© 2006 Society of Systematic Biologists
Molecular Phylogenetics of the Exoneurine Allodapine Bees Reveal an Ancient and Puzzling Dispersal from Africa to Australia
Edited by Thomas Buckley: Associate Editor
1 Biological Sciences, Flinders University GPO Box 2100, Adelaide, SA, 5001, Australia E-mail: Michael.Schwarz{at}flinders.edu.au
2 Evolutionary Biology Unit, South Australian Museum Adelaide South, Australia 5000, Australia
3 Centre for Evolutionary Biology and Biodiversity, The University of Adelaide South Australia 5005 Australia
 |
Abstract |
|---|
 TopNotes Abstract MethodsResultsDiscussionReferences |
|---|
Previous phylogenetic studies of the bee tribe Allodapini suggested a puzzling biogeographic problem: one of the key basal divergences involved separation of the southern African and southern Australian clades at a very early stage in allodapine evolution, but no taxa occur in the Palaearctic or Asian regions that might suggest a Laurasian dispersal route. However, these studies lacked sufficient sequence data and appropriate maximum likelihood partition models to provide reliable phylogenetic estimates and enable alternative biogeographic hypotheses to be distinguished. Using Bayesian and penalized likelihood approaches and an expanded sequence and taxon set we examine phylogenetic relationships between the Australian, African, and Malagasy groups and estimate divergence times for key nodes. We show that divergence of the three basal Australian clades (known as the exoneurines) occurred at least 25 Mya following a single colonization event, and that this group diverged from the African + Madagascan clade at least 30 Mya, but actual divergence dates are likely to be much older than these very conservative limits. The bifurcation order of the exoneurine clades was not resolved and analyses could not rule out the existence of a hard polytomy, suggesting rapid radiation after colonization of Australia. Their divergence involved major transitions in life history traits and these placed constraints on the kinds of social organization that subsequently evolved in each lineage. Early divergence between the African, Malagasy, and Australian clades presents a major puzzle for historical biogeography: node ages are too recent for Gondwanan vicariance hypotheses, but too early for Laurasian dispersal scenarios. We suggest a scenario involving island hopping across the Indian Ocean via a series of now largely submerged elements of the Kergulen Plateau and Broken Ridge provinces, both of which are known to have had subaerial formations during the Cenozoic.
Keywords: Bayesian; biogeography; dispersal; Gondwana; Kerguelen Plateau; penalized likelihood
Received November 30, 2004; Revised March 4, 2005; Accepted May 23, 2005
Many Australian animal and plant taxa are thought to have either a Gondwanan origin or else arrived by dispersal from the north after the collision of the Australian and southeast Asian plates in the mid-Miocene or later. However, some dispersals into Australia appear to be too early for the collision between the Australian and Laurasian plates, but too recent to be explained by Gondwana vicariance models. Sampson et al. (1997) and Cooper et al. (2001) suggested a possible Kerguelen landbridge, linking Africa and Madagascar to Antarctica until some 80 Mya, as an explanation of abelisaurid and ratite distributions. A number of other studies are more puzzling, because they suggest distributions that are too recent to be explained by Gondwana vicariance models, but where divergences are too ancient for the mid-Miocene interchange. Baum et al. (1998) suggested long-range transoceanic dispersal from Africa/Madagascar to Australia for the baobabs. Danforth et al.'s (2004) study on halictine bees suggested dispersal into Australia about 30 Mya, and Leys et al. (2002) suggested a dispersal of carpenter bees into Australia 30+ Mya, when there was still a significant water barrier between Australia and southern Asia (Smith et al., 1994). Barker et al. (2004) also found evidence for multiple dispersals of passerines out of Australasia beginning in the Eocene. These studies suggest that traditional models, involving Gondwanan vicariance or Australian-Laurasian interchange in the mid-Miocene, may not account for some major biogeographic patterns and highlight more general patterns suggesting that dispersal events in the southern hemisphere may have been much greater than previously thought (Sanmartín and Ronquist, 2004).
There is a very clear need to investigate divergence times for taxa that span major biogeographic regions in the southern hemisphere but where Gondwanan vicariance hypotheses are not applicable. The recent development of DNA-based analytic procedures (e.g., Sanderson, 2002; Thorne et al., 1998; Thorne and Kishino, 2002) make such studies possible, and these have had major impacts on our understanding of historical processes underlying current biogeographic patterns (e.g., Richardson et al., 2001; Cooper et al., 2001).
The allodapine bees are largely restricted to sub-Saharan Africa, southern and southeast Asia, and Australia. The greatest abundance and diversity of these bees occurs in southern Africa (seven genera) and southern Australia (five genera). There is only one rare genus, Exoneuridia, which occurs in the southern Palearctic, and the only genus in southern Asia is Braunsapis, which occurs throughout the southern Old World. Braunsapis is a relatively recently derived genus among allodapines, with a likely origin in Africa, followed by dispersal into Asia and then from Asia into Australia in the middle-late Miocene (Schwarz et al., 2004).
The origin of the Australian "exoneurine" genera (Exoneura, Exoneurella, Brevineura, and the parasitic genus Inquilina) has been problematic. This group forms one of the most basal allodapine clades (Schwarz et al., 2003; Bull et al., 2003) and, with the exception of one arid-zone species, Exoneurella eremophila (Houston 1976), its species are entirely limited to southern semi-arid and temperate regions of Australia (Michener, 1965). The nonparasitic exoneurine genera are interesting in that each of the genera differ from each other in major life history and social features, yet these are largely invariant within genera, suggesting phylogenetic inertia following initial generic-level divergences (Tierney et al., 2000). Bull et al. (2003) attempted to resolve the phylogeny of the exoneurine genera and provide a hypothesis for how they arrived in southern Australia so early on in the allodapine radiation. Their analyses were based on 1049 nucleotides from two mitochondrial genes, COI and Cyt b, and 457 nucleotides from the F1 copy of the nuclear gene EF-1
, and they used maximum parsimony and maximum likelihood methods. Weak to moderate support for monophyly of the exoneurine genera was found, but bifurcation order among the nonparasitic genera could not be resolved. They suggested that this may have been due to very rapid diversification after an early colonization of Australia, so that there was little opportunity for phylogenetically informative signals to be "trapped" during divergences of the genera. Such rapid divergences would lead to an effectively "hard" polytomy. Alternatively, it is possible that inability to resolve relationships among this group was due to a combination of small sequence lengths, extreme AT bias for mitochondrial genes, and the inability of current software to search for trees while fitting separate models to gene partitions with clearly different evolutionary dynamics.
Bull et al. (2003) also discussed the biogeographic origin of the exoneurines and presented three hypotheses: (i) dispersal from Africa via southern Asia, followed by extinction of remnant species in the Arabian Peninsula, Asia, and northern Australia; (ii) a Gondwana vicariance explanation; and (iii) a southern dispersal event over the Indian Ocean, perhaps via island stepping stones from the now largely submerged Kerguelen Plateau. A Gondwanan origin seems very unlikely because it would require that allodapine divergences were occurring shortly after the origin of bees per se, something that seems unlikely given the derived nature of allodapines in the family Apidae (Engel, 2001; Michener, 2001), which is itself a nonbasal bee clade. Bull et al. (2003) were unable to assess the two remaining scenarios because they were not able to estimate the relative times of divergence events in their phylogeny and were therefore unable to link the divergence between the exoneurines and the African clades to either the collision between Australia and Laurasia or submergence of the Kerguelen Plateau.
