• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

The Origin and Radiation of Macaronesian Beetles Breeding in Euphorbia: The Relative Importance of Multiple Data Partitions and Population Sampling

Posted on: Thursday, 2 December 2004, 03:00 CST

Abstract.-

Species-level phylogenies derived from many independent character sources and wide geographical sampling provide a powerful tool in assessing the importance of various factors associated with cladogenesis. In this study, we explore the relative importance of insular isolation and host plant switching in the diversification of a group of bark beetles (Curculionidae: Scolytinae) feeding and breeding in woody Euphorbia spurges. All species in the genus Aphanarthrum are each associated with only one species group of Euphorbia (succulents or one of three different arborescent groups), and the majority of species are endemic to one or several of the Macaronesian Islands. Hence, putative mechanisms of speciation could be assessed by identifying pairs of sister species in a phylogenetic analysis. We used DNA sequences from two nuclear and two mitochondrial genes, and morphological characters, to reconstruct the genealogical relationships among 92 individuals of 25 species and subspecies of Aphanarthrum and related genera. A stable tree topology was highly dependent on multiple character sources, but much less so on wide population sampling. However, multiple samples per species demonstrated one case of species paraphyly, as well as deep coalescence among three putative subspecies pairs. The phylogenetic analyses consistently placed the arborescent breeding and West African-Lanzarote-distributed species A. armatum in the most basal position in Aphanarthrum, rendering this genus paraphyletic with respect to Coleobothrus. Two major radiations followed, one predominantly African lineage of succulent feeding species, and one island radiation associated with arborescent host plants. Sister comparisons showed that most recent divergences occurred in allopatry on closely related hosts, with subsequent expansions obscuring more ancient events. Only 6 out of 24 cladogenetic events were associated with host switching, rendering geographical factors more important in recent diversification. [Biogeography; data combinability; Euphorbia; host plant use; molecular systematics; Scolytinae; speciation; taxon sampling.]

Reconstruction of species-level phylogenies is a powerful tool in the study of macroevolutionary patterns of diversification, where the assessment of putative sister species provides a quantitative measure of the factors associated with cladogenetic events. These measures can only be as accurate as the phylogenies themselves, however, and various methodological problems may reduce the precision needed for such studies. The best practical approach to overcome these problems involves reconstructions under various optimality criteria, sampling every species and multiple populations of the focus group, and sampling characters from as many independent character sources as necessary to stabilize the phylogenetic hypothesis. Most studies on higher level phylogenies so far have sacrificed details for scale by including only one (or zero) individual per species, using only a single or a few sources of phylogenetic characters (Barraclough and Nee, 2001). Our aim in this study is to establish a robust phylogenetic hypothesis to explore important factors in the diversification of a group of bark beetles associated with Euphorbia spurges, by sampling five different character partitions and nearly all species and populations in the reconstruction of the phylogeny. During the process of exploring these data, we investigate the effect of partial sampling of characters and taxa, providing empirical results to complement recent theoretical discussions on these issues (Cummings et al., 1995; Poe and Swofford, 1999; Rosenberg and Kumar, 2.001; Pollock et al, 2002).

Putative diversifying factors associated with cladogenetic events may be of two main categories, geographical isolation or ecological shifts (Barraclough and Nee, 2001). The relative strength of the comparative approach to assess speciation modes, however, depends logically on how accurately geographical and ecological differences can be defined, and how stable these are over evolutionary time scales (Losos and Glor, 2003). Perhaps the most precise model systems to explore patterns of colonization and geographic speciation are groups of species endemic to different oceanic islands, where each island in an archipelago has very similar barriers of open water between them (Juan et al, 2000; Emerson, 2002; Gillespie and Roderick, 2002). Similar or even stronger barriers to gene flow, in terms of ecological isolation, occur in organisms that specialize on narrow resources for feeding or reproduction. Some of the best examples on ecological isolation involves plants that serve as the sole host for herbivores, especially when the herbivores complete their entire life cycle within a single resource (Kelley and Farrell, 1998; Machado et al., 2000; Pellmyr and LeebensMack, 2000; Scheffer and Wiegmann, 2000; Despres et al., 2002).

Whereas many phylogenetic studies to date have focused on either geographical or ecological factors in speciation, rather few have tried to include both in a simultaneous study (Joy and Conn, 2001; Jordan et al., 2003; Percy, 2003). This paper explores the potential influence from both of these factors in the diversification of crypturgine bark beetles breeding in arborescent Euphorbia spurges on the Macaronesian islands. Macaronesia is made up of four archipelagos west of the African coast: the Azores (which do not host any crypturgine beetles), Cape Verde, and the Canary and Madeiran island groups. The oldest islands in each archipelago (Fuerteventura, Porto Santo, and the eastern Cape Verde islands) are known to be at least 18 My, with younger islands gradually decreasing in age westward, with the most recent island only 1.1 My (El Hierro; see Hess et al., 2000, and Brown et al., 2001). Hence, these islands provide opportunities for diversification over an extended period of time, similar to the Hawaiian and Galapagos archipelagos, which have played prominent roles as evolutionary model systems (e.g., Gillespie and Roderick, 2002).

Arborescent growth forms have evolved several times independently in Euphorbia (Molero et al., 2002), and are particularly frequent on islands and coastal areas, as in many other oceanic plant groups (Bohle et al., 1996; Panero et al., 1999). Their prevalence on the Macaronesian islands and in drier parts of the African mainland has provided an unique opportunity for niche shifts and radiations in bark beetles of the weevil subfamily Scoly tinae (jordal et al., 2003). The largest clade of beetles associated with Euphorbia are found in the tribe Crypturgini, where all species of Aphanarthrum (19) and ColeobothYUS (4) breed exclusively in dead or moribund parts of these plants. The majority of species are found on one or several of the Macaronesian islands (17) or western Africa (5), with the remaining species (5) reaching eastern parts of Africa and India. Five species have also been divided into two or three subspecies based on genitalic differences and allopatric island distributions (Israelson, 1972), suggesting that speciation is currently an ongoing process. The closest relatives to Aphanarthrum and Coleobothrus are most likely Cisurgus and Crypturgus whose species are morphologically very similar to each other (Wood, 1986). However, all species of the mainly Holarctic Crypturgus (10) breed in conifers only, whereas species of the Mediterranean Cisurgus (6) breed in various woody herbs, including two species breeding in succulent euphorbs in Morocco and the Canary Islands. The last two genera in Crypturgini are the monotypic Dolurgus (breeding in North American pines) and Deropria (associated with Rubus in the Canary Islands), believed to be distantly related to the other four genera (Wood, 1986).

