A Genus-level Supertree of the Dinosauria

Downloaded from on April 26, 2011 A genus-level supertree of the Dinosauria D. Pisani, A. M. Yates, M. C. Langer and M. J. Benton Proc. R. Soc. Lond. B 2002 269, 915-921 doi: 10.1098/rspb.2001.1942 Supplementary data Data Supplement l Article cited in: Receive free email alerts when new articles cite thi
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  doi: 10.1098/rspb.2001.1942, 915-921 269 2002 Proc. R. Soc. Lond. B   D. Pisani, A. M. Yates, M. C. Langer and M. J. Benton  A genus-level supertree of the Dinosauria  Supplementary data l  Data Supplement References   Article cited in: Email alerting service   here right-hand corner of the article or clickReceive free email alerts when new articlescite this article - sign up in the box at the top go to:  Proc. R. Soc. Lond. B  To subscribe toThis journal is © 2002 The Royal Society  on April 26, 2011rspb.royalsocietypublishing.orgDownloaded from   Received  24 September 2001  Accepted  3 December 2001 Published online 9 April 2002  A genus-level supertree of the Dinosauria Davide Pisani 1,2 * , Adam M. Yates 1 , Max C. Langer 1 and Michael J. Benton 1 1 Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK  2 Department of Zoology, The Natural History Museum, Cromwell Road, London SW7 5BD, UK  One of the ultimate aims of systematics is the reconstruction of the tree of life. This is a huge undertakingthat is inhibited by the existence of a computational limit to the inclusiveness of phylogenetic analyses.Supertree methods have been developed to overcome, or at least to go around this problem by combiningsmaller, partially overlapping cladograms. Here, we present a very inclusive generic-level supertree of Dinosauria (covering a total of 277 genera), which is remarkably well resolved and provides some clarityin many contentious areas of dinosaur systematics. Keywords: supertrees; phylogeny; dinosaurs; fossil; Mesozoic 1. INTRODUCTION In the last few decades, advances in theoretical system-atics, as well as in computer sciences, have provided bio-logists and palaeontologists with tools to investigate thephylogenetic relationships among organisms, yet phylo-genetic inference is computationally expensive and analy-ses of large datasets hardly feasible (Graham & Foulds1982). This problem is exacerbated in palaeontology,where missing data can lead to an exponential increase inthe number of most-parsimonious solutions permitted bya given dataset. However, ‘partial phylogenies’ (i.e. phy-logenies including only a subset of the taxa known tobelong to a certain group) are accumulating at an increas-ing rate, and the Dinosauria represent a typical, if notextreme, example of this trend.Supertree methods, formally introduced by Gordon(1986), can be implemented to combine such ‘partial phy-logenies’ and obtain more inclusive estimates without theneed to pool the srcinal datasets. Accordingly, whenevera collection of ‘partial phylogenies’ exists, supertrees canrepresent an alternative to a direct analysis of the primarydata (i.e. the pooled dataset).Several supertree methods have been developed (e.g.Sanderson et al. 1998; Wilkinson et al. 2001 for review),but only those using matrix representation with parsimony(MRP methods) have software implementation (Swofford1998; Thorley & Page 2000) and these are currently theonly viable choice in supertree reconstruction. Neverthe-less, their characterization is still incomplete. In particular,when the source trees are incompatible (i.e. they disagreeabout the relationships of the taxa that overlap), theirbehaviour is not completely understood. In these cases,MRP methods can be inaccurate. It is debatable whetherMRP estimates are accurate enough to be useful tree-reconstruction methods (Purvis 1995 a ; Wilkinson et al. 2001). Yet, there is evidence (Bininda-Emonds & San-derson 2001) that, if the set of source trees is largeenough, the MRP supertree(s) are accurate represen- * Author for correspondence ( Proc. R. Soc. Lond. B (2002) 269 , 915–921 915  2002 The Royal SocietyDOI 10.1098/rspb.2001.1942 tations of the information conveyed by the trees in theinput set.The srcinal MRP method of Baum (1992) and Ragan(1992), referred to as component coding-MRP (CC-MRP) in this paper, was used to combine 126 ‘partial phy-logenies’ of dinosaurs (covering a total of 277 genera) withthe aim of obtaining a single, genus-level supertree. CC-MRP makes use of the fact that a tree can be representedin a matrix by means of a set of binary ‘characters’. Fol-lowing Baum & Ragan (1993), we refer to these ‘charac-ters’ as matrix elements. Matrix representations of different source trees can be readily combined with miss-ing entries in the matrix elements corresponding to taxathat are not present in the matching source tree. Parsi-mony analysis of the combined matrix representationsyields one or more most parsimonious trees (MPTs), theCC-MRP supertree(s), which comprise all the includedtaxa (see Baum 1992; Ragan 1992 for more details). 2. MATERIAL AND METHODS Potential source trees were identified from online searches.Two electronic literature databases were searched for publicationswith relevant source trees, BFV (Bibliography of Fossil Verte-brates) online ( F form.html)and Web of Science ( BFV online doesnot allow the enquirer to confine his search to a limited numberof years and it was thus screened over the whole time-span thatit covers (1509–1993), while Web of Science was searched forthe years 1993–2001. All publications that were likely to includea cladogram were examined. Because the aim of this analysiswas to obtain a consensus representation of cladistic views ondinosaur systematics, the earliest of which date back to thebeginning of the 1980s (Benton 1990), publications presentedbefore the year 1980 have been excluded a priori  . Publicationsafter this date were retained when they presented cladogram(s)resulting from an srcinal study, or from the modification of apre-existing dataset. The list of all retained publications is avail-able (see electronic appendix). For each selected cladogram, aNexus (Swofford 1998) tree file was produced using M ac C lade (Maddison & Maddison 1997). Some terminal genera wereomitted a priori  (see electronic appendix, available on The RoyalSociety’s Publications Web site, for the list of the omitted taxa).  on April 26, 2011rspb.royalsocietypublishing.orgDownloaded from   916 D. Pisani and others A supertree of Dinosauria These were either nomina dubia , unconstrained taxa, or both.The former are taxa based on material inadequate for diagnosisand for proper phylogenetic analysis. The latter may be validtaxa but they appear only in one source tree and as membersof a polytomy. Accordingly, their inclusion adds nothing to theanalysis but an increase in the number of possible MPTs. Mostsrcinal studies included a variable proportion of supragenericterminal taxa. For analytical proposes, their monophyly wasassumed and a standard taxon substitution (Wilkinson et al. 2001) was performed. Each suprageneric taxon was substitutedwith a star tree (i.e. an unresolved polytomy) including either:(i) the taxa embodied in each of them when explicitly given inthe srcinal study; or (ii) only its non-controversial members inthe case that no such de fi nition was given (see electronic appen-dix for the minimal non-controversial de fi nition of each supra-generic taxon). When multiple MPTs were presented, strictreduced consensus (RC; Wilkinson 1994) was implementedusing R  ad C on (Thorley & Page 2000) and the tree in the RCpro fi le with the highest cladistic information content (Thorley et al. 1998) was retained for the CC-MRP analysis.Five CC-MRP matrices (see electronic appendix), one scoringall of the 277 included taxa (Matrix I), one for Sauropodomor-pha (Matrix II), one for Ornithischia (Matrix III) and two forTheropoda (Matrices IV and V) were produced using R  ad C on .These have been analysed separately, resulting in one compre-hensive (Matrix I) and four compartmentalized (Matrices II  –  V)supertrees. It is important to note that Matrices II  –  V are matr-ices in their own right, not subsamples of Matrix I. They havenot been obtained by pruning the excluded taxa, e.g. all thero-pods and all ornithischians (in the case of Matrix II) from MatrixI, but by pruning the source trees and then recoding them in anew CC-MRP