Volume 58 Number 6, pp. 547–661 Systematic Biology Volume 58, Number 6 December 2009 CONTENTS Systematic Biology REGULAR ARTICLES Species Tree Discordance Traces to Phylogeographic Clade Boundaries in North American Fence A JOURNAL OF THE Lizards ( Sceloporus ) Adam D. Leaché ...... 547 Society of Systematic Biologists The Use and Validity of Composite Taxa in Phylogenetic Analysis Véronique Campbell and François-Joseph Lapointe ...... 560 Radiation of Extant Cetaceans Driven by Restructuring of the Oceans Mette E. Steeman, Martin B. Hebsgaard, R. Ewan Fordyce, Simon Y. W. Ho, Daniel L. Rabosky, Rasmus Nielsen, Carsten Rahbek, Henrik Glenner, Martin V. Sørensen, and Eske Willerslev ...... 573 Taxon Selection under Split Diversity Bui Quang Minh, Steffen Klaere, and Arndt von Haeseler ...... 586 Estimating Trait-Dependent Speciation and Extinction Rates from Incompletely Resolved Phylogenies Richard G. FitzJohn, Wayne P. Maddison, and Sarah P. Otto ...... 595 Reticulation, Data Combination, and Inferring Evolutionary History: An Example from Danthonioideae (Poaceae) Michael D. Pirie, Aelys M. Humphreys, Nigel P. Barker, and H. Peter Linder ...... 612

Heritability of Extinction Rates Links Diversifi cation Patterns in Molecular Phylogenies and Fossils SYSTEMATIC BIOLOGY Daniel L. Rabosky ...... 629 Testing Species-Level Diversifi cation Hypotheses in Madagascar: The Case of Microendemic Brookesia Leaf Chameleons Ted M. Townsend, David R. Vieites, Frank Glaw, and Miguel Vences ...... 641

BOOK REVIEWS Stephen Jay Gould: Refl ections on His View of Life David A. Morrison ...... 657 Systematics and Taxonomy of Australian Birds R. Terry Chesser ...... 659 December 2009 December Cover Illustration: Much less familiar than their larger and more colorful cousins, the diminutive leaf chameleons of the genus Brookesia are nonetheless one of the most speciose reptile groups in Madagascar. Although these leaf-litter inhabitants are distributed across most of the island, individual species often have very restricted distributions. In this issue, Townsend and co-authors use distributional data combined with a dated multi-gene phylogeny of this group to test recent hypotheses proposed to explain the generally high level of faunal and fl oral microendemism on the island. Their results fi nd no support for watershed-contraction or riverine-barrier mediated diversifi cation in this group, although a montane-refuge hypothesis is consistent with some fi ndings. However, species-level divergences in this group are much older than the Pliocene–Pleistocene ages previously proposed. Photos by Frank Glaw.

VOLUME 58 ONLINE ISSN 1076-836X DECEMBER 2009 PRINT ISSN 1063-5157 oxford NUMBER 6 Published on behalf of the Society of Systematic Biologists http:// systbiol.org/ Syst. Biol. 58(6):641–656, 2009 c The Author(s) 2009. Published by Oxford University Press, on behalf of the Society of Systematic Biologists. All rights reserved. For Permissions, please email: [email protected] DOI:10.1093/sysbio/syp073 Advance Access publication on November 11, 2009

Testing Species-Level Diversification Hypotheses in Madagascar: The Case of Microendemic Brookesia Leaf Chameleons

1, 2 3 4 TED M.TOWNSEND ∗,DAVID R.VIEITES ,FRANK GLAW , AND MIGUEL VENCES 1Department of Biology, San Diego State University, 5500 Campanile Drive, San Diego, CA 92182-4164, USA; 2 Museo Nacional de Ciencias Naturales, Consejo Superior de Investigaciones Cient´ıficas,C/ Jos´eGutierrez Abascal n◦2, 28006, Madrid, Spain; 3Zoologische Staatssammlung, M¨unchhausenstr. 21, 81247 M¨unchen,Germany; and 4Technical University of Braunschweig, Zoological Institute, Spielmannstr. 8, 38106 Braunschweig, Germany; ∗Correspondence to be sent to: Department of Biology, San Diego State University, 5500 Campanile Drive, San Diego, CA 92182-4164, USA; E-mail: [email protected].

Abstract.—Madagascar’s flora and fauna are remarkable both for their diversity and supraspecific endemism. Moreover, many taxa contain large numbers of species with limited distributions. Several hypotheses have been proposed to explain this high level of microendemism, including 1) riverine barrier, 2) mountain refuge, and 3) watershed contraction hypotheses, the latter 2 of which center on fragmentation due to climatic shifts associated with Pliocene/Pleistocene glaciations. The Malagasy leaf chameleon genus Brookesia is a speciose group with a high proportion of microendemic taxa, thus making it an excellent candidate to test these vicariance scenarios. We used mitochondrial and nuclear sequence data to construct a Brookesia phylogeny, and temporal concordance with Pliocene/Pleistocene speciation scenarios was tested by estimating divergence dates using a relaxed-clock Bayesian method. We strongly reject a role for Pliocene/Pleistocene climatic fluctuations in species-level diversification of Brookesia. We also used simulations to test the spatial predictions of the watershed contraction model in a phylogenetic context, independent of its temporal component, and found no statistical support for this model. The riverine barrier model is likewise a qualitatively poor fit to our data, but some rela- tionships support a more ancient mountain refuge effect. We assessed support for the 3 hypotheses in a nonphylogenetic context by examining altitude and species richness and found a significant positive correlation between these variables. This is consistent with a mountain refuge effect but does not support the watershed contraction or riverine barrier models. Finally, we find repeated higher level east-west divergence patterns 1) between the 2 sister clades comprising the Brookesia minima group and 2) within the clade of larger leaf chameleons, which shows a basal divergence between western and eastern/northern sister clades. Our results highlight the central role of phylogeny in any meaningful tests of species-level diversification theories. [Biogeography; Chamaeleonidae; phylogeny; Pleistocene glaciation; relaxed clock; speciation; Squamata.]

Once part of the southern supercontinent of Gond- Nussbaum et al. 1998; Vences and Glaw 2002, 2003; wana, Madagascar, is currently situated some 400 km Wood et al. 2007). In other groups, the major evolu- off the southeastern shore of Africa and is famous tionary splits consistently separate a clade occurring for a remarkable flora and fauna that are increasingly roughly in the northern fifth of Madagascar from a more threatened by loss of habitat and other human-induced southern clade, with examples found in chameleons, changes. Madagascar’s last subaerial connections to geckos, skinks, and mouse lemurs (Yoder and Heckman Africa and India date to approximately 160 and 90 mil- 2006; Boumans et al. 2007). As a general pattern, the lion years ago, respectively (Storey et al. 1995), although spatial distribution of species richness in some higher limited dispersal to and from other continents may have taxonomic groups may have been shaped by a latitudi- been possible over land bridges or small sea barriers in nal and altitudinal middomain effect (Lees et al. 1999). the Late Cretaceous or Paleocene (Briggs 2003; Noonan Nonetheless, even within the major bioclimatic regions, and Chippindale 2006; Van Bocxlaer et al. 2007; van species-level local endemism is high, and a major goal der Meijden et al. 2007). This long history of isolation of researchers has been to uncover the causes of this has contributed greatly to the remarkable degree of pattern (Goodman and Benstead 2003 and references floral and faunal endemism of Madagascar (Goodman therein; Vences et al. 2009). and Benstead 2003), which amounts to 100% in native amphibians and 92% in nonmarine nonavian reptiles (Glaw and Vences 2007). Potential Climatic and Geographic Barriers In terrestrial vertebrates, the majority of species diver- In their study of biogeographic patterns in the leaf sity corresponds to a limited number of endemic clades chameleons (genus Brookesia) of northern Madagascar, that colonized most of the available bioclimatic regions, Raxworthy and Nussbaum (1995) recognized 5 rainfor- including the eastern rainforests, central highlands, est regions delimited largely by altitudinal differences western dry deciduous forests, and southern arid spiny and intervening dry forests and characterized by a high forests. There are several sister-taxon pairs that follow degree of endemism in these lizards. Raxworthy and an east-west pattern and possibly arose by specializa- Nussbaum (1995) hypothesized that Pleistocene climatic tion to these different climatic conditions (the “eco- fluctuations had caused fragmentation of the rainforest, graphic constraint hypothesis”; see Yoder and Heckman resulting in multiple allopatrically distributed sister- 2006), from groups such as skinks, snakes, boophine taxon pairs (hereafter referred to as the mountain refuge treefrogs, and spiders (Nussbaum and Raxworthy 1998; [MR] hypothesis). These conclusions were not based