Dead or moribund plant material serves simultaneously as both the food resource and as breeding space for adults and larvae of these beetles, where they complete the entire life cycle as in all other Scolytinae bark beetles. The level of host plant fidelity is high, and each species show distinct preference for only one of the four main groups of Euphorbia which have proven suitable as host plants (Table 1; Jordal, submitted). All host plants produce various forms of latex (Carter and Smith, 1988; Seigler, 1994) that are still present during the early stage of colonization, and are further distinguished by growth form and structural differences in tissue (Webster, 1994) as illustrated by the cactus-like succulents (e.g., E. canariense, E. handiense), and the shrub-like groups consisting of E. lamarckii complex (includes the artropurpurcacomplex), E. balsnmifera, and E. longifolia. The four groups are also deeply separated by molecular data (Molero et al., 2002), further confirming the different nature of potential host plants. The wide and largely overlapping distributions among each of the three first and most important host groups, together with the beetle's distinct preference for one of the groups, make cladogenetic events associated with host switching less ambiguous in t\his model system.

Current classification of Crypturgini, with reciprocal monophyly of all genera, implies at least three origins of the Macaronesian species and two origins of Euphorbia associations. The aim of this study is to establish a new and well-supported phylogeny to allow a more exact estimation of the rate and direction of island-mainland transitions and host plant switching, and hence, to estimate the relative frequency of allopatric divergence versus host shifts in diversification. The largely overlapping distribution within several guilds of ecologically similar species, and the lack of externally diagnostic characters between most of these (Israelson, 1972), may suggest that not all of the species are reciprocally monophyletic. Recent splits often reveal problems with paraphyletic species, or worse, lineage sorting of ancestral polymorphism or even reticulate evolution under low frequency hybridization. To be able to assess these problems, if they do occur, we included multiple specimens per species, from several different islands, and characters from five different partitions. We used sequences from two genes each from the nuclear and the mitochondrial genomes, and morphological characters, to avoid biased gene tree representation of the species tree, and explored the relative contribution of each partition to the various depths of the combined data phylogeny.

MATERIAL AND METHODS

Taxou Sampling

We collected and sequenced 92 specimens, including all Macaronesian and African species of Aphanarthrum and Coleobothrus (Table 1), two species each of the outgroups Crypturgus and Cisurgus, and Dolurgus pumilus (Mannerheim). Preliminary analyses of 839 EF-Ia coding characters, from 12 crypturgine and 45 additional species of Scolytinae and Platypodine, confirmed the previously suggested monophyly of Crypturgini (Wood, 1986), and that Dolurgus, and Crypturgus plus Cisurgus, are nested sets of outgroups suitable for all further analyses of a monophyletic Aphanarthrum plus Coleobothrus. An identical outgroup topology also resulted from the analysis of 316 aligned characters from the large ribosomal subunit (28S) for 23 crypturgine taxa and 3 scolytine outgroups.

Collected specimens were compared to the original descriptions or type material in the Natural History Museum, London, and Zoologische Staatssammlung, Munich. Based on these comparisons, three species were treated as synonyms as follows: A. saturaturn Peyerimhoff and A. duongi Villiers = A. mairei Peyerimhoff, and A. mododi Paulian & Villiers = A. armatum Wollaston. Also Cisurgus resinifcrac Peyerimhoff was treated as a synonym of Ci. occidentalis Payerimhoff. Three Aphanarthrum species described from India (Wood, 1988) were not available for DNA extraction, but we note that these species are very similar to the two Ugandan species included. Multiple specimens per species were included for all species possible, with coverage of nearly all islands within their known range. Current studies on the population genetics of the Canary Island species confirmed that our samples were unbiased with respect to the populations included here.

TABLE 1. Species and populations included in the phylogenetic analyses. Samples not collected by the senior author are indicated by collectors name (in parentheses).

TABLE 1. Species and populations included in the phylogenetic analyses. Samples not collected by the senior author are indicated by collectors name (in parentheses).

Character Sampling

Morphological and behavioral characters.-Twenty-eight morphological characters were coded and included in the parsimony analyses (Appendices 1, 2). We also included Dcropria elongata (Eggers) in this matrix to test the monophyly of all other crypturgine genera. Host plant use and geographical distribution were based on literature records (lsraelson, 1972, 1976, 1979) and corrected and augmented after intensive sampling throughout the Macaronesian islands (Jordal, in preparation). Previous records from multiple Euphorbia groups were probably based on rare stray specimens on the less preferred host, as documented during recent sampling. Geographical distributions of the beetles were largely congruent with literature records, with only minor corrections needed.

DNA amplification and sequencing.-We sequenced portions of two mitochondrial and two nuclear genes, using mostly the same primer for sequencing as for PCK amplification. Mitochondrial cytochrome oxidase 1 was amplified with the primers si 718 5'-CGA GGA TTT GGA AAT TGA TTA GTT CC-3' (forward), a2237 5'-CCG AAT GCT TCT TTT TTA CCT CTT TCT TG-3' (reverse), or a2411 5'-GCT AAT CAT CTA AAA ACT TTA ATT CCW CTW G-3' (reverse) (Normark et al., 1999). PCR reactions were performed in a 25-L volume containing 0.2 M of each primer, 0.25 mM of each dNTP, 0.5 unit of Bioline BioTaq DNA polymerase, 1 buffer with MgCl2 to a final concentration of 1.5 mM. Typical PCR cycles consisted of 90 s initial denaturing at 94C, followed by 38 cycles of 94C for 30 s, 46C annealing for 60 s, and 72C extension for 60 s, followed by a final extension for 7 min.

The mitochondrial large ribosomal subunit, 16S, was amplified under the same PCR conditions, using the primers 16S 5'-TTT AAT CCA ACA TCC AGG-3' (forward) and 16S 5'-CGC CTG TTT AAC AAA AAC AT-3' (reverse) (Cognato and Vogler, 2001).

The nuclear gene encoding Elongation Factor Iα was amplified with the primers efsl49 5'-ATC GAG AAG TTC GAG AAG GAG GCY CAR GAA ATG GG-3' (forward) and efa 1043 5'-GTA TAT CCA TTC GAA ATT TGA CCN GGR TGR TT-3' (reverse) (Normark et al., 1999). We used a touchdown profile consisting of 43 cycles where all cycles had 72C extension for 40 s and 94C denaturing for 30 s, with the first 16 cycles having decreasing annealing temperature from 60C to 47C by 2C (last 1C) every second cycle, with the final 27 cycles at 46C for 60 s.

The nuclear encoded Enolnse gene (2-phospho-D-glycerate hydrolasc) (Tracy and Hedges, 2000; Farrell et al., 2001) was amplified with nested PCRs using initially the primers ens57c 5'- GAC TCC CGT GGN AAC CCC ACN GTG GAG GT-3' (forward) and ena886 5'- CCA GTC RTC YTG RTC RAA XGG-3' (reverse), with the latter replaced by ena780 5'-TCT TGA ACT CCA AAT CGT A-3' (reverse, sequencing primer) in the nested PCR. Initial reactions were performed under similar conditions as for the mtDNA genes, but with 2.5 mM MgCL: concentration and 44C annealing temperature. The nested PCR reactions used 1.5 mM MgCl^sub 2^ concentration, with 1.0 g bovine serum albumin added. Cycles consisted of 90 s initial denaturing at 94C, followed by 40 cycles where all cycles had 72C extension for 60 s and 94C denaturing for 30 s, with the first 16 cycles having increasing annealing temperature from 40C to 55C by 2C (last IC) every second cycle, with the final 12 cycles at 550C for 60 s. Purified PCR products were sequenced using a standard protocol for big-dye version 3 (Perkin Elmer).

