Mitochondrial and nuclear DNA reveals reticulate evolution in hares (Lepus spp., Lagomorpha, Mammalia) from Ethiopia

For hares (Lepus spp., Leporidae, Lagomorpha, Mammalia) from Ethiopia no conclusive molecular phylogenetic data are available. To provide a first molecular phylogenetic model for the Abyssinian Hare (Lepus habessinicus), the Ethiopian Hare (L. fagani), and the Ethiopian Highland Hare (L. starcki) and their evolutionary relationships to hares from Africa, Eurasia, and North America, we phylogenetically analysed mitochondrial ATPase subunit 6 (ATP6; n = 153 / 416bp) and nuclear transferrin (TF; n = 155 / 434bp) sequences of phenotypically determined individuals. For the hares from Ethiopia, genotype composition at twelve microsatellite loci (n = 107) was used to explore both interspecific gene pool separation and levels of current hybridization, as has been observed in some other Lepus species. For phylogenetic analyses ATP6 and TF sequences of Lepus species from South and North Africa (L. capensis, L. saxatilis), the Anatolian peninsula and Europe (L. europaeus, L. timidus) were also produced and additional TF sequences of 18 Lepus species retrieved from GenBank were included as well. Median joining networks, neighbour joining, maximum likelihood analyses, as well as Bayesian inference resulted in similar models of evolution of the three species from Ethiopia for the ATP6 and TF sequences, respectively. The Ethiopian species are, however, not monophyletic, with signatures of contemporary uni- and bidirectional mitochondrial introgression and/ or shared ancestral polymorphism. Lepus habessinicus carries mtDNA distinct from South African L. capensis and North African L. capensis sensu lato; that finding is not in line with earlier suggestions of its conspecificity with L. capensis. Lepus starcki has mtDNA distinct from L. capensis and L. europaeus, which is not in line with earlier suggestions to include it either in L. capensis or L. europaeus. Lepus fagani shares mitochondrial haplotypes with the other two species from Ethiopia, despite its distinct phenotypic and microsatellite differences; moreover, it is not represented by a species-specific mitochondrial haplogroup, suggesting considerable mitochondrial capture by the other species from Ethiopia or species from other parts of Africa. Both mitochondrial and nuclear sequences indicate close phylogenetic relationships among all three Lepus species from Ethiopia, with L. fagani being surprisingly tightly connected to L. habessinicus. TF sequences suggest close evolutionary relationships between the three Ethiopian species and Cape hares from South and North Africa; they further suggest that hares from Ethiopia hold a position ancestral to many Eurasian and North American species.

For hares (Lepus spp., Leporidae, Lagomorpha, Mammalia) from Ethiopia no conclusive molecular phylogenetic data are available. To provide a first molecular phylogenetic model for the Abyssinian Hare (Lepus habessinicus), the Ethiopian Hare (L. fagani), and the Ethiopian Highland Hare (L. starcki) and their evolutionary relationships to hares from Africa, Eurasia, and North America, we phylogenetically analysed mitochondrial ATPase subunit 6 (ATP6; n = 153 / 416bp) and nuclear transferrin (TF; n = 155 / 434bp) sequences of phenotypically determined individuals. For the hares from Ethiopia, genotype composition at twelve microsatellite loci (n = 107) was used to explore both interspecific gene pool separation and levels of current hybridization, as has been observed in some other Lepus species. For phylogenetic analyses ATP6 and TF sequences of Lepus species from South and North Africa (L. capensis, L. saxatilis), the Anatolian peninsula and Europe (L. europaeus, L. timidus) were also produced and additional TF sequences of 18 Lepus species retrieved from GenBank were included as well. Median joining networks, neighbour joining, maximum likelihood analyses, as well as Bayesian inference resulted in similar models of evolution of the three species from Ethiopia for the ATP6 and TF sequences, respectively. The Ethiopian species are, however, not monophyletic, with signatures of contemporary uni-and bidirectional mitochondrial introgression and/ or shared ancestral polymorphism. Lepus habessinicus carries mtDNA distinct from South African L. capensis and North African L. capensis sensu lato; that finding is not in line with earlier suggestions of its conspecificity with L. capensis. Lepus starcki has mtDNA distinct from L. capensis and L. europaeus, which is not in line with earlier suggestions to include it either in L. capensis or L. europaeus. Lepus fagani shares mitochondrial haplotypes with the other two species from Ethiopia, despite its distinct phenotypic and microsatellite differences; moreover, it is not represented by a species-specific mitochondrial haplogroup, suggesting considerable mitochondrial capture by the other species from Ethiopia or species from other parts of Africa. Both mitochondrial and nuclear sequences indicate close phylogenetic relationships among all three Lepus species PLOS
For hares from East Africa published molecular phylogenetic data are only available for two mtDNA sequences, one from L. habessinicus and one from L. starcki Pierpaoli et al. [33]; their very close phylogenetic relationship suggests either shared ancestral polymorphism or recent introgression, conspecificity, incorrect phenotypical species identification, or a mix-up of samples. In this study we analyze for the first time evolutionary relationships among L. habessinicus, L. fagani, and L. starcki from Ethiopia, based on molecular data obtained from phenotypically assigned specimens and their phylogenetic relationships with other Lepus species from Africa, Eurasia, and North America. Specifically, we used mitochondrial ATP synthase subunit 6 (ATP6) sequences that have proved useful for molecular phylogenetic resolution at the intraspecific and species levels (Ben Slimen et al., unpubl. data; see also Smith et al. [53]) and intron sequences of one nuclear gene, transferrin (TF) [8]. Given the occurrence of reticulate evolution in several Lepus species (see above) and the close phylogenetic relationship between the two mtDNA sequences of L. habessinicus and L. starcki published by Pierpaoli et al. [33], we expected introgressive hybridzation or shared ancestral polymorphism in hares from Ethiopia. Thus, we produced microsatellite genotypes for the three species from Ethiopia to obtain indications of genetic species distinction compared to phenotype variation and to evaluate nuclear introgression and extent of contemporaneous gene flow across species (e.g., Andersson et al. [54], Kryger [55], Thulin et al. [56], Suchentrunk et al. [40,41], Liu et al. [9], Wu et al. [27], Melo-Ferreira et al. [10]).

