Recent and Rapid Speciation with Limited Morphological Disparity in the Genus Rattus

Abstract

Recent and rapid radiations provide rich material to examine the factors that drive speciation. Most recent and rapid radiations that have been well-characterized involve species that exhibit overt ecomorphological differences associated with clear partitioning of ecological niches in sympatry. The most diverse genus of rodents, Rattus (66 species), evolved fairly recently, but without overt ecomorphological divergence among species. We used multilocus molecular phylogenetic data and five fossil calibrations to estimate the tempo of diversification in Rattus, and their radiation on Australia and New Guinea (Sahul, 24 species). Based on our analyses, the genus Rattus originated at a date centered on the Pliocene–Pleistocene boundary (1.84–3.17 Ma) with a subsequent colonization of Sahul in the middle Pleistocene (0.85–1.28 Ma). Given these dates, the per lineage diversification rates in Rattus and Sahulian Rattus are among the highest reported for vertebrates (1.1–1.9 and 1.6–3.0 species per lineage per million years, respectively). Despite their rapid diversification, Rattus display little ecomorphological divergence among species and do not fit clearly into current models of adaptive radiations. Lineage through time plots and ancestral state reconstruction of ecological characters suggest that diversification of Sahulian Rattus was most rapid early on as they expanded into novel ecological conditions. However, rapid lineage accumulation occurred even when morphological disparity within lineages was low suggesting that future studies consider other phenotypes in the diversification of Rattus . [Adaptive radiation; Australia; chromosomal rearrangements; Molecular systematics; Murinae; New Guinea .]

Radiations of species are of particular interest to evolutionary biologists because they provide a window into the mechanisms underlying speciation processes ( Givnish and Sytsma 1997 , Schluter 2000 ). In most examples involving recent and rapid radiations of species (e.g., Darwin's finches, Grant 1986 ; Hawaiian honeycreepers, Freed et al. 1987 ; Hawaiian silverswords, Baldwin and Sanderson 1998 ), the species involved are sympatric and ecologically or morphologically well differentiated. Models for understanding adaptive radiations ( Gavrilets and Losos 2009 , Schluter 2009 ) through a process of ecological speciation have primarily been developed based on these types of cases; indeed, some authors have argued that adaptive radiations should only be recognized where the component species are exceptionally diverse morphologically or ecologically ( Losos and Miles 2002 ). In contrast, most examples of nonadaptive radiations have involved deeply divergent and allopatric species without clear differentiation of ecological or morphological characters (e.g., Achatinella land snails, Gittenberger 1991 ; plethodontid salamanders, Kozak et al. 2006 ; Wake 2006 ; Rundell and Price 2009 ). Alternatively, structural genomic changes (chromosomal rearrangements and polyploidization) can also drive the formation of reproductively isolated species without overt ecological divergence among taxa ( White 1978 , King 1993 , Searle 1993 ), though chromosomal rearrangements also may facilitate adaptive divergence ( Kirkpatrick and Barton 2005 ).

Within Rodentia, which accounts for more than half of all mammalian species, some groups have diversified notably faster than others, principally within the superfamily Muroidea ( Steppan et al. 2004 ). As a whole, rodents exhibit a vast range of morphologies, are found worldwide, and occur in virtually every habitat. However, rapidly diversifying groups within Rodentia often exhibit a narrow range of morphological disparity without clear ecological divergence among species (e.g., Rattus, Microtus, Peromyscus, Oryzomyinae). Moreover, they commonly show chromosomal rearrangements among closely related taxa, in some cases associated with hybrid sterility (e.g., Mus , Hauffe and Searle 1998 ). However, the significance of these structural changes for diversification rates, generally, remains unclear ( Wilson et al. 1974 , Bush et al. 1977 , Patton and Sherwood 1983 , Rieseberg 2001 ).

The genus Rattus represents a classic example of a rapid morphologically limited but chromosomally disparate radiation. With 66 recognized species, it is the most diverse genus of rodents ( Musser and Carleton 2005 ). Rattus species are typically distinguished by a variety of subtle morphological contrasts ( Taylor and Horner 1977 , Taylor et al. 1982, Musser and Holden 1991 ); hence, they lack the ecomorphological diversity observed in most classic adaptive radiations. Chromosomal rearrangements among species are common and laboratory studies have demonstrated that these rearrangements can lead to malsegregation and reduced fertility among closely related species ( Yosida 1980 , Baverstock et al. 1983 ). Given these observations, we might assume that Rattus speciation has been driven by nonadaptive processes. However, our ability to examine this hypothesis in any depth is limited by the paucity of robust phylogenetic studies within the genus.

Although Rattus are now represented on every continent, their natural distribution prior to human transport was restricted to Asia and the western Pacific, from Mongolia south to Australia and from the Philippines west to central Asia ( Musser and Carleton 2005 ). The origin of the genus Rattus almost certainly occurred on continental Asia and most likely Southeast Asia. All other species of the tribe Rattini ( LeCompte et al. 2008 ), in which Rattus is phylogenetically nested, are restricted to Southeast Asia, which is the center of Rattus diversity, and the oldest Rattus fossils have been recovered from from the same place ( Chaimanee 1998 ). Recent molecular studies have estimated an origin of Rattus in the mid Pliocene around 3 Ma ( Furano and Usdin 1995 , Verneau et al. 1998 , Jansa et al. 2006 , Robins et al. 2008 ).

Despite their recent Asian origin, over one-third of Rattus species are found on the southern islands of Australia and New Guinea ( Musser and Carleton 2005 ) which, during Quaternary glaciations, were united to form the “island continent” known as Sahul. Murine rodents, alone among Asian nonvolant mammal lineages, succeeded in crossing the multiple ocean channels separating Asia from Sahul. This occurred at least two times—an early episode of dispersal, leading to the “Old Endemic” murine radiation of Sahul ( Rowe et al. 2008 ), and a second, more recent, crossing to spawn an endemic Sahulian Rattus radiation. An age of 1–2 Ma is often cited for the origin of Sahulian Rattus ( Watts and Aslin 1981 , Breed and Ford 2007 ) but is not well substantiated due to a very sparse fossil record spanning this time period. The genus is abundantly represented in numerous Australian and New Guinean fossil localities of latest Pleistocene and Holocene age ( Wakefield 1960 , Marshall 1973 , Hope et al. 1977 ) including, to date, the earliest dated occurrence of Rattus in Sahul at 149,000 years ago at Mt Etna, Queensland ( Hocknull 2005 ). In contrast, Rattus are conspicuously absent in deposits of late Pliocene and earliest Pleistocene age, including localities that have produced abundant murine remains referable to genera of the Old Endemic radiation ( Godthelp 1990 , 1994 ; Whitelaw 1991 , Tedford et al. 1992, Aplin 2006 , Piper 2007 ). In effect, the time of entry of Rattus to Sahul is currently unconstrained by fossil evidence other than to say that it falls somewhere between the Early and Late Pleistocene.