Sequences were assembled and edited with the Lasergene software (DNASTAR, Inc.). There are multiple copies known for EF-Iα and Enolase and we used the single-intron copy (Cl) of EF-Iα and intron free copy of Enolase (Enol) in this study (Farrell et al., 2001; Jordal, 2002). Sequences can be downloaded from GenBank under the accession numbers AY500900 to AY500991 (EF-Iα), AY514121 to AY514211 (Enolase), AY514212 to AY514303 (16S), AY514904 to AY514995 (COI).

Phylogenetic Analyses

Alignments.-For Enolase and the EF-Iα coding region, alignments were unambiguous due to the lack of any insertions or deletions in the coding region. Three COI sequences had indels (in multiples of 3 bp), but none of these putative pseudogenes had stop codons and were cautiously included; these sequences grouped together with conspecific indel free sequences (A. aeoni El Cercado; Co. allouai, Morocco). Sequences of 16S and the EF-Iα intron were aligned in ClustalX (Thompson et al., 1997) under a wide range of parameters to assess alignment stability (gap cost 2, 4, 6, 8, 12 16; gap extension equal or half the cost of gap opening; transitions weighted equally or half the cost of transversions). Preliminary alignments were chosen based on their topological congruence with the topology resulting from the elision matrices which combines all alignments from each partition into one matrix, providing a gradual down weighting of ambiguous sites (Wheeler et al., 1995; Lee, 2001). Final alignments were produced under gap opening to extension cost of 6:3, with transitions half the cost of transversions, resulting in 68 ambiguous sites for 16S, and 54 ambiguous sites in the much more variable EF-Iα intron. These alignments were also the most congruent with the unambiguously aligned portions of the EF-Iα exon and 16S, respectively. A secondary structure model for insects (Buckley et al., 2000) was used to assist in editing the final 16S alignment. All indel ambiguous regions were located in the terminal loops of helices 68 (9 bp), 75 (48 bp, including all but 5 bp of the helix), 81' (8 bp), and 84' (4 bp). Various alignments (or inclusion/ exclusion of ambiguous sites) had no influence in the analyses of combined data. Neither did gaps as a fifth character change the combined data topology. Hence we scored gaps as missing characters in the parsimony analyses, to make comparisons to likelihood analyses more realistic.

Sequence pruning.-To assess the relative importance of sampling multiple sequences per species, and to increase the speed in likelihood analyses, we pruned one sequence from each of the species which had an excess of sequences before each heuristic parsimony search. The phylogenetic depth and number of node switches was noted for each partition, genome, and the combined data, on their respective strict consensus trees.

Tree searching.-Paup* 4.0 (Swofford, 1999) was used in most phylogenetic analyses (the complete data matrix andFigure 1 can be downloaded from Tree-Base, accession number SN1709, http:// www.treebase. org/treebase/). Under the maximum-parsimony (MP) criterion, we performed 500 random addition replicates of heuristic searches for each of the data partitions and the combined matrix. Bootstrap support for individual nodes was assessed by 200 bootstrap replicates (Felsenstein, 1985) of 10 random additions per replicate, with uninformative characters excluded. Bremer support was calculated as the number of additional steps necessary to collapse a node (Bremer, 1994), using TreeRot v.2 (Sorenson, 1999). As DeBry (2001) recently pointed out, the significance of this metric is dependent on the number of informative characters assigned to each node. We have used minimum branch lengths as a proxy estimate of informative characters, and applied DeBry's Table 3 to assess approximate significance of each Bremer support value.

Maximum-likelihood (ML) analyses were performed on the pruned data matrix only (54 terminals), with parameters estimated through the application of Modeltest 3.06 (Posada and Crandall, 1998). This application tests a nested set of substitution models using either the log-likelihood ratio test (LRT) or the Akaike information criterion (AIC). When the two statistics disagreed, we applied the AIC criterion to select the best model for each partition, or for all data combined. Ten random additions were used in each of the final likelihood searches. ML branch lengths were also optimised with the molecular clock enforced, and likelihood scores were compared to those obtained with no constraints using the LRT.

MrBayes 3.1 (Ronquist and Huelsenbeck, 2003) was used in the Bayesian analyses of the combined data, with a GTR +Γ+ model of substitution (selected by Modeltest) fitted to each molecular partition, and a rate variable model fitted to the morphological data. Parameters were estimated on each of the genes and morphology (5 partitions), or to each coding position, EF-Iα intron, 16S, and morphology (12 partitions). Initial searches of 500,000 generations were used to assess convergence in likelihood values obtained from four running Markov chains, sampling trees every 500 generations. Trees from the first 200,000 generations were discarded in the final analysis of 1,000,000 generations, as a 'burn-in' before stability in likelihood values was achieved.

lncongruence.-The incongruence length difference test (ILD) (Farris et al., 1995) has been frequently applied to assess combinability of data partitions. However, both empirical (Yoder et al., 2001) and simulated data (Barker and Lutzoni, 2002, and references therein) have demonstrated the potential failure of this test to allow combination of data when they should be. Potential negative effects from combining statistically incongruent gene partitions can now be tested in a Bayesian framework, by analyzing all data under a uniform model of substitution versus a mixed model where each partition fits their own model. In a parsimony framework, incongruence can further be explored by accumulating data from an increasing number of partitions, aiming towards a predictive hierarchy; that is, it remains stable with the addition of more data (Nixon and Carpenter, 1996). Incongruence may then be quantified by the level of node support in conflicting nodes across different partitions. A node-specific approach to detect conflicting taxa and partitions simultaneously was implemented through the calculation of partitioned Bremer support (Baker and DeSaIIe, 1997; Baker et al, 1998), using TreeRot 2.0 (Sorenson, 1999). We estimated the relative contribution from each data partition to the combined analysis, using default settings with 20 random addition sequences for each node in each of the partitions.

FIGURE 1. Strict consensus of 48 most parsimonious trees of length 5014 steps (CI = 0.35, Rl = 0.82) resulting from the unweighted analysis of all data. Bootstrap support values are shown above the nodes, with minimum branch length and Bremer support values below (ml:BS, 'significant' support shown in bold). Capital letters in bold indicate nodes discussed in the text. Geographic origin of each terminal is indicated by capitals: SV, Sao Vicente; SA, Santo Antao (Cape Verde); Mad, Madeira; PS, Porto Santo (Madeira); EH, El Hierro; LP, La Palma; LG, La Cornera; T, Tenerife; GC, Gran Canaria; F, Fuerteventura; L, Lanzarote (Canary Islands); Mor, Morocco; Ug, Uganda, US, United States of America; N, Norway. The line drawing illustrates Aphanarthrum armatiim.

