Understanding the principals that govern the generation and maintenance of functional genetic diversity is fundamental to evolutionary biology. Large reductions in population size, through bottleneck or founder events, result in a loss of genetic diversity 1 which may affect the ability of populations to adapt and survive in changing environments 1 2 . However, genes of ecological adaptive importance may maintain variation through a severe reduction in population size through processes such as balancing selection 3 4 . The Major Histocompatibility Complex (MHC) is such a functional locus, and has been extensively studied in both model and non-model species 5 6 7 .

The MHC is a multigene family involved in the vertebrate immune response 8 , and is the most polymorphic set of genes known in vertebrates 9 10 . MHC variation is driven by an arms race between host and pathogen, where balancing selection maintains alleles in the population. An extensive repertoire of alleles enables the population to respond rapidly to changing or novel pathogens 11 12 13 . The highly variable peptide binding region (PBR) encoded by MHC class II ß exon 2 (MHCIIß) ensures the binding of a large number of conformationally different peptides 8 . The PBR of MHC molecules is involved in antigen recognition and as such may be under strong balancing selection when compared with the non-PBR sites 14 . Although the major driving force behind MHC diversity is host-pathogen co-evolution 11 15 , sexual selection and selection against deleterious mutations also play a role in the maintenance of MHC variation 16 17 18 .

Like many multi-gene families, MHC is governed by the birth-and-death model of evolution where new genes are generated through gene duplication. Some of these genes are maintained for long periods and even through population divergence events, while others lose function (pseudogenes) or are lost completely. MHC variation is also governed by gene conversion, where homologous recombination occurs between duplicated genes (paralogous genes), thus homogenising sequences between different loci 6 19 . In passerine birds, the MHC is characterised by multiple gene copies, pseudogenes and long introns, and is exceptionally diverse and complex compared to other birds and vertebrate species 20 21 22 . Gene duplication events of MHC can be traced phylogenetically in most lineages, because duplicated genes evolve independently. This can be seen in the phylogenetic grouping of orthologous genes, rather than in a species-specific grouping 19 23 24 . Alternatively, recent duplication and concerted evolution of genes (through gene conversion) can result in species-specific clustering 6 22 25 26 . Due to the high rate of gene duplication and loss, and the confounding effect of gene conversion, it is notoriously difficult to re-construct avian MHC phylogenies 6 .

Following a bottleneck or founder event, the genetic diversity of a population is reduced to only a subset of the original variation. As the population adapts to its new environment, the MHC allelic diversity will be made up of a combination of ancestral polymorphism and novel genetic variation. Trans-species evolution 27 or ancestral polymorphism 28 refers to the long-term maintenance of ancestral alleles in populations and species 29 30 . This process is governed by balancing selection 31 and is seen when related species or subspecies share similar or the same MHC alleles despite local selection pressure. This pattern is common in mammals which do not often show concerted evolution, thus orthologous loci can be recognized between distantly related taxa such as mice and humans 24 . The high levels of concerted evolution in birds often make it difficult to distinguish between orthologous and paralogous loci 25 , although isolated cases have been reported e.g. 5 32 . Novel genetic diversity is introduced in populations either through dispersal or mutations. Mutational processes include gene duplication, point mutations and gene conversion e.g. 26 33 . Gene conversion is known to occur frequently in birds at the highly duplicated MHC genes 6 26 34 35 . The rate of gene conversion has been shown to be far greater than that of point mutations, thus may be a very important mechanism for generation of variation in bottlenecked populations 9 26 .

In the present study, we assess MHC variation in the Nesospiza bunting species complex and its putative sister taxon, Rowettia goughensis. Evaluation of the MHC in Nesospiza and R. goughensis is interesting for several reasons. Nesospiza and R. goughensis are considered sister taxa and are presumed to have arrived at Tristan da Cunha and nearby Gough Island with the same colonization event 36 . Mitochondrial cytochrome b sequences are reciprocally monophyletic between island systems, and neutral microsatellite markers show substantial genetic differentiation between species 37 38 . It is thus interesting to compare the MHC differentiation and allele sharing in Nesospiza and R. goughensis and determine the level of ancestral polymorphism between these species. Further, Nesospiza buntings have undergone an ecological adaptive radiation in parallel on two islands 37 . Both Nightingale and Inaccessible islands are inhabited by large- and small-billed Nesospiza buntings. The two species on Nightingale Island (N. questi and N. wilkinsi) co-occur with little, if any, interbreeding, probably due to the availability of two discrete seed sizes within a single habitat. Inaccessible Island has three lineages of N. acunhae buntings: large-billed N. a. dunnei, and two colour morphs of the small-billed bunting, N. a. fraseri and N. a. acunhae 37 39 . Hybridisation occurs between all three forms across an ecotone on the eastern plateau of Inaccessible Island. This is probably due to a large variation of seed sizes occurring at low densities, which favours greater diversity in bill-sizes 37 . A single Nesospiza species inhabited the main island of Tristan, but was driven to extinction shortly after the arrival of humans at the archipelago. Genetic structure analysis based on neutral microsatellite markers show little or no hybridization between species on Nightingale, and strong differentiation between Nightingale Nesospiza and those on Inaccessible Island 37 38 . Despite ongoing hybridization on Inaccessible Island, a strong association has been found between bill morphology, habitat choice and genetic differentiation suggesting that both natural and sexual selection may maintain differentiation 37 38 . Thus, it is possible that these selective pressures will result in species-specific patterns of MHC variation. However, an alternative hypothesis is that balancing selection has maintained most of the MHC variation across the species complex. Here we aim to 1) test for signatures of selection at the MHCIIß in Nesospiza buntings, and 2) investigate the extent of ancestral polymorphism between Nesospiza, its putative sister taxon Rowettia goughensis, and other passerine species 5 32 34 35 40 41 .

This study describes 23 MHCIIß alleles representing at least four functional loci and two pseudogenes in the Nesospiza bunting species complex. Many MHCIIβ alleles were shared between Nesospiza taxa as well as between Nesospiza and its putative sister taxon R. goughensis. This pattern of ancestral polymorphism suggests that the observed gene duplications occurred prior to the phylogenetic split of the species, and subsequent unusually low selective pressure at the loci has prevented allelic divergence between species. The MHC nuclear genetic diversity in Nesospiza on Inaccessible (π = 0.11) was comparable to that of outbred passerine species (e.g. 0.15 in Luscinia svecica; 5 ), and despite the low sample size for Nightingale, allele numbers and nucleotide diversity were higher than in the severely bottlenecked Chatham Island robin population (0.05) 35 . We have screened 14 Nesospiza individuals for MHC variation, which is similar to some previous Passerine MHC studies using cloning and sequencing e.g. 34 35 43 47 . However, because larger sample sizes would have been necessary to cover the variation of each population sufficiently, we will not discuss population-level MHC variation further.

Patterns of both ancestral polymorphism and concerted evolution among Nesospiza and Rowettia populations are evident from our results. Ancestral polymorphism, found here for Nesospiza and R. goughensis, as well as in other species (e.g. great reed warbler, house finch, vegetarian finch, red-winged blackbird and common yellowthroat), can be seen in the sharing of the same or similar alleles between species (Figures 2 and 3). Of the 23 Nesospiza alleles, 15 were found in species from both islands. All seven alleles occurring in R. goughensis are shared with Nesospiza (Neso5, 9, 13–15, 17, 23) and these alleles are found in all three Nesospiza groups in the gene tree (Figures 2 and 3). The estimated minimum number of putatively functional gene copies in Nesospiza (i.e. 4 loci) suggests that the three Nesospiza allele groups are not necessarily locus-specific, despite their divergent clustering. Group 3 may represent a single locus, since only one or two alleles from this cluster occur in each individual. However, this is not the case for R. goughensis, where three of these alleles occur in one individual. Two highly supported clusters are seen within Group 2 (Figure 3), which is also the cluster containing the most alleles, suggesting that this cluster is likely to represent more than one gene copy. A likely explanation for the clustering of alleles from different gene loci is the genetic homogenization caused by gene duplication events with subsequent gene conversion.

The highly supported branches of sequences forming Groups 2 and 3 in the gene tree contain only Nesospiza and R. goughensis alleles. Although several species were included due to the similarity between their MHCIIβ alleles and those of Nesospiza, the observed divergent clustering of Group 2sequences could be explained by a lack of closely related species in the analysis. Alternatively, the species-specific clustering of Nesospiza may be attributed to their long divergence time from the other passerines sampled 48 . The deep divergence of Groups 2 and 3, and the clustering of Group 3 close to the distantly related species of common yellowthroat, New Zealand robins, Florida scrub jay, and vegetarian finch, however, provide strong support for preserved ancestral polymorphism. These patterns suggest that extant MHC variation in Nesospiza and R. goughensis can be explained by shared ancestral polymorphism during colonisation which has since been maintained. It is possible that the additional variation has been generated by gene conversion events, which is the most likely method of generating variation from the few alleles remaining in a population following a population bottleneck 26 .

Amino acid sequences are more similar between Groups 1 and 3 (Figure 4). This could either represent evidence of recombination with the pseudogenes, producing a new group of functional sequences, or perhaps more likely indicate that the pseudogenes resulted from gene duplication events of Group 3 sequences. Copying errors during gene duplication and recombination events may result in non-functional genes (pseudogenes) and the subsequent lack of functional constraint on evolutionary processes (such as mutation) acting on the pseudogenes result in rapid sequence divergence 49 . This is evidently the case for the two presumably non-functional alleles, Neso22 and Neso23, which form a well supported group with a red-winged blackbird pseudogene, clustered sister to all the functional passerine sequences. However, some pseudogenes (e.g. red-winged blackbird APAF030994 and vegetarian finch PCAY064439) do not show evidence of rapid divergence (Figure 3), perhaps due to ongoing recombination with functional genes that is leading to sequence conservation. Alternatively, there may have been insufficient time for the genes to become highly diverged since they became non-functional.

