Taxa

We selected megachiropteran taxa in order to test monophyly of Harpyionycterinae as currently defined (Harpyionycteris, Dobsonia, and Aproteles) and investigate relationships of these taxa to other genera of megabats as previously proposed in the literature. Specifically, we sought to test (1) the association of Boneia, Rousettus, and Harpyionycteris proposed in the description of the latter (Thomas, 1896) versus the association of Boneia with Rousettus (Andersen, 1912; Bergmans and Rozendaal, 1988) and Harpyionycteris with Dobsonia (Andersen, 1912; Giannini et al., 2006); (2) the sister relationship of Aproteles and Dobsonia (Colgan and da Costa, 2002; Giannini and Simmons, 2003, 2005; Menzies, 1977; Romagnoli and Springer, 2000); and (3) the inclusion of Dobsonia and their relatives in a pteropodine clade (Romagnoli and Springer, 2000) versus a rousettine clade (Andersen, 1912; Bergmans, 1997).

Megachiropteran terminals were chosen from each main pteropodid clade following Giannini et al. (2006): Nyctimene albiventer and N. vizcaccia (Nyctimeninae), Cynopterus sphinx and Ptenochirus jagori (Cynopterinae), Boneia bidens, Eonycteris spelaea, Rousettus amplexicaudatus, and R. aegyptiacus (Rousettinae), Myonycteris torquata and Megaloglossus woermanni (Epomophorinae, Myonycterini), Epomops franqueti and Epomophorus wahlbergi (Epomophorinae, Epomophorini), Macroglossus minimus and Syconycteris australis (Macroglossinae), Melonycteris fardoulisi and Melonycteris melanops (Pteropodinae, Notopterini), and Pteropus hypomelanus and P. tonganus (Pteropodinae, Pteropodini). Harpyionycterinae megabats (sensu Giannini et al., 2006) included are: Aproteles bulmerae, Harpyionycteris whiteheadi, H. celebensis, and a member of each of the four currently recognized Dobsonia species groups (Simmons, 2005), namely Dobsonia minor (minor species group), inermis (viridis group), D. moluccensis (moluccensis group), and D. peronii (peronii group). Here D. magna is considered a junior synonym of D. moluccensis following Helgen (2007; see also Byrnes, 2005; Koopman, 1994).

Trees were rooted using Rhinopoma hardwickii (Rhinopomatidae) and Artibeus jamaicensis (Phyllostomidae). Rhinopomatids are currently thought to belong to Rhinolophoidea, which is the sister clade of megabats in multigene phylogenetic analyses, while phyllostomids represent the other major clade of echolocating bats, Yangochiroptera (Eick et al., 2005; Teeling et al., 2005; Miller-Butterworth et al., 2007).

Sequences

We used sequences of exon 28 of the von Willebrand factor gene (vWF, 1231 bp) and cytochrome b (cytb, 1140 bp) from Giannini et al. (2006), and generated new sequences for harpyionycterine species for this study. We also incorporated sequences of two other nuclear coding genes, the recombination activating gene 1 (RAG1, 1084 bp), and the recombination activating gene 2 (RAG2, 760 bp) from Giannini et al. (2008) for the same taxa, and similarly generated new sequences for harpyioncterines. In order to include relevant species whose samples were not available to us, we used published sequences of the mitochondrial 12S rDNA (12S, ca. 970 bp) and Valine tDNA (tVal, ca. 70 bp) and contributed new sequences for these genes. For additional analyses (see below), we incorporated published sequences of the mitochondrial 16S rDNA (16S, ca. 1560 bp for complete taxa) and NADH dehydrogenase subunit 2 (ND2, 1044 bp) genes, and the oocyte maturation factor Mos proto-oncogene (c-mos, ca. 480 bp). Composition of the data set, accession numbers of all sequences used, and voucher information for new sequences are given in appendix 1.

Total DNA was obtained from preserved tissue samples with the DNeasy tissue kit (QIAGEN). PCR amplification was carried out using previously published primers (12S-tVal: Springer et al., 1995; cytb: Bastian et al., 2002; RAG1 and RAG2: Teeling et al., 2000; vWF: Porter et al., 1996). To obtain both forward and reverse sequences for each gene region, additional internal primers were developed for the RAG1 and vWF genes (table 1). All sequences were obtained with an automated ABI 3730XL sequencer. Sequence editing and prealignment were done with the Sequencher 4.2 software (Gene Codes). Accession numbers for the new sequences are FJ218461-84. Although for some species more than one sample was available, only one was included in each analysis. The aligned dataset is provided at  ftp://ftp.amnh.org/pub/group/mammalogy/downloads/.

Phylogenetic Analyses

Sequences of our coding and ribosomal genes were submitted to parsimony analysis. In the present study, we used data for the genes for which we contributed new sequences (i.e., 12S-tVal, cytb, RAG1, RAG2, and vWF) to conduct three principal sets of phylogenetic analyses: (1) individual gene partitions; (2) nuclear and mitochondrial partitions; and (3) all genes combined. Here we report results of the main partitions and the combined set, and comment on results from the individual genes where relevant. In additional analyses, we used genes not sequenced in our study (16S, ND2, and c-mos in individual and combined sets; see details below).

Analyses involving only coding genes (individual analyses for c-mos, cytb, RAG1, RAG2, vWF, and the nuclear partition) were performed under conventional parsimony (static alignment). The tree search strategy consisted of 500 replicates of random addition sequences of taxa (RAS) each followed by tree bisection reconnection branch swapping (TBR). An additional round of TBR was done on optimal trees so obtained. Clade support was assessed using Bremer or decay values (Bremer, 1994) and character resampling (Goloboff et al., 2003). We calculated Bremer values via incremental sampling of suboptimal trees (see Giannini and Bertelli, 2004). Briefly, we saved up to 2000 suboptimal trees one step longer than the previous optimum in successive stages. That is, we first searched for suboptimal trees 1 step longer than the optimal tree length, next saving suboptimals at every step from 2 through 10 steps longer than the optimal trees. Second, group frequency (based on unbiased symmetric resampling; Goloboff et al., 2003) was calculated on the basis of 5000 replications. These analyses were executed in TNT 1.0 (Goloboff et al., 2004, 2008).

For all analyses involving ribosomal and transfer RNA genes (i.e., individual 12S-tVal partition, the mitochondrial partition, the combined set, and additional analyses using 16S), we used direct optimization (DO; Wheeler, 1996). In this approach, alignment is viewed as dynamic (i.e., varying across topologies in tree space) and is therefore linked to tree search. The transformation cost of whole unaligned sequences on a topology is calculated via optimization. Costs of each transformation type (indel or substitution) were established a priori—as in every alignment procedure. According to these costs, the length of the tree is calculated as the sum of transformations on each internal node in the downpass optimization, which minimizes indels and substitutions (steps) of whole sequences on the candidate tree. Minimum-length trees are chosen among those visited. In practice, this is done by way of tree building and branch swapping. We set equal costs of indels and substitutions, and our search strategy consisted of replicated swapping + refinements (e.g., Giannini and Simmons, 2003, 2005). Briefly, 100 RAS + TBR branch swapping were run, collecting all optimal trees from each replication. Those trees were submitted to tree fusing (Goloboff, 1999). In analyses combining ribosomal and coding genes, the latter were treated as prealigned; i.e., no indels were allowed in coding genes (static alignment). All DO analyses, including searches to calculate Bremer support values, were performed using POY 4 (Varon et al., 2007).

Additional phylogenetic analyses were conducted to take advantage of additional published data. First, we added published 16S and c-mos sequences to our combined set to provide a “total-evidence” analysis based on sequences available for the majority of our taxonomic samples. Second, we ran a specific analysis of ND2 with the few sequences of megabats available in GenBank (which unfortunately had low overlap with our taxonomic sampling) to provide a separate test of the affinities of Boneia. Data available included ND2 sequences of Dobsonia minor among the harpyionycterines and also Boneia bidens. This analysis was important because our data on Boneia originated from a single individual (ZMA 23100) from which we extracted DNA and generated all sequences, whereas the ND2 sequences originated from DNA that was amplified from a different individual (GenBank accession number AY504581).

In addition to the parsimony analyses, we performed a maximum-likelihood (ML) tree search for the combined nuclear data set using a GTR + Γ + I partitioned model (i.e., parameters were estimated from the data for each gene separately) using the program RAxML (Stamatakis, 2006; Stamatakis et al, 2008). Our search strategy consisted of executing 100 rapid bootstrap inferences and thereafter a thorough ML search. Statistical support was obtained with 100 bootstrap replications.

Biogeographic Analysis

Our biogeographic analysis consisted of mapping onto the optimal tree obtained from our total-evidence phylogenetic analysis a single biogeographic character composed of six states, each of which consisted of known distribution areas of terminals (see below). We performed a partial (downpass-only) Fitch (1971) optimization in order to recover the pattern of minimal biogeographic events required by the tree. In this case, we did not intend to provide wide coverage of Megachiroptera; our aim was instead to identify the main dispersal-vicariance events of our ingroup, the Harpyionycterinae—those events that led to the origin of currently recognized genera. This attempt is necessarily incomplete as we did not include all Dobsonia species. Therefore we restricted our analysis to the sampled harpyionycterines and their areas of occurrence, which were coded in a single multistate, unordered character whose states consisted of broad biogeographic subregions of the eastern Austromalayan region as recognized by Corbet and Hill (1992): Sulawesi (SUL), the Philippines (PHI), Lesser Sunda Islands (LSI), the Moluccas (MOL), New Guinea (NGU), Australia (AUS), and the Solomon Islands (SOL). Taxa occurring in more than one area were scored as polymorphic (for instance, Dobsonia minor was scored NGU/SUL).

Nuclear Partition

Parsimony analysis of nuclear data (RAG1, RAG2, and vWF sequences) resulted in 10 most-parsimonious trees of 1922 steps (strict consensus tree in fig. 1). In this tree, Megachiroptera is highly supported (Bremer support value, BS  =  38). Mutually monophyletic Nyctimeninae and Cynopterinae form a clade that corresponds with Andersen's (1912) Cynopterus section, and is sister to a poorly supported multichotomy of four clades containing all other megachiropterans. The four groups are: a reduced macroglossine clade (Macroglossus + Syconycteris); a reduced pteropodine clade including Melonycteris and Pteropus (see Bergmans, 1997; Giannini et al., 2008); a clade composed of rousettine (Rousettus and Eonycteris) and epomophorine megabats, including Myonycterini (Megaloglossus and Myonycteris) and Epomophorini (Epomophorus and Epomops); and a harpyionycterine clade (sensu Giannini et al., 2006) that includes Boneia as sister to Harpyionycteris, and both sister to Dobsonia. Our ML analysis (fig. 2) recovered the same major groupings as the parsimony analysis (fig. 1), resolving the backbone polytomy of the parsimony result (cf. fig. 1) with varying degrees of support (bootstrap values 32–82). This topology is identical to one of the optimal topologies obtained under parsimony (i.e., in this analysis, parsimony is conservative in reflecting conflicts in the data). Regarding harpyionycterine megabats, the branching pattern is the same in both ML and MP analyses (cf. figs. 1 and 2).

Figure 1

Strict consensus tree resulting from a parsimony analysis of three nuclear coding genes combined (RAG1, RAG2, and vWF). Bremer support values are given above branches and resampling (jackknife) frequencies are given below branches.

Figure 2

Maximum likelihood tree resulting from the analysis of the nuclear dataset (RAG1, RAG2, and vWF; gene partitions modeled using GTR + Γ + I; ML score -11,136.227759). Values near nodes represent bootstrap estimates of clade support based on 100 replications.

Mitochondrial Partition

The analysis of the 12S-tVal + cytb sequences under direct optimization produced a single tree at 3220 steps (fig. 3). Megachiroptera is highly supported (BS  =  36), but much of the backbone of the megachiropteran subtree is not very well supported (2 ≤ BS ≤ 6). Melonycteris appears as sister taxon to Nyctimene, Syconycteris, and a large group composed of the remainder of megachiropterans. The latter is further subdivided into a clade containing (Pteropodini + Cynopterinae), the previously reported rousettine-epomophorine clade, and another version of the harpyionycterine clade including Aproteles, Boneia, Dobsonia, Harpyionycteris, and, surprisingly, Macroglossus. In this clade, all intergeneric relationships are weakly supported (1 ≤ BS ≤ 2).

Figure 3

Single optimal tree resulting from a direct optimization, parsimony analysis of three mitochondrial genes combined (12S-tVal + cytb). Bremer support values are given below branches.

Combined and Additional Analyses

A combined parsimony analysis using both the nuclear genes (RAG1, RAG2, and vWF) and our mitochondrial data (cytb and 12S-tVal) resulted in three trees of 4451 steps (strict consensus in fig. 4). In this analysis, Megachiroptera is highly supported (BS  =  97). Melonycteris is sister to a trichotomy including Macroglossus, Syconycteris, and a poorly supported (BS  =  3) group containing the remainder of terminals, which is further subdivided into a harpyionycterine clade ((Harpyionycteris + Boneia) + (Aproteles + Dobsonia)), and another group containing the rousettine-epomophorine clade and a composite clade (Pteropodini (Nyctimeninae + Cynopterinae)).

Figure 4

Strict consensus tree of a parsimony analysis of RAG1, RAG2, vWF, cytb and 12S-tVal genes combined. Bremer support values are given below branches.

Adding to this data set published sequences of the mitochondrial 16S and the nuclear oncogene c-mos genes produces a single tree of 6119 steps (fig. 5). In this tree, Nyctimene is sister to all other megachiropterans, and this tree partition appeared highly supported (BS  =  19). Successive clades recovered were (Pteropodini + Cynopterinae), the rousettine-epomophorine clade, and a highly supported harpyionycterine clade (BS  =  23) in which Dobsonia was sister to Aproteles, Boneia, and Harpyionycteris. Several important dispersal-vicariance events are suggested by mapping distribution of terminals onto the optimal tree of our combined phylogenetic analysis. These patterns are discussed below.

Figure 5

Single optimal tree of a parsimony analysis of RAG1, RAG2, vWF, cytb, 12S-tVal, 16S, and c-mos genes combined. Bremer support values are given below branches.

Finally, we obtained four trees of 1717 steps in a separate analysis based on ND2 sequences from pteropodid species available on GenBank. The strict consensus tree (fig. 6) was generally poorly supported but placed Boneia bidens and Dobsonia minor as sister taxa. This result is compatible with those of previous analyses that separately support an association of Boneia with harpyionycterine megabats—D. minor among the latter.

Figure 6

Strict consensus tree of a parsimony analysis of the ND2 gene. Bremer and jackknife values (above and below branches, respectively) are reported for the harpyionycterine clade recovered in this analysis. Numbers in parenthesis indicate number of terminals included in the cynopterine and pteropodine clades recovered.

Phylogenetic Relationships and Classification

The phylogenetic analyses of nuclear data recovered a harpyionycterine clade composed of Harpyionycteris, Aproteles, Dobsonia, and, rather unexpectedly, Boneia. Although Macroglossus was added to this group in the analysis of the mitochondrial partition (with negligible support, 1 ≤ BS ≤ 2; fig. 3), the group membership of the nuclear analysis was recovered in the total-evidence analysis with high support (BS  =  23; fig. 5). This represents strong evidence that Boneia is not a synonym of Rousettus but a distinct genus that belongs in Harypionycterinae. We also confirmed the inclusion of Aproteles in this group (see discussion in Giannini et al., 2006). Thus, we formally expand Harpyionycterinae Miller, 1907, to include Boneia Jentink, 1879, Harpyionycteris Thomas, 1896, Dobsonia Palmer, 1898, and Aproteles Menzies, 1977.

Resolving the composition of Harpyionycterinae has proven difficult throughout the history of megachiropteran systematics. Thomas (1896) recognized the difficulties of finding allies of Harpyionycteris among the megachiropterans known at the time of its description. Thomas (1896: 243) attributed the similarities in canine morphology with Harpyia ( =  Nyctimene) to either homoplasy (“accident” in his words) or plesiomorphy (i.e., “common descent from the (presumably) cuspidate-toothed ancestor of the Pteropodidae”). Thomas (1896: 243) suggested that “On the whole … [Harpyionycteris] may be most conveniently placed near Xantharpyia [ =  Rousettus] and Boneia, with which it shares certain external characters, an indical claw, and the cheek-tooth formula of P. 3/3, M. 2/3.” The index finger claw and the cheek-tooth formula are likely plesiomorphic for Megachiroptera (see Giannini and Simmons, 2005, 2007), and the external characters were not specified (but see the external resemblance of Harpyionycteris celebensis and Boneia bidens in fig. 7). Nevertheless, it seems clear that Thomas (1896) did correctly perceive a link between these forms.

Figure 7

External appearance of live specimens of Harpyionycteris celebensis (top) and Boneia bidens (bottom). Photos: courtesy Jan Haft (with permission).

Despite Thomas's (1896) intuition and the molecular support discovered in the current study, to our knowledge, no author subsequent to Thomas considered a subfamily or tribal group inclusive of both Harpyionycteris and Boneia. This is reflected in a complete lack of potential morphological synapomorphies for such grouping in the literature. In recent systematic studies involving Boneia, discussion has centered on the nature of its relationship with Rousettus (e.g., Bergmans, 1994; Bergmans and Rozendaal, 1988). This is not surprising given the prominent role that Rousettus has played in megachiropteran systematics as a result of long being considered close to the ancestral form of megabats following Andersen (1912). Although this assessment has been refuted by all cladistic studies since the early 1990s (e.g., Colgan and da Costa, 2002; Giannini and Simmons, 2003, 2005; Kirsch et al., 1995; Romagnoli and Springer, 2000; Springer et al., 1995), most megachiropteran genera have been compared with Rousettus on these grounds, and Boneia was no exception. Given the lack of unusual external features, unremarkable skull shape, dental formula close to the generalized condition of megabats, and the habit of roosting gregariously in caves, Boneia was considered so close an ally of Rousettus as to be included in Rousettus as a subgenus (see Bergmans and Rozendaal, 1988). The morphological evidence discussed by the authors in favor of this association includes both morphometric and qualitative data and is rather robust, which makes our finding of a Boneia-Harpyionycteris clade more surprising. As shown in fig. 8, the skulls of Boneia and Harpyionycteris bear little mutual resemblance. In Boneia, the rostrum is thin and strongly deflected relative to the basicranial axis, the dentition is simplified but otherwise typically pteropodid, the zygomatic arch is weak, and in overall appearance it resembles Rousettus. By contrast, Harpyionycteris shows a strong “dobsonine” skull with the most peculiar rostrum and dentition seen in megabats (see comments in Giannini et al., 2006).

Figure 8

Lateral (upper row) and ventral (lower row) views of the skulls of Rousettus amplexicaudatus (A, AMNH 226391), Boneia bidens (B, AMNH 254542), and Harpyionycteris whiteheadi (C, AMNH 142766). Scale  =  5 mm.

However difficult it may seem to reconcile the current composition of the harpyionycterine clade with the morphological variation traditionally considered in megabat systematics, our phylogeny does suggest some interesting scenarios of morphological and ecological evolution. One particularly significant example involves the dentition. Harpyionycterines exhibit dental formulae that vary across taxa, but are similar in the sense that one formula can be parsimoniously transformed into any other in one-step losses or gains. A most relevant case is the first upper incisor, which is the sole tooth missing from the generalized pteropodid formula in Boneia (a variably deciduous tooth in this taxon; Bergmans and Rozendaal, 1988). This is also the incisor demonstrably lacking in Dobsonia and hypothesized to be absent in Harpyionycteris (see Giannini and Simmons, 2007: table 2). Aproteles lacks both upper incisors. Our phylogeny, once it is accepted as a supported hypothesis, allows us to suggest that the loss or reduction of I1 is effectively homologous across harpyionycterine taxa, and that the loss of incisors in Aproteles likely happened sequentially (I1 first lost first, followed by I2).

Table 2

Distribution of Each of the 18 Species of Harpyionycterinae Miller, 1907

Biogeographic regions follow Corbet and Hill (1992).

Biogeographic and Ecological Patterns

Harpyionycterinae sensu Giannini et al. (2006; i.e., composed of Harpyionycteris, Dobsonia, and Aproteles), expanded here to include Boneia, comprises 18 species whose joint distribution encompasses the Philippine, Wallacean, and Papuan subregions of the Austromalayan area, plus two species that narrowly escape the biogeographic boundaries of those subregions (fig. 9; table 2). The latter two taxa are D. peronii, which also occurs in Bali (Sundaic subregion) and D. moluccensis, which also occurs in Queensland, Australia (Corbet and Hill, 1992; Simmons, 2005). Keeping in mind that our biogeographic interpretation rests upon (a) our limited taxonomic sampling (cf. table 2), and (b) the optimal tree from our total-evidence analysis, our biogeographic analysis reconstructed the Papuan subregion as the unambiguous area of origin of harpyionycterines (fig. 9). From this region, the ancestor of Harpyionycteris and Boneia crossed the Lyddekker's and Weber's lines westward to establish populations and speciate in Sulawesi, where differentiation of the two genera occurred. Later, an ancestral Harpyionycteris species split into a lineage that colonized the Philippines (crossing the Wallace line northward) giving rise to H. whiteheadi, and into another lineage that remained in Sulawesi to become H. celebensis.

Figure 9

Joint distribution of species in harpyionycterine megabat genera in the Austromalayan Region (A), and biogeographic patterns in harpyionycterine megabats (B). References for distributions: Aproteles (dots), Boneia (hatched area), Dobsonia (perimeter lines), and Harpyionycteris (shaded areas). Indicated are the Wallace, Weber, and Lydekker lines. Arrows indicate the distribution of Dobsonia species that escape the boundaries of the Wallacean and Papuan subregions: (A) Dobsonia chapmani in Negros and Cebu Islands (Philippine subregion); (B) Dobsonia peronii in Bali (Sundaic subregion, Javan Division; D. peronii is widespread in the Lesser Sunda Islands); (C) Dobsonia moluccensis in Queensland, Australia (D. moluccensis, inclusive of D. magna, is widespread in the Moluccas and New Guinea). Biogeographic patterns: Each species is assigned one or more of the following areas, each of which is a state in a single unordered biogeographic character: AUS Australia, LSI Lesser Sunda Islands, NGU New Guinea, PHI Philippines, SOL Solomon Islands, SUL Sulawesi. Biogeographic events are inferred by partial (downpass only) optimization of distribution areas of species in the harpyionycterine clade in the optimal tree from the most complete, combined analysis (B; see fig. 4).

The dispersal-vicariance events (sensu Ronquist, 1997) that led to the present-day diversity within Dobsonia can only be matter of speculation given our limited sample (but see Byrnes, 2005, for a detailed account). In our analysis, the genus also originated in New Guinea. In a more conventional scenario in which Aproteles is sister to Dobsonia as in the tree in fig. 4, the area of the ancestral harpyionycterine bat is reconstructed as Sulawesi + New Guinea. Accepting a Papuan (or Papuan-Wallacean) origin as probable (fig. 9), and taking together the distribution of all species, several Dobsonia lineages apparently diverged in various directions: northeastward to colonize the Bismarck Archipelago and the Solomon Islands; southward to reach Cape York Peninsula (Australia), and westward to reach the Moluccas, Lesser Sunda Islands, Sulawesi, and beyond the Wallace line in Bali (populations of D. peronii) and the Philippines (D. chapmani, restricted to Negros and Cebu Is.; Alcala et al. 2004; Heaney et al., 1998; Paguntalan et al., 2004). Other speciation events took place in land-bridge islands of New Guinea (Byrnes, 2005).

The optimal tree from our total-evidence analysis also suggests some interesting ecological patterns including insights into the evolution of roosting habits (see also discussion in Byrnes, 2005). Roost availability is one of the limiting factors of bat populations (Kunz and Pierson, 1994). Megachiropterans are either cavity dwellers (occupying structures from tree holes to large caves) or foliage dwellers (a habit that varies from taking solitary refuge on leaves or vines to living in larger groups in actively defoliated trees; Kunz and Pierson, 1994). The roosting habits of Harpyionycteris are unknown, but this genus is nested within successive sister clades that include many typical cave-dwelling bats: Boneia (Bergmans and Rozendaal, 1988), Aproteles (Flannery, 1995; Menzies, 1977), and several Dobsonia species including D. moluccensis (Bonaccorso, 1998; Flannery, 1995; Helgen, 2007), and probably also D. peronii (Hutson et al., 2008) and D. inermis. By contrast, D. minor has been reported to roost both in foliage (Flannery, 1995) and in caves (Boeadi and Bergmans, 1987). The prediction we draw from reconstructing roosting habits (foliage vs. cavities) in the optimal tree of our most complete analysis is that Harpyionycteris could be also a cavity dweller, as are many members of the harpyionycterine clade (fig. 10). It is interesting that the harpyionycterine clade appeared in our analysis as sister to the rousettine-epomophorine clade (with high support in our analysis, BS  =  21). Taken together, these two major megabat clades share a common ancestor that is also reconstructed as a cavity dweller (fig. 10). Once again, limited taxonomic sampling prevents further speculation, but we risk the hypothesis that cavity dwelling may have had a major impact in the diversification of megabats.

Figure 10

Reconstruction of roosting habits (cavity versus foliage) in the most parsimonious tree from the total dataset of this study (fig. 4). Hatched lines indicate ambiguity (in this case, terminals for which both roosting habits were documented). Cavity dwelling is indicated in thick branches. Question marks indicate lack of data for Harpyionycteris. Note that this genus is nested in a clade of cavity dwellers (see text).