MBE Advance Access originally published online on June 1, 2005
Molecular Biology and Evolution 2005 22(9):1869-1886; doi:10.1093/molbev/msi180
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Research Article |
A Nuclear DNA Phylogenetic Perspective on the Evolution of Echolocation and Historical Biogeography of Extant Bats (Chiroptera)
Department of Zoology, Stellenbosch University, Stellenbosch, South Africa
E-mail: geeta.eick{at}gmail.com.
 |
Abstract |
|---|
 TOP Abstract IntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Bats (Order Chiroptera), the only mammals capable of powered flight and sophisticated laryngeal echolocation, represent one of the most species-rich and ubiquitous orders of mammals. However, phylogenetic relationships within this group are poorly resolved. A robust evolutionary tree of Chiroptera is essential for evaluating the phylogeny of echolocation within Chiroptera, as well as for understanding their biogeographical history. We generated 4 kb of sequence data from portions of four novel nuclear intron markers for multiple representatives of 17 of the 18 recognized extant bat families, as well as the putative bat family Miniopteridae. Three echolocation-call characters were examined by mapping them onto the combined topology: (1) high-duty cycle versus low-duty cycle, (2) high-intensity versus low-intensity call emission, and (3) oral versus nasal emission. Echolocation seems to be highly convergent, and the mapping of echolocation-call design onto our phylogeny does not appear to resolve the question of whether echolocation had a single or two origins. Fossil taxa may also provide insight into the evolution of bats; we therefore evaluate 195 morphological characters in light of our nuclear DNA phylogeny. All but 24 of the morphological characters were found to be homoplasious when mapped onto the supermatrix topology, while the remaining characters provided insufficient information to reconstruct the placement of the fossil bat taxa with respect to extant families. However, a morphological synapomorphy characterizing the Rhinolophoidea was identified and is suggestive of a separate origin of echolocation in this clade. Dispersal-Vicariance analysis together with a relaxed Bayesian clock were used to evaluate possible biogeographic scenarios that could account for the current distribution pattern of extant bat families. Africa was reconstructed as the center of origin of modern-day bat families.
Key Words: Chiroptera • extant bat • DIVA • molecular clock • Laurasiatheria • mammal • phylogeny
|
Introduction |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Bat systematics has experienced great upheaval in recent years with the advent of large-scale molecular studies and application of explicit phylogenetic methodologies. First, the superorder Archonta (Novacek and Wyss 1986
; Novacek 1992) comprising Chiroptera (bats), Dermoptera (flying lemurs), Primata (primates), and Scandentia (tree shrews) is erroneous. Several molecular studies have shown that Chiroptera belong to the Laurasiatheria (represented by carnivores, pangolins, cetartiodactyls, eulipotyphlans, and perissodactyls) and are only distantly related to dermopterans, scandentians, and primates (Nikaido et al. 2000; Lin and Penny 2001; Madsen et al. 2001; Murphy et al. 2001a, 2001b; Van Den Bussche and Hoofer 2004). Second, molecular studies have challenged the notion, based largely on morphological characters, that bats are diphyletic with megabats being more closely related to primates than to microbats (Smith and Madkour 1980; Hill and Smith 1984; Pettigrew 1986; Pettigrew et al. 1989). Instead, these studies have consistently shown high support for a monophyletic Chiroptera (Nikaido et al. 2000; Murphy et al. 2001a, b; Volleth et al. 2002; Van Den Bussche and Hoofer 2004). Finally, the division of the Chiroptera into two suborders, the Megachiroptera (nonecholocating bats) and Microchiroptera (echolocating bats) (Gray 1821 fide Miller 1907) has been challenged by some molecular studies that place Old World fruit bats and rhinolophoid microbats (excluding nycterids) in one clade, with all other microbats in another (Hutcheon, Kirsch, and Pettigrew 1998; Teeling et al. 2000, 2002, 2003, 2005; Springer et al. 2001; Hutcheon and Kirsch 2004; Van Den Bussche and Hoofer 2004).
Given the disparity between molecular and non-molecular phylogenies, acceptance of microchiropteran diphyly has not been widespread (Schintzler, Kalko, and Denzinger 2004; Simmons and Conway 2003) with authors citing the need for more comprehensive and robust molecular phylogenies based on increased sampling at the family level (Teeling et al. 2002; Simmons and Conway 2004). Within the framework of monophyly of Chiroptera, paraphyly of microbats either requires loss of echolocation along the pteropodid lineage, only to reevolve in Rousettus albeit in a more primitive form (Holland, Waters, and Rayner 2004), or two independent origins of echolocation within the chiropteran lineage. Thus, more evidence for or against the paraphyly of microbats is essential for evaluating the evolution of echolocation within the monophyletic Chiroptera.
By using increased taxon sampling (Pollock and Bruno 2000; Pollock et al. 2002; Zwickl and Hillis 2002), we were interested in exploring the utility of novel nuclear intron sequences for recovering higher level systematic relationships among extant chiropteran families. This study was prompted by the availability of a suite of unique nuclear DNA intron markers useful for recovering phylogenetic information at several different taxonomic levels (Matthee and Davis 2001; Matthee et al. 2001, 2004; Willows-Munro, Robinson, and Matthee 2005).