Selection tests showed no consistent evidences of balancing or positive selection at the PBR or non-PBR regions of MHCIIβ exon 2 in Nesospiza and Rowettia. The short fragment length of our sequences excludes some of the PBR sites, and therefore there is a chance that some sites that may be under selection were excluded from the analyses. However, selection tests were done according to two different PBR characterizations 44 45 , and tested on the entire data set as well as all species individually, and the three clusters independently. Ratios of d_N/d_S were non-significant in all cases (Table 1), and additional selection tests showed weak evidence of selection with only one site likely to be under positive selection (Table 2). New MHC variation can be generated by point mutations or through recombination between alleles, giving rise to a new allele 26 33 . The latter process, known as gene conversion, has been documented in some natural avian populations 22 25 26 and has been suggested to be essential in generating genetic variation at MHC after a bottleneck 26 . During gene conversion events, synonymous substitutions may hitchhike with non-synonymous variation 26 and this may be a reason why d_N/d_S ration tests fail to detect positive selection. We found, however, no evidence of recombination in our data, but recombination can be difficult to verify with short sequences.

Despite the lack of significant evidence for selection, ratios of d_N/d_S > 1.0 that we observe in Rowettia and all Nesospiza populations indicate that the loci are under weak balancing selection, or perhaps more likely, that ancestral balancing selection acted on the loci before colonisation of the islands. Lack of strong positive selection may reflect a decreased pathogen load in both Nesospiza and R. goughensis. Passerines generally are less parasitised by lice and ectoparasites than other avian orders e.g. 50 . This is particularly true of small populations on isolated oceanic islands (R Palma pers. comm.). Myrsidea lice occur at extremely low prevalence (6.4%) across 12 species of Darwin’s finches at the Galápagos Islands 50 . On Tristan da Cunha and Gough Island, different louse species (order Phthiraptera) have been found on 20 bird species, including the Tristan thrush (Nesocichla eremita) 51 , yet careful inspection of Nesospiza buntings yielded no lice, with hippoboscid flies and feather mites the only ectoparasites (PG Ryan unpubl. data). The absence of parasites could be due to an uninfected founding population (“missing the boat”) 52 , or subsequent extinction from the host after colonisation. The high level of ancestral polymorphism between R. goughensis and Nesospiza suggest that the former is more likely, where a single uninfected founding population colonization both Tristan da Cunha and Gough Island.

Some shortcomings of the cloning and sequencing method employed in the study may result in underestimation of MHC variation. Firstly, the large number of gene copies and the high level of convergence between loci make it difficult to amplify a single MHC locus at a time. Thus, most MHC studies on non-model vertebrates amplify alleles from multiple gene copies simultaneously. This increases the risk of chimera formation during the PCR, which in turn leads to overestimation of levels of gene recombination 53 . In addition, PCR products are prone to point mutations e.g. 54 , although these are relatively easy to detect since mutation rates are relatively low and are unlikely to occur in more than one sequence 55 56 . In this study, we compensate for these problems by only accepting alleles that occur in at least two individuals e.g. 57 58 . Secondly, the amplification of a multi-gene family is necessarily problematic since not all loci and not all alleles at a locus will be detected using a single primers set. The primers employed in this study were designed for non locus-specific amplification of exon 2 of MHCIIß in zebra finch (Taeniopygia guttata) 59 and have been successfully employed in other passerine MHC studies (H Westerdahl pers. comm.). A regression analysis of the number of clones sequenced per individual found that more individuals and sequences will be necessary to estimate true MHC variation per individual. Finally, sequences were obtained for only half of the variable MHCIIβ exon 2 gene. Although not all the variation has been analysed in this study, this is often the case with such complex multi-gene systems 58 and does not preclude our finding of ancestral polymorphism between species and within the Nesospiza species complex. More comprehensive studies of population level variation of MHC would require that more individuals and sequences were analysed. However, the present study focuses on selection and levels of shared polymorphism, and for such analyses the present data is sufficient.

The extent of shared alleles and ancestral polymorphism between Nesospiza and R. goughensis suggests that both originated from the same colonization wave. We find that similar or the same alleles are maintained between species due to the recent species divergence and low levels of (local) selection acting on PBR. The additional variation found within the Nesospiza species complex may be due to gene conversion, which is likely the most prominent mechanism for generating new variation after a bottleneck event 26 . The extant genetic variation is not likely to change rapidly, unless there is a drastic geographic or environmental change leading to strong selection at the MHC. One such situation would be the introduction of pathogens, since populations with low MHC diversity are often more susceptible to novel pathogens 35 60 . In the absence of strong selection, MHC is expected to diverge over time between islands and populations due to drift, with the generation of new haplotypes through point mutations or gene conversion. Ongoing gene flow between populations and subspecies on Inaccessible Island can maintain genetic variation to some extent. The potential role of MHC dependent sexual selection 22 61 to drive divergence between populations even further remains open to study, and would require wider sampling over the entire geographic range to cover the details of geographic- and species-specific variation.

AJvR carried out the molecular lab work, statistical analyses, and drafted the manuscript. BH, PB, and PGR conceived of the study and participated in its design. BH participated in the coordination of the study and helped to draft the manuscript. All authors read and approved the final manuscript.