Specimens, phenotypic and taxonomical assignments
We based our study on 120 hares collected from 26 locations in Ethiopia (Fig 1) between February 2010 and December 2012 and on 47 hares of three other Lepus species from North and South Africa, the Anatolian Peninsula, and Europe (cape hare, L. capensis sensu lato, scrub hare, L. saxatilis, brown hare, L. europaeus) with supposed close phylogenetic relationships to hares from Ethiopia. All hare samples used in this study were collected with the permit and approval (No. 163/2008) given by the Ethiopian Wildlife Conservation Authority (EWCA) to Zelalem Tolesa (ZT; 2009 to 2014) in the context of his PhD work. The hares were shot by professional hunters based on the permission stated above and tissue samples were collected for further molecular analyses. For outgroup comparison we used two Swiss mountain hares (L. timidus varronis) that were considered evolutionarily divergent from all the species mentioned above (e.g., Pierpaoli et al. [33], Alves et al. [24]). For some few individuals, however, not all molecular data could be obtained, due to insufficient sample quality (see Table 1 for details). For phylogenetic sequence comparison of our presently obtained TF sequences, we included all reliable TF sequences of 18 Lepus species from GenBank.

Laboratory procedures
DNA extraction, PCR amplification, and sequencing. We used the GenElute™ Mammalian Genomic DNA Miniprep kit (Sigma-Aldrich) to extract total genomic DNA from ear tissues and followed Smith et al. [53] to PCR amplify a 416 bp segment of the mitochondrial ATPase sub-unit 6 (ATP6, from site 8142 to 8594; see also Arnason et al. [66]). We amplified a 434 bp fragment of the transferrin gene (TF, between exons 6 and 7) according to Alves et al. [8] and included TF sequences of five rabbits (Oryctolagus cuniculus) from the breeding station at the Wildlife Research Institute of the University of Veterinary Medicine in Vienna (Austria). Moreover, we retrieved 111 TF sequences of 13 Lepus species plus one European rabbit two of Sylvilagus floridanus from GenBank for phylogenetic comparison (S1 Table). However, our ATP6 primers did not yield satisfactory results for the Oryctolagus samples, and no ATP6 sequences were available for Sylvilagus on GenBank.
We used the enzymatic clean-up process using Exonuclease I and Shrimp Alkaline Phosphatase (Werle et al. [67]) for PCR product purification and performed cycle sequencing of both strands with an ABI 3130xl genetic analyzer. We used BioEdit vers. 7.1.3.0 (Hall [68]) for sequence alignment and checked alignments by eye. We used composite sequences as obtained from both strands for phylogenetic analyses. We used the Phase 2.1.1 algorithm [69,70], as implemented in DnaSP vers. 5.10.01 Labrado and Rozas [71], to reconstruct haplotypes (e.g., Flot et al. [72], Garrick et al. [73]) for TF sequences with more than one ambiguity.

Phylogenetic analyses of ATP6 and TF sequences
We used DnaSP to calculate numbers of variable sites, phylogenetic informative sites and singletons as well as haplotype (h) and nucleotide (π) diversity, and mean numbers of pairwise differences (k) to indicate levels of polymorphism.
We constructed phylogenetic haplotype networks for both sets of sequences by using the median-joining (MJ) algorithm (with default parameter settings) implemented in the Network software, version 4.6.1.1 [74]. Contrary to tree-based inference, networks can uncover possible alternative evolutionary pathways of haplotypes; hence, the depiction of evolutionary relationships among haplotypes is not reduced to bifurcating events and displaying taxa only at terminal positions. This may facilitate inference on the presence of shared ancestral polymorphism or recent introgression, particularly for clusters with low bootstrap support (e.g., Hall [75], Morrison[76]).
In addition, we used MEGA vers. 5.0 Kumar et al. [77], Tamura et al. [78] to analyze phylogenetic relationships among haplotypes by Neighbor Joining (NJ), Saitou and Nei [79], Tamura and Kumar [80] and Maximum Likelihood (ML), Kimura [81] modelling. We also included 111 TF seqeuences of 18 Lepus species from different parts of the world, one European rabbit and two Sylvilagus floridanus (Eastern Cottontail Rabbit) sequences, downloaded from Gen-Bank, but accepted only those with a maximum of one nucleotide base ambiguity in our alignment, in order to exclude the chance of picking up problematic ones. Prior to NJ analyses, we checked sequences for suitability by calculating average Jukes Cantor (JC) distances (criterion: average pairwise JC distance < 1.0; Nei and Kumar [82], see e.g., Hall [75]). We used the Maximum Composite Likelihood model for NJ trees, as recommended by Kumar et al. [77], (also Hall [75]), and assessed confidence in resulting nodes by 1000 bootstrap replicates. For ML analyses, we applied the most appropriate evolutionary model based on BIC values (e.g., Posada [83]), according to the model test option in MEGA with 1000 bootstrap replicates. We also used MEGA to calculate net mean between group p distances for ATP6 and TF sequences.
We used MrBayes v. 3.2 Ronquist et al. [84], that estimates phylogenetic relationships by seeking the most likely tree given our actual sequence data and the chosen substitution model. We began our Bayesian inference (BI) with random starting trees and ran it for 10 million generations, with Markov chains sampled every 1000 generations. We checked the average standard deviation of split frequencies for parameter convergence. The first 2.5 million generations were excluded as burn-in. We conducted the BI twice to ensure that the analyses were not trapped in local optimum [85,86]. The remaining trees from both analyses were used to create a majority rule consensus tree where the percent of samples recovering the same clade represented the posterior probability value of that clade. For tree viewing and editing we used FigTree v.1.4.2 [87].
Population genetic statistics. Allelic variability, Hardy Weinberg, and Linkage disequilibrium We calculated allele frequencies, mean number of alleles (A), observed (Ho) and expected (He) heterozygosity for each locus and species with GENETIX 4.05. 2 Belkhir [93]. We tested each locus, separately for each species, for deviation from Hardy-Weinberg equilibrium (HWE) and pairs of loci for linkage disequilibrium (LD) using the Markov chain method implemented in GENEPOP version 3.4 (Raymond and Rousset [94], with default parameter settings and accounted for multiple tests by strict Bonferroni corrections at α = 0.05 Rice [95],; the latter was used for all further test series. Due to a significant LD signal in L. habessinicus (see Results), we excluded the Lsa2 locus from all further calculations. We calculated speciesspecific Weir and Cockerham [96] estimators of population level "inbreeding coefficients" due to non-random mating (F is ) and associated significance levels for deviation from zero by permutation tests (10.000 permutations) by using GENETIX.
Evaluating possible homoplasy We could not exclude size homoplasy, particularly at highly polymorphic loci (e.g., Garza and Freimer [97]; but see Ben Slimen et al. [37], for hares from Europe and Africa). Under pronounced homoplasy, genetic differentiation is expected to be higher in lesser polymorphic loci (e.g., Ben Slimen et al. [37]). We used S-PLUS 6.2 [98] to run a generalized least squares regression model (gls) to test locus-specific f st values (across the three Ethiopian species, respectively) for variation due to locus-specific numbers of alleles and sample sizes (fixed factors), by accounting for occasionally missing data. Significantly higher f st values due to lower numbers of alleles per locus would concord with a homoplasy effect. Additionally, we counted genotypes shared by two or all three species, separately for each locus. Low genotype sharing among species would suggest negligible homoplasy, particularly, if the shared genotypes occurred either at low or at very divergent frequencies among species.
