Syst. Biol. 58(2):257–270, 2009 Copyright c Society of Systematic Biologists DOI:10.1093/sysbio/syp025! Advance Access publication on June 9, 2009 Networks, Trees, and Treeshrews: Assessing Support and Identifying Conflict with Multiple Loci and a Problematic Root 1,2,3, 4,5 1,2 TRINA E. ROBERTS ∗,ERIC J. SARGIS , AND LINK E. OLSON 1University of Alaska Museum and 2Institute of Arctic Biology, University of Alaska Fairbanks, Fairbanks, AK 99775, USA; 3National Evolutionary Synthesis Center, Durham, NC 27705, USA; 4Department of Anthropology and 5Division of Vertebrate Zoology, Peabody Museum of Natural History, Yale University, New Haven, CT 06520, USA; ∗Correspondence to be sent to: NESCent, 2024 West Main Street, Suite A200, Durham, NC 27705, USA; E-mail: [email protected]. Abstract.—Multiple unlinked genetic loci often provide a more comprehensive picture of evolutionary history than any single gene can, but analyzing multigene data presents particular challenges. Differing rates and patterns of nucleotide substitution, combined with the limited information available in any data set, can make it difficult to specify a model of evolution. In addition, conflict among loci can be the result of real differences in evolutionary process or of stochastic variance and errors in reconstruction. We used 6 presumably unlinked nuclear loci to investigate relationships within the mammalian family Tupaiidae (Scandentia), containing all but one of the extant tupaiid genera. We used a phylogenetic mixture model to analyze the concatenated data and compared this with results using partitioned models. We found that more complex models were not necessarily preferred under tests using Bayes factors and that model complexity affected both tree length and parameter variance. We also compared the results of single-gene and multigene analyses and used splits networks to analyze the source and degree of conflict among genes. Networks can show specific relationships that are inconsistent with each other; these conflicting and minority relationships, which are implicitly ignored or collapsed by traditional consensus methods, can be useful in identifying the underlying causes of topological uncertainty. In our data, conflict is concentrated around particular relationships, not widespread throughout the tree. This pattern is further clarified by considering conflict surrounding the root separately from conflict within the ingroup. Uncertainty in rooting may be because of the apparent evolutionary distance separating these genera and our outgroup, the tupaiid genus Dendrogale. Unlike a previous mitochondrial study, these nuclear data strongly suggest that the genus Tupaia is not monophyletic with respect to the monotypic Urogale,evenwhenuncertaintyaboutrootingistakenintoaccount.Thesedataconcurwith mitochondrial DNA on other relationships, including the close affinity of Tupaia tana with the enigmatic Tupaia splendidula and of Tupaia belangeri with Tupaia glis. We also discuss the taxonomic and biogeographic implications of these results. [Mixture model; partitioned model; Southeast Asia; splits network; treeshrew; Tupaia; Urogale.] As multilocus genetic data proliferate in systematic We use this combination of methods to help us deter- studies, the question of how best to analyze them be- mine where individual methods may be suggesting a comes more important than ever. Early analyses almost solution with inflated confidence and to identify poten- always involved simply concatenating data under a sin- tial sources of conflict in our data. gle model of evolution or optimality criterion. More The mammalian order Scandentia (treeshrews) is no- recently, it has become easy to partition data—by locus, table both for its close affinity to primates and for the codon position, or any other criterion that appeals to lack of recent attention that has been paid to its evolu- the investigator’s knowledge of the data—and assign tionary history. Treeshrews have long been considered adifferentmodeltoeachpartition.However,itcanbe to be among the closest living relatives of primates and difficult to know what the best set of partitions is, and have been well represented in recent large-scale studies increasing the number of partitions means that each of mammalian interordinal relationships (e.g., Murphy contains fewer data from which to estimate para- et al. 2001; Scally et al. 2001; Reyes et al. 2004; Janecka meters. An alternative is mixture modeling, in which data et al. 2007). As a result, some species have become im- are not partitioned but in which multiple models are portant model organisms in biomedical research (e.g., averaged across all sites (Gelman et al. 2004; Pagel and Czeh et al. 2001; Bahr et al. 2003; von Weizsacker et Meade 2004). Other