Several developments since 1Bull et al.'s (2003) study now allow us to reexamine the phylogeny of the exoneurine genera more effectively: (i) use of new primers has allowed us to extend the sequence length of COI by 662 nucleotides as well as sequence a potentially more useful nuclear gene, the F2 copy of EF-1 (Danforth et al., 2004); (ii) we have increased the number of sequenced exoneurine taxa; and (iii) Bayesian software allows separate modeling of gene partitions concurrent with exploring tree space, and at the same time allows extensive variation in model parameters for each partition. Schwarz et al. (2004) have recently shown that, for allodapine bees, the inability of current maximum likelihood (ML) software to separately model partitions with very different base compositions and evolutionary dynamics while simultaneously searching tree space can lead to highly implausible topologies even for the recently derived allodapine genus Braunsapis. However, both ML analyses that excluded 3rd mitochondrial codon positions and Bayesian analyses that included this partition but fitted separate models to partitions both lead to plausible topologies, but Bayesian analyses avoided the problem of disregarding data that was useful for resolving recent divergences.
There are several compelling reasons for attempting to resolve the phylogeny of the exoneurines. First, identifying bifurcation order would allow transitions in key social and life history traits to be mapped and compared with possible causal factors. Alternatively, finding that generic divergences effectively form a hard polytomy would suggest explosive radiation following colonization of a new continent, accompanied by major radiation in key life history and social traits. Lastly, being able to estimate approximate times for some major divergences is important because it may help address the question of how early allodapine divergences might fit with alternative biogeographic scenarios.
|
Methods |
|---|
|
TopNotesAbstractMethodsResultsDiscussionReferences |
|---|
Taxa Used
We used two halictid bees, Lasioglossum lanarium (AF103956 [GenBank] , AF264793 [GenBank] ) and Agapostemon tyleri, (AF102835 [GenBank] , AF140320 [GenBank] ) from a study by Danforth et al. (2004) as the outgroup, and included members of two other apid tribes, Apis mellifera (Apini) (U72278 [GenBank] , AJ581105 [GenBank] , AF015267 [GenBank] ) and three species of Ceratina (Ceratinini) along with our allodapine taxa. The two halictid taxa and Apis mellifera were included so that in dating analyses (see below) a minimum age could be set for the root node for the Apidae, namely the divergence between the lineage leading to the Xylocopinae and the lineage leading to all other apids (Engel, 2001). The three Ceratina species were included to help resolve basal bifurcations in the allodapines and were chosen because the Ceratinini is the extant sister tribe to the Allodapini.
Most allodapine and ceratinine taxa in our analyses have been represented in previous studies by Schwarz et al. (2003, 2004, 2005) and Bull et al. (2003), and collecting localities and GenBank accession numbers are given in those publications. Additional sequences, including those for newly added taxa, used in this study have GenBank accession numbers DQ149653 [GenBank] to DQ149724. For this study we included three taxa additional to previous studies, comprising two undescribed Exoneura species and one undescribed Inquilina species. The Inquilina species was found inhabiting nests of an undescribed Exoneura species in the Adelaide Hills region, South Australia, and we refer to these species as Exoneura Adelaide sp. and Inquilina Adelaide sp., respectively. The other unidentified Exoneura species was collected from Port Arthur in Tasmania and is referred to here as Exoneura Tasmania sp. It is possible that this species is a Tasmanian form of the widespread species E. angophorae. The species-level taxonomy of allodapines is poorly known, with many undescribed species throughout Africa, Asia, and Australia, and with little prospect for taxonomic revision in the near future.
DNA Extraction, Amplification, and Sequencing Methods
Specimen handling and DNA extraction techniques are described by Schwarz et al. (2003, 2005). All DNA extractions were from mesosomal or metasomal sections of ethanol-preserved specimens following a modified protocol of Doyle and Doyle's (1990) CTAB method or using DNAzol Genomic DNA Isolation Reagent (Chomczynski et al., 1997, 1998). Ten microliters of proteinase K (20 mg/ml) was added to the extraction prior to incubation (CTAB: 55°C for 2 h; DNAzol: 37°C overnight). DNA pellets were resuspended in 50 to 100 µ l of TE buffer and stored frozen. Working solutions were subsequently diluted in distilled H2O (1:5 or 1:10), depending on initial concentration of extraction.
One nuclear and two mitochondrial gene regions were amplified and sequenced (bidirectionally). The mitochondrial regions were from the protein-coding genes cytochrome oxidase b (Cyt b; 428 bp sequenced) and cytochrome oxidase I (CO1; 1279 bp) and the nuclear exon region from the F2 copy of elongation factor 1 (EF-1 F2; 772 bp). Earlier studies by Schwarz et al. (2003) and Bull et al. (2003) used the F1 copy of EF-1, but the number of parsimony-informative sites in that gene fragment was very low, and following Danforth et al. (2004) we use here the F2 copy instead. The primers used for PCR amplification of the Cyt b and COI regions and amplification conditions for these regions are given in Schwarz et al. (2004). EF-1 occurs as two copies, EF-1 F1 and EF-1 F2, in bees, which are expressed at different stages of development (Danforth and Ji, 1998). The primers used for PCR amplification of the EF-1 F2 region included the F2-specific forward primer (HaF2For1: 5'-G GGY AAA GGW TCC TTC AAR TAT GC-3') designed by Danforth (personal communication) and the F2-specific reverse primer (F2-Rev1) designed by Danforth and Ji (2001) for halactid bee species. Amplification conditions for the EF-1 F2 region included an initial hot start of 94°C for 9 min, followed by 35 cycles of denaturation at 94°C for 30 s; annealing at 54°C for 45 s, extension at 72°C for 1 min; and then a final extension step of 72°C for 6 min.
PCR products were purified using Ultraclean PCR Clean-up columns (MO BIO Laboratories, Inc.), and 
50 ng of product was sequenced in 20-µl reaction volumes using the Big Dye Sequencing Ready Reaction kit Version 3.1 (Applied Biosystems), with the original PCR primers used as sequencing primers. Reaction products were purified by isopropanol precipitation and sequenced on a capillary DNA sequencer. Forward and reverse sequences were compared for each gene fragment, and sequences were manually edited and aligned using SeqEd 1.03 (Applied Biosystems). The intron region of EF-1 F2 was excluded from analyses because large sections of sequence were unalignable.
Phylogenetic Methods
Maximum parsimony
We undertook two maximum parsimony (MP) analyses, one with all nucleotides included and another with mitochondrial 3nts excluded. We used a heuristic search with 50 random sequence stepwise additions, holding five trees at each step and with TBR branch swapping. Support was estimated for nodes using the same procedure with 500 bootstrap pseudoreplicates.
Bayesian inference
Schwarz et al. (2004) and Tierney (2004) found that for allodapines partitioned, Bayesian analyses produced much more plausible topologies than single-model maximum-likelihood analyses, probably because they permit much more feasible models to be fitted to partitions while allowing simultaneous tree searching. We used a similar protocol here, with a general time-reversible model (Nst = 6) applied to each of the six partitions. Felsenstein and Churchill (1996) have argued that rates of change among neighboring sites may be a more realistic assumption than completely independent rates of change. We checked for possible autocorrelation for the three codon positions of COI + Cyt b and EF-1, using the number of different nucleotides at each site, across our taxa set as the dependent variable and with correlations calculated for lags of 1 to 30. The data set was generated using PAUP* (Swofford, 1999) and autocorrelation was carried out using SPSS v11 for MacOSX. This suggested autocorrelations for both 1nt and 2nts for both COI + Cyt b and EF-1. However, subsequent MrBayes runs using an autodiscrete gamma (adgamma) setting for the 1nt and 2nt mitochondrial and nuclear partitions indicated no significant autocorrelation, with correlation values close to zero. We therefore used a gamma-shaped distribution with a proportion of invariant sites (invgamma) for subsequent analyses. We allowed partition models to vary by unlinking gamma shapes, transition matrices, the proportions of invariable sites, base compositions, and correlations for the 1nd and 2nt partitions. We otherwise used the MrBayes default priors. MCMC chains were run for 3 x 106 generations, trees were sampled every 500th generation, and stationarity was assessed by graphing log likelihood values against generation number. The number of burn-in generations was much greater than when apparent stationarity was reached. To check that Bayesian runs were converging on similar parameters we carried out three separate runs and compared parameter estimates, including topologies, for consistency.
Hard versus soft polytomy
Previous studies (Schwarz et al., 2003; Bull et al., 2003) were unable to resolve bifurcation order among the three main exoneurine lineages and phylogenetic analyses from the study here also do not produce well-supported resolution of these nodes (see Results). We investigated the possibility that the continued lack of resolution, despite the increased sequence data in our current study, may reflect a hard polytomy using the method outlined by Slowinski (2001). Slowinski's approach is based on the expectation that a hard polytomy would be reflected by branch lengths that are not statistically different from zero, whereas soft polytomies should be accompanied by non-zero branch lengths.
Slowinski's method requires that branch lengths be assessed using log-likelihood-ratio tests. Ideally, such tests would be carried out for single-model ML analyses applied to the entire data set, thereby making fullest use of the available nucleotide data. However, Schwarz et al. (2004) have shown that single models applied to all codon positions for mitochondrial data give highly unlikely results because of the very different evolutionary dynamics and saturation problems for mitochondrial codon positions. Furthermore, our analyses indicate quite different dynamics for the EF-1 codon positions as well (see below). We therefore carried out the following procedure.
First, we fitted ML models to six partitions, comprising codon positions for both COI and EF-1. Models were fitted with ModelTest 3.06 applied to the consensus Bayesian tree (see above and Results). Because MP and Bayesian analyses were equivocal regarding bifurcation orders of Exoneurella, Exoneura + Inquilina, and Brevineura, and there are three possibilities for ordering these clades other than a hard polytomy, we tested all three alternatives for zero lengths in the most basal branches (i.e., the branch that prevents collapse into a polytomy). For tests of these critical basal branches, we employed phylogenies (based on the Bayesian consensus tree) that differed only in bifurcation order of these principal exoneurine clades, and for each analysis branch lengths were fitted separately for each partition. Branch length was then assessed for each partition for each of the alternative non-polytomous topologies. Following Austin et al. (2003) we used Goldman and Whelan's (2001: Table 1) corrected values for these tests because the appropriate degrees of freedom should comprise a 50:50 mixture of 0 and 1 degrees of freedom rather than a single degree of freedom used in PAUP*. However, Goldman and Whelan's table does not allow precise estimate of significance for each comparison and this can make it difficult to employ adjustments of alpha levels to reduce the likelihood of type I error rates. We therefore considered possible type I errors in a case by case manner.
|
Dating methods
We aimed to test several possible biogeographic scenarios for the origin of the Australian exoneurines: (i) a vicariance origin involving the separation of Africa and Madagascar from Australia, requiring a divergence of the exoneurines from other allodapines in the early Cretaceous; (ii) dispersal from Africa to Australia via Asia, requiring a colonization of Australia during or after the mid-Miocene collision between the Australian and Laurasian masses; and (iii) dispersal across the Indian Ocean via a Kerguelen land bridge or remnant island stepping stones of this, requiring colonization of Australia prior to the early Miocene. In order to assess these alternative scenarios we estimated dates for key nodes involving divergences between African, Malagasy, Asian, and Australian clades. We used two methods to estimate relative times: Sanderson's (2002) penalized likelihood and Thorne et al.'s (1998) Bayesian approach.
There are no known allodapine fossils, so that trees could not be calibrated internally. However, the fossil tribe Boreallodapini from Baltic amber is reliably dated at 45 Mya and there is strong morphological evidence that this tribe forms a sister group to the Allodapini and that the extant tribe Ceratinini forms the next most basal clade (Engel, 2001). The Baltic amber boreallodapine fossils comprise three to four species and, given that preservation as an amber fossil is likely to be a very rare event, it seems likely that the Boreallodapini was already speciose by 45 Mya. We were therefore able to use the Boreallodapini to provide a minimum-age calibration point for divergence between the Ceratinini and the lineage leading to both the Allodapini and Boreallodapini. In our analyses we used a minimum date of 45 My for this split but note that this is a highly conservative minimum given that the Boreallodapini is morphologically very distinct and apparently speciose.
We also used deeper node to calibrate our tree, namely the divergence between clades leading to the xylocopines and the corbiculates. This is the oldest divergence among extant apid clades (Engel, 2001). The oldest reliably identified and dated bee fossil is a Maastrichtian amber fossil of a meliponine corbiculate, Cretotrigona, dated at 65 to 70 Mya (Grimaldi, 1999; Engel, 2001). The Apini diverged from the Meliponini before this date (Engel, 2001) and the existence of plants that are today tightly associated with the Apinae is known from deposits dated at 90 Mya (Crepet and Nixon, 1998). Because the Xylocopinae show an early divergence from the clade leading to corbiculates, including the Apinae (Engel, 2001), we set a minimum age for the root of our phylogeny (node separating Apis mellifera from the Ceratinini + Allodapini of 90 Mya), but note that this date, for the earliest divergence within the Apidae, is likely to be a conservative minimum value.
Lastly, Sanderson's penalized likelihood method requires that the root node for the clade of interest is obtained by including an outgroup when calculating phylograms. This outgroup is then pruned prior to estimating relative divergence times. We therefore included two halictid species, Agapostemon tyleri and Lasioglossum lanarium, to obtain the root node for the Apidae.
Penalized likelihood
Sanderson's (2002) penalized likelihood (PL) method is a semiparametric technique that allows variation in rates along branches using a roughness penalty. We used Sanderson's r8s version 1.5 to analyze our data. Smoothing values were estimated using cross-validation and the truncated Newton method was used to find optima for the objective function. The outgroup (the two halictid species) was pruned prior to rate and divergence time estimations. Sanderson's PL method presents some problems for our data because it allows input of only one gamma-shape estimate and assumption of equal rates is not indicated by Bayesian results (see below). We therefore estimated a gamma shape for the combined data using PAUP* commands enabled by ModelTest, but we point out that our Bayesian analyses indicated quite different gamma shapes for the various partitions when models were fitted separately.