Testing for reduced effective population size We used BOTTLENECK vers. 1.2.02 Cornuet and Luikart [99], Luikart and Cornuet [100],) to run one-tailed Wilcoxon sign rank tests (10.000 iterations), separately for each species, to test for reduced effective population size in the recent past that could have influenced genetic variability and differentiation among the Ethiopian species. Thereby, we applied the stepwise mutation model (S.M.M), the infinite allele model (I.A.M.), and a two-phased model (T.P.M., with default settings) of microsatellite evolution. Additionally, we calculated the mean ratio of the number of alleles to the range of allele size (M ratios) separately for each species, to infer possible genetic bottlenecks in the more distant past [101].
Genetic differentiation and contemporary introgression We calculated pariwise F st values of relative genetic differentiation between species and tested significance by 10.000 permutations by using GENETIX. We used GeneAlex v. 6.5 [102] to calculate pairwise G´s t Hedrick [103], and D est (i.e., Jost´s D; Jost [104]) and ssociated significance levels (10.000 randomizations); the latter two indices account for high heterozygosity, whereas F st does not, which may underestimate differentiation. Moreover, we used MSA 4.05 [105] to calculate pairwise Cavalli-Sforza and Edwards chord (CSE) distances and constructed a NJ dendrogram using PHYLIP 3.6.9.5 [106]. Individuals clustering with non-conspecifics might highlight recently introgressed individuals. Further, we used Arlequin v. 3.11 [107] for an analysis of molecular variance (AMOVA) among the three species.
We used GENECLASS 2 [108] to determine the likelihood of an individual being a "first generation migrant" between the phenotypically determined species using the criterion of Paetkau et al. [109] and the Bayesian simulation algorithm of Paetkau et al. [110] with 10.000 simulated individuals and default allelic frequencies of 0.01. "First generation migrants" detected within a phenotypically determined species would in principle indicate incorrect phenotypic species determinations or individuals being massively introgressed by genes of a different species. In addition, we estimated the likelihood of an individual's multilocus genotype to be assigned to the phenotypically determined species from which it was sampled by the Bayesian method of Rannala and Mountain [111] and the resampling (100.000 simulated indivuduals) algorithm of Paetkau et al. [109].
We used STRUCTURE [112,113] to assess the most likely number of population groupings (i.e., k genetic clusters) compatible with the observed genotypic distribution across all samples. We run models for all 107 individuals based on 12 loci, without and with prior population (i.e., species) information under admixture models and correlated allele frequencies. The likelihood when assuming different numbers of genetic clusters (k = 1 to 11) was calculated by 150.000 MCMC following a burn-in of 50.000, and ten iterations per k. Mean and standard deviation of likelihood values were calculated for each k and plotted together with Evanno´s [114] ad hoc statistic by using the STRUCTURE HARVESTER on-line option [115]. The model runs were repeated with only eight loci (i.e., excluding Sol33, Lsa3, Sat2, Sat8), as null alleles at higher frequencies (up to 30%) could not be excluded for at least one species at those loci.
We used MIGRATE v.3.0 [116] for coalescent-theory based ML estimates of current migration rates between the three species, as additional indication of contemporary hybridization between species. We used the following specifications for our "Brownian Motion" model assuming constant mutation rates for 12 loci: 10 short chains with 5.000 recorded steps, 500.000 visited genealogies, burn-in of 10.000; 2 long chains with 50.000 recorded steps and 10.000.000 visited genealogies; for the set of eight loci (see above) the specifications were: 10 short chains with 1.000 steps, 200.000 visited genealogies, burin-in of 10.000; 2 long chains with 1.000 recorded steps and 100.000 visited genealogies.

Mitochondrial ATP6 sequence variation and phylogenetic relationships
We obtained ATP6 sequences of 104 hares from Ethiopia (L. habessinicus, L. starcki, L. fagani) and of 49 specimens of four species (L. capensis, L. saxatilis, L. europaeus, L. timidus) from Africa, the Anatolian Peninsula, and Europe (Table 1). All new variable sequences were deposited on GenBank (see also S2 Table). All sequences could be translated into proteins (hence, no suspect of pseudogenes) and the composite alignment encompassing 416 base pairs revealed 64 haplotypes. Overall haplotype diversity (h) was 0.952, overall nucleotide diversity (π) was 0.054, and average number of nucleotide difference (k) was 22.36.
The median joining network (Fig 2) grouped all North and South African cape hares clearly separate from all other hares. The two L. timidus sequences resulted in one haplotype at a position far distant from all hares studied presently, and connected to a node between the South and North African L. capensis phylogroups on the one hand and all other hares studied on the other. For L. habessinicus two phylogroups were revealed, phylogroup A consisting of haplotypes from southwestern Ethiopia and the southern Rift Valley in Ethiopia, and phylogroup B comprising haplotypes from central and eastern Ethiopia, and the northern Rift Valley in Ethiopia. Phylogroup A was basal to phylogroup B, as revealed by their positions relative to L. timidus and L. capensis in the network. Phylogroup A, however, also contained sequences of eight L. fagani, one L. starcki, one L. saxatilis, and one morphologically unidentified subadult Lepus individual from western Ethiopia (AS, see Fig 1). Except for the one L. starcki sequence in phylogroup A, all other L. starcki sequences clustered into a single-evolutionarily young-phylogroup ("L. starcki phylogroup") closely related to phylogroup B of L. habessinicus. Most sequences of L. fagani clustered into phylogroup A of L. habessinicus, two clustered with the L. starcki phylogroup, and one sequence represented a single haplotype between phylogroup A of L. habessinicus and the L. starcki phylogroup. Three L. habessinicus sequences clustered within the L. starcki phylogroup. All 14 L. europaeus sequences clustered into two closely related phylogroups, a European and an Anatolian Peninsula phylogroup. Neither the Ethiopian L. fagani nor the South African L. saxatilis sequences formed distinct phylogroups; rather, they were scattered across three phylogroups, two Ethiopian (phylogroup A of L. habessinicus, L. starcki phylogroup) and the European phylogroup, consisting of L. europaeus sequences from Europe.