641 642 SYSTEMATIC BIOLOGY VOL. 58 on explicit phylogenetic hypotheses; sister-taxon pairs Brookesia species have very small ranges (Raxworthy were assumed based on general morphological and Nussbaum 1995; Raselimanana and Rakotoma- similarities. malala 2003), with almost half known essentially from Wilme´ et al. (2006) analyzed Madagascar’s striking single localities (Carpenter and Robson 2005). Brookesia microendemicity in the context of watersheds. Using can be divided into 3 main groups based on morphol- museum data for more than 35,000 georeferenced land ogy. One is represented by just 2 species (B. nasus and vertebrate specimens, they found that watersheds with B. lolontany) that are highly divergent from other Brooke- low-elevation headwaters tended to define centers of sia by both molecular (Raxworthy et al. 2002; Townsend endemism (COEs), whereas those with connections to and Larson 2002) and morphological measures (long the 3 highest summits in Madagascar (each at >2000 m) snouts, laterally compressed bodies; Raxworthy and tended to contain more widespread species. Wilme´ et al. Nussbaum 1995). A second larger group is composed of (2006) reasoned that during periods of Late Tertiary approximately 18 species with more robust bodies and and Pleistocene glacial maxima, aridification caused blunted snouts. The remaining 6 described species have contraction of previously widespread mesic environ- taken the already diminutive body form of Brookesia ments. Lower elevations were more severely affected to an extreme, with total lengths of about 28–45 mm, than higher elevations, leading to fragmentation and making them some of the world’s smallest vertebrate isolation of low-elevation watersheds. In contrast, areas species (Glaw and Vences 2007). For ease of reference, with riverine connections to high-elevation source wa- these 3 groups will hereafter be referred to as the B. nasus ters were buffered from this effect and thus served as group, the “typical” Brookesia, and the B. minima group, “retreat–dispersion watersheds” (RDWs), providing a respectively. means of refuge and recolonization during dry and wet cycles, respectively. The authors presented this model of Pliocene–Pleistocene fragmentation and refugia (here- after referred to as the watershed contraction [WC] Hypothesis Testing hypothesis) as the main generator of Madagascar’s In this study, we use Brookesia to test the temporal famously microendemic biota. The Wilme´ et al. (2006) and spatial predictions of 3 species-level diversification study was based entirely on extant species distributions hypotheses for Madagascar (MR, WC, and RB; Vences coupled with climatological, hydrological, and topo- et al. 2009). The first step in this process is to infer a graphical variables and incorporated no phylogenetic phylogeny of Brookesia to identify statistically supported data. Pearson and Raxworthy (2009) found some sup- sister-species pairs using DNA sequence data from mul- port for the WC hypothesis in lemurs and day geckos tiple mitochondrial and nuclear protein-coding genes. of the genus Phelsuma (a climatological gradient effect Next, we use divergence dating to statistically test the was also cited for these lizards), but once again relied temporal prediction of both the MR and the WC hy- on distribution patterns alone. potheses that recent (Pleistocene or possibly Pliocene) Finally, the presence of large rivers flowing from the climatic cycles are a major force promoting Brookesia highlands either toward the west or the east has also species diversification. been discussed as a factor influencing species formation Major climatic cycles have of course occurred repeat- and microendemism (hereafter referred to as the river- edly throughout the earth’s history, and each of these ine barrier [RB] hypothesis). In western Madagascar, 3 general hypotheses make spatial predictions that can these rivers coincide with the limits of distribution areas be tested independently of any temporal predictions of species or phylogeographic lineages within species (e.g., Miocene WCs could generate the same general of lemurs and are therefore likely to constitute signifi- fragmentation patterns as the proposed Pleistocene con- cant barriers to gene flow for these animals (Pastorini tractions). Specifically, the Wilme´ et al. (2006) WC model et al. 2003). A similar situation seems to exist in east- states that contraction of mesic habitats within adjacent ern Madagascar and may especially influence species lowland watersheds during periods of glaciation cre- occurring at low elevations where rivers are widest ated gaps in ancestral species’ distributions, leading to (Goodman and Ganzhorn 2004; Louis et al. 2005). For allopatric differentiation/speciation. Assuming stability a more detailed overview of the processes inherent in the relative geographic positions of populations, this to the MR, WC, and RB hypotheses, see Vences et al. model therefore predicts that sister species should gen- (2009). erally occupy adjacent COEs or possibly a COE and an adjacent RDW (although the species in the RDW would be expected to have a wider geographic range). Because Brookesia Leaf Chameleons most of the major watersheds are composed of several The chameleon genus Brookesia is an excellent group smaller drainages, this model could also be construed to test hypotheses of species-level diversification and to predict that sister-species pairs would occupy the microendemism on Madagascar. These lizards, which same drainage. Likewise, the MR model of Raxworthy appear to form the sister taxon of all other chameleons and Nussbaum (1995) predicts that sister taxa will tend (Rieppel 1987; Townsend and Larson 2002), constitute to occupy montane forested areas separated by lower one of the largest Malagasy reptile groups (Raxworthy altitude dry forests. Finally, the RB model (Pastorini and Nussbaum 1995; Glaw and Vences 2007). Most et al. 2003) predicts species’ ranges to be delimited by 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 643 major lowland rivers. To evaluate these hypotheses, we TABLE 1. Nonchamaeleonid outgroup sampling for all phyloge- first reconstruct species ranges using a georeferenced netic analysesa database of Brookesia distribution records and adjust Higher taxon Representative taxonb these ranges (when sample sizes permit) through en- Rhyncocephalia Sphenodon punctatus vironmental niche modeling. We then use simulations Squamata to test the fit of our data to the spatial predictions of Dibamidae Dibamus Gekkonidae Gekko gecko the WC hypothesis, and we assess the qualitative fit of Xantusiidae Xantusia vigilis our data to the MR and RB hypotheses (all in a phylo- Scincidae Mabuya genetic context). Cordylidae Cordylus Teiidae Teiinae Finally, each of the 3 hypotheses suggests predictions Amphisbaenia Trogonophidae about the altitudinal distribution of species richness. Lacertidae Lacertidae The WC and RB models both predict a greater number Serpentes Dinodon Shinisauridae Shinisaurus crocodilurus of lowland species; low-headwater watersheds are Iguanidae Liolaemus more likely than high-headwater watersheds to become Uromastycinae Uromastyx fragmented from their neighbors during periods of in- Leiolepidinae Leiolepis creased aridity, and rivers are widest at lower elevations aAll phylogenetic analyses included these outgroup taxa as well as all and therefore more likely to constitute significant bar- non-Brookesia chamaeleonid taxa from Figure 2. riers to gene flow as altitude decreases. Conversely, the bComposite taxa with more than 1 species represented among the dif- MR model predicts higher species diversity in mon- ferent gene partitions are called by the name of the most exclusive higher taxon possible. tane areas because it is here that populations tended to become isolated during periods of forest contrac- tion; lowlands were subsequently recolonized by some species. We use these predictions to evaluate relative support for these hypotheses in a nonphylogenetic con- merase chain reaction products were sequenced with text by correlation of species richness with altitudinal ABI 3100 PRISMTM automated sequencers, and con- range. tigs were assembled using Sequencher 4.1.2 (Gene Codes Corporation, Ann Arbor, MI). Primer sequences are given in Table 2. We used the Clustal algorithm MATERIALS AND METHODS (Thompson et al. 1994) implemented in the program Taxa and Gene Regions Sampled DAMBE (Xia and Xie 2001) to align all sequences by A previous phylogenetic study (Townsend and Larson their amino acid translations and adjusted alignments 2002) identified 9 clades with uncertain interrelation- by eye using MacClade 4.03 (Maddison D.R. and ships that diverged from each other early in the his- Maddison W.P. 2000). Ambiguously aligned regions tory of Chamaeleonidae. Because current taxonomy were excluded from all analyses (see Results section). places 2 of these major clades within the genus Brookesia, Gaps were treated as a separate (binary) character parti- outgroup samples from all 7 non-Brookesia chamaeleonid tion in the MrBayes analyses and as missing data in the clades were included in this study. Sampling within BEAST and maximum-likelihood (ML) analyses. Brookesia was broad, including more than three-quarters Partitioned Bayesian and likelihood methods were of the named species as well as several undescribed used to infer phylogenies. BEAST (Drummond et al. forms, and species were sampled from multiple locali- 2006) implements a Bayesian relaxed-clock method that ties whenever possible. To facilitate divergence-dating allows the simultaneous estimation of topology and calibrations, outgroup sampling included representa- divergence times. The incorporation of a relaxed-clock tives of all major squamate lineages (Table 1). constraint into the phylogenetic inference procedure has All specimens were sequenced for 2 nuclear (RAG1, intuitive appeal, as it seems likely to model biological 1500 bp and CMOS, 800 bp) and 3 mitochondrial reality better than the 2 extreme alternatives of either a (∼ND1, 70 bp; ND2, ∼1035 bp; and ND4, 700 bp) strict clock or no clock at all (i.e., the unrooted method protein-coding∼ genes;∼ we included several ∼ND4 se- of Felsenstein 1981). The relaxed clock is also expected quences from Raxworthy et al. (2002) that expanded our to have greater statistical power due to the smaller taxonomic (3 species) or geographic (6 species) cover- number of estimated parameters relative to the un- age. Our final Brookesia sampling covered approximately rooted method (Drummond et al. 2006). MrBayes v3.1.2 28 species, including all but 2 of the named species. (Huelsenbeck and Ronquist 2001) and BEAST v1.4.6 Museum/collection and GenBank numbers of speci- (Drummond and Rambaut 2007) were used to conduct mens can be found in the online Supplementary Material Bayesian analyses under unrooted and relaxed-clock (available from http://sysbio.oxfordjournals.org). models, respectively. The data were divided a priori into partitions by gene and codon position, with mod- els chosen using the Akaike information criterion as implemented in MrModelTest (Nylander 2004). Analy- Molecular Data and Phylogenetic Analyses ses were performed under several a priori partitioning Genomic DNA was extracted and amplified using schemes, and the harmonic means of likelihood scores standard protocols (see Townsend et al. 2004). Poly- from the posterior distribution were compared using 644 SYSTEMATIC BIOLOGY VOL. 58

TABLE 2. Amplification and sequencing primers used in this study Directiona Location Sequence Reference

Forward ND1 50-CAACTAATACACCTACTATGAAA-30 Macey et al. (1997) Forward ND1 50-CGATTCCGATATGACCARCT-30 Kumazawa and Nishida (1993) Ala Reverse tRNA 50-AAAATRTCTGRGTTGCATTCAG-30 Macey et al. (1997) Forward ND4 50-TGACTACCAAAAGCTCATGTAGAAGC-30 Raxworthy et al. (2002) Leu Reverse tRNA 50-CATTACTTTTACTTGGATTTGCACCA-30 Raxworthy et al. (2002) Forward RAG1 50-TCTGAATGGAAATTCAAGCTGTT-30 Groth and Barrowclough (1999) Forward RAG1 50-CCACTTGGAAAAATACTCCCTGA-30 This study Reverse RAG1 50-GTCATCAACCAAATGTTGTATGCCTG-30 This study Reverse RAG1 50-GTGTCYACTGGGTARTCATC-30 Townsend et al. (2004) Forward CMOS 50-ATTATGCGATCMCCTMTTCC-30 This study Forward CMOS 50-TCTGGAATTTTCTCCWTCTGT-30 This study Reverse CMOS 50-GCTACCACAGARTASAGTACA-30 This study Note: tRNA = transfer RNA. aMitochondrial forward and reverse primers extend the light and heavy strands, respectively.