The overall support from different data partitions may also be asymmetric, such that one partition perform significantly worse on the topologies resulting from the other partitions, but not vice versa. We applied the Wilcoxon sign rank test (parsimony) in Paup* to test this asymmetry, with the critical value Bonferroni corrected for 49 consecutive comparisons among partitions. For comparison, we also applied the Shimodaira-Hasegawa test (Shimodaira and Hasegawa, 1999) on the molecular partitions under the likelihood criterion, using 1000 bootstrap pseudoreplicates.

Lineages that were topologically unstable across different analyses were also examined for potential deviation in their base frequencies. We used the base heterogeneity test as implemented in PAUP*, and performed tests on all sites, and on informative sites only, to detect confounding opposite bias in invariant sites (Buckley et al, 2001). Log-determinant distance analysis (Lockhart et al., 1994) is the only tree reconstruction method that is specifically designed to avoid attraction of unrelated lineages with similar base composition. We used this method to explore the possibility that biased base frequencies had any effect on conflicting or unstable nodes.

Character evolution.-MacClade 3.04 (Maddison and Maddison, 1992) was used for editing matrices, calculating substitution frequencies, and tracing character transformations over the most parsimonious or likely tree topologies from the combined data analyses. We applied the "chart state changes" option to count the minimum and maximum transformations in host plant use and island-mainland distribution under ACCTRAN and DELTRAN optimisations. Although accelerated character transformations may be preferred due to conforming more closely to primary homology statements (by implying reversals rather than parallel evolution of characters), there is no particular reason to expect one or the other for labile island-mainland transitions or host switching.

TABLE 2. Tree statistics from the parsimony analyses of the various data partitions (54 taxa). Informative characters for the ingroup only (Aphnnnrthnirn and Coleobothrus) are given in parenthesis, tncongrucncc refers to the number of node shifts compared to the combined data tree topology.

Ancestral character states were also reconstructed using maximum likelihood, incorporated in the computer program Discrete (Pagel, 1994,1999). The likelihood approach has the advantage (over parsimony methods) of estimating the probability of each ancestral character state, for each ancestor at a time, taking character frequencies and branch lengths (but not prior node estimates) into account.

RESULTS

Combined Analysis of All Data

The parsimony analysis of all 2688 characters (2660 aligned base pairs, 28 morphological characters; Table 2), resulted in two trees differing only in the placement of the two El Hierro specimens of A. jubae (Fig. 1). Aphanarthrum plus Coleobothrus were monophyletic in this analysis, with the latter nested within Aphanarthrum. The predefined hypothesis on reciprocal monophyly of Aphanarthrum and Coleobothrus added 55 steps to the tree length, and was firmly rejected by the Kishino-Hasegawa and Wilcoxon tests (f = 4.01, z = - 3.99, P < 0.001).

All but one of the species with multiple individuals sampled were monophyletic, and most sister species were clearly defined. Aphanarthrum glabrum was paraphyletic with respect to A. subglabrum which formed the sistergroup to the La Palma subspecies of A. glabrum (nudum). Among the higher level relationships, only the positions of A. jubae, and the internal arrangement of two Coleobothrus species with respect to A. mairei, were uncertain as shown by minimum Bremer and bootstrap supports. Aphanarthrum armatum was the basal lineage, with all species associated with succulent Euphorbia in one well supported clade (C), and all Macaronesian species breeding in arborescent euphorbs making up a highly supported sistergroup (clade D). All clades that received maximum bootstrap support had also significantly high Bremer support, with only two significantly supported nodes having bootstrap values below 80%. There was a generally high correlation between the two node support indices for the combined data.

Outgroup manipulation.-Because the Dolurgus pumilus sequences were relatively more diverged from the ingroup than the Crypturgus and Cisurgus sequences, we excluded this taxon in the remaining analyses. We also excluded Cr. borealis because this taxon had the entire Enolase partition missing. The exclusion of these taxa did not change the topology resulting from the combined parsimony analysis (or any of the molecular, nucDJSIA, or EP-1α topologies). Support for the most basal nodes (A to G) increased after exclusion of D. pumilus and Cr. borealis, further suggesting that the analyses should be improved by using more narrowly defined outgroups.

Sequence Pruning

Rather few contradicting changes occurred with increasingly pruned data matrices in each of the gene partitions (Table 3); all data combined, nucDNA, and the 16S data did not produce any contradictory nodes with increased pruning of sequences. Most changes occurred in the two genes with highest substitution rates, Eno/ase and COI (cf. Fig. 4), but these changes were unstable and sometimes reversed to the original topology in t\he next step of pruning. Overall, there were only marginal effects of pruned matrices in each gene partition, down to three sequences per species, and all bootstrap supported nodes (>50%) were recovered when only one or two sequences per species or subspecies were retained. Hence, we used two sequences per species or subspecies in all further comparative analyses of the various gene partitions.

Separste Analyses-Parsimonry

Morphology.-The analysis of morphological characters resulted in 294 equally parsimonious trees of length 55 steps (Table 2). The inclusion of all outgroups and Deropria elongata resulted in an identical topology for the ingroup (Aphanarthrum plus Coleobothrus), and confirmed the distant relationship of D. elongata in Crypturgini (Pig. 2) as previously suggested (Wood, 1986). Our morphological data did not enable us to distinguish Crypturgiis andCisiirgus, but grouped Aphanarthrum and Coleobothrus with high support (including two uniquely derived characters), with the latter genus nested within Aphanarthrum. Topological congruence with the combined data tree was further exemplified by the basal placement of A. armatiim, the close relationship between A. mairei, A. orientalis, and Coleobothrus, and the well-supported clade of Macaronesian shrub feeders (clade D). The lack of resolution in the latter clade was apparently due to the low number of informative characters at this topological level.

TABLE 3. Number of shifted interspecific nodes (total/deep only) after deletion of sequences. Only those taxa with excess sequences had sequences deleted, one per species (or subspecies) before each search. The first column shows the number of sequences retained per species. Deep nodes were defined as clades A to D (see Fig. 1).

FIGURE 2. Strict consensus of 432 MP trees of length 73 steps each (CI = 0.7I7 RI = 0.87), resulting trom the parsimony analysis of morphological characters. Deropria clongata was included in this analysis to test the monophyly of the ingroup. Bootstrap support values are shown above the nodes. Black bars marked by numbers in squares indicate characters with synapomorphic character states (except autapomorphies also shown in Deropria and Dolurgus, see Appendices 1 and 2). The line drawing illustrates Aphniinrthrum armatum.