nonconcatenation methods treat al. 2004). However, knowledge of the evolutionary his- multiple gene trees as having possibly separate evolu- tory of this group is woefully incomplete, and the last tionary histories but use the information in each gene thorough revision of treeshrew taxonomy was pub- tree to infer a species tree through consensus methods lished nearly a century ago (Lyon 1913). In many ways, (e.g., Holland et al. 2006; Huson and Bryant 2006) or treeshrews epitomize the legacy left by Victorian era ty- using the properties of the coalescent (e.g., Degnan and pological taxonomy—of the more than 120 species and Salter 2005; Edwards et al. 2007). Because these meth- subspecies described between 1820 and 1920, only 20 ods make different uses of the same data, require dif- species are currently recognized, although taxonomists ferent assumptions, and have different strengths and readily acknowledge this as an underestimate (Corbet weaknesses, using them together can allow complemen- and Hill 1992; Helgen 2005). The order is generally tary interpretations of a given data set. Within a general Southeast Asian (Fig. 1), with a distribution almost framework of Bayesian phylogenetics, we use 3 meth- perfectly delimited by Wallace’s line (Wallace 1860, ods to examine phylogenetic hypotheses in a poorly 1876) and extending west into India (Helgen 2005). studied mammalian group: mixture modeling, parti- This region encompasses several major conservation tioned analyses, and splits-network consensus analysis. hot spots (Myers et al. 2000; Sodhi et al. 2004) with high 257 258 SYSTEMATIC BIOLOGY VOL. 58 biodiversity and endemism, and understanding evo- of Tupaia, and Dendrogale murina,whichweusedasthe lutionary and biogeographic processes in the area is outgroup for all analyses (Olson et al. 2005); we lack critical. The broad distribution of treeshrews makes Ptilocercus lowii, Dendrogale melanura, Anathana elliotti, them excellent candidates for regional-scale research in and 6 species of Tupaia.WeextractedDNAusingthe Southeast Asia, but questions about the biogeography, PureGene protocol for animal tissue (Gentra Systems, ecology, ecogeography, behavior, and morphological Valencia, CA). We sequenced DNA from 6 nuclear evolution of this group cannot be answered in a phylo- genes: brain-derived neurotrophic factor (BDNF, 566 genetic context without some knowledge of systematic bp; coding), the 3" untranslated region (UTR) of cAMP relationships within the order. responsive element modulator (CREM,469bp;noncod- One contentious question within treeshrews is the ing), the 3" UTR of phospholipase C beta 4 (PLCB4, 334 status of Urogale,amonotypicgenusendemictothe bp; noncoding), tyrosinase (Tyr,440bp;coding),recom- Mindanao Faunal Region of the Philippines (Heaney bination activating gene 2 (RAG2,450bp;coding),and et al. 1998). Urogale everetti was originally described as a Exon 28 of the von Willebrand factor (vWF,574bp;cod- species of the more widely distributed and speciose ing), for a total of 2833 bp. Although we have no way genus Tupaia (Thomas 1892) but was later elevated to be sure where in any given treeshrew genome these (Mearns 1905) on the basis of its unique morphology; genes fall, they do not appear to be adjacent on the cur- modern evidence for its generic status is weak at best. rent shotgun assembly of the Tupaia belangeri genome DNA–DNA hybridization data (Han et al. 2000) sug- (National Center for Biotechnology Information locus gest that Urogale is nested within Tupaia, as do some AAPY01000000), and we therefore have good reason to morphological and morphometric analyses (Steele 1973; believe that none of them are closely linked. Olson et al. 2004; Sargis 2004). Other analyses have Amplification followed standard polymerase chain supported the distinction between Tupaia and Urogale reaction (PCR) methods, in general 35 cycles of denat- (Butler 1980; Luckett 1980), although results have been uration, annealing, and extension, with a magnesium shown to vary depending on analytical methods and concentration of 1.5–4.0 mM, an annealing temperature assumptions in character coding (Olson et al. 2004). Re- of 50–65◦C, and the primers shown in Table 1. PCR cently, Olson et al. (2005) used mitochondrial DNA to products were purified with either GeneClean (Bio101, infer phylogenetic relationships within Scandentia but Vista, CA), following the manufacturer’s instructions, found little resolution for most intergeneric relation- or Exo/SAP (USB Corp., Cleveland, OH) using one- ships in the family Tupaiidae, which includes Tupaia, quarter of the recommended volume of each reagent. Dendrogale, Anathana, and Urogale. Urogale has the east- We sequenced PCR products using 0.5–1.0 µLofABI ernmost distribution of any treeshrew, and