Among all 125 polymorphic sites, 17 were singletons, and 108 were phylogenetically informative. Under the BIC criterion, the most appropriate model of sequence evolution was the HKY+G (Γ = 0.31). A matrix of net mean between group (i.e., species) p distances for all 153 hares sequenced presently is displayed in S3 Table. The BI, ML, and NJ analyses produced in essence concordant tree topologies. The average standard deviation of the split frequencies of the BI was 0.004 when runs were added. The BI majority-rule consensus tree is shown in Fig 3   Fig 2. Median joining network of ATP6 haplotypes. Haplotypes (pies) are proportional to total sample number, taxon assignments of single haplotypes (pie slices) represent percentages of taxa per haplotype. Black dots indicate inferred haplotypes, not revealed presently, numbers associated with lines give numbers of substitutions between any two haplotypes/inferred haplotypes, if more than one (single mutational steps between any two haplotypes are not indicated). Evolutionary distances between haplotypes are only roughly in proportional scale. Taxa

Transferrin (TF) sequence variation and phylogenetic relationships
We obtained TF sequences of 154 hares (110 hares from Ethiopia, 44 hares of four Lepus species from Africa, Asia Minor, and Europe and of five European rabbits (Oryctolagus cuniculus) ( Table 1). All new variable TF sequences are available from GenBank (for accession numbers and details see S4 Table and for pairwise net p distances among taxa see S5 Table). Among our TF sequences, we observed one insertion (1bp) in four L. starcki, in ten L. europaeus, and in the two L. timidus at the same position as in several of the downloaded sequences. Our presently produced sequences had a maximum of five ambiguities, and resulted in an overall alignment length of 427 base pairs, when the downloaded sequences were included. The overall alignment included six indels with a total length of 17 sites. However, all were excluded from phylogenetic analyses. Phasing of the overall alignment resulted in 92 haplotypes without ambiguous resolution and 110 polymorphic sites. Overall h was 0.942, π amounted to 0.02, and k was 8.15.
The network topology reflected only roughly the major geographic origin of the species, with some haplotypes, however, being shared by species from different continents (Fig 4). It was partitioned into two major phylogroups (A and B) in addition to the Oryctolagus/Sylvilagus outgroup. Phylogroup A appeared closer to the outgroup than phylogroup B. This was also indicated by the BI tree, whereas the NJ and ML trees did not recover that topology (Fig 5).  sequences connected the outgroup haplotypes with a major haplotype of phylogroup A. That latter haplotype was found in L. habessinicus, L. starcki, and North African L. capensis, as well as in L. timidus and L. yarkandensis. It gave rise to a number of haplotypes of species of Asian (Chinese L. capensis, L. mandshuricus, L. sinensis, L. yarkandensis) and African origin (North and South African L. capensis, L. saxatilis), and further to a closely related haplogroup, typical for Ethiopian species and including also L. saxatilis and one L. capensis from South Africa. Sequence evolution was best described by the K2+G (Γ = 0.5389) model under the BIC criterion. The overall TF alignment (n = 377 sequences of 159 individuals presently analyzed and 111 individuals from GenBank) revealed 110 variable sites, 77 phylogenetically informative sites, and 33 singletons (for net mean between group p distances see S5 Table). Whereas the BI tree topology (average s.d. of the split frequencies of the BI was 0.007 when runs were added) reflected by and large the network topology; the NJ and ML analyses, however, resulted in somewhat different tree topologies.  Table 2, as well as species-specific H e , H o , A, numbers of private alleles and F is . Significant deviations from HWE were found in each species. Our MICRO-CHECKER runs did not exclude presence of null alleles at the Sol33, Lsa3, Sat8, Sat2 loci for one or more species, with estimated frequencies exceeding somewhat 20%. Null alleles with frequencies up to ca. 20-30% do not seriously confound population genetic results Chapuis and Estoup [118]; however, for confirmation of the results based on 12 loci, we repeated several analyses (Bottleneck, Geneclass, Migrate, M ratios, Structure) by excluding the Sol33, Lsa3, Sat2, Sat8 loci.
Homoplasy signal. Our gls model indicated only a tendency for slightly higher f st values (across the three species) for lesser polymorphic loci compared to higher polymorphic loci, when accounting for varying sample sizes (coefficient = -0.0233, F 1,7 = 3.813, p = 0.083). That tendency was, however, driven by only one locus (Lsa 6). Moreover, our genotype comparisons between the species revealed that only 18.7% of all 337 recovered genotypes at 12 loci were shared by two species, and only 1.8% was shared by all three species. Mean species-specific frequencies of shared genotypes at highly polymorphic loci (37-51 genotypes) averaged at 8.3% (median = 6.6%, range: 2.9-42.1%). For lesser polymorphic loci (6-30 genotypes) the frequencies averaged at 12.5% (median = 7.8%, range: 2.4-77.5%). At Sol30, the locus with most (51) genotypes, only 9.8% of all genotypes were shared by two species, with a mean species- Na-number of alleles per locus, R-allelic size range in bp, he-locus-specific expected heterozygosity, ho-locus-specific observed heterozygosity, separately for each species, He-species-specific expected heterozygosity, Ho-observed species-specific heterozygosity, A-mean number of alleles per locus, F IS −species-specific "inbreeding coefficient". The number of private alleles for each species is given in parentheses associated with the species name. Significant deviation from HWE is indicated by * with F IS .
https://doi.org/10.1371/journal.pone.0180137.t002 specific frequency of 13.0%, a median of 6.6%, and a range between 4.8 and 42.1%; the maximal value of 42.1% resulted from the very high frequency (80.8%) of a genotype in L. starcki but a very low corresponding frequency (3.4%) in L. habessinicus (suggesting also no homoplasy problem). Overall, potential homoplasy seemed to be of no major concern for our data as already observed by Ben Slimen et al. [37]. Reduction of genetic population size. Wilcoxon sign rank tests did not return any significant signals of a recent bottleneck, when accounting for multiple tests. The M ratios (L. habessinicus: 0.786/0.788; L. starcki: 0.56/0.518; L. fagani: 0.7/0.738 for 12 and 8 loci, respectively), however, indicated a reduction of the effective population size for L. starcki in the more distant past.
Genetic differentiation, molecular species assignment, recent introgression Overall F ST was 0.275 (95% c.i.: 0.208-0.364); Table 3 displays pairwise F st values between species, along with pairwise G´s t and Jost´s D. The AMOVA revealed significant partitioning of relative genetic variability among the three species at 27.44% (variance comp. = 1.360, p < 0.00001), among individuals within species at 14.3% (variance comp. = 0.709, p < 0.00001), and within individuals at 58.26% (variance comp. = 2.890, p < 0.00001). Within L. habessinicus, only 3.59% of relative genetic variability were due to partitioning between the ancestral and the recent ATP6 phylogroups A and B.