A PL approach also entails some problems for estimating measures of confidence for node ages. These are often estimated by bootstrapping sequence data, fitting ML models to each of the bootstrap pseudoreplicates, and then using the resulting phylograms as separate tree inputs for r8s. However, an earlier study of allodapines (Schwarz et al., 2004) and results from our Bayesian analysis (see below) indicate that fitting a single model to the codon position/gene fragment partitions in our study is likely to lead to spurious results. Although we could have used a bootstrap approach by treating each partition separately, this would have led to comparatively small data sets for each model, exacerbated by the fact that bootstrapped replicates use only a proportion of available data and multiple partition-specific estimates raise type II error problems.
Instead, we used the following procedure. We filtered the post-burn-in trees from a Bayesian analysis using the Bayesian consensus tree as a constraint tree in PAUP*. This led to a subset of trees, each with identical topology but not identical branch lengths, and these were then used in r8s to estimate mean node ages and the central 95% of the age distributions. These 95% limits are therefore based only on those Bayesian MCMC post-burn-in generation trees that are compatible with the constraint topology. However, there is a potential problem regarding the independence of trees in Bayesian analyses and this could impact on estimating confidence for specific parameters. Bayesian MCMC runs involve perturbations of a few parameters at each step and this means that the parameter locations from one generation to the next are not independent (Holder and Lewis, 2003). Most Bayesian studies attempt to avoid this lack of independence by sampling a small percentage of the generations (typically one in every hundred) but do not check for dependence in the parameters of interest. For our study we sampled one in 500 generations and then checked for dependence in the parameters of interest (key node ages) using autocorrelation. We used lag times of 1 to 5 sequential trees from the Bayesian saved trees, corresponding to 500 to 2500 sampled generations in steps of 500, and autocorrelation was assessed using SPSS version 11 for Macintosh OSX.
Bayesian dating using Multidivtime
We also used a probabilistic approach for dating relative divergence times (Thorne et al., 1998) implemented with the Estbranches and Multidivtime software package (Thorne and Kishino, 2002). Posterior distributions of branch rates and divergence times are derived with an MCMC procedure and the method allows multiple substitution models for different partitions, something that we regarded as highly desirable given the heterogeneity in evolutionary dynamics of the gene and codon partitions for other studies of allodapine bees (Schwarz et al., 2004; see Results). We used six partitions in this analysis, comprising the three codon positions for both COI and EF-1. We excluded Cyt b data because these were entirely lacking for the outgroup. Models for each partition were estimated using PAML and fitted to the partitioned Bayesian tree (see above and Results). Three taxa (Ceratina (Neoceratina) sp., Ceratina minutula and Brevineura ploratula) were missing EF-1 data but Multidivtime is able to incorporate taxa with some missing partitions (Thorne et al., 1998). The same calibration points were used for the PL method (above). MCMC chains were sampled at every 100th generation up to 1 x 106 generation, with a burn-in of 100,000. Confidence in node ages was assessed using 95% credibility intervals. We investigated robustness of the chronogram to varying priors by multiple runs where we varied the evolutionary rate at the root (values of 0.001, 0.002, and 0.003) and "minab" (0.8, 1.0, and 1.2). Given a likely minimum age of 90+ Mya for divergence between the Apidae and Halictidae, the prior for the mean of the root node (divergence between Apidae and Halictidae) was set at 95 (My) and its standard deviation was set at 10. Thorne and Kishino (2002) recommend that the prior for brownmean (and its standard deviation) be set at a value so that when multiplied by the approximate time from the root to the present the product is between 1 and 2, and we therefore set brownmean at 0.015 and set its standard deviation at the same value.
Biogeographic Analysis
We used Ronquist's (1996, 1997) DIVA 1.1 program to infer likely vicariance and dispersal events that shaped the current distribution of allodapines. To simplify analysis we used genera instead of species and recorded genera as being present in any of six regions: Africa, Madagascar, southern Asia, the Middle East, northern Australia (Torresian and northern Eyrean regions), and southern Australia (Bassian region). Only two allodapine genera occur in more than one of these areas. Macrogalea occurs in Africa and Madagascar, but molecular phylogenetic analyses have shown that the Madagascan fauna has resulted from a single, very recent (approximately 2 Mya) colonization event and is nested within the African clades (Tierney, 2004), which occur from west Africa as far south as Namibia across to east Africa as far north as Ethiopia (Michener, 2001). Because Macrogalea is holophyletic (Tierney, 2004), we followed Ronquist's (1996) recommendation and coded Macrogalea as being limited to its presumed area of origin, Africa. Braunsapis also occurs in all of the above regions except the Middle East and its distribution was scored accordingly. The sister group to the allodapines, Ceratinini, has a worldwide distribution and occurs in all of the regions where allodapines are now found.
DIVA allows one to specify the maximum number of regions that could be simultaneously occupied by hypothetical ancestral lineages, ranging from two to the number of regions currently occupied by terminal taxa (Ronquist, 1996). The value that is set will depend on assumptions about how widespread ancestral populations could have been. We carried out separate analyses for each of the five possible values that this parameter could take.
|
Results |
|---|
|
TopNotesAbstractMethodsResultsDiscussionReferences |
|---|
We recovered 1282, 427, and 774 base pairs of COI, Cyt b and the exon region of EF-1
respectively, of which 406, 151, and 213 base pairs were parsimony informative.
Maximum Parsimony
The bootstrap consensus tree for the analysis excluding 3rd mitochondrial positions, showing nodes compatible with a 50% majority rule, is given in Figure 1 with bootstrap support values indicated. This figure indicates strong support (100%) for monophyly of the allodapines and a sister-group relationship between Allodapini and Ceratinini, as expected (Engel, 2001). It also provides strong support for a sister-group relationship between Macrogalea and all other allodapines combined, consistent with previous studies (Schwarz et al., 2003; Bull et al., 2003). There was moderate support (85%) for monophyly of the exoneurines, but very low support (< 68%) for bifurcation order among any genera in the non-Macrogalea clade. The bootstrap analysis for the total data set including 3rd mitochrondrial positions (not shown here) showed an identical topology at the generic level except for the placement of Exoneuridia, which was recovered as the next most basal clade to Macrogalea, rather than part of the clade containing Allodapula and Compsomelissa. It is very likely that this difference is due to long-branch attraction. Bayesian analyses (see below) indicated a longer branch leading to Exoneuridia than for any other ingroup terminal branch and this was also true for both the MP analyses (phylograms not shown here). Schwarz et al. (2004) showed that for the allodapine genus Braunsapis long branch attraction leads to some very unlikely topological features when 3rd mitochrondrial positions were included in MP analyses, but that these features were not supported when 3rd positions were removed, or when codon positions were modeled separately in Bayesian analysis.
Bayesian Analyses
All three runs converged on highly similar model parameters and consensus phylograms and a randomly chosen phylogram is given in Figure 2, along with posterior probability (PP) values where these were less than 100%. Partition model parameters (Q matrix elements, gamma shapes, proportion of invariant sites) were similar to those reported by Schwarz et al. (2004) for another allodapine study and are not shown here. This concensus phylogram is largely topologically concordant with the MP bootstrap analysis, except for bifurcation order among some Exoneura species. PP values are mostly higher than for the bootstrap analysis, with 97% PP support for monophyly of the exoneurines, but some nodes still received relatively low support, with a PP value of 75% for bifurcation order among the major exoneurine clades (grouping Exoneura with Brevineura) and a value of 68% for bifurcation order among the Allodapula representatives (grouping Allodapula with Compsomelissa). The terminal branch leading to Exoneuridia hakkariensis was longer than for any other ingroup terminal branch, not surprising given that we had only one species of this rare genus and were therefore unable to subdivide this lineage.
|
|
|
Hard versus Soft Polytomy for the Three Major Exoneurine Clades
Of the 18 tests of nonzero branch lengths, two comparisons resulted in significance values < 0.05, disregarding any adjustment of alpha levels for type I error (Table 1). One comparison involved a partition (EF-1
1nt) that showed a significant non-zero branch length for the topology (Exoneurella + (Exoneura + Brevineura)), which is supported by both Bayesian and MP analyses. However, another partition of the same gene (EF-1 3nt) also gave a significant nonzero branch, which supports an alternative bifurcation (Exoneura + (Exoneurella + Brevineura)). Both results would still be significant if alpha levels are adjusted to 0.01 and only one test would be significant if levels were set at 0.005. Given these findings, we regard this evidence for non-zero branch lengths as being both very weak and contradictory.
