UTKU PERKTAS, GEORGE F. BARROWCLOUGH* and JEFF G. GROTH

Transcription

1 Biological Journal of the Linnean Society, 2011, 104, With 6 figures Phylogeography and species limits in the green woodpecker complex (Aves: Picidae): multiple Pleistocene refugia and range expansion across Europe and the Near East UTKU PERKTAS, GEORGE F. BARROWCLOUGH* and JEFF G. GROTH Department of Ornithology, American Museum of Natural History, Central Park West at 79th Street, New York, NY 10024, USA Received 15 April 2011; revised 3 June 2011; accepted for publication 3 June 2011bij_1750 The green woodpecker complex consists of the green woodpecker (Picus viridis), distributed from Western Europe to the Caucasus and Iran, and the related LeVaillant s woodpecker (P. vaillantii), distributed in north-western Africa from central Morocco to Tunisia. Much of the habitat of green woodpeckers in Central and Northern Europe was covered by ice, tundra, steppe or other unsuitable habitat during the Pleistocene; consequently, they must have come to occupy most of their current range during the past years. We used complete mitochondrial ND2 sequences from populations throughout the range to investigate the genetic structure and evolutionary history of this complex. Three well-differentiated clades, corresponding to three biogeographical regions, were recovered; 89% of the total genetic variance was distributed among these three regions. The populations in North Africa were sister to those of Europe and, within Europe, Iberia was sister to the rest of Europe and the Near East. This suggests that the post-glacial colonization of most of Europe occurred from a refuge east of Iberia, probably in Italy or the Balkans; there was no substantial divergence among these regions. In addition, a population sample from Iran was genetically distinct from those of Western Europe, indicating a history of genetic isolation and an additional Pleistocene refuge east of the well-known Balkan refugia and south of the Caucasus. Within Europe, northern populations were less genetically variable than southern ones, consistent with recent colonization. There was significant isolation-by-distance across Europe, indicating restricted gene flow; this was particularly apparent between western populations and those of the Caucasus and Iran. We recognize four species in the complex.. ADDITIONAL KEYWORDS: isolation by distance mitochondrial DNA ND2 Picus vaillantii Picus viridis population genetics. INTRODUCTION With the recognition that many temperate zone organisms could not have occupied their current distributional ranges during long periods of the Pleistocene, there has been interest in reconstructing the history and dynamics of the recolonization process of these regions (e.g. Pielou, 1991). For North America *Corresponding author. gfb@amnh.org Current address: Department of Biology (Zoology Section), Faculty of Science, Hacettepe University, 06800, Beytepe, Ankara, Turkey and Europe, southern refugia have been proposed that would have provided suitable climates and potential sources from which plant and animal populations could have expanded, advanced northwards and come to occupy their present ranges. If multiple refugia were suitable for a particular organism, then, during periods of glaciation, populations restricted to allopatric refugia might have differentiated in neutral genetic traits, phenotype or both. On expansion, if introgression were limited as a result of reproductive isolation, assortative mating or restricted vagility, the populations expanding from the allopatric refugia might be taxonomically recognized as different 710

2 PHYLOGEOGRAPHY OF GREEN WOODPECKERS 711 species or subspecies (e.g. Mengel, 1964), and regions in which multiple taxa from different refugia meet might be recognized as suture zones (e.g. Remington, 1968). In addition, whether or not taxonomically ranked as divergent, populations that had been restricted to different refugia for evolutionarily long periods of time would carry the genetic signatures of isolation simply as a result of genetic drift. The current patterns of these genetic signatures allow the reconstruction of the number and position of refugia and the historical process of post-glacial range expansion and occupation of the north temperate zones (e.g. Hewitt, 2000). In Europe, continental geography is particularly amenable to this reconstruction because, during the height of glaciation, Northern Europe would have been covered by ice, tundra and steppe. Southern Europe would have been broken into relatively discrete temperate refugia separated by seas and mountainous regions. In particular, relatively warm and hospitable parts of the Iberian Peninsula were separated from similarly suitable portions of the Italian Peninsula by the Pyrenees, the Alps and the western Mediterranean. Likewise, possible refugia in the Balkan Peninsula and Anatolia would have been isolated from an Italian refuge by the Alps and the Adriatic. An additional refuge may have existed east of the Balkans near the Caucasus and the Caspian Sea (Hewitt, 1999). Phylogeographical studies of a number of European organisms restricted to deciduous and mixed deciduous coniferous forest have recovered genetic signatures of (usually) three and (occasionally) four major glacial refugia in the aforementioned regions. However, the inferred patterns of expansion from these refugia and the subsequent reoccupation of Northern Europe have varied in idiosyncratic fashion from organism to organism (Taberlet et al., 1998). Hewitt (1999) referred to three common expansion patterns as the grasshopper, bear and hedgehog patterns, for cases tracing contemporary Northern European populations to predominantly Balkan, Balkan plus Iberian or Balkan plus both Iberian and Italian sources, respectively. In the canonical examples, hybrid zones were found where populations originating in each of the