Molecules.-The four molecular partitions contributed from 154 (16S) to 272 (EF-Iα) informative characters (Table 2), and produced partly resolved topologies for each partition, except for the fully resolved COI data (Fig. 3). Although several discrepancies appeared among the four consensus trees (Table 2), only three topological differences were supported by a bootstrap support above 70%, and only barely so (EF-Iα: A. orientalis, A. bicinctum, and A. canescens. COI: A. bicolor). Major differences of low or no support were mostly observed in the COI topology where the basal structure was strikingly absent when compared to the combined data tree. Other noteworthy differences were seen in Enolase, which differed by the changed position of the otherwise basal A. armatum and the more basal position of A. orientalis, and in EF-Iα, which differed in their basal position of A. euphorbiae plus A. affine in clade D.

The Enolase and COI genes had also significant biases in base composition across taxa, caused by nonstationarity in third positions (Table 4). EF-Iα had a similarly significant bias in third position, but all parsimony informative sites combined were not significantly biased for this gene. For EF-Iα and Enolase, the outgroups and A. armatum had generally intermediate base frequencies, and all succulent feeding taxa, and all remaining shrub feeding taxa, had divergent base frequencies that were more similar within each of these clades. The pattern was less clear in COI, and varied much more between close relatives. We further noticed that A. orientalis had frequencies more similar to the Coleobothrus clade than to the A. canariense clade for the two nuclear genes. However, a LogDet distance analysis of the nuclear DNA partitions did not remove A. orientalis from the Coleobothrus clade. One exception to the within-clade correlation in base composition was found for the two basal shrub feeding taxa in the EF-Iα analysis, A. affine and A. euphorbiae, which had EF-Iα base composition similar to the succulent taxa and the outgroups. LogDet analysis of these data placed A. bicinctum as the most basal taxon in this clade, consistent with most other analyses.

We applied the consistency index (CI) and the retention index (RI) as measures of homoplasy (Table 2). These two correlated well with each other, and also reflected the different distributions of substitution frequencies in the four genes (Fig. 4) such that those genes with the fewest substitutions per character also had the highest consistency and synapomorphy (RI) indices.

We also used the partitioned Bremer support index (PBS) to measure the relative support of each partition to the various depths of the phylogeny (Baker and DeSaIIe, 1997). All four genes contributed positively to the combined data tree, although COI contributed consistently less at higher level nodes (Fig. 5, Table 5). This trend was particularly clear when PBS was corrected for minimum tree lengths for each partition.

Separate Analyses-Maximum Likelihood

All models selected by Modeltest were derivatives of a general time reversible model, where all assumed at least two different transition rates and two different transversion rates (Table 6). Major differences were found between the nuclear and mitochondrial genes in the guanine base frequencies and guanine-thymine substitution frequencies. Also, the shape of the gammadistributed substitution rates was higher in the Enolase partition, indicating a more uniform distribution of character changes for variable sites (cf. Fig. 4). When all sites were considered (including invariable sites), all gene partitions had nearly identical shape parameters (Table 4).

Overall, the ML analyses confirmed our results from the parsimony analyses, with a few noteworthy exceptions (Fig. 6). In both of the Enolase and 16S partitions, A. canescens was moved to the basal position in clade D, and the position of A. orientalis was changed. Compared to the combined data topology, several improvements occurred in the COI partition, with monophyly of A. euphorbiae and A. affine, and of all Coleobothms species (including A. mairei). Aphanarthrum annatum moved closer to the base of the COI tree, and moved to the 'correct' basal position in the Enolase tree. However, apparently worse results also occurred, with paraphyly of A. c. canariense (EF-Iα), A. wollastoni (Enolase), and the otherwise monophyletic A. wollastoni plus A. aeoni (COI). Overall, the number of topological differences compared to the likelihood analysis of all data combined was higher than for the equivalent parsimony analyses, with 6, 4, 6, and 12 incongruent nodes found in the EF- 1α, Enolase, 16S, and COI partitions, respectively.

Dynamics of Data Combination

Parsimony.-Combination of characters from the same genome resulted in two consensus trees which only differed from the combined data tree by two (nucDNA) or three (mtDNA) topological positions (Fig. 7). All nodes supported by more than 70% bootstrap support in each of the two genomic partitions were fully congruent with each other and with the combined data tree. Bootstrap support in the nucDNA partition incorporated complementary support from each of the Enolase and EF-Iα partitions, whereas some support in the mtDNA tree was lost, particularly for clades B and C, compared to the separate 16S analysis. The combination of all molecular partitions (Fig. 8) resulted in an overall increased bootstrap support for the most basal nodes (A to C), as well as some of the more derived clades (E to G). Adding morphological characters did not influence tree topology, or node support, at all.

None of the separate or combined data partitions were incongruent when forced upon the combined data tree (Table 7). Only the COI data were incongruent on the nucDNA topology, whereas all nucDNA partitions were incongruent upon the mtDNA topology. EF-Iα had the most congruent topology with only two partitions (mtDNA, COI) being significantly incongruent. Conversely, COI was the most incongruent topology, only congruent with the mtDNA data. The two mitochondrial partitions were also less congruent than the two nuclear partitions as measured by the ILD test (COI versus I6S, P = 0.02; EF-Iα versus Enolase, P = 0.08; mtDNA versus nucDNA, P = 0.001).

FIGURE 3. Strict consensus trees resulting from the parsimony analyses of each gene partition. Bootstrap support values are shown above the nodes, transitions in habitat selection under ACCTRAN optimisation is marked below nodes (A, arborescent; S, succulent). see Table 2 for further details. Selected taxa that varied in their position across analyses of different partitions are indicated by different symbols and fonts.

TABLE 4. Summary of sequence statistics for the various molecular character partitions. Tests on deviation from base stationarily were performed on all sites, or informative sites only, and Benferroni corrected for 11 consecutive and nonoverlapping data partitions (critical P = 0.0045, underlined). The transition to transversion rate (Ti:Tv) and the shape of the Γ parameter (α) were estimated under a HKY+Γ model of substitution, with invariant sites artificially set to zero for comparisons between partitions.

FIGURE 4. Frequency distributions of character changes in the four molecular partitions. Invariable characters are not shown on the graphs. In 16S we distinguished between sites in alignment ambiguous and unambiguous regions.

Maximum Likelihood.-The combination of all molecular data resulted in one tree which only differed in two node positions c\ompared to the parsimony analysis of combined data; A. orientalis was the sister group to Coleobothrus, with Co. allouai as the sister species of A. mairei (Fig. 8). These nodes had very short internal branches in the likelihood analyses, a possible explanation for the unstable positions of these taxa.

Shimodaira-Hasegawa tests of all molecular partitions (as listed in Table 7) produced quite different results from the Wilcoxon sign rank tests, with six new comparisons significantly incongruent, whereas six comparisons that were incongruent in the parsimony analyses were insignificantly different under likelihood optimization. The most striking difference was observed in comparisons involving 16S data that were congruent with all partitions except COI. With parameters optimized for the COI data, this partition was incongruent when forced upon all other topologies, except mtDNA. Furthermore, the EF-Ia topology was also much more incongruent in the SH analyses, now incongruent with all data combined, mtDNA, Enolase and COI, but not 16S and nucDNA.