Our aims were fourfold. First, to provide an independent assessment of evolutionary relationships within Chiroptera based on comprehensive taxon sampling and phylogenetic analyses of a nuclear intron supermatrix. Second, we were interested in examining the evolution of echolocation in bats using the phylogenetic framework developed in this study. Third, we aimed to reevaluate the morphological characters identified by Simmons and Geisler (1998) in terms of our molecular phylogeny to assess their utility for reconstructing relationships of fossil bat taxa to extant groups. This aided our interpretation of the evolution of echolocation. Last, given the complex biogeographical distribution as well as the scant paleontological record of bats, we were interested in providing a timescale for the radiation of extant chiropteran families using a relaxed Bayesian clock. This allowed us to evaluate possible biogeographical scenarios that explain the current distribution of bats.
|
Materials and Methods |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Taxonomic Sampling
Multiple representatives of 17 of the 18 currently recognized extant bat families (excluding the monotypic Crasonycteridae) (Simmons, in press
) were included in this study (table 1). We also included species representatives of the putative bat family Miniopteridae (Mein and Tupinier 1977; Gopalakrishna and Chari 1983; Pierson 1986; Tiunov 1989; Kawai et al. 2002; Van Den Bussche and Hoofer 2004; Hutcheon and Kirsch 2004). The two species comprising the vespertilionid genus Cistugo Thomas 1912 were included as this taxon may be sufficiently different to warrant separate family-level status (M. Ruedi, personal communication). We included only three representatives of Laurasiatheria (Nikaido et al. 2000; Lin and Penny 2001; Madsen et al. 2001; Murphy et al. 2001a, b; Van Den Bussche and Hoofer 2004), namely the horse (Order Perissodactyla), genet (Order Carnivora), and pangolin (Order Pholidota) as outgroups (table 1) because recent comprehensive mammalian phylogenies have unambiguously demonstrated the monophyly of Chiroptera (Nikaido et al. 2000; Lin and Penny 2001; Madsen et al. 2001; Murphy et al. 2001a, b; Van Den Bussche and Hoofer 2004).
|
Data Collection
Total genomic DNA was extracted from 95% ethanol or dimethyl sulfoxide-preserved tissue using a phenol-chloroform-isoamyl procedure (Sambrook, Fritsch, and Maniatis 1989
). Introns from four nuclear genes (PRKC1, SPTBN, STAT5A, and THY) were targeted using primers designed previously (Matthee et al. 2001, 2004). To increase the success of amplification across all taxa included, chiropteran-specific primers were designed (supplementary table 1, Supplementary Material online), and used in various combinations with previously published primers. Sequence data for the horse, Equus caballus, was obtained from GenBank (Matthee et al. 2001; table 1), whereas intron sequence data for the other two out-group taxa were generated in this study. The cycling profile and subsequent purification and automated sequencing followed protocols outlined in Matthee et al. (2004). In most instances, both strands were sequenced to ensure accuracy of the sequence data, and the sequence identity was checked by GenBank BlastN searches. Heterozygous changes occurred in a limited number of nuclear fragments, and these were assigned an IUBMB ambiguity code. Nuclear intron sequences and alignments generated in this study have been deposited in EMBL (STAT5A: AJ865389–AJ865445, ALIGN_000804; THY: AJ865634–AJ865690, ALIGN_000805; PRKC1: AJ866286–AJ86636, ALIGN_000806; SPTBN: AJ866337–AJ86639, ALIGN_000807).
Sequence Alignment
Sequences were initially aligned using the multiple alignment program T-COFFEE (Notredame, Higgins, and Heringa 2000) and thereafter optimized manually in MacClade 4.0 (W. P. Maddison and D. R. Maddison 1989), using the conserved exon sequences on either the 5' or 3' end of the sequence to anchor the alignments. Insertion deletion (indel) events were observed among taxa, and gaps were introduced (by T-COFFEE or manually) to maintain the alignment. Although there were some areas where alternative alignments were likely, our investigations showed that minor changes to the alignment in these regions did not significantly alter relationships among taxa (data not presented). Two regions within the SPTBN intron (region 796–815 and region 832–1190; ALIGN_000807) could not be aligned, and these were therefore excluded in all analyses. All alignment gaps were treated as missing characters in phylogenetic analyses. Following the suggestions of Matthee et al. (2001) only indels longer than 2 bp were mapped onto the tree obtained from analysis of the supermatrix.
Data Set Characterization and Phylogenetic Analyses
Base composition was estimated using MEGA v2.1 (http://www.megasoftware.net), and base frequency stationarity was evaluated using a 
2 test implemented in PAUP* v4.0b10 (Swofford 2002). Incongruence between data sets was evaluated using the recommendations of De Queiroz (1993) by determining whether there were any consistently strongly (
70% bootstrap support, 0.95 Bayesian posterior probability) supported nodes in one data set that conflicted with strongly supported nodes in another. As no such incongruent nodes were present, the four introns were concatenated to form a supermatrix. The genes were analyzed separately as well as combined using maximum parsimony (MP) and maximum likelihood (ML) in PAUP* v4.0b10 (Swofford 2002), and Bayesian inference (BI) as implemented by MrBayes 3.0b4 (Huelsenbeck and Ronquist 2001). For MP searches, trees were generated using equal weighting and the heuristic search option with tree-bisection-reconnection (TBR) branch swapping and stepwise addition of taxa using 1,000 random sequence addition replicates, with one tree retained per stepwise addition replicate. For ML analyses, nucleotide substitution models were selected using Modeltest v 3.06 (Posada and Crandall 1998), and searches were performed under both likelihood ratio test (LRT) and Akaike Information Criterion (AIC) optimal models if these were different. In all ML analyses, starting trees were obtained by neighbor-joining followed by TBR branch swapping. Nodal support for the MP analyses was assessed from 1,000 nonparametric bootstrap replicates (full heuristic search; two random stepwise addition of taxa). Nodal support for individual intron ML analyses was assessed from 100 ML bootstrap replicates using NNI as the branch-swapping algorithm. Due to computational demands, ML bootstrap runs for single nuclear introns were performed with a constraint topology in which all families with more than one taxon representative were constrained to be monophyletic (as supported in Murphy et al. 2001b). For the supermatrix, ML bootstrap support was determined from 75 replicates with TBR branch swapping and no phylogenetic constraints imposed. BI was implemented setting the prior model to that specified by Modeltest for each data set. If the model suggested by AIC and LRT differed, two separate runs were performed. The supermatrix was analyzed using both a single model and in a partitioned manner to allow the selection of different optimal parameters for each partition (Huelsenbeck and Ronquist 2001). A random tree generated by MrBayes was used as a starting tree for each Markov chain. Four Markov chains were run for 1 million generations, comprising one cold chain and three incrementally heated chains. Tree sampling was performed every 50 generations, thereby generating 20,000 sample points. The sump command was used to generate plots of generation number versus the log probability of observing the data, and samples taken during the first 25,000 cycles of the chain were discarded as "burn-in" (Huelsenbeck 2002). Posterior probabilities were based on the remaining 19,500 trees. Three independent Bayesian runs with different random starting trees were performed to ensure convergence on the same topology (Huelsenbeck and Ronquist 2001). Nodes that received 70% bootstrap support or those with 0.95 Bayesian posterior probability were considered well supported. To investigate the amount of phylogenetic signal in the data sets, the number of unique topologies in the 95% posterior interval was estimated for all data sets (Buckley et al. 2002). Alternative tree topologies were compared with the optimal ML tree topology using the approximately unbiased (AU) test (Shimodaira 2002) implemented in Consel V1.0g (Shimodaira and Hasegawa 2001).