refugia met. Phylogenetic surveys of European birds whose ranges include all of the southern refugial regions have not been common. For example, some studies that would be relevant, e.g. Dendrocopos major and Sitta europaea (Zink, Drovetski & Rohwer, 2002, 2006), did not include samples from the presumptive southern refugial areas of Europe; similarly, a study including European Ficedula did not include multiple samples from glaciated and refugial areas (Outlaw & Voelker, 2006). Consequently, there are few studies of avian phylogeography in Europe that have sampled most or all of the relevant geography. In one such survey, of the tawny owl (Strix aluco), three strongly supported mitochondrial DNA (mtdna) control region clades were found (Brito, 2005). Two of these were principally restricted to Iberia and Italy, respectively, but the third clade occurred in both the Balkans and throughout Northern Europe. This suggested that most of Europe had been colonized by owls radiating out of an eastern Balkan refuge, whereas Iberian and Italian refugia had a much more local influence; this interpretation corresponds to Hewitt s (2000) grasshopper pattern. In addition, a basal clade of North African haplotypes was found to be sister to all of Europe. Other phylogeographical surveys of widespread European birds have not uncovered such a distinctive signature of multiple southern refugia. For example, Baker and colleagues investigated the genetic structure of two finches, the greenfinch (Carduelis chloris) and the chaffinch (Fringilla coelebs), throughout Western Europe (Merilä, Björklund & Baker, 1997; Griswold & Baker, 2002). In neither instance did they find clades of haplotypes with geographically disjunct occurrences. Similarly, a study of the great tit (Parus major) found that a single, common haplotype occurred across Iberia, Corsica and most of Northern Europe, with locally distributed, closely related haplotypes showing little geographical structure (Kvist et al., 2003). In the light of the differences in geographical structuring between the owl and the three passerines, we further investigated the generality of the canonical patterns of European phylogeography for temperate zone birds. In particular, we investigated a second widespread and sedentary nonpasserine, the green woodpecker (Picus viridis Linnaeus, 1758), together with its African congener, LeVaillant s woodpecker (P. vaillantii). The green woodpecker is distributed from Iberia, Italy and the Balkans, north to Britain and the southern reaches of Scandinavia, east to coastal Turkey, western Russia, the Caucasus and western Iran. Like the tawny owl, this woodpecker is nonmigratory; although it is present in Britain, it is not found on Mediterranean islands; this suggests that it possesses limited dispersal propensity. LeVaillant s woodpecker is restricted to a narrow range in the Atlas Mountains across North Africa, from central Morocco to Tunisia (Cramp, 1985). Both species are residents of deciduous and mixed woodlands (Cramp, 1985). Four subspecies of the green woodpecker are generally recognized today, whereas LeVaillant s woodpecker is considered to be monotypic (Dickinson, 2003); however, some authors (e.g. Vaurie, 1959, 1965; Short, 1982; del Hoyo, Elliott & Sargatal, 2002) have considered LeVaillant s woodpecker to be a subspecies of the green woodpecker.

3 712 U. PERKTAS ET AL. Pons et al. (2011) investigated the genetic variation in the green woodpecker complex using short segments of the mitochondrial cyt-b gene and a nuclear intron, for samples predominantly distributed in the western portion of the species range. Our sampling complements that study with a complete alternative mtdna gene and a much denser population sampling in the eastern and southern portions of the range. MATERIAL AND METHODS SAMPLES We sampled P. vaillantii and all four generally recognized subspecies of P. viridis. Older taxonomic revisions of the species complex recognized several additional subspecies of P. viridis in Europe and Asia (e.g. Peters, 1948; Vaurie, 1959) that are no longer treated as valid (Dickinson, 2003). We sampled all subspecific taxa recognized by Peters (1948) and subsequent authors, with the exception of P. v. bampurensis; this latter taxon was described from Baluchistan (south-east Iran) in 1911, but no specimens are known to exist and the subspecies is probably extinct (del Hoyo et al., 2002). We attempted to sample populations covering as much of the range of the green woodpecker as possible, including all putative southern refugia. Most of the DNA samples from P. viridis and P. vaillantii used in this study were extracted from toe pads from traditional specimens in the collections of the American Museum of Natural History (AMNH: 58), the Museum of Natural History of Vienna (NMW: 9) and the Yale University Peabody Museum of Natural History (YPM: 5). A few additional samples were available from recently collected tissue samples in the collections of AMNH (4), NMW (4) and YPM (2). We attempted to obtain five or more samples per locality, where localities were defined as geographical areas up to several hundred kilometres in diameter, within a single country, that were not fragmented by major barriers such as mountain ranges or large bodies of water. In addition, samples from Bosnia and Herzegovina, Macedonia and Serbia were combined into a sample labelled Western Balkans, and samples from Azerbaijan, Georgia and south-western Russia (Krasnodar and Ossetia) were combined into a sample labelled Caucasus. The approximate locations of the sampled localities are shown in Figure 1. All of the P. viridis and P. vaillantii ND2 sequences produced in this research have been deposited in GenBank (JN JN022018); these accessions include both catalogue numbers of specimen vouchers and the geographical origin of the samples. One additional sequence was available in GenBank (DQ479146: Benz, Robbins & Peterson, 2006); this sequence was added to our sample for the Caucasus. LABORATORY We sequenced the complete mitochondrial ND2 gene for all samples used in this study. Each fragment was sequenced in both directions. Because more than half the specimens were over 50 years old and some were over 100 years old, the quality of the DNA in the toe pad samples often was poor; therefore, a novel set of polymerase chain reaction (PCR) primers was designed, specifically for woodpeckers, to amplify small fragments of the ND2 gene. These new primer sequences are described in Figure 2. Toe pad samples were extracted and initial PCRs were prepared in a separate laboratory to prevent contamination from PCR products from other sources. Standard DNA extraction, taq-dna polymerase-based PCR amplification and sequencing, and fluorescent dye labelling were used on the samples; varying the annealing temperature and reaction time was sufficient for PCR troubleshooting. Products were run on an ABI 3730xl DNA analyser. We did not attempt to sequence nuclear genes for these woodpeckers because in excess of 85% of the samples were taken from old (in many cases, very old) skin specimens; nevertheless, as pointed out by Zink & Barrowclough (2008), mtdna is a leading indicator of differentiation vis-à-vis nuclear DNA and often is used to guide subsequent analysis with nuclear loci (Barrowclough & Zink, 2009). ANALYSIS We estimated standard measures of within- and among-population genetic variation. These included Nei s (1987) nucleotide diversity (p), mismatch distributions (Slatkin & Hudson, 1991) and Tajima s (1989) D statistic. Confidence intervals on p were estimated using a bootstrap procedure over individuals (1000 replicates). In some cases, we combined sequences across localities in estimating the mismatch distributions in order to obtain patterns for regions of particular interest: Morocco and Tunisia were combined to provide a North African distribution, Spain and Portugal to provide an Iberian distribution, and three Italian populations to provide an overall Italian pattern. The significance of the mismatch distributions was estimated by comparison of the observed distribution with that of a Poisson distribution, with an identical mean, using a Kolmogorov Smirnov onesample test. The significance of Tajima s D was estimated by comparison with a b distribution. We also calculated the percentage of base pairs that were variable within a population sample, as well as the

4 PHYLOGEOGRAPHY OF GREEN WOODPECKERS 713 Picus viridis Sweden England P i c u s v. v Austria i r i d i s Romania Portugal Morocco i s h a r p e Spain v a i l l a n t i i Veneto Tuscana Lazio Tunisia W Balkans k a Greece r e Thrace l i n i Anatolia Caucasus innominatus Iran Figure 1. Distribution and genetic structure of green woodpeckers in North Africa, Europe and the Near East. The ranges of the generally recognized subspecies are shown in various shades of green. Populations sampled in this study are indicated by pie diagrams and regional names; the area of each pie is proportional to the sample size. The colours of the pie sections indicate the proportion of sample corresponding to the haplotype groupings shown in Figure 4. frequency of haplotypes that were unique to each population sample (private haplotypes). The components of genetic variance distributed among populations were estimated using Holsinger & Mason-Gamer s (1996) hierarchical G st statistic; its significance was estimated using a bootstrap procedure, over individuals, with 1000 replicates. This statistic was also estimated for various hierarchies of localities in the green woodpecker complex, as well as for a matrix of genetic divergences between all population pairs. A matrix of approximate geographical distances between all pairs of localities was estimated from maps, independent of possible physical barriers to dispersal. Mantel s test was used to determine the statistical significance of the correlation between the genetic and geographical distance matrices ( random perturbations). We investigated patterns of genetic isolation-by-distance by plotting normalized genetic divergence between population pairs against the logarithm of geographical distance between the same pairs (Slatkin, 1993; Rousset, 1997); we also computed a semi-logarithmic regression for this same plot to evaluate the magnitude of the geographical dependence. We used PAUP* (Swofford, 2001) to infer minimum length trees for the ND2 sequences. We used eight ND2 sequences from closely related woodpeckers as outgroups; these included two new sequences (Picus canus canus, GenBank JN022020; P. erythropygius, GenBank JN022019) and six sequences previously archived in GenBank (P. canus hessei, EU327647; P. chlorolophus, DQ361293; P. flavinucha, DQ361289; P. mentalis, DQ479188; Campethera cailliautii, DQ479168; C. nivosa, DQ479167). These last six

5 714 U. PERKTAS ET AL. Pv11a Pv1 Pv3 Pv5 Pv7 Pv9 Pv11 Pv13 Pv15 Pv17 M ND2 W Pv2 Pv4 Pv6 Pv8 Pv10 Pv12 Pv14 Pv16 Pv18 Pv12a Pv1 GCTATCGGGCCCATACCCCGAA Pv2 CGGATGCGGCTGCTTGGACTAGGAA Pv3 GCAACCACTGAGCGATAGCTTGAAC Pv4 CCTAGTTTTATGGCGATTGCGGCTGT Pv5 GCATCCGCCTCCATCCTGTTCTC Pv6 GATGTGAGGAGGAGAATGGCGA Pv7 CTAGGCCTAGCACCATTCCACTT Pv8 CTTAAGTGGGAGATGGAAGAGAA Pv9 CCTCCTCACATCAAACTCCCTCAA Pv10 GATGGTTGAAGTAAGGAGGGTGTA Pv11 TCTTCCATCTCCCACTTAGGATG Pv12 CAATCCGGCTAGGGAGAGAAGGGT Pv11a CTCCTCCATCTCCCACTTAGGGTG Pv12a GTAATCCAGCAAGGGAGAGGAAGGT Pv13 CTTTACTTCAGCCATCTTCCTCTC Pv14 GGTAGAAGAATAGGCTTAGTAGGGA Pv15 CCTTTATCGGCTTCCTACCCAAATG Pv16 GATAGGGGGGCTAGGATGGCGAT Pv17 CCGCCTCACCTACTTTTCATCCAT Pv18 CTCTTGTTTAAGGCTTTGAAGGCCT Figure 2. Polymerase chain reaction (PCR) primers used to amplify and sequence mitochondrial DNA (mtdna) from the ND2 gene for green woodpeckers in this study. All primer sequences are described in the 5 to 3 direction. The approximate positions of each primer along the ND2 gene and its two flanking t-rnas are indicated. sequences were first reported by Benz et al. (2006), Fuchs et al. (2007) and Wright et al. (2008). The sequences for P. flavinucha and P. chlorolophus lacked one and four base pairs, respectively, at the beginning of the gene; these were coded as missing in phylogenetic analyses. Trees were found using a heuristic search procedure in which sequences were randomly added to the network in a stepwise fashion, followed by tree bisection reconnection (TBR) branch swapping; the random addition of sequences was repeated ten times. The stability of the phylogenetic relationships was investigated using a bootstrap procedure (1000 replicates). A minimum spanning network of haplotypes was inferred using the haplotype incompatibility method described by Bandelt, Forster & Röhl (1999). We used the program MDIV (Nielsen & Wakeley, 2001) to distinguish between historical isolation vs. continuous, but reduced, gene flow for pairs of populations that might have been restricted to alternative refugia during glacial periods. We estimated both the magnitude of gene flow between the population pairs, as well as the time to their most recent common ancestor (T mrca). MDIV uses a Markov chain Monte Carlo (MCMC) procedure to explore parameter space and can use either an infinite alleles model or an HKY model of base pair substitution. We assumed an HKY model of DNA evolution for all pairwise comparisons, as well as an upper bound on migration of 10 female migrants per generation (mn ef) and a maximum time to most recent ancestor of 30N ef generations. We used steps in the Markov chains, of which were discarded as burn-in. Analyses were repeated with several random numbers to ascertain the stability of the results. RESULTS SEQUENCES We obtained 1041 base pairs of mtdna sequence from 83 individuals representing 17 populations of green and LeVaillant s woodpeckers from large fractions of their ranges (Fig. 1). The sequences comprise the entire ND2 gene, including the start and stop codons. They could be aligned without indels, and 39 unique haplotypes were identified among the 83 sequences; thus, the ratio of haplotypes to individuals was nearly one-half (0.48). Within the green LeVaillant complex, variation occurred at 116 positions; there were 105 transitions and 12 transversions (at one position there were both). Sixty-four of the transitions were C/T substitutions and 41 were A/G. In one individual from Spain, we observed A and G peaks of approximately equal intensity at a single site in sequences obtained using both

6 PHYLOGEOGRAPHY OF GREEN WOODPECKERS 715 primers; we scored this site as an R and interpreted it as an incidence of rare heteroplasmy (i.e. one of bp examined). There were 22, 12 and 83 substitutions observed at first, second and third codon positions, respectively; consequently, the variation appeared to be consistent with expectations based on amino acid coding redundancy. If degraded DNA from the old skin specimens had resulted in random sequencing errors, we would have expected to observe a more nearly uniform distribution of changes across codon positions. Based on the mtdna genetic code, the observed substitutions within the green woodpecker complex (excluding outgroups) would have resulted in amino acid changes at 32 codons; this is 9.2% of the 347 codons in the 1041-bp ND2 gene. VARIATION WITHIN POPULATIONS The sample sizes, number of haplotypes observed, percentage of variable sites and nucleotide diversities (with 95% confidence intervals) for the 17 populations are presented in Table 1. Sample sizes varied from four to eight individuals for 14 of the populations; two populations were represented by single individuals. With the exception of the last two singletons, each population showed genetic variation. Variable sites within populations ranged from 0.1% to nearly 0.8%; however, this statistic is strongly influenced by sample size. Nucleotide diversity is less subject to sample size artefacts (Nei, 1987); it averaged across populations and varied by an order of magnitude, from to , among the samples. In particular, the more northerly samples, taken from localities that were probably covered by ice, tundra or steppe during Pleistocene glaciations, had nucleotide diversities ranging from to With the exception of Portugal (0.0006), the European samples taken from localities that probably possessed habitat suitable for green woodpeckers during the Pleistocene (glacial refugia) had higher nucleotide diversities ranging from to We used Tajima s (1989) D statistic to determine whether the variation observed within populations was consistent with neutral expectations or required selection, population bottlenecks or departures from the infinite alleles model for explanation. The results are reported in Table 1; none of the populations differed significantly from neutral expectations, albeit given the small sample sizes. We used mismatch distributions (Slatkin & Hudson, 1991) to investigate whether variation within populations was consistent with constant population size or suggested recent population growth. The distributions from different populations have different modes, largely consistent with refugial (Iberia, Italy and Balkan Peninsula plus Anatolia) vs. nonrefugial geographical origin, but none of the distributions was significantly different from that expected for a Poisson distribution with an identical mean (Fig. 3). Table 1. Mitochondrial DNA variation within populations of the green woodpecker complex, based on ND2 sequences Population Number of individuals Number of haplotypes % private haplotypes % variable sites Nucleotide diversity 95% confidence interval Tajima s D North Africa Morocco * Tunisia * Iberia Portugal * Spain * Eurasia Italy: Veneto na Italy: Tuscana * Italy: Lazio * Anatolia * England * Austria * Western Balkans * Romania * Sweden * Caucasus * Iran * *P > 0.05.

7 716 U. PERKTAS ET AL. frequency Austria Sweden North Africa Iberia Italy differences HAPLOTYPE RELATIONSHIPS Caucasus Figure 3. Sample mismatch distributions for several refugial and northern populations of green woodpecker, based on mitochondrial DNA (mtdna) sequences of the ND2 gene. A parsimony analysis of the 39 green and LeVaillant s woodpecker haplotypes, plus the eight outgroup sequences, yielded a set of 418 trees with a length of 606 steps. The consistency index for these trees, excluding uninformative sites, was In all the trees, the North African haplotypes from LeVaillant s woodpecker formed a monophyletic clade that was sister to all green woodpecker (sensu stricto) haplotypes. Within this latter clade, all Iberian green woodpecker haplotypes from Portugal and Spain formed a monophyletic clade that was sister to all European and Near East haplotypes. These three clades were separated by long branches and, in a bootstrap analysis, each of the three occurred in 100% of the replicates. In addition, the association of the Iberian with the European/Near East clade also occurred in 100% of the replicates. The outgroups rooted these woodpeckers between the African and European clades in each of the 418 shortest trees. These three clades are consistent with two vicariant events in green woodpeckers, with the division between North Africa and Europe preceding the split between Iberia and the rest of Europe. A minimum length network of the ingroup haplotypes (Fig. 4) indicates that the three clades identified in the parsimony analysis are separated from each other by long internal branches of greater than 20 substitutions. For the non-iberian European region, the network indicates a general pattern of a few common haplotypes at high frequency with less frequent daughter haplotypes one to two substitutions away; such a pattern is typical of growing populations. The pattern of the relationship shown (Fig. 4) among haplotypes within the three major clades was invariant across all 418 alternative trees; however, the position of the branch attaching the North African clade to the rest of the network moved among North African haplotypes in alternative trees. The most frequent position is shown in the figure. The relationships among the four haplotypes (brown circles) separated from other European haplotypes by four to five steps (right side of Fig. 4), and restricted to Anatolia, the Caucasus and Iran, are complicated by homoplasy; a summary of their alternative possible relationships has been illustrated using the method of Bandelt et al. (1999). They form a monophyletic clade in all 418 minimum length trees and have 49% bootstrap support; this suggests genetic isolation between Europe and the Near East and may represent the signature of an Iranian (or nearby) refuge. GEOGRAPHICAL VARIATION A summary of the distribution of related haplotypes and their frequencies is shown in Figure 1. The North African and Iberian clades are restricted to these respective regions. The remaining Eurasian haplotypes are heterogeneously distributed, with nearly 20 closely related haplotypes (yellow in Fig. 1) in Western Europe, a small clade of two haplotypes (orange in Fig. 1) restricted to the Caucasus, and four more divergent haplotypes (brown in Fig. 1) only found in Anatolia, the Caucasus and Iran. With a ratio of haplotypes to sequences of nearly 50%, it is not surprising that many localities had a high frequency of private haplotypes (Table 1). For example, no haplotypes were shared between any North African localities and, in Iberia, there were no widespread haplotypes at high frequency. However, in Central Europe, a single, common haplotype was widespread and represented 28% of the total sequences found among sampled Eurasian individuals; it occurred from Italy and England east to Romania and the Western Balkans. Three of four Iranian individuals possessed a single haplotype not found elsewhere, as did half of the eight individuals from the Caucasus; common, private haplotypes such as these suggest isolation or limited gene flow (Slatkin, 1985; Novembre & Slatkin, 2009). An analysis of overall genetic variation provides a quantitative perspective on the pattern of

8 PHYLOGEOGRAPHY OF GREEN WOODPECKERS Root Figure 4. Minimum-length network of ND2 haplotypes for green woodpeckers. Parsimony bootstrap proportions for the three major clades are shown below the branches; the numbers of inferred steps between clades are indicated on the branches. Major haplotype groupings are indicated by different colours; the area of each circle is proportional to the haplotype frequency. Black dots indicate the inferred position of unobserved haplotypes within clades. geographical variation (Table 2). For the 14 populations represented by four or more individuals, 93% of the total genetic variance was distributed among the samples; only 7% of the total genetic variance was distributed within populations. That is, most of the genetic variation was attributable to geography. We investigated the major sources of this variation using a series of hierarchical analyses. As indicated in Table 2, essentially all of the variation among the 14 populations exists at the level of variation among four regions: North Africa, Iberia, Europe and Iran. In addition to the dominant fraction of the total genetic variation distributed among regions, substantial portions of the residual, within-region, variation was still distributed among populations within regions (Table 2); this indicates that, within regions, there was additional isolation and reduced gene flow, although of a much smaller absolute magnitude than the primary, four-region effect. We investigated the significance and magnitude of this phenomenon by comparing genetic and geographical differences. Table 2. Hierarchical estimates of among-population components of genetic variance (G st) for the green woodpecker complex, based on ND2 sequences of mitochondrial DNA (mtdna). Bootstrap 95% confidence intervals and number of populations sampled within regions are shown in parentheses Hierarchy G st (95% confidence interval) Among 14 populations 0.93 ( ) Among three regions [North 0.89 ( ) Africa/Iberia/Eurasia] Among four regions [North 0.92 ( ) Africa/Iberia/Europe/Iran] Between Europe [except 0.42 ( ) Iberia] and Iran Among populations within regions North Africa (2) 0.22 ( ) Iberia (2) 0.19 ( ) Eurasia [except Iberia] (10) 0.50 ( ) Eurasia [except Iberia and 0.38 ( ) Iran] (9)

9 718 U. PERKTAS ET AL. Genetic distance [Gst/(1-Gst)] Europe Caucasus Iran Geographical distance [km] Figure 5. Isolation-by-distance among pairs of European (non-iberian) and Near East populations of green woodpecker. Semi-logarithmic least-square regression lines between normalized genetic distances and straight-line geographical distances are shown for all localities (all dots; dotted line), within Europe and Europe to the Caucasus (black plus grey dots; broken line) and within Europe only (black dots; full line). ISOLATION-BY-DISTANCE The among-population genetic divergences between non-iberian localities in Western Europe, as measured by G st, were positively and significantly correlated with the geographical distances between the same localities, indicating significant isolation-bydistance. The correlation coefficient between the respective genetic and geographical distance matrices was 0.53; Mantel s test of the correlation was significant (P = 0.008) for all these localities, and nearly so for the same localities, except Iran (P = 0.068). We have illustrated the patterns of isolation-by-distance, with and without Iran and the Caucasus, in Figure 5. The geographical distances between the European and Iranian samples were relatively large and their genetic variance components all exceeded 0.5; hence, the regression for all locality pairs including Iran was steep. For the same locality pairs except Iran, the regression was still positive, indicating isolation-bydistance, but the slope was not as steep. For locality pairs excepting both Iran and the Caucasus, the regression line had a slope similar to that of the regression line without only Iran. Bayesian Posterior Probability Density Bayesian Posterior Probability Density Iberia-Italy Italy-Turkey Turkey-Iran Gene Flow (mn ef ) Turkey-Caucasus Italy-Turkey Turkey-Caucasus Turkey-Iran Iberia-Italy Time to most recent common ancestor (2Nef) Figure 6. Coalescent-based estimates of probability density for gene flow (A, number of female migrants per generation) and divergence times (B, mitochondrial coalescent times) between some southern populations of green woodpecker. GENE FLOW AND DIVERGENCE TIMES Our phylogenetic results indicated that there was moderate homoplasy in the ND2 sequence data (consistency index, 0.54); consequently, we used an HKY model of substitution (rather than an IA model) in MDIV. In addition, the nearly ten-fold ratio of transitions to transversions, approximately equal ratio of purine to pyrimidine transitions and unequal base frequencies all suggested that an HKY model was appropriate for MDIV analysis. We used our estimates of gene flow and time to most recent common ancestor, based on the MDIV results, to investigate the status of several southern population samples as potential glacial refugia (Fig. 6). In the case of Iberia vs. Italy, the results strongly support the argument for Iberia as a long-isolated refuge separate from the rest of Europe. All of the probability mass for gene A B

10 PHYLOGEOGRAPHY OF GREEN WOODPECKERS 719 flow between the two localities occurs at values of gene flow much less than one female per generation (Fig. 6A). This level of gene flow is too low to prevent genetic divergence (Crow & Kimura, 1970). The modal estimate of T mrca was approximately 7.08 coalescent times (Fig. 6B), where the expected time to reciprocal monophyly is two coalescent times. Consequently, the Iberian population of green woodpeckers has been isolated from those of the rest of Europe long enough for reciprocal monophyly, which, in fact, was found; in addition, there is little evidence for any significant gene flow between Iberia and the rest of Europe even now, more than years since the Pyrenees have been ice free. In contrast with the Iberian comparisons, the estimates of T mrca were less than 2N e for comparisons of Italy with Turkey plus Greece and Turkey plus Greece with the Caucasus; only the Turkey plus Greece with Iran comparison yielded an estimate greater than two (2.37), suggesting the existence of a separate refuge that gave rise to the Iranian population. The estimate of gene flow between Turkey plus Greece and the Caucasus had a significant probability density of between one and four females per generation (Fig. 6A); this suggests that gene flow has occurred between the two populations, but at a level sufficiently low such that some genetic divergence was to be expected. DISCUSSION GEOGRAPHICAL DIFFERENTIATION AND REFUGIA The pattern of haplotype relationships (Fig. 4) indicates that North African populations of green woodpecker (LeVaillant s woodpecker) have been isolated from European populations for a very long time; a subsequent event resulted in a split between Iberian populations and those of the rest of Europe. Both of these events were sufficiently distant in the past that the populations have become reciprocally monophyletic, i.e. have coalesced. The fraction of the total genetic variation distributed among these regions was of the order of ten times the variation within the regions (Table 2). Pons et al. (2011) obtained a similar result for some mitochondrial and nuclear loci using two birds from Morocco. The pattern of differentiation was less clear within Europe and the Near East. In contrast with the pattern in many organisms (Hewitt, 2000), there was no clear signature of haplotype differentiation between Italy and the Balkans/Anatolia, which are widely accepted to represent two distinct refugia. Instead, most of non-iberian Europe was occupied by a group of a few common haplotypes, plus their daughters, that varied in proportion, from place to place, in an idiosyncratic fashion; Pons et al. (2011) also obtained this pattern. It appears likely that Italy and the Balkan region shared a common, or closely related, pool of haplotypes during the most recent glacial maximum. Nevertheless, the Alps must be somewhat of a barrier to dispersal as the estimate of gene flow between Italy and Turkey has most of its probability mass at less than two females per generation (Fig. 6A). An alternative to a single, common Italian/Balkan refuge would be the Pleistocene extinction of woodpeckers in one of these two regions (probably Italy), followed by a post-pleistocene recolonization from the other. However, the mismatch distribution for Italy, and the corresponding estimate of nucleotide diversity, are unlike those of northern population samples from areas that clearly were recently recolonized. Further east, our sample from the Caucasus region consisted of eight individuals, seven of which possessed haplotypes found nowhere else. Five of these seven formed a small clade of private haplotypes two steps away from the large European clade. The high frequency of a private clade of haplotypes suggests restricted gene flow between Europe and the Caucasus (Fig. 6A), or a weakly divergent refuge; the latter possibility is contradicted by the vanishingly small estimate of time to most recent common ancestor between Turkey and the Caucasus (Fig. 6B). The isolation-by-distance plot (Fig. 5) indicates that the Caucasus to Europe comparisons fall on the graph at approximately the position at which one might have expected them, given the overall pattern of genetic vs. geographical regression. The Caucasus population was originally described as a weakly divergent taxon (P. v. saundersi) that is no longer widely recognized (e.g. Vaurie, 1959). The sample was nearly diagnosable on the basis of mtdna, but not with morphology. Thus, the evidence for a separate glacial refuge in the Caucasus is weak; it is more likely that that population has been subject to considerable isolation-bydistance and restricted gene flow. Pons et al. (2011) obtained two samples from just north of the Caucasus range, they possessed haplotypes similar to those of Northern Europe. Our sampling included individuals from localities further east and south, and suggests substantially more differentiation than observed by Pons et al. (2011). A similar situation arose in our sample from the Zagros Mountains of Iran. The small sample from this range formed a clade in some, but not all, of the minimum length trees. All the haplotypes (none shared with any other locality) were four or more steps from those of Europe, but, in some trees, were made paraphyletic by haplotypes from Anatolia and the Caucasus. The plumage of this population is slightly divergent from that of other green woodpecker populations, and the population is generally recognized taxonomically as P. v. innominatus. The

11 720 U. PERKTAS ET AL. present range is completely allopatric with respect to the rest of the complex. In an isolation-by-distance plot, the comparisons of this population with those of Europe were more genetically differentiated than one would have anticipated on the basis of their geographical separation (Fig. 5); this suggests a complete absence of gene flow for a substantial period of time, rather than isolation-by-distance. Nucleotide diversity in this population was low in comparison with other locations in Southern Europe, suggesting a relatively small effective population size, and G st between Iran and Europe (excluding Iberia) was The latter statistic suggests that there has been no gene flow between Iran and populations to the north and west for a long period of time; most of the probability mass for gene flow between Iran and Turkey is consistent with less than one female exchanged per generation (Fig. 6A). Therefore, it is likely that the present population of the Zagros Mountains was isolated in a glacial refuge somewhere in that region during the most recent glacial episodes; however, the resulting isolation does not extend as far back in time as that separating Iberia from the rest of Europe, or of Europe from North Africa. The separation between Iberia and Europe was about seven expected coalescent times; that between Turkey and Iran was only slightly more than two (Fig. 6B). Pons et al. (2011) obtained no Iranian samples in their survey. HISTORICAL EXPANSION All of the samples obtained from regions of Northern Europe that would not have had a reasonable green woodpecker habitat during glacial epochs had haplotypes that were closely related to those found in Italy and the Balkan/Anatolian region. No Iberian haplotypes were found east of Spain, given our sampling. Furthermore, nucleotide diversities in the Northern European samples were substantially lower than those of Iberia, Italy and Anatolia, and mismatch distributions from the northern samples were characterized by modal values closer to the ordinate than those from Iberia, Italy and North Africa (Fig. 3). All of these observations, and those of Pons et al. (2011), are consistent with the colonization of Northern Europe from a southern refuge in either Italy or the Balkan/Anatolian region, or a combination of the two, after the retreat of ice and the occupation of Northern Europe by temperate vegetation. Iberian populations did not make it far beyond the Pyrenees and representatives of the Iranian population did not get as far west as Europe. This pattern of expansion is similar to that observed for the tawny owl, but with ambiguity with respect to the role of Italian vs. Balkan refugia. It is strikingly divergent from the patterns reported for three passerines: the great tit, chaffinch and greenfinch. In the case of the chaffinch, Fringilla coelebs, Griswold & Baker (2002) did not find clades of haplotypes corresponding to Southern European refugia, or even to North Africa, for that matter. Rather, three common haplotypes were distributed nearly everywhere North Africa to Scandinavia and Portugal to Greece. Only by estimating the time to most recent common ancestor of each sampled population were the authors able to distinguish between old and possibly refugial populations in Iberia, Corsica, Greece and North Africa, and newer, principally northern, populations. The authors concluded that Northern European localities had been colonized from a combination of North African, Iberian and Grecian refugia. Merilä et al. (1997) examined mitochondrial variation in the greenfinch (Carduelis chloris) using samples from North Africa, Iberia, Sicily and several localities in Northern Europe. All of the haplotypes observed were closely related; one was present at high frequencies everywhere and at frequencies greater than 50% throughout Northern and Eastern Europe. Three other haplotypes occurred locally at moderate frequency in Sicily, Iberia and North Africa, but these were only one step removed from the most common haplotype. In the case of the great tit (Parus major), a common haplotype occurred throughout Western Europe; additional haplotypes, only one to two changes from the common haplotype, were more locally distributed. Consequently, a distinctive signal of Pleistocene refuge structure in Southern Europe, with subsequent expansion, was not present in any of these three passerines, unlike the cases of the tawny owl and green woodpecker. Chaffinches and greenfinches are common, partially migratory and quite vagile (e.g. present on all the major Mediterranean islands, the Canaries and the Azores). Thus, in these two species, it is plausible that there was continuing gene flow among populations whilst in separate refugia or that post-glacial gene flow has been so extensive as to obliterate whatever glacial signatures were present. The great tit, although largely nonmigratory, occasionally shows eruptive southward population movements (e.g. Cramp, 1993); these might have obscured previous geographical structure. The striking differences between all three passerines and the tawny owl and green woodpecker may stem from the increased admixture among populations as a result of the dispersal and gene flow associated with migratory or eruptive behaviour. The owl and woodpecker demonstrate geographical structure more typical of the canonical cases, for sedentary organisms, summarized by Hewitt (2000).

12 PHYLOGEOGRAPHY OF GREEN WOODPECKERS 721 SPECIES LIMITS AND TAXONOMY We found extensive genetic structure among populations in the green woodpecker complex; this suggests a long history of effective reproductive isolation among regions. The North African populations have been recognized as a separate species, LeVaillant s woodpecker (P. vaillantii), by several authors, largely based on differences between the male moustache between the African and Eurasian birds. Our ND2 results indicated that the clade of haplotypes found in the African populations was approximately 6.7% divergent (100% bootstrap) from the Eurasian clades of haplotypes; the most commonly used molecular clock calibration would indicate a divergence time of approximately 3.3 million years for this divergence (Lovette, 2004). Thus, LeVaillant s woodpecker appears to have been geographically isolated from the European populations for a long time and is completely diagnosable on the basis of both plumage and mtdna data; it meets the criteria for species recognition from the points of view of both the biological and phylogenetic species concepts (Zink & McKitrick, 1995; Johnson, Remsen & Cicero, 1999). The green woodpecker populations on the Iberian peninsula differ from those of the rest of Eurasia on the basis of their facial and moustachial patterns; they have long been recognized as the subspecies P. v. sharpei. Our results indicate that the Iberian haplotypes formed a clade that was reciprocally monophyletic with respect to the clade of haplotypes from the rest of Eurasia. These clades had 100% bootstrap support and were 2.4% divergent from each other. Thus, the Iberian population is 100% diagnosable and has been isolated for perhaps a million years; it is recognizable as a phylogenetic species. The salient plumage difference between sharpei and viridis, the colour of the malar stripe, is a sexually dimorphic character and hence might be important in mate choice. Vaurie (1965) stated that the colour differences between viridis and sharpei were not bridged by intergradation, which suggests that the latter is probably isolated. Consequently, the status of the two taxa in terms of the biological species concept is suggestive of species. In a previous molecular study, Pons et al. (2011) found mtdna divergences nearly identical to those reported here between the North African taxon and the European taxa, and between the Iberian taxon and the rest of Europe. They used the cyt-b gene, whereas we used the ND2 gene; thus, their results are consistent with ours. In addition, Pons et al. (2011) found the same pattern of reciprocal monophyly between these three taxa for several nuclear introns. In addition, in a third study, based on very small samples, the Iberian P. v. sharpei and European P. viridis were distinct (1.0 Bayesian posteriors) for several nuclear introns (Fuchs et al., 2008). Consequently, the considerable divergences reported here have been corroborated by nuclear results. The limited sample of innominatus available to us from the Zagros Mountains in western Iran was composed of two haplotypes that were not observed elsewhere; they formed a clade, which left viridis paraphyletic, in some but not all of the shortest trees. Thus, innominatus is 100% diagnosable, but not reciprocally monophyletic with respect to viridis. Its current geographical distribution (Fig. 1) is completely allopatric with respect to other taxa in the complex; again, given the caveats associated with sample size, it comprises a phylogenetic species. The plumage differences between innominatus and viridis comprise subtle variations in colour, not qualitative differences in pattern (Vaurie, 1965); probably few authors would recognize innominatus as a biological species. In summary, we recognize four species in the green woodpecker complex, P. viridis Linnaeus (1758), P. sharpei (Saunders, 1872), P. vaillantii (Malherbe, 1847) and P. innominatus (Sarudny & Loudon, 1905), because these populations are all genetically divergent and diagnosable. Importantly, they are avatars of a historical pattern of geographical isolation that should generalize across many organisms and should represent distinct entities in biogeographical and phylogenetic investigations. Our data indicate that closely related haplotypes were found throughout Southern (except Iberia) and Northern Europe. There was no genetic differentiation that corresponded to a division between the subspecies P. v. viridis and P. v. karelini. Vaurie (1965) found the morphological differences between these, and across non-iberian Europe, to be slight. Most of the range of nominate viridis was covered by ice and tundra until years ago. The estimate of the divergence time between the Turkish and Caucasus populations was less than that between the Turkish and Italian populations (Fig. 6B). Consequently, we recognize no intraspecific taxa in Western Europe because there are no differentiated historical entities. ACKNOWLEDGEMENTS We are grateful to the curators and collection managers at the American Museum of Natural History (Paul Sweet), the Peabody Museum of Natural History at Yale University (Kristof Zyskowski) and the Naturhistorisches Museum Wien (Anita Gamauf) for assistance with tissue samples from critical specimens used in this research. Bob Zink and three anonymous reviewers offered useful suggestions on an earlier version of the manuscript. Post-doctoral research funding for