Our initial likelihood analysis of "first generation migrants" using GENECLASS indicated that one phenotypical L. starcki neonate collected in the Abijata-Shala National Park in the Rift Valley (AB, Fig 1) has been misidentified, in fact belonging to L. habessinicus (likelihoods: L. starcki = -37.08; L. habessinicus = -16.367; L. fagani = -28.931). After allocating that specimen to L. habessinicus, Bayesian species assignments of individuals matched in all cases the phenotypically determined species, when based on 12 or occasionally slightly fewer loci (due to missing genotypes); however, one (phenotypic and molecular) L. habessinicus was assigned to L. fagani when using only eight loci.
The NJ tree (S1 Fig) based on pairwise individual CSE distances revealed a set of seven minor or more comprehensive clusters of L. habessinicus individuals very closely related to each other, one distinct L. fagani clade, closely related to two of the L. habessinicus clades, and a distinct L. starcki clade, slightly more separate from the L. habessinicus clades and most distant from the L. fagani clade. Two L. starcki and two L. fagani individuals clustered within  Thus, we considered those latter two individuals as recent hybrids, rather than as drift signals of ancestral alleles. A few more hares might be viewed as offspring of recent hybrids, but the admixture results were not concordant in all STRUCTURE models. Our panel of eight loci, however, did not fully qualify for identifying modern hybrids. Notably, our STRUCTURE models suggested significant substructures within L. habessinicus and L. fagani without much admixture, respectively.
Current gene flow between species as estimated by MIGRATE ranged between 0.311 and 2.447 individuals per generation for the 12 loci data set and between 0.265 and 3.360 individuals per generation for the 8 loci data set, respectively (Table 3). Whereas migration was rather balanced between L. habessinicus and L. starcki, it was asymmetric between L. habessinicus and L. fagani, and between L. starcki and L. fagani, with clearly less gene flow from L. fagani to either species.

Phenotypes and molecular species assignment
Given the possibility of regionally varying scenarios of reticulate evolution (e.g., Thulin and Tegelström [119], Alves et al. [8], Liu et al. [9], Melo-Ferreira et al. [10,11]) and the large phenotypic variance within and among species or taxa Angermann [2], molecular phylogenetic studies of the genus Lepus must be accompanied at least by external phenotype analysis in regions of species sympatry or parapatry. This should also help to prevent ambiguities and future confusions of molecular data archived in repositories.
All presently collected hares from Ethiopia could be unambiguously assigned to Ethiopian hares Lepus fagani, Abyssinian hares, L. habessinicus, or Ethiopian Highland hares, L. starcki, based on their external phenotype, except for one subadult individual from Assosa (AS), western Ethiopia, close to the North Sudan border, and one neonate L. habessinicus from the Abijata-Shalla National Park (AB) in the central Ethiopian Rift Valley (Fig 1). The remains of the external phenotype and the ATP6 haplotype of the former (from AS) corresponded by and large to that of typical L. fagani, but due to technical reasons we repeatedly failed to obtain meaningful microsatellite data for its final molecular assignment. The neonate from AB, however, featured an external phenotype typical for L. starcki, but was undoubtedly assigned to L. habessinicus by all our molecular markers. That mismatch was not due to possible sample confusion, because that individual featured a unique multi-locus microsatellite profile, when cross-checked for its relationship with all other L. habessinicus individuals by Queller and Goodnight [120] r xy indices (using IDENTIX 1.0/1.0.2 [121]). Moreover, no hint of possible introgression by L. starcki was detected in that neonate, as revealed by allele/genotype comparisons, STRUCTURE, and the Bayesian assignment based on its full microsatellite set. Generally, within Lepus phenotype-based species determination of neonates is deemed impossible, though-as to our knowledge-not systematically studied. However, the conspicuous outer pinnae pattern and the often entire white upper surface of the tail represent a combination of external characteristics of L. starcki that cannot be confused with phenotypes of other hares occurring in Ethiopia (e.g., Yalden et al. [60]). Notably, the single neonate with a L. starcki phenotype from the Abijata-Shalla National Park in the Rift Valley did display those two characteristics very distinctly. No specific results are available to interpret that phenotype/molecular data mismatch. Possibly, few introgressed genes underlying the addressed external characteristics (e.g., coat colour genes; Koutsogiannouli et al. [122]) were not reflected by our overall microsatellite set. Notably, despite distinct coat colour differences, only shallow molecular divergence was observed among North African cape hares (L. capensis) and hares from Israel, cf., L. capensis and L. europaeus Ben Slimen et al. [21,37], as well as between Mandshurian hares (L. mandshuricus) and the blackish "L. melainus" form from eastern Asia considered conspecific with L. mandshuricus [44,123]. Even certain alleles of single coat colour genes (TYR, MC1R) did not predict the two winter coat types in Japanese hares, L. brachyurus [45].
The vast majority (99.1%) of all our comparisons between phenotype and molecular data, nevertheless, resulted in unambiguous species determination by external phenotypes alone, without doubtful intermediate forms hinting towards possible hybrids. External phenotypes of Somalian cape hares, L. capensis, as determined by Azzaroli-Puccetti [57], however, are very similar to our L. habessinicus (unpub. data of one of the authors (FS) collected at the Museum of Zoology and Natural History, "La Specola", Florence, Italy). Moreover, several skins in the collection of "La Specola" (Florence) assigned to L. crawshayi (= L. victoriae) by M. L. Azzaroli-Puzzetti are virtually indistinguishable from standard "L. habessinicus" phenotypes (pers. observ. of FS). Similarly, external phenotypes of the currently investigated L. fagani may not differ much from those of some forms of East African savanna hares, L. victoriae (e.g., Angermann [1], Flux and Flux [117], see also Yalden et al. [60], Flux and Angermann [5]). All these ambiguities call for thorough examination of phenotypic character variation, including skull and dental characters as well (e.g., Palacios et al. [39]) in the context of molecular phylogenetic studies of hares (from Africa).

