Phylogeography of the endangered darkling beetle species of Pimelia endemic to Gran Canaria (Canary Islands)

Transcription

1 Molecular Ecology (2003) 12, doi: /j X x Phylogeography of the endangered darkling beetle species Blackwell August 12 8Original PHYLOGEOGRAPHY H. G. CONTRERAS-DÍAZ 2003 Article Publishing Ltd. OF GRAN ET AL. CANARIAN PIMELIA of Pimelia endemic to Gran Canaria (Canary Islands) HERMANS G. CONTRERAS-DÍAZ,* OSCAR MOYA,* PEDRO OROMÍ and CARLOS JUAN* *Departament de Biologia, Universitat de les Illes Balears, Palma de Mallorca, Spain, Departamento de Biología Animal, Facultad de Biología, Universidad de La Laguna, La Laguna, Tenerife, Spain Abstract Phylogenetic and geographical nested clade analysis (NCA) methods were applied to mitochondrial DNA sequences of Pimelia darkling beetles (Coleoptera, Tenebrionidae) endemic to Gran Canaria, an island in the Canary archipelago. The three species P. granulicollis, P. estevezi and P. sparsa occur on the island, the latter with three recognized subspecies. Another species, P. fernandezlopezi (endemic to the island of La Gomera) is a close relative of P. granulicollis based on partial Cytochrome Oxidase I mtdna sequences obtained in a previous study. Some of these beetles are endangered, so phylogeographical structure within species and populations can help to define conservation priorities. A total of about 700 bp of Cytochrome Oxidase II were examined in 18 populations and up to 75 individuals excluding outgroups. Among them, 22 haplotypes were exclusive to P. granulicollis and P. estevezi and 31 were from P. sparsa. Phylogenetic analysis points to the paraphyly of Gran Canarian Pimelia, as the La Gomera P. fernandezlopezi haplotypes are included in them, and reciprocal monophyly of two species groups: one constituted by P. granulicollis, P. estevezi and P. fernandezlopezi (subgenus Aphanaspis), and the other by P. sparsa sensu lato. The two species groups show a remarkably high mtdna divergence. Within P. sparsa, different analyses all reveal a common result, i.e. conflict between current subspecific taxonomic designations and evolutionary units, while P. estevezi and P. fernandezlopezi are very close to P. granulicollis measured at the mtdna level. Geographical NCA identifies several cases of nonrandom associations between haplotypes and geography that may be caused by allopatric fragmentation of populations with some cases of restriction of gene flow or range expansion. Analyses of molecular variance and geographical NCA allow definition of evolutionary units for conservation purposes in both species-groups and suggest scenarios in which vicariance caused by geological history of the island may have shaped the pattern of the mitochondrial genetic diversity of these beetles. Keywords: Canary Islands, Coleoptera, phylogeography, conservation genetics, diversification, mitochondrial DNA Received 9 December 2002; revision received 12 March 2003; accepted 9 April 2003 Introduction The Canary Islands present a high level of endemic taxa in their flora and invertebrate fauna (between 50 and 80% depending on particular groups) despite their proximity to the African continent (Oromí et al. 1991; Juan et al. 2000). The archipelago forms an island chain of seven main islands in a band about 500 km long and 200 km wide Correspondence: Carlos Juan. Fax: ; cjuan@uib.es northeast to southwest, 110 km offshore from Cape Juby in northwest Africa (see inset in Fig. 1). The emerged parts of the islands were formed by volcanism with maximum eruptive activity during the Miocene with the oldest parts of each block having emerged in the range of 21 1 million years ago (Ma) in a temporal cline from northeast to southwest, respectively (Anguita & Hernán 1975; Ancochea et al. 1990). Speciation and biodiversity in the archipelago has clearly been promoted by habitat diversity due to a combination of factors. These include the independent origin of each island, the high altitude in five of the seven 2003 Blackwell Publishing Ltd

2 2132 H. G. C O N T R E R A S - D Í A Z E T A L. Fig. 1 Map of Gran Canaria indicating geographical locations of the samples. P. granulicollis; P. estevezi; P. sparsa sparsa; P. sparsa serrimargo; P. sparsa albohumeralis. The codes for each population correspond to the ones listed in Table 1, in which sample numbers are specified. Localities unsuccessfully surveyed in the present study for P. sparsa sparsa and P. sparsa serrimargo, but with samples in entomological collections are indicated with the corresponding symbols without codes. On La Gomera (inset of the Canary Islands) the locality where P. fernandezlopezi occurs is indicated with an arrow. islands, and marked climatic differences between north and south slopes of the volcanoes (Oromí et al. 1991). Data on the geological history of Gran Canaria (the third island in size) based on K-Ar age determinations indicate that it emerged at about 15 Ma (Hoernle et al. 1991). Two eruptive periods, one beginning at about 14 Ma, and another between the last Ma, the latter forming the Roque Nublo agglomerate complex, were very explosive and produced extensive aerial emissions (Pérez-Torrado et al. 1995). It has been suggested that these events caused massive extinctions on the island and could explain the lower DNA sequence divergence of species occurring in the mountainous areas of Gran Canaria compared to those near the coast (Emerson 2002). This would also explain lower species richness in this island compared to others of similar size and habitat diversity such as Tenerife, or even smaller like La Gomera (Brown & Pestano 1998; Emerson et al. 1999, 2000a,b). During the Quaternary, volcanism has mostly been based in the northeast half of Gran Canaria, with lava flows and cones being active as recently as 3000 bp (Funck et al. 1996). These facts have been invoked to explain Gran Canaria s geographical patterns based on mtdna genealogies in beetles (Rees et al. 2001) and skinks (Pestano & Brown 1999).