TABLE 5. Partitioned Bremer support for each of the data partitions, at interspecific, specific, and subspecific levels. PBS was also corrected by the minimum possible length (mL) of each separate partion.

Bayesinn analysis.-The implementation of a mixed model that included separate estimation of parameters for each coding position, intron, rDNA, and morphology (12 partitions) produced an identical topology to the ML (molecular data only) and Bayesian analyses under a uniform model (Fig. 8). When models were fitted to each gene and morphological data (5 partitions), three shallow nodes shifted their position within clade G (see Fig. 1), moving A. affine and A. euphorbiae to the most basal position in that clade, followed by A, jubae.

FIGURE 5. Distribution of partitioned Bremer support values in relation to the distance measured from the node tips to the root (all molecular data, clock enforced). Stippled line marks the transition between positive and negative contribution to the combined data topology.

Speciation Rates

Likelihood ratio tests showed a significant departure from a clock-like substitution rate in the combined molecular data as well as for the mtDNA data for 54 taxa. With the outgroups Cisurgus and Crypturgus pruned from the mtDNA matrix, however, a clock-like rate could not be rejected at the 1 % significance level (2InLR = 73.02, df = 49, P = 0.023). Further deletion of the three putative COI pseudogene sequences (A. aeoni [I], Co. allaudi [2]) resulted in even more clock-like rates (2InLR = 64.66, df = 46, P = 0.04). Hence, we used the mtDNA data with clock-like rates to estimate distances from the tips to the root of the similarly pruned combined data parsimony tree.

TABLE 6. Maximum-likelihood parameters estimated by Modeltest and selected by the Akaike information criterion (AIC). All base frequencies are empirical.

We plotted the cumulative number of lineages of Aphanarthrum and Coleobothrus on the relative time elapsed since the most recent common ancestor for all species of these genera (Fig. 9). This plot shows a more or less straight line through time, indicating a nearly constant speciation to extinction rate.

Geographical Patterns and Host Plant Use

Island colonization.-Within Aphanar thrum and Coleobothrus combined, only one unambiguous colonization of the islands was found using the "chart state changes" option in MacClade (accelerated and delayed transformations indifferent). Five unambiguous back colorizations to Africa were found on the MP topology (Fig. 10), and three on the ML topology. The alternative placements of A. orientalis and A. mairei did not influence these results. However, when A. armatum was forced to be monomorphic African, implying recent dispersal to the Canary islands in this species, the numbers of unambiguous back colorizations were reversed to one for each topology, with up to five possible island colorizations (one unambiguous). A maximum of seven transitions were required by the current data, at least three of which were recent intraspecific dispersals.

Likelihood estimates of ancestral distributions (African mainland versus islands) supported the general trends found in the parsimony optimizations. The ancestor of Aphanarthrum was equally likely African or island distributed. However, all remaining supraspecies nodes supported island ancestry, involving at least six back colorizations to the mainland (Fig. 11). When predominantly African distributed species were assumed originally African (excluding island distributions), the result indicated a putative minimum of two island colorizations and three back colorizations. Under all analyses, the island-mainland distributions observed in A. bicinctum and A. affine are most likely due to back colonization to the continent.

FIGURE 6. The most likely tree topologies estimated from each gene partition, using the best model selected by Modeltest (Table 6). see Figure 3 for further details.

FIGURE 7. Comparison of the most parsimonious mtDNA and nucDNA consensus trees. Bootstrap support higher than 50% is given above each node. see Table 2 and Figure 3 for further details.

Host switching.-The transition to breeding in Euphorbia occurred twice in Crypturgini (Fig. 10). Within Aphanarthrum and Coleobothrus, a single origin of breeding in succulents occurred after the transition to Euphorbia. Likelihood estimation of the ancestral character state in Aphanarthrum suggested that feeding on arborescent shrubs was more than twice as likely as feeding on succulent euphorbs (Fig. 10). The number of transitions to E. balsamifera (one to three) and E. lamarckii complex (one to two) was more uncertain, with one and three origins of E. lamarckii and E. balsamifera feeding, respectively, under ACCTRAN optimization. Likelihood estimation of ancestral states in the shrub feeding clade also supported three separate origins of E. balsamifera, with all supraspecies nodes obtaining at least twice the probability for E. lamarckii as the ancestral host in the shrub feeding clade.

TABLE 7. Percentage increase in parsimony tree lengths when optimized on competing topologies from other partitions or combinations thereof. Bold numbers mark significant difference as measured by the Wilcoxon signed rank test (Bonferroni corrected significance levels for 49 comparisons, 0.001).

Speciation modes.-Comparisons of sister subspecies, species, and higher groups identified five cases of allopatric splits in five subspecies pairs, four allopatric and three sympatric splits in seven species pairs, and three allopatric and nine sympatric splits among higher level clades (Table 8). Only one shift in host plant group was found in the sister species or subspecies comparisons, but this event occurred in allopatry (A. euphorbiae).

FIGURE 8. Comparison of the most parsimonious (left) and Bayesian/ ML (right) tree topologies resulting from the analyses of all molecular data (12-parameter mixed model and uniform model indifferent). Addition of morphological data did not change these topologies in the parsimony or mixed-model Bayesian analyses. Posterior probabilities above 95% are indicated by a star in the Bayesian topology (12 parameters). The line drawing illustrates Aphanarthrum armatum.

DISCUSSION

Phylogenetic Analyses

Data combinability.-The simultaneous analysis of five different data partitions resulted in a well-resolved tree with most nodes receiving higher branch support than in any of the separate analyses. Some of the improvement in branch support was likely due to the increased number of informative characters in the combined data matrix. Variable-length bootstrap resampling of each partition showed an overall (but modest) correlation between the number of resampled characters and branch support for nodes congruent with the combined data (except COI), a pattern observed in numerous other studies (e.g., Cummings et al., 1995; Springer et al., 2001; Koepfli and Wayne, 2003; Rokas et al., 2003). However, the choice of character sets sampled is not trivial in resolving conflicting nodes, due to apparent idiosyncrasies of single genes. Location dependent processes in the genome, including biased base frequencies as observed in the COI and Enolase genes (see Table 4), may allow individual partitions to converge towards wrong topologies (e.g., Cummings et al., 1995; Mitchell et al, 2000). Sampling characters from many independent sources will most likely reduce the negative effect of nonindependence when location specific idiosyncrasies are diluted in combined analyses. This could be a possible explanation for the largely increased support in the combined data observed for the Coleobothrus clade and the clades F and G (cf. Fig. 1) that were incongruent in some of the separate analyses. Empirical evidence so far have unequivocally shown that an equal number of characters sampled from smaller fragments throughout the genomes resolve the species tree more quickly than in a gene by gene procedure (e.g. Cummings, 1995; Rokas et al., 2003). Our sample of four gene fragments and morphological data may not yet constitute a balanced representation of the entire genome, explaining why all nodes are not yet fully resolved. Nevertheless, our results illustrate well how the process of combining multiple sources of data leads to complementary support in various hierarchical levels of the tree topology, and falls in line with several other studies which have emphasized the importance of using multiple and independent sources of phylogenetic characters (Baker and DeSaIIe, 1997; Mitchell et al., 2000; Baker et al., 2001; Cameron and Mardulyn, 2001; Cognato and Vogler, 2001; Koepfli and Wayne, 2003; Rokas et al., 2003).