Mapping of Echolocation Characters
MacClade version 4.0 (W. P. Maddison and D. R. Maddison 1989) was used to optimize characters related to echolocation behavior on the supermatrix topology using parsimony as an optimality criterion. Reconstruction of characters was examined using both delayed transformations (Deltran) and accelerated transformations (Actran) optimization, but no differences in the reconstructions for the ancestral nodes of interest were observed. Echolocation characters were evaluated under two scenarios as the current gene trees suggest that echolocation of extant chiropteran families have either two independent origins or one origin and one loss of this character in the Old World fruit bat lineage. The three echolocation characters mapped onto the phylogeny were (1) high-duty versus low-duty cycle echolocation calls (Fenton et al. 1995), (2) low-intensity versus high-intensity echolocation calls (Arita and Fenton 1997; DeBaeremaker and Fenton 2003), and (3) nasal echolocation versus oral echolocation (Pederson 1993, 1995, 1998). Two families of bats, the Hipposideridae and Rhinolophoidae, as well as the mormoopid species Pteronotus parnellii were classified as high-duty cycle echolocators (Fenton 1999; Jones 1999). Five families of bats were classified as comprising bats producing mainly low-intensity echolocation calls, namely Thyropteridae (Fenton et al. 1999a), Nycteridae (Aldridge et al. 1990), Megadermatidae, Phyllostomidae (Schintzler and Kalko 2001), and Furipteridae (Fenton et al. 1999b). Vespertilionidae was coded as having both low-intensity and high-intensity echolocating bats as some vesper species produce low-intensity echolocation calls, for example, Myotis emarginatus (Schumm, Krull, and Neuweiler 1991) and Plecotus auritus (Waters and Jones 1995). Six families of bats were classified as nasal echolocators, namely Rhinolophidae, Hipposideridae, Megadermatidae, Nycteridae, Phyllostomidae, and Rhinopomatidae. Although it is uncertain whether rhinopomatids emit their echolocation calls through their mouths or nostrils, Pederson (1993) classified Rhinopoma muscatellum as a nasal echolocator based on cephalometric characters. Similarly Göbbel (2002) reports similarities between Rhinopoma hardwickei, rhinolophids and megadermatids (the latter two both unambiguous nasal echolocators) in the external nasal cartilage. Although some vespertilionids possess rudimentary nose leaves that are often associated with nasal emission, for example, Nyctophilus, Pharotis, and Antrozous (Nowak 1999), it is unclear if these bats emit echolocation signals orally or nasally; therefore vespertilionids were not coded as polymorphic for oral-nasal echolocation.
Morphology, Molecular Scaffolds, and Fossil Taxa
The 195 morphological characters identified by Simmons and Geisler (1998) were reevaluated in light of the paraphyly of microbats using the "trace character" option in MacClade. Morphological characters were characterized as either homoplasious or nonhomoplasious by mapping them on the gene tree comprising two clades—one containing the fruit bats and rhinolophoid microbats, and the other comprising all remaining bat families. Relationships among families within these two clades were collapsed to polytomies. Craseonycteris was placed within the fruit bat rhinolophoid lineage on the basis of morphological and molecular evidence suggesting a close affiliation with this group (Simmons and Geisler 1998; Hulva and Horacek 2002). Only morphological characters in congruence with the gene tree and scored for at least one fossil taxon were used to reconstruct the relationships of the fossil bats to the extant taxa. Although character polarity is dependent on the choice of out-groups (in this instance representatives of Scandentia and Dermoptera), the aim of this exercise was merely to reexamine the placement of the fossil taxa as suggested by Springer et al. (2001) because we argue that the exclusion of a significant amount of homoplasious characters can potentially alter the conclusions reached by these authors. The molecular scaffold used by Springer et al. (2001) was used as one of two backbone constraints in MP analyses with selected morphological characters and all the taxa included in the data set of Simmons and Geisler (1998) and was as follows: (Scandentia, Dermoptera, (Pteropodidae (Hipposideridae, Megadermatidae)), (Molossinae, Emballnouridae, Phyllostomidae)). This scaffold was congruent with the intron supermatrix topology generated in our study. The second scaffold used constrained Kerivoulinae, Murinae, Myotinae, Antrozoidae, Tomopotinae, and Vespertilionidae to form a monophyletic group (Teeling et al. 2002; Hoofer and Van Den Bussche 2003; Hutcheon and Kirsch 2004) as well as members of the superfamily Noctilionoidea (Phyllostomidae, Mormoopidae, Furipteridae, Mystacinidae, Noctilionidae, and Thyropteridae) and the Old World fruit bat—rhinolophoid microbat clade (Pteropodidae, Rhinolophidae, Hipposideridae, Megadermatidae, Rhinopomatidae, and Craseonycteridae) based on consensus between the results from this and other studies (Teeling et al. 2000, 2002, 2003, 2005; Hoofer and Van Den Bussche 2003; Van Den Bussche and Hoofer 2004). Unweighted parsimony analysis with 1,000 random addition sequence replicates was performed, setting the maximum number of trees saved to 20,000. Nonparametric bootstrap estimates were based on 1,000 replicates, with a maximum of 1,000 trees saved per replicate.