3 PHYLOGEOGRAPHY OF GRAN CANARIAN PIMELIA 2133 Pimelia (Coleoptera, Tenebrionidae) is a speciose genus with a circum-mediterranean distribution constituted by saprophagous flightless species with presumably low individual dispersal. There are 14 recognized taxa (species and subspecies) in the Canaries, each one endemic to a single island, except P. lutaria which is exclusive to both Lanzarote and Fuerteventura. Five Pimelia taxa occur in Gran Canaria (Oromí 1979, 1982): P. granulicollis and P. estevezi are of similar size and morphological traits, the former is present in most of the coastal dunes of the island, while the latter has a very narrow distribution in an isolated palaeo-dune to the west (see Fig. 1). There are also three subspecies of P. sparsa showing different morphologies, each found in different habitats: high altitude (P. sparsa sparsa), coastal dunes in two isolated localities with halo-psammophilous habitat (P. sparsa albohumeralis) and lowland dry open areas (P. sparsa serrimargo). The main morphological differences between these three subspecies are integument sculpture and pilosity, and in particular body size: P. s. sparsa is bigger with less marked sculpture, while P. s. serrimargo and P. s. albohumeralis are much smaller (the smallest among all Canary Pimelia) and show more conspicuous granules, denticles and carinae. The latter two subspecies have the same body size and other external differences are variable. Populations of P. granulicollis and the two lowland subspecies of P. sparsa are sympatric in some locations, the only sympatry case among Canary Pimelia, coinciding with a remarkably smaller size in these two subspecies of P. sparsa (the allopatric P. sparsa sparsa is much closer to granulicollis in size). Surprisingly, P. granulicollis but not the isolated population of P. estevezi, has been placed in the Catálogo Nacional de Especies Amenazadas (Spanish List of Endangered Species). However, the local authorities include these two species in the Catálogo de Especies Amenazadas de Canarias (List of Endangered Species in the Canaries) in the most threatened species group (endangered species) and P. sparsa albohumeralis in the second most threatened group (i.e. species sensitive to habitat alteration). The genetic diversity and origin of the species P. fernandezlopezi is also of particular interest, because of its conservation importance, as represented by a unique, highly isolated population occurring on fossil dunes in La Gomera; it is also included in the regional protection list as sensitive to habitat alteration and was previously shown to be genetically close to the Gran Canarian Pimelia (Juan et al. 1995). P. fernandezlopezi, P. granulicollis and P. estevezi in fact, share enough distinctive morphological features to have been proposed for a different subgenus, Aphanaspis (see Español 1971; Machado 1979; Oromí 1990). The objective of the present study is to use a representative sampling of all Pimelia populations present on Gran Canaria and apply standard phylogenetic and Nested Cladistic Analysis (NCA) to the mitochondrial haplotype distributions obtained, in order to deduce population structure and test hypotheses about the population history of each species. NCA constructs haplotype networks into a series of hierarchical groups using statistical parsimony, and performs permutational contingency tests and geographical distance analyses based on the nested design, to test for alternative hypotheses of nonrandom spatial distribution of haplotypes (Templeton et al. 1995; Templeton 1998). Comparing the genetic diversity among populations and clarifying the evolutionary units (Avise 2000; Crandall et al. 2000) in the Gran Canarian Pimelia populations may help to make decisions regarding conservation management and recovery strategies for these beetles. Materials and methods Sample collection Pimelia specimens (P. granulicollis, P. estevezi, P. s. sparsa, P. s. albohumeralis and P. s. serrrimargo) were collected on Gran Canaria during 2001 and The related taxon P. fernandezlopezi was collected in 2001 on La Gomera in Puntallana Nature Reserve, the only known site where this taxon occurs. Figure 1 shows the geographical range of Pimelia taxa based on previous collections plus the locations and codes of the Gran Canaria samples obtained in the present study. Table 1 gives details about collection localities, their sample sizes, UTM co-ordinates and habitat. Samples were stored in absolute ethanol until the DNA extractions were performed. Sequences of P. lutaria endemic to Fuerteventura and Lanzarote, as well as P. canariensis and P. radula endemic to Tenerife were included in the analyses to test the monophyly of Gran Canarian taxa. DNA extractions, PCR amplification and sequencing A leg or head from each specimen was used for DNA purification with a standard protocol, with tissue homogenization in a Tris-EDTA-phosphate buffer, protein precipitation in a saline solution, phenol:chloroform extractions and ethanol DNA precipitation. Pellets were resuspended in 50 µl of Tris-EDTA buffer and 1 µl of the dilution was used for PCR amplification of the fragment of Cytochrome Oxidase II (COII) using 1 unit of Taq DNA polymerase (Ecogen). The primers used were a modified TL-J-3037 (5 - TAATATGGCAGATTAGTGCATTGGA-3 ) and a longer TK-N-3785 (5 -GAGACCATTACTTGCTTTCAGTCATCT- 3 ; Simon et al. 1994; Gómez-Zurita et al. 2000). PCR conditions were as follows: 4 min at 95 C followed by 35 denaturation cycles at 95 C for 30 s, annealing at 50 C for 1 min, and extension at 72 C for 2 min, with a final single extra extension step at 72 C for 10 min. PCR products were checked in a 1% agarose gel and the products of the expected length were precipitated with ammonium