TABLE 8. Summary of putative spcciation modes associated with cladogenetic events among subspecies, species and higher clades.

FIGURE 9. Lineage through time plot showing ne\arly constant speciation rate in Aphanarthrum. Splits between subspecies were included, but not splits between other populations. Length from the root was calculated as the number of mitochondrial substitutions per site, using the best mtDNA likelihood model with clock enforced.

Although topological incongruence with the combined data ranged from 1 (16S) to 14 (COI) conflicting nodes in the separate parsimony analyses, these conflicting nodes had low, if any, branch support. Hence, the analyses of separate data did not help to clarify any of the less resolved relationships from the combined analyses; neither did any of the separate gene trees suggest any strongly supported alternative hypotheses. Still, measures of incongruence (ILD, Wilcoxon sign rank test, SH test) indicated significant differences among partitions, especially between nuclear and mitochondrial partitions. A general weakness of these tests is that they do not consider the nature of incongruence, such that one rogue taxon may cause the entire data set to be incongruent. Also, the likelihood- and parsimony-based illcongruence tests produced very different results, further reducing confidence in such methods. The use of partitioned branch support (partitioned Bremer support sensu Baker, 1997), on the other hand, is a very useful tool in the localization and quantification of node conflicts. Although the significance of Bremer's (1994) node decay index is dependent on the branch length (DeBry, 2001), the way that the PBS measure is used here only considers the relative contribution (or conflict) to each node separately. The largest cost of any data partitions enforced on the combined topology was four extra steps in one Enolase node (see Fig. 5), which is nevertheless insignificant when branch length is taken into account (cf. Table 3 in DeBry, 2001). The heterogeneity measured between the nuclear and mitochondrial partitions was hardly visible in the PBS analyses, with only three nodes having a negative contribution from both mtDNA partitions (clades including A. affine, A. euphorbiae, A. jubae; A. hesperidum, A. piscatorium; all of these plus A, bicolor). However, these three nodes were characterized by very short ML branch lengths in separate and combined analyses (cf. Figs. 6, 8), and the apparent conflict may reflect random sorting events during a short period of elevated speciation rates (cf. Fig. 9).

Potential negative effects of combining 'statistically' incongruent partitions involve less resolved or less supported tree topologies. However, exclusion or sequential down weighting of the most incongruent partitions, COI (entire gene fragment or third positions only) or morphology, did not change the combined data topologies in neither of the parsimony or Bayesian analyses. Further exclusion of alignment ambiguous regions still resulted in the same topology, with node support remaining essentially the same, or slightly lower, through all these analyses. Moreover, the application of a mixed model in the Bayesian analyses did not produce different results from the uniform Bayesian analysis. Hence, the Bayesian results suggest that the effect of fitting separate models to subpartitions become less critical with an increased number of independent character sources.

Overall, our data have demonstrated how significant measures of incongruence can mislead when conflict in fact is very low (Baker et al., 2001). The incongruence observed between the separate data partitions then possibly reflects the smaller number of informative characters in each of them, which makes tree estimation more susceptible to homoplasious changes, in particular those partitions experiencing the highest substitution rates.

Phylogenetic relationships.-This study clearly demonstrated the paraphyly of Aphanarthrum with respect to Coleobothrus. The basal position of A. armatum was supported by all data sets except COI, in either parsimony or likelihood analyses, or both. COI performed much worse than any other partition, and had the highest level of homoplasy, significant nonstationarity in base composition, the fewest supported nodes, and the lowest branch support per informative character. The quite different properties (and congruence with the combined data tree) of the completely linked 16S data emphasize that the poor performance of the COI data is not genome specific, and confirm previous findings that COI is not very useful as a stand alone data source for higher level insect phylogenetics (Durando et al, 2000; Baker et al, 2001; Jordal et al, 2002; Danforth et al, 2003). However, it is also possible that our COI data included conserved mitochondrial pseudogenes integrated in the nuclear genome. Although none of these sequences contained stop codons, several recent studies have detected multiple integrations of indel-free mitochondrial pseudogenes (Bensasson et al., 2001, but see Williams and Knowlton, 2001, for an extreme example). Until more conclusive evidence is obtained in favor or disfavor of the pseudogene hypothesis, we caution against any strong interpretations of the COI data. We note, however, that exclusion of the COI data did not change the topology, or the average bootstrap support, rendering these data harmless in the face of potential paralogy.

FIGURE 10. Parsimony (branches) and likelihood (pies) reconstruction of host plant use on the combined parsimony topology. Succulents include E. canaricnsc, E. liandiensc, E. echinus, E. take; the 'E. latnnrckii' complex includes E. regis-jnbac, E. lamarckii, E. artropiirpuren, E. tuckeyana, E. piscatoria. Pies show the probability of ancestral host plant groups (succulents versus arborescent shrubs). The right hand matrix displays the geographical distribution of each species or subspecies: V, Cape Verde archipelago; M, Madeira archipelago; H, El Hierro; P, La Palma; G, La Gomera; T, Tenerife; C, Gran Canaria; F, Fuerteventura; L, Lanzarote (all Canary Islands); Aw, western Africa (Morocco, Senegal, and Gambia); A^sub e^, eastern Africa (Uganda, Kenya, Sudan, Ethiopia); X, Holarctic region.

FIGURE 11. Likelihood estimation of the geographical distribution of ancestors (mainland, black; Macaronesian islands, white). Terminal boxes indicate the current distribution. The relative proportion in each pie indicates the probability of each ancestral character state. The alternative reconstruction shown on the right assumes recent African to Macaronesian dispersal in some species, including the basal A. armatum.

The inclusion of A. mairei in the Coleobothrus clade, and the basal position of A. orientalis (ML/Bayesian) in this clade, reinforced the evidence against the monophyly of Coleobothrus. Although the two Aphanarthrum species had base compositions more similar to the Coleobothrus species than to the other Aphanarthrum species, LogDet distance analyses of the nuclear DNA data, or all molecular data, did riot shift the position of these species. Thus, biased base frequencies can not alone explain these relationships (DeBry, 2003). Also several suits of morphological characters associate A. orientalis and A. mairei, with Coleobothrus (Fig. 2). Male genitalia are nearly identical in these five species, all with a hair-pinned ventral sclerite and asymmetrical tegmen (Appendices 1,2). Consequently, Coleobothrus must be treated as a synonym of Aphanarthrum, thus rejecting Menier's proposed generic status of Coleobothrus (Menier, 1973). In the other taxa of the succulent breeders, A. canariense and A. pygmeum, genitalia are highly specialized, or reduced, showing how rapid genitalic characters can diverge in secondarily sympatric species. However, molecular data left no doubt about the monophyly of all species breeding in succulent euphorbs.

The 12 remaining Macaronesian Aphanarthrum, which all breed in shrubs, made up one strongly supported clade. The support for A. bicinctum and A. canescens as the two most basal taxa was also strong. Even though the separate analyses disagreed upon which of these taxa should be most basal, most of them, as well as the combined parsimony, likelihood and Bayesian analyses, placed A. bicinctum in that position. All 12 species are externally very similar to each other, but the shape of the transverse elytral fasciae distinguish the two basal taxa from the remaining ten. Whereas most nodes in this clade were strongly supported, two short internodes had the lowest possible branch support (two basal nodes in A. glabrum's sister group; Fig. 1). The two nodes were also those that most frequently changed under various combinations of partitions (and various optimality criteria) and we consider these unresolved until more data are collected. However, it seems a difficult task to achieve well supported resolutions for these nodes, based on the low branch support provided by all five partitions. The low resolution was also reflected in short internal branch lengths in each of the underlying gene trees (cf. Fig. 6), corresponding to a small but sudden increase in speciation rate (cf. Fig. 9). Under such circumstances, only large amounts of data would provide a sufficient number of informative characters needed to improve phylogenetic accuracy and confidence.

Given the extensive overlap and general lack of good diagnostic external characters, the strongly supported monophyly of each species was strikingly clear. With only one exception, in the paraphyleticg/ofrram complex, these species have existed as separate evolutionary entities for millions of years (all interspecific COI divergences above 10%). The deep coalescence between ecologically and morphologically similar species thus refute the hypotheses that hybridization or recent diversification act as possible confusing factors in the evolution of morphological variation.

Subspecies Formation

The paraphyly of A. glabnim with respect to A. subglabrum, and the very little genetic divergence between the two s\pecies, illustrate a case of very recent speciation. The paraphyletic signature is a predicted one under circumstances where new species originate by peripheral isolation, and does not preclude species status of the two species involved (Harrison, 1998). Some speciation events must necessarily be recent and too little time has elapsed for complete and reciprocal monophyly to evolve for every single gene tree. Population sizes in these species, as in most species of Aphanarthrum, are enormous and probably require millions of generations to obtain exclusive groups. Nevertheless, the Enolase gene strongly supported reciprocal monophyly of each of the two glabnim subspecies and A. subglabnim, suggesting that selection on certain nuclear genes may follow the assumed reproductive isolation between the two species on La Palma. More samples are needed to explore the exact nature of this incipient speciation process, and a more detailed investigation is currently under way. What this example has clearly demonstrated though is the need for sampling multiple populations and markers to discover the paraphylctic nature of species.

Sampling of multiple populations also provided information on the reality of proposed subspecies (Israelson, 1972). Overall, most species showed clear geographical divergences corresponding to the suggested subspecific classification. Although we did not manage to sample one of the A. canescens subspecies, and A. bicinctum obsitum from Gran Canaria and A. jubac tuberculatum could not be judged by the limited sample included, all other subspecies were clearly distinct and each of the gene trees showed reciprocal monophyly of A. canescens and A. canarience subspecies, whereas Enoiase and the combined data showed the same for A. bicinctum. Very deep coalescence characterized the first two subspecies splits, which were even deeper than some of the splits between morphologically distinct species (e.g., A. Mcoior versus A. hesperidum versus A. piscatorium, or A. wollasfom versus A. aeoni). Taken together with the almost straight line LTT plot (Fig. 9), the recent formation of subspecific lineages fits the general picture of a group undergoing active evolution and continuous speciation.

Factors Associated with Speciation

Host plant switching has only played a minor role in recent spcciation events in crypturgine beetles. Among the total of 24 cladogenetic events in the Aphanarthrum clade (Fig. 10), only 6 of these were associated with major changes in host plant use, significantly lower than expected from a random distribution of character states (8 to 12, average 10.3). The use of succulent euphorbs was particularly conservative, with no reversals to other host groups within Aphanarthrum. Given that this lineage has existed for a long time, and that both succulents and shrub-like euphorbs coexist in most Canary islands and in Morocco, stasis in host plant use may be interpreted as strong constraints on host switching. Switching between the various shrub-like euphorb groups was more frequent, but even the most recent of these (to E. longifofia) involved considerable sequence divergence (12.8% to 13.5% COT, uncorrected), and occurred in allopatry.

Many recent sister lineages are allopatrically distributed, suggesting that these diverged by geographical isolation on different islands. The degree of sympatry increases with increasing phylogenetic depth, however, suggestive of secondary range expansion (Barraclough et al., 1998; Barraclough and Vogler, 2000). Although allopatric differentiation may be the most universal speciation mode in island archipelagos, as has also been shown in some other insect groups (Jordan et al., 2003; Percy, 2003), several studies have reported a more even contribution (Kambysellis and Craddock, 1997; Joy and Conn, 2001), or even favored ecological specialization as the most prominent isolating factor on islands (Shaw, 2002).

Colonization Patterns

Several lines of evidence suggest that the relatively short distance between the African continent and the Atlantic islands has not been an important barrier to dispersal in this group of insects. Contrary to the single-island colorizations seen in many flightless beetles (Juan et al., 2000; Gillespie and Roderick, 2002), the high frequency of geographic interchange in Aphanarthrum beetles seems connected to their flight capacity. Recent exchange has taken place within A. mairei, A. affine, and A. bicinctum, and seems equally likely in two more species, A. armatum and A. allouai, which have previously been collected in western Africa and Fuerteventura, respectively (Fig. 10). Although the most parsimonious or likely solution of African-Macaronesian exchange suggests a Macaronesian origin of Aphanarthrum and five back colonisations to the mainland, this is dependent on the ancestral distribution of the basal A. armatum. Given the frequent interchange and high dispersal capacity in these beetles, A. armatum could have originally been restricted to Africa only, implying up to three ancient colonisations of Macaronesia. A reasonable consensus area of origin may include the Moroccan coast and the oldest (eastern) Canary islands. After all, the most basal taxon is currently found in this area, and the ancestor of at least one of the eastern African species, Co. germeauxi, most likely occurred in the same area.

Whereas the geographical origin of Aphanarthrum remains uncertain, the monophyletic clade of Macaronesian shrub breeding species implies a single common ancestor from the Canary Islands. Furthermore, the two species endemic to the Cape Verde and Madeira islands are nested within this clade, related to species endemic to a restricted number of islands. Aphanarthrum hesperidum is related to A. piscatorium and A. bicolor, a group of mainly western

Source: Systematic Biology

More News in this Category