Molecular phylogenetic relationships among hares from Ethiopia and other closely related species
Somewhat unexpectedly, both our ATP6 and TF sequences indicate generally very close phylogenetic relationships between L. fagani, L. habessinicus, and L. starcki, despite their clear phenotypic and genetic distinction as reflected by microsatellites. All three species are paraphyletic both in their mitochondrial and nuclear sequences, which is not uncommon in the genus Lepus (e.g., Alves et al. [8], Wu et al. [19], Melo-Ferreira et al. [18], Ben Slimen et al. [21], Melo-Ferreira et al. [22], Liu et al. [9]). Particularly during short periods of adaptive radiation and with large effective sizes of ancestral populations incomplete lineage sorting is likely to result in shared ancestral polymorphism in modern species (e.g., Degnan and Rosenberg [124]). Although modern hares and jackrabbits (Lepus) may represent evolutionarily young offshoots of (now extinct) forms given currently genera ranks in paleontological taxonomy, likely dating back to the late Miocene (see e.g., Matthee et al. [125]), their major and rapid adaptive radiation has happened only in the recent past, as inferred from paleontological (e.g., Lopes-Martinez [126]) and molecular studies (Robinson and Matthee [127], and references therein); supposed pre-Pliocene Lepus finds in Africa, need a thorough revision [128]. Effective sizes of ancestral hare populations with little ecological adaptation to different niches and little behavioural specifications may have been large, especially at shallow genetic divergence across large ranges (comp. Campos et al. [129] for Ne estimates of extant L. europaeus) and this might have favored incomplete lineage sorting. Regarding potential incomplete lineage sorting of the currently studied MT-ATP6 sequences would not exclude (temporal and spatial) scenarios of positive or negative selection [10].

Relationships between Lepus habessinicus and L. capensis
Our mtATP6 data (Figs 2 and 3) indicate that L. habessinicus and the other two species from Ethiopia are clearly evolutionarily separated from South and North African L. capensis, without any hint of shared ancestral polymorphism. Thus, for L. habessinicus, the straightforward evolutionary interpretation based on mtDNA would suggest a species quite distinct from L. capensis, in agreement with Azzaroli-Puccetti [57,58]. On the other hand, in our evolutionary network model the most ancestral haplotype of L. habessinicus holds a position close to where the lineages of L. europaeus, L. timidus, and L. capensis meet and close to the L. europaeus phylogroup. This may suggest its relatively ancestral origin together with L. europaeus. Notably, one haplotype of L. saxatilis (scrub hare) is ancestral to both L. habessinicus and L. europaeus and shares one relatively old haplotype with North African L. capensis and L. europaeus, though the latter species does not occur in Africa, and L. saxatilis occurs only in South Africa, where it does obviously not share haplotypes with the sympatric L. capensis. The most ancestral L. habessinicus haplotype is also shared by L. saxatilis. This scenario might suggest that lineages that are currently found in L. saxatilis, L. europaeus, the ancestral L. habessinicus phylogroup, and occasionally in North African L. capensis are amongst the most ancestral of all currently studied species. Thus, alternative to the straightforward phylogenetic interpretation above, we hypothesize that L. habessinicus represents a comparatively old descendant of an "ancestral L. capensis form" (a precursor of modern L. capensis sensu lato), probably originating from eastern or central Africa, that has given rise to lineages now found in modern South and North African L. capensis. This hypothesis is supported by our transferrin (TF) sequences: 1) they indicate a close evolutionary relationship between South and North African cape hares and 2) all African L. capensis haplotypes are descendants of the most ancestral L. habessinicus haplotype that is also the most ancestral among all haplotypes found in the currently studied African taxa and all other Lepus species from America and Eurasia, except for one haplotype of L. californicus from America and one of L. hainanus from Southeast Asia (Fig 4); the latter two haplotypes link to the outgroup genera Oryctolagus and Sylvilagus.
Our hypothesis of L. habessinicus representing a close taxon to an "ancestral L. capensis form" would help to explain the close phenotypic and morphological relationships among L. habessinicus and all South and North African forms of L. capensis ("L. capensis sensu lato": Angermann [1,2], Yalden et al. [60], Flux and Angermann [5], see also Azzaroli-Puzzetti [57,58]). It could further explain the big divergence of mitochondrial control region sequences between South and North African cape hares Suchentrunk et al. [6], despite their clearly lower mtDNA divergence as estimated by RFLPs Ben Slimen et al. [52], and their quite close genetic relationship as estimated by allozymes Suchentrunk et al. [6] and microsatellites Ben Slimen et al. [37]. Obviously, mitochondrial control region sequences can reach remarkably high divergence levels in hares with close nuclear gene pool relationships, as in South and North African cape hares Kryger [55], Ben Slimen et al. [31], Suchentrunk et al. [40,41].
Mitochondrial and nuclear sequences of phenotypical East African cape hares (e.g., Flux and Flux [117]) should allow testing our hypothesis of the ancestral status of L. habessinicus and its close relationship to East African L. capensis connecting cape hares from South and North Africa. Results based on few available mtDNA and nuclear sequences of African L. capensis [130,33,8,125,21], however, are ambiguous. For instance, European brown hares (L. europaeus), Iberian hares (L. granatensis), and several American Lepus species appear ancestral to (North) African cape hares in cyt b sequences [8]. On the contrary, mitochondrial control region sequences of South African cape hares are basal to all other taxa studied, whereas control region sequences of North African cape hares are closely related to those of L. europaeus and L. saxatilis Ben Slimen et al. [21]; however, in all those analyses basal nodes were not supported, and different sets of taxa were used for the different phylogeny models.
Our TF sequence results indicate that L. europaeus appears evolutionarily younger than all African L. capensis and L. habessinicus, whereas South African scrub hares, L. saxatilis, may hold a relatively ancestral position. The evolutionary relationship between South African L. saxatilis and African savanna hares, L. victoriae, however, is not fully understood. Microsatellite data Kryger [55] indicate high gene flow between L. saxatilis from SW South Africa and the range of L. victoriae in N and NE South Africa, Botswana, Namibia, and parts of Zimbawe, where L. saxatilis has traditionally been considered absent. A conspecific status of those two taxa as proposed by Robinson and Dippenaar [51] (see also Collins[50]), for instance with clinal genetic divergence (i.e., isolation by distance) and a resultant species range from South to North Africa of L. saxatilis (being hypothetically conspecific with L. victoriae) could significantly influence the evolutionary interpretation of the currently revealed ATP6 haplotypes shared by L. saxatilis and hares from Ethiopia. The generally not well supported basal nodes of mtDNA-based phylogenies of the genus Lepus published so far are congruent with a supposed quick radiation of the genus Lepus during the Pleistocene (e.g., Matthee et al. [125]). This calls for numerous samples from geographically dispersed origins and comprehensive molecular marker sets to unravel detailed phylogenetic relationships of African hares.
Phylogenetic position of Lepus starcki. The Ethiopian highland hare, Lepus starcki, has traditionally been viewed as closely related to L. capensis or L. europaeus, or possibly representing a relict form of one or both of them (e.g., Petter [59], Angermann [1]; see also Yalden et al. [60,61], Azzaroli-Puccetti [57,58], Flux and Angermann [5]). Phenotypically, it is very distinct from all other East African species; its external phenotype superficially resembles that of L. europaeus, a species absent from Africa. All but one mitochondrial lineages of L. starcki occur in a derived haplogroup ("L. starcki haplogroup"; see Fig 2), related to the modern haplogroup of L. habessinicus; this suggests its rather recent evolution from L. habessinicus. The two mitochondrial L. starcki and L. habessinicus sequences published by Pierpaoli et al. [33], our microsatellite and TF sequences results are in principle not contradictory to that interpretation. The presence of three haplotypes of the "L. starcki haplogroup" in individuals of L. habessinicus does agree with this interpretation as well, and the occurrence of a single L. starcki haplotype in the ancestral L. habessinicus haplogroup may be viewed as shared ancestral polymorphism, or ancient or modern introgression.
However, L. starcki is the only African hare species that harbors haplotypes in all three (ancestral and modern) major TF phylogroups. Notably, it shares one TF haplotype with the Holarctic L. arcticus, L. othus, L. timidus (i.e., L. timidus complex) and with L. townsendii. On the contrary, one L. timidus individual shares the most ancestral haplotype among all African species with L. starcki, L. habessinicus, North African cape hares, and one central Asiatic specimen of L. yarkandensis. Lepus arcticus, L. othus, L. timidus, and L. townsendii are all closely related forms (e.g., Flux and Angermann [5], Melo-Ferreira et al. [28]) absent from Africa, but adapted to cold Palearctic or Nearctic climate regions in the Palearctic or Nearctic. Taking into consideration the results of all three currently studied molecular marker systems, the most parsimonious phylogenetic interpretation would be that L. starcki has acquired that relatively modern TF haplotype shared with the cold adapted Palearctic and Nearctic hares by ancient introgression during an early phase of adaptive radiation starting from a (Holarctic?) precursor form that has led to the modern L. timidus complex (and to L. townsendii). This could have happened during one of the cold phases of the middle or late Pleistocene under a range expansion of either L. starcki or the (hypothetical) Holarctic precursor, or of both; it would help to explain the adaptation of L. starcki to cooler environments in tropical East Africa, i.e., its restriction to the Ethiopian highlands, east and west of the Rift Valley (e.g., Yalden et al. [60], Flux and Angermann [5]). The surface of the tail is often fully white in L. starcki [60,5], in congruence with many forms of the L. timidus complex and L. townsendii; and that might hint towards such a hypothesized ancient introgression by an arcto-alpine precursor of the modern L. timidus complex and L. townsendii.
An alternative interpretation could be that L. starcki represents an ancestral (relict) species that has relatively recently been massively introgressed by expanding L. habessinicus, as observed for mtDNA of Chinese L. mandshuricus that has been entirely replaced by that of L. timidus or L. sinensis [9]. Our microsatellite data, however, do not readily support this interpretation, though the significant signal of a reduction of its effective population size in the more distant past is not incongruent with such an evolutionary scenario. Future more geographically spaced L. starcki and L. timidus samples, together with multiple nuclear sequences, should enable evaluating the likelihood and extent of ancient introgression of ancestral L. timidus and allied cold-adapted taxa into L. starcki, similar to the scenario revealed in hares from the Iberian Peninsula (e.g., Alves et al. [8], Melo-Ferreira et al. [28]) or the massive introgression of L. timidus into Chinese L. capensis [9].
Another potential interpretation could be that the TF sequence shared by L. starcki and forms of the L. timidus complex and L. townsendii represents in fact a L. habessinicus haplotype that has given rise to the haplotypes present in the modern L. timidus complex and L. townsendii, as well as in L. starcki, but has so far not been detected in L. habessinicus itself, due e.g., to our restricted geographical sampling. The one L. timidus individual that shares the most ancestral transferrin haplotype of all L. habessinicus haplotypes lends support for this hypothesis, but without comprehensive geographical TF data for both L. habessinicus and L. starcki we consider it rather speculative.
Still more evolutionary scenarios are conceivable for L. starcki, such as ancient introgression by L. europaeus that has in turn been introgressed by L. timidus; the 1bp-long transferrin insertion shared by L. starcki, L. europaeus, and L. timidus, as well as similar external phenotypes, skulls, and dental characters of L. starcki and L. europaeus might suggest this. However, a thorough evaluation of such alternative hypotheses needs a much more comprehensive arrangement of samples and molecular makers.
Lepus fagani-Young or ancestral species?. The Ethiopian hare, Lepus fagani, represents one of the least known Lepus forms currently given species rank (e.g, Flux and Angermann [5], Hoffmann and Smith [12], Alves and Hackländer [13]). As to our knowledge no molecular data have so far been published on it. Equally, no modern literature on phenotypes and morphometry or other aspects of biology does exist for that taxon, nor is its range reasonably described. Traditionally, it is considered either a subspecies of the African savanna hare, L. victoriae, or as a separate species closely related to the latter based on earlier descriptions of fur, skull, and dental characteristics (e.g., Angermann [1], Yalden et al. [60,61], Azzaroli-Puccetti [57,58], Flux and Angermann [5]).
Our microsatellite results and TF sequences concordantly indicate a very young evolutionary descent of L. fagani from L. habessinicus, strikingly contradicting the traditional systematic view. The mitochondrial ATP6 sequences are also not incongruent with the hypothesis of a very recent evolution from the most basal L. habessinicus haplotype or with the hypothesis of shared ancestral polymorphism and relatively recent or ancestral introgression by lineages of the (more derived) haplogroup in hares from Ethiopia that is otherwise typical for L. starcki. Alternatively, the ATP6 network of L. fagani haplotypes may be interpreted as representing an ancestral lineage system that has totally been substituted by L. habessinicus lineages and the "L. starcki haplogroup" (i.e., "mitochondrial capture"). In fact, a "L. fagani-typical ATP6 haplogroup" or even haplotype does not exist, and one haplotype that occurs exclusively in one single L. fagani individual holds an ancestral position equal to the most ancestral L. habessinicus haplotype; those findings are not incongruent with the hypothesis of an ancestral position of L. fagani and its mitochondrial capture by L. habessinicus/L. starcki. Nevertheless, we consider this hypothesis as less likely, because neither microsatellites nor the TF sequences suggest ancestral states compared to L. habessinicus: TF haplotypes of L. fagani are all younger than the most ancestral L. habessinicus haplotype, and multilocus microsatellite data indicate a very recent differentiation of L. fagani compared to a somewhat earlier differentiation of L. starcki from L. habessinicus (in accordance with ATP6 sequences). Given the ancestral bottleneck signal for L. starcki, however, the multilocus genetic divergence between L. starcki and L. habessinicus might have been somewhat inflated due to a possible drift effect in the former species. The TF sequences suggest a rather similar recent evolutionary divergence of either species from L. habessinicus. As 33 of all found microsatellite alleles are autapomorhic to L. fagani, the hypothesis of full mitochondrial and nuclear substitution of an ancestral L. fagani species by L. habessinicus would mean that 22.6% of all currently found alleles may in fact represent modern or ancestral L. habessinicus alleles, but do not any longer exist in modern L. habessinicus (or have at least not been found currently); that assumption seems, however, fairly unlikely. The allelic contribution of other African hare species (L. victoriae; L. saxatilis) in earlier phases of evolution to L. fagani is unknown; however, a late Pleistocene distribution of L. saxatilis across large parts of Africa and at least regional admixture with L. victoriae is conceivable, and such a scenario might have provided gene pool sources for L. fagani. This would explain the phenotypic similarity of L. fagani with L. victoriae (e.g., Angermann [1], Yalden et al. [60], Flux and Angermann [5]) that is considered closely related to or even conspecific with L. saxatilis. The hypothesis of a hybridogenic origin of L. fagani can only be explored by including many L. victoriae / L. saxatilis data into the analysis.

Contemporaneous introgressive hybridization in hares from Ethiopia
Already Halanych et al. [130] noticed that Lepus species that are known to hybridize in the wild, such as L. europaeus and L. timidus, may nevertheless show marked mtDNA divergence, indicative of "good species". He hypothesized that geographic, ecological, or behavioural isolation mechanisms may be driving speciation in the genus Lepus rather than genetic incompatibility, also in view of the very little chromosomal variation within the genus (e.g., Robinson et al. [131,132], Azzaroli-Puccetti et al. [133]). Later studies are not incongruent with this hypothesis (e.g., Alves et al. [8], Liu et al. [9], Wu et al. [27], Melo-Ferreira et al. [10,11], see also Thulin et al. [7], and Robinson and Matthee [127]).
Indeed, our microsatellite admixture results revealed two cases (1.9%) of contemporaneous introgressive hybridization in the hares from Ethiopia, albeit external phenotypes did not suggest hybrids. One nuclear genome of a L. habessinicus individual was introgressed by L. fagani and one nuclear genome of a L. starcki individual was introgressed by L. habessinicus. A few more hares might have been introgressed recently in their nuclear genomes, but inconsistent Bayesian admixture models did not confirm that. The absence of intermediate phenotypes conforms to the relatively rare occurrence of current nuclear introgressive hybridization. However, given that only a relatively small proportion (25%) of hares were collected from sympatric areas, introgressive hybridization might actually occur at somewhat higher species-wide levels, respectively. That is suggested by our coalescence theorybased gene flow results, both for the full and the reduced set of loci: for L. habessinicus and L. starcki estimated numbers of individuals migrating per generation are all above 1.0; theoretically, that level of migration is sufficient to compensate genetic drift or reduce interspecific divergence between species/populations under an island population model [134]. For L. fagani, however, we observed asymmetric gene exchange with L. habessinicus and L. starcki, with clearly less than one L. fagani individual per generation introgressing into the other two species, respectively. Nevertheless, we observed one L. habessinicus with its nuclear DNA having been introgressed in one of its recent ancestors by L. fagani. However, all these migration estimates may vary according to the geographical distribution of specimen samples. i.e., the proportion of hares collected sympatrically, pararapatrically, or allopatrically. Hence, a more developed discussion of geographic variation of nuclear introgression frequencies requires geographically more comprehensive data. Overall, ongoing nuclear DNA introgression seems to occur at low level, possibly similar to that found between L. granatensis and L. europaeus in northern Iberia Melo-Ferreira et al. [11] and L.europaeus and L. timidus [17,26].

Conclusions and systematic remarks
The present study represents the first comprehensive molecular phylogenetic investigation of hares (Lepus) from Ethiopia. It reveals a complex molecular evolutionary scenario: both mitochondrial and nuclear DNA exhibit shared ancestral polymorphism among hares from Ethiopia and other species from Africa and Europe. Moreover, nuclear TF sequences indicate shared ancestral polymorphism among African, Eurasian, and American hare species (see also Awadi et al. [135]). Both our microsatellite and mtDNA data indicate very close evolutionary relationships of all three species collected from Ethiopia, the Abyssinian hare, L. habessinicus, the Ethiopian hare, L. fagani, and the Ethiopian Highland hare, L. starcki, despite their very good phenotypic distinction, without intermediate forms that might suggest recent hybridization. Nevertheless, occasional hybridization does occur in hares from Ethiopia, possibly at somewhat higher frequencies in areas of sympatry, as suggested by our coalescence theory-based gene flow results. Unexpectedly, and in strong contradiction to traditional taxonomy based on phenotypic and morphological characters, all our molecular markers indicate recent descent of L. fagani from L. habessinicus. Almost complete mitochondrial substitution of (an ancestral form of) L. fagani by L. habessinicus is not inconsistent with our molecular data, but is deemed unlikely for nuclear DNA. Massive ancient introgression of ancestral forms of modern African savanna hares, L. victoriae, and/or scrub hares, L. saxatilis, into parts of an ancestral form of L. habessinicus with secondary separate evolution into modern L. fagani might represent an additional alternative evolutionary hypothesis for L. fagani, but remains to be tested by many geographically meaningful samples and additional molecular markers. Also unexpected is the evolutionary position of L. starcki close to L. habessinicus. It might have received ancient gene flow from currently allopatric L. timidus and/or L. europaeus, possibly during periods of range expansion of those species. The inclusion of many phenotypical L. capensis sensu lato from eastern Africa should help to understand the relationship between L. habessinicus and L. capensis sensu lato. Despite the complicated reticulate evolutionary scenario, ongoing gene exchange, their close evolutionary relationships, and pending data on evolutionary relationships particularly of cape hares, L. capensis sensu lato, and forms of the L. saxatilis/L. victoriae complex, our results are not incongruent with the currently acknowledged separate species status of L. fagani, L. habessinicus, and L. starcki from Ethiopia. This conclusion conforms to the current Lepus taxonomy (e.g., Alves and Hackländer [13]), and to a "relaxed biological species definition" (e.g., Coyne and Orr [136]) that allows for some gene flow between species, provided main evolutionary trajectories for species are not corrupted.  Table. Accession numbers, frequencies, and taxon names of the ATP6 sequences produced in this study; haplotype numbers per taxon are given in parentheses.

Supporting information
(DOC) S3 Table. Net mean between group (i.e., species) p distances of ATP6 sequences. (DOC) S4 Table. Accession numbers, frequencies, and taxon names of the phased TF sequences produced in this study; haplotype numbers per taxon are given in parentheses.