matrix. This approach, implemented here for the fi rst time, to our knowledge, is important because each matrixelement is representative of a node in a tree. Thus, pruning ataxon from an MRP matrix will create a matrix that is not rep-resentative of the real topology of the pruned tree ( fi gure 1).Because objective functions for implementing a correct,unequal character-weighting scheme in MRP analyses are stillunknown (Bininda-Emonds et al. 1999; Bininda-Emonds &Sanderson 2001), equally weighted parsimony was used to ana-lyse the fi ve CC-MRP matrices.P aup ∗ (Swofford 1998) was used to analyse the data matriceswith parsimony and, in order to avoid the problem of being ‘ bogged down ’ in sub-optimal tree islands, the following tree-search protocol was applied: (1) 100 heuristic searches werecarried out using tree bisection  –  reconnection (TBR) branchswapping with the ‘ multrees ’ option not in effect; random step-wise addition was used to obtain the starting trees for these repli-cates; (2) the MPTs found were retained and swapped usingTBR branch swapping with the ‘ multrees ’ option in effect.A potentially serious problem of MRP supertree methods istheir ability to resolve con fl ict among the source trees by thegeneration of novel clades ( sensu Bininda-Emonds & Bryant1998). Bininda-Emonds et al  . (1999) and Bininda-Emonds &Sanderson (2001) have likened the appearance of such clades inMRP to their appearance in Total Evidence analyses. However,as shown by Pisani & Wilkinson (2002), this cannot be the caseand such novel clades should always be collapsed. No novelclades ( sensu Bininda-Emonds & Bryant 1998) are present in thedinosaur supertree.A poorly understood problem in MRP supertree building ishow to evaluate the support for the resulting clades. For thosewho feel that an MRP supertree is naked without some measure Proc. R. Soc. Lond. B (2002) of support, following Bininda-Emonds et al. (1999) and Liu et al  . (2001), decay indices (Bremer 1988) are provided. Decayvalues have been calculated for Matrices II  –  IV, but because of the size of the resulting supertree, they have not been calculatedfor Matrix I. However, it is our opinion that this measure cannotbe considered really representative of the support for the cladesrecovered by the supertree. Accordingly, decay values should beinterpreted with care. Consider, for example, that novel clades( sensu Bininda-Emonds & Bryant 1998), despite being unsup-ported (Pisani & Wilkinson 2002), have a positive decay index.Thus, new support measures for MRP supertrees are neededand are under development (D. Pisani, unpublished data), butthey will not be considered any further in this paper. 3. RESULTS A total of 34 900 MPTs was found from the analysis of Matrix I before the search had to be aborted because of memory limitations. We suggest that this sample of MPTs, although less than the set of all possible supertrees,is representative of the whole. This is because the topologyof their strict consensus agrees perfectly with those of thestrict consensus trees summarizing the results obtainedfrom the compartmentalized analyses (Matrices II, III andV), which have been run to completion. It is thus possibleto argue that increasing the number of retained MPTsfrom the analysis of Matrix I will result in a more compre-hensive collection of alternative solutions for the polytom-ies seen in fi gure 2 and not in the collapse of its resolvedareas. The strict consensus of the 34 900 MPTs from theanalysis of Matrix I ( fi gure 2) therefore represents themost inclusive estimate that has ever, to our knowledge,been presented of the phylogeny of the Dinosauria.Contentious issues in dinosaur systematics are resolvedin the present supertree. Sauropodomorpha is composedof  Saturnalia and monophyletic Prosauropoda and Sauro-poda (Sereno 1999; Benton et al. 2000; contra Gauthier1986). Within Prosauropoda, a monophyletic Melanoro-sauridae ( Riojasaurus , Camelotia and Melanorosaurus )emerges as the sister group of Plateosauria (  Masso-spondylus , Yunnanosaurus and Plateosauridae). WithinSauropoda, Barapasaurus is the sister group to all otherEusauropoda (Upchurch 1998; contra Wilson & Sereno1998) and Omeisaurus is more closely related to Neosauro-poda than it is to Shunosaurus (Wilson & Sereno 1998; contra Upchurch 1998). The early dinosaurs Eoraptor  andthe Herrerasauridae are basal theropods (Sereno et al  .1993; contra Langer et al. 1999). Ceratosauria is a mono-phyletic group containing Coelophysoidea and Neocerato-sauria (Sereno 1999; Holtz 2000; contra Forster 1999),while torvosaurids are more closely related to derivedtetanurans (Avetheropoda) than to spinosaurids (Holtz2000; contra Sereno et al  . 