Bayes factors to choose a final partitioning scheme (see et al. 2001), 2) the oldest known stem scincomorphs Brandley et al. 2005). (here defined as the clade containing the crown taxa We used the ML program RAxML (Stamatakis 2006) Scincidae, Cordylidae, and Xantusiidae) (Evans 1993, via its Web server (Stamatakis et al. 2008) to estimate a 1998), 3) the oldest known anguimorph (Evans 1994, phylogeny and conduct nonparametric bootstrap anal- 1998), 4) the oldest known amphisbaenian (Gao and yses under the same partitioning scheme favored in Nessov 1998), and 5) the oldest teiids (Winkler et al. the Bayesian analyses, except that all partitions were 1990; Nydam and Cifelli 2002). Finally, the Comoros assigned a general time reversible model (currently the archipelago is home to the chameleon species Furcifer only option) with a proportion of invariable sites es- polleni and Furcifer cephalolepis, which are endemic to timated and branch lengths optimized separately for Mayotte and Grand Comoro, respectively. The oldest of each data partition. For simplicity, further references the 2 islands (Mayotte) formed 10–15 million years ago to the unrooted Bayesian, relaxed-clock Bayesian, and (Nougier et al. 1986), and the older of these dates served maximum-likelihood analyses will be referred to by the as 6) a gamma-distributed maximum age constraint for initials UB, RCB, and ML, respectively. the most recent common ancestor (mrca) of these taxa. We were concerned about the effects that substi- tutional saturation at deeper levels might have on Divergence Time Estimates divergence time estimates, especially with regard to In most cases, fossil constraints should place the max- mitochondrial DNA (mtDNA) data (Hugall et al. 2007). imum probability near the estimated age of the fossil, However, the mtDNA data were needed to confidently with probabilities dropping off rapidly for younger resolve more recent branching points (see below), sug- ages and more gradually for older ages (Benton and gesting also that branch lengths within Brookesia would Donoghue 2007), effectively placing a hard bound on be better estimated by including mtDNA data. We there- the minimum age and a soft bound (Yang and Rannala fore conducted analyses using 1) all data, 2) nuclear data 2006) on the maximum age. A translated lognormal plus first and second positions from the mtDNA data, (TL) distribution models this situation well (Hedges and 3) nuclear data alone to test the robustness of our and Kumar 2004; Drummond et al. 2006; Ho 2007) and divergence time estimates. We also tested our fossil cal- was used for most of the calibrations in this study. Be- ibrations using the cross-validation method of Near et cause overestimation of divergence times could lead to al. (2005), which identifies potentially problematic fos- false rejection of Pliocene/Pleistocene speciation scenar- sils that cause incongruent age estimates of other dated ios, we also ran analyses under a more conservative set nodes in the tree. of normally distributed fossil constraints in place of the All BEAST analyses were run for a sufficient number more geologically appropriate lognormal constraints. of generations to achieve an effective sample size of at Fossils or geologic data suitable for constraining least 200 for all estimated parameters, and 3 replicate nodes within Brookesia are not available, and we there- runs were conducted for each analysis. Initial analyses fore used several external calibration points for our were run without data to check the influence of the pri- analyses. For fossil-based calibrations, we constructed ors on the results. BEAST output was examined for ev- TL distributions in BEAST with the lower bound of the idence of proper mixing and convergence using Tracer 95% highest probability density (HPD) at a point 1% (Rambaut and Drummond 2004), runs were combined more recent than the estimated fossil age, and we left using LogCombiner (part of BEAST package), and max- (somewhat arbitrarily) long probability tails for the soft imum credibility trees with divergence time means and maxima. The following fossil-based calibration points 95% HPDs were produced using Tree Annotator (part were used (Table 3): 1) the rhynchocephalian–squamate of BEAST package). All BEAST files used are included split within lepidosaurs (Sues and Olsen 1990; Evans in the online Supplementary Material. 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 645

TABLE 3. Divergence date calibration priors Node Relevant fossila Median (95% CI) TL zero offset (million years ago) (mean, SD) Rhyncocephalian–squamate Indeterminate taxon 231 (225–301) 224 (2.0, 1.2) Scincomorph–iguanian Balnealacerta 167 (162–204) 161 (1.8, 1.0) Anguimorph–iguanian Parviraptor 167 (162–204) 161 (1.8, 1.0) Lacertid–amphisbaenian Hodzhhakulia 102 (97–182) 97 (1.7, 1.4) Teiid–(lacertid, amphisbaenian) Ptilotodon 116 (110–187) 110 (1.8, 1.3) Furcifer polleni–Furcifer cephalolepis N/A 6.3 (1.7–16) 0.0 (3.5, 2.0)b Notes: SD, standard deviation; N/A, not available. aSee text for references. bNonfossil calibration based on geologic data and modeled with a gamma distribution: zero offset (γ-shape, γ-scale) (see text).

Distribution Mapping and Diversity Analyses the model values are reduced until they reach zero. Locality data for all Brookesia species of Madagas- We used the following parameters: t = 40, b = 40, and car were gathered from museum data, our own global f = 80, as they provide a conservative scenario to remove positioning system readings, and pertinent literature. biogeographic overprediction areas (see Kremen et al. Single-species maps that almost fully agree with the 2008). data used herein are reproduced in Glaw and Vences To calculate values of spatial species richness and en- (2007). Small sample sizes of locality records are not demism, the distributional data (potential distribution appropriate to obtain reliable estimates of potential dis- models and point distribution data) were transformed tribution by modeling. For 9 species, more than 6 (range into a grid cell data set and plotted on a one-quarter de- 7–39) locality records were available. For these species, gree square grid covering the entire island. When more we defined distributions by potential distribution mod- than 2 records were available per species, we drew a els (pruned for overprediction), and for the rest of the minimum convex polygon between localities, and grid species, we used point locality data. cells within the polygon were considered to contain the For the models, we used 23 variables as predictors: species. We used Worldmap v. 4.20.18. (Williams 2002) to potential evapotranspiration, yearly water balance (an- calculate species richness as the total number of species nual rainfall minus annual evapotranspiration), number per grid cell. Endemism was calculated as a measure of months with a positive water balance (a measure of of range-size rarity, expressed as the percentage aggre- drought), percentage of forest cover in 2000, and 19 gated reciprocal range size for all species per grid cell. climatic variables from the WorldClim database ver- sion 1.4 (Hijmans et al. 2005): annual mean tempera- ture; mean diurnal temperature range; isothermality Hypothesis Testing (monthly/annual temperature range); temperature sea- Temporal concordance with the Wilme´ et al. (2006) sonality (standard deviation across months); maximum and the Raxworthy and Nussbaum (1995) Pliocene/ temperature of warmest month; minimum temperature Pleistocene speciation scenarios (WC and MR hypothe- of coldest month; annual temperature range; mean tem- ses, respectively) was statistically tested by comparing perature of wettest, driest, warmest, and coldest quar- the most recent tail of the 95% HPD for each sister- ters; annual precipitation; precipitation of wettest and species pair in our phylogeny to the oldest limits of driest months; precipitation seasonality (coefficient of these geological epochs. The RB hypothesis was not variation); and precipitation of wettest, driest, warmest, formulated with an explicit time frame and thus could and coldest quarters. not be tested in this manner. We used Maxent v2.3 (Phillips et al. 2006) for dis- The Wilme´ et al. (2006) WC model predicts that sis- tribution modeling, as it performs well in predicting ter species should generally occupy adjacent COEs or species distributions (Elith et al. 2006), even when sam- possibly a COE and an adjacent RDW. To test the fit of ple sizes are small (Hernandez et al. 2006). Analyses our Brookesia data to the spatial aspect of this model, followed Vieites et al. (2008), using 1000 randomly se- we first plotted the distributions of all Brookesia sister- lected data points across Madagascar as background species pairs identified in our phylogenetic analyses pseudoabsence data. All models had areas under the onto the Wilme´ et al. (2006) watershed map (Fig. 1) receiver operating characteristic curve (Hanley and and counted the number of pairs matching the model’s McNeil 1982) higher than 0.7, suggesting that the mod- predictions. Next, we randomly assigned each species els were good at discriminating between presence and from all sister-species pairs to one of the watersheds absence sites (Fielding and Bell 1997). We applied a collectively occupied by them and counted the number pruning algorithm to remove areas of overprediction of pairs fitting the model’s predictions. We repeated from the mean model. This algorithm thresholds the this 10,000 times to generate a null distribution of ex- Maxent output models using a user-defined probabil- pected number of fits to the model if sister species are ity of occurrence (t), defining a convex hull around all actually distributed randomly with respect to relative occupied regions, and buffering this hull by a given watershed positions. If less than 5% of the simulation number of grid cells (b) and width (f ), within which rounds resulted in a number of matches equal to or 646 SYSTEMATIC BIOLOGY VOL. 58

Several factors make it impractical to test this MR hy- pothesis with our data using the statistical framework outlined above for the WC hypothesis. First, these AOEs comprise only a fraction of the land area of Madagascar (see Fig. 1), which would necessitate the exclusion of several sister-species pairs that are not confined to them (see Results section) and thus reduce statistical power. Also, there are multiple potential vicariant relationships among most of these areas (e.g., the northeastern region could have once been connected by suitable habitat to all the other AOEs), so the number of sister-species pairs consistent with the MR hypothesis under the null model (i.e., no real effect of forest contraction on speciation) would be high. Furthermore, as stated in Raxworthy and Nussbaum (1995), there is no clear bio- geographic boundary between the northeastern and the eastern AOEs, making interpretation of relationships among taxa from these areas unclear. However, there is intuitive appeal to arguments that forest-fragment contraction promoted vicariant speciation in geolog- ically complex northern Madagascar. We thus exam- ine congruence of several potential examples with our new phylogenetic hypothesis (some of the proposed examples of Raxworthy and Nussbaum were flawed because they compared nonsister taxa; see Results section). The RB hypothesis (e.g., Pastorini et al. 2003) pre- dicts sister taxa to be separated by major rivers. A few Malagasy rivers can be unambiguously defined as “major” (e.g., the Betsiboka, Mangoro, and Mana- nara rivers) because they are quite long, with headwa- ters at relatively high elevations (Wilme´ et al. 2006), FIGURE 1. Map showing proposed centers of endemism (COEs; dark gray), retreat-dispersion watersheds (RDWs; white), and the and because previous works (e.g., Pastorini et al. 2003) 3 highest peaks in Madagascar. Light gray areas are designated as have identified them as potential barriers for lemurs. RDWs, with possibly some function as COEs. Modified from Wilme´ However, for most rivers, an objective assessment of et al. (2006). their potential as barriers would require combining parameters such as length, width, headwater eleva- greater than the number of matches from the real data, tion, and permanence. These factors, combined with we rejected the null in favor of the Wilme´ et al. (2006) the difficulty of determining the size and shape ofthe WC hypothesis. We repeated this analysis under the in- test areas, preclude a formalized test of the RB hypoth- terpretation that allopatric sister species occupying the esis. However, we examine the spatial relationship of same watershed also fit the model. The MR hypothesis of Raxworthy and Nussbaum Brookesia sister taxa with several of the largest rivers to (1995) states that rainforest contraction around high- look for any evidence of a major role in species-level elevation massifs during cooler periods isolated popu- diversification. Our geographic sampling within some lations on either side of unsuitable intervening habitat, species also allows some evaluation of the role that leading to allopatric speciation. Referencing Figure 1, rivers might play in intraspecific genetic structuring the 5 proposed areas of endemism (AOEs) of Raxwor- within Brookesia. thy and Nussbaum (1995, see their fig. 7) correspond The WC, RB, and MR hypotheses all make general to 1) the Montagne d’Ambre mountain range in the far predictions about relative species abundance at differ- north (straddling the border between COEs 1 and 12); 2) ent altitudes. Following Wollenberg et al. (2008), we di- a northwestern region west from Tsaratanana to Nosy vided Madagascar into a coarser grid with cell sizes of Be and Presqu’Ile d’Ampasindava; 3) the Tsaratanana 82 63 km = 5166 km2 and used the ArcView extension × massif and surrounding high-altitude environs; 4) a “Endemicity Tools” (provided by N. Danho) to calcu- northeast region extending east from Tsaratanana and late Brookesia species richness and corrected weighted encompassing parts of the RDWs A and a2, the southern endemism (Crisp et al. 2001) for each cell. We then tested tip of COE 1, and the northern section of COE 2, with for nonparametric correlation of these values with al- the Antongil Bay watershed as its southwestern border; titudinal range, defined as maximum–minimum eleva- and finally 5) an eastern region corresponding to at least tions per grid cell (excluding grids with no occurrence the southern section of COE 2. of any Brookesia species). 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 647

RESULTS bonsi, and B. brygooi. These species are restricted to the Data Characteristics and Phylogenetic Analyses dry deciduous forests of the west and/or southwest and together form the sister taxon of the remaining typical Alignments were unambiguous across chameleons Brookesia. The eastern rainforest sister species Brookesia for all partitions. However, alignment of the 3’ ends therezieni and Brookesia superciliaris (Clade 2) and the of ND1 ( 70 bp) and ND2 ( 90 bp), as well as part northern rainforest species Brookesia ebenaui (Clade 3) of ND4 ( ∼60 bp), was problematic∼ across more distant ∼ are sister taxa to progressively less inclusive clades. outgroups due to high levels of amino acid replacement Clade 4 contains several populations that show sub- and multiple indels. We therefore coded these regions stantial sequence divergence from each other (11.1% as missing data for nonchamaeleonid taxa. Maximum- average uncorrected ND4 divergence among specimens likelihood analyses of near-complete ND1 and ND2 data from distinct localities) but whose interrelationships are across several chameleon species found very similar mostly poorly supported. The eastern rainforest species evolutionary model parameter values for these 2 genes Brookesia thieli is paraphyletic, although basal diver- (results not shown). To avoid the statistical problem of gences are poorly supported. One subclade of B. thieli low data/parameter ratios, the ND1 sequence ( 70 bp) ∼ appears to be the sister taxon of the morphologically was combined with the ND2 for partitioning purposes. distinctive Brookesia vadoni, suggesting the possibility of For the same reason, data from the 3 transfer RNA genes cryptic species within B. thieli. Finally, Clade 5 comprises separating ND1 and ND2 (and from which the loop re- several species (Brookesia antakarana, Brookesia ambreen- gions were impossible to align confidently) were not sis, Brookesia valerieae, Brookesia griveaudi, and Brookesia used in this study. Bayes factor analysis on various par- stumpffi) largely restricted to evergreen rainforest of titioning schemes (2, 3, 6, 9, and 12 partitions) strongly the north, northeast, and northwest (B. stumpffi is ex- favored the 12-partition scheme (nuclear and mtDNA ceptional among Brookesia in its adaptation to both dry genes each partitioned by codon position; Bayes factor deciduous and evergreen forest). Brookesia valerieae is 178 in all comparisons). All results reported here are ≥ strongly supported by all analyses as the sister taxon of from this partitioning scheme. B. griveaudi. Brookesia stumpffi is weakly supported as the Within each of the ML, UB, and RCB analysis sets, sister taxon of a strongly monophyletic B. antakarana–B. phylogenetic results were broadly congruent across ambreensis clade in the RCB analysis (Fig. 2). However, the different genomes and genes, with topological con- the other 2 methods find weak (ML, 54% bootstrap) flicts involving only poorly supported nodes (i.e., ML to strong (UB, 96% PP) support for B. stumpffi as the bootstraps < 70%, Bayesian posterior probabilities [PPs] sister taxon of the B. griveaudi–B. valerieae clade, with < 95%). One exception to this pattern involves strongly this larger clade as the sister group to the partially conflicting nuclear and mitochondrial results for the sympatric species pair B. antakarana and B. ambreensis placement of 1 specimen of Brookesia brygooi. This dis- (Fig. 2). These latter 2 species exhibit very low levels crepancy is probably due to ancient mtDNA introgres- of divergence from one another (0.6–1.3% uncorrected sion, and the nuclear topology almost certainly reflects ND4 distance; see Fig. 2 inset) and may not be recip- the true organismal history (see online Supplementary rocally monophyletic. There do appear to be 2 distinct Material). Mitochondrial data for this specimen were morphologies in this lineage (see photos in Glaw and therefore excluded in all further phylogenetic analy- Vences 2007), but mtDNA data from several additional ses. In general, the mtDNA data alone gave strong specimens suggest that either incomplete lineage sort- support for relationships within more terminal clades, ing or mitochondrial introgression is a major factor at but poor support for basal nodes, and the nuclear data this site (results not shown). showed the opposite pattern. All results presented here Generally, the RCB method gave a more precise phy- are from analyses of the combined data set (Fig. 2). logenetic estimate than did the UB method (83% vs. 72% This data set and trees from this study can be found at of nodes significantly supported in Fig. 2). The increase www.treebase.org under accession number SN4455. in support may be a consequence of the expected in- The mrca of the B. nasus–B. lolontany clade and all crease in statistical power inherent to the relaxed-clock other Brookesia appears early in the history of the group approach. Alternatively, model misspecification may (although monophyly of neither Brookesia nor this sub- play a role in the difference. clade is significantly supported; Fig. 2). The remainder of the genus is clearly monophyletic and is divided into 2 well-supported clades. One of these clades corre- Geographical Centers of Diversity and Endemism in sponds to the extremely miniaturized B. minima group. Brookesia The other major clade, representing the typical Brooke- For the 9 modelable species, all distribution models sia morphotype, consists of a series of 5 maximally predicted known localities, although comparison with supported subclades (one of which is a single species) museum records indicated some overprediction. The whose interrelationships are strongly supported by the pruning algorithm removed some (but not all) areas RCB analysis but less so by the UB and ML analyses appearing to have suitable environmental niches for the (Fig. 2). species, but which were located far from the known Referencing the numbered clades in Figure 2, Clade distribution ranges. This suggests that nonautecological 1 contains Brookesia perarmata, Brookesia decaryi, Brookesia (e.g., biogeographical, community ecological) factors 648 SYSTEMATIC BIOLOGY VOL. 58

FIGURE 2. Combined mitochondrial (ND1/ND2/ND4) and nuclear (RAG1/CMOS) data, 12-partition relaxed-clock Bayesian (RCB) clado- gram with maximum-likelihood (ML) phylogram insert. Analyses included all ougroup taxa, although for clarity, most nonchameleon taxa are not shown. Major clades are color coded, and major clades of “typical” Brookesia are numbered (see text). RCB/unrooted Bayesian (UB) PP >95% are denoted by asterisks (PP between 90% and 95% are shown as actual values) above branches, and ML bootstrap values >50% are shown below branches. Black diamond indicates alternate attachment point for Brookesia stumpffi clade in the UB and ML analyses. Dashed terminal branches mark individuals represented by ND4 data only. 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 649

FIGURE 3. Maps illustrating (a) differential Brookesia species diversity (richness) across Madagascar, quantified as numbers of species present within one-quarter degree square grid cells covering the entire island and (b) degree of endemism in different parts of Madagascar. Endemism is calculated as a measure of range-size rarity, expressed as the percentage aggregated reciprocal range size for all species per one-quarter degree square grid cell. are needed to explain the restricted microendemic dis- (95% confidence interval [CI] from 63 to 81 million tribution of some species. years ago; Fig. 4), which is possibly∼ older∼ than any Brookesia distributions span the island, although they divergence within non-Brookesia chameleons (see Sup- are absent from many large areas in the central high- plementary Material). Generally, divergences between lands and the south. Three main centers of diversity recognized species groups and even sister taxa appear to (Fig. 3a) are found in the northeast, north, and north- be quite deep. Within the typical Brookesia clade, aside west, respectively. These areas are also COEs, with the from the problematic B. ambreensis–B. antakarana com- highest percent values of range-size rarity (Fig. 3b). plex, the most recent sister-species split (B. bonsi–B. However, there are several areas, mainly in the west decaryi) has a 95% CI extending no more recently than and northwest regions of the island, with a high de- the Middle Miocene (Fig. 4). Within the B. minima clade, gree of endemism corresponding to several species mean estimates for sister-taxon divergences are even only known from single locality records. older (the most recent divergence time mean is at the The 3 main clades recovered in our phylogenetic Eocene–Oligocene boundary), although the 95% CIs analyses (Figs. 2 and 4) show distinct biogeographic within the 2 clades overlap. Thus, we can confidently patterns. The basally diverging B. nasus clade contains reject the temporal component (i.e., Pliocene–Pleistocene 2 species, B. nasus in the southeast and B. lolontany in the time frame) of both the MR (Raxworthy and Nussbaum north, suggesting an old north-south connection. The B. 1995) and the WC (Wilme´ et al. 2006) hypotheses. minima clade shows the highest degree of endemism; Additional analyses using alternative calibration dis- most of the 11 species are restricted to very small areas tributions and data sets had little effect on divergence in the north and north-central regions. The third and time estimates. Analyses performed with normally dis- largest clade (typical Brookesia) spans most of the island, tributed calibrations (with standard deviations arbitrar- although centers of both diversity and endemism in this ily set at 10% of the mean) gave average intrachameleon clade are in the north. Within this clade, B. brygooi, B. divergences about 10% more recent than those using bonsi, B. decaryi, and B. perarmata (all western species; lognormally distributed calibrations. Likewise, analy- Clade 1, Fig. 2) form the sister taxon of the remaining ses that excluded mtDNA third codon positions and all species, none of which are restricted to the western mtDNA data gave respective divergence estimates on forests. average about 15% and 18% more recent than the full- data analysis. In all these analyses, Brookesia sister-taxon divergences remained solidly Miocene in age, and thus, Timing of Diversification in Brookesia our above conclusions are unaffected. The RCB 12-partition analysis places the basal split Using the method of Near et al. (2005), we found no within Brookesia at approximately 72 million years ago significant incongruence among our fossil calibrations 650 SYSTEMATIC BIOLOGY VOL. 58

FIGURE 4. Chronogram from the 12-partition (ND1/ND2, ND4, CMOS, and RAG1, each partitioned by codon position) RCB phylogenetic analysis. Time units on scale bar in millions of years ago. See Supplementary Materials for exact divergence time means and 95% CIs for all nodes. Maps above chronogram illustrate patterns of species richness (quantified as numbers of species present within one-quarter degree square grid cells) for the 3 main clades (1 = Brookesia nasus group, 2 = Brookesia minima group, and 3 = “typical” Brookesia group). 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 651 with one exception; the root calibration (representing predict. Nonetheless, accepting these 3 pairs as poten- the rhynchocephalian/squamate split) was a signifi- tial fits to the model, we performed 10,000 rounds of cant outlier (P = 0.017). Following Near et al. (2005), random allocations of the 20 relevant species (i.e., those this calibration should be discarded as potentially mis- belonging to sister-species pairs) to the watersheds leading. Exclusion of this calibration point draws all collectively occupied by them and found that 12% of extrachamaeleonid divergences to markedly more re- the simulation rounds allocated the members of 3 or cent dates, and the estimated root age varies from ap- more sister-species pairs to adjacent watersheds (i.e., proximately 170–150 million years ago. However, fossil P = 0.12). When allopatric taxa occupying the same rhynchocephalians are known from multiple Laurasian watershed were included as potential matches in the and Gondwanan localities by the Late Triassic (Evans simulations (there were none of these in the real data), et al. 2001). Thus, even allowing for reasonable error in P = 0.32. Thus, we found no support for the spatial fossil dates, Middle to Late Jurassic estimates for the component of the WC hypothesis. Rhynchocephalia–Squamata split are clearly untenable Unlike the WC hypothesis, the MR hypothesis of (see Marshall 2008 for further discussion of the poten- Raxworthy and Nussbaum (1995) was formulated with tial pitfalls of excluding outliers). We therefore favor the respect to 5 AOEs in the north and east of Madagascar results from the full constraints analysis. That said, we (see Materials and Methods section and Fig. 1), as op- note that even in analyses performed without the root posed to watersheds covering the entire island. Of the constraint, in no case did the 95% HPDs for Brookesia 10 Brookesia sister-species pairs from our phylogeny sister-taxon divergences reach the Pliocene. (Fig. 2), only a subset of these taxa are distributed in these areas and therefore useful for evaluation of this hypothesis. Specifically, the species pairs B. antakarana– Fit of Spatial Hypotheses B. ambreensis, B. griveaudi–B. valeriae, Brookesia karchei– We compared the distribution of Brookesia with the Brookesia peyrierasi, and B. tuberculata–B. sp. “Montagne 12–15 COEs recently proposed by Wilme´ et al. (2006) de Francais” are wholly confined to these AOEs. Of (Fig. 1). Of approximately 30 Brookesia species, 18 are these pairs, B. griveaudi (northeast) and B. valeriae (north- limited to single COEs (3 of these are also found in 1 west) both inhabit forests less than about 900 m alti- RDW), 9 are found in 2 or more separate COEs (1 is tude and are absent from the higher altitude (>1600 m) also found in an RDW), and 2 additional species are forests of the intervening Tsaratanana massif. As pro- limited to single RDWs. However, these numbers are posed by Raxworthy and Nussbaum (1995), these areas difficult to interpret for several reasons. First, Brookesia may once have been connected by an east-west corri- distributions are heavily skewed toward the northern dor of even lower altitude forest that was lost during a end of the island, and certain COEs are occupied by period of aridification. Tsaratanana itself is home to B. disproportionate numbers of species. For example, 22 lolontany, likely sister taxon of the only other Brookesia of 30 species (73%) are distributed across just 6 COEs, species with populations known to inhabit this alti- and 7 of these species are restricted to a single COE (S. tudinal range, B. nasus of southeastern Madagascar. Bemarivo/N. Mangoro [2] on the eastern and northeast- Although none of the other species pair distributions ern coast) (Fig. 1). Furthermore, the level of microen- are consistent with this spatial model, 2 interesting demism within Brookesia is extreme: 18 of 30 species higher level patterns within the B. minima group clade are essentially known from single localities (Glaw and are potential fits. First, B. minima is restricted to the Vences 2007). Thus, most species of Brookesia will be northwest, whereas its sister clade (B. tuberculata + confined to single COEs regardless of the mechanism B. sp. “Montagne des Francais”) is restricted to the of their isolation. But perhaps most importantly, sim- Montagne d’Ambre region. Second, the clade formed by ple tabulations of species counts across proposed COEs these 3 species is found only to the west of Tsaratanana, ignores phylogeny completely. whereas its sister clade (B. cf. karchei + B. peyrierasi) is To test for spatial concordance with the WC hy- found only to the east of Tsaratanana. pothesis, we first plotted the distributions of the10 Finally, a comparison of Brookesia distribution and strongly supported (PP > 95%) Brookesia sister-species phylogeny with major rivers yields no clear support pairs identified in our phylogenetic analysis (i.e., 20 for the RB hypothesis. The Betsiboka River in western species, see Fig. 2, also Supplementary Table S1) onto Madagascar, which has been shown to be a major barrier the Wilme´ et al. watershed map (Fig. 1). Although for several lemur populations (Pastorini et al. 2003), may none of these sister-species pairs are strictly confined possibly lay between the sister species B. exarmata (to to adjacent COEs, Brookesia tuberculata is found on the southwest) and B. dentata (to the northeast). How- Montagne d’Ambre, which straddles COEs 1 and 12, ever, our B. dentata is the only known specimen from and its sister species Brookesia sp. “Montagne de Fran- Ankaranfantsika National Park, which actually strad- cais” occupies COE 1. Two sister-species pairs (Brookesia dles the river. At any rate, B. exarmata is confined to the exarmata [8]/Brookesia dentata [G; also 9] and Brookesia Tsingy de Bemaraha approximately 300 km (and sev- sp.“Betampona” [2]/Brookesia ramanantsoai [B]) follow eral more rivers) away to the southwest. Similarly, B. the adjacent COE/RDW pattern (Figs. 1 and 2). How- bonsi is found at a single locality (Namoroka) about 75 ever, neither of these RDW residents is a wide-ranging and 100 km to the west of the large Mahavavy and species, as the Wilme´ et al. (2006) hypothesis would Betsiboka rivers, respectively, and its sister species 652 SYSTEMATIC BIOLOGY VOL. 58

B. decaryi is known only from Ankarafantsika. Once sity is constrained and biased by the relative availabi- again, the highly restricted and disjunct distributions of lity of climatological and geological data from different these species make any inference of a river barrier as periods in the earth’s history. In general, Pleistocene the isolating mechanism highly speculative. No other climatic changes (specifically, glaciation cycles that have sister-species pair (or higher level sister-clade pair) from predictable effects on sea level, aridity, and tempera- our phylogeny is divided by what might reasonably be ture) are much better understood than those from considered a major river. the Pliocene, Miocene, and earlier (Wells 2003). This River barriers can of course play a role in intraspecific may explain their prominence in many glaciation- genetic structuring (e.g., Pastorini et al. 2003; Kozak et al. related diversification theories. Results from phylo- 2006; Lemmon et al. 2007), and there are a few Brookesia geographic studies have challenged the importance of species with distributions that span large rivers (B. na- the most recent Pleistocene and Holocene glaciations sus, B. superciliaris, and B. thieli in the east; B. stumpffi as major drivers of species-level diversification for sev- and possibly B. ebenaui in the north; and B. brygooi in eral groups (Klicka and Zink 1997; e.g., Avise et al. the west). To fully study the phylogeographic effects of 1998; Rull 2008). In Madagascar, the existence of Pleis- these rivers will require much more thorough sam- tocene glacial activity and dynamic vegetational shifts pling. However, our sampling within some species does has been documented (Burney 1995; Vidal-Roman´ı et al. allow at least some preliminary comments. The western 2002), but well-studied examples indicating the depen- B. brygooi has an extensive distribution, including pop- dence of species diversification on these processes are ulations on both sides of the Betsiboka River. However, lacking. the deepest split within B. brygooi separates a population Most phylogeny-based biogeographical studies of to the south of the river from multiple populations span- Madagascar have centered on the relative importance of ning both sides of the river. Likewise, the distribution of vicariance and dispersal as the source of Madagascar’s B. stumpffi spans the Sambirano, Mahavavy, and Manan- many endemic higher taxa (Yoder and Nowak 2006 jeba rivers (among others) in northwestern Madagascar. and references therein). Given the long-standing inter- Although the deepest split does separate a popula- est in Madagascar’s high level of micorendemicity, it tion to the south of the Sambirano River from several is surprising that only a few phylogenetic studies have more northern populations, relationships among the explicitly estimated sibling-species divergence times remaining populations do not support the Mahavavy (e.g., Vences et al. 2002; Raxworthy et al. 2008; Wirta et and Mananjeba rivers as barriers to gene flow. Finally, al. 2008). To our knowledge, none of these studies have B. nasus is found both to the north and to the south of found species-level divergence times compatible with the large Mananara River in southeastern Madagascar. Pleistocene–Pliocene radiations. Other mtDNA studies However, an individual from the northernmost pop- of various groups including geckos, skinks, spiders, and ulation is nested within a clade of individuals from lemurs (Yoder et al. 2000; Vences et al. 2002; Schmitz et populations to the south of the potential barrier. See al. 2005; Wood et al. 2007; Jackman et al. 2008) have re- Supplementary Material for figures illustrating these ported genetic divergences that are likewise incompat- scenarios. Thus, although further sampling could reveal ible with widespread Pleistocene–Pliocene speciation, a more complex pattern (possibly also involving smaller using standard rates of mtDNA sequence evolution (but rivers), our data do not suggest a major river effect on see Wirta and Montreuil 2008 for a possible Pliocene genetic structuring within any of these species. radiation of beetles). Thus, although well-sampled phy- Species richness of Brookesia was negatively correlated logenies of many more groups are needed, existing with the minimum altitude (P < 0.05) and positively evidence (including this study) suggests that recent cli- correlated with the maximum altitude (P < 0.001) in matological cycles have had relatively minor effects on each grid cell. Species richness and corrected weighted levels of species richness and associated microendemic- endemism of Brookesia furthermore showed a signif- ity on Madagascar. icant correlation with altitudinal range per grid cell Theoretical considerations (e.g., Hewitt 1996) and (P < 0.001 and P < 0.05, respectively). These results are empirical examples from other groups have suggested not congruent with the RB and WC hypotheses, both of the importance of earlier glaciation cycles on the North which predict a higher species richness and endemism American and European flora and fauna (Veith et al. at lower elevations where rivers are widest and where 2003; Ayoub and Riechert 2004; Hewitt 2004; Kadereit river basins forming COEs have their headwaters. In et al. 2004; Turgeon et al. 2005). Unfortunately, histori- contrast, the results are consistent with the MR hypoth- cal species distributions in Madagascar have been very esis, which predicts more species will be restricted to difficult to determine due to the paucity of Tertiary fos- montane regions where there is a higher probability of sil deposits earlier than the Quaternary (Wells 2003). isolation during dry periods. This has in turn greatly limited the ability to infer pa- leoclimates and historical vegetation cover that might help explain existing complex microendemic distribu- tion patterns. Nonetheless, assuming relative stability of DISCUSSION drainage systems and topographical features, resultant Our ability to evaluate past processes that produced phylogenetic patterns could be very similar between current patterns of organismal distribution and diver- earlier and more recent climatic cycles. An evaluation 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 653 of the competing hypotheses thus depends largely on and leaf litter microhabitat, they could quite plausibly our knowledge of current distribution, habitat, and float across rivers on debris washed away during heavy phylogeny. rains.

Watershed Contractions, Mountain Refuges, and Higher Level Phylogenetic and Biogeographical Patterns Riverine Barriers Wells (2003) hypothesized that the northward drift Our test of the spatial component of the Wilme´ et al. of Madagascar from its pre-Paleocene position south of WC hypothesis is potentially biased in that phylogenet- the subtropical arid zone precipitated the aridification of ically closer species might be expected to have distribu- most or all the island, causing the loss or great reduction tions geographically closer than more distantly related of much of the existing biodiversity. Later entry into the species. This could result in a nonrandom association “trade wind” zone of southeasterly winds (most likely between sister taxa and adjacent watersheds, even if WC sometime in the Eocene) brought rains to the east coast, was not the cause of their separation. Thus, our failure causing the formation of the eastern humid forests. to reject a random watershed/sister-taxon association, According to Wells (2003), this increased moisture, combined with a negative correlation between altitude combined with flow of the warm southern equatorial and species richness, strongly suggests that this mecha- current through the widening Mozambique Channel, nism cannot explain current Brookesia distributions. On also led to the formation of the western deciduous the other hand, because species-level divergences in this forests. Correlating this scenario with our dated phy- group are relatively deep (Late Eocene to Late Miocene), logeny allows some speculation on drivers of diversifi- there is the possibility that extinctions, climate/habitat cation and distribution at higher levels within Brookesia. changes, and range expansions have obscured patterns Specifically, aridification may have fragmented the an- that would support this hypothesis. These are problems cient B. nasus clade into disjunct high montane areas, inherent to the study of any older radiation (Losos and with expansion of B. nasus into lower altitudes of the Glor 2003). Unless new relevant fossils (including those eastern rainforest during the Eocene–Oligocene. Most indicative of past climates) are discovered, the first 2 diversification within both the B. minima and the typ- problems are largely unavoidable. However, Brookesia ical Brookesia clades occurred during this same period may be less prone than many groups to the last poten- and was thus likely tied to the same expansion of mesic tial problem due to the probable low vagility of these habitats. The striking size disparity between members leaf litter inhabitants. of these 2 clades, combined with largely overlapping The Raxworthy and Nussbaum (1995) MR model, distributions, suggests some sort of niche partitioning although restricted to northern Madagascar, seems to (most likely prey size and/or microhabitat usage), but fit to our data reasonably well. Our study reveals that data supporting or refuting this hypothesis are lacking. some of the sister-taxon pairs originally cited in favor Our geographical analyses confirm that northern of this model (B. ambreensis/B. valerieae and B. stumpffi Madagascar is the center of species richness and en- /B. griveaudi) are actually not sister taxa but only more demism of Brookesia (Fig. 3), as postulated by Raxworthy distantly related. Nonetheless, the Tsaratanana massif and Nussbaum (1995). However, within the clade of typ- and associated montane areas do separate 2 sister taxa ical Brookesia, the most basal splits do not divide taxa as well as 2 multispecies clades, and another clade is restricted to the north. Instead, the first 2 splits sepa- split between the Northwest and Montagne d’Ambre rate a western and eastern clade, respectively, from the regions. We noted earlier that for any given forest frag- remainder, which itself contains mostly northern and a ment, multiple potential vicariance relationships often few more nested eastern species. This pattern suggests exist. This fact may account for some of the apparent ancient east-west dispersal or vicariance with subse- fit to this model. However, independent support for quent diversification within these areas and a later di- this model is provided by the significant positive cor- versification in the north. Although the B. minima group relation of altitude and altitudinal range with species is speciose, it is split at its base into a north-restricted richness and endemism, which is the pattern expected if clade and a more southern clade. Thus, hypotheses of most speciation occurs by fragmentation and isolation northern or southern origins for this group are equally in montane forest fragments. parsimonious. However, within the southern clade, Rivers appear to have played a substantial role in the there is a clear basal split between 1 western (dry de- diversification of several lemur groups (Pastorini etal. ciduous) and 2 eastern (rainforest) groups. Interestingly, 2003; Goodman and Ganzhorn 2004; Louis et al. 2005), the repeated east-west splits within Brookesia represent some frogs (Vieites et al. 2006), and possibly even some a type of within-landmass biome shift in one of the lin- chameleons of the genus Furcifer (Raselimanana and eages, a phenomenon that was found to be quite rare in Rakotomamalala 2003). However, this does not seem to a recent global-wide study of plants (Crisp et al. 2009). be true for leaf chameleons, at least at the species level In summary, higher level diversification within Brooke- and above. It seems unlikely that individual Brookesia sia was quite possibly tied to an Eocene–Oligocene shift disperse over large distances, and they certainly can- to more mesic habitats on Madagascar, and our analyses not swim across rivers. However, given their small size support a substantial role for rainforest fragmentation 654 SYSTEMATIC BIOLOGY VOL. 58

(especially in the elevationally heterogeneous north) in Benton M.J., Donoghue P.C.J. 2007. Paleontological evidence to date species-level diversification. A true test of any model the tree of life. Mol. Biol. Evol. 24:26–53. purported to generally explain the complex patterns of Boumans L., Vieites D.R., Glaw F., Vences M. 2007. Geographical pat- terns of deep mitochondrial differentiation in widespread Mala- diversity in the Malagasy fauna will require at least a gasy reptiles. Mol. Phylogenet. Evol. 45:822–839. combination of distributional and current climatic data Brandley M.C., Schmitz A., Reeder T.W. 2005. Partitioned Bayesian such as that provided by Wilme´ et al. (2006) and a se- analyses, partition choice, and the phylogenetic relationships of ries of well-supported phylogenies (with reasonably scincid lizards. Syst. Biol. 54:373–390. Briggs J.C. 2003. The biogeographic and tectonic history of India. accurate estimates of divergence times) from a vari- J. Biogeogr. 30:381–388. ety of taxonomic groups. As paleoclimatological and Burney D.A. 1995. Theories and facts on Holocene environmen- paleovegetational reconstructions become available, tal change before and after human colonization. Environmental they will also be very valuable because Malagasy change in Madagascar. In: D. Rakotondravony et al. eds., Fiovan’ny terrestrial fossil deposits from the Tertiary are virtually tontolo iainana eto Madagasikara. Chicago: Field Museum of Natural History. p. 108–109. nonexistent. Carpenter A.I., Robson O. 2005. A review of the endemic chameleon genus Brookesia from Madagascar, and the rationale for its listing on CITES Appendix II. Oryx. 39:375–380. SUPPLEMENTARY MATERIAL Crisp M.D., Arroyo M.T.K., Cook L.G., Gandolfo M.A., Jordan G.J., McGlone M.S., Weston P.H., Westoby M., Wilf P., Linder H.P. Supplementary material can be found at http:// 2009. Phylogenetic biome conservatism on a global scale. Nature. www.sysbio.oxfordjournals.org/. 458:754–756. Crisp M.D., Laffan S., Linder H.P., Monro A. 2001. Endemism in the Australian flora. J. Biogeogr. 28:183–198. Drummond A.J., Ho S.Y.W., Phillips M.J., Rambaut A. 2006. Relaxed FUNDING phylogenetics and dating with confidence. PLoS Biol. 4:e88. This work was supported by grants from the Deutsche Drummond A.J., Rambaut A. 2007. BEAST: Bayesian evolutionary analysis by sampling trees. BMC Evol. Biol. 7:214. Forschungsgemeinschaft and the Volkswagen Foun- Elith J., Graham C.H., Anderson R.P., Dudik M., Ferrier S., Guisan dation to M.V. and F.G. and by U.S. National Science A., Hijmans R.J., Huettmann F., Leathwick J.R., Lehmann A., Li J., Foundation grants (postdoctoral bioinformatics fellow- Lohmann L.G., Loiselle B.A., Manion G., Moritz C., Nakamura M., ship DBI-0204451 to T.M.T., Tree of Life NSF 1-10135 Nakazawa Y., Overton J.M., Peterson A.T., Phillips S.J., Richardson grant to D.R.V., and Tree of Life grant EF 0334923 to Tod K., Scachetti-Pereira R., Schapire R.E., Soberon J., Williams S., Wisz M.S., Zimmermann N.E. 2006. Novel methods improve prediction Reeder). of species’ distributions from occurrence data. Ecography. 29:129– 151. Evans S.E. 1993. Jurassic lizard assemblages. In: Buffetaut E., Mazin ACKNOWLEGMENTS J.-M., editors. Proceedings of the Second Georges Cuvier Sym- posium, Revue de Paleobiologie, Special Volume 7. Montbeliard We are grateful to Daniel Scantlebury for assistance (France). p. 55–65. in the lab and to Edward Louis of the Henry Doorly Evans S.E. 1994. A new anguimorph lizard from the Jurassic and Zoo’s Center for Conservation and Research for provid- Lower Cretaceous of England. Palaeontology. 37:33–49. Evans S.E. 1998. Crown-group lizards from the Middle Jurassic of ing collecting opportunities for T.M.T. in Madagascar. Britain. Palaeontographica A. 250:1–32. Akira Mori generously provided important additional Evans S.E., Prasad G.V.R., Manhas B.K. 2001. Rhynchocephalians (Di- tissue samples. Franco Andreone, Parfait Bora, Kathrin apsida: Lepidosauria) from the Jurassic Kota Formation of India. Glaw, Fabio Mattioli, Roger Daniel Randrianiaina, and Zool. J. Linn. Soc. 133:309–334. Jasmin E. Randrianirina assisted M.V., F.G., and D.R.V. Felsenstein J. 1981. Evolutionary trees from DNA sequences: a maximum-likelihood approach. J. Mol. Evol. 17:368–376. during fieldwork. Katharina C. Wollenberg helped Fielding A.H., Bell J.F. 1997. A review of methods for the assessment with the geographical analyses. Scott Kelley gave help- of prediction errors in conservation presence/absence models. ful Python coding advice. This research was carried Environ. Conserv. 24:38–49. out in the framework of collaboration accords with the Gao K.Q., Nessov L.A. 1998. Early Cretaceous squamates from the Kyzylkum Desert, Uzbekistan. Neues Jahrb. Geol. Palaontol. Abh. Departement´ de Biologie Animale, Universite´ d’Antana- 207:289–309. narivo, and the Association Nationale des Aires Protegees´ Glaw F., Vences M. 2007. A fieldguide to the amphibians and reptiles ANGAP. We are indebted to the Malagasy authorities of Madagascar. 3rd ed. Koln¨ (Germany): Vences & Glaw. for issuing permits for research and export of specimens Goodman S.M., Benstead J.P. editors. 2003. The natural history of and samples. Tod Reeder, Bradford Hollingsworth, Ben Madagascar. Chicago (IL): The University of Chicago Press. Goodman S.M., Ganzhorn J.U. 2004. Biogeography of lemurs in the Lowe, Adam Leache,´ Marshal Hedin, Jack Sullivan, humid forests of Madagascar: the role of elevational distribution and an anonymous reviewer gave helpful comments on and rivers. J. Biogeogr. 31:47–55. earlier versions of the manuscript. Groth J.G., Barrowclough G.F. 1999. Basal divergences in birds and the phylogenetic utility of the nuclear RAG-1 gene. Mol. Phylogenet. Evol. 12:115–123. Hanley J.A., McNeil B.J. 1982. The meaning and use of the area under a REFERENCES receiver operating characteristic (Roc) curve. Radiology. 143:29–36. Avise J.C., Walker D., Johns G.C. 1998. Speciation durations and Pleis- Hedges S.B., Kumar S. 2004. Precision of molecular time estimates. tocene effects on vertebrate phylogeography. Proc. R. Soc. Lond. Trends Genet. 20:242–247. Ser. B Biol. Sci. 265:1707–1712. Hernandez P.A., Graham C.H., Master L.L., Albert D.L. 2006. The Ayoub N.A., Riechert S.E. 2004. Molecular evidence for Pleistocene effect of sample size and species characteristics on performance glacial cycles driving diversification of a North American desert of different species distribution modeling methods. Ecography. spider, Agelenopsis aperta. Mol. Ecol. 13:3453–3465. 29:773–785. 2009 TOWNSEND ET AL.—BROOKESIA PHYLOGENETICS AND BIOGEOGRAPHY 655

Hewitt G.M. 1996. Some genetic consequences of ice ages, and their (Reptilia, Squamata, Gekkonidae). Misc. Publ. Mus. Zool. Univ. role in divergence and speciation. Biol. J. Linn. Soc. 58:247–276. Michigan 186:1–26. Hewitt G.M. 2004. Genetic consequences of climatic oscillations in Nydam R.L., Cifelli R.L. 2002. Lizards from the Lower Cretaceous the Quaternary. Philos. Trans. R. Soc. Lond. Ser. B Biol. Sci. 359: (Aptian-Albian) Antlers and Cloverly formations. J. Vert. Paleon- 183–195. tol. 22:286–298. Hijmans R.J., Cameron S.E., Parra J.L., Jones P.G., Jarvis A. 2005. Very Nylander J.A.A. 2004. MrModelTest v.2. Distributed by the author. high resolution interpolated climate surfaces for global land areas. Uppsula (Sweden): Evolutionary Biology Centre, Uppsala Int. J. Climatol. 25:1965–1978. University. Ho S.Y.W. 2007. Calibrating molecular estimates of substitution rates Pastorini J., Thalmann U., Martin R.D. 2003. A molecular approach and divergence times in birds. J. Avian Biol. 38:409–414. to comparative phylogeography of extant Malagasy lemurs. Proc. Huelsenbeck J.P., Ronquist F. 2001. MRBAYES: Bayesian inference of Natl. Acad. Sci. USA. 100:5879–5884. phylogenetic trees. Bioinformatics. 17:754–755. Pearson R.G., Raxworthy C.J. 2009. The evolution of local endemism in Hugall A.F., Foster R., Lee M.S.Y. 2007. Calibration choice, rate Madagascar: watershed versus climatic gradient hypotheses evalu- smoothing, and the pattern of tetrapod diversification according to ated by null biogeographic models. Evolution. 63:959–967. the long nuclear gene RAG-1. Syst. Biol. 56:543–563. Phillips S.J., Anderson R.P., Schapire R.E. 2006. Maximum en- Jackman T.R., Bauer A.M., Greenbaum E., Glaw F., Vences M. tropy modeling of species geographic distributions. Ecol. Modell. 2008. Molecular phylogenetic relationships among species of 190:231–259. the Malagasy-Comoran gecko genus Paroedura (Squamata: Rambaut A., Drummond A.J. 2004. Tracer v1.3. Available from Gekkonidae). Mol. Phylogenet. Evol. 46:74–81. http://beast.bio.ed.ac.uk/Tracer. Kadereit J.W., Griebeler E.M., Comes H.P. 2004. Quaternary diversifi- Raselimanana A.P., Rakotomamalala D. 2003. Chamaeleonidae, cation in European alpine plants: pattern and process. Philos. Trans. Chameleons. In: Goodman S.M., Benstead J.P., editors. The natu- R. Soc. Lond. Ser. B Biol. Sci. 359:265–274. ral history of Madagascar. Chicago (IL): The University of Chicago Klicka J., Zink R.M. 1997. The importance of recent ice ages in specia- Press. p. 960–972. tion: a failed paradigm. Science. 277:1666–1669. Raxworthy C.J., Forstner M.R.J., Nussbaum R.A. 2002. Chameleon Kozak K.H., Blaine R.A., Larson A. 2006. Gene lineages and eastern radiation by oceanic dispersal. Nature. 415:784–787. North American palaeodrainage basins: phylogeography and spe- Raxworthy C.J., Nussbaum R.A. 1995. Systematics, speciation and bio- ciation in salamanders of the Eurycea bislineata species complex. geography of the dwarf chameleons (Brookesia, Reptilia, Squamata, Mol. Ecol. 15:191–207. Chamaeleontidae) of northern Madagascar. J. Zool. 235:525–558. Kremen C., Cameron A., Moilanen A., Phillips S.J., Thomas C.D., Been- Raxworthy C.J., Pearson R.G., Zimkus B.M. Reddy S., Deo A.J. Nuss- tje H., Dransfield J., Fisher B.L., Glaw F., Good T.C., Harper G.J., baum R.A., Ingram C.M. 2008. Continental speciation in the tropics: Hijmans R.J., Lees D.C., Louis E., Nussbaum R.A., Raxworthy C.J., contrasting biogeographic patterns of divergence in the Uroplatus Razafimpahanana A., Schatz G.E., Vences M., Vieites D.R., Wright leaf-tailed gecko radiation of Madagascar. J. Zool. 275:423–440. P.C., Zjhra M.L. 2008. Aligning conservation priorities across taxa Rieppel O. 1987. The phylogenetic relationships within the in Madagascar with high-resolution planning tools. Science. 320: Chamaeleonidae, with comments on some aspects of cladistic 222–226. analysis. Zool. J. Linn. Soc. 89:41–62. Kumazawa Y., Nishida M. 1993. Sequence evolution of mitochondrial Rull V. 2008. Speciation timing and neotropical biodiversity: the transfer-RNA genes and deep-branch animal phylogenetics. J. Mol. Tertiary-Quaternary debate in the light of molecular phylogenetic Evol. 37:380–398. evidence. Mol. Ecol. 17:2722–2729. Lees D.C., Kremen C., Andriamampianina L. 1999. A null model for Schmitz A., Brandley M.C., Mausfeld P., Vences M., Glaw F., species richness gradients: bounded range overlap of butterflies Nussbaum R.A., Reeder T.W. 2005. Opening the black box: phy- and other rainforest endemics in Madagascar. Biol. J. Linn. Soc. logenetics and morphological evolution of the Malagasy fossorial 67:529–584. lizards of the subfamily “Scincinae”. Mol. Phylogenet. Evol. 34:118– Lemmon E.M., Lemmon A.R., Cannatella D.C. 2007. Geological and 133. climatic forces driving speciation in the continentally distributed Stamatakis A. 2006. RAxML-VI-HPC: maximum likelihood-based trilling chorus frogs (Pseudacris). Evolution. 61:2086–2103. phylogenetic analyses with thousands of taxa and mixed models. Losos J.B., Glor R.E. 2003. Phylogenetic comparative methods and the Bioinformatics. 22:2688–2690. geography of speciation. Trends Ecol. Evol. 18:220–227. Stamatakis A., Hoover P., Rougemont J. 2008. A rapid bootstrap algo- Louis E.E., Ratsimbazafy J.H., Razakamaharauo V.R., Pierson D.J., Bar- rithm for the RAxML Web servers. Syst. Biol. 57:758–771. ber R.C., Brenneman R.A. 2005. Conservation genetics of black and Storey M., Mahoney J.J., Saunders A.D., Duncan R.A., Kelley S.P., Cof- white ruffed lemurs, Varecia variegata, from southeastern Madagas- fin M.F. 1995. Timing of hot spot-related volcanism and the breakup car. Anim. Conserv. 8:105–111. of Madagascar and India. Science. 267:852–855. Macey J.R., Larson A., Ananjeva N.B., Fang Z.L., Papenfuss T.J. 1997. Sues H.D., Olsen P.E. 1990. Triassic vertebrates of Gondwanan aspect Two novel gene orders and the role of light-strand replication in from the Richmond basin of Virginia. Science. 249:1020–1023. rearrangement of the vertebrate mitochondrial genome. Mol. Biol. Thompson J.D., Higgins D.G., Gibson T.J. 1994. Clustal-W: improving Evol. 14:91–104. the sensitivity of progressive multiple sequence alignment through Maddison D.R., Maddison W.P. 2000. MacClade 4: analysis of phy- sequence weighting, position-specific gap penalties and weight logeny and character evolution. Version 4.03. Tucson (AZ): Sinauer. matrix choice. Nucleic Acids Res. 22:4673–4680. Marshall C.R. 2008. A simple method for bracketing absolute diver- Townsend T.M., Larson A. 2002. Molecular phylogenetics and mi- gence times on molecular phylogenies using multiple fossil calibra- tochondrial genomic evolution in the Chamaeleonidae (Reptilia, tion points. Am. Nat. 171:726–742. Squamata). Mol. Phylogenet. Evol. 23:22–36. Near T.J., Meylan P.A., Shaffer H.B. 2005. Assessing concordance of Townsend T.M., Larson A., Louis E., Macey J.R. 2004. Molecular phylo- fossil calibration points in molecular clock studies: an example us- genetics of Squamata: the position of snakes, amphisbaenians, and ing turtles. Am. Nat. 165:137–146. dibamids, and the root of the squamate tree. Syst. Biol. 53:735–757. Noonan B.P., Chippindale P.T. 2006. Vicariant origin of Malagasy rep- Turgeon J., Stoks R., Thum R.A., Brown J.M., McPeek M.A. 2005. Si- tiles supports Late Cretaceous Antarctic land bridge. Am. Nat. multaneous Quaternary radiations of three damselfly clades across 168:730–741. the Holarctic. Am. Nat. 165:E78–E107. Nougier J., Cantagrel J.M., Karche J.P. 1986. The Comores archipelago Van Bocxlaer I., Roelants K., Biju S., Nagaraju J., Bossuyt F. 2007. Late in the western Indian Ocean: volcanology, geochronology, and geo- Cretaceous vicariance in Gondwanan amphibians. PLoS One. 1:e74. dynamic setting. J. Afr. Earth Sci. 5:135–145. doi:10.1371/journal.pone.0000074. Nussbaum R.A., Raxworthy C.J. 1998. Revision of the genus Ebe- van der Meijden A., Vences M., Hoegg S., Boistel R., Channing A., navia Boettger (Reptilia: Squamata: Gekkonidae). Herpetologica. 54: Meyer A. 2007. Nuclear gene phylogeny of narrow-mouthed toads 18–34. (Family: Microhylidae) and a discussion of competing hypotheses Nussbaum R.A., Raxworthy C.J., Pronk O. 1998. The ghost geckos of concerning their biogeographical origins. Mol. Phylogenet. Evol. Madagascar: a further revision of the Malagasy leaf-toed geckos 44:1017–1030. 656 SYSTEMATIC BIOLOGY VOL. 58

Veith M., Schmidtler J.F., Kosuch J., Baran I., Seitz A. 2003. Palaeo- Winkler D.A., Murry P.A., Jacobs L.L. 1990. Early Cretaceous (Co- climatic changes explain Anatolian mountain frog evolution: a manchean) vertebrates of central Texas. J. Vert. Paleontol. 10: test for alternating vicariance and dispersal events. Mol. Ecol. 12: 95–116. 185–199. Wirta H., Montreuil O. 2008. Evolution of the Canthonini Longitarsi Vences M., Andreone F., Glaw F., Kosuch J., Meyer A., Schaefer H.C., (Scarabaeidae) in Madagascar. Zool. Scr. 37:651–663. Veith M. 2002. Exploring the potential of life-history key innova- Wirta H., Orsini L., Hanski I. 2008. An old adaptive radiation of tion: brook breeding in the radiation of the Malagasy treefrog genus forest dung beetles in Madagascar. Mol. Phylogenet. Evol. 47:1076– Boophis. Mol. Ecol. 11:1453–1463. 1089. Vences M., Glaw F. 2002. Molecular phylogeography of Boophis Wollenberg K.C., Vieites D.R., van der Meijden A., Glaw F., Cannatella tephraeomystax: a test case for east-west vicariance in Malagasy anu- D.C., Vences M. 2008. Patterns of endemism and species richness in rans (Amphibia, Anura, Mantellidae). Spixiana. 25:79–84. Malagasy cophyline frogs support a key role of mountainous areas Vences M., Glaw F. 2003. Phylogeography, systematics and conserva- for speciation. Evolution. 62:1890–1907. tion status of boid snakes from Madagascar (Sanzinia and Acran- Wood H.M., Griswold C.E., Spicer G.S. 2007. Phylogenetic relation- tophis). Salamandra. 39:181–206. ships within an endemic group of Malagasy ‘assassin spiders’ Vences M., Wollenberg K.C., Vieites D.R., Lees D.C. 2009. Madagas- (Araneae, Archaeidae): ancestral character reconstruction, con- car as a model region of species diversification. Trends Ecol. Evol. vergent evolution and biogeography. Mol. Phylogenet. Evol. 45: 24:456–465. 612–619. ´ Vidal-Romanı J.R., Mosquera D.F., Campos M.L. 2002. A 12,000 yr Xia X., Xie Z. 2001. DAMBE: software package for data analysis in BP record from Andringitra Massif, (southern Madagascar): post- molecular biology and evolution. J. Hered. 92:371–373. glacial environmental evolution from geomorphological and sedi- Yang Z.H., Rannala B. 2006. Bayesian estimation of species divergence mentary evidence. Quat. Int. 93–4:45–51. times under a molecular clock using multiple fossil calibrations Vieites D.R., Chiari Y., Vences M., Andreone F., Rabemananjara F., with soft bounds. Mol. Biol. Evol. 23:212–226. Bora P., Nieto-Roman S., Meyer A. 2006. Mitochondrial evidence for distinct phylogeographic units in the endangered Malagasy Yoder A.D., Heckman K.L. 2006. Mouse lemur phylogeography poison frog Mantella bernhardi. Mol. Ecol. 15:1617–1625. revises a model of ecographic constraint in Madagascar. In: Flea- Vieites D.R., Nieto-Roman´ S., Vences M. 2008. Towards understand- gle J., Lehman S.M., editors. Primate biogeography: progress and ing the spatial pattern of amphibian diversity in Madagascar. In: prospects. New York: Kluwer Press. p. 255–268 Andreone F., editor. A conservation strategy for the amphibians Yoder A.D., Nowak M.D. 2006. Has vicariance or dispersal been the of Madagascar. Turin (Italy): Monografie del Museo Regionale di predominant biogeographic force in Madagascar? Only time will Scienze Naturali di Torino. p. 397–410 tell. Annu. Rev. Ecol. Evol. Syst. 37:405–431. Wells N.A. 2003. Some hypotheses on the Mesozoic and Cenozoic pa- Yoder A.D., Rasoloarison R.M., Goodman S.M., Irwin J.A., Atsalis S., leoenvironmental history of Madagascar. In: Goodman S.M., Ben- Ravosa M.J., Ganzhorn J.U. 2000. Remarkable species diversity in stead J.P., editors. The natural history of Madagascar. Chicago (IL): Malagasy mouse lemurs (Primates: Microcebus). Proc. Natl. Acad. The University of Chicago Press. p. 16–34. Sci. USA. 97:11325–11330. Williams P.H. 2002. Using WORLDMAP. Priority areas for biodiver- sity v4.20. Distributed by the author. London: The Natural History Received 19 January 2009; reviews returned 3 April 2009; Museum. Wilme´ L., Goodman S.M., Ganzhorn J.U. 2006. Biogeographic evolu- accepted 21 September 2009 tion of Madagascar’s microendemic biota. Science. 312:1063–1065. Associate Editor: Marshal Hedin