Relative Divergence Times
Penalized likelihood
The cross-validation procedure indicated a smoothing value of 1.45 and all runs gave highly similar chronograms. A chronogram from a randomly chosen run is given in Figure 3 and estimated divergence times for the key nodes of interest are given in Table 2. Five hundred thirty-three post-burn-in trees were found that were consistent with the Bayesian consensus tree. The 95% central distribution limits, calculated by removing 2.5% of cases from the tails, is also given in Table 2.
|
Autocorrelations of age for each of the five key nodes was examined for lag intervals of 1 to 5 (corresponding to 500 to 2500 sampled Bayesian generations). The P value of the most significant of the 25 correlation coefficients that were calculated was P = 0.073. Given the number of tests undertaken (25) with the expected effect on type I error rates, this does not suggest a lack of independence for node age estimates between successively sampled generations. That is, node age estimates based on each sampled Bayesian generation were effectively independent across successively sampled Bayesian generations.
Bayesian dating analysis
Despite variation in priors, all analyses returned chronograms (Fig. 4) that were virtually identical except for (i) minor differences in estimated absolute times, and (ii) values of minab lower than 0.07, which led to analyses that failed to converge. Recommendations for setting minab are for values of 1.0 or slightly above or below this value (Rutschmann, 2004). Variation in node ages were greatest for the root node (divergence between the lineages leading to the corbiculates and the Xylocopinae) but this only varied from 89 to 93 Mya. Estimated node ages and 95% credibility intervals are given in Table 2 for an analysis where minab was set to 0.8 and the rate at the root node was set at 0.0012.
The two different methods for producing chronograms gave quite different estimated ages for all of the key nodes, with point estimates from Bayesian dating analyses exceeding the PL values by about 10 My. We cannot be sure why these discrepancies occurred, but two possibilities seem likely. First, the Bayesian approach allows the root node (divergence between corbiculates and xylocopines) to exceed 90 Mya, whereas this node age is fixed for the PL analyses. Second, the Bayesian approach is influenced by branch lengths connecting the outgroup to the ingroup, whereas the outgroup is pruned for PL. Divergence between the outgroup (short-tongued bees) and the ingroup (long-tongued bees) probably occurred shortly after the origin of bees per se (Engel, 2001) and patterns of molecular evolutionary change may have been very different during this initial radiation, than more recent divergences. However, we note that 95% minimum node age estimates are very similar for both the Bayesian and PL methods, differing by a maximum of 2 My.
|
Biogeographic Analyses
Six analyses were performed, varying the maximum number of regions (henceforth "maxareas") that could be occupied simultaneously by an ancestral lineage. Because the Ceratinini, extant sister tribe to the Allodapini, have a worldwide distribution, the inferred distributions for the root node connecting these tribes contained nearly all possible combinations of the regions currently occupied by allodapines and are therefore not useful for inferring ancestral distributions of allodapines. However, the optimized distributions for internal nodes were highly informative and are shown in Figure 5a. All solutions (maxareas ranging from 2 to 6) for the root allodapine node indicated an African distribution only. For all four analyses where maxareas > 2, all solutions were identical and contained two possible alternatives for the common ancestor of the Australian and non-Macrogalea allodapines, namely Africa + southern Australia and Africa + Madagascar + southern Australia. When the maximum span was set at 2, this internode had only a single solution, Africa +southern Australia. Both solutions imply vicariance events involving Africa and Madagascar with one also involving Australia. However, such vicariance events are impossible because the oldest possible dates for divergence between the African and Malagasy fauna based on our Bayesian dating analysis (95% upper credibility limit of 60 Mya) and PL method (upper 95% central distribution limit of 44 Mya) are very much younger than when the Madagascan/Indian block began moving away from Africa about 165 Mya, and when it finished moving away about 120 Mya (Rabinowitz et al., 1983). Because vicariance is therefore clearly precluded here, these inferred ancestral distributions must instead indicate dispersal events from one region to another. Yuan et al. (2005) similarly found that impossible vicariance events were inferred by DIVA and noted that this was because DIVA regarded wide distribution and vicariance as a default optimization, but one that was clearly precluded by their dating analyses. Because our DIVA analyses indicated only a single region for the root allodapine node, namely Africa, then both the constrained (maxareas = 2) and less constrained (maxareas = 3–6) analyses must involve two early dispersal events, one from Africa to Madagascar and one from Africa to Australia.
|
Discussion |
|---|
|
TopNotesAbstractMethodsResultsDiscussionReferences |
|---|
Monophyly of the Australian Exoneurines
Michener (1977) and Reyes (1998) both used morphological traits to argue that the Australian exoneurine genera formed a monophyletic group. However, such cladistic arguments, based on relatively small data suites, can be, and occasionally have been, misleading in allodapines because of problems in codification and information content relative to the number of taxa and because of the possibility of convergence (Schwarz et al., 2003). Bull et al. (2003) also examined phylogenetics of the exoneurines using maximum parsimony and maximum likelihood analyses of DNA sequence data from three gene fragments, comprising a total of 1506 bp. That study did not provide strong support for monophyly of the exoneurines, with bootstrap support ranging from < 50% to 62% (Bull et al., 2003), but the available sequence lengths may have been insufficient if early divergences in the allodapines were both rapid and ancient. Our analyses, based on a larger sequence data set and utilizing the more informative F2 copy of EF-1
, do indicate monophyly of the Australian exoneurines with relatively strong MP bootstrap support (85%) and a high Bayesian PP (97%). Our findings therefore suggest a single colonization event of Australia by the ancestor of the extant exoneurines. Support for Macrogalea being the sister group to all other allodapines was high for both MP and Bayesian analyses (92% and 100%, respectively), but support for the non-Macrogalea African genera forming a monophyletic group is very weak, with bootstrap support of only 25% and PP support of 89%. This means that there is not strong support for a bifurcation between Australian exoneurines and a combined African/Malagasy clade, and that either of the African, Malagasy, or Australian taxa could potentially comprise the sister group to Macrogalea.
The lack of strong support for bifurcation order in these more basal nodes of the non-Macrogalea clade is interesting. The non-Macrogalea genera are all characterized by presence of larval appendages, a presumed adaptation to living in nests without physical barriers between larvae (Michener, 1977). Macrogalea is the only lineage that exhibits the plesiomorphic absence of larval appendages and is notable in its lack of deep intraclade divergences despite its relative age. The deep divergences in the non-Macrogalea clade contrast strongly with this. It is possible that the almost explosive deep radiation in this latter clade could be associated with evolutionary elaborations following the origin of larval appendages and highly modified setae, but the most basal divergences are also associated with biogeographical disjunctions among Australia, Madagascar, and Africa. Of course, hypotheses that this early radiation of the major non-Macrogalea clades are due to either biogeographical events (dispersal or vicariance) or developments in larval morphology need not be mutually exclusive.
|
) where the inferred distribution was either AF+MD+SA or AF+SA. Because the inferred vicariance between the African and Malagasy or southern Australian groups requires an origin of allodapines prior to the origin of their family, the Apidae, both solutions must imply two dispersal events, one from Africa to Madagascar, and one from Africa to southern Australia. (b) Orthographic azithumal map centered on the southern Indian ocean at 30 Mya. Plates, terranes, and oceanic plateaus are indicated by light shading and current shorelines (0 Mya) projected onto the past plate reconstructions are indicated with black lines. Not all shaded areas were necessarily above sea level 30 Mya. Dark shaded areas indicate the current distribution of the allodapines, excluding the derived genera Exoneuridia and Braunsapis. The two arrows represent alternative hypothetical dispersal pathways, one via Laurasia (top) and the other via elements in the southern Indian Ocean. Regions included in the DIVA analysis are indicated as: AF = Africa; MD = Madagascar; ME = Middle East; AS = Asia; NA = northern Australia; and SA = southern Australia. Map reconstruction used the methods of Hard versus Soft Polytomy for the Exoneurines
Bull et al. (2003) were unable to resolve bifurcation order among the three main exoneurine lineages, but could not discern whether this reflected a hard or soft polytomy. Our expanded data set (2483 versus 1506 nucleotides), which increased the number of parsimony-informative sites by over 50% (770 versus 512 sites), did not lead to well-supported resolution of bifurcation order, suggesting a hard polytomy or at least very rapid early divergence. Log-likelihood-ratio tests indicate that although there may be some support for Exoneurella being the sister group to the remaining exoneurines, this support comes from only one of the six partitions and slightly weaker support for an alternative bifurcation order came from another partition. Although the expanded data set and analyses presented here do not allow us to demonstrate, without doubt, that a hard polytomy is implicated, they do suggest that radiation following colonization of Australia was very rapid.
Rapid divergence of lineages following colonization of a major new habitat, as appeared to happen once allodapines reached Australia, is not surprising given the diversity of new niches that would be encountered. However, for the exoneurine allodapines, such divergence is very interesting because the three main clades differ strongly in life-history traits that have major impacts on social evolution (Tierney et al., 2000). These traits include voltinism, egg-production schedules, and brood development patterns, all of which determine the kinds of sociality that can be expressed and the nature of roles available to potential worker-like castes. Importantly, life history traits in these lineages show strong signs of phylogenetic inertia after their divergence (Tierney et al., 2000). Our results therefore suggest that important differences in sociality could be influenced by rapid and major changes in life history traits associated with colonization of a new habitat.
Origin of the Exoneurines
Our DIVA analyses indicated an origin of the allodapines in Africa, a result that concords with Africa being the center of diversity for this tribe. Although Macrogalea, which is the sister group to all other allodapines, also occurs in Madagascar, the Malagasy clade is derived from within the African clade and colonized that island less than 5 Mya (Smith, 2004; Tierney, 2004). The only two allodapine genera that occur in Laurasian regions, Braunsapis and Exoneuridia, do not form basal clades in the Allodapini, arguing against an origin in that region that might otherwise be suggested by the finding of boreallodapine bees in Baltic amber (Engel, 2001). Although our DIVA analyses indicated an early vicariance event involving African and Madagascan allodapines, this can be dismissed because geological separations between these land masses is much older than the maximum likely divergence times in our dating analyses. Indeed, a vicariance event involving the separation of Madagascar from Africa (beginning 165 Mya and finishing about 120 Mya; Rabinowitz et al., 1983) would have required an origin of the allodapines that predated divergence of the major bee families (Engel, 2001); and a vicariance event involving Australia and Africa would have been even earlier. Therefore, the origin of the exoneurines must have involved dispersal from Africa to Australia, raising the question of what dispersal route was followed. One possibility involves a northern dispersal route from Africa to Australia via southern Asia (Fig. 5b), but we believe that this scenario is very unlikely for three reasons:
- Our DIVA analyses did not infer either a vicariance or dispersal event for the exoneurines or their ancestral lineage that involved both Asia and Australia. This reflects the lack of basal lineages in either Asia or northern Australia.
- The fact that the exoneurines do not have a sister-group relationship with any allodapines from Laurasian or tropical Australian regions suggests their ancestors did not pass through these regions or that all intervening lineages became extinct. This is in strong contrast to the recently derived (ca. 20 Mya) genus Braunsapis, which shows evidence of a typical Indian Ocean Rim dispersal with an origin in Africa followed by dispersal into southern Asia and then northern Australia (Schwarz et al., 2004; Fuller et al., 2005), with major diversification in both of those regions.
- Both the time of divergence between the exoneurines and the African/Malagasy clade (95% minimum ages of 31 and 30 Mya for Bayesian and PL analyses) and radiation of the Australian exoneurines (24 and 25 Mya) occurred before the closure of the Tethys Sea and before the mid-Miocene interchange between the Australian and Laurasian landmasses.
In fact, the minimum age estimates from our analyses are likely to strongly underestimate divergence times for the basal allodapine nodes for three reasons. First, our root node age of 90 Mya assumed that divergence between corbiculate bees and the xylocopine bees coincided with the earliest evidence for corbiculates (Crepet and Dixon, 1998), yet the divergence between these groups forms the most basal node in the Apidae, so that the true age of this node is likely to be much older. Second, we set the minimum age of the node separating the Ceratinini from the Allodapini at 45 Mya, corresponding to the age of the boreallodapine amber fossils. But this calibration point assumes that the allodapine and boreallodapine tribes had diverged from the Ceratinini simultaneously and that the Boreallodapini had become diverse enough for at least three species to survive as amber fossils, all in zero amount of time. It is much more likely that these tribal divergences followed by diversification of the Boreallodapini required substantial time. Lastly, we have based our arguments on the 95% lower limit (i.e., youngest) of estimated node ages, and the actual node ages are likely to be older than this.
A northern dispersal via Asia therefore seems very unlikely, which leaves only two hypotheses for the colonization of Australia: (i) an amphinotic dispersal via Antarctica, or (ii) migration across the Indian Ocean. We now discuss these possibilities.
Amphinotic dispersal, from Africa to Australia via Antarctica, is concordant with the southern distribution of exoneurines and the separation of Antarctica from Australia is comparatively recent (Eocene). However, Africa was the first major land mass to separate from Gondwana and moved rapidly away from its initial departure point with the Antarctic land mass by the late Cretaceous (Smith et al., 1994). Indeed, the separation between Africa and Antarctica in the late Cretaceous, which corresponds to our 95% maximum ages for divergence between the exoneurines and the African + Malagasy clade, is not substantially greater than the current distance between these land masses (Smith et al., 1994).
In contrast, dispersal across the Indian Ocean during the Eocene (Fig. 5b) may not have required transport over ocean barriers as large as those separating Africa from Antarctica. Although Krause et al. (1997) and Cooper et al. (2001) have argued that a possible Kerguelen landbridge, linking Africa and Madagascar to Antarctica until some 80 Mya, may explain ratite and abelisaurid distributions, our maximum age estimates for divergence between exoneurines and African/Malagasy clades are substantially more recent than this and we are not aware of any evidence that such a land bridge existed after the late Cretaceous. However, some evidence raises the possibility that elements of the northern Kerguelen Plateau (NKP) and Broken Ridge igneous provinces may have provided more recent island stepping stones in the eastern Indian Ocean. Volcanism has been widespread in this region since about 130 Mya and is related to the breakup of eastern Gondwana and sea floor spreading (Coffin et al., 2002). Furthermore, recent ocean deep sea drilling indicates subaerial eruption for many elements in the NKP and Broken Ridge regions in the Cenozoic, up until about 25 Mya (Duncan, 2002; Wallace et al., 2002). Although the number of drilling sites for these areas is still very limited, the extent of these igneous provinces during the late Eocene (Fig. 5b), when volcanism was still occurring, allows the possibility of island stepping stones across the Indian Ocean.
There is some evidence that allodapines are able to cross moderately large ocean expanses. Fuller et al. (in press) showed that that allodapine genus Braunsapis successfully crossed the Mozambique Channel, about 450 km wide, at least twice in the last 20 My and Macrogalea has also crossed this barrier in the last 5 My (Tierney, 2004; Smith, 2004), indicating that a seawater distance of this size comprises a substantial, but not insuperable, barrier. Dispersal across now-submerged stepping stones in the Indian Ocean for organisms able to cross moderately large water expanses could have also been aided by the West Wind Drift as well as by the low global sea levels from 30 to 25 Mya (Haq et al., 1988).
Interestingly, there has also been a single colonization event of Australia by the halictine bees of the genus Lasioglossum (Danforth and Ji, 2001), and this gave rise to a major bee faunal element in Australia, with some subsequent northwards dispersal into the indo-Papuan region. Bayesian dating analyses (Danforth et al., 2004) suggest that the colonization of Australia occurred in the Oligocene, about 30 Mya. Leys et al. (2002) also suggested an early dispersal of xylocopine bees (leading to the Australian and Papuan subgenus Lestis) into Australia > 30 Mya. Both inferred dispersal times precede the mid-Miocene interchange between Laurasia and Australia, but concord with our estimates for the timing of dispersal of allodapines to Australia. Unlike the exoneurines, the Australian halictines and the xylocopine subgenus Lestis also occur in tropical regions, but the timing of colonization does not fit well with standard models for interchange between Australian and Laurasian elements. Danforth et al.'s (2004) study does not include Asian halictine taxa that could comprise potential ancestral clades containing the Australian clade, so that the possibility of a Laurasian dispersal route cannot be evaluated at this time. Similarly, current data (Leys et al., 2002) do not allow us to determine whether or not Lestis shows affinities to southern Asian subgenera that could indicate a Laurasian dispersal route.
Studies of southern hemisphere historical biogeography have revealed level of complexity that has challenged older, largely vicariance models and evidence suggests a much greater role for dispersal (Sanmartín and Ronquist, 2004) than had been expected. Nevertheless, some distributions seem to require extraordinarily long dispersals, such as the baobabs (Baum et al., 1998), or else occurred at times that do not match well-understood paleogeographic scenarios, such as the Oligocene colonization of Australia by halictine bees. Current understanding of some aspects of paleogeology, especially in the eastern Indian Ocean, limits our ability to evaluate the likelihood of some dispersals that do not fit more standard models. Results such as ours, presented above, indicate that surprises may be in store and that yet-to-be uncovered dispersal corridors may help explain some currently puzzling similarities among African and Australian biota.
|
Acknowledgements |
|---|
We thank Ishbel Kerkez, Meg Schwarz, and John Zammit for help in the field, often under arduous conditions, and Bryan Danforth for valuable advice on EF-1
primers. We thank Michael Sanderson for helpful advice on using r8s, Mike McLeish for his expert help in preparing the map reconstruction figure, and Kathy Saint for her tireless help with all aspects of sequencing. This work was supported by Australian Research Council grants to M. Schwarz, S. Cooper, and B. Crespi and Flinders University OSP and Program Grant support to M. Schwarz. |
Notes |
|---|
|
TopNotesAbstractMethodsResultsDiscussionReferences |
|---|
4 Current Address: School of Natural Resource Sciences, Queensland University of Technology, GPO Box 2434, Brisbane, Qld, 400, Australia
|
References |
|---|
|
TopNotesAbstractMethodsResultsDiscussionReferences |
|---|
-
Austin J. D., Lougheed S. C., Moler P. E., Boag P. T. Phylogenetics, zoogeography, and the role of dispersal and vicariance in the evolution of the Rana catesbeiana(Anura: Ranidae) species group. Biol. J. Linn. Soc. (2003) 80:601–624.[CrossRef][Web of Science]
Barker K. F., Cibois A., Schikler P., Feinstein J., Cracraft J. Phylogeny and diversification of the largest avian radiation. Proc. Natl. Acad. Sci. USA (2004) 101:11040–11045.
Baum D. A., Smith R. L., Wendel J. F. Biogeography and floral evolution of baobabs (Adansonia, Bombacaceae) as inferred from multiple data sets. Syst. Biol. (1998) 47:181–207.
Bull N. J., Schwarz M. P., Cooper S. J. B. Phylogenetic divergence of the Australian allodapine bees. Mol. Phylogenet. Evol. (2003) 27:212–222.[CrossRef][Web of Science][Medline]
Chomczynski P., Mackey K., Drews R., Wilfinger W. DNAzol®: A reagent for the rapid isolation of genomic DNA. BioTechniques (1997) 22:550–553.[Web of Science][Medline]
Chomczynski P., Wilfinger W., Mackey K. Isolation of genomic DNA from human, animal and plant samples with DNAzol reagents. In: Biotechnology International (1998) San Francisco, CA: Universal Medical Press, Inc. 185–188.
Coffin M. F., Pringle M. S., Duncan R. A., Gladczenko T. P., Storey M., Müller R. D., Gahagan L. A. Kerguelen hotspot magma output since 130 Ma. J. Petrol. (2002) 43:1121–1139.
Cooper A., Lalueza-Fox C., Anderson S., Rambaut A., Austin J., Ward R. Complete mitochondrial genome sequences of two extinct moas clarify ratite evolution. Nature (2001) 409:704–707.[CrossRef]
Crepet W. L., Nixon K. C. Fossil Clusiaceae from the late Cretaceous (Turonian) of New Jersey and implications regarding the history of bee pollination. Am. J. Botany (1998) 85:1122–1133.[Abstract]
Danforth B. N., Brady S. G., Sipes S. D., Pearson A. Single-copy nuclear genes recover Cretaceous-age divergences in bees. Syst. Biol. (2004) 53:309–326.
Danforth B. N., Ji S. Elongation factor-1 occurs as two copies in bees: Implications for phylogenetic analysis of EF-1 sequences in insects. Mol. Biol. Evol. (1998) 15:225–235.[Abstract]
Doyle J. J., Doyle J. L. Isolation of plant DNA from fresh tissues. Focus (1990) 12:13–15.
Duncan R. A. A time frame for construction of the Kerguelen Plateau and Broken Ridge. J. Petrol. (2002) 43:1109–1119.