Molecular Clock
The relaxed Bayesian clock method (Thorne, Kishino, and Painter 1998; Thorne and Kishino 2002) following the methodology outlined in Matthee et al. (2004) was used to date the evolution of the various chiropteran lineages. As priors we used 65 Myr (standard deviation [SD] = 65 Myr) between the tip and the root and 0.003 (SD = 0.003) substitutions per site per Myr for the rate at the root node. The conservative prior by 65 million years was chosen based on a strict interpretation of the Explosive model of placental diversification, which places extant placental ordinal diversification in the early Paleocene (Archibald and Deutschman 2001). The value of the substitutions per site at the root rate was determined by using a median amount of evolution (substitutions per site) among genes separating roots and tips; this value was divided by the 65 million years that was believed to be a reasonable age for the diversification of the Chiroplera. The value of the rate of evolution at the root node was varied, and it was found that even large changes to the root rate had little influence on clock estimates. In addition, large differences between the prior and posterior time estimates were observed, tending to support the notion that most of the molecular dating information was based on the concatenated DNA markers and not the priors. Equus caballus was designated as the out-group, and to obtain reasonably narrow posterior distributions for divergence times, six time constraints were incorporated from the fossil record. The first pair of constraints was a minimum of 34 MYA and a maximum of 55 MYA for the split between Megaderma and Rhinopoma (McKenna and Bell 1997; Teeling et al. 2003). The second pair of constraints was a minimum of 37 MYA and a maximum of 55 MYA for the split between the hipposiderids and rhinolophids (McKenna and Bell 1997; Teeling et al. 2003). A minimum of 63 MYA was used for the carnivore pangolin split based on fossil data (McKenna and Bell 1997), and a minimum of 55 MYA was used for the appearance of bats based on the oldest bat fossils (Icaronycteris and Australonycteris) discovered to date (Jepsen 1966; Hand et al. 1994). The highest possible divergence for the in-group was set at 100 Myr. Analyses were repeated removing one constraint per run to estimate the sensitivity of the molecular clock to any one particular constraint. To examine the "clock-like" signal within each intron, divergence estimates were also estimated for each intron separately.
The International Union of Geological Sciences International Stratigraphic Chart (available at http://www.iugs.org/iugs/pubs/intstratchart.htm) was used in conjunction with McKenna and Bell (1997) for delineating epochs.
Biogeographic Analyses
Dispersal-vicariance analysis (Diva) (Ronquist 1997), as implemented in the computer program DIVA v. 1.1 (Ronquist 1996), was used to reconstruct ancestral distributions of extant Chiroptera. The supermatrix phylogeny with families as terminal taxa was used, and seven biogeographical areas were recognized based on continental designations: Africa (A), Asia (B), Australia (C), Europe (D), North America (E), South America (F), and New Zealand (G). Families were coded for their current distributions based on distribution tables from Walker's Mammals of the World (Nowak 1999). Although Myzopoda currently only occurs in Madagascar, early Pleistocene fossil records have been found in East Africa (McKenna and Bell 1997), hence this family was coded as present in Africa. The analysis was run with no constraint on the number of "maxareas."
|
Results |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Sequence Data
The combined intron supermatrix comprised 4,002 alignment positions for 58 taxa (table 2). Despite multiple amplification attempts, the SPTBN region did not produce homologous sequences to the targeted intron for Tadarida aegyptiaca, Mormopterus petrophilus, Chaerephon ansorgei, and Cistugo seabrai. Likewise, no PRKC1 data could be generated for Thyroptera lavali, Thyroptera tricolor, Myzopoda aurita, Otomops martiensenni, Hipposideros commersoni, and Cynopterus sphinx. Retention index (RI) values were relatively high (>0.6), indicating low levels of homoplasy in the intron data sets. Base frequencies did not deviate from stationarity across all lineages for all taxa for three of the four nuclear introns used in this study (PRKC1:
2 = 82.3, df = 153, P = 0.99; STAT5A: 2 = 181, df = 171, P = 0.29; THY: 2 = 95.1, df = 171, P = 0.99). For SPTBN, there was seemingly significant deviation from stationarity for all taxa (2 = 204.13, df = 159, P = 0.009), although this appears to be due to a paucity of adenines in the SPTBN intron of Cistugo lesueuri. When the latter taxon was removed the result became nonsignificant (2 = 158.11, df = 156, P = 0.44). Despite this, C. lesueuri was not excluded from the analyses as its phylogenetic placement was invariant across all introns and all types of analyses.
|
Insertions and Deletions
A total of 34 potentially phylogenetically informative indels were identified in the four intron data sets. Of these, 22 indels provided confirmation for the monophyly of several bat families, while only four supported associations at the higher taxonomic level (fig. 1). The eight remaining indels were homoplasious when mapped onto the supermatrix topology and are not shown. Large autapomorphic insertions (>100 bp) were present in the introns of some taxa, and BlastN searches of these revealed no significant homology to any annotated gene sequences available in GenBank. The only exception was the 199-bp insertion in T. aegyptiaca, which showed significant homology to a published Myotis myotis microsatellite (Castella and Ruedi 2000
).
|
|
Model Selection
For two of the five data sets (STAT5A and the supermatrix) the same optimal model was chosen under hierachical likelihood ratio test (hLRT) and AIC criteria (supplementary table 2, Supplementary Material online). For PRKC1, SPTBN, and THY analyses, analyses were performed twice using both models, respectively. These alternative models did not influence any of the topologies, and for consistency, only the ML and Bayesian topologies based on hLRT will be presented.
Separate Analysis of Nuclear Introns
All nodes defining the monophyly of chiropteran families and also those defining the associations among chiropteran families were labeled A through to Q (fig. 1). Only three nodes (fig. 1, A, N, and P) were consistently obtained when each intron was analyzed separately (table 3). However, no incongruent nodes recovered by any of the analytical methods were supported by high bootstrap or posterior probability values. The STAT5A and THY introns showed the most congruence with the supermatrix topology, with 65%–88% of the nodes present recovered for the former and 71%–94% of the nodes recovered for the latter (table 3). The lack of resolving power when each intron is analyzed separately is emphasized by the large number of trees (>11,000 trees) present in the 95% confidence interval of each intron (table 2).
Combining Data
Combining the introns into a supermatrix dramatically improved the resolving power of the data. Single MP and ML trees were recovered, and the number of trees in the 0.95 Bayesian confidence interval decreased by approximately 16-fold compared with the analysis of each intron separately (table 2). Whereas only one node (Node A) received robust support across all analyses when introns were analyzed separately (table 3), seven nodes received significant support using all three methods of analysis when the data were combined, and an additional three nodes received significant support from two of the three phylogenetic analyses (fig. 1). Bayesian analyses under a single model or using a partitioned model resulted in identical topologies, and Bayesian posterior probability values for nodes differed by 0.02–0.03 at most. The number of trees found in the 95% posterior interval was slightly lower under a partitioned model than under a combined model.
Intron Supermatrix and Chiropteran Phylogenetics
The single ML tree and Bayesian consensus topology recovered from analysis of the intron supermatrix were identical (fig. 1). The MP tree differed from this topology in only one respect: M. aurita was placed as sister taxon to Miniopteridae, rather than basal to Miniopteridae, Vespertilionidae, Molossidae, and Natalidae.
Nuclear intron data support paraphyly of the microbats (Hutcheon, Kirsch, and Pettigrew 1998; Teeling et al. 2000, 2002, 2003, 2005; Hutcheon and Kirsch 2004; Van Den Bussche and Hoofer 2004). The association between Old World fruit bats (Pteropodidae) and rhinolophids, hipposiderids, megadermatids, and rhinopomatids (hereafter referred to as "rhinolophoid microbats") received significant statistical support from all three methods of phylogenetic analyses (table 4). Furthermore, a phylogenetic hypothesis in which the microbats are monophyletic was rejected at the 95% confidence level by the AU test (table 4). The remaining microbats grouped within a clade that received significant bootstrap support from all three methods of phylogenetic analysis and was supported by a 15-bp deletion in the STAT5A intron. This corresponds to the suborder Yangochiroptera of Teeling et al. (2002, 2003, 2005), and if this holds, our data suggest that the families Furipteridae, Myzopodidae, Thyropteridae, and Miniopteridae, which were not included in previous studies, can now also be included in this subordinal rank. Following the recommendations of Hutcheon and Kirsch (2004), we refer to the two suborders of chiropterans as "Pteropodiformes" (comprising the Pteropodidae, Rhinolophidae, Hipposideridae, Megadermatidae, and Rhinopomatidae) and "Vespertilioniformes" (remaining microbat families). The association of the families Noctilionidae, Furipteridae, Phyllostomidae, Mormoopidae, Mystacinidae, and Thyropteridae in a single, robustly supported clade (fig. 1) corresponds to the expanded Noctilionoidea of Van Den Bussche and Hoofer (2004). Within this clade, the sister taxon relationship between Mormoopidae and Phyllostomidae is supported both by robust bootstrap and Bayesian posterior probability values as well as a unique indel in THY (Kirsch et al. 1998; Van Den Bussche and Hoofer 2000, 2001, 2004; Teeling et al. 2003, 2005). Noctilionoidae and Furipteridae are sister taxa, based on nodal support values as well as tests of alternative topologies (fig. 1; table 4), as supported by the mitochondrial-based studies of Van Den Bussche and Hoofer (2001, 2004) and Hoofer et al. (2003). Nycteridae and Emballonuridae are sister taxa as supported by Teeling et al. (2002, 2003, 2005), with the intron data rejecting a hypothesis in which these two groups are separated from each other (table 4). Miniopterids appear most closely related to the vespertilionids, with the molossids sister taxon appearing most closely related to the miniopterids and vespers (fig. 1). The grouping of vespertilionids, miniopterids, molossids, and natalids corresponds to the Superfamily Vespertilionoidea of Teeling et al. (2002, 2005), although they included no miniopterids in their study. In our study, there is a lack of strong statistical support for this group in contrast to the high bootstrap and Bayesian support (>90%) found by Teeling et al. (2002, 2003, 2005). Alternative hypotheses that miniopterids and vespertilionids are not sister taxa and that miniopterids and molossid are more closely related could not be rejected by the intron data (table 4). Myzopoda aurita, characterized by a relatively long branch, was placed basal to the Vespertilionoidea. However, this finding should be treated with caution as there was no significant nodal support for this placement. Furthermore, evaluation of alternative topologies in which M. aurita was associated with the nycterids and emballonurids (fig. 1, Node I), Noctilionoidea (fig. 1, Node N) or a clade comprising nycterids, emballonurids and the Noctilionoidae (fig. 1, Node O) could not be rejected by the data (table 4).
|
Within the Pteropodiformes, phylogenetic relationships were congruent with those found in previous studies, with rhinolophids and hipposiderids sister taxa and megadermatids and rhinopomatids sister taxa (Teeling et al. 2002
, 2003). Unfortunately, because Craseonycteris thonglonyai was not included in this study, the hypothesis that this family is allied to the rhinolophoid microbats (Hulva and Horacek 2002) could not be evaluated. The concept of a clade Rhinolophoidea sensu Simmons and Geisler (1998) comprising rhinolophids, hipposiderids, megadermatids, rhinopomatids, and nycterids based on morphological characters was rejected by the intron data (table 4). Similarly, Nataloidea sensu Simmons and Geisler (1998) (comprising Furipteridae, Natalidae, Thyropteridae, and Myzopodidae), again a clade strongly supported by morphological characters, was also rejected by the intron data (table 4). The genus Cistugo appears distinct from other vespertilionids. Whereas all other vespers are characterized by a unique deletion in SPTBN, the two species of Cistugo lack this indel character and are instead characterized by a unique insertion in PRKC1, which in turn is not present in the other vespertilionids sequenced (fig. 1). This observation indicates that there may be some validity to the hypothesis that Cistugo is a distinct family. However, more comprehensive taxonomic sampling within the Vespertilionidae needs to be conducted before this question can be adequately addressed.
Evolution of Echolocation
The following discussion is based on the assumption that there was no change in echolocation characters between their time of origin and their present state in extant families. Based on mapping of echolocation characters on the supermatrix topology, if one origin of echolocation in the bat lineage is assumed, the ancestral crown group protobat probably produced echolocation calls of low-duty cycle and high intensity. Whether the ancestral bat was an oral or nasal emitter is equivocal. If two separate origins of echolocation are inferred, the ancestor of Vespertilioniformes was likely an oral emitter producing low-duty cycle, high-intensity calls. The ancestor of the Rhinolophoidae, on the other hand, was probably a nasal emitter producing calls of high intensity. Whether this ancestor produced high-duty or low-duty cycle calls is equivocal because of the sister taxa status of the rhinolophid-hipposiderid and megadermatid-rhinopomatid clades. Regardless of whether echolocation evolved once or twice in the bat lineage, high-duty cycle echolocation has evolved at least twice, at least once in the Pteropodiformes, and once in the Vespertilioniformes. In the latter lineage, it is a derived form of echolocation evolving from a low-duty cycle ancestor.
Low-intensity echolocation calls have evolved independently at least six times in the bat lineage, with high-intensity echolocation pulses the ancestral condition. Similarly, there may have been up to three independent origins of nasal echolocation in Chiroptera—once in the rhinolophoid lineage, once in the nycterids, and once in the phyllostomids, assuming two origins of echolocation or one origin of echolocation with an oral-emitting protobat. There have been at least two independent transitions from oral to nasal echolocation within the Vespertilioniformes lineage: once in the nycterids and once in the lineage leading to the phyllostomids.
Reevaluation of Morphological Characters and Placement of Fossil Bat Taxa
Fifty-nine nonhomoplasic morphological characters were identified that did not contradict microbat paraphyly. These represent 31 soft tissue and 28 hard tissue characters. Of these, a total of 24 were scored for at least one fossil taxon and were used in subsequent analyses. Regardless of the backbone molecular scaffold used, the strict consensus topology was unresolved with respect to placement of the fossil taxa relative to extant taxa (supplementary fig. 1, Supplementary Material online). Reevaluating the morphological characters in light of the intron supermatrix topology, character 82 of Simmons and Geisler (1998) was found to be a synapomorphy for Rhinolophoidea as defined in this study. Rhinolophids, hipposiderids, megadermatids, rhinopomatids, and Craseonycteris are all characterized by ossification of their first costal cartilage, which is fused to the manubrium as well as to the first rib, whereas all other bat families do not have this character. Another synapomorphy of the Rhinolophoidea is the presence of one pair of pubic nipples in females (character 176 of Simmons and Geisler 1998). The presence of a triangular flange on the anteromedial edge of the scapula was a morphological synapomorphy for the clade comprising vespers, miniopterids, molossids, natalids, and M. aurita. No projections or flanges are present on the anteromedial edge of the scapula of other bat families.
Molecular Clock and Biogeography
The molecular clock hypothesis was rejected for the combined DNA data set by a LRT under the general time reversible (GTR) + model with invariant sites + 
model: 
= 2 x (ln L unconstrained – ln L clock) = 2 x [–36 114.20 – (–36 356.72)] = 242.52; df = 56; P < 0.0001. This indicates that there is extensive rate variation among lineages that precludes the application of the linearized tree method (Takezaki, Rzhetsky, and Nei 1995). Thus, the use of a relaxed molecular clock approach designed to accommodate rate variation is preferable for estimating divergence ages with this data set.
Independent Markov chains initiated from different starting points converged on the same divergence times. Estimates of divergence dates based on posterior estimates were characterized by much smaller SDs and narrower credibility intervals than the prior estimates (table 5). Allowing genes to evolve with independent rates or allowing them to evolve with correlated rates resulted in almost identical divergence estimates. Estimates of divergence time were also remarkably robust to removal of single time constraints. The largest change observed was when removing the constraint on the upper divergence of Megaderma and Rhinopoma, which resulted in a systematic increase in divergence date estimates, although the values were still well within the 95% credibility intervals obtained using all six time constraints. Divergence estimates based on individual introns displayed far wider credibility intervals and larger SDs than the divergence estimates based on the supermatrix (data not shown). Interestingly, in all 11 independent runs based on the complete intron supermatrix, the null hypothesis that STAT5A versus SPTBN, and STAT5A versus PRKC1 evolve independently was rejected.
|
The molecular clock applied herein suggests that the first divergence among chiropterans dates back to approximately 62 MYA ± 4 MYA (table 5). It also suggests that by 33 MYA at least 17 of the 18 extant chiropteran families were present (table 5).
Diva analysis suggests an African origin for the ancestor of extant Chiroptera, with 39 dispersal events and three vicariant events required to explain the current distribution of chiropteran families (fig. 2).
|
|
Discussion |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Resolving Power of Nuclear Intron Sequences
The limitation of using single markers for resolving family-level chiropteran relationships was highlighted by the results obtained from analyses of individual nuclear intron fragments (table 3). The large number of equally parsimonious trees found for PRKC1, STAT, and THY and the large number of trees present in the 95% posterior probability interval further emphasizes the poor resolving power of individual nuclear markers (table 2). In short, these results agree with previous studies that emphasize the importance of a supermatrix approach for improving phylogenetic resolution (Baker and DeSalle 1997
; Murphy et al. 2001a, 2001b; Buckley et al. 2002), particularly when the radiation was rapid resulting in short internal branches (Matthee et al. 2001, 2004; Willows-Munro, Robinson, and Matthee 2005).
Chiropteran Phylogeny
Despite different degrees of taxonomic sampling, both within and among families (Hutcheon, Kirsch, and Pettigrew 1998; Teeling et al. 2000, 2002, 2003, 2005; Hutcheon and Kirsch 2004; Van Den Bussche and Hoofer 2004), the paraphyly of the microbats is a consistent finding across studies, separating bats into the superordinal clades Pteropodiformes and Vespertilioniformes.
Another area of agreement among studies is the superfamily Noctilionoidea (Springer et al. 2001; Teeling et al. 2003, 2005; Van Den Bussche and Hoofer 2004). Within this superfamily, the sister taxon relationship between Phyllostomidae and Mormoopidae (Kirsch et al. 1998; Van Den Bussche and Hoofer 2000, 2001, 2004; Teeling et al. 2003) and Furipteridae and Noctilionidae (Van Den Bussche and Hoofer 2001, 2004; Hoofer et al. 2003, Teeling et al. 2005) are well supported. Similarly, the sister taxon status of the emballonurids and nycterids reported by others (Teeling et al. 2000, 2002, 2003, 2005) and the superfamily Vespertilionoidea (Van Den Bussche 2004, miniopterids) is also supported by our study. Within the Pteropodiformes, there is widespread support for a clade comprising rhinolophids, hipposiderids, megadermatids, and rhinopomatids (Rhinolophoidea) based on both molecular (Teeling et al. 2000, 2002, 2003, 2005) as well as two morphological synapomorphies, namely ossification of the first costal cartilage which is fused to the manubrium as well as to the first rib and the presence of one pair of pubic nipples in the female. Furthermore, the rhinolophid-hipposiderid + megadermatid-rhinopomatid sister taxon relationship is well supported by this and other studies (Teeling et al. 2000, 2002, 2003, 2005), with the Old World fruit bat lineage consistently reconstructed as the sister taxon to this rhinolophoid microbat clade, indicating microbat diphyly. Thus, microbats are paraphyletic.
Areas of phylogenetic uncertainty include branching patterns within the superfamily Vespertilionoidea, with the association of the miniopterids with the vespers or molossids being uncertain. However, the basal position of the natalids to these three families, presented in this study, also receives support from previous studies (Hoofer et al. 2003; Teeling et al. 2003). A clade comprising the nycterids-emballonurids sister to the superfamily Noctilionoidea received moderate nodal support in our study, in agreement with Teeling et al. (2002, 2003, 2005), but is absent from other studies, indicating that this node may not be completely stable. The precise placement of Mystacinidae within the superfamily Noctilionoidea, as well as the branching order of the families within this clade requires further investigation. A further area of uncertainty concerns the precise placement of M. aurita. This taxon is characterized by a long branch and was placed basal to the superfamily Vespertilionoidea in this study. The association of M. aurita with this clade in the intron supermatrix, although not well supported by molecular characters, is also supported by the synapomorphic triangular anteromedial flange on the anteromedial edge of their scapula. A possible association between Myzopodidae and Natalidae was previously suggested based on the resemblance of the myzopodid skeletal structure (Miller 1907). However, in a study based on Rag 2 nuclear data (Hoofer et al. 2003), M. aurita was basal to all Vespertilioniformes, and in a study by Teeling et al. (2005) M. aurita was associated with the Noctilionoidea. Additional taxonomic sampling of vespertilionids and the inclusion of different types of data, such as SINE insertions (Kawai et al. 2002; Murphy, Pevzner, and O'Brien 2004; Pecon-Slattery et al. 2004), are clearly needed to resolve this issue.
Although Springer et al. (2001) and Teeling et al. (2005) found phylogenetic resolution among extinct and extant bat taxa, our analyses suggest that much of this resolution was obtained from homoplasious characters. While we recognize the importance of incorporating morphological data into phylogenetic reconstructions, especially for fossil data which can provide vital information (Wiens 2004), the high level of parallel evolution when using morphological characters is problematic. This convergence most likely results from the constraints imposed by the key innovations of flight coupled to echolocation, as well as adaptation to particular ecological niches (Ruedi and Mayer 2001).
Evolution of Echolocation
The unresolved position of the fossil taxa casts doubt on explanations of the evolution of echolocation that are based on the basal position of these fossil bat taxa (Springer et al. 2001). Although chiropteran monophyly provides strong evidence that flight has evolved only once (Simmons and Geisler 1998), microbat paraphyly makes the evolution of echolocation less clear. The molecular phylogeny presented here is consistent with either two independent origins of echolocation within the chiropteran lineage or a single origin (Teeling et al. 2000, 2002, 2003, 2005). A single origin of echolocation requires that echolocation was secondarily lost in the pteropid lineage and regained in rudimentary form in Rousettus. Two independent origins of echolocation propose that it arose in the lineage leading to the Rhinolophoidea, as well as in the ancestor of the Vespertilioniformes.
The mapping of echolocation-call design onto our phylogeny does not appear to resolve the question of whether echolocation had one or two origins. Although the echolocation-call design and processing in rhinolophids and hipposiderids, who use high-duty cycle echolocation calls and Doppler-shift compensation along with an acoustic fovea to separate call and echo in frequency rather than time, is fundamentally different from the low-duty cycle calls used in all Vespertilionformes (except one), rhinopomatids and megadermatids also use low-duty cycle echolocation. High-duty cycle echolocation is not therefore a synapomorphy for the Rhinolophoidea, which would have provided strong support for two independent origins of echolocation.
Much like morphological characters, echolocation seems to be highly convergent. For example, high-duty cycle echolocation and the associated mechanical and neurological tuning required to exploit Doppler-shifted echoes has evolved convergently in the mormoopid P. parnelli (Kossl et al. 1999). Similarly, low-intensity echolocation calls, often associated with gleaning (Faure, Fullard, and Barclay 1990; Miller and Treat 1993), have evolved independently at least six times in both Vespertilioniformes and Pteropodiformes lineages (i.e., in the Phyllostomidae, Thyropteridae, Nycteridae, Furipteridae, Vespertilionidae, and Megadermatidae). The nasal-oral emitting dichotomy also shows no phylogenetic pattern, with at least two switches from oral emission to nasal emission within the Vespertilioniformes (Nycteridae and Phyllostomidae) and three independent origins of nasal echolocation within the Chiroptera. The latter is supported by morphological data. Whereas phyllostomid skulls lack resonating chambers and instead are characterized by large olfactory fossae and a well-developed voremonasal complex, nycterid skulls possess resonating chambers situated external to the bony nasal cavity. This in turn is different to the skull morphology of megadermatids, rhinolophids, and hipposiderids, where resonating chambers are formed by the nasal cavities and lie within the skull. The olfactory fossae and voremonasal complex are also distinctly smaller in the rhinolophoids and nycterids compared to the phyllostomids (Pederson 1993, 1995, 1998). Convergent evolution is also evident in the co-occurrence of low-intensity echolocation calls and nasal emission in three lineages—Megadermatidae, Nycteridae, and Phyllostomidae. This might indicate some association between nasal emission and the inability to produce echolocation calls of high intensity (Pederson 1993, 1995), which in turn may constrain foraging options. This may hold true particularly for the phyllostomids—they possess neither resonating chambers nor a "tuned" rostrum, so their emission is muffled in the nasal passages, resulting in low-intensity echolocation calls (Pederson 1998). Further investigation is required, however, to evaluate if the low-intensity echolocation calls produced by nycterids and megadermatids, both of which possess resonating chambers, can be linked to distinctive morphological features of their pharynx and skulls.
It is perhaps unsurprising that there appears to be no phylogenetic patterning to any of the three dichotomies summarizing echolocation behavior in bats. Echolocation-call design in bats arises as a result of strong selective pressures intimately linked to the ecological and environmental conditions bats are exposed to when navigating or searching for food (Harbesetzer 1981; Schintzler and Kalko 2001; Schintzler, Moss, and Denzinger 2003; Von Helversen and von Helversen 2003). This results in remarkable congruence in echolocation behavior and call design among bats foraging in similar habitat types (e.g., uncluttered space, background-cluttered space, or highly cluttered space) independent of phylogenetic associations (Surlykke et al. 1993; Schintzler and Kalko 2001; Denzinger, Kalko, and Jones 2004; Schintzler, Kalko, and Denzinger 2004).
Although the three echolocation characters discussed above do not lend unambiguous support for two independent origins of echolocation, one of the morphological synapomorphies characterizing the Rhinolophoidea suggests a separate origin of echolocation in this clade. All rhinolophoids have an ossified first costal cartilage fused to the manubrium and first rib. This may be an adaptation for decreasing the energetic costs associated with echolocation from a stationary position (Speakman, Anderson, and Racey 1989; Speakman and Racey 1991; Speakman et al. 2004), which strongly suggests that in this clade, echolocation may have developed in a perch-hunting, gleaning ancestor. The protorhinolophoid was likely a perch-hunting, flying, nocturnal or crepuscular small mammal which used passive cues such as prey-generated sound and vision to localize and detect prey on the substrate and flight to get from branch to branch. Ossification of the first costal cartilage and fusion of this to the rib and manubrium would have allowed energetically inexpensive production of echolocation calls while stationary (Speakman et al. 2004). Rhinolophoids (rhinolophids, hipposiderids, megadermatids) that echolocate while stationary also share modifications with other bats such as nycterids, which also echolocate while stationary; for example, they all possess a first rib at least twice the width of other ribs (character 81 Simmons and Geisler 1998) and have a second rib that articulates with the manubrium with no contact between the rib and mesosternum (character 83 Simmons and Geisler 1998), both of which presumably play some role in decreasing the costs of echolocation while stationary. Given that the two groups are not closely related, these modifications have evolved convergently in the nycterids and gleaning rhinolophoids. This is in contrast to character 82 of Simmons and Geisler (1998). This character is absent in nycterids (and other Vespertilioniformes families) but present in all rhinolophoid microbats. Thus, the presence of this character in all rhinolophoid families and the absence in all Vespertilioniformes indicate that echolocation may have had two independent origins within Chiroptera.
In conclusion, although the echolocation characteristics evaluated on the intron supermatrix do not provide clear support for two independent origins of echolocation in bats, the presence of a unique synapomorphy linked to echolocation in rhinolophoid microbats suggests that there may have been two independent origins of echolocation in bats. Furthermore, the advantages that echolocation confers upon an organism makes it unlikely that echolocation, once evolved, would have been lost in the pteropodids, only to be regained by Rousettus, as required by a single origin of echolocation (Arita and Fenton 1997; Speakman 1999, 2001). Assuming parallel evolution of echolocation in the Vespertilioniformes and rhinolophoid microbats, the numerous similarities in echolocation in these two groups are presumably due to homologous developmental pathways underlying the ability to echolocate.
Biogeography and a Molecular Clock
On the basis of their current distribution patterns and thermoregulatory abilities, it has been hypothesized that bats originated somewhere in the Old World tropics (Legendre 1980; Hand 1984; Hall 1989; Hall and Woodside 1989; Hand et al. 1994), and paleontological evidence suggests that bats had their origins within the tropical forests of Laurasia (Cracraft 1973; Hand 1984). Sige (1991), however, hypothesized that modern bat groups evolved from isolated immigrant archaic groups somewhere in the Southern Hemisphere. Unfortunately, the bat fossil record is depauperate (Hand 1984) and is biased towards Europe and to a lesser extent North America, with very few early Asian, African, and South American examples (Savage and Russell 1983; McKenna and Bell 1997). Teeling et al. (2005) recently proposed that bats originated in Laurasia based on a phylogeny including fossil bats and extant taxa. However, we found little phylogenetic resolution between fossil taxa and extant bats using a similar approach where we limited our choice of morphological characters to those that did not contradict microbat paraphyly. Therefore in this study, we focused on evaluating biogeographic hypotheses for extant taxa only.
The Diva reconstruction of Africa as the center of origin of modern-day bat families is in accord with a Southern Hemisphere origin (fig. 2). The relaxed date estimates from our study are congruent with those of previous studies that place the diversification of extant Chiroptera at the Cretaceous-Tertiary boundary approximately 65 MYA (Springer et al. 2003; Teeling et al. 2003; Delsuc, Vizcaino, and Douzery 2004). This was rapidly followed by diversification of the superfamilies, and from our analyses it seems evident that extant bat families appear to have radiated fairly rapidly, with all families having evolved before the late Eocene. This deep, rapid radiation is supported by short internal branches near the base of the radiation (fig. 1).
Th