Estimating evolutionary rates using discrete morphological characters: a case

study with birds

Luke Barrett Harrison

Department of Biology

McGill University, Montreal

April 2013

A thesis submitted to McGill University in partial fulfillment of the requirements

of the degree of Doctor of Philosophy

© Luke Harrison 2013 DEDICATION

I dedicate this thesis to my wife, Soo Bin Chun. I would not have been able to do it without you nor would I have wanted to.

semper fidelis & love always TABLE OF CONTENTS

ABSTRACT / RÉSUMÉ 1

ACKNOWLEDGEMENTS 5

PREFACE 7

GENERAL INTRODUCTION 10

CHAPTER 1: Estimating Evolutionary Rates of Discrete Morphological Characters Introduction 14 Rates of Phenotypic Evolution 15 Traits, Characters, and Discrete Morphological Characters 16 Heterogeneity in Rates Between Discrete Morphological Characters in Phylogenetic Analysis 18 Absolute Rates of Evolution of Discrete Morphological Characters 20 Westoll (1949) 22 Forey (1988) 22 Cloutier (1991) 23 Wagner (1997) 23 Bromham et al. (2002) 24 Ruta et al. (2006) 24 Brusatte et al. (2008) 25 Roelants et al. (2011) 25 Lloyd et al. (2012) 27 Summary of Previous Methods and Future Directions 28 Appropriate Null Models 28 Likelihood-based Methods for Estimating Morphological Evolutionary Rates: Potential Advantages 29 An Ideal Model-based Framework for Estimating Absolute Rates of Evolution of Discrete Morphological Characters 30 Conclusions 31 CONNECTING TEXT 32

CHAPTER 2: Among-Character Rate Variation in Phylogenetic Analysis of Discrete Morphological Characters: Prevalence and Bayesian Model Selection Introduction 33 Materials and Methods Data Sets 35 Bayesian Analysis 36 Topology Comparisons 40 Parsimony-based Estimation of Rate Distributions 40 Optimal Number of Discrete Rate Categories 41 Focal Phylogenetic Analysis 42 Results Equal Rates Models and Unequal Rates Models 43 Gamma and Lognormal Rate Distributions 43 Parsimony Analysis 45 Optimal Number of Discrete Rate Categories 45 Effect on Phylogenetic Topology and Branch Lengths: Sidlauskas and Vari (2008) 46 Discussion Rate Heterogeneity in Data Sets of Discrete Morphological Characters 46 Gamma and Lognormal Rate Distributions 47 Optimal Number of Discrete Rate Categories 49 Effects on Estimations of Phylogenetic Topology 51 Alternative Approaches to Model ACRV 53 Conclusions 54 Figures 56 Tables 70 Appendices 73

CONNECTING TEXT 86 CHAPTER 3: Embracing Uncertainty: A Distribution of Evolutionary Timescales for Modern Bird Lineages Introduction 87 Materials and Methods 90 Sequence Data 91 Phylogenetic Analysis 92 Divergence Time Analysis 93 Comparisons with the Jetz et al. (2012) Distribution of Chronograms 94 Results and Discussion Phylogenetic Analysis 95 Divergence Time Analysis 96 Comparisons to Other Divergence Time Studies – Higher-Level Divergences 99 Comparison to Jetz et al. (2012) – Lower Level Divergences 100 Implications for Avian Evolution 101 Conclusions 102 Figures 104 Tables 108 Supplementary Methodology 122 Supplementary Figures 150 Appendices 156

CONNECTING TEXT 220

CHAPTER 4: Estimating Absolute Rates of Evolution using Discrete Morphological Characters across an Uncertain Phylogeny: Rates of Anatomical Evolution in Modern Birds Introduction 221 Materials and Methods 225 Morphological Data 225 Chronogram Distributions 226 Reconstruction of Morphological Evolution 226 Estimating Rates of Evolution: MCC Chronograms 227 Estimating Rates of Evolution: Chronogram Distributions 228 Rates of Evolution of Monophyletic Clades 229 Visualizing and Testing Absolute Rates of Evolution Through Time 231 Aggregate Morphological and Molecular Rates Through Time 232 Results Among-character Rate Heterogeneity and Partition Testing 233 Rates of Evolutionary Change on MCC Trees 233 Rates of Evolution of Focal Clades 235 Absolute Rates of Evolutionary Change Through Time 236 Correlation Test for the Early-Burst Hypothesis 238 Discussion Modular Evolution in the Livezey and Zusi (2007) Data Set 239 Estimating Rates of Evolution of Discrete Morphological Characters 241 Implications for the Evolution of Modern Birds 242 Limitations of the Present Analysis 244 Broad-scale Patterns of Molecular and Morphological Evolutionary Rates 246 Future Research 246 Conclusions 248 Figures 250 Tables 262 Supplementary Methodology 268 Supplementary Figures 276 Appendix 298

SUMMARY AND CONCLUSIONS 304

BIBLIOGRAPHY 309 ABSTRACT

The rate of evolution is a fundamental unifying concept in evolutionary biology and sets the stage for the investigation of genotypic, phenotypic and taxonomic biodiversity. This thesis specifically examined the rate of phenotypic evolution using discrete morphological characters, which are relatively understudied for this purpose compared to continuously-valued characters and traits. I first focused on heterogeneity in rates among characters in phylogenetic analysis. I used Bayesian model selection tools and 77 matrices of discrete morphological characters to show that a) models incorporating rate-heterogeneity among characters in phylogenetic analysis were preferred over equal-rates models in 80–88% of matrices, suggesting rate heterogeneity is a common property of these data sets, and b) although most data sets were equivocal, there was some weak support for a recently formulated hypothesis that the lognormal distribution is more appropriate to model such variation relative to the commonly used gamma distribution. I then focused on estimating absolute rates of evolution of discrete morphological characters in a phylogenetic context. I extended previous methods to better incorporate phylogenetic and divergence time uncertainty using distributions of dated phylogenies derived from independent data. I used modern birds as a case study and performed a large Bayesian divergence time study of a comprehensive sample of 310 modern bird genera to provide a posterior sample of 10 000 dated trees to estimate absolute rates of evolution. This analysis, based on 23 fossil calibrations and a multigene molecular supermatrix of existing sequences, although qualified by uncertainty in estimated relationships and divergence times, estimated that the basal radiation of Neoaves occurred within a relatively short interval in the Late Cretaceous. Many lineages were estimated to cross the Cretaceous–Paleogene (K–Pg) boundary while within order diversification of crown groups was nearly exclusively in the Cenozoic. Finally, I employed this tree distribution along with another recently published tree distribution to estimate absolute rates of phenotypic evolution using both

1 maximum parsimony and likelihood-based methods using an existing comprehensive data set of discrete avian anatomical characters. Incorporating phylogenetic and divergence time uncertainty, estimated rates of evolution were found to be highly variable and had a complex multimodal distribution through time when visualized across 10 000 dated trees. Combined with an analysis of rates of evolution across clades, maximum clade credibility trees, and a correlation test of rates against time, the results were complex, but in aggregate, were consistent with the hypothesis of an early-burst of higher rates of phenotypic evolution in modern birds.

2 RÉSUMÉ

Le taux d'évolution est un concept fondamental unificateur en biologie évolutive, et ouvre la voie à l'étude de la biodiversité génotypique, phénotypique et taxonomique. La présente thèse a examiné de manière spécifique le taux d'évolution phénotypique à l'aide de charactères morphologiques discrètes, qui sont relativement peu étudiés dans cette optique comparativement aux traits et charactèrs à valeur continue. Premièrement, je me suis penché sur l'hétérogénéité des taux dans les caractères d'analyse phylogénétique. Les outils de sélection du modèle Bayesien ainsi que 77 matrices de charactères morphologiques discrets ont été utilisé afin de démontrer que a) les modèles incorporant l'hétérogenéité des taux dans les charactères d'analyse phylogénétique étaient préférées des modèles à taux égaux dans 80 à 88% des matrices, ce qui suggère que l'hétérogénéité est une charactéristique commune dans les ensembles de données, et b) bien que la plupart des ensembles de données étaient équivoques, il y avait un faible appui pour l'hypothèse formulée récemment que la distribution log-normale est plus appropriée pour modéliser les variations relatives que la distribution gamma couramment utilisée. Ensuite, je me suis concentré sur l'esmination des taux absolus d'évolution des charactères morphologiques discrets dans un contexte phylogénétique. J'ai étendu des méthodes existantes afin de mieux incorporer les incertitudes phylogénétique et de divergence temporelle en utilisant des distributions de phylogénies datées extraits de données indépendantes. J'ai utilisé les oiseaux modernes comme étude de cas et j'ai effectué une grande étude Bayesien de divergence temporelle d'un échantillon exhaustif de 310 genera d'oiseaux modernes pour y extraire un échantillon postérieur de 10 000 arbres datées, dans le but d'arriver à une estimation absolue des taux d'évolution. Cette analyse, qui est basée sur vingt-trois étalonnages de fossils et une supermatrice multigène moléculaire de séquences existantes, bien que qualifié par une incertitude dans les relations estimées et divergences temporelles, estime que le rayonnement de base a eu lieu dans un laps de temps relativement court dans la

3 fin du Crétacé. De nombreuses lignées ont été estimés à traverser la frontière Crétacé-Paléogène (K–Pg) tandis que la diversification des groupes couronnes (« crown groups ») à l'intérieur du groupe était presque exclusivement dans le Cénozoïque. Finalement, j'ai utilisé cette distribution avec une autre distribution publiée récemment afin d'estimer les taux absolus de l'évolution phénotypique en utilisant la parcimonie maximale et les méthodes basées sur les probabilités en utilisant un ensemble de données compréhensif de charactères anotomiques discrèts d'oiseaux. Intégrer les incertitudes phylogénétiques et de divergence temporelle, les taux d'évolution estimés se sont révélés être très variables et ont fait preuve d'une distribution multimodale complexe lorsque visualisés à travers 10 000 arbres datés. Combiné avec une analyse des taux d'évolution à travers les clades, des arbres de clade à crédibilité maximale, et un test de corrélation entre les taux en fonction du temps, les résultats se sont avérés complexes, mais dans l'ensemble, étaient compatibles avec l'hypothèse d'une rupture précoce de la hausse des taux d'évolution phénotypique chez les oiseaux modernes.

4 ACKNOWLEDGEMENTS

DISCUSSION AND SUPPORT

Firstly, I would like to thank my supervisor Dr Hans Larsson and the members of my supervisory committee, Dr Ehab Ebouheif and Dr Thomas Bureau, for guidance and helpful discussions. I am greatly indebted to Dr Larsson's support, both financial and moral, and for allowing me the freedom to pursue my research interests. I would also like to thank the members of the Larsson Lab for insightful discussions about this work. I would like to thank Dr Matthew Vavrek, Dr Alex Dececchi, Trina Du, Dr Victoria Dickenson and W.C. Chun for critical readings of each chapter of this thesis. Finally, I would like to thank my wife Soo Bin Chun for much moral support and encouragement, aiding in the French translation of the abstract and for proofreading each chapter of this thesis.

FUNDING

Funding for this project was provided by an NSERC CGS-D postgraduate fellowship, a McGill Provost's Graduate Fellowship and a Department of Biology Top-Up Award. I would also like to sincerely thank Rosalie McKenna Hamer and Dr Arthur McDonald for honouring Dr André Hamer's memory with the NSERC André Hamer Prize, of which I was a truly humbled recipient.

DATA AND SOFTWARE

I am greatly indebted to the researchers whose work to sequence thousands of gene sequences and whose choice to make the data available in public databases made this project possible. I am further indebted to the truly exhaustive work to characterize avian morphology by Dr Bradley Livezey and Dr Richard Zusi; theirs is truly a work without precedent in both quantity and quality. I was

5 saddened to learn of Dr Livezey's untimely passing in 2011; the scientific community has lost an important figure. Although Dr Livezey may not have approved of the methods I used to analyze his and Dr Zusi's data set or the sets of phylogenies used, I hope he would not have minded too much. I would also like to thank the authors and contributors to the BEAST, RAxML and MrBayes software packages. These advanced software tools created the infrastructure and foundation for this thesis and it would not have been possible without them. Without question, the phylogenetic analyses presented in this thesis rest on the shoulders of these giants.

COMPUTATIONAL RESOURCES

The analyses presented in this thesis would not have been possible without the computational resources provided by CalculQuébec/CLUMEQ. CLUMEQ is an indispensable resource for researchers in Québec, particularly those where high performance computing is not a resource that is commonly available. I hope that the value-added and force multiplier that CalculQuébec/CLUMEQ provides is recognized with continued funding. I would also like to thank the support staff at CLUMEQ/Guillimin and CLUMEQ/Colosse for their help in troubleshooting software and analyses: their responses were always helpful and always timely.

6 PREFACE

CONTRIBUTIONS OF AUTHORS

The first chapter of this thesis “Estimating Evolutionary Rates of Discrete Morphological Characters” was written by me for the purpose of inclusion in this thesis. My supervisor Dr Hans Larsson provided editorial input and contributed some elements of the discussion on the distinction between traits and characters. The second chapter of this thesis, “Among-character Rate Variation in the Phylogenetic Analysis of Discrete Morphological Characters: Prevalence and Bayesian Model Selection” was prepared as a co-authored manuscript with my supervisor, Dr Larsson, as second author for submission to Systematic Biology. I implemented all of the modifications to the MrBayes software package, performed all of the analyses and wrote the manuscript in its entirety. Dr Larsson supervised, provided editorial input and supported the project. The third chapter of this thesis, “Embracing Uncertainty: A Distribution of Evolutionary Timescales for Modern Bird Lineages” was prepared as a co-authored manuscript with Dr Larsson as the second author for submission to Proceedings of the Royal Society of London Series B. I collected all of the molecular sequence data and performed all of the analysis under Dr Larsson's supervision. Dr Larsson contributed editorial changes and supported the project. The manuscript and all figures were my own work. The fourth chapter, “Estimating Absolute Rates of Evolution using Discrete Morphological Characters across an Uncertain Phylogeny: Rates of Anatomical Evolution in Modern Birds” was prepared as a co-authored manuscript, with Dr Larsson as the second author, for submission to Evolution. I performed all of the modifications to the RAxML and MrBayes software packages, and created all of the R scripts used to conducted the analysis. I further conducted, under Dr Larsson's guidance, all of the analyses presented and created all of the figures. Dr Larsson provided editorial support and contributed to the chapter's discussion. All other sections of this thesis, including the general

7 introduction and conclusions, were written solely for inclusion in this thesis and were written by me in their entirety.

STATEMENT OF DISTINCT CONTRIBUTIONS TO KNOWLEDGE AND ORIGINAL

SCHOLARSHIP* * Please refer to the conclusions of this thesis for a more complete summary

This thesis provides several distinct contributions to knowledge and original scholarship. First, in the second chapter, I estimated the prevalence of rate heterogeneity in a diverse sample of matrices of discrete morphological characters by comparing support for models with equal- and unequal-rates in a Bayesian phylogenetic framework. I modified an existing software package to implement the previously proposed discrete lognormal model for among-character rate heterogeneity. I then used this modified software to test support for this model using the collected data sets in a further series of Bayesian phylogenetic analyses. This chapter represents an original and distinct contribution to our empirical understanding of the degree of rate heterogeneity in matrices of discrete morphological characters and how to accommodate it in phylogenetic analysis. In the third chapter, I compiled existing molecular sequence data for 447 species of birds and performed a phylogenetic and Bayesian divergence time analysis. I reviewed the literature and selected 23 fossils to calibrate the analysis. This constitutes one of the largest data sets used to investigate the timing of bird evolution and is a distinct contribution to our understanding of the early evolution of modern birds. In the fourth chapter, I extended current approaches to estimate absolute rates of evolution using discrete morphological characters to accommodate phylogenetic and divergence time uncertainty using distributions of dated phylogenies. This required the creation of extensive, original computer programs in the R language and the modification of an existing maximum likelihood software package to implement the discrete lognormal model for rate heterogeneity among characters, building on the results of the second chapter.

8 Finally, I applied these methods to estimate the absolute rate of evolution of discrete morphological characters across a sample of modern birds, using the timescale derived in the third chapter and a large existing data set of discrete morphological characters. I estimated rates of evolution through time across the early radiation of modern birds and tested the hypothesis that high rates characterized the early evolution of birds. Phenotypic evolutionary rates in this clade had not previously been estimated across all birds using such an extensive sampling of anatomical phenotype. In summary, this thesis provides original scholarship and distinct contributions to knowledge on approaches to estimate rates of evolution using discrete morphological characters and to our understanding of the evolution of modern birds.

9 GENERAL INTRODUCTION

Extant biological diversity is the product of millions of years of evolutionary processes varying in space, time and among lineages and populations of organisms. Rates of evolution are a unifying concept to approach the study of the origin and maintenance of biodiversity at all its levels (Larsson et al., 2012). Among rates of evolution, the study of the tempo and mode of phenotypic change is fundamental to major questions in evolutionary biology (Lloyd et al., 2012). For example, are phenotypic evolutionary rates related to diversification rates and potentially higher during adaptive radiations? (e.g. Adams et al., 2009; Gavrilets and Losos, 2009; Rabosky and Adams, 2012). As a consequence, are overall rates of phenotypic evolution through time uniform, or do they follow a punctuated pattern related to speciation? (Gould and Eldredge, 1977). Are rates of phenotypic evolution related to rates of molecular evolution and if so, how? (e.g. Omland, 1997; Bromham et al., 2002; Davies and Savolainen, 2006). These questions can be investigated using morphological characters, but approaches differ according to the type of character: continuously- or discretely-valued. Study of the evolution of morphological characters is generally divided into methods applied to discretely-valued characters and those for continuously-valued characters, although this distinction is somewhat artificial and new approaches are beginning to close the gap (e.g. Felsenstein, 2012). Discrete morphological characters have been used extensively to estimate phylogenetic relationships and remain necessarily dominant in the paleontological literature. Compared to their utility to infer phylogeny, discrete morphological characters are underexploited in terms of their utility to examine absolute rates of evolution, although methodology and studies do exist (e.g. Forey, 1988; Cloutier, 1991; Ruta et al., 2006; Brusatte et al., 2008; Roelants et al., 2011; Lloyd et al., 2012). This is particularly true compared to continuously-valued characters and traits (Lloyd et al., 2012) that have benefited from considerable attention using comparative data (e.g. Ricklefs, 2004; Adams et al., 2009; Harmon et al., 2010; Burbrink et al., 2012) and recent

10 advances in likelihood-based phylogenetic methods to model evolutionary rates (e.g. O'Meara, 2006; Thomas et al., 2006; Eastman et al., 2011; Revell et al., 2012). The relative lack of attention paid to discrete morphological characters is perhaps unexpected given their long history of use in systematics (Lloyd et al., 2012). Furthermore, in the last decade, data sets of discrete morphological characters have grown in size and now exceed thousands of characters scored for hundreds of taxa, encompassing whole morphological phenotypes (e.g. Livezey and Zusi, 2007; O'Leary et al., 2013). Indeed, two thousand discrete characters coded for all species related by a phylogenetic tree likely contain significant information about the aggregate tempo and mode of phenotypic evolution across phylogeny (Larsson et al., 2012). With the comparative lack of attention to rates of discrete morphological characters, the overall objective of this research project was to empirically investigate rate heterogeneity in an explicitly phylogenetic context using discrete morphological characters. Brown and van Tuinen (2011) argued that Bayesian methods have created a new paradigm for phylogenetic methodology and analysis that embraces the uncertainty in data: complecto errorem. Throughout this thesis, a conscious attempt was made to apply this paradigm; although it is clearly not possible to integrate across all forms of uncertainty simultaneously and model all sources of error, I focused especially on integrating across phylogenetic and divergence time uncertainty. A first objective of this project was to empirically examine rate variation between discrete characters in data matrices used for phylogenetic analysis and to assess rate heterogeneity models. A second objective was to extend current methods to estimate absolute rates of evolution using discrete characters by better incorporating phylogenetic and divergence time uncertainty using independently derived distributions of dated phylogenies. I used modern birds as a case study. In the first chapter, I briefly reviewed the concept of measuring rates of phenotypic evolution and focused specifically on discrete morphological characters. I further reviewed salient phylogenetic studies that estimated absolute rates of morphological evolution from discrete morphological characters,

11 discussed commonalities, which included reliance of a single or handful of phylogenetic topologies, and proposed steps towards an ideal approach to the problem. In the second chapter, I focused on methods to model among-character rate heterogeneity in matrices of discrete morphological characters used for constructing phylogenetic hypotheses. Current models to accommodate such heterogeneity in likelihood-based methods were directly adopted from molecular evolutionary models, but are relatively little investigated for discrete morphological characters. Using a Bayesian model testing approach and a diverse set of 77 published matrices of discrete morphological characters, I first examined the prevalence of rate heterogeneity among characters by comparing equal-and unequal-rates models, integrating across phylogenetic uncertainty. I then tested Wagner's (2012) recent hypothesis that the lognormal distribution, rather than the currently used gamma distribution, is more appropriate to model variation in rates. I also attempted to characterize the behavior of both models under their discrete approximations, different priors and also examined the effect of rate model choice on a phylogenetic analysis. In the third and fourth chapters, I focused on the estimation of absolute rates of evolution using discrete morphological characters in phylogenetic context. I extended current methods to integrate across phylogenetic and divergence time uncertainty using an independently derived distribution of dated phylogenies. The use of distributions of phylogenies is becoming a more common approach in comparative analysis to incorporate uncertainty in phylogenetic relationships and timescales (e.g. Arnold et al., 2010; de Villemereuil et al., 2012; Jetz et al., 2012). I took advantage of the rich data available for modern birds, both in terms of morphology and molecular sequence data, as a case study. The Livezey and Zusi (2006; 2007) data set of discrete morphological characters is one of the largest ever assembled (2954 characters, 188 taxa). Because of its size, comprehensive anatomical coverage of all major groups of modern birds and lack of missing data, it is an ideal data set to estimate evolutionary rates. In order to characterize the absolute rate of morphological evolution along branches in a phylogeny, an

12 estimate of branch temporal durations is required. In the third chapter of this thesis, taking advantage of considerable interest in avian systematics and evolution (e.g. Sibley and Ahlquist, 1990; Fain and Houde, 2004; Ericson et al., 2006; Hackett et al., 2008; Jetz et al., 2012; Wang et al., 2012), I sampled existing molecular sequence data and used Bayesian divergence time dating methods to estimate a distribution of dated phylogenies, or chronograms, for a sample of extant genera spanning the Livezey and Zusi (2006; 2007) data set. This provides a framework upon which to model rates of character evolution. Until very recently (i.e. Jetz et al.'s [2012] composite distribution of chronograms for all extant birds), such a framework did not exist, and because of controversy (e.g. Pagel, 2012) surrounding the Jetz et al. analysis, there is value in this single, consistent Bayesian analysis. In the fourth chapter, I extended existing approaches for estimating absolute rates of evolution using discrete morphological characters to integrate across phylogenetic and divergence time uncertainty applying both parsimony and likelihood-based methods. I used the distribution of chronograms derived in the third chapter and as well as Jetz et al.'s (2012) distribution to estimate absolute rates of evolution using the Livezey and Zusi (2006; 2007) data set across clades and through time. I used likelihood-based and parsimony-based methods to examine the hypothesis that the rate of phenotypic evolution in birds was highest during their early evolution, consistent with predictions of an adaptive radiation (e.g. Gavrilets and Losos, 2009).

13 CHAPTER 1

ESTIMATING EVOLUTIONARY RATES OF DISCRETE

MORPHOLOGICAL CHARACTERS

Rates of evolution of variation are principle drivers in evolutionary biology and at the core of formulating and testing hypotheses regarding the origin and maintenance of biodiversity. Major advances have been made in measuring genotypic, phenotypic and cladogenic evolution, but issues remain concerning how to quantify rates of change within and across these very different levels of biology (e.g. Hendry and Kinnison, 1999; Bokma, 2008; Gingerich, 2009; Lanfear et al., 2010b). In spite of difficulties, rates of evolution remain a key concept to unite understanding of the intertwined processes of evolution at each biological level (Larsson et al., 2012). The definition of rate of evolution is broad and may refer to many related and overlapping quantities of interest to evolutionary biologists. Indeed, Simpson (1949: 205) wrote: “the expression of 'rate of evolution' has so many possible meanings as to be almost meaningless without further qualification” A significant body of literature estimates the tempo and mode of diversification and the birth and death of lineages (e.g. Rabosky and Lovette, 2008; Alfaro et al., 2009; Jetz et al., 2012). A further area of investigation is the rate of molecular evolution and its relationship to life history variables (e.g. Martin and Palumbi, 1993; Gillooly et al., 2005; Nabholz et al., 2008; Lanfear et al., 2010b). Phenotypic evolutionary rates have classically been examined in the paleontological literature (e.g. Simpson, 1944; Haldane, 1949). The advent of new, explicitly phylogenetic likelihood-based methods, particularly for continuously-valued phenotypic traits and comparative data, has focused attention on modern groups (e.g. Harmon et al., 2010). Linkages between these different evolutionary rates have also been examined in the literature: for example, the

14 relationship between molecular evolutionary rates and diversification (e.g. Lanfear et al., 2010a; Goldie et al., 2011), the relationship between phenotypic and molecular evolutionary rates (e.g. Omland, 1997; Bromham et al., 2002; Davies and Savolainen, 2006; Seligmann, 2010; Janecka et al., 2012) and the relationship between rates of phenotypic evolution and diversification (e.g. Adams, et al., 2009; Harmon et al., 2010; Rabosky and Adams, 2012). Noting Simpson's (1949) suggestion quoted above, this review focuses specifically on methods to estimate rates of phenotypic evolution. Phenotypic rates can be measured using continuously or discretely valued characters/traits, although the former has received the greatest attention. The primary focus in this review is discretely-valued morphological characters. First, selected measures to estimate rates of phenotypic evolution for continuously-measured variables are briefly reviewed. Then, methods to estimate and model rates of evolution of discrete morphological characters are reviewed, focusing first on rate heterogeneity between discrete characters used in phylogenetic analysis and then on phylogenetic methods to estimate absolute rates of evolution of phenotypes from discrete characters.

RATES OF PHENOTYPIC EVOLUTION

The measurement of rates of phenotypic evolution has a long history, strongly influenced by Simpson's (1944) Tempo and Mode in Evolution. This work was extended by Haldane (1949) with the introduction of a standardization for rates of continuously-valued traits which he termed the darwin (d). The darwin is defined as the difference in natural log units between the means of a trait over a unit of time (e.g. millions of years). This standardizes changes into units of proportional change, allowing comparisons across traits and organisms. Originally conceived by Haldane (1949) to examine rates of evolution using ancestor-descendant series in the fossil record, darwins have been subsequently used to measure evolutionary rates between contemporary populations (see Hendry and Kinnison, 1999 and

15 references therein). Though straightforward to calculate, the use of the darwin is not without concerns, primarily with respect to the time interval of measurement and choice of the natural logarithm as a unit (e.g. Gingerich, 1993). Building on Haldane's (1949) work to include population variation, Gingerich (1993) developed what he termed the haldane. The haldane differs from the darwin in that it standardizes differences in the natural log of traits means by the pooled standard deviations of those traits, and uses numbers of generations as the denominator, creating a dimensionless and dimension-independent unit (Gingerich, 1993). The haldane therefore quantifies rates in terms of standard deviations rather than units of the natural logarithm, and the choice of which unit to use, darwin or haldane, is context and hypothesis specific (Hendry and Kinnison, 1999). Ackerly (2009) has recently introduced a new measure termed the felsen, which is based on phylogenetic sister comparisons. Using comparative data and methods to reconstruct ancestral traits (e.g. Brownian motion), the felsen is an increase of one unit of variance in natural log-transformed traits per million years. Phylogenetic methods to estimate evolutionary rates of continuously- valued comparative data using likelihood-based approaches have recently been elaborated and can estimate rate heterogeneity between clades to test hypotheses or estimate rate inflection points (e.g. Thomas et al., 2006; O'Meara et al., 2006; Eastman et al., 2011; Revell et al., 2012). In spite of the diversity of methods to examine phenotypic evolutionary rates using continuously-valued data, a wealth of anatomical phenotypes are enumerated as discrete morphological characters. Before continuing with a discussion on estimating evolutionary rates using discrete characters, a discussion on what they are is necessary to make clear the issues concerning the estimation of their evolutionary rates.

TRAITS, CHARACTERS, AND DISCRETE MORPHOLOGICAL

CHARACTERS

16 Traits can be defined as any quantity or quality of an operational taxonomic unit (OTU) and coded continuously or discretely. Traits can include geographic ranges, song types or ecological variables. Traits are frequently analyzed in a phylogenetic comparative framework individually or a handful at a time. Many authors have focused on the evolution of traits in the context of, for example: reconstructing ancestral traits (e.g. Cunningham et al., 1999), estimating character histories (e.g. Ronquist, 2004), estimating correlations between traits (e.g. Pagel, 1994; Minin and Suchard, 2008), modeling the relationship between traits and diversification (e.g. Maddison et al., 2007; Magnuson-Ford and Otto, 2012), and estimating associations between traits and molecular evolutionary rates (e.g. O'Connor and Mundy, 2009; Smith and Donaghue, 2008; Mayrose and Otto, 2011). A trait concerning an OTU's phenotype could range from linear dimensions (e.g. body size) to physiology (e.g metabolic rate) to behaviour (e.g. arboreality). Each of these, in turn, can be described by many other traits. Characters, however, are a special subset of traits. A good working definition of characters was presented by Lewontin as quasi-independent units of evolution (1978: 230). The 'quasi-independence' of characters enforces specific delineations because characters should be as equivalent to and independent from each other as possible. Because of this, characters are a unique transect through trait space that optimizes this independence, equivalence, and enough degree of modularity to encompass homologous evolutionarily variable units. Additionally, many measurable traits, such as bird wing aspect ratios, could in fact be considered as an assembly of many characters that describe relatively independent evolving units within the wing. Among phenotypic characters, morphological characters describe anatomical variation. Discrete morphological characters (DMCs) are anatomical 'units' of evolution that are coded as qualitative descriptive morphological character states. These states may range from absence-presence values, meristic values, shapes, textures, colours, relative sizes, ranges of ratios, descriptions of anatomical contacts, and relative anatomical positions. The analytic treatment of matrices of

17 DMCs can be distinguished operationally from those applied to individual traits or characters in several ways. First, matrices of DMCs are usually codifications of large portions of organismal morphology. Second, matrices of DMCs are usually analyzed with many (usually n > ~25) characters. Third, matrices of DMCs are generally used to reconstruct phylogenies and to provide aggregate information about evolutionary processes at the lineage or clade level.

HETEROGENEITY IN RATES BETWEEN DISCRETE MORPHOLOGICAL

CHARACTERS IN PHYLOGENETIC ANALYSIS

It has long been recognized that morphologies evolve at different rates and have varying degrees of evolutionary and developmental covariances (e.g. Simpson, 1944). These covariances underly the concepts of modularity and mosaic evolution (e.g. Schlosser and Wagner, 2004). High heterogeneity among rates of evolution of molecular characters can mislead phylogenetic analysis, especially when simple models are used (e.g. Sullivan and Swofford, 2001). An extensive literature on rate heterogeneity among sites in molecular sequences has developed over the past several decades. Rate heterogeneity among sites is now known to be common in molecular sequences and, if not accounted for, the effect of phylogenetic inference of topology and branch lengths may be compromised (reviewed in Yang, 1996). Numerous approaches exist in the literature to accommodate among-site variation in molecular sequences using likelihood-based models: for example, using the discrete gamma model (Yang, 1994), assuming a proportion of invariant sites (Gu et al., 1995), partitioning sites (e.g. Shapiro et al., 2006), or a combination of approaches (reviewed in Yang, 2006). Left unaccounted for, ignoring rate heterogeneity in matrices of DMCs could be as much an issue as it is in molecular sequences (Lewis, 2001). This may be particularly important for DMCs, as state transitions in DMCs, although typically weighted equally, are clearly not as equivalent between characters as transitions between nucleotides (Goloboff, 1997). When a phylogeny is reconstructed from

18 discrete morphological characters, the treatment of rate variation differs between the most commonly used methods: maximum parsimony and likelihood-based approaches. Under maximum parsimony, there is no explicit model of evolution or defined evolutionary rates. The most common implementation is unweighted parsimony which treats all character state changes equally between characters and branches (Hennig, 1966; Felsenstein, 2004; Lee and Worthy, 2012). However, character weighting schemes in parsimony methods for molecular sequences or morphological characters (e.g. Farris, 1969) provide a way to lower the consideration given to “unreliable” characters with greater homoplasy, mainly due to higher evolutionary rates (Felsenstein, 2001). These weighting schemes include both a priori character weights (e.g. Neff, 1986) and more complicated data-driven procedures (e.g. Goloboff, 1997). Recent likelihood-based phylogenetic analyses using matrices of DMCs are usually conducted under the Mk model of evolution (Lewis, 2001). The basic Mk model assumes constant rates among sites. Lewis (2001) suggested that the discrete gamma model of Yang (1994), commonly used for modeling among-site rate variation in molecular sequences be applied to likelihood-based analyses of matrices of DMCs. The discrete gamma method integrates character transition probabilities across a set of equiprobable rate categories with rates drawn from the gamma distribution (Yang, 1994). The shape of distribution is estimated from the data and accommodates both highly unequal rates and equal rates of evolution (Pupko and Mayrose, 2010). Several studies have used Bayesian model selection to compare the fit of the discrete gamma model of rate heterogeneity to an equal rates model for DMCs in phylogenetic analyses. These studies have nearly universally supported modeling rate heterogeneity using a discrete gamma distribution over equal-rates models (e.g. Wiens et al. 2005; Müller and Reisz, 2006; Clarke and Middleton, 2008; Ayache and Near, 2009; Prieto-Márquez, 2010), suggesting that rate heterogeneity is a feature of matrices of discrete morphological characters. Müller and Reisz (2006) noted that phylogenies estimated under equal-rates models and without the inclusion of autapomorphies showed weaker support and were generally incongruent with accepted relationships. Clarke and Middleton (2008) also found support for anatomical

19 character partitioning strategies coupled with gamma distributed rates. Wagner (2012) recently argued that the lognormal distribution rather than the gamma distribution is a better approach to model among-character rates of state change and derivation in DMCs because their rates are the product of hierarchical integration, selection and probabilistic processes. Wagner (2012) further supported his hypothesis with a likelihood-based analysis of several invertebrate data matrices. A test of this hypothesis under standard analytical conditions across different data matrices is the subject of the second chapter of this volume.

ABSOLUTE RATES OF EVOLUTION OF DISCRETE MORPHOLOGICAL

CHARACTERS ON A PHYLOGENY

Absolute rates of morphological evolution have classically been studied in the paleontological literature using putative ancestor-descendant relationships without reference to phylogeny (e.g. Simpson, 1944). However, as briefly introduced above, the availability of time-calibrated phylogenies, coupled with comparative data has led to interest in reconstructing rates of phenotypic evolution on branches of phylogenetic trees. Reconstructing rates of evolution on phylogeny permits the exploration of historical patterns of evolutionary rates even in groups with sparse fossil records (e.g. Harmon et al., 2010). For continuously- valued traits, methods exists to estimate rates of phenotypic evolution across whole trees (e.g. Garland, 1992) and to test for differences in rates between clades or groups of taxa (e.g. O'Meara et al., 2006; Thomas et al., 2006). Recent progress includes Bayesian methods to estimate inflection points in evolutionary rates without a priori hypotheses (Eastman et al., 2011; Revell et al., 2012). However, relative to these developments for continuously-valued traits, DMCs, and the rich resource of anatomical information they quantify, have been relatively underused to estimate rates of phenotypic evolution (Lloyd et al., 2012). Because matrices of DMCs can quantify large proportions of organismal phenotype, one advantage of

20 parameterizing phenotypic rates of evolution using DMCs is to approximate rates of evolution of whole organisms. Simpson (1944: p. 15) argued: "Direct determination of rate of evolution for whole organisms, as opposed to selected characters of organisms, would be of the greatest value for the study of evolution" Simpson (1949) also expressly divided the study of rates of evolution into the study of relative rates and absolute rates. Relative rates of evolution of DMCs have been examined by many authors (e.g. Ruta et al., 2006; Dececchi and Larsson, 2009) using patristic distance methodologies, among other methods. Patristic distance analyses focus on phylogenetic pathways and examine the sum of character changes per branch (e.g. as estimated by maximum parsimony), independent of branch durations (Wagner, 1997). Wagner (1997) extended patristic distance to calculate patristic dissimilarities by correcting distance by the number of comparable characters. Although patristic rates can provide information on the tempo of morphological evolution relative to cladogenic events, estimates of the absolute rates of evolution require that changes be measured over absolute time, not numbers of branches (Ruta et al., 2006). Reconstructed changes of DMCs along branches in a phylogenetic tree from a maximum parsimony analysis or branch lengths calculated from a likelihood- based analysis of DMCs represent the product of two confounded variables: the rate of evolution and the time duration of the branch. Estimating the absolute tempo of morphological evolution requires the uncoupling of these two variables (Brusatte, 2011). This is analogous to a divergence time dating where the goal is to deconfound molecular rates from branch durations under an assumption of either a constant rate of evolution (e.g. the molecular clock, Zuckerland and Pauling, 1965) or variable rates of evolution (e.g. relaxed clocks: Drummond et al., 2006). Because there are an infinite possible combinations of rates and durations, it is necessary to use external information to estimate rates and durations (Yang, 2006). For this reason, until recently, most analyses of the absolute rates of evolution of DMCs have been paleontological, because the age of fossils and their stratigraphic ranges provided the necessary timescale to separate rates from durations (e.g. Cloutier, 1991, Ruta et al., 2006; Brusatte et al., 2008). Brusatte (2011) provided a detailed primer and step-by-step

21 guide to paleontological analysis of rates of evolution using discrete morphological characters. These analyses perhaps also reflect the comparative popularity of cladistic analyses using DMCs in the paleontological literature as well as the available temporal data, the history of time series analyses, and macroscopic scope within the paleontological literature. Recent availability of dated phylogenies derived from molecular clock analyses can now provide the external time scale necessary to separate morphological evolution reconstructed from DMCs on phylogenies of modern taxa into estimates of evolutionary rates (e.g. Roelants et al., 2011; see below). The increasing attention given to evolutionary rates estimated from continuously-valued characters suggest that rate studies using discrete morphological characters are due for a renaissance. Below, selected focal studies using discrete morphological characters to reconstruct absolute rates of evolution are reviewed, summarized, and a road map towards using DMCs for robust rate analyses is proposed.

Westoll (1949) In the first study of absolute rates of evolution using DMCs, Westoll (1949) examined the tempo of lungfish evolution. He coded osteology into discrete states, ordered in degree of primitiveness and considered the lineage from ancestral lungfish to modern forms. Like many early paleontological studies of rates of morphological evolution, this was a non phylogenetic analysis that only examined putative ancestor–descendant relationships. He then plotted the total degree of primitiveness of taxa against fossil age on a geological timescale and examined the slopes connecting the data points to estimate rates of evolution. He recovered higher rates of evolution during the early history of lungfish with a decline towards the present. Westoll (1949: 173) also recognized that equivalence of character state changes between characters was problematic and applied what he described as a “crude and admittedly subjective reweighing” to the characters.

Forey (1988) In one of the first phylogenetic studies of DMC rates (see also Derstler, 1982),

22 Forrey (1988) first estimated a cladogram of 16 coelacanth genera using 30 characters using maximum parsimony. He used the estimated ages of the fossils to estimate the ages of nodes in the phylogeny assuming, arbitrarily, that divergences occurred 5 Ma before the earliest occurrence in either sister lineage for that node. He then divided the reconstructed character changes estimated from the same DMCs on the branch separating nodes by the estimated branch duration. Concerned primarily with the distribution of branch rates through time, he noted that rates were highest in the Triassic (~200 Ma ago) and that higher rates appeared to be associated with speciation.

Cloutier (1991) Following Forrey (1988), Cloutier (1991) also examined rates of evolution using DMCs in coelacanths. Using 75 discrete morphological characters, he estimated phylogeny using maximum parsimony. Using a similar method to Forey (1988), Cloutier (1991) determined branch durations by adding 2 Ma to the earliest sister occurrence age. Using the same characters, he then applied three rate metrics to estimate rates along the backbone of the phylogeny to the extant taxon Latimeria: dividing branch changes by estimated branch durations or by binning changes into time intervals and geological periods, and then dividing by total bin time to examine rates through time. Finally, he used correlation tests and recovered a punctuated pattern of evolution in branch rates and an overall decline in rates of evolution through time towards the present.

Wagner (1997) In a study of disparity and rates of evolution among the Rostroconchia (molluscs), Wagner extended previous approaches to measuring branch changes based on estimated character changes with a measurement he termed patristic dissimilarity. This measure corrected the number of reconstructed character changes on a branch by the number of characters comparable between nodes. He used a total of 128 characters to first estimate phylogeny, incorporating measures of stratigraphic fit. He then used the phylogeny and a subset of the characters to

23 estimate pairwise patristic dissimilarities between taxa. By binning taxa according to stratigraphic occurrence, he examined rates of evolution over time across several phylogenetic hypotheses. In contrast to other studies, he did not recover a signal for increased rates early in Rostroconch evolutionary history.

Bromham et al. (2002) Bromham et al. (2002) tested the relationship between morphological evolutionary rates and molecular evolutionary rates. In one of three tests they applied, Phylogenetic Analysis of Rate Estimates (PARE), they reconstructed morphological evolution using parsimony analysis of 13 data sets of DMCs under the ACCTRAN optimization. They used a single molecular phylogeny for each data set and applied non-parameter rate smoothing, without calibrations, to convert the molecular phylograms into ultrametric trees with relative, not absolute, timescales. Dividing parsimony changes by relative branch times, the authors estimated rates of evolution. The authors used these rates to investigate correlations with molecular rates and found no association using this method.

Ruta et al. (2006) Ruta et al. (2006) examined evolutionary patterns in early tetrapods using a data set of 339 DMCs coded for 95 species. They used a phylogeny estimated by Ruta and Coates (2006) using maximum parsimony from the same characters. They then calculated reconstructed patristic dissimilarities for each branch using the DMCs. They used the first out of 324 most parsimonious trees (MPTs) but reported that results were similar between trees. They estimated branch durations using a procedure based on chronostratigraphic distance. They separately considered cases where a taxon was younger than its sister taxon: in that case, the branch duration to that taxon was simply the stratichronographic separation of the sister taxa (see Ruta et al., 2006: Figure 1). In the case where sister taxa were recovered from the same horizon, they used inferred character change frequencies to weight the temporal separation between the next older outgroup taxon into branch durations (see Ruta et al., 2006 and the comprehensive guide in Brusatte,

24 2011 for details on this method). Patristic dissimilarities were then divided by corresponding estimates of branch durations to provide branch rate estimates. Using a nonparametric correlation test, they recovered a significant pattern of decreasing rates of evolution in fossil tetrapods through time. They argued that these results are consistent with an early rapid exploration of morphospace related to reduced intrinsic constraints or ecological restrictions in the group's early history, with a possible later increase in ecological restrictions and constraints.

Brusatte et al. (2008) In a study of both rates of evolution and disparity, Brusatte et al. (2008) examined the evolution of dinosaurs and crurotarsans. Using 437 DMCs scored for 64 taxa, Brusatte et al. (2008) used patristic dissimilarities (see Wagner, 1997, above) calculated from character changes optimized by maximum parsimony under ACCTRAN and DELTRAN. Although not considered by earlier studies, parsimony optimization of character changes onto branches is frequently ambiguous; ACCTRAN and DELTRAN resolve ambiguities by polarizing character changes either as early or as late in the phylogeny as possible, respectively (Maddison, 1994). They used a single phylogeny estimated by employing a tree from a previous study that used 178 of the same discrete characters and grafting missing relationships using a phylogeny estimated from the full 437 character data set. They constructed a time-calibrated phylogeny following a similar method to Ruta et al. (2006; Brusatte, 2011; see above) but they divided durations evenly between contemporary sister taxa rather than weighting by character changes. The authors binned branches by geological age and by clade to calculate estimates of absolute rates of evolution for time periods and groups. Among other patterns, they recovered high rates of character evolution early in the evolutionary history of archosaurs which they interpreted as being consistent with the theory that the rate of evolution is high during morphological radiations.

Roelants et al. (2011)

25 In a study of anuran tadpoles, Roelants et al. (2011) reconstructed evolutionary rates using 131 DMCs for 81 ingroup taxa. Unique among the other analyses considered here, they conducted their analysis on comparative data from an extant group: anuran tadpoles. Using a multi-step analysis, they first constructed a molecular scaffold tree for 164 amphibian species at least congeneric with their morphological sampling using both maximum likelihood and Bayesian analyses and retaining only highly supported nodes. They then constructed a phylogenetic tree using their morphological data set, constrained by the molecular scaffold tree using maximum parsimony and Bayesian inference. They then conducted a Bayesian divergence time analysis on a single most parsimonious tree (first out of 1,760) to provide lineage durations and confidence ranges for the taxa sampled in the morphological analysis. Finally, to calculate morphological evolutionary rates, they divided total changes by corresponding total branch durations for clades, and also plotted each branch by duration and character changes in a cumulative manner to visualize rates. They performed these calculations under the ACCTRAN and DELTRAN parsimony optimization using five randomly chosen parsimonious trees (out of 1,760 MPTs) and the branch lengths from the Bayesian analysis. This represents the first use, to the author's knowledge, of likelihood- based morphological branch lengths to estimate absolute rates of morphological evolution using DMCs. Roelants et al. (2011) recovered highly congruent results from both parsimony optimizations and from the Bayesian analysis. To provide confidence intervals for estimated rates of evolution of clades, they incorporated rate optimization uncertainty by performing a non-parametric bootstrap of the character matrix, and also by using the lower and upper 95% credibility intervals for the total clade durations, as determined by the divergence time analysis. They further generated 500 stochastically simulated character matrices under a model of constant rate of evolution through time to test whether branches were more quickly evolving than expected by stochastic variation under constant rates through time. They recovered a pattern of high rates of evolution at the base of the anuran radiation, associated with phylogenetic diversification and lower rates of evolution in more modern radiations.

26 Lloyd et al. (2012) In a recent paper, Lloyd et al. (2012) developed new approaches to identify rate heterogeneity among branches and clades on phylogenetic trees using DMCs and applied them to the fossil record of lungfish. They proposed three tests for rate heterogeneity: a branch randomization test, a branch likelihood test and a clade likelihood test. Although their statistical approach was likelihood-based, they estimated branch rates following closely the methods of Ruta et al. (2006) and Brusatte et al. (2008): they estimated rates in terms of patristic dissimilarity per Ma by using maximum parsimony under ACCTRAN and DELTRAN optimizations with branch durations estimated using the approach of Ruta et al. (2006). They used a phylogenetic tree estimated from the same DMC data set, using a majority-rule consensus procedure that yielded two resolved phylogenies. To incorporate phylogenetic uncertainty, the authors examined both trees as well as 10 additional trees derived from their consensus method. The authors also introduced a further randomization procedure to accommodate uncertainties in fossil ages by sampling from uniform distributions bounded by the uncertainty in each fossil's age across 1000 iterations. Lloyd et al. (2012) used their approach to examine the evolution of lungfish (Dipnoi), comparing with Westoll's (1949) analysis. They tested reconstructed rates to determine whether rates departed from expectations of constant rates of evolution and equal rates between branches and clades. Although results from the separate tests were not always congruent, they demonstrated that rates of evolution were consistently lower than expected along the phylogenetic backbone and that Dipnoan evolution was characterized by multiple clade-level rate increases. By binning branch rates according to geological time periods, they recovered a pattern of high rates of evolution early in the evolutionary history of Dipnoi, declining towards the present but with several “kinks” during the Permian and Cretaceous.

27 SUMMARY OF PREVIOUS METHODS AND FUTURE DIRECTIONS

The studies described above were diverse in their approaches to estimating evolutionary rates using DMCs and no standard method has yet emerged. Commonalities in the above approaches included the application of parsimony- based methods to estimate character evolution and use of either a single or restricted set of phylogenetic hypotheses which were typically derived from the same data (but see Bromham et al., 2002). That reconstructions of morphological evolution, and therefore rates calculated from them can be sensitive to topology (e.g. Dececchi and Larsson, 2009; Brusatte, 2011), constitutes a serious caveat to these analyses. Furthermore, because most of the studies are paleontological, phylogenetic topologies were estimated using the same or similar DMCs: because maximum parsimony estimates topology to specifically minimize overall character changes, this may have non-negligible effects on inferred rate patterns. With the exception of Roelants et al.'s (2011) study, the studies considered here relied exclusively on maximum parsimony to reconstruct numbers of character changes on branches. Parsimony-based methods may underestimate changes on long branches because they cannot infer multiple changes per character per branch (Bromham et al., 2002). Finally, in most cases, point estimates of branch durations were applied, which may significantly underestimate variability in branch durations and thus rates, a point emphasized by Lloyd et al. (2012), who used fossil age randomization procedures, and Roelants et al. (2011), who used duration credibility intervals from Bayesian divergence time dating.

Appropriate Null Models The choice of null model for examining rates of evolution of discrete morphological characters is less clear than for continuously valued traits (Larsson et al., 2012). The studies above, where such a model was considered, have adopted a model assuming equal rates of evolution among branches or through time (e.g. Ruta et al., 2006; Roelants et al., 2011; Lloyd et al., 2012) and Larsson

28 et al. (2012) went through some length to defend the choice of a constant rates stochastic null model. In their branch randomization test, Lloyd et al., (2012) compared recovered rates against rates estimated by simulated character changes under a constant rate uniform distribution. Finally, Roelants et al. (2011) employed stochastic simulations under constant rates to assess the significance of recovered branch rates.

Likelihood-based Methods for Estimating Morphological Evolutionary Rates: Potential Advantages In the context of reconstructing absolute rates of change from matrices of discrete morphological characters, there are several reasons that likelihood-based approaches (sensu Lewis, 2001) may offer a complimentary and in some ways advantageous approach over parsimony-based methods for estimating the quantity of morphological change per branch. The first is that likelihood-based methods model branch lengths as a function of all the characters, accounting for rate heterogeneity among characters using probabilistic models (Lewis, 2001; Clarke and Middleton, 2008). Parsimony methods do not use branch lengths: character changes are optimized onto branches separately for each character, regardless of branch length. The likelihood approach of treating branch lengths as structural model parameters might lead to more representative estimates of aggregate evolutionary change per branch, especially if combined with more complex models of evolution, like character partitioning (e.g. Clarke and Middleton, 2008). Second, likelihood-based methods may be less sensitive to the node density effect (Venditti et al., 2006), wherein changes along long branches are underestimated because likelihood methods, unlike parsimony, allow multiple changes of the same character per branch (Bromham et al., 2002). Furthermore, likelihood-based models, especially employed in a Bayesian framework, can more intuitively account for uncertainty in parameter estimates. When the number of changes per branch is estimated using parsimony, ambiguously optimized character states fall into the extremes of accelerated transformation (ACCTRAN) and delayed transformation (DELTRAN), which can

29 have different outcomes on rates analysis (e.g. Lloyd et al., 2012). A likelihood- based Bayesian method would report the posterior distribution of branch lengths, rather than two discrete realizations of character state optimization, while a maximum likelihood-based method would report the point estimate of branch lengths that maximizes the probability of observing the data.

An Ideal Model-based Framework for Estimating Absolute Rates of Evolution of Discrete Morphological Characters Given the above discussion of the issues surrounding the estimation of evolutionary rates using DMCs, a set of ground rules can be outlined to optimize a methodology to estimate these rates. First, and foremost, the DMCs that are part of the rate analysis should not be part of the phylogenetic tree estimation. This only leads to tautology and should be avoided. Furthermore, Brusatte (2011) argued that the ideal DMCs for rate analysis are evenly sampled across the phenotype, including autapomorphies, and these characters may not be ideal for cladistic analysis. Trees could be either estimated from other phenotypic characters, molecular characters, or both. The phylogenetic hypotheses should also accommodate tree topology and divergence time uncertainties. Bayesian methods allow for efficient selections of trees that span these uncertainties. Ideally, rather than estimating morphological branch lengths or changes and dividing these a posteriori by branch durations, a fully model-based approach with branch durations as input parameters would be used with explicit assumptions of the evolution of the rate of evolution. Additionally, rate heterogeneity across characters must be considered and methods to accommodate varying degrees of non-independence of characters would ideally be included. Considering the ideal model would integrate across phylogenetic, divergence time and uncertainty in modeling the evolution of the DMCs, a fully Bayesian approach seems ideal. Recent Bayesian methods to incorporate fossil taxa directly into divergence time analysis (Pyron, 2011; Ronquist et al., 2012a; Wood et al., 2013) offer a promising approach where morphological evolutionary rates might be simultaneously estimated with molecular rates, topology and divergence times.

30 Although this violates the above argument that DMCs for rate analysis should not be used to estimate topology, these methods further allow the intuitive incorporation of DMCs from extant and fossil taxa in the same analysis. These studies did in fact estimate rates of morphological evolution of DMCs as part of the divergence dating procedure but this was not their focus. In the latter two studies, the authors expressly linked morphological evolutionary rates to molecular rates, which may or may not be appropriate (e.g. Bromham et al., 2002; Davies and Savolainen, 2006), and which complicates interpretation of morphological rates.

CONCLUSIONS

Although methodology for the estimation of rates of morphological evolution of continuous characters on phylogenies has advanced rapidly in the last ten years, equivalent progress for discrete morphological characters is lagging. Some progress has been made with the introduction of statistical testing frameworks and likelihood-based methods have also been used to estimate absolute rates using matrices of discrete morphological characters. All of the studies considered above used fixed topologies or considered a handful of alternative phylogenetic relationships. An ideal method was described to explicitly model the rate of evolution of morphology integrating across uncertainties in both relationships and divergence times. Between the current methods described here and the unrealized goal of this ideal approach, there are a diversity of possible extensions to current approaches to better accommodate the factors listed above (see Chapters 2 and 4).

31 CONNECTING TEXT

In the first chapter of this thesis, I reviewed methods to estimate rates of evolution of discrete morphological characters. I introduced the concept of heterogeneity in rates of evolution between discrete morphological characters in a phylogenetic analysis. Not accounting for rate-heterogeneity among characters may adversely affect the estimation of phylogenetic topology and branch lengths. In the first chapter, I also noted a recent hypothesis formulated by Wagner (2012) that a lognormally-distributed rates model may better describe rates of change among discrete morphological characters than the usually employed gamma- distributed rates model. In the following chapter of this thesis, I performed an empirical test of Wagner's hypothesis by modifying the commonly used MrBayes software package to implement the lognormal model. I used recently developed Bayesian phylogenetic model selection techniques to explore the prevalence of rate heterogeneity by testing equal-rates and unequal-rates models on a diverse sample of 77 published matrices of discrete morphological characters. I then tested Wagner's (2012) hypothesis by comparing relative support for the lognormal and gamma-distributed rates models In both cases, using these Bayesian methods, I was able to integrate across uncertainty in other parameters, including phylogenetic topology.

32 CHAPTER 2

AMONG-CHARACTER RATE VARIATION IN PHYLOGENETIC ANALYSIS

OF DISCRETE MORPHOLOGICAL CHARACTERS: PREVALENCE AND

BAYESIAN MODEL SELECTION

Likelihood-based methods for phylogenetic analysis using discrete morphological characters (DMCs) are becoming more commonly used for estimating evolutionary relationships among extinct and extant taxa (e.g. Wiens et al., 2005; Müller and Reisz, 2006; Lee and Worthy, 2012). This class of methods has also been used for "total-evidence" studies of phylogeny, which use both molecular sequence data and DMCs to simultaneously infer phylogeny (Nylander et al., 2004; Asher and Hofreiter, 2006; Wiens et al., 2010; Müller et al., 2011; Dávalos et al., 2012; O'Leary et al., 2013; among others) and more recently, divergence times and phylogeny (Pyron, 2011; Ronquist et al., 2012a; Wood et al., 2013). DMCs are typically modeled using continuous-time Markov chains (CTMCs) in a framework nearly identical to that used for molecular sequences (Lewis, 2001). However, DMCs are not nucleotides. Their unique properties, which include lack of constant characters, arbitrary state labeling and character matrix order (when n > 1), enforce restricted substitution models and modifications to the likelihood calculation in order to maintain statistical consistency and accurate branch length estimation (Lewis, 2001). The most commonly used model of character evolution is Lewis' (2001) Mk(V) model, implemented in popular maximum likelihood and Bayesian software packages for phylogenetic inference (e.g. MrBayes: Ronquist et al., 2012b; and RAxML: Stamatakis, 2006). However, there have been comparatively few model selection studies in likelihood-based phylogenetic analysis of DMCs (but see especially Clarke and Middleton, 2008). The comparative lack of attention given to likelihood-based models of DMC evolution may be due to lingering concerns over

33 the sufficiency of time-homogeneous Markov models to model morphological evolution (Clarke and Middleton, 2008; see also e.g. Goloboff and Pol, 2006; Spencer and Wilberg, in press; Sterli et al., in press). The increasing size of data sets of DMCs, now exceeding thousands of characters across greater than 50 taxa (e.g. Livezey and Zusi, 2006; 2007; Naish et al., 2012; O'Leary et al., 2013), suggests that an exploration of model choice and of more complex models is warranted (Clarke and Middleton, 2008; Larsson et al., 2012). As with most phylogenetic studies using molecular sequences, most likelihood-based analyses using DMCs allow rates to vary across characters to account for among-character rate variation (ACRV). Lewis (2001) originally suggested the use of the discretized gamma distribution to model among-character rate heterogeneity and this has been nearly universally adopted. The discretized gamma distribution was a reasonable choice for a number of reasons: it is commonly employed for modeling among-site rate heterogeneity in molecular sequences, is computationally tractable, and is an inherently flexible model, accommodating data sets with little or extensive rate variation (Yang, 2006; Pupko and Mayrose, 2010; see also Waddell et al., 1997; Mayrose et al., 2005; Huelsenbeck and Suchard, 2007; Izquierdo-Carrasco et al., 2011 for alternatives). Several Bayesian phylogenetic analyses of DMCs have tested whether the discrete gamma-distributed rates model improves fit over an equal rates model and the results are nearly universally positive, indicating that ACRV is likely a feature of DMC matrices (e.g. Wiens et al., 2005; Müller and Reisz, 2006; Ayache and Near, 2009; Fröbisch and Shoch, 2009; Prieto-Márquez, 2010). Wagner (2012) recently argued on a theoretical basis that the lognormal distribution should theoretically be a better model to accommodate ACRV for DMCs if evolutionary rates are the product of an underlying probabilistic process (like nucleotides), selection and hierarchical interactions between characters. He further demonstrated empirical evidence for his hypothesis across several invertebrate groups using a customized maximum likelihood model. However, it remains uncertain if his hypothesis is generalizable beyond invertebrates and whether the same pattern would be observed with widely-used likelihood-based

34 software packages (e.g. MrBayes; Ronquist et al., 2012b) and models (e.g. Mk; Lewis, 2001) for phylogenetic analysis in a typical analytical setting. Wagