Demography, migration and natural selection are predominant processes affecting the distribution of genetic variation among natural populations. Many studies use neutral genetic markers to make inferences about population history. However, the investigation of functional coding loci, which directly reflect fitness, is critical to our understanding of species' ecology and evolution. Immune genes, such as those of the Major Histocompatibility Complex (MHC), play an important role in pathogen recognition and provide a potent model system for studying selection. We contrasted diversity patterns of neutral data with MHC loci, ELA-DRA and -DQA, in two southern African plains zebra (Equus quagga) populations: Etosha National Park, Namibia, and Kruger National Park, South Africa. Results from neutrality tests, along with observations of elevated diversity and low differentiation across populations, supported previous genus-level evidence for balancing selection at these loci. Despite being low, MHC divergence across populations was significant and may be attributed to drift effects typical of geographically separated populations experiencing little to no gene flow, or alternatively to shifting allele frequency distributions driven by spatially variable and fluctuating pathogen communities. At the DRA, zebra exhibited geographic differentiation concordant with microsatellites and reduced levels of diversity in Etosha due to highly skewed allele frequencies that could not be explained by demography, suggestive of spatially heterogeneous selection and local adaptation. This study highlights the complexity in which selection affects immune gene diversity and warrants the need for further research on the ecological mechanisms shaping patterns of adaptive variation among natural populations.
Citation: Kamath PL, Getz WM (2012) Unraveling the Effects of Selection and Demography on Immune Gene Variation in Free-Ranging Plains Zebra (Equus quagga) Populations. PLoS ONE 7(12): e50971. https://doi.org/10.1371/journal.pone.0050971
Editor: Norman Johnson, University of Massachusetts, United States of America
Received: July 4, 2012; Accepted: October 29, 2012; Published: December 14, 2012
This is an open-access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Funding: This research was supported by a National Institutes of Health Ecology and Evolution of Infectious Disease Grant (GM083863) to WMG and a National Science Foundation Doctoral Dissertation Improvement Grant (MCINS-20091291) to PLK. The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: Wayne M. Getz currently serves as a PLOS ONE Editorial Board member. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.
Spatiotemporal fluctuations in pathogens generate complex patterns in the distribution of gene variation among host populations. Pathogens affect hosts not only by driving changes in population size, but also by eliciting an evolutionary response in host immune genes. Stochastic demographic factors (e.g. chance events associated with births, deaths and movement) influence, in the same way, rates of gene flow and genetic drift at all loci in an organism's genome , . Thus, neutral genetic data is informative for studying demography , for example, to examine relatedness, mating behavior, dispersal patterns, changes in population size, population genetic structure and speciation –. It is functional gene variation, however, that reflects natural selection, fitness and the potential to adapt to changing environments, making it of critical importance to the study of evolution, ecology and conservation , . Pathogen-mediated selection is expected to affect specific genomic regions, depending on gene function and the magnitude of the fitness consequence to the host. Contrasting patterns of diversity at neutral and immune response loci can be particularly informative for understanding how hosts adapt to pathogens, by illuminating the relative effects of selection versus demography on gene variation and population dynamics –.
The Major Histocompatibility Complex (MHC) is a family of genes that plays a critical role in vertebrate immune function. Class II MHC genes encode cell-surface glycoproteins, with peptide binding regions (PBRs) that are responsible for recognizing foreign antigens from extracellular pathogens (e.g. eukaryotic parasites and bacteria) and presenting them to helper T-lymphocytes to elicit an immune response. These genes typically exhibit elevated levels of polymorphism and highly divergent variants have been shown to persist over long time periods, even through the course of speciation events , a phenomenon largely explained by balancing selection . Variation in the MHC has been shown to be concentrated within the PBR , suggesting that pathogen recognition is a selective driving mechanism, with increased allelic diversity allowing for recognition of a broader spectrum of pathogens . Beyond implications for modulating disease resistance, MHC variants are also known to influence other biological traits such as mate preference, kin recognition and maternal-fetal interactions (reviewed in –). As a well-studied system, the MHC remains an important model with which to test hypotheses regarding the influence of selection on host genetic diversity.
The mechanism by which pathogen-mediated balancing selection maintains diversity in MHC genes has been extensively debated in the literature , , –. At present, there are three primary hypotheses that have been widely discussed: (i) overdominant selection , (ii) frequency-dependent selection  and (iii) fluctuating selection , . The theory of overdominant selection (or heterozygote advantage) makes the assumption that heterozygous individuals recognize a broader range of pathogens and therefore have a fitness advantage over homozygotes . Whereas, under frequency-dependent selection (or rare-allele advantage) the advantage of a particular allele is suspected to vary with its frequency, such that as pathogens evolve to evade the more common host alleles, rare alleles may become advantageous by conferring host resistance , . In reality, however, natural populations are exposed to fluctuating environmental conditions and, subsequently, host-pathogen interactions are similarly expected to vary spatiotemporally . Empirical genetic evidence for geographic heterogeneity in selective pressures has been shown in fish , birds ,  and mammals , . A selection model demonstrated that temporal variation in pathogen resistance may be sufficient to maintain polymorphism in the absence of both heterozygote and rare-allele advantage . While it is generally accepted that any or all of these proposed mechanisms can play a role in shaping the distribution of MHC variation, distinguishing between them in natural populations can be challenging due to similarities in the expected genetic outcomes .
Plains zebra (Equus quagga) experience spatiotemporal variability in pathogen pressures across localities (Table 1). In Etosha National Park, Namibia, E. quagga are the main host species of anthrax, a deadly disease caused by the gram-positive bacterium, Bacillus anthracis, which occurs in severe and consistent annual outbreaks , . In contrast, the zebra population of Kruger National Park, South Africa, suffers from sporadic and less intensive anthrax infections, with outbreaks occurring on an approximate decadal cycle . MHC genes are known to be involved in immune response to bacterial pathogens, and associations have been documented between polymorphisms in classical human leukocyte antigen haplotypes (HLA-DR-DQ) and heterogeneity in host immunoglobulin G antibody response to the B. anthracis protective antigen following administration of the anthrax vaccine .
Other than anthrax, zebra in Etosha are appreciably infected by gastrointestinal (GI) parasites, found in the population at nearly 100% prevalence . Gastrointestinal nematode prevalence is similarly high in the Kruger population, but there is a greater nematode species richness than in Etosha . Differences in parasite species richness between these populations is also apparent with regards to the ectoparasite community, with higher Ixodidae tick species diversity in Kruger  than in Etosha . Recent studies have provided evidence for balancing selection acting on two MHC (or Equine Lymphocyte Antigen; ELA) genes, DRA and DQA, over the history of the genus Equus, as well as positive selection acting at sites responsible for binding foreign peptides , . MHC variation has been explored in a captive population of E. przewalski , however no studies, to our knowledge, have examined the evolution of MHC genes in wild equid populations. The differential pathogen pressures in Etosha and Kruger provide an excellent natural system for examining the effects of selection and demography on MHC genes.
Contrasting the levels and distribution of genetic diversity at MHC genes with those expected from neutral expectations within and among populations has been frequently employed to elucidate the effects of selection on MHC genes –, , . In this study, we examined the distribution of genetic variation in the ELA- DRA and DQA PBR within zebra populations differing in parasite richness and severity (Table 1) to shed light on how selection may affect MHC gene variation. A baseline for demography, including genetic structure and inferences regarding changes in population size, was established through analyses combining the use of neutral microsatellite and nuclear intron data. We employed tests for selection and demography on both functional and neutral data to reveal differences in selective pressures occurring across loci and zebra populations. We expect to discover a signature of balancing selection in ELA genes across zebra populations, given their functional role in immune response. We also predict to find evidence of heterogeneous selection pressures acting across populations, given the differences in parasite species richness in these two localities. These results provide valuable insight for understanding the importance of immune genes in local adaptation and for identifying candidate alleles that may play an important role in pathogen resistance.
Materials and Methods
This study was approved by the Namibian Ministry of Environment and Tourism (Permit #1220/2007) and South African National Parks, South Africa (Contract, Project title: “The role of host genetics in susceptibility to anthrax among Burchell's zebra of southern Africa”). All field sampling was conducted in accordance to the requirements and guidelines outlined. The sample collection protocol was also approved by the Animal Care and Use Committee (Protocol #R217-0510B) at the University of California, Berkeley.
Study populations and sample collection
Blood, tissue and fecal samples were collected from two E. quagga populations of southern Africa: Etosha National Park, Namibia (n = 84), and Kruger National Park, South Africa (n = 89). Fecal samples were collected immediately following defecation and could be attributed to individual zebra. Sample preservation and genomic DNA extraction protocols are outlined in Kamath & Getz . DNA extracted from fecal material occasionally resulted in failed PCR-amplifications and was attributed to enzyme degradation, hydrolytic and oxidative damage. Therefore, we often used only subset of the data in downstream genetic analyses (between 41–71% of samples were successfully typed per locus; Table S1).
From 2000 through 2012, Etosha zebra population abundance was estimated from aerial survey data as being relatively stable, fluctuating between approximately 13,000 to 16,000 individuals [Namibian Ministry of Environment and Tourism, unpublished data]. In Kruger, zebra population surveys from 1980 to 1993 similarly indicate that abundance was stable, ranging between 21,454 to 33,164 individuals per year, estimates that may only represent 60–85% of the true population size , .
ELA amplification, cloning and sequencing
The exon 2 coding regions of the ELA-DRA and -DQA were amplified from genomic DNA following PCR and sequencing reaction protocols described in Kamath & Getz . Amplified fragments were 246 and 205 bp and encompassed the PBR of the DRA and DQA, respectively. We sequenced loci in forward and reverse directions to confirm heterozygous base positions. Sequence chromatograms were edited and aligned manually using Geneious 5.0 . Allelic phase for DRA heterozygotes was determined with the haplotype reconstruction program, PHASE v2.1 , which has been shown to perform well, even with MHC sequences comprised of many heterozygous sites . We conducted two independent runs with different initial random seeds and upheld a threshold posterior probability of 0.80 for haplotype confirmation, which is higher than the suggested cutoff . For all but two individuals that were included in further analyses, haplotype posterior probabilities fell between 0.95 and 1.
The DQA locus is known to be highly variable and our data may represent alleles from two loci , . Although our primers appeared to preferentially amplify only one locus, for the sake of simplicity we herein refer to all DQA sequences as ‘alleles’ despite the possibility they may come from the different loci. To identify alleles in heterozygous individuals, molecular cloning was carried out using a TOPO-TA® cloning kit with Mach 1™ T1R competent cells (Invitrogen). We minimized polymerase errors, heteroduplex and chimera formation by using a high fidelity polymerase (Platinum® Taq DNA Polymerase High Fidelity, Invitrogen) in the initial PCR. Amplicons were run on a 1% agarose gel and visualized using GelStar® Nucleic Acid Gel Stain (Cambrex Bio Science Rockland, Inc.) and the Dark Reader transilluminator system (Clare Chemical) which does not damage DNA and also increases the transformation efficiency during subsequent cloning procedures. Bands were excised and purified using QIAquick Gel Extraction Kits (Qiagen), ligated into pCR®4 TOPO vectors and transformed into E. coli competent cells. Following an overnight incubation at 37°C, sixteen to twenty-three positive clones were picked per individual and clones were directly sequenced using the same primers as in the initial PCR. Alleles were confirmed with at least two observations (i.e. amplified in at least two independent PCRs from the same individual or seen in two different individuals). Sequences not meeting these criteria were considered to be erroneous and not considered in subsequent analyses.
β-Fibrinogen intron amplification
The Fibrinogen beta chain gene, intron 7 (β-Fibr) appears to be evolving neutrally and has been demonstrated to be an informative marker for systematic and evolutionary studies in both birds and mammals –. We amplified 668 bp of β-Fibr with primers developed using the publicly-available horse genome and Primer3 : β-Fibr10 (5′-CAGTAGTATCTGCCGTTTGG-3′) and β-Fibr11 (5′-GAGGGCGACAAATACCAAC- 3′) (Protocol S1). Amplified products were cleaned using Exo-SAP-IT (USB Corporation) and sequenced in both directions on an ABI 3730 sequencer (Applied Biosystems). Cycling conditions were followed as described for ELA loci and haplotype determined using PHASE v2.1 . Due to the presence of a length polymorphism, we used the program OLFinder  to resolve heterozygous genotypes when an insertion/deletion (indel) event was evident.
We genotyped 15 microsatellite loci previously isolated from the horse (Equus callabus) (Protocol S1, Table S2). Forward primers were modified at the 5′-end by the addition of a fluorescent label: HEX, 6-FAM (Invitrogen), NED or PET (Applied Biosystems). Allele fragments were scored for size against the LIZ-500 size standard through electrophoresis using an ABI3730 DNA Analyser, followed by visualization with GeneMapper v.4.0 (Applied Biosystems) software.
A portion of the samples were derived from feces, and thus were expected to be subject to high genotyping error rates and allelic dropout. To account for these issues, we used a comparative genotyping protocol  (Figure S1), as a modification of the multi-tubes approach . This protocol has been shown to efficiently reduce error rates by minimizing the number of PCRs necessary to arrive at a consensus genotype . Furthermore, we characterized and quantified genotyping error using paired blood and fecal samples (n = 42) obtained from zebras in Etosha.
Fisher's exact tests for Hardy-Weinberg (H-W) equilibrium and genotypic linkage disequilibrium (LD) between pairs of loci were conducted in GENEPOP 4.0  to test that microsatellite loci followed the assumptions of neutrality. Significance of exact tests was determined with a Markov chain algorithm  using default parameters, and corrected for multiple comparisons through a sequential Bonferroni procedure . Null allele frequency (NAF) (frequency of non-amplified alleles resulting in an apparent homozygote) was estimated per locus (10,000 pseudoreplicates) using the expectation maximum algorithm implemented in FreeNA . Two loci, Asb23 and Lex33, had low amplification success in samples from Kruger, with 63% and 71% missing data, respectively, and were excluded from all further analyses on both populations.
Intra-population genetic diversity
General patterns of intra-population diversity were assessed by calculating the average number of alleles (A) and expected heterozygosity (HE) at all loci using GenAlEx 6.2 . Allelic richness, being sensitive to differences in sample size, was corrected for by rarefaction (ACORR) in HP-RARE 1.1 . Allelic variation in zebra populations was quantified at sequence-based loci (β-Fibr, DRA, DQA) in terms of number of segregating sites (S: ), haplotype diversity (HD: ), nucleotide diversity (π: ) and the mean number of pair-wise nucleotide differences (k: ) in DNAsp v5 . Empirical distributions were generated to determine sampling variance and standard deviations of parameter estimates.
Given that the DQA sequences may be derived from two unresolved loci, we were not able to assign specific alleles to a locus. Therefore, we estimated allele frequencies as the number of individuals carrying a particular allele out of the total number of alleles (see , ). Homozygotes were assumed to have two copies of the observed allele, but in cloned heterozygotes (with ≥2 sequences identified) each allele observed was counted only once. We recognize that this method may underestimate the frequency of common alleles and likewise overestimate rare allele frequency. Therefore, we alternatively assessed intra- and inter-population variability at the DQA locus using measures independent of allele frequency: mean number of alleles per individual, total number of alleles per population and average percent difference (APD). APD was calculated based on the average percentage of sequences that differ among all possible individual pair-wise comparisons, as outlined in Yuhki and O'Brien , and can be used as a reliable measure of within-population genetic variation from multi-locus data . To facilitate comparisons across loci, we similarly estimated APD for both the DRA and β-Fibr loci. As diversity data was not normally distributed, we used a non-parametric Kruskal-Wallis rank randomization test to test the null hypothesis of group-mean equality. Differences in variance between population samples were addressed by a Levene test based on absolute residuals of each observation to the respective population mean. A Wilcoxon sign-rank test (Z test) was performed to account for unequal variance between population samples in group-mean comparisons. In addition, we tested for differences in the distribution of DQA allele number per individual across populations through goodness-of-fit contingency analyses, with significance assessed through calculation of the Chi Square statistic (χ2). All statistical analyses were performed using JMP 4.0 (SAS Institute Inc.) software. As we found greater than two alleles in only a small number of individuals (n = 10), we also conducted all analyses excluding these individuals to confirm observed patterns in the data.
Population structure and clustering
Selection may affect the distribution of mutations and frequency of alleles within and among populations. Therefore, comparing the population differentiation observed at functional and neutral loci may be informative for understanding the nature of selection , . Under the effects of balancing selection, population divergence is expected to be lower at MHC loci relative to neutral loci, due to a more even spatial distribution of genetic variation, whereas if the mode of selection varies across populations, we may expect MHC divergence to be higher than estimates based on neutral data. We contrasted the partitioning of genetic variation at ELA loci to that of neutral data to control for the confounding effects of migration and population size, and shed light on selection heterogeneity across populations. Population differentiation was assessed in Arlequin v3.5  by conventional F-statistics (FST: ). The FST estimator, ΦST , was calculated for sequence data, using a Kimura-2 parameter distance matrix, with significance determined by 1,000 permutations. Ninety-five percent confidence intervals were generated for microsatellite FST estimates by bootstrapping (10,000 replicates) in FSTAT v2.9  and ELA values falling outside this interval were considered to be significantly different . ELA estimates were directly compared to FST values derived from the nuclear intron, β-Fibr. Population divergence was further assessed by global exact tests of differentiation (Markov chain steps = 100,000, dememorization steps = 10,000).
As the variability of a genetic marker increases, estimates of FST and its family of analogs have been shown to approach zero (i.e. equivalent to 100% genetic similarity), sometimes even among fully differentiated subpopulations . This observation suggests these estimators may not be comparable across the markers and, therefore, may not represent ‘true’ estimates of the level of similarity among populations. To account for this, we calculated Jost's D actual differentiation estimator, Dest  in SMOGD . For microsatellite-based estimates of Dest, we reported the harmonic mean across all loci.
At the DQA locus, population allele frequencies used in FST computations are potentially biased due to overestimation of rare allele frequencies (as discussed previously). Therefore, we calculated a measure of differentiation F′ (analogous to FST), to account for this bias in allele frequencies. This measure is derived from the similarity index  and because it uses percent similarity among pair-wise sequences of individuals to estimate population sub-division, it is analogous to APD. Standard errors of F′ were estimated by applying a Taylor expansion approximation . Estimates of F′ were also estimated for the DRA and β-Fibr to validate comparisons of differentiation across loci.
Zebra population structure was further assessed using the Bayesian clustering algorithm in the program STRUCTURE v2.3 , with analyses conducted on neutral data (13 microsatellite loci and the β-Fibr intron), and both ELA loci independently. The genetic clustering algorithm maximizes Hardy-Weinberg proportions, under a model allowing for population admixture and the assumption that alleles were correlated. We considered models that specified K = 1 through 5 genetic clusters and conducted 10 independent runs for each value of K, each totaling 1,000,000 MCMC steps (burnin = 100,000). We inspected both the maximum log probability for a given model, Pr(X|K), and the rate of change in the log probability of the data between successive K values, ΔK, to determine the most probable number of genetic clusters given the dataset . Results from the 10 replicate runs were summarized for the most probable K using the program CLUMPP v1.1  and results were visualized using Distruct .
We constructed mismatch distributions using the β-Fibr sequence data in Arlequin 3.5  to test the null hypothesis of recent increase in population size. Mismatch analyses compared the observed frequency distribution of pair-wise differences to the expected unimodal distribution of a population that has undergone a sudden expansion, generated through coalescent simulations , . Multimodality and deviations from the expected distribution may be indicative of a stationary population at demographic equilibrium. Goodness-of-fit tests for population expansion were conducted by calculating the sum of squared deviation (SSD) and raggedness index (RI) , and significance determined by 10,000 coalescent simulations.
Historical changes in zebra population sizes were characterized with a coalescent-based model in LAMARC v2.1.5  using microsatellite data. We conducted two independent Bayesian runs under the Brownian motion approximation mutation model , with a sampling routine of 40,000 parameter sets at intervals of 80 increments, and a burnin of 4,000. Demographic parameters were jointly estimated from the posterior sampling distributions: (1) Theta (θ), a measure of diversity proportional to the effective population size (Ne) and mutation rate (μ) such that (θ = 4Neμ) and (2) the exponential growth rate parameter (g, with units of 1/generations) relative to the neutral mutation rate representing the direction and magnitude of change in population size. A Metropolis-coupled MCMC approach was employed for each run (using one cold chain and four heated chains). Acceptance rates fell between 5 and 40%, probability density functions were inspected for unimodality, and effective sample size (ESS) were confirmed to be >200 using Tracer v1.5 .
Past population dynamics were inferred using intron data with the Bayesian Skyline Plot (BSP) model in BEAST v1.6 . A posterior distribution of effective population size through time was generated using a MCMC sampling scheme. Two independent analyses were run for 100,000,000 generations (sampling every 10,000 and 10% burnin) under a HKY substitution model, assuming a strict molecular clock with an approximated evolutionary rate of 0.002 substitutions/site/my for β-Fibr , . We applied the piece-wise linear change model with 10 internodes. Skyline reconstruction was performed in Tracer v1.5 , and the median and 95% credibility interval were plotted as a function of time.
Departures from neutrality and selection analyses
Hypothesis testing was conducted to check for significant departures from neutrality at ELA loci using approaches that reflect both recent and historical processes. All tests were conducted in Arlequin 3.5  and significance determined with 10,000 simulations. First, we used Slatkin's Markov-Chain Monte Carlo (MCMC) implementation  of the Ewens-Watterson (E-W) Test ,  to test for recent selection or demographic events (i.e. rapid population expansion or bottlenecks) affecting allele frequency patterns. The E-W test compared the observed homozygosity (Fobs) with the expected homozygosity (Fexp), based on a random sample of the same size consisting of the same number of alleles, simulated under the assumption of neutrality. Tajima's D  and Fu's FS  were calculated for each population to test for departures from the null hypothesis of neutral evolution and population equilibrium. D and FS are analogous to the E-W test, but because they are based on sequence data can also reflect historical selective pressures or demographic changes. FS measures the probability of observing a certain number of alleles given the average pair-wise sequence divergence (θ), whereas Tajima's statistic contrasts the observed θ to that which would be expected under neutrality given the number of segregating sites (S). Negative values for both statistics imply population expansion or purifying selection, and positive values suggest either a population bottleneck or balancing selection. To tease apart signatures of demography and selection, we contrasted neutrality test results to those performed on β-Fibr. In addition, allele frequency distributions at neutral and functional loci were statistically compared across populations with contingency analyses in JMP 4.0 (SAS Institute Inc.).
To characterize molecular-level evidence of selection, we estimated rates of non-synonymous to synonymous mutations (ω = dN/dS) at coding genes (DRA, DQA) using maximum likelihood models of codon-substitution in the CODEML subroutine of PAML . Input starting trees were generated using PhyML 3.0  assuming the K80 +Γ6 (Kimura 2 parameter with 6 gamma-distributed rate categories) nucleotide substitution model as determined by jModelTest  and applying the subtree pruning and re-grafting (SPR) tree searching algorithm. We performed maximum likelihood ratio tests (LRTs) to test for significant positive selection by comparing the likelihoods of models of neutral evolution (M1a, M7) to those incorporating positive selection (M2a, M8). We also compared models assuming one evolutionary rate across codon sites (M0) to those allowing for heterogeneous rates (M3).
Phylogenetic allele networks
We inferred phylogenetic relationships among sequence haplotypes by constructing a median-joining haplotype network  using maximum parsimony in Network 4.6.1 . Allele frequency information and population proportion were incorporated into the visualization of the network. Unequal sample sizes were adjusted through rarefaction . Sequences from the horse (E. callabus) were incorporated [GenBank: AY726647, L47174, M60100, L47172, AJ575295, FJ716134, L33909, U92505, U92506, U92507, U92508] to evaluate the ancestral state of E. quagga haplotypes.
Microsatellite loci indices
Microsatellite loci evaluated in E. quagga were highly polymorphic, with a mean of 9.0 alleles per locus (range: 3–15) in the Etosha zebra population and 7.1 alleles per locus (range: 3–12) in Kruger (Table S3). Mean expected heterozygosity was 0.76 and 0.73 over all samples genotyped in Etosha and Kruger, respectively.
We found evidence for significant LD (p<0.001) between Hmb1 and Htg14 in Etosha, but not in Kruger. In addition, results from exact tests by locus and population indicated significant (p<0.001) heterozygote deficiency at Hms7 and Htg15, in Etosha only. We note that these observations may be due to the presence of null alleles. However, mean null allele frequency (NAF) by locus was estimated to be low (2.3%, range: 0–6.8%; Table S4). Total genotyping error varied by locus (range: 0–7%) and mean error rate was estimated to be 0.3% and 2.5% in blood and fecal samples genotyped, respectively. Breakdown of error contributions suggested allelic dropout only accounted for ∼0.5% of the error in fecal samples genotyped. Additional PCRs, following the comparative approach of Frantz et al.  lowered the error rate to 0%. Therefore, we are confident our genotyping approach yielded accurate results and that the assumptions made in subsequent analyses were upheld.
Population genetic variation
Ten novel alleles were discovered at the β-Fibr locus in E. quagga [GenBank: KC109106–KC109115], with 8 polymorphic sites and two indel mutations totaling 14 base pairs. All β-Fibr alleles differed from a sequence previously reported in E. callabus [GenBank: AY726647]; However, β-Fibr*01 only differed by one nucleotide from the E. callabus sequence. Nine DRA alleles (Eqbu-DRA*01, *03-*05, *07-*11) [GenBank: AJ575299, EU930126, EU930121, EU930118, HQ637392–HQ637396] were recovered in Etosha and Kruger zebra populations. Twenty-one DQA alleles (Eqbu-DQA*01-02, *04-*22) [GenBank: EU935837, EU935829, EU935834–EU935836, EU935832, EU930130, HQ637397–HQ637409, KC109105] were also found. Population patterns of allelic richness were analogous with and without the correction applied for unequal sample sizes.
At the DQA locus, there was on average 1.56 (range: 1.31–1.80) and 1.77 (range: 1.46–2.07) alleles per individual in the Etosha and Kruger zebra populations, respectively. Less that 15% of samples cloned were found to have more than two alleles, and therefore the majority of our DQA data is likely derived from one locus. There was no evidence for significant differences in copy number frequency distributions across populations (χ2 = 5.578, p = 0.1341; Figure S2).
At ELA loci, indices of sequence diversity (HD, π and k) were high when contrasted to β-Fibr intron data (Figure 1, Table 2). Comparisons of indices across populations revealed significantly depressed sequence diversity at the DRA locus in Etosha relative to Kruger (Figure 1). This pattern was consistent across sequence-based (HD, π and k) and allele frequency-based (HE) indices (Etosha: HD = 0.748, π = 0.0088, k = 2.158, HE = 0.743; Kruger: HD = 0.874, π = 0.0097, k = 2.393, HE = 0.860) (Table 2). This is also contradictory to what we would expect given the diversity at intron and microsatellite data— neutral diversity estimates were significantly higher in Etosha than Kruger. Diversity estimates at the DQA locus were similar across populations. Furthermore, results from non-parametric statistical analyses of APD, an unbiased diversity measure, corresponded with this inter-population pattern of variability (Figure 1, Table 3); APD in Etosha was significantly higher at the β-Fibr and lower at the DRA locus, whereas we could not reject the null hypothesis of APD mean equality across populations at the DQA locus. Thus, despite the potential bias in standard DQA diversity measures, the correspondence of APD patterns across loci and populations with other diversity estimates indicates our results are robust.
Genetic diversity indices plotted by population (Etosha: black, Kruger: gray) and locus. Diversity indices shown are: (A) average percent difference (APD), and (B) haplotype diversity (HD). Significant difference (p<0.05) between population means are indicated by an asterisk (*) and non-significance by NS.
Estimates of population pair-wise FST revealed evidence for significant, albeit low, levels of genetic differentiation at ELA loci across populations (DRA: FST = 0.045, p<0.001; DQA: FST = 0.02, p = 0.027; Table 4). DQA population structure was significantly lower than that observed at neutral loci, and the estimated FST fell outside of the microsatellite 95% CI (0.026–0.053). In contrast, DRA differentiation was not significantly different from FST estimates observed at microsatellite loci (FST = 0.038, p<0.001). Analyses of β-Fibr genetic structuring indicated substantially higher differentiation across populations (FST = 0.140, p<0.001) and supported the observation that low differentiation at ELA loci may be indicative of balancing selection acting on these genes. These results were corroborated by estimates of the FST analog, F′ (Table 4). The unbiased estimator of differentiation (Dest) followed a similar pattern to that indicated by the patterns of FST across loci (Table 4) with of β-Fibr displaying higher levels of differentiation than both estimates based on microsatellites and ELA loci.
Clustering analyses in STRUCTURE based on neutral data revealed that a model of K = 2 populations had the highest likelihood (Ln = −5518.11) and highest ΔK, with individuals from Etosha clustering separately from those in Kruger (Figure 2, Table S5). At the DQA locus, the best-fit model was similarly K = 2 genetic clusters but, in contrast, clustering did not correspond to population. At the DRA locus, K = 1 had the highest log likelihood, but the highest ΔK suggested K = 4 populations. However, ΔK is not able to find the best K when the true K = 1  and inspection of STRUCTURE plots under models of K≥2 showed that clustering assignments broke down within individuals.
Individual genetic clustering assignments under the model of K = 2 based on (A) neutral loci (13 microsatellites and the β-Fibr intron), (B) ELA-DRA, and (C) ELA-DQA. Percent assignment to each of two genetic clusters are shown for each individual genotype in plots. Population where an individual was sampled is indicated by the bar to the left of each plot (Etosha: black, Kruger: gray). The mean posterior probability (L(K)) ± 95% confidence interval and the rate of change in the log probability of the data between successive K values (ΔK) from K = 1 to 5, for each marker type are shown below each corresponding STRUCTURE plot.
Demographic inference from neutral data
Mismatch distribution analyses of the neutral intron indicated a weak signature of population expansion in the Etosha zebra population, as the null model could not be rejected (SSD = 0.014, p = 0.058). The model of sudden population growth was rejected in Kruger, but with marginal significance (SSD = 0.029, p = 0.043). As significance is borderline for both populations, mismatch results are inconclusive. However, in both Etosha and Kruger the population expansion model could not be rejected as indicated by relatively low, non-significant estimates of the Harpending's raggedness index (Etosha: RI = 0.085, p = 0.11; Kruger: RI = 0.128 p = 0.39).
Outcomes from demographic analyses in LAMARC similarly indicated low positive growth, of similar magnitude in both populations. The most probable estimate for the exponential growth rate parameter (95% confidence interval) in Etosha was 0.302 (0.253–0.549) generations−1, whereas, in Kruger, g was estimated to be 0.388 (0.317–0.667) generations−1. Although positive, the magnitude of g is extremely small and may indicate populations that are in the process of stabilization. The most probable estimate of the diversity measure, θ, was identical across populations, estimated to be 99.3 (84.8–100.2) and 99.3 (83.0–100.1) in Etosha and Kruger, respectively. Bayesian skyline plots, based on an independent dataset, confirmed that these zebra populations are both large, and have similarly experienced stability in the recent past (Figure 3).
Tajima's D and Fu's FS tests for departures from neutrality were not significant (p>0.05) at all loci and populations (Table 5). Although non-significant, test statistics were negative at the intron, indicating a potential weak effect of population expansion. Positive values observed at ELA loci contrasted neutral data and may be representative of positive selection acting on the site frequency spectrum. Results from Slatkin's E-W test revealed significantly lower homozygosity (Fobs) than would be expected under a neutral model, supporting the hypothesis that balancing selection is occurring at ELA loci. However, when multi-locus genotypes are excluded at the DQA, these results become non-significant. Significance of E-W tests, but not D or FS, implies that selection at these loci is possibly a relatively recent phenomenon.
Skewness in population allele frequency distributions may be interpreted as a signature of demographic change and/or selection, and was used to tease apart findings from neutrality tests. ELA allele frequency distributions were inspected relative to that at the β-Fibr by population. The most outstanding observation was that DRA allele frequencies were evenly distributed in Kruger, but skewed in Etosha (Figure 4), with contingency analyses indicating significant differences in frequency distributions across populations (χ2 = 15.4, df = 7, p = 0.03). In particular, the DRA*03 allele was predominant in Etosha, representing approximately 44% of all alleles observed which could potentially be driving the decreased diversity found at this locus. Contingency analyses also suggested significantly different allele frequency distributions at the DQA locus (χ2 = 62.1, df = 20, p<0.001). DQA distributions differed in the large number of rare and private alleles found in each population (Figure 4, Figure S3). In both populations, the DQA*01 allele was present at a significantly greater frequency than any other allele (25–29%). β-Fibr allele frequency distributions were also significantly different (χ2 = 51.9, df = 9, p<0.001) between populations, with Kruger exhibiting a more skewed distribution (Figure 4).
Alleles presented in order of descending frequency at the (A) ELA-DRA, (B) ELA-DQA, and (C) β-Fibr intron by population, with Etosha shown in black and Kruger in gray. Allele sample sizes are as follows: Etosha (β-Fibr: 72, DRA: 144, DQA: 66) and Kruger (β-Fibr: 47, DRA: 62, DQA: 46). DQA allele frequencies were calculated excluding individuals with multi-locus genotypes.
Molecular-based analyses revealed no evidence for positive selection or variable rates of selection at the molecular level of the DRA locus in both zebra populations (Table 6). In contrast, at the DQA locus, likelihood ratio tests (LRTs) comparing models of invariable versus variable evolutionary rates across codon sites (M3 vs. M0) and models of positive selection to neutral evolution (M7 vs. M8) were significant (p<0.05) in both populations. While estimates of ω across the DQA gene region were close to 1 in both populations (Etosha: ω = 1.11, Kruger: ω = 0.71), approximately 40% of codon sites were determined to be under significant positive selection (ω = 2.52).
Phylogenetic relationships among alleles
Phylogenetic median-joining haplotype networks of ELA loci showed a lack of geographical allele structuring by population (Figure 5). The β-Fibr allele from E. callabus differed by one mutational step from the most frequent allele observed in E. quagga (β-Fibr*01). At the DRA, four E. callabus alleles were ancestral to the Eqbu-DRA*10 and Eqbu-DRA*11 alleles in E. quagga. However, one allele was shared between the two species (Eqbu-DRA*08 = Eqca-DRA*04). At the DQA, there were large mutational differences observed between alleles, and E. callabus alleles were found distributed throughout the network.
Circle size is proportional to haplotype frequency. Proportion of total alleles found in Etosha (black) and Kruger (gray) are shown, with sample sizes corrected by rarefaction. Haplotypes from Equus callabus are shown in white (GenBank accession numbers– β-Fibr: AY726647, DRA: L47174, M60100, L47172, AJ575295, FJ716134, DQA: L33909, U92505, U92506, U92507, U92508). Differences between haplotypes that were greater than one mutational step are notated in red italics. Insertion-deletion mutational events are represented by a red perpendicular line.
This study juxtaposed MHC with neutral data to disentangle the co-occurring effects of selection and neutral evolutionary processes (e.g. mutation, gene flow, changes in population size, and drift) on the distribution of immunogenetic variation across E. quagga populations. Our results suggested that selection is acting variably across MHC genes and populations in zebra.
Demography and migration dynamics may confound our ability to detect selection. We used neutral data to elucidate population history and extricate the role of these processes in shaping immune gene variation. Population size estimates from plains zebra in Kruger (1980–1993) ,  and Etosha (2000–2012) [Namibian Ministry of Environment and Tourism, unpublished data] suggest that zebra populations in both regions have remained relatively stable. This is in agreement with Bayesian skyline plots and demographic estimates (g) from this study that indicated effective population trajectories are similar in Etosha and Kruger zebra and suggested historically stabilized populations with the possibility of low, positive growth. Given census estimates, we expected neutral genetic diversity to be higher in Kruger due to a population size that is approximately twice as large as Etosha. In contrast, we found the opposing pattern, with neutral diversity being higher in Etosha. This discrepancy could be due to sampling artifacts from differences in survey approaches and observer bias across parks, or it is possible there is subdivision within the Kruger zebra population and our genetic data represent only a portion of this population. To date, however, there has been no genetic study in zebra documenting the degree of population mixing throughout Kruger.
At microsatellite loci, genetic differentiation across zebra populations was relatively low (FST = 0.04). This finding is in agreement with a recent phylogeographic study on five subspecies of E. quagga that showed zebra population structuring is among the lowest of 17 different savannah-adapted ungulate species in Africa (mtDNA: ΦST = 0. 17, microsatellites: θST = 0.05) . Population structure at the β-Fibr intron (FST = 0. 14, ΦST = 0. 13), however, was higher than that based on microsatellites (FST = 0.04). This discrepancy between nuclear markers suggests further inquiry into the possibility that our assumption of the neutrality at the β-Fibr intron has been violated. Alternatively, the differences observed between structure as inferred from microsatellite and intron loci may be due to differences in the mutational modes and rates of these markers .
Under balancing selection, due to similar pathogen regimes across populations, we would expect low population differentiation at MHC loci relative to neutral loci due to theoretically greater effective migration rates . This has been shown to occur even among small populations with little gene flow that are highly susceptible to drift effects . In contrast to this, if the mode of selection varies among populations, we may expect MHC divergence to be higher than that estimated from neutral markers. Consistent with the prediction of a gene under balancing selection, we found that FST measures at the DQA locus were seven to eight times lower than at the β-Fibr intron, and two to three times lower than microsatellite loci. At the DRA locus, FST was two times lower than that estimated at the intron. It is possible these results are driven by high heterozygosity at this locus, given that FST estimates may decrease as locus variability increases , . Unbiased estimates of differentiation (Jost's D) mirrored FST results in comparisons of ELA loci with the β-Fibr intron, suggesting that the pattern may also be upheld. Furthermore, STRUCTURE results using neutral data revealed clustering of individuals by population, in contrast to the lack of geographic clustering observed at both ELA loci.
Differentiation at the DRA locus was comparable to estimates at microsatellite loci. A similar observation was made at the DRA locus among domestic donkey (E. asinus) populations . Also, previous studies have contradicted expectations with discoveries of comparable or elevated population structure at MHC versus microsatellite loci , , , . This observation may be attributed to either (i) local adaptation, (ii) weak selection acting on the MHC locus, or (iii) an artifact of using genetic markers evolving under different mutational modes and rates . Our finding could be due to one or a combination of the aforementioned causes. Reduced structure at the DRA relative to the β-Fibr locus, akin to our observations at the DQA locus, indicates that a difference in the mutation model may be responsible for the discrepancy. But, the lack of a similar observation at the DQA may alternatively suggest that local adaptation is occurring at this locus. We note, however, that our inferences regarding demography based on a single locus must also be taken with caution.
Despite being low, ELA population differentiation remained significant, as indicated by pair-wise FST and global exact tests. The evolutionary process of genetic drift may be affecting ELA differentiation, albeit weakly, by stochastically shifting allele frequency distributions within populations. This is parsimonious with an expectation of little to no recent gene flow between these two populations, as they are contained in parks and separated by more than 2,000 km of a human-dominated matrix. An alternative explanation (to that of genetic drift) is that variable pathogen pressures across the localities and over time may be driving distributional shifts in ELA allele frequencies across populations.
The E. quagga population in Etosha exhibited higher genetic diversity than Kruger at all neutral loci. This pattern was fairly consistent, regardless of the diversity index evaluated. In contrast, DQA population diversity patterns are generally similar across populations. It is possible that balancing selection may be preserving similar levels of DQA diversity due to equivalently high pathogen diversity across populations. Interestingly, DRA genetic diversity across populations is strikingly incongruent with that observed at neutral loci, revealing depressed variability in Etosha. Allele frequency distributions showed that the paucity of diversity at this locus in Etosha is driven by the presence of one predominant allele. In contrast, Kruger zebra possessed many of the same alleles, but frequencies were relatively evenly distributed. Opposing DRA and neutral diversity patterns across populations provides evidence for geographically heterogeneous selection at the DRA. Pathogen-mediated selective pressure may differentially affect these two loci because of differences in their functional role (i.e. differences in the specific pathogens that these antigens are capable of recognizing).
Specifically, the marked difference in DRA patterns across populations, with respect to neutral data, highlights the possibility that zebra in Etosha may be subject to local selective pressure by pathogens at this locus. Matthee et al.  found that gastrointestinal helminth communities differed significantly among African equids and even between Etosha and Kruger zebra populations, signifying that the specialization of parasite communities has followed host population divergence. Notably, Etosha zebra had a considerably lower Stronglyinae nematode species richness— with only one species as opposed to six identified in Kruger . Despite low species diversity in Etosha, nematode prevalence has been found to be extremely high (>98%: ). Etosha's particularly arid climate (rainfall<500 mm/year) relative to Kruger's (rainfall 550–650 mm/year) may play an important role in limiting macroparasite diversity .
Besides intestinal parasites, zebra in Etosha are also known to be severely affected by anthrax, a lethal infectious bacterial disease , . Zebra in the park exhibit the highest recorded incidence of anthrax in southern Africa and the disease has been implicated as one of the primary causes of adult mortality . In contrast, Kruger zebra experience infrequent and less severe anthrax outbreaks . Given the role of MHC class II genes in recognizing extracellular pathogens, we speculate that the combined low, but prevalent parasite community and repeated severe anthrax outbreaks may present significant selective pressures limiting MHC diversity in Etosha.
Tajima's D and Fu's FS tests were unable to reject the hypothesis of neutral evolution; However, we recognize that the ability of these tests to detect selection depends on the nature of the mutational process and the duration, strength and timing of selection . Demographic processes may obscure signatures of selection and it is unclear whether this signature has been preserved in the current variation in these genes. Despite lack of significance, the sign of D and FS test statistics were generally consistent with positive selection at ELA genes and opposed values observed at the intron, indicating a putative weak selection signal. As these statistics reflect selection operating to change the site frequency spectrum, accumulation of such mutations may require long time periods extending beyond the history of a population. Whereas, the significant E-W tests recovered here support the conclusion that relatively recent balancing selection plays an important role in shaping allele frequency distributions within zebra populations. At the DQA locus, molecular selection analyses also suggested significant heterogeneity across the gene region and positive selection at approximately 40% of codon sites. Thus, while balancing selection is likely maintaining diversity at these genes, it is also possible that selection signatures have been obscured over time by fluctuating selective pressures due to changing pathogen communities or demographic events.
Simulations by Ejsmond et al.  demonstrated that allele frequency distributions are not always predictable under balancing selection, depending on the specific mode in effect (e.g. overdominance, negative frequency-dependence). Under overdominant selection, allele frequency distributions are consistently more even than those observed under neutrality. Whereas, with negative frequency-dependence, distributions are unpredictable and can be skewed, even, or indistinguishable from that observed under neutrality, depending upon at what point during the host-parasite cycle sample collection occurred. The conclusion is that it is impossible to indisputably infer mode of selection from neutrality tests based on departures of allele frequency distributions from neutral expectations (e.g. the E-W test). Therefore, we cautiously conclude balancing selection is occurring at these loci based on results from E-W tests, and do not make any further interpretation regarding the strength or nature of selection.
The hypothesis that balancing selection preserves MHC diversity is well accepted and supported in the literature , . Furthermore, recent studies focused on long-term evolutionary patterns (through evaluation of dN/dS ratios and MHC gene phylogenies) have provided genus-level evidence for balancing selection acting on both the DRA and DQA loci in equids , . Our results provided several lines of evidence to further support its occurrence over shorter evolutionary time scales in zebra (i.e. over the history of the species or population), including (i) elevated ELA diversity over neutral diversity within populations, (ii) low genetic structure across populations relative to that observed at neutral loci, (iii) significant rejection of the null hypothesis of neutrality based on the assumption of mutation-drift equilibrium, and (iv) evidence for heterogeneous rates of evolution across codon sites, with significant positive selection occurring at specific sites (only at the DQA). These combined findings provide convincing evidence for balancing selection operating on ELA genes in zebra. However, the somewhat differing patterns observed at the DRA across populations also suggest that there may be heterogeneous selection pressures and local adaptation operating at this locus.
MHC studies in natural populations are critical for understanding adaptation and evolutionary potential, as variation in these genes reflect biologically relevant processes significant to fitness. Patterns of MHC variation are shaped by a complex interplay of selective and demographic factors, which may be challenging to disentangle, but possible to achieve through the amalgamation of multiple lines of evidence. Our data suggest that selection on MHC genes may vary spatially, and also differ by locus. Balancing selection over evolutionary time scales may act cumulatively to retain MHC diversity, but this selection signature may be obscured due to fluctuating and diverse pathogen communities. We found evidence for balancing selection at MHC genes in zebra populations, but we also conclude heterogeneous selection may be acting across populations at the DRA locus— two findings which are compatible when considering different time and spatial scales. These results highlight the importance of integrating neutral and adaptive data over different scales to uncover the relative effects of demography and selection in shaping functional diversity. Future ecological studies are warranted that investigate the link between host immunogenetic diversity and pathogen community structure to better understand the mechanisms underlying adaptation.
Polymerase Chain Reaction protocols for neutral loci. PCR reagents and cycling conditions used to amplify the β-Fibrinogen, intron 7 and microsatellite loci: Aht21, Asb23, Cor014, Hmb1, Hms7, Htg7, Htg9, Htg14, Htg15, Lex20, Lex33, Lex52, Ucdeq505, Um011, and Vhl47.
Flowchart of comparative microsatellite genotyping approach. This approach involves comparing two to three initial replicate PCRs, with heterozygotes confirmed in two PCRs and homozygotes in three PCRs. If a disagreement is found (e.g. first PCR results in heterozygote, and second in homozygote for one allele) additional PCRs were performed until each allele is observed a minimum of two times. In the event that no consensus was found, an individual was either scored as having a missing genotype or given a half-locus genotype, by assigning one allele as missing data. In summary, a minimum of 2 PCRs is required to confirm a heterozygote genotype and 3 PCRs for a homozygote, with a maximum of up to 7 PCRs conducted. Adapted from Hansen et al. .
Frequency distribution of ELA-DQA copy number. DQA copy number frequency distribution in individuals from Etosha (black) versus Kruger (gray). One to four alleles were observed in each individual. Contingency goodness-of-fit analyses revealed no significant difference between population frequency distributions (χ2 = 5.578, p = 0.1341).
ELA-DQA allele frequency distributions, including alleles from multi-locus genotypes. Alleles are presented in descending order and by population, with Etosha in black and Kruger in gray.
Populations sampled and individuals successfully genotyped at each locus. Loci investigated include microsatellites (μsats), β-Fibrinogen intron 7 (β-Fibr), ELA-DRA exon 2 (DRA), and ELA-DQA exon 2 (DQA). The number of confirmed alleles found in each population is reported, with that for μsats reported as the mean (standard error) of 15 loci. Over both populations, typing success ranged from 41–71% per locus.
Description of microsatellite loci. Includes total number of alleles, observed size range in base pairs (bp), fluorescent label used for genotyping analyses, annealing temperature used in polymerase chain reaction (PCR), and the reference study in which it was initially discovered in E. callabus.
Microsatellite diversity by locus and population. For each population, diversity by locus and total mean diversity are reported. Diversity is described in terms of number of alleles (A), observed heterozygosity (HO), and expected heterozygosity (HE). Sample sizes (N) are reported for each locus and population. Significant departures from Hardy-Weinberg equilibrium are indicated in boldface.
Microsatellite null allele frequency and genotyping error rates. Null allele frequency (NAF) was estimated by population and locus. Total genotyping error rates, broken down by allelic dropout and false allele generation, were determined by paired genotyping of blood and fecal samples from individual zebra captures. All rates marked as “0” represent values of 0.000.
Results from clustering analyses in STRUCTURE. Analyses were conducted over 10 runs of K = 1 to 5, combining all neutral data (13 microsatellites and the β-Fibr intron) and, separately, at each ELA locus. The mean posterior log probability (Mean L(K)), standard deviation (SD L(K)), mean difference between successive likelihood values of K (L′(K)), absolute value of the difference between successive values of L′(K) (|L″(K)|), and second order rate of change of the likelihood function with respect to K (ΔK) are reported. The best-fit models are highlighted in bold, and were selected based on mean posterior probabilities and following the approach described by Evanno et al. .
We would like to thank three anonymous reviewers and members of the Getz and Bowie Labs at UC Berkeley for valuable comments on this manuscript. We are also grateful to Olivier Putzeys, Holly Ganz, Wendy Turner, Carrie Cizauskas for help collecting samples; Miriam Tsyaluk, Kaitlin Hall and Elyse DeFranco for assistance in the lab; Kari R. Goodman, Margarita Hadjistylli and Emily Rubidge for valuable discussions on methods and analyses. Field work and sampling was made possible by permission and facility resources provided by the Namibian Ministry and Environment and Tourism (MET) and South African National Parks (SANParks). We further thank the scientific staff at the Etosha Ecological Institute (EEI), particularly Werner Kilian, who provided field support for this study.
Conceived and designed the experiments: PLK WMG. Performed the experiments: PLK. Analyzed the data: PLK. Contributed reagents/materials/analysis tools: PLK WMG. Wrote the paper: PLK WMG.
- 1. Lewontin RC, Krakauer J (1973) Distribution of gene frequency as a test of theory of selective neutrality of polymorphisms. Genetics 74: 175–195.
- 2. Nielsen R (2005) Molecular signatures of natural selection. Annual Review of Genetics 39: 197–218.
- 3. Avise JC (2004) Molecular Markers, Natural History, and Evolution. Sunderland, MA: Sinauer Associates, Inc.
- 4. Blouin MS, Parsons M, Lacaille V, Lotz S (1996) Use of microsatellite loci to classify individuals by relatedness. Molecular Ecology 5: 393–401.
- 5. Paetkau D, Waits LP, Clarkson PL, Craighead L, Strobeck C (1997) An empirical evaluation of genetic distance statistics using microsatellite data from bear (Ursidae) populations. Genetics 147: 1943–1957.
- 6. Balloux F, Lugon-Moulin N (2002) The estimation of population differentiation with microsatellite markers. Molecular Ecology 11: 155–165.
- 7. van Tienderen PH, de Haan AA, van der Linden CG, Vosman B (2002) Biodiversity assessment using markers for ecologically important traits. Trends in Ecology & Evolution 17: 577–582.
- 8. Sommer S (2005) The importance of immune gene variability (MHC) in evolutionary ecology and conservation. Frontiers in Zoology 2: 1–18.
- 9. Landry C, Bernatchez L (2001) Comparative analysis of population structure across environments and geographical scales at major histocompatibility complex and microsatellite loci in Atlantic salmon (Salmo salar). Molecular Ecology 10: 2525–2539.
- 10. Miller KM, Kaukinen KH, Beacham TD, Withler RE (2001) Geographic heterogeneity in natural selection on an MHC locus in sockeye salmon. Genetica 111: 237–257.
- 11. Cutrera AP, Lacey EA (2006) Major histocompatibility complex variation in Talas tuco-tucos: The influence of demography on selection. Journal of Mammalogy 87: 706–716.
- 12. Mona S, Crestanello B, Bankhead-Dronnet S, Pecchioli E, Ingrosso S, et al. (2008) Disentangling the effects of recombination, selection, and demography on the genetic variation at a major histocompatibility complex class II gene in the alpine chamois. Molecular Ecology 17: 4053–4067.
- 13. Loiseau C, Richard M, Garnier S, Chastel O, Julliard R, et al. (2009) Diversifying selection on MHC class I in the house sparrow (Passer domesticus). Molecular Ecology 18: 1331–1340.
- 14. Miller HC, Allendorf F, Daugherty CH (2010) Genetic diversity and differentiation at MHC genes in island populations of tuatara (Sphenodon spp.). Molecular Ecology 19: 3894–3908.
- 15. Klein J (1986) Natural History of the Major Histocompatibility Complex. New York: Wiley & Sons.
- 16. Takahata N (1990) A simple genealogical structure of strongly balanced allelic lines and transspecies evolution of polymorphism. Proceedings of the National Academy of Sciences of the United States of America 87: 2419–2423.
- 17. Hughes AL, Nei M (1989) Nucleotide substitution at Major Histocompatibility Complex class-II loci - Evidence for overdominant selection. Proceedings of the National Academy of Sciences of the United States of America 86: 958–962.
- 18. Hedrick P, Kim T (1998) Genetics of complex polymorphisms: parasites and maintenance of MHC variation. In: Singh RS, Krimbas CB, editors. Genetics, Evolution, and Society. Cambridge: Harvard University Press. pp. 205–233.
- 19. Edwards SV, Hedrick PW (1998) Evolution and ecology of MHC molecules: from genomics to sexual selection. Trends in Ecology & Evolution 13: 305–311.
- 20. Bernatchez L, Landry C (2003) MHC studies in nonmodel vertebrates: what have we learned about natural selection in 15 years? Journal of Evolutionary Biology 16: 363–377.
- 21. Piertney SB, Oliver MK (2006) The evolutionary ecology of the major histocompatibility complex. Heredity 96: 7–21.
- 22. Apanius V, Penn D, Slev PR, Ruff LR, Potts WK (1997) The nature of selection on the major histocompatibility complex. Critical Reviews in Immunology 17: 179–224.
- 23. Spurgin LG, Richardson DS (2010) How pathogens drive genetic diversity: MHC, mechanisms and misunderstandings. Proceedings of the Royal Society B-Biological Sciences 277: 979–988.
- 24. Meyer D, Thomson G (2001) How selection shapes variation of the human major histocompatibility complex: a review. Annals of Human Genetics 65: 1–26.
- 25. Doherty PC, Zinkernagel RM (1975) Enhanced immunological surveillance in mice heterozygous at H-2 gene complex. Nature 256: 50–52.
- 26. Bodmer WF (1972) Evolutionary significance of Hl-a system. Nature 237: 139–&.
- 27. Hill AVS, Allsopp CEM, Kwiatkowski D, Anstey NM, Twumasi P, et al. (1991) Common West African HLA antigens are associated with protection from severe malaria. Nature 352: 595–600.
- 28. Hedrick PW, Thomson G, Klitz W (1987) Evolutionary genetics and HLA - Another classic example. Biological Journal of the Linnean Society 31: 311–331.
- 29. Slade RW, McCallum HI (1992) Overdominant vs frequency-dependent selection at MHC loci. Genetics 132: 861–862.
- 30. Takahata N, Nei M (1990) Allelic genealogy under overdominant and frequency-dependent selection and polymorphism of Major Histocompatibility Complex loci. Genetics 124: 967–978.
- 31. Alcaide M, Edwards SV, Negro JJ, Serrano D, Tella JL (2008) Extensive polymorphism and geographical variation at a positively selected MHC class IIB gene of the lesser kestrel (Falco naumanni). Molecular Ecology 17: 2652–2665.
- 32. Ekblom R, Saether SA, Jacobsson P, Fiske P, Sahlman T, et al. (2007) Spatial pattern of MHC class II variation in the great snipe (Gallinago media). Molecular Ecology 16: 1439–1451.
- 33. Oliver MK, Lambin X, Cornulier T, Piertney SB (2009) Spatio-temporal variation in the strength and mode of selection acting on major histocompatibility complex diversity in water vole (Arvicola terrestris) metapopulations. Molecular Ecology 18: 80–92.
- 34. Kloch A, Babik W, Bajer A, Sinski E, Radwan J (2010) Effects of an MHC-DRB genotype and allele number on the load of gut parasites in the bank vole Myodes glareolus. Molecular Ecology 19: 255–265.
- 35. Hedrick PW (2002) Pathogen resistance and genetic variation at MHC loci. Evolution 56: 1902–1908.
- 36. Ebedes H (1976) Anthrax epizootics in Etosha National Park. Madoqua 10: 99–118.
- 37. Lindeque PM, Turnbull PCB (1994) Ecology and epidemiology of anthrax in the Etosha National Park, Namibia. Onderstepoort Journal of Veterinary Research 61: 71–83.
- 38. de Vos V (1990) The ecology of anthrax in Kruger National Park, South Africa. Salisbury Medical Bulletin (Supplement) 68: 19–23.
- 39. Pajewski NM, Parker SD, Poland GA, Ovsyannikova IG, Song W, et al. (2011) The role of HLA-DR-DQ haplotypes in variable antibody reponses to Anthrax Vaccine Adsorbed. Genes and Immunity 12: 457–465.
- 40. Turner WC, Getz WM (2010) Seasonal and demographic factors influencing gastrointestinal parasitism in ungulates of Etosha National Park. Journal of Wildlife Diseases 46: 1108–1119.
- 41. Matthee S, Krecek RC, McGeoch MA (2004) A comparison of the intestinal helminth communities of equidae in Southern Africa. Journal of Parasitology 90: 1263–1273.
- 42. Horak IG, Devos V, Deklerk BD (1984) Parasites of domestic and wild animals in South-Africa .17. Arthropod parasites of Burchell zebra, Equus burchelli, in the Eastern Transvaal Lowveld. Onderstepoort Journal of Veterinary Research 51: 145–154.
- 43. Horak IG, Anthonissen M, Krecek RC, Boomker J (1992) Arthropod parasites of springbok, gemsbok, kudus, giraffes and Burchell and Hartmann zebras in the Etosha and Hardap Nature-Reserves, Namibia. Onderstepoort Journal of Veterinary Research 59: 253–257.
- 44. Janova E, Matiasovic J, Vahala J, Vodicka R, Van Dyk E, et al. (2009) Polymorphism and selection in the major histocompatibility complex DRA and DQA genes in the family Equidae. Immunogenetics 61: 513–527.
- 45. Kamath PL, Getz WM (2011) Adaptive molecular evolution of the Major Histocompatibility Complex genes, DRA and DQA, in the genus Equus. BMC Evolutionary Biology 11.
- 46. Hedrick PW, Parker KM, Miller EL, Miller PS (1999) Major histocompatibility complex variation in the endangered Przewalski's horse. Genetics 152: 1701–1710.
- 47. Owen-Smith N, Mason DR (2005) Comparative changes in adult vs. juvenile survival affecting population trends of African ungulates. Journal of Animal Ecology 74: 762–773.
- 48. Ogutu JO, Owen-Smith N (2003) ENSO, rainfall and temperature influences on extreme population declines among African savanna ungulates. Ecology Letters 6: 412–419.
- 49. Drummond A, Ashton B, Cheung M, Heled J, Kearse M, et al.. (2010) Geneious v5.0. Available: http://www.geneious.com. Accessed 1 November 2012.
- 50. Stephens M, Smith NJ, Donnelly P (2001) A new statistical method for haplotype reconstruction from population data. American Journal of Human Genetics 68: 978–989.
- 51. Bos DH, Turner SM, Dewoody JA (2007) Haplotype inference from diploid sequence data: evaluating performance using non-neutral MHC sequences. Hereditas 144: 228–234.
- 52. Harrigan RJ, Mazza ME, Sorenson MD (2008) Computation vs. cloning: evaluation of two methods for haplotype determination. Molecular Ecology Resources 8: 1239–1248.
- 53. Fraser DG, Bailey E (1998) Polymorphism and multiple loci for the horse DQA gene. Immunogenetics 47: 487–490.
- 54. Matocq MD, Shurtliff QR, Feldman CR (2007) Phylogenetics of the woodrat genus Neotoma (Rodentia : Muridae): Implications for the evolution of phenotypic variation in male external genitalia. Molecular Phylogenetics and Evolution 42: 637–652.
- 55. Wickliffe JK, Hoffmann FG, Carroll DS, Dunina-Barkovskaya YV, Bradley RD, et al. (2003) Intron 7 (FGB-I7) of the fibrinogen, B beta polypeptide (FGB): a nuclear DNA phylogenetic marker for mammals. Occasional Papers Museum of Texas Tech University 219: 1–6.
- 56. Prychitko TM, Moore WS (1997) The utility of DNA sequences of an intron from the beta-fibrinogen gene in phylogenetic analysis of woodpeckers (Aves: Picidae). Molecular Phylogenetics and Evolution 8: 193–204.
- 57. Rozen S, Skaletsky H (2000) Primer3 on the WWW for general users and biologist programmers. Methods in Molecular Biology 132: 365–386.
- 58. Dixon CJ (2010) OLFinder-a program which disentangles DNA sequences containing heterozygous indels. Molecular Ecology Resources 10: 335–340.
- 59. Frantz AC, Pope LC, Carpenter PJ, Roper TJ, Wilson GJ, et al. (2003) Reliable microsatellite genotyping of the Eurasian badger (Meles meles) using faecal DNA. Molecular Ecology 12: 1649–1661.
- 60. Taberlet P, Griffin S, Goossens B, Questiau S, Manceau V, et al. (1996) Reliable genotyping of samples with very low DNA quantities using PCR. Nucleic Acids Research 24: 3189–3194.
- 61. Hansen H, Ben-David M, McDonald DB (2008) Effects of genotyping protocols on success and errors in identifying individual river otters (Lontra canadensis) from their faeces. Molecular Ecology Resources 8: 282–289.
- 62. Raymond M, Rousset F (1995) Genepop (Version-1.2) - Population-Genetics Software for Exact Tests and Ecumenicism. Journal of Heredity 86: 248–249.
- 63. Guo SW, Thompson EA (1992) Performing the exact test of Hardy-Weinberg proportion for multiple alleles. Biometrics 48: 361–372.
- 64. Holm S (1979) A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics 6: 65–70.
- 65. Chapuis MP, Estoup A (2007) Microsatellite null alleles and estimation of population differentiation. Molecular Biology and Evolution 24: 621–631.
- 66. Peakall R, Smouse PE (2006) GENALEX 6: genetic analysis in Excel. Population genetic software for teaching and research. Molecular Ecology Notes 6: 288–295.
- 67. Kalinowski ST (2005) HP-RARE 1.0: a computer program for performing rarefaction on measures of allelic richness. Molecular Ecology Notes 5: 187–189.
- 68. Nei M (1987) Molecular evolutionary genetics. Molecular evolutionary genetics: Columbia University Press, New York & Guildford, England. pp. i–x, 1–512.
- 69. Nei M, Tajima F (1981) DNA polymorphism detectable by restriction endonucleases. Genetics 97: 145–163.
- 70. Tajima F (1983) Evolutionary relationship of DNA sequences in finite populations. Genetics 105: 437–460.
- 71. Rozas J, Sanchez-DelBarrio JC, Messeguer X, Rozas R (2003) DnaSP, DNA polymorphism analyses by the coalescent and other methods. Bioinformatics 19: 2496–2497.
- 72. Yuhki N, O'Brien SJ (1990) DNA variation of the mammalian Major Histocompatibility Complex reflects genomic diversity and population history. Proceedings of the National Academy of Sciences of the United States of America 87: 836–840.
- 73. Excoffier L, Lischer HEL (2010) Arlequin suite ver 3.5: a new series of programs to perform population genetics analyses under Linux and Windows. Molecular Ecology Resources 10: 564–567.
- 74. Wright S (1951) The genetical structure of populations. Annals of Eugenics 15: 323–354.
- 75. Weir BS, Cockerham CC (1984) Estimating F-Statistics for the analysis of population structure. Evolution 38: 1358–1370.
- 76. Goudet J (1995) FSTAT (Version 1.2): A computer program to calculate F-statistics. Journal of Heredity 86: 485–486.
- 77. Hedrick PW (2005) A standardized genetic differentiation measure. Evolution 59: 1633–1638.
- 78. Jost L (2008) G(ST) and its relatives do not measure differentiation. Molecular Ecology 17: 4015–4026.
- 79. Crawford NG smogd: software for the measurement of genetic diversity. Molecular Ecology Resources 10: 556–557.
- 80. Lynch M (1990) The similarity index and DNA fingerprinting. Molecular Biology and Evolution 7: 478–484.
- 81. Lynch M, Crease TJ (1990) The analysis of population survey data on DNA sequence variation. Molecular Biology and Evolution 7: 377–394.
- 82. Pritchard JK, Stephens M, Donnelly P (2000) Inference of population structure using multilocus genotype data. Genetics 155: 945–959.
- 83. Evanno G, Regnaut S, Goudet J (2005) Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study. Molecular Ecology 14: 2611–2620.
- 84. Jakobsson M, Rosenberg NA (2007) CLUMPP: a cluster matching and permutation program for dealing with label switching and multimodality in analysis of population structure. Bioinformatics 23: 1801–1806.
- 85. Rosenberg NA (2004) DISTRUCT: a program for the graphical display of population structure. Molecular Ecology Notes 4: 137–138.
- 86. Rogers AR, Harpending H (1992) Population-growth makes waves in the distribution of pairwise genetic-differences. Molecular Biology and Evolution 9: 552–569.
- 87. Hudson RR, Slatkin M, Maddison WP (1992) Estimation of levels of gene flow from DNA-dequence data. Genetics 132: 583–589.
- 88. Harpending HC, Batzer MA, Gurven M, Jorde LB, Rogers AR, et al. (1998) Genetic traces of ancient demography. Proceedings of the National Academy of Sciences of the United States of America 95: 1961–1967.
- 89. Kuhner MK (2006) LAMARC 2.0: maximum likelihood and Bayesian estimation of population parameters. Bioinformatics 22: 768–770.
- 90. Beerli P, Felsenstein J (2001) Maximum likelihood estimation of a migration matrix and effective population sizes in n subpopulations by using a coalescent approach. Proceedings of the National Academy of Sciences of the United States of America 98: 4563–4568.
- 91. Rambaut A, Drummond A (2007) Tracer v1.4. Available: http://tree.bio.ed.ac.uk/software/tracer/. Accessed 1 November 2012.
- 92. Drummond AJ, Rambaut A (2007) BEAST: Bayesian evolutionary analysis by sampling trees. BMC Evolutionary Biology 7.
- 93. Lerner HRL, Meyer M, James HF, Hofreiter M, Fleischer RC (2011) Multilocus resolution of phylogeny and timescale in the extant adaptive radiation of Hawaiian honeycreepers. Current Biology 21: 1–7.
- 94. Yu L, Zhang YP (2005) Phylogenetic studies of pantherine cats (Felidae) based on multiple genes, with novel application of nuclear beta-fibrinogen intron 7 to carnivores. Molecular Phylogenetics and Evolution 35: 483–495.
- 95. Slatkin M (1994) An exact test for neutrality based on the Ewens sampling distribution. Genetical Research 64: 71–74.
- 96. Ewens WJ (1972) Sampling theory of selectively neutral alleles. Theoretical Population Biology 3: 87–&.
- 97. Watterson GA (1978) Homozygosity test of neutrality. Genetics 88: 405–417.
- 98. Tajima F (1989) Statistical-method for testing the neutral mutation hypothesis by DNA polymorphism. Genetics 123: 585–595.
- 99. Fu YX (1997) Statistical tests of neutrality of mutations against population growth, hitchhiking and background selection. Genetics 147: 915–925.
- 100. Yang ZH (2007) PAML 4: Phylogenetic analysis by maximum likelihood. Molecular Biology and Evolution 24: 1586–1591.
- 101. Guindon S, Gascuel O (2003) A simple, fast, and accurate algorithm to estimate large phylogenies by maximum likelihood. Systematic Biology 52: 696–704.
- 102. Posada D, Crandall KA (1998) MODELTEST: testing the model of DNA substitution. Bioinformatics 14: 817–818.
- 103. Bandelt HJ, Forster P, Rohl A (1999) Median-joining networks for inferring intraspecific phylogenies. Molecular Biology and Evolution 16: 37–48.
- 104. Fluxus-engineering website. Available: http://www.fluxus-engineering.com/. Accessed 1 November 2012.
- 105. Hurlbert SH (1971) Nonconcept of species diversity - Critique and alternative parameters. Ecology 52: 577–&.
- 106. Lorenzen ED, Arctander P, Siegismund HR (2008) High variation and very low differentiation in wide ranging plains zebra (Equus quagga) insights from mtDNA and microsatellites. Molecular Ecology 17: 2812–2824.
- 107. Hedrick PW (1999) Perspective: Highly variable loci and their interpretation in evolution and conservation. Evolution 53: 313–318.
- 108. Schierup MH, Vekemans X, Charlesworth D (2000) The effect of subdivision on variation at multi-allelic loci under balancing selection. Genetical Research 76: 51–62.
- 109. Vranova M, Alloggio I, Qablan M, Vyskocil M, Baumeisterova A, et al. (2011) Genetic diversity of the class II major histocompatibility DRA locus in European, Asiatic and African domestic donkeys. Infection Genetics and Evolution 11: 1136–1141.
- 110. Garrigan D, Hedrick PW (2003) Perspective: Detecting adaptive molecular polymorphism: Lessons from the MHC. Evolution 57: 1707–1722.
- 111. Ejsmond MJ, Babik W, Radwan J (2010) MHC allele frequency distributions under parasite-driven selection: A simulation model. Bmc Evolutionary Biology 10: 9.
- 112. Krecek RC, Malan FS, Reinecke RK, de Vos V (1987) Nematode parasites from Burchell's zebras in South Africa. Journal of Wildlife Diseases 23: 401–411.
- 113. Pienaar UD (1967) Epidemiology of anthrax in wild animals and control of anthrax epizootics in the Kruger National Park, Souht Africa. Fed Proc 26: 1497–1502.