Ecologically, Sahulian Rattus are more diverse than their Asian counterparts, having colonized every available habitat including closed canopy forests, savannah, open grasslands, montane meadows, and deserts. Yet, despite this ecological breadth, they remain morphologically conservative, varying minimally in body size, pelage texture and color, and tail length ( Taylor and Horner 1977 , Taylor et al. 1982). Every species of Sahulian Rattus is broadly sympatric with at least one other species including examples of syntopy (e.g., Rattus leucopus and R. fuscipes ) with no evidence of hybridization. Thus, despite their morphological similarities, species boundaries are maintained by mechanisms other than geographic isolation. Many species of Sahulian Rattus are distinguishable based on karyotypic differences, primarily Robertsonian fusions ( Baverstock et al. 1977 , 1983 ; Dennis and Menzies 1978 ). However, most crosses between species produce fertile offspring and only crosses between two species, R. colletti (2 N = 42) and R. villosissimus (2 N = 50), result in reduced fertility that can be explained by the malsegregation of chromosomes. This case has been particularly notable because the two species are indistinguishable at 55 electrophoretic protein characters ( Baverstock et al. 1986 ).

The Sahulian Rattus , therefore, provide an exemplary recent and rapid radiation, but one that does not fit clearly into classic models of overt adaptive or nonadaptive radiations. Ecological divergence among Rattus species has not been considered in a phylogenetic context and the tempo and mechanisms underlying the rapid diversification of Rattus warrant further examination. Given the intensive study of R. norvegicus as a model organism in biomedical science, the diversification of Rattus also offers the potential to integrate genomic and phenotypic variation in the context of a recent rapid radiation and to examine the evolution of characters other than morphology that may underlie the rapid diversification of species ( Jacob 1999 ; Aitman et al. 2008 ).

Our objectives in this study are 1) to resolve phylogenetic relationships among Sahulian Rattus , 2) to estimate rates of net diversification for Sahulian Rattus , for the genus Rattus, and for the subfamily Murinae, 3) to examine the relationships between ecological transitions and lineage accumulation in the Sahulian Rattus , and 4) to test the relationship between morphological disparity and speciation in Sahulian Rattus .

MATERIALS ANDMETHODS

Specimens and Genetic Sequencing

Our analyses included data from 69 specimens from 13 species of Sahulian Rattus ( Supplementary Data available from http://www.sysbio.oxfordjournals.org ). These included all six species (13 of 14 subspecies) of Australian Rattus, and 7 of 10 species of New Guinean Rattus (and 4 of 5 subspecies of R. leucopus ) recognized by Taylor et al. (1982; Fig. 1 , Supplementary Data ). All species and subspecies sampled, except R. verecundus and R. fuscipes greyi , were represented by at least two individuals from two sampling localities with up to nine individuals per subspecies ( R. fuscipes coracius and R. leucopus cooktownensis ). In their recent review of Murinae, Musser and Carleton (2005) recognized all of Taylor's species but split three New Guinean species into an additional five species and suggested that four species (two undescribed) of Moluccan Rattus are part of the Sahulian radiation. This view follows Flannery in his reviews of New Guinean and Moluccan mammals ( Flannery 1995a , 1995b ). Unfortunately, these species have not been formally described nor are any tissues available for these additional species. In all analyses of taxonomic diversity, we considered both the 16 species of Taylor and the 24 species of Musser and Carleton. Rattus norvegicus and R. rattus were used as outgroups. Taxonomic identification of specimens was confirmed by careful examination of vouchers (by K.P.A. and K.C.R.) and misidentifications were corrected before genetic analyses (corrections in Supplementary Data ).

FIGURE 1.

Approximate historical distributions of Rattus species examined in this study. Approximate tissue localities are indicated by asterisks. Closed forest species are depicted with black distributions. Open habitat species are depicted with gray distributions. Rattus novaeguineae have been collected in both grassland and forest habitats. Where relevant, subspecies are identified.

Specimens were sequenced for nine unlinked autosomal nuclear loci ( ANT1, Atp5a1, Fgb, Bzrp, BRCA1, DHFR, Fn1, IRBP, RAG1 ) and the mitochondrial control region ( Supplementary Data ). DNA extraction, amplification, and sequencing followed standard procedures described previously ( Rowe et al. 2008 ). Amplification and sequencing primers are listed in Supplementary Data . All loci were sequenced in forward and reverse directions with BigDye v.3 sequencing chemistries using an ABI 3730 capillary sequencer. At least one sequence of each gene was obtained for each species and subspecies, except Fn1 for R. fuscipes fuscipes ( Supplementary Data ). Rattus norvegicus sequences were obtained by a BLAST-N search to the R. norvegicus genome. All loci matched unequivocally to a single region of the R. norvegicus genome. To utilize fossil calibrations for dating diversification events in Rattus, sequences from four nuclear loci ( Bzrp, BRCA1, IRBP, and RAG1 ) were obtained from GenBank for one deomyine rodent, one gerbilline rodent, and 48 species of murine rodents representing most major divisions ( Musser and Carleton 2005 ) and tribes ( LeCompte et al. 2008 ). All new DNA sequences generated for this study were deposited in GenBank (accession numbers HQ334261–HQ334891; Supplementary Data ). Sequences were aligned using CodonCode v2.0.6 (CodonCode Corporation) and manually inspected in MacClade v.4.08 (Maddison D.R. and Maddison W.P. 2003). Alignments for the Rattus data set were unambiguous for all nuclear loci examined. Alignment of the mitochondrial control region was also unambiguous except at two indel regions (bases 198–202 and 214–242) that were excluded from all analyses. Alignment of the four locus Rattus+ Murinae data was obtained by manual alignment of the Rattus sequences in MacClade to a previously published alignment of Murinae ( Rowe et al. 2008 ). Coding regions were determined in MacClade by alignment of mRNA sequences from the R. norvegicus genome. Heterozygous positions (defined here as two clear, nearly equally sized peaks at the same position) were observed at less than 0.05% of sites in any individual. Haplotypes were resolved statistically using PHASE 2.1.1. To reduce computational time we included a single randomly selected haplotype from each individual in all subsequent analyses. The complete alignments used in this study are available on the TreeBase Web site (Study Accession URL: http://purl.org/phylo/treebase/phylows/study/ TB2:S10869).