Engel M. S. A monograph of the Baltic amber bees and evolution of the Apoidea. Bull. Am. Mus. Nat. Hist. (2001) 259:1–192.[CrossRef]
Felsenstein J., Churchill G. A. A hidden Markov approach to variation among sites in rate of evolution. Mol. Biol. Evol. (1996) 13:93–104.[Abstract]
Fuller S, Schwarz M. P., Tierney S. M. Phylogenetics of the allodapine bee genus Braunsapis: historical biogeography and long-range dispersal over water. J. Biogeog. (2005) 32:2135–2144.[CrossRef]
Goldman N., Whelan S. Statistical tests of gamma-distributed rate heterogeneity in models of sequence evolution in phylogenetics. Mol. Biol. Evol. (2000) 17:975–978.
Grimaldi D. A. The co-radiation of insects and angiosperms in the Cretaceous. Ann. Missouri Bot. Gard. (1999) 86:373–406.[CrossRef][Web of Science]
Haq B. U., Hardenbol J., Vail P. R. Chronology of fluctuating sea levels since the Triassic. Science (1987) 235:1156–1167.
Hay W. W., DeConto R., Wold C. N., Wilson K. M., Voigt S., Schulz M., Wold-Rossby A., Dullo W.-C., Ronov A. B., Balukhovsky A. N., Soeding E. Alternative global cretaceous paleogeography. In: The evolution of Cretaceous ocean/climate systems, Geological Society of America Special Paper 332—Barrera E., Johnson C., eds. (1999) Boulder, CA: Geological Society of America. 1–47.
Houston T. F. New Australian bees (subgenus Exoneurella Michener) and their immatures. Trans. Roy. Soc. S. Austr. (1976) 100:15–28.
Huelsenbeck J. P., Ronquist F. MrBayes: Bayesian inference of phylogeny. Bioinformatics (2001) 17:754–755.
Leys R., Cooper S. J., Schwarz M. P. Molecular phylogeny and historical biogeography of the large carpenter bees, genus Xylocopa. Biol. J. Linn. Soc. (2002) 77:249–266.[CrossRef][Web of Science]
Michener C. D. The lifecycle and social organization of bees of the genus Exoneura and their parasite Inquilina. Univ. Kansas Sci. Bull. (1965) 46:317–358.
Michener C. D. Discordant evolution and the classification of allodapine bees. Syst. Zool. (1977) 26:32–56.
Michener C. D. The bees of the world (2000) Baltimore, Maryland: John Hopkins University Press.
Pauly A., Brooks R. W., Nilsson L. A., Persenko Y. A., Eardley C. D., Terzo M., Griswold T., Schwarz M., Patiny S., Munzinger J., Barbier Y. Hymenoptera Apoidea de Madagascar et des iles voisines. Ann. Sci. Zool. (2001) 286:1–390.
Rabinowitz P. D., Coffin M. F., Falvey D. The separation of Madagascar and Africa. Science (1983) 220:67–69.
Reyes S. G. A cladistic analysis of the bee tribe Allodapini (Hymenoptera: Apidae: Xylocopinae). Philipp. Entomol. (1998) 12:55–84.
Richardson J. E., Weitz F. M., Fay M. F., Cronk Q. C. B., Linder H. P., Reeves G., Chase M. W. Rapid and recent origin of species richness in the Cape flora of South Africa. Nature (2001) 412:181–183.[CrossRef]
Ronquist F. DIVA version 1.1. Computer program and manual available by anonymous FTP from Uppsala University (1996) ftp.uu.se or ftp.systbot.uu.se.
Ronquist F. Dispersal-vicariance analysis: A new approach to the quantification of historical biogeography. Syst. Biol. (1997) 46:195–203.
Rutschmann F. Bayesian molecular dating using PAML/multidivtime. In: A step-by-step manual (2004) Switzerland: University of Zurich.
Sampson S. D., Witmer L. M., Forster C. A., Krause D. W., O'Connor P. M., Dodson P., Ravoavy F. Predatory dinosaur remains from Madagascar: Implications for the Cretaceous. Science (1998) 280:1048–1051.
Sanmartín I., Ronquist F. Southern hemisphere biogeography inferred by event-based models: Plant versus animal patterns. Syst. Biol. (2004) 53:216–243.
Schwarz M. P., Bull N. J., Cooper S J. B. The molecular phylogenetics of allodapine bees, with implications for the evolution of sociality and progressive rearing. Syst. Biol. (2003) 52:1–14.
Schwarz M. P., Bull N. J., Hogendoorn K. Evolution of sociality in the allodapine bees: A review of sex allocation, ecology and evolution. Insectes Soc. (1998) 45:349–368.[CrossRef][Web of Science]
Schwarz M. P., Tierney S. M., Cooper S. J. B., Bull N. J. Molecular phylogenetics of the allodapine bee genus Braunsapis: A-T bias and heterogeneous substitution parameters. Mol. Phylogenel. Evol. (2004) 32:110–122.[CrossRef]
Schwarz M. P., Tierney S. M., Zammit J., Schwarz P. M., Fuller S. Brood provisioning and colony composition of a Malagasy species of Halterapis: implications for social evolution in the allodapine bees. Ann. Entomol. Soc. Am. (2005) 98:126–133.[CrossRef][Web of Science]
Slowinski J. B. Molecular polytomies. Mol. Phylo. Evol. (2001) 19:114–120.[CrossRef][Web of Science][Medline]
Smith A. G., Smith D. G., Funnell M. Atlas of Mesozoic and Cenozoic coastlines (1994) Cambridge, UK: Cambridge University Press.
Swofford D. L. PAUP. Phylogenetic analysis using parsimony (*and other methods) (1999) Sunderland, Massachusetts: Sinauer Associates. Version 4.0b.
Terzo M. Révision du genre Exoneuridia Cockerell, 1911. Belg. J. Entomol. (1999) 1:137–152.
Thorne J. L., Kishino H. Divergence time and evolutionary rate estimation with multilocus data. Syst. Biol. (2002) 51:689–702.
Thorne J. L., Kishino H., Painter I. S. Estimating the rate of evolution and the rate of molecular evolution. Mol. Biol. Evol. (1998) 15:1647–1657.[Abstract]
Tierney S. M. Evolution of African allodapine bees (2004) Adelaide: School of Biological Sciences, Flinders University. PhD thesis.
Tierney S. M., Cronin A. L., Loussert N., Schwarz M. P. The biology of Brevineura froggatti and phylogenetic conservatism in Australian allodapine bees (Apidae, Allodapini). Insectes Soc. (2000) 47:96–97.[CrossRef][Web of Science]
Wallace P. J., Frey F. A., Weis D., Coffin M. F. Origin and evolution of the Kerguelen Plateau, Broken Ridge and Kerguelen Archipelago. J. Petrol. (2002) 43:1105–1108.
Yuan Y., Wohlhauser S., Möller M., Klackenberg J., et al. Phylogeny and biogeography of Exacum (Gentianaceae): A disjunctive distribution in the Indian Ocean Basin resulted from long distance dispersal and extensive radiation. Syst. Biol. (2005) 54:21–35.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  T. J. Bradley, A. D. Briscoe, S. G. Brady, H. L. Contreras, B. N. Danforth, R. Dudley, D. Grimaldi, J. F. Harrison, J. A. Kaiser, C. Merlin, et al. Episodes in insect evolution Integr. Comp. Biol., November 1, 2009; 49(5): 590 - 606. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||