1994). Therizinosauroidea is thesister group of Oviraptorosauria (Russell & Dong 1993;Holtz 2000; contra Sereno 1999), but Caudipteryx is abasal member of Paraves: it is excluded from both Ovira-ptorosauria ( contra Sereno 1999) and Avialae ( contra Ji et al  . 1998), while Metornithes is monophyletic and con-tains Ornithothoraces, Avimimus (Chatterjee 1991; contra Holtz 1994) and Alvarezsauridae (Chiappe et al  . 1996; contra Sereno 1999).Within Ornithischia, Minmi  and Gargoyleosaurus arebasal Ankylosauridae (Sereno 1999; contra Kirkland  on April 26, 2011rspb.royalsocietypublishing.orgDownloaded from    A supertree of Dinosauria D. Pisani and others 917 ABCEFnode 1node 2node 3node 4nodes: N1 N2 N3 N4A 0 0 0 0B 0 0 0 1C 0 0 1 1D 0 1 1 1E 1 1 1 1F 1 1 1 1nodes: N1 N2 N3 N4A 0 0 0 0B 0 0 0 1C 0 0 1 1E 1 1 1 1F 1 1 1 1nodes: N1  N3  N4A 0 0 0B 0 0 1C 0 1 1E 1 1 1F 1 1 1(tree A)Matrix obtained when pruningtaxon D from the CC-MRPrepresentation of tree A (i.e.using the P AUP option ‘Delete-Restore taxa’). Note that thenew matrix is not a correct CC-MRP representation of tree B.  The deletion of taxon D implies thesuppression of node 2.Accordingly, the correspondingmatrix element (N2) is not presentin the correct CC-MRP representationof tree B.ABCFnode 1node 3node 4(tree B)CC-MRP representation of tree A pruning taxon DCC-MRP representation of tree BDE Figure 1. A tree on six leaves (tree A) and its corresponding pruned version (tree B), from which taxon D has been deleted.Pruning taxon D from the CC-MRP representation of tree A results in a matrix that is not the correct CC-MRP matrixrepresenting tree B. 1998). A monophyletic Iguanodontidae (Norman 1998; contra Sereno 1999) containing Iguanodon , Altirhinus and Ouranosaurus is recovered, while the Hadrosauridaeappears more closely related to Probactrosaurus than toIguanodontidae. Leptoceratops and Udanoceratops areexcluded from Coronosauria (Protoceratopsidae plusCeratopsoidea) and a monophyletic Protoceratopsidaecontaining Bagaceratops , Breviceratops , Graciliceratops and Protoceratops is recovered (Sereno 2000; contra Chinnery &Weishampel 1998).The present supertree contains three important poly-tomies. One lies within Sauropoda and the other twowithin Theropoda: one at the base of Coelurosauria andthe other within Coelurosaria, at the base of the Eumanir-aptora.The polytomy within Sauropoda involves the taxa morederived than Shunosaurus and is due to both the existenceof con fl icting hypotheses concerning the relationships of  Proc. R. Soc. Lond. B (2002) three long-necked Chinese taxa, Omeisaurus , Mamenchi-saurus and Euhelopus , and to the poorly constrained pos-ition of  Cetiosaurus . The fi rst problem re fl ects differencesin the results obtained by different authors on the relation-ships of those taxa. Among the source trees, two classesof cladogram exist. Upchurch (1995, 1998) includes allthree Chinese taxa in a clade nested outside Neosauro-poda, while Wilson & Sereno (1998; also Sereno 1999)do not include Mamenchisaurus and consider Omeisaurus and Euhelopus to be distantly related. In the supertree, Omeisaurus and Euhelopus are nested according to thesecond of these two hypotheses, but Mamenchisaurus isfree to cluster with either of the two . Implementing RC,two ‘ rogue ’ taxa ( Cetiosaurus and Mamenchisaurus ) areidenti fi ed. With their exclusion ( fi gure 3 a ), the followingtaxonomic statements are true in all the supertrees: (i) Haplocanthosaurus falls within Neosauropoda (de fi ned bythe node connecting Diplodocus and Saltasaurus ) and is  on April 26, 2011rspb.royalsocietypublishing.orgDownloaded from 
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks