia and class Ib sequences, at least for the sequences of the α3 domains. This is supported with very strong significance by the groupings of the tamarin and hamster CD94L molecules with their respective class Ia counterparts rather than with their respective primate and rodent CD94L functional homologues. The clade containing all the murine MHC molecules, class Ia, M3 and CD94L also provides a further clear proof of the intra-species homogenisation of the α3 domains, although, in this latter case, the H2-D^d molecule clusters with the rat MHC class Ia sequences rather than with those of other mouse class Ia.

We then constructed a phylogenetic tree for the first 180 amino acids that constitute the α1–α2 codomain (Fig 5B). The most remarkable difference between the tree shown in Figure 4B and this one is that the removal of the species-specific sequences of the α3 domain results in the rodent and primate CD94L molecules (blue) falling much closer together, and away from the clades of their respective class Ia counterparts. This finding already suggests that these two sets of class Ib molecules may well share a common ancestor.

The presence of the rat and mouse M3 molecules (green) on the same branch as the group of rodent CD94L could indicate that the M3 and CD94L loci arose by gene duplication in a rodent ancestor before the rat-mouse divergence. An alternative explanation is that the CD94L and M3 loci have coexisted long enough for concerted evolution to drive their homogenisation. Yet another interpretation – turning the tree on its head – is that the M3 molecules may not only go back a long time, but the structural requirements imposed by their function may be sufficiently tight that their evolution could have been much slower than that of the other molecules on the tree. In this case, these sequences would be closer to an 'ancestral' MHC sequence, with the CD94L sequences moving slowly away from it, and those of class Ia molecules much faster. The clustering of the mouse H2-M3 α3 domain sequence with those of hamster sequences rather than that of other mouse sequences could be seen as a further, albeit weak, argument in support of this argument.

Given the observation that certain residues of the α1–α2 codomain that are situated outside the PBR tend to evolve by species, we carried out separate comparisons for those 67 residues that are generally considered as being involved in the PBR, and for the 113 residues outside the antigen recognition site (ROARS; see Materials and Methods for the precise list).

The tree obtained by comparing the ROARS (Fig. 6A) has roughly the same overall structure as that derived from comparing the three domains together (Fig. 4B), with the rodent sequences clearly separated from the primates. The class Ia molecules (red) have a strong tendency to cluster by species, and the class Ib (blue) by function within each order, the exception being Saoe E that clusters with HLA-C. This clustering of class Ia loci by species is probably in good part explained by gene duplication and contraction, as proposed in the birth and death model [18]. Given our observation of certain species-specific residues harboured by both class Ia and CD94L molecules, however, we believe that concerted evolution within species could occur significantly for these ROARS, thereby also contributing to the sequences' tendency to group by species.

Gene conversion can result in the transfer of as few as half a dozen nucleotides to as many as several hundred [37]. For the sequences coding the α1–α2 codomain, the interspersed arrangement of PBR and ROARS residues must greatly influence the size of the DNA segments that can be exchanged to result in ROARS homogenisation. For class Ib molecules, the need to preserve their defined antigen-binding specificity would mean that PBR residues could not undergo gene conversion with other loci. The size of the DNA segments being exchanged would then be limited to those fitting in between PBR residues, resulting in a slower rate of evolution than for the α3 domain, where the size restriction would not apply. For class Ia molecules, however, the very intense evolutionary processes that drives their PBR residues towards heterogeneity would result in an acceleration of the homogenisation of ROARS compared to the α3 domain residues.

Comparison of this tree (Fig. 6A) to that for α3 domains (Fig. 5A) indeed hints that class Ia ROARS sequences homogenise even faster than their α3 domains. Firstly, mouse H2-D^d falls among rat sequences in the α3 domain analysis but clusters with the other mouse class Ia molecules for the ROARS. Similarly, the Mamu A01 sequence clusters with human and chimpanzee A sequences for the α3 domain but with the Mamu B04 sequence for the ROARS (albeit with low support values).

When the 67 PBR residues are compared (Fig. 6B), all CD94L (blue) fall in one clearly defined group. This grouping could be explained either by convergent evolution, or by the fact that all eight CD94L genes derive from a common ancestor. Since the murine and human CD94L loci occupy conserved syntenic positions within the MHC, and since concerted evolution can explain the homologies of MHC class I sequences found overall within species, we conclude that all eight CD94L molecules almost certainly derive from a common ancestor, without the need to call upon convergent evolution. A further argument supporting this conclusion is the arrangement of the CD94L sequences on this tree that matches exactly the divergence of the species (Fig. 4A). Comparison of the PBR of the class Ia molecules (red) gives an indication of their exuberant evolution, with some primate sequences (Saoe 4, Saoe 6, Saoe 8 and Mamu B04) intermingled with rodent sequences (with, however, exceedingly low support values for the branches), and Mamu-A01 resembling most closely the pig class Ia molecule we picked as an outlier. From this type of observation, we conclude that phylogenetic trees are not necessarily the best tool to decipher the evolution of residues that undergo extremely rapid evolution such as the PBR residues of class Ia molecules.

The PBR residues of primate and rodent CD94L molecules have been actively conserved from a common ancestral sequence

Comparing the rates of synonymous versus non-synonymous (ds/dn) substitutions in nucleic acid sequences has been used to great effect to investigate the nature of the selective pressures that drive the evolution of MHC molecules, particularly to demonstrate that the PBR residues of class Ia molecules are under positive selective pressure [17]. If our conclusion that the rodent and primate CD94L molecules derived from a common ancestor is correct, and the antigen presenting function of these molecule has been conserved for over 50 million years, then we predict that their PBR residues would show signs of negative selective pressure.

Although the statistical interpretation of this distribution is difficult, it is certainly consistent with our conclusion that all eight CD94L most likely derive from a common CD94L ancestor molecule, and have all evolved under similar global evolutionary pressures since the rodent-primate radiation that took place over 100 million years ago.

Concerted evolution to reinterpret previous data

Since the polymorphism of the HLA-E locus apparently predates most HLA-A and HLA-B polymorphism [10], HLA-E can be considered as quite an ancient locus, which fits with our conclusion that it must have been present in the rodent/primate ancestor. In 1989, Hughes and Nei pointed to a probable interlocus exchange of the whole exon 4, coding for the α3 domain, between HLA-A and HLA-E [38]. The sequence alignment shown in Figure 1 of their paper (see Additional file 1), however, is more consistent with gene conversion than with an overall exon swap by recombination between HLA-A and HLA-E. Over the first 21 nucleotides of exon 4 (red box), HLA-E is much more closely related to HLA-B, HLA-C and the Patr B sequence than to HLA-A or Patr A. Over the following 93 nucleotides (blue box), all six sequences are almost completely homologous to one another. The relatedness of the HLA-E α3 domain to those of HLA-A and Patr A is, in fact, due almost entirely to eight sites in the subsequent 99 nucleotides where the HLA-B, HLA-C and Patr B sequences diverge from these three sequences. The following 12 nucleotides are completely homologous between the six sequences (blue box), and over the last 48 nucleotides of exon 4 (red box), the HLA-E sequence matches once again better with HLA-C than with HLA-A. Given that the length of the DNA segments exchanged by gene conversion can concern any number from less than 10 to several hundred nucleotides [37], this 'mosaicism' of the human and chimpanzee sequences is, therefore, much more suggestive of events of gene conversion than of exon swapping, and the extensive occurrence of such events appears to be the most likely source of the concerted evolution we witness for the α3 domain.

The fact that gene conversion contributes to the diversity of MHC molecules is broadly accepted. The frequency of past gene conversion events in human and chimpanzee MHC class I sequences was even evaluated statistically by Kuhner et al. [39], and matching of the data on polymorphism obtained in mammals to mathematical simulations has suggested that the rate of allelic conversion is relatively high [40]. Although gene conversion can contribute to the generation of new MHC class I alleles by shuffling DNA sequence motifs, thereby enhancing heterogeneity, the very same mechanism can also result in homogenisation of sequences within a multigene family. Since gene conversion does not actually mutate DNA but simply shuffles motifs between existing loci, we can see that this phenomenon has no intrinsic effect on increasing or decreasing heterogeneity of individual residues. What gene conversion actually does is to accelerate the rate of evolution of MHC molecules, and it is the selective pressures on the function of MHC molecules that must drive certain portions of these molecules towards homogeneity or heterogeneity.

Concerted evolution and protein function

Several hypotheses have been proposed to explain the fact that MHC molecules are the most diverse and fastest evolving genes in mammalian genomes, including pathogen driven evolution, high spontaneous mutation rate, frequent gene conversion, mating preferences and materno-foetal incompatibility [41]. These hypotheses are not mutually exclusive and probably all contribute to a certain extent to MHC diversity, but the main driving force is clearly the stringent selective pressure imposed by infectious pathogens. Several models have been put forward to explain how the pressure of fast evolving pathogens drives MHC diversity, for example balancing selection, over dominant selection and rare allele advantage All these models are reviewed and discussed in a recent review by Jeffery and Bagham [42].

In their turn, many pathogens have devised several clever ways to block antigen presentation by making inhibitors that bind either directly to MHC molecules outside the PBR or to other molecular partners involved in antigen processing, presentation and recognition [43, 44], providing another level of selective pressure by pathogens on the immune system. Many of these proteins [e.g. the transporter for antigen processing (TAP), tapasin, calnexin, calreticulin, beta 2 microglobulin (β2M), CD8 and the T cell receptor for antigen (TCR)] must, therefore, also be under considerable evolutionary pressure to evade these inhibitory mechanisms.

If a pathogen evolves to block the antigen presentation/recognition pathway, individuals carrying mutations that allow them to evade this blockage will have a sizeable advantage to fight off the infection. Such advantageous mutations might occur either in one of the multiple loci coding for MHC class I molecules, or in one of the genes encoding proteins involved in the antigen presentation/recognition pathway (PRIAP). If a mutated form of a PRIAP arises as a result of pathogen selection, one potential consequence would be that the pool of MHC class I molecules in that species would no longer be optimally suited to interact with that PRIAP mutant. Evolutionary pressures would, in turn, favour individuals harbouring MHC molecules adapted to interact better with these new PRIAP forms. Such co-evolution has been reported for the genetically linked TAP and MHC class Ia loci in the rat [45] and the chicken [46]. In these circumstances gene conversion could spread advantageous motifs between MHC molecules within a given species, ultimately resulting in their concerted evolution.

In strong support of this model, in our study, all the positions evocative of species-specific evolution correspond to residues that would be readily accessible to interactions with other proteins, be they PRIAPs or inhibitors of antigen presentation derived from infectious pathogens (or both). Regarding the remarkable congregation of species-specific residues in the portion of the α3 domain furthest away from the PBR, this portion of MHC class I molecules is known to be involved with interactions with the peptide-loading machinery; position 222 is involved in interactions with tapasin [47] and position 227 with calreticulin [48]. Residue 182, which also undergoes intra-species homogenisation, is located next to the site bound by the inhibitory protein US2 from human cytomegalovirus [49]. Our observation that three positions in the α1–α2 codomain that clearly lie outside of the PBR (1, 90 and 91) tend to be conserved among CD94L molecules could indicate sites of interactions with partners specific to those molecules, possibly with the CD94-NKG2 receptors themselves.

NK receptors are amazingly diverse both within and between species, and they have almost certainly co-evolved with the MHC molecules they recognise [50–53]. Certain residues of MHC class I molecules that evolve in a species-specific manner may therefore be involved in interactions with NK receptors.

Our results suggest that, in MHC molecules, different subsets of residues are under different selective pressures. As previously shown by Hughes and Nei, the PBR residues of class Ia molecules, which present highly variable antigens, are under positive selective pressure, favouring heterogeneity of their antigen-binding properties [17]. Conversely, in class Ib molecules, which present precisely defined antigens such as N-formylated peptides or the leader peptides of class Ia molecules, the same PBR residues are under negative selective pressure to ensure maintenance of function. Similarly, for class Ia molecules, the ROARS residues of the α1 and α2 domains undergo frequent gene conversion and evolve fast, whereas in class Ib molecules, which are less variable, the ROARS residues evolve more slowly. We predict that the residues of the α3 domain will evolve at similar rates in class Ia and class Ib molecules, between those of the fast class Ia and the slow class Ib ROARS.

By contributing to a better understanding of how MHC molecules evolve, we hope that our results may help to decipher where and how our adaptive immune system arose, and keeps evolving in the face of the permanent challenge of infectious organisms or, as the case may be, the lack of them that may well favour allergic or auto-immune conditions.

Reviewer's report 1

Reviewer's report 2

Reviewer's report 2