Phylogenetic Analyses

Phylogenetic analyses were conducted on the 10 locus Rattus data set using Bayesian methods as implemented in MrBayes ver. 3.1.2 ( Huelsenbeck et al. 2001 , Ronquist2003). Bayesian analyses of the Rattus data set were conducted on the full-concatenated data, the nuclear concatenated data, and the mitochondrial data alone. We used the full-concatenated data to evaluate eight partitioning strategies ( Table 1 and Supplementary Material Online, available from http://www.sysbio.oxfordjournals.org ); 1) single partition, 2) partitioned into exon and noncoding regions, 3) partitioned into nuclear exon, nuclear noncoding and mtDNA, 4) partitioned into three codon positions and noncoding regions, 5) partitioned into nuclear codon positions, nuclear noncoding regions, and mtDNA, 6) partitioned into genes, 7) partitioned into genes, exon, and noncoding regions, and 8) partitioned into genes, codon positions, and noncoding regions. To simplify partitioning models, only intron regions from Fgb, Bzrp, DHFR , and Fn1 were included in the analyses. The short exon regions that were excluded from these loci accounted for a total of nine parsimony-informative characters. The results of these partitioning strategies were compared using log base 10 Bayes factors calculated in Tracer v1.4 ( Table 1 ; Rambaut and Drummond 2007 ). For each partition, we calculated likelihood scores for 24 hierarchical substitution models in PAUP v.b4.10 ( Swofford 2002 ) using the “modelblock” file in MrModeltest ( Nylander 2004 ) and selected model parameters based on Akaike information criteria from MrModeltest. In all MrBayes runs, we unlinked parameters for each partition and allowed branch lengths to vary proportionately across partitions using the ratepr = variable setting. For all partitions in the concatenated data set, we ran two independent sets of four chains for 20 million generations with trees and parameters recorded every 500 generations. To correct for Bayesian inflation of branch-length estimation ( Marshall 2010 ), we used the equation of Brown et al. (2010) to set an appropriate value for our λ prior. To get an approximation for the total tree length, we used the web-based RaxML Black Box software ( Stamatakis 2006 ) to estimate a maximum likelihood (ML) topology and to provide ML bootstrap support values. We estimated branch lengths on the consensus topology using PAUP v.b4.10 ( Swofford 2002 ) and a GTR+I+Γ model with all rates estimated. Based on the average branch lengths from this tree we set our λ prior to 323.9. For the best partition model, we then ran MrBayes for a total of 50 million generations on the full-concatenated data, the nuclear concatenated data, and the mtDNA data. Convergence and stationarity of runs were estimated by examination of likelihood plots, split frequencies and by means of diagnostics from are we there yet? (Wilgenbusch et al. 2004).

TABLE 1.

Comparison of eight partitioning strategies run on the full concatenated data set reported as Log base 10 Bayes factors

	Single partition	Exon/ noncoding	Exon/ noncoding + mtDNA	Coding position	Coding position + mtDNA	Gene	Gene + exon/ noncoding	Gene + coding position
Number of partitions	1	2	3	4	5	10	12	22
ln P(model | data)	− 28393.051	− 28210.199	− 27921.203	− 28187.448	− 28020.142	− 28069.79	− 27830.489	− 28382.12
SE	±0.231	±0.692	±0.185	±0.254	±0.229	±0.229	±0.227	±0.34
No -artition	—	− 79.412	− 204.921	− 89.292	− 161.953	− 161.953	− 244.318	− 4.745
Exon/noncoding		–	− 125.509	− 9.881	− 82.541	− 82.541	− 164.906	74.667
Exon/noncoding + mtDNA			–	115.629	42.969	− 64.531	− 39.396	200.177
Coding -osition				–	− 72.66	− 72.66	− 155.025	84.548
Coding position + mtDNA					–	− 21.562	− 82.365	157.208
Gene						–	− 103.927	135.646
Gene + exon/noncoding								239.573
Gene + coding position							− 239.573	−

Bayesian Estimation of Species Trees

To incorporate effects of coalescent variance among loci, we also conducted phylogenetic analyses using Bayesian Estimation of Species Trees v2.2 (BEST; Edwards et al. 2007 ; Liu and Pearl 2007 ; Edwards 2009 ). Unlike concatenation methods, which do not allow individual genes to have independent topologies, BEST estimates gene trees individually by a modification of the Bayesian methods implemented in MrBayes. After estimation of gene trees, BEST then integrates across the posterior distributions of these gene trees to estimate the most likely species tree given the gene trees. To estimate the species tree, BEST requires the user to assign samples to species. We defined species as currently recognized except for R. fuscipes and R. leucopus , where we additionally specified allopatric subspecies as species. Preliminary analyses suggested that these subspecies were as genetically divergent as other species of Sahulian Rattus and because these subspecies are allopatric we were confident that they did not violate the main assumption of BEST that there is no gene flow between species. Because BEST analyses estimate phylogenies for each gene individually, missing loci can have unpredictable effects on the placement of taxa; thus, we consider only samples for which data were available from all 10 loci. Preliminary runs including this subset of the data set failed to reach stationarity in likelihood scores after several hundred million generations and several hundred hours of processor time. Thus, to reduce computational time and to achieve stationarity, we included a single haplotype for each locus from one sample of each species and one sample of each subspecies of R. fuscipes and R. leucopus for which there was complete data ( Supplementary Data ) for a total of 14 samples and subspecies in the analysis. The three species/subspecies that were not included in the BEST analysis were R. sordidus , R. fuscipes fuscipes , and R. novaeguineae . Data were partitioned by gene, and initial model parameters were as for the Bayesian analyses. Priors were set to defaults accept the theta prior was set to invgamma (3, 0.03). Two independent sets of two chains were run for 250 million generations with trees and parameters recorded every 10,000 generations. Stationarity was estimated by examination of likelihood plots and convergence was determined by examination of split frequencies. The accuracy of branch-length estimates for each resultant gene tree from BEST were evaluated based on comparison with branch lengths estimated on the same topology using the appropriate likelihood model in PAUP v.b4.10 ( Swofford 2002 ).

Molecular Dating and Net Diversification Rate

To provide fossil calibration points, we combined an alignment of Murinae containing the 4 loci Bzrp, BRCA, IRBP, and RAG1 with our complete 10 locus Rattus data set ( Supplementary Data ). We also included IRBP data for three additional species of Rattus ( R. exulans, R. everretti, and R. tanezumi ) We used five murine fossil calibrations to estimate the time to most recent common ancestor ( tmrca ) of Murinae, of Rattus and for key divergence events within Sahulian Rattus . Using only the four loci shared between the two data sets, we recovered similar estimates for the tmrca's for Murinae, Rattus , and the Sahulian Rattus (data not shown). We applied both normal distributions and lognormal distributions around the means of our fossil calibrations. For the normal distributions, we assigned broad standard deviations (SD) around the means to allow for uncertainty in the age and placement of fossil calibrations. For the lognormal distributions, we set the zero offset such that the median value was equal to the mean of fossil calibration. As there was no appreciable difference between the two analyses, we report only the results of the lognormal estimates. Fossil calibrations were assigned as follows. To allow for uncertainty in our fossil dates and placements, we assumed broad SD around the mean ages of each fossil.

• 1) We assigned the age of transition from Antemus to Progonomys at 12.1 Ma (±2.0 SD; Jacobs and Downs 1994 ) to the node splitting Phloeomys/Batomys from the remaining Murinae as has been argued previously ( Steppan et al. 2004 ).

• 2) We assigned the oldest fossils of the lineage leading to the genus Apodemus at 11 Ma (±2.0 SD) in the Early Vallesian to the split between Apodemus / Tokudaia clade and the Mus / Praomys / Mastomys clade ( Martin Suarez and Mein 1998 , Vangegeim et al. 2006).

• 3) We assigned the earliest record of the African Arvicanthis lineage at 6 Ma (±1.0 SD) to the divergence between Otomys / Paratomys and the A rvicanthis / Rhabdomys / Grammomys / Oenomys clade ( Winkler 2002 ).

• 4) We assigned the earliest record of the modern genus Mus ( Mus auctor ) at 5.7 Ma (±1.0 SD) to the split between Mus musculus and Mus pahari ( Jacobs and Downs 1994 ) that reflects the earliest divergence among subgenera of Mus ( Lundrigan et al. 2002 ).

• 5) We assigned the oldest confirmed record of the “old endemic” rodents of Australia at 3.4 Ma (±1.0 SD; Tedford et al. 1992, Aplin 2006 ) to the node uniting Uromys , Melomys , Paramelomys , Conilurus , Mesembriomys , Leporillus , Pseudomys , Mastacomys , Zyzomys , and Leggadina . Notably, this date (3.4 Ma) was also recovered by a recent molecular divergence estimate using calibrations 1 and 3 above (clade V in Rowe et al. 2008 ).

Divergence dates were estimated using a relaxed molecular clock approach in BEAST v1.5.2 ( Drummond et al. 2006 , Drummond and Rambaut 2007 ). We implemented an uncorrected lognormal model of rate variation, a Yule speciation process for branching rates, and the GTR + I + Γ model of nucleotide substitution with four rate categories. Data were partitioned using the strategy selected by Bayes factor analysis. Substitution models and clock models were unlinked among partitions. All analyses were run for 80 million generations and logged every 1000 steps. We used Tracer v 1.2 ( Rambaut and Drummond 2007 ) to determine convergence and to calculate the mean and 95% credible intervals for the tmrca estimates.

We used the tmrca estimates from BEAST to calculate the net diversification rate (speciation–extinction) in Murinae, Rattus, and Sahulian Rattus using the method described in Magallon and Sanderson (2001) and as implemented in the R package LASER v.2.3. ( Rabosky 2006 ). For the number of extant Rattus species, we used 66 species as recognized by Musser and Carleton (2005) . For the Sahulian Rattus , we used both the 16 species recognized by Taylor et al. (1982) and the 24 species (including Moluccan taxa) recognized as part of the Sahulian radiation by Musser and Carleton (2005) . For the number of extant murine species, we used 561 species as reported by Musser and Carleton (2005) . We estimated a range of rates by considering the 95% credible intervals from the BEAST analyses. To evaluate the accumulation of lineages through time (LTT), we estimated a LTT plot in LASER v.2.3 using an ultrametric tree of Sahulian Rattus subspecies by pruning the majority-rule consensus phylogeny recovered by MrBayes analysis of the full-concatenated data set to one specimen per subspecies of Sahul Rattus. To obtain an ultrametric tree, we transformed branch lengths using nonparametric rate smoothing weighted across the root. For the LTT, we used a log transformation of lineage number. To test whether the rate of accumulation of lineages has slowed to the present, we estimated the γ statistic and corrected for incomplete sampling using LASER v.2.3 and the methods proposed by Pybus and Harvey (2000) .

Ancestral State Reconstruction

To examine the role of ecological and biogeographic transitions in the diversification of Sahulian Rattus , we reconstructed ancestral states for two ecological characters, habitat and elevation, and one biogeographic character, distribution in Australia and New Guinea. In general, Sahulian Rattus are readily divided into two habitat categories—closed (rainforest, eucalyptus forest, coastal scrub, subalpine woodland) and open (grassland, desert, alluvial floodplains, heathland)—and into two elevational categories—lowland (not observed above 1500 m) and highland (observed above 1500 m). Geographic assignments were determined based on published accounts (Taylor and Horner 1973; Taylor et al. 1982; Flannery 1995a ; Van Dyck and Strahan 2008 ). Ancestral states were reconstructed in Mesquite v. 1.11 (Maddison W.P. and Maddison D.R. 2006) using a likelihood reconstruction method and the Mk1 rate model ( Lewis 2001 ). Rattus novaeguineae has been collected in both the forest and grassland. Unfortunately, characters cannot be treated as polymorphic in Mesquite, so we assigned R. novaeguineae an uncertain habitat state that treats the characters as equally likely to be either state. To account for topological uncertainty, we reconstructed ancestral states on both the majority-rule consensus phylogeny recovered by MrBayes analysis of the full-concatenated data set and on the phylogeny recovered by BEST analysis of the full data.

Morphological Disparity through Time

To evaluate the rate of morphological evolution in relation to speciation in Sahulian Rattus , we calculated morphological disparity through time (DTT) following the procedure described in Harmon et al. (2003) . DTT analyses compare phenotypic disparity simulated under a model of brownian motion with observed phenotypic disparity among and within subclades relative to total disparity at all time steps in a phylogeny. Values near zero indicate that disparity is partitioned among subclades with limited disparity within clades, whereas values near one indicate that disparity is high within subclades relative to total disparity across the entire phylogeny. We obtained from the literature the mean values for 20 morphological characters for each subspecies in our phylogeny ( Taylor and Horner 1977 , Taylor et al. 1982). The 20 morphological measurements included head and body length, tail length, hind foot length, 15 cranial measurements, and two molar measurements ( Supplementary Data ; ranges summarized in Table 2 ). All values were log transformed before analyses. A principal component analysis, as implemented with the prcomp command in R v. 2.8.1, was used to reduce dimensionality of the data set and to account for correlations among characters due to overall body size. The first four principal components accounted for 95% of all variation and were retained. We calculated the DTT plot using average squared Euclidean distances as implemented in the R-package GEIGER v.1.3.1 ( Harmon et al. 2008 ). We calculated the morphological disparity index (MDI) to assess whether disparity within lineages is less than or greater than the median expectations of the null model.

TABLE 2.

Morphological characters and ranges of measurements used in DTT analyses

	Minimum value (mm)	Species with minimum value	maximum value (mm)	Species with maximum value
Head and body length	120.1	R. niobe	195.9	R. praetor
Tail length	98.9	R. giluwe	175.2	R. praetor
Hind foot length	28.1	R. niobe	39.4	R. leucopus dobodurae
Occipitonasal length of skull	32.6	R. niobe	44.6	R. praetor
Condylobasal length	30	R. niobe	42.3	R. praetor
Basal length	27.3	R. niobe	39.2	R. praetor
Zygomatic arch	15.2	R. niobe	21.3	R. praetor
Interorbital width	4.06	R. fuscipes greyi	6.5	R. leucopus dobodurae
Interparietal length	3.81	R. lutreolus velutinus	5.8	R. praetor
Interparietal width	9.4	R. lutreolus velutinus	10.9	R. praetor
Braincase width	14.2	R. niobe	17.3	R. praetor
Mastoid width	12.2	R. niobe	14.8	R. praetor
Nasal length	12.2	R. niobe	16.5	R. praetor
Nasal width	3.4	R. niobe	5.1	R. novaeguineae
Palatal length	16.7	R. niobe	24.2	R. praetor
Incisive foramina length	4.4	R. niobe	8.31	R. villosissimus
Incisive foramina width	1.68	R. lutreolus velutinus	2.9	R. praetor
Inside m width	2.62	R. tunneyi culmorum	4.3	R. praetor
Outside m width	6.7	R. niobe	9.3	R. praetor
Bulla length	4.7	R. niobe	8.8	R. tunneyi tunneyi

RESULTS

Concatenated Data

Concatenation of the nine nuclear loci and one mitochondrial locus from the Rattus samples resulted in a data set consisting of 10,457 aligned base pairs, 1128 variable sites, and 1023 parsimony-informative sites. Excluding the outgroups, R. norvegicus and R. rattus , resulted in 892 variables sites and 780 parsimony-informative sites (434 from the nine nuclear loci and 346 from the one mitochondrial locus; Supplementary Data ). Bayes factors analysis of eight partitioning strategies overwhelmingly selected the gene-exon noncoding regions as the best strategy by a minimum of 39.4 log base 10 Bayes factors ( Table 1 ).

Bayesian phylogenetic analyses of three data sets using the Bayes factors selected partitioning strategy produced highly resolved phylogenies (nine nuclear loci + mtDNA concatenation, Fig. 2 ; nine nuclear loci concatenation, Figure S2; and mtDNA alone, Figure S1). All data sets supported a monophyletic Sahulian Rattus (Clade A) and the following six main lineages within the Sahulian Rattus : Clade E: R. niobe ; Clade F (recent New Guinean Rattus ): R. giluwensis , R. verecundus , R. praetor , R. steini , R. novaeguineae ; Clade G: R. leucopus subspecies; Clade H: R. fuscipes subspecies; and Clade I (recent Australian Rattus ): R. lutreolus , R. tunneyi , R. sordidus , R. colletti , and R. villosissimus . All analyses supported a sister relationship between Clade H and Clade I (Clade D; Australian Rattus ). The full data and mtDNA alone both recovered weak support for a sister relationship (Clade C; Fig. 2 ; Figure S1) between R. leucopus (Clade G) and the Australian Rattus (Clade D; Bayesian posterior probability = 0.75 and 0.63 respectively). The nuclear loci alone recovered weak support for a sister relationship (Clade M, Figure S2, Bayesian posterior probability (BPP) = 0.85) between R. leucopus (Clade G) and the recent New Guinean Rattus and was equivocal with regard to the relationships among this group (Clade M), R. niobe (Clade E), and the Australian Rattus (clade D). RaxML analyses of the full data set resolved an identical topology to the Bayesian phylogeny (bootstrap values in Fig. 2 ).

FIGURE 2.

Bayesian phylogeny of Sahulian Rattus estimated by MrBayes from the partitioned concatenation of nine nuclear loci and one mitochondrial locus. Nodes discussed in the text are labeled A–I. Nodal support is indicated by Bayesian posterior probabilities above branches and maximum likelihood bootstrap values below branches.

Bayesian Estimation of Species Trees

Both the full data and nuclear loci only BEST analyses recovered the same topology except that the nuclear only data (not shown) recovered a polytomy at the base of the Sahulian Rattus , whereas the full data recovered a split between R. niobe and Clade L discussed below. Both BEST analyses largely supported the phylogenetic relationships recovered in the concatenated data sets, and resolved the same Clades A, D, F, G, H, I, and K ( Fig. 3 ). Clade E ( R. niobe ) in the concatenated analyses was represented by a single sample in the BEST analyses and therefore could not be recovered as a clade. Both BEST analyses recovered the same Clade M as the nuclear concatenated data set and in contrast to the full-concatenated and mtDNA data sets. Although the nuclear concatenated data and the nuclear only BEST analyses were both equivocal with respect to Clade B recovered in the full-concatenated analyses, the full data BEST analyses recovered a Clade L that split R. niobe from all other Sahulian Rattus . Although the BEST analyses resolved a topology largely consistent with the concatenated data sets, posterior probabilities were lower at all nodes.

FIGURE 3.

Sahulian Rattus species tree estimated using BEST. Nodes are labeled as in Figure 2 with posterior probability values above branches. Nodal support is indicated by posterior probabilities above branches. The likelihood of ancestral states for three binary characters, elevation, habitat, and geography are represented by pie charts at each node. Character states for terminal taxa are indicated at the tips of the tree.

Molecular Dating and Net Diversification Rate

We estimated the time to most recent common ancestor ( tmrca) of Murinae (including Phloeomys and Batomys ) at 13.34 Ma (confidence interval ([CI] 11.84–15.10 Ma) in the mid-to-late Miocene, and the tmrca of all Rattus at 2.44 Ma (CI 1.84–3.17 Ma) in the late Pliocene ( Table 3 ; Fig. 4 ; Fig. S3). As found in previous analyses ( Verneau et al. 1998 , Robins et al. 2007 ), the common ancestor of Rattus marked the divergence between Sahulian Rattus and all other Rattus examined. However, the origin of extant Sahulian Rattus lineages is much younger than this split and we estimated a tmrca for Sahulian Rattus in the mid-Pleistocene 1.05 Ma (CI 0.85–1.28 Ma). We also estimated the tmrca for subspecies of two widespread species, R. leucopus (Clade G) and R. fuscipes (Clade H), and compared these with estimates for other major lineages within New Guinea and Australia. For R. leucopus subspecies, we estimated a tmrca of 0.55 Ma (CI 0.41–0.70 Ma), nearly identical to our tmrca estimate of 0.54 Ma (CI 0.32–0.77 Ma) for the recent New Guinean Rattus (Clade F) that comprise five ecologically disparate and morphologically identifiable species. A similar pattern was observed in the Australian Rattus (clade D). We estimated a tmrca for R. fuscipes subspecies of 0.73 Ma (CI 0.57–0.90 Ma) comparable with our estimate of 0.78 Ma (CI 0.63–0.94 Ma) for the five other Australian Rattus that form an ecologically disparate clade (Clade I).

TABLE 3.

Estimated time to most recent common ancestor in millions of years before present for Rattus and key clades of the Sahulian Rattus based on five fossil calibrations from the rodent subfamily Murinae

	Calibration 1: Murinae	Calibration 2: Mus-Apodemus	Calibration 3: Arvicanthis	Calibration 4: Mus	Calibration 5: Australian old endemic murines	Rattus
Mean	13.37	10.40	7.46	5.39	3.58	2.44
Median	13.34	10.32	7.34	5.30	3.58	2.39
95% CI: lower	11.84	10.03	5.42	4.75	3.01	1.84
95% CI: upper	15.10	10.95	9.50	6.22	4.16	3.17
Effective sample size	975.34	10500.00	196.71	3064.42	273.94	204.89

	Clade A: Sahulian Rattus	Clade D: Australian Rattus excl. R. leucopus	Clade F: New Guinean Rattus excl. R. niobe and R. leucopus	Clade G: R. leucopus	Clade H: R. fuscipes	Clade I: Australian Rattus excl. R. fuscipes and R. leucopus
Mean	1.06	0.92	0.54	0.55	0.74	0.78
Median	1.05	0.91	0.54	0.55	0.73	0.78
95% CI: lower	0.85	0.76	0.32	0.41	0.57	0.63
95% CI: upper	1.28	1.09	0.77	0.70	0.90	0.94
Effective sample size	112.22	112.22	152.73	197.40	86.83	87.64

FIGURE 4.

Chronogram inferred using BEAST analyses trimmed to show only Rattus species in the analysis (full murine phylogeny in Figure S3). The topology is the same as the MrBayes phylogeny in Figure 2 . Nodes are labeled as in Figure 2 . Gray bars around nodes reflect 95% credibility intervals around the tmrca estimates from BEAST. Karyotypes are presented beside taxa where known. Karyotype results for “ Rattus ruber” could apply to R. novaeguineae , R. steini , and/or R. praetor . Predicted Robertsonian fusions (RF), metacentric fissions (MF), and pericentric inversions (PI) are indicated along branches. Chromosome data from Baverstock et al. (1983) and Dennis and Menzies (1978) . The likelihood of ancestral states of three binary characters, elevation, habitat and geography are represented by pie charts at each node. Character states for terminal taxa are indicated at the tips of the tree. Ancestral states presented were reconstructed using the phylogeny estimated by MrBayes ( Fig. 2 ).

Rattus as a whole and Sahulian Rattus in particular have diversified at exceptionally high rates ( Table 4 ). With a late Pliocene origin, the vast majority of Rattus speciation occurred in the Pleistocene. For the subfamily Murinae, we estimated a net diversification rate (ND) of 0.42 species/lineage/myr (CI = 0.37–0.48). In contrast, we estimated an ND for Rattus of 1.46 (CI = 1.10–1.90). For Sahulian Rattus, we estimated an ND of 1.98 (CI = 1.61–2.45) assuming 16 extant species as recognized by Taylor et al. (1982) or 2.41 (CI = 1.96–2.97) assuming 24 extant species as recognized by Musser and Carleton (2005) . The (LTT) plot indicated that ND in Sahulian Rattus has not remained constant over time ( Fig. 5 ). The LTT plot exhibited a concave-down shape characteristic of the rapid accumulation of lineages early in the radiation (up to the formation of clade D, E, F, G, H, and I) with slower accumulation of lineages more recently in the phylogeny ( Rüber and Zardoya 2005 ; McPeek 2008 , PP08). We calculated a γ statistic of − 3.10 for the Sahulian Rattus . Given our incomplete sampling of species and assuming the taxonomy of Musser and Carleton (2005) , simulations in LASER estimated that the critical value for the γ statistic is − 2.04. Our estimated γ statistic is therefore, significantly less than zero ( P = 0.001) and supports a slowing of the net diversification rate in Sahulian Rattus .

TABLE 4.

Estimates of net diversification rates using tmcra estimates from BEAST and the method of Magallon and Sanderson (2001)

	Median (sp/line/myr)	95% lower limit	95% upper limit
Net diversification estimates
Murinae (561 extant species)	0.42	0.37	0.48
Rattus (66 extant species)	1.46	1.10	1.90
Sahul Rattus (16 extant species)	1.98	1.61	2.45
Sahul Rattus (24 extant species)	2.41	1.96	2.97

FIGURE 5.

a) Log LTT plot for Sahulian Rattus . Lineages accumulate rapidly early in the radiation and slow to the present. b) Morphological DTT plot for Sahulian Rattus (solid line) against the median result of brownian model simulations. Time is relative to phylogenetic depth from the base of the phylogeny on the left to the terminal tips on the right. c) Ultrametric phylogeny used for LTT and DTT analyses. The first four splits in the phylogeny are labeled on the LTT and DTT plots and correspond to the colonization of the primary ecological and geographic states modeled.

Ancestral State Reconstruction

To examine how ecological transitions relate to diversification dynamics, we reconstructed ancestral states for elevation, habitat, and geography. To account for topological uncertainty, we reconstructed states on the full-concatenated MrBayes topology (MB; Fig. 4 ) and the full data BEST topology ( Fig. 3 ). On the MB topology, the ancestral elevation state at node A was equivocal (62% low), whereas the BEST topology suggests a high likelihood of a high elevation ancestral state (94%). Both topologies, however, recovered a split between high- and low-elevation lineages early in the phylogeny; between Clades B and C on the MB topology and both the R. niobe /Clade L split and the Clade F/Clade G split on the BEST topology. Both the MB and BEST topologies recovered a high likelihood of a closed habitat ancestor at Node A (94% and 99% respectively), and a clear transition to open habitat in the common ancestor of Clade I (recent Australian Rattus ). The MB topology is equivocal with regard to the geographic distribution of the ancestor at Node A (54 % New Guinea), whereas the BEST topology produced a high likelihood of a New Guinean ancestor (96%) at Node A. On both topologies, Clade D is clearly supported as an early Australian lineage separate from other early New Guinean lineages (MB—Clade B; BEST— R. niobe, Clade M and Clade F). Both topologies recovered the same ancestral states at Nodes F, H, and I. These three lineages reflect (F) a high elevation, closed habitat, and New Guinean lineage; (H) a low elevation, closed habitat, and Australian lineage; and (I) a low elevation, open habitat, and Australian lineage. Node E represents R. niobe , a high elevation, closed habitat, and New Guinean species. Node G represents R. leucopus, a low elevation, and closed habitat species of New Guinea and Australia. High-elevation open habitat lineages appear at the terminal branches and only in New Guinea (similar habitats are not present in Australia), with the earliest origin along the branch leading to R. giluwensis. In short, our ancestral state reconstructions show that early in the phylogeny Rattus had diverged into Australian and New Guinean lineages, low- and high-elevation lineages and closed and open habitat lineages. In most lineages, with the exception of the recent New Guinean Rattus (Clade F), these ancestral environmental attributes are retained to the present.

Morphological DTT

To relate speciation dynamics to phenotypic evolution, we examined morphological DTT ( Fig. 5 ). At most points in the phylogeny, the proportion of total morphological disparity within clades was less than expected by a brownian motion model. This is reflected in the negative MDI statistic for Sahulian Rattus ( − 0.07). These results suggest that early in the diversification of Sahulian Rattus , lineages occupied a limited proportion of total morphological space, with most variation partitioned among not-within lineages.

DISCUSSION

The diversification rates for Rattus and Sahulian Rattus are among the highest reported for vertebrates ( Supplementary Data ) and are comparable with or exceed “explosive” rates reported for other prominent rapid radiations (e.g., Darwin's finches, Grant 1986 ; Hawaiian silverswords, Baldwin and Sanderson 1998 ; Dendroica warblers, Lovette and Bermingham 1999 ; Plethodon salamanders, Kozak et al. 2005 ; zosteropid birds, Moyle et al. 2009 ). Our median estimate for the net diversification rate among Sahulian Rattus is 4.7 to 11 times higher than for rodents of the family Muridae (0.36; Stanley 1998 ), for Mammalia (0.22; Stanley 1998 ) and for Murinae (0.42). Although our incomplete and nonrandom taxon sampling precludes an explicit test of whether Sahulian Rattus diversified more rapidly than the background Rattus rate ( Pybus and Harvey 2000 , BN01), our ND estimates are sufficient to suggest a remarkably rapid diversification.

The radiation of Sahulian Rattus shares certain characteristics with other adaptive radiations (e.g., early burst of diversification and environmental partitioning of lineages) but is notable for characteristics that are more consistent with nonadaptive radiations (i.e., conserved morphology). As is often observed in rapid radiations, the LTT plots and γ statistic for the Sahulian Rattus recovered a burst of speciation early in the radiation following colonization ( Lovette and Bermingham 1999 ; Harmon et al. 2003 ; Kozak et al. 2006 , McPeek 2008 , Phillimore and Price 2008 ). Ecological speciation theory predicts that radiations driven by ecological opportunity should exhibit an early burst of speciation with declining speciation rates, once novel environments are occupied ( Simpson 1953 ; Gavrilets and Losos 2009 ). However, there is considerable debate about whether or not ecological opportunity is the only mechanism that can produce this early burst pattern ( McPeek 2008 , R09a). Some authors have argued that niche conservatism and ensuing allopatric divergence could lead to a similar pattern of early diversification ( Wiens 2004 ; Rundell and Price 2009 ). The radiation of North American Plethodon salamanders ( Kozak et al. 2006 ) and Dendroica warblers ( Lovette and Bermingham 1999 ) provide two prominent empirical examples of early burst diversification attributed to filling of ecologically similar but patchily distributed habitats. In short, ecological models predict that early diversification results from species adapting to and filling divergent niches, whereas nonecological or niche conservatism models predict that early diversification results from filling of spatially isolated patches of similar habitat. Support for either of these models often rests on whether diversification of species is most coincident with ecological or spatial discontinuities ( Graham et al. 2004 , Kozak and Wiens 2006 )

In this paper, we reconstructed ancestral states of two very simple ecological characters: elevation (low and high) and habitat (closed and open). The expansion of Rattus between these simple environmental states coincides well with their early burst of speciation. Despite some inconsistency among phylogenetic reconstruction at deeper nodes, both concatenation ( Fig. 4 ) and species tree methods ( Fig. 3 ) recovered the same ancestral ecological states of lineages F, G, H and I. These nodes reflect the colonization of both Australia and New Guinea and the acquisition of the primary ecological states in our phylogeny: 1) high-elevation closed habitat (F), 2) low-elevation closed habitat (G & H), and 3) low-elevation open habitat (I). Up to colonization of these environments and of Australia and New Guinea, lineage accumulation remained rapid and fairly constant, reflecting the early burst of speciation ( Fig. 5 ). In addition, there are almost no reversals in these ecological states suggesting that they were driven by unique adaptations retained in descendant lineages. For instance, the recent Australian Rattus represent a single clear transition to low-elevation open habitats (Clade I) from low-elevation closed habitats (Clade D). All descendant species, except one subspecies of R. lutreolus , are constrained to these states, albeit in otherwise ecologically different habitats (e.g., wetlands vs. tropical grasslands). Within the recent New Guinean Rattus , similar transitions to high-elevation open habitats occurred twice with descendant lineages retaining these habitat associations. This pattern of early transition and retention of habitat associations supports an ecological diversification of Rattus species.

Studies of parapatric species interactions also support the hypothesis that Sahulian Rattus are partitioned by their performance in different environments. For instance, the lineages leading to R. fuscipes (Clade H), a closed habitat species, and R. lutreolus (Clade I), an open habitat species, diverged during the early burst of diversification. These two species are broadly sympatric along the eastern margin of Australia but locally are competitively partitioned into forest and wetland habitats, respectively ( Maitz and Dickman 2001 ). Similarly, R. fuscipes also diverged from R. leucopus during the early burst of diversification and the two closed forest species partition space elevationally ( Winter 1997 ) with syntopy at middle elevations. A similar pattern of elevational partitioning is observed between R. leucopus , R. niobe and species of the recent New Guinean Rattus clade (Taylor et al. 1982, 1985). These examples suggest that inherent ecological differences between Rattus species have promoted their divergence, despite overall morphological conservation.

Morphological differences among species have played a prominent role in discussions of adaptive and nonadaptive diversification. In most adaptive radiations studied, the diversification of species into ecologically disparate habitats is manifest as morphological divergence among species. Conversely, the lack of clear morphological divergence has been taken to suggest that species diverged nonadaptively ( Lovette and Bermingham 1999 , Kozak et al. 2006 ). Like Plethodon and Dendroica , Rattus, in general, retain a very conserved body plan with little overt ecomorphological specialization among species. DTT plots showed that early in diversification of Sahulian Rattus morphological disparity within lineages is less than expected under a brownian model ( Fig. 5 ). This pattern is often interpreted to reflect divergence in ecomorphology ( Harmon et al. 2003 ; Kozak et al. 2005 ; Burbrink and Pyron 2010 ). Thus, despite their fairly conserved morphology, Rattus are in fact, diverging morphologically early in their radiation. However, even this belies the whole truth because rapid lineage accumulation continues even after morphological disparity within lineages is low, and before lineages have diversified into the primary ecological zones. In essence, partitioning of morphological variation occurred early in the diversification of Sahulian Rattus consistent with ecomorphological divergence of lineages, but continued rapid ecological diversification did not depend on morphological disparity alone.

Nonecological models of rapid radiations predict that limited dispersal and the isolation of species in ecologically similar environments can also drive rapid diversification ( Wiens 2004 , K06; Rundell and Price 2009 ). The strongest cases for significant genetic divergence in ecologically similar allopatric habitats occur among allopatric subspecies of R. leucopus and R. fuscipes. Divergence levels and tmrca estimates among allopatric subspecies of R. leucopus and R. fuscipes are of the same scale as those among ecologically and morphologically definable species in Clades F (recent New Guinean Rattus ) and I (recent Australian Rattus ). The recent Australian Rattus species (Clade I) are also karyotypically diverse, whereas subspecies of R. fuscipes are not. Rattus fuscipes and R. leucopus are both widespread mesic forest species, whereas the species in clades F and I have overlapping geographic ranges but occupy a variety of habitats. Subspecies of R. lutreolus and R. tunneyi are also distributed across xeric biogeographic gaps in their preferred habitat without apparent speciation (Fig. 1; R. lutreolus across the Burdekin gap in central Queensland and R. tunneyi across the Carpenteria gap in western Queensland). Thus, isolated populations of Rattus in similar environments have not diverged into morphologically and chromosomally recognized species as have populations under divergent ecological conditions.

In our phylogeny, physiographic barriers are not good predictors of phylogenetic splits. The central highlands of New Guinea are the highest island mountains on Earth; however, there is no phylogenetic split between southern and northern lowland Rattus species. In Australia, there are few overt physiographic barriers like the high mountain ranges of New Guinea; however, the aridification of Australia over the last 20 myr has resulted in the dramatic retraction of mesic environments ( Byrne et al. 2008 ). Despite their ecological differences, all Sahulian Rattus require mesic environments for their reproduction. Even R. villosissimus is associated with mesic springs of the Great Artesian Basin in the arid interior and is only abundant and widespread during infrequent wet periods. The environment that Rattus encountered when it first colonized Australia was the culmination of nearly 20 myr of aridification with the first stony deserts emerging 3–2 Ma prior to the colonization of Rattus ( Byrne et al. 2008 ). Over the Pleistocene, glacial–interglacial cycles produced dramatic climatic oscillations across Australia with extreme changes in precipitation leading to large variability in the spatial and temporal distribution of mesic habitats with a peak in amplitude between 0.4–0.2 Ma ( Byrne et al. 2008 ; Hocknull 2005 ). This pattern of expanding and contracting habitats is similar to those observed in North America that may explain the rapid diversification of Plethodon and Dendroica ( Lovette and Bermingham 1999 , K06). How Pleistocene expansion and contraction of mesic habitats has affected Rattus diversification warrants further analysis.

The radiation of Sahulian Rattus is all the more remarkable because it has occurred on a continental scale across habitats already occupied by a large diversity of phylogenetically related and ecologically similar species. Most rapid radiations, in contrast, have involved species that have diversified in habitats unoccupied by ecologically similar species ( Grant 1986 , FCF87, M93, BS98, LBR01). When Rattus first colonized Sahul, they encountered a diverse group of “old endemic” murine rodents that had colonized the continent previously in the early Pliocene ( Rowe et al. 2008 ). Today, every Sahulian Rattus species shares habitat with multiple species of old endemic rodents. Most notably, all Rattus, except R. villosissimus , are sympatric with species of the genus Melomys that are similar to Rattus in diets, morphological attributes, and space use patterns ( Flannery 1995a , 1995b ; Van Dyck and Strahan 2008 ). Whereas net diversification rates for the old endemics (∼1.0) are slower than Rattus, the old endemics are also considerably more diverse with nearly 160 species. Given that rates within Rattus are already slowing, they may never be as diverse as the old endemics and their rapid diversification may reflect their nonequilibrium state following recent colonization ( Rabosky 2009b ). Thus, the rate of diversification and the diversity of Rattus compared with the old endemics are not particularly noteworthy. However, the ability of Rattus to colonize and diversify at a nonequilibrium rate in a community of rodents that appears to be ecologically saturated emphasizes the remarkable ability of Rattus to capitalize on niche space underutilized by other rodents.

Chromosomal rearrangements present an alternative model for the nonecological diversification of species ( White 1978 , K93, S93). The Australian Rattus species ( Fig. 4 ; Clade D), in particular, are all diagnosable by their karyotypes with diploid numbers varying from 32 in R. sordidus to 50 in R. villosissimus. However, no evidence of similar karyotypic diversity has been found in the remaining Sahulian Rattus. Most of the chromosomal variation among the Australian Rattus is restricted to the three species of the sordidus group that is a deeply nested clade in our phylogeny, and only two species, R. colletti and R. villosissimus , have chromosomal differences that are expected to lead to severe meiotic malsegregation (Baverstock et al. 1983). Although most species of Australian Rattus maintain fairly stable population sizes, the three species of the sordidus group experience irruptive fluctuations in population size ( McDougall 1944 ; Wood 1971 ; Taylor and Horner 1973; Redhead 1979 ; Madsen et al. 2006). To capitalize on the erratic and infrequent periods of favorable conditions in their respective habitats, these species have evolved extremely high reproductive potentials, with nipple numbers and litter sizes, twice those of other Australian Rattus ( Breed and Ford 2007 ). Large fluctuations in population sizes could lead to small effective population sizes which are the most conducive to the establishment of karyotypic mutations via drift (Bush et al. 1977, Baker and Bickham 1986 ). Reproductive adaptations that may have allowed these species to colonize marginal and irregularly favorable habitats may have secondarily also increased the likelihood of karyotypic mutation. Thus, whether chromosome change is a cause or a correlate of rapid diversification in this group remains uncertain.

Most of the rapid radiations reported are explained by overt morphological novelties that allow species to exploit separate niches. Species of Rattus represent a clear example of recent, “explosive” diversification where the mechanisms isolating species are not defined by overt morphological differences. What key innovations have allowed Rattus to diversify in ecosystems that are saturated with ecologically similar and phylogenetically related species? Many factors that are often difficult to measure, such as behavior, physiology, and reproductive biology, are more likely candidates than morphology alone. Given the deep understanding of genome–phenotype connections in the “lab rat” ( R. norvegicus ), the diversification of Rattus has tremendous potential for uncovering nonmorphological adaptation in the context of recent and rapid speciation.

SUPPLEMENTARYMATERIAL

Supplementary material , including data files and/or online-only appendices, can be found at http://www.sysbio.oxfordjournals.org/ .

FUNDING

This work was supported by 0502375 to K.C.R. from the National Science Foundation, International Research Fellowship Program.

The authors thank Stephen Donnellan (South Australia Museum), Terry Bertozzi (South Australia Museum), and Catherine Nock (Centre for Animal Conservation Genetics, Southern Cross University) for tissue loans; Sandy Ingleby (Australian Museum), Steve Van Dyck (Queensland Museum), Heather Janetzki (Queensland Museum), and David Stemmer (South Australia Museum) for access to voucher specimens; and Karen Rowe for assistance in the field and in preparing specimens. This manuscript was greatly improved by critical comments from Elizabeth Jockusch, Jack Sullivan, two anonymous reviewers, Sonal Singhal, Ricardo Pereira, and members of the Moritz lab.

Author notes

Associate Editor: Elizabeth Jockusch