4 2134 H. G. CONTRERAS-DÍAZ ET AL. Table 1 Summary of the Pimelia sampling in Gran Canaria and La Gomera. Taxa, localities and their codes, habitat characteristics, UTM co-ordinates, haplotype distribution (number of individuals with the same haplotype in parenthesis) and number of individuals collected are given for each sample. Habitat codes are as follows: A high altitude; B coastal dunes; C lowland dry areas; D basaltic sand; E fossil dunes Taxa Locality Code Habitat UTM Haplotypes N P. sparsa sparsa Cuevas Blancas CB A b1(2), b2, b3, b4 5 P. sparsa albohumeralis Arinaga AI-1 B b5, b6, b7, b8, b9 5 Dunas de Maspalomas DU-1 B b10(2), b11(2) 4 P. sparsa serrimargo Aldea de S. Nicolás AL C b12(2), b13, b14, b15(2), b16 7 El Confital CO C b17(3), b18, b19 5 Lomo Gualapa GU C b31(2), b21, b22 4 Sardina del Norte SA C b14, b23 2 Tufia TU-1 C b24, b8, b25, b26, b27, b28, b29, b30 8 Bco. Tio Vicente VI C b20, b31(4) 5 P. granulicollis Arinaga AI-2 B a1, a2 2 Arguineguín AG D a3(2), a4, a5 4 Playa del Burrero BR B a6, a7 2 Playa de las Burras BU B a8(2), a9 3 Charco de la Aldea CH D a10(2) 2 Dunas de Maspalomas DU-2 B a11, a12, a13, 3 Guanarteme GT B a13(2), a14, a15 5 Tufia TU-2 B a1, a3, a16, a17, a18, a19, a20 7 P. estevezi Punta de las Arenas AR E e1(2), e2 3 P. fernandezlopezi Puntallana E f1(2), f2 3 (LA GOMERA) acetate 5 m and isopropanol. The forward strand was cycle-sequenced using an ABI Prism DYE Terminator Cycle Sequencing Reaction Kit sequenced in an ABI 377 automated sequencer, all the low frequency haplotypes were confirmed by sequencing in the opposite direction. A fragment of about 700 bp was sequenced in all samples. Nucleotide sequences were deposited at EMBL under accesion numbers AJ AJ and the alignment of all the sequences used in this study is found under accession number ALIGN_ Phylogenetic analyses The mitochondrial haplotypes obtained were used for a maximum parsimony (MP) analysis using paup* version 4.0 (Swofford 2002). Heuristic searches used 100 replications of random addition of taxa and TBR branch swapping on all trees, using the steepest descent option and equally weighting all nucleotides. Support for monophyly of particular clades was assessed by Bremer support values using treerot version 2b to generate constraint trees (Sorenson 1999) and bootstrap values with 1000 replicates (full heuristic search) using paup*. modeltest version 3.06 (Posada & Crandall 1998) was used to select the substitution model that best describes the data. Given the computer time involved, no Maximum Likelihood (ML) analysis was carried out and, instead, a Neighbour-Joining (NJ) analysis of the ML distances obtained using the parameter estimates derived from modeltest was performed. Bootstrap values for this analysis were obtained from 1000 replications. Population genetic variability, population subdivision and Nested Clade Analyses Nucleotide diversitiy, (average number of nucleotide differences per site between two sequences), was calculated using dnasp 3.53 (Rozas & Rozas 1999). Analysis of Molecular Variance (amova) was used to partition molecular variance into different hierarchical levels. We used arlequin version 2000 for amova analyses (Schneider et al. 2000) in which we tested for differences between sampling sites (nested within regions) and differences between regions. The significance of the variance components was computed using a nonparametric permutation test (Excoffier et al. 1992). Populations were grouped either according to geography, based on the clades deduced by phylogeny, or on taxonomic considerations. The grouping that maximized the among regions variation was assumed to be the most plausible geographical subdivision (Paulo et al. 2002). For nested clade analysis, the network based on the most parsimonious connections of haplotypes was obtained using tcs version 1.06 (Clement et al. 2000). Nested design was based on the rules proposed by Templeton et al. (1995). Finally, nested geographical distance analysis was carried out with the geodis software version 2.0 (Posada & Crandall 2000). This method uses geographical distances between the sampled locations

5 PHYLOGEOGRAPHY OF GRAN CANARIAN PIMELIA 2135 and estimates two measurements: Dc and Dn. Dc or clade distance is the mean geographical distance among individuals having haplotypes within a particular nesting group. Dn or nested clade distance is the mean distance among individuals having haplotypes within a particular nesting group and the geographical centre of the higher nesting group. The distinction between tip (with only one connection to the remaining network) and interior (with two or more connections) haplotype groups in the context of coalescent theory allows testing of the hypothesis of random geographical distribution by permutational tests. Using the updated inference key of Templeton et al. (1995) we can deduce which factor(s) caused spatial and/or temporal significant association between haplotypes. In this way, historic or nonrecurrent events such as range expansion, long distance colonization or population fragmentation can be differentiated from contemporaneous or recurrent ones, i.e. restriction of gene flow (Gómez- Zurita 2002). The method uses geographical distances between sample localities estimated from their geographical co-ordinates. When constructing the geographical distance matrix for P. granulicollis, we used minimum separations between localities calculated along the coast due to the mostly coastal distribution and habitat requirements (calcareous dunes) of this and allied taxa; since the dispersion across unsuitable inland environments is very unlikely in these nonflying insects. The analysis could also be affected by the fact that the haplotypes of P. fernandezlopezi, a species exclusive to La Gomera, are located at a greater geographical distance from the other haplotypes in the total network (all of them of Gran Canaria). La Gomera is of independent volcanic origin. It is over 130 Km from Gran Canaria and separated by waters over 2000 m deep, so the presence of the species on the former is likely to be the result of a dispersal across the sea from Gran Canaria, although the converse possibility can not entirely be ruled out. For this reason, we compared the results obtained including or excluding the haplotypes of P. fernandezlopezi from La Gomera as part of the geographical range of the P. granulicollis species group. Permutation contingency tests were applied to test the association of habitat characteristics (see Table 1) as qualitative categories with particular nested haplotype groups using permutations. The program chiperm 1.2 (Posada 2000) was used to perform this analysis. Results Cytochrome Oxidase II sequences Among all individuals examined for a total of 722 aligned nucleotide positions of Gran Canaria Pimelia, we found 53 unique haplotypes, of which 22 were exclusive to P. granulicollis and P. estevezi, while 31 corresponded to P. sparsa. The best model for the data based on the Akaike information criterion was GTR + G (General Time Reversible with a gamma distribution shape parameter G = ), with a lnl = This model gave sequence divergences ranging from a minimum of 0.1% (between different haplotypes of P. sparsa) to a maximum of 21.4% (between haplotypes of P. sparsa and P. granulicollis). The latter is a surprisingly high distance for related species endemic to the same island, and shows a deep divergence between the two taxa. Among the entire COII data set, 24 first (19.7%) and 90 third (73.7%) codon positions vary, only 8 (6.6%) substitutions were detected at second positions. Phylogenetic analyses For phylogenetic analyses we included sequences of the related species P. fernandezlopezi endemic to La Gomera in addition of the unique haplotypes from Gran Canarian taxa. COII sequences of other Pimelia species alien to Gran Canaria were obtained in order to be used as outgroups: the members of Tenerife sister clade P. radula radula, P. radula oromii and P. canariensis, and the Fuerteventura and Lanzarote endemism P. lutaria, which is considered basal to all Canary Pimelia supported both by COI mtdna and highly repetitive satellite DNA phylogenies (Juan et al. 1995; Pons et al. 2002). Of the total 722 bp examined, 231 positions were variable for the ingroup sequences, of which 108 were parsimony informative. The parsimony analysis rooting at P. lutaria with all positions equally weighted rendered 67 most parsimonious trees (501 steps, consistency index CI = 0.62; rescaled consistency index = 0.55; retention index = 0.89) whose strict consensus showed monophyly of P. sparsa and for a single clade including P. granulicollis, P. estevezi and P. fernandezlopezi, with the latter in an unresolved polytomy. A single round of reweighting according to rescaled CI resulted in 92 parsimonious trees of CI = 0.81, resolving the polytomy between P. granulicollis and P. fernandezlopezi (Fig. 2). MP and NJ trees obtained using the ML distances were highly congruent and indicate paraphyly of Gran Canarian Pimelia as they include the La Gomera P. fernandezlopezi. Bootstrap (MP and NJ analyses) and Bremer (MP analysis) values indicate that the clade including all the P. granulicollis, P. estevezi (Gran Canaria) and P. fernandezlopezi (La Gomera) haplotypes (clade A in Fig. 2, Bremer support > 10) is monophyletic. The haplotypes of the latter taxon showed a mean divergence of 2.6% (range %) with respect to its Gran Canarian relatives, which confirms them as closely related. In addition, all P. sparsa haplotypes formed a well supported clade (Bremer support = 5, clade B in Fig. 2), composed of two quite divergent sister groups (7.9% mean divergence between the two): one including haplotypes of some of P. s. serrimargo samples from the east of Gran Canaria plus all P. s. albohumeralis haplotypes from two

6 2136 H. G. CONTRERAS-DÍAZ ET AL. Fig. 2 Strict consensus of the most parsimonious trees obtained after a round of re-weighting according to rescaled consistency index. Nodes having Bremer support (first number above branches) and bootstrap values (second number above branches) higher than 60% are indicated. Particular nodes mentioned in the text are labelled (A, B, B1, B2, C1, C2). isolated localities in the south and southeast (clade B1, Bremer support = 5). The other group is made up of sequences of P. s. sparsa from the central, high altitude populations (where only this subspecies occurs) and samples of P. s. serrimargo from the west side of the island (clade B2, Bremer support = 7). Within this clade, there is a further highly supported group (clade C2, Bremer support > 10) which includes haplotypes obtained from P. s. serrimargo samples from three localities in the northwest (Fig. 2). For calibrating branch lengths a rough estimate of 2 2.3% per million year is generally assumed for arthropod mtdna (Brower 1994). Although calibration data in beetles is scarce, Ribera et al. (2001) showed that for Agabus species (Dytiscidae, Coleoptera) the 2%/Ma estimation (based on COI + 16S sequences) is in agreement with the rate reported by Gómez-Zurita et al. (2000) for Timarcha leaf beetles based on 16S rrna. A molecular clock was rejected because the constrained and unconstrained analyses were significantly different so the nonparametric rate smoothing (NPRS) method of Sanderson (1997) was applied to produce an ultrametric tree. ML distance branches were transformed by NPRS, which smoothes local transformations in rate as rates change over the tree. Considering the maximum divergence of 21.4% for the Gran Canarian Pimelia, an estimated age of Ma for the most recent common ancestor of the entire Gran Canarian clade (including P. fernandezlopezi) can be estimated. Using this age as callibration point, a maximum age of Ma can be deduced for the diversification of the P. granulicollis species-group (node A in Fig. 2) and about 3 Ma for the nodes C1 and C2. Population structure in Pimelia granulicollis and closely related taxa A total of 30 sequences were obtained from nine sampled

7 PHYLOGEOGRAPHY OF GRAN CANARIAN PIMELIA 2137 Table 2 Analysis of molecular variation (amova) among populations of Gran Canarian Pimelia grouped into different associations. Variance explained by differences among groups and its significance using probabilities derived from 1000 permutations are indicated. Phi ST, Phi SC and Phi CT are fixation indices within populations, among populations within groups and among groups of populations in the species, respectively Groups Phi ST Phi SC Phi CT groups % among P P. granulicollis < granulicollis > < estevezi > < 0.01 < north > < east > < south > < west > < 0.01 < north > < east > < south > < southwest > < west > < 0.01 < north > < east > < BU > < DU > < southwest > < west > < 0.05 < north > < east > < BU > < DU > < southwest > < CH > < AR > < 0.01 P. sparsa < northwest > < central-west > < east > < 0.05 < west > < central > < east > < 0.01 < albohumeralis > < sparsa > < serrimargo > < 0.01 < northwest > < northeast > < east-south > < central-west > n.s < northwest > < northeast > < east > < south > < central-west > < 0.01 < northwest > < northeast > < east > < albohumeralis > < central-west > n.s < northwest > < northeast > < TU > < AI > < DU > < central-west > n.s < VI > < GU > < SA > < northeast > < TU > < AI > < DU > < central-west > n.s populations of P. granulicollis + P. estevezi clade (A), whose haplotype distribution is shown in Table 1. The nucleotide diversity among all sequences was relatively low, = ± ( = ± considering only P. granulicollis haplotypes). No particular geographical trend was observed in the sequence variation among these samples, except for a reduced diversity in AG and in the eastern locations of AI-2 + TU-2. amova was performed considering geographical regions on the coast of Gran Canaria by dividing the sampling localities of P. granulicollis into groups based on geographical distribution. One of the schemes considered the eastern populations (AI-2 + BR + TU-2) as one group and the remaining six populations as separate additional groups (see Table 2). This design maximized the Phi CT value (fixation index among groups), Phi CT = 0.420, so it was chosen as best describing geographical subdivision (Table 2). Population structure in Pimelia sparsa Nine populations of P. sparsa were sampled with a total of 45 individuals (corresponding to clade B in Fig. 2), with the haplotype distribution shown in Table 1. The nucleotide diversity obtained ( = ± 0.001) was considerably higher than in P. granulicollis. amova was used as above to assess molecular variance among P. sparsa populations (Table 2). This analysis confirmed a high population differentiation, with a geographical grouping showing a maximum Phi CT = considering northwest (SA + GU + VI), central-west (CB + AL), northeast (CO), east (AI-1 + TU-1) and south (DU-1) localities as different population groups. Nested clade analyses: Pimelia granulicollis Statistical parsimony networks were constructed (Templeton et al. 1992) for P. granulicollis and P. estevezi with haplotypes connected by 11 substitutions, which is the number of steps having a 95% probability of being linked without homoplasy in our data set (Fig. 3). A single network was produced when haplotypes of La Gomera P. fernandezlopezi were included in the analysis. This resulted in more information and statistical power than their exclusion, despite breaking the continuous terrestrial geographical range of mtdna haplotypes. The total network is made up of 24 sampled haplotypes distributed into thirteen 2-step, five 3-step and two 4-step haplotype groups. The network was alternatively rooted at haplotype groups 4.1 or 4.2, the two analyses gave similar results at low nesting levels but differed at higher ones. A contingency test showed a significant level of geographical association of genotypes for groups 4.1 (χ 2 = 20.00; P < 0.01), 4.2 (χ 2 = 14.22; P < 0.01) and for the total network (χ 2 = 27.58; P < 0.01), but not at lower nesting categories (not shown). The inference results for the geographical NCA analysis are summarized in Table 3. Within 2-step levels, past fragmentation is the common outcome of the analysis, mainly affecting haplotypes from the southeast of the island. Other significant deviations from a random geographical distribution are observed in the haplotype groups at the four-step level, caused by restricted gene flow with isolation by distance (haplotype group 4.2) and contiguous range expansion (haplotype group 4.1). Finally, the total network showed a significant deviation from a random geographical distribution of

8 2138 H. G. CONTRERAS-DÍAZ ET AL. Fig. 3 Network obtained using statistical parsimony for P. granulicollis, P. estevezi and P. fernandezlopezi COII mtdna sequences. Sampled haplotypes are in circles (except the most frequent one in a square) with designations as in Table 1, each connection represents a mutational step. Intermediate missing haplotypes are represented by empty circles. Nested haplotype groups with increasing number of steps are enclosed in rectangles and numbers for each nesting level are indicated. Symbols as in Fig. 1 indicate the taxon from which the haplotypes were obtained. haplotypes inferred to be caused by contiguous range expansion rooting at 4.2 and restriction to gene flow with isolation by distance, if 4.1 is considered interior and 4.2 tip as the alternative hypothesis. Nested clade analyses: Pimelia sparsa P. sparsa haplotypes were arranged in three separate networks having connections of less than 11 substitutions within each of them, based on the limits of parsimony of Templeton et al. (1992). Of these three networks, C1 and C2 are connected by 19 steps, while C1 and B1 by 26 (Fig. 4). Network B1 includes 17 sequences coming from eastern localities exactly equivalent to the corresponding clade in the phylogenetic tree (see above and Fig. 2), belonging to all P. s. albohumeralis haplotypes and part of P. s. serrimargo ones. Network C2 includes five haplotypes from northwestern populations of P. s. serrimargo. Network C1 is made up of nine haplotypes from central-western populations belonging to all P. s. sparsa and a fraction of P. s. serrimargo haplotypes. Nested contingency tests for networks C1 and C2 did not detect a significant level of geographical association at any hierarchical level. The same analysis for network B1 detected significance for group 4.1 (χ 2 = 21.00; P < 0.01) and group 3.2 (χ 2 = 20.28; P < 0.01) and for the total network (χ 2 = 86.02; P < 0.01) but not at lower nesting levels. Accordingly, geographical NCA outcome (Table 4) gave nonsignificant or inconclusive inferences at low nesting levels. At higher levels, inference for haplotype group 3.2 is restricted gene flow but with some dispersal over intermediate areas not occupied by the species of south (DU-1) from eastern (AI-TU-1) localities. Moreover, the northeastern CO locality seems to be allopatric from the remaining areas included in the haplotype group 4.1. This was an expected result as this area of La Isleta was formed independently from the rest of the island but subsequently connected by a sand bar (Pestano & Brown 1999). Finally, for the total network, different hypotheses of interior/tip status for the haplotype groups involved were tested, all of them giving the same conclusion of allopatric fragmentation between the three main clades. Table 4 shows the results for the total network rooting at group 4.1. A schematic representation of P. sparsa NCA inferences overlayed on a map of Gran Canaria showing the lava flows of the last volcanic cycle (3 3.5 Ma) (based on Marrero & Francisco-Ortega 2001) is shown in Fig. 5. Testing association of evolutionary lineages with ecological traits The populations of Pimelia in Gran Canaria show ecological differences with regard to habitat characteristics, e.g. P. estevezi occurs in a fossil dune and P. granulicollis populations either in coastal calcareous dunes or in basaltic sand (see Table 1). Permutation contingency analyses were performed to test the association between lineages and habitat traits. They showed a significant departure from the null hypothesis of no association at the total cladogram and 4.2 levels only (P < 0.005; permutations). Similarly, permutation contingency analyses relating genetic lineages

9 PHYLOGEOGRAPHY OF GRAN CANARIAN PIMELIA 2139 Table 3 Results of the geographical nested cladistic analyses for Pimelia granulicollis, P. estevezi and P. fernandezlopezi mtdna haplotypes. The values of Dc and Dn and the test of interior vs. tip clades (I vs. T) are given in the cases of geographical and haplotype variation and for instances of clades with significantly larger or shorter geographical ranges than expected under random distribution. For the other clades the null hypothesis of no geographical association of haplotypes (or clades) can not be rejected. Interior and tip clades, based on the nesting design or in independent evidence are indicated with subscript i and t, respectively. Significant values (P < 0.05), based on permutational contingency tests, are highlighted in bold. Superscript L and S indicate values for clades with geographical distribution significantly larger and shorter than expected, respectively. N refers to the number of individuals in each level. The followed key chain and the inferred process obtained (from the updated inference key in Templeton et al. 1995) is specified Dc Group 2.9 (N = 10) Clade 1.18 i Clade 1.19 t S L Clade 1.22 t Clade 1.23 t I vs. T NO Past fragmentation Group 2.13 (N = 7) Clade 1.27 i Clade 1.30 t Clade 1.31 t I vs. T L NO Past fragmentation Group 4.1 (N = 12) Clade 3.1 t S S Clade 3.2 t L Clade 3.3 i S I vs. T NO Contiguous range expansion Group 4.2 (N = 22) Clade 3.4 i Clade 3.5 t S I vs. T L NO Restricted gene flow with isolation by distance Total Cladogram rooting at 4.2 (N = 34) Clade 4.1 t Clade 4.2 i S S I vs. T S NO Contiguous range expansion Total Cladogram rooting at 4.1 Clade 4.1 i Clade 4.2 t S S I vs. T L L NO Restricted gene flow with isolation by distance Dn in P. sparsa and habitat characteristics (high altitude, lowland dry areas and coastal dunes) showed that the association of the three main genetic breaks within the total cladogram with the ecological traits (and subspecies status) was statistically significant (P < 0.001; permutations). These results suggest that some of the main genetic lineages in Gran Canarian Pimelia show adaptive variation not only as a result of historical or recent limitations to gene flow but also to differential selection regimes. Discussion Phylogeny of Gran Canaria species and populations of Pimelia The mtdna COII phylogeny obtained shows that Gran Canarian Pimelia form a paraphyletic group, with sequences of P. fernandezlopezi from La Gomera embedded in the lineage of P. granulicollis. Two divergent clades are highly supported; one representing the species P. sparsa and the other including P. granulicollis, P. fernandezlopezi, and the P. estevezi haplotypes, the two latter being paraphyletic with respect to P. granulicollis. The results obtained for the two divergent species-groups allow us to examine the divergence values of the Gran Canarian endemic taxa in relation to the complex geological history of the island, marked by at least two major volcanic cycles (Pérez- Torrado et al. 1995). There is more genetic diversity within P. sparsa than within P. granulicollis and allied taxa. Within the former, the maximum genetic distance obtained by the Maximum Likelihood substitution model is around 10.5%, while in P. granulicollis and allied taxa it is 4.0%. Assuming the molecular clock calibrated at about 2% per million year as a rough estimate, the diversification of Gran Canarian Pimelia could have been much later than the first catastrophic volcanic episode dated at 14 Ma and surviving through the Roque Nublo eruptive period, dated at Ma (Pérez-Torrado et al. 1995). On the other hand, P. granulicollis lineage diversification could have begun around the last volcanic episode. It has been noted elsewhere that most animal and plant groups associated with the forest ecosystems of the island are recent, postdating the last eruptive period, while those occurring exclusively outside the laurel and pine-wood ecosystems, predate and persisted trough the violent Roque Nublo episode (Emerson 2002). The minimum divergence levels within P. granulicollis are very similar to those shown by the two P. fernandezlopezi haplotypes (0.15%); besides this, no shared haplotype has been found between the two islands of Gran Canaria and La Gomera. These facts rule out a recent human introduction of P. granulicollis on La Gomera and suggest a long distance colonization from one island to the other. The weakness of our phylogenetic analysis is that both

10 2140 H. G. CONTRERAS-DÍAZ ET AL. Fig. 4 Networks obtained under the limits of statistical parsimony for P. sparsa. The three subnetworks named C1, C2 and B1 are concordant with the main clades in the tree in Fig. 2. Numbers of mutational steps connecting the three subnetworks are indicated. Sampled haplotypes are in circles with designations as in Table 1, symbols for taxa are the same as in Fig. 1. Haplotype b8 is shared by two P. serrimargo and P. albohumeralis individuals. Nested haplotype groups of increasing number of steps are enclosed in rectangles and numbers for each nesting level indicated. Table 4 Results of the geographical nested cladistic analyses for Pimelia sparsa network B1. Dc, Dn and the test of interior vs. tip clades (I vs. T), their significances and the inferences obtained are given as outlined for Table 3 Dc Dn Group 3.2 (N = 17) Clade 2.3 t L Clade 2.4 i Clade 2.5 t S Clade 2.6 t Clade 2.7 t S S I vs. T L L (too few clades) 7 8 YES Restricted gene flow but with some dispersal over intermediate areas not occupied by the species. Group 4.1 (N = 22) Clade 3.1 t S L Clade 3.2 i S S I vs. T L S YES Allopatric fragmentation Total Cladogram (N = 27) Clade 2.10 t S S Clade 3.3 i S S Clade 4.1 i L I vs. T L L YES Allopatric fragmentation possibilities are equally likely; although a P. granulicollis haplotype seems to be basal for the entire clade, the hypothesis of P. fernandezlopezi at the base of the lineage only requires an extra step under parsimony. Nevertheless, given the predominant sea currents in the archipelago from east to west, the more recent formation of La Gomera (10 Ma) compared to Gran Canaria (15 Ma) and the fact Fig. 5 Schematic representation of the P. sparsa NCA inferences overlayed on a map of Gran Canaria showing the lava flows of the last volcanic cycle (3 3.5 Ma) (modified from Fig of Marrero & Francisco-Ortega 2001). The broken lines represent allopatric fragmentation and the curved arrow restricted gene flow, but with some dispersal over intermediate areas not occupied by the species. Letters and numbers refer to the particular nesting levels involved and symbols to subspecies of P. sparsa as indicated in Fig. 1. that no extant relatives or fossils are known of this radiation (subgenus Aphanaspis) on the intermediate island of Tenerife, we favour the possibility of colonization of La Gomera from Gran Canaria. P. sparsa mtdna genealogy shows two sister groups, B1 and B2, having considerable

11 PHYLOGEOGRAPHY OF GRAN CANARIAN PIMELIA 2141 sequence divergence with respect to each other, and a further clade within B2 corresponding to some, but not all, haplotypes of P. s. serrimargo. There is no support at the mtdna sequence level for any of the three proposed subspecies based on morphology, since none of them appears as monophyletic. Phylogeographical patterns amova, an analysis based on Wright s F, estimates population differentiation directly from molecular divergences, arranging hierarchically the population data and computing mean squares for groups at different levels. This allows for hypothesis testing of between-group and within-group differences at several hierarchical levels. The amova results for Gran Canarian Pimelia detected subdivision among population groups. In addition, nested clade analysis (NCA), an increasingly popular method for studying intraspecific gene genealogies (Templeton et al. 1995; Templeton 1998) was applied to the Pimelia mtdna data set. For our data, the use of parsimony and nested clade analyses gave, as expected, coincidental clades and haplotype groups, in part corroborated by amova. mtdna geographical distribution of P. granulicollis and allied taxa points to past population fragmentation at low nesting levels, although these inferences are weak given the low sample sizes involved. At higher levels, contemporaneous or recurrent events were inferred, such as restricted gene flow with isolation by distance and contiguous range expansion. For P. sparsa all analyses converge in showing three major lineages, roughly dividing into eastern from western haplotypes, and within the latter, northwest from west and central mtdna sequences (see Figs 4 and 5). These three groups, in conflict with taxonomic designations, define three disjunctive statistical parsimony networks in the NCA analysis. No significant association between haplotypes and geography was found within the networks coincidental with clades C1 and C2. This is best explained by low genetic diversity in these clades, i.e. insufficient resolution for the analysis to be performed or simply due to their small sample sizes (2 8 individuals per locality). Within network B1 and including the three clades in the analysis, NCA results suggest fragmentation of populations as the main cause of the geographical association of genotypes of the distinct clades (haplotype group 4.1 and total network), which is compatible with the fact that P. sparsa sensu lato populations are much older than the last main volcanic cycle. We hypothesize that the pattern observed has been produced by habitat fragmentation due to interruption by lava flows. Indeed, there has been opportunity for habitat fragmentation by the volcanic emissions in the northern part of the island either around Ma or more recently (Funck et al. 1996; Pérez-Torrado et al. 1995) and the proposed lava covered areas could well have acted as the common geological event producing population subdivision (Marrero & Francisco-Ortega 2001). A remarkably similar general pattern was found in the Gran Canarian skink Chalcides sexlineatus using 12S mtdna sequences: three deep lineages corresponding to northern, southeastern and southwestern parts of the island with low levels of mitochondrial introgression for two of these clades (Pestano & Brown 1999). This phylogeography was explained by population vicariance during volcanic eruptions in the past 2.8 Ma in the north of the island, plus ongoing differential adaptation and selection against hybrids in the transition zone between the humid northern and arid southern areas of Gran Canaria. Although the predictions based on the geological history of the island seem corroborated by NCA, and coincidental patterns in different organisms reinforce the inferences obtained in the case of P. sparsa, there are limitations to the use of single gene trees and statistical phylogeographical methodologies to deduce a species history. The a priori assumption that a long term barrier to gene flow explains any genealogical break can be incorrect for species with low dispersal and population sizes; in this situation phylogeographical breaks will form even without geographical barriers, especially for matrilinear genealogies (Irwin 2002). Knowles & Maddison (2002) have shown that the stochasticity of population genetic processes and the lack of a statistical framework to test likelihood of a particular recurrent (or historical) process over alternative ones can compromise the NCA results. These studies show that although the field of statistical phylogeography has made much progress towards providing objective explanations for the observed patterns, much work remains to be done to distinguish alternative hypotheses. Implications for managing Gran Canaria Pimelia populations The Canary Islands Regional List of Endangered Species (BOC 2001) includes P. granulicollis, P. estevezi and P. s. albohumeralis from Gran Canaria, and P. fernandezlopezi from La Gomera. Population density is very low in some of the P. s. albohumeralis geographical ranges (Arinaga and Playa del Cabrón) which are in clear regression; and for P. estevezi (Punta de Las Arenas) a small population present in a Natural Reserve area (the situation is similar in La Gomera for P. fernandezlopezi). The main threats to these endemic beetles, especially in the case of P. s. albohumeralis, are habitat reduction or loss by human action caused by the expansion of tourist resorts, sand extraction activities, rubbish dump sites in distribution areas, etc. Pimelia s. sparsa occurs on a ridge from Los Pechos to Tamadaba (ranging from 1900 to 1200 m in altitude) and surveys indicate that it is constituted by small populations. P. granulicollis and P. s. serrimargo have wider distributions

12 2142 H. G. CONTRERAS-DÍAZ ET AL. in the lowland and coastal areas and have larger population sizes. However, in the case of P. granulicollis the distribution is much more fragmented and the two more isolated populations AL and GT (both in genetic and geographical terms) are in a worse situation due to habitat reduction. The present protection of Gran Canaria Pimelia beetles is based on morphological and/or biogeographical considerations alone, with no assessment of the relationships between morphospecies and genetic diversity. Management units for conservation purposes at the intraspecific level should incorporate adaptive differences as evidenced by both genetic and ecological data (Crandall et al. 2000). Genetic and ecological exchangeability (sensu Templeton 1989) should be used as null hypotheses to test for population distinctiveness. The results of Pimelia population analyses using different analytical methodologies show that distinct genetic and ecological units exist on Gran Canaria, in part differing from the currently accepted taxonomic views. Within P. granulicollis, the populations on the eastern coast (AI-2, BR and TU-2) are genetically differentiated from the remaining populations due to a restriction of gene flow, including that of the genetically close but geographically localized P. estevezi. Permutation contingency analyses performed to test the association between lineages and ecology showed significant departures from the null hypothesis of no association in P. granulicollis and P. sparsa at the deeper genetic breaks. The haplotypes within groups 4.1 and 4.2 and at the lower levels 3.5 and 3.4 of P. granulicollis and closely related taxa are candidates to be treated as different management units. In spite of being morphologically the most differentiated within the Aphanaspis group, P. fernandezlopezi seems to be part of the P. granulicollis haplotype diversity, but with a narrow and isolated geographical distribution in a fossil dunes habitat in the island of La Gomera. The two nominal species P. fernandezlopezi and P. estevezi clearly deserve conservation effort despite lacking a long history of reproductive isolation as they occupy localized geographical areas with no connections with surrounding regions by suitable habitat, and so they are currently isolated. The three recognized subspecies of P. sparsa do not reflect the matrilinear genealogies obtained and the populations having mitochondrial haplotypes within the C1, C2 and B1 should be managed separately. Multiple gene trees are necessary to assess if the pattern obtained in P. sparsa has been produced by mtdna introgression or might be better explained by stochastic lineage sorting. Acknowledgements We thank Dr Jesús Gómez-Zurita for help in nested clade analysis and for his critical reading of the paper. The comments of Dr Eduard Petitpierre and permission by Dr Brent Emerson to cite his unpublished data (now in press) are also acknowledged. Drs Alfried Vogler, Ignacio Ribera and three anonymous referees made comments which helped in improving a previous version of the paper. Guido Jones kindly corrected the English style of the manuscript. Elena Morales, Sonia Martín, Alexis Galindo and especially Heriberto López helped us collect specimens. Families López Hernández, Amador Medina and Medina Montesdeoca provided lodging facilities during fieldwork journeys. This study was supported by the research funds REN /GLO of the Spanish Ministerio de Ciencia y Tecnología and a Conservation Project on Pimelia of the Canary Islands Viceconsejería de Medio Ambiente. The corresponding permits to collect protected species were obtained from Cabildo de Gran Canaria and Viceconsejería de Medio Ambiente. References Ancochea EJ, Fuster M, Ibarrola E et al. (1990) Volcanic evolution of the island of Tenerife (Canary Islands) in the light of new K-Ar data. Journal of Volcanology and Geothermal Research, 44, Anguita F, Hernán F (1975) A propagating fracture model versus a hot spot origin for the Canary Islands. Earth and Planet Science Letters, 27, Avise JC (2000) Phylogeography. Harvard University Press, Cambridge, Massachusets. BOC (2001) Catálogo Regional de Especies Amenazadas. Boletín Oficial de Canarias, 97, Brower AVZ (1994) Rapid morphological radiation and convergence among races of the butterfly Helicanius erato inferred from patterns of mitochondrial DNA evolution. Proceedings of the Natural Academy of Science USA, 91, Brown RP, Pestano J (1998) Phylogeography of skinks (Chalcides) in the Canary Islands inferred from mitochondrial DNA sequences. Molecular Ecology, 7, Clement MD, Posada D, Crandall KA (2000) tcs: a computer program to estimate gene genealogies. Molecular Ecology, 9, Crandall KA, Bininda-Emonds ORP, Mace GM, Wayne RK (2000) Considering evolutionary process in conservation biology. Trends in Ecology and Evolution, 15, Emerson BC (2003) Genes, geology and biodiversity: faunal and floral diversity on the island of Gran Canaria. Animal Biodiversity and Evolution, in press. Emerson BC, Oromí P, Hewitt GM (1999) mtdna phylogeography and recent intra-island diversification among Canary Island Calathus beetles. Molecular Phylogenetics and Evolution, 13, Emerson BC, Oromí P, Hewitt GM (2000a) Tracking colonisation and diversification of insect lineages on islands: mitochondrial DNA phylogeography of Tarphius canariensis (Coleoptera: Colydiidae) on the Canary Islands. Proceedings of the Royal Society of London B, 267, Emerson BC, Oromí P, Hewitt GM (2000b) Colonisation and diversification of the species Brachyderes rugatus (Coleoptera) on the Canary Islands: evidence from mitochondrial DNA COII gene sequences. Evolution, 54, Español F (1971) Las Pimelia de las Islas Canarias. Anuario Estudios Atlánticos, 7, Excoffier L, Smouse PE et al. (1992) Analysis of molecular variance inferred from metric distances among DNA haplotypes: application to human mitochondrial DNA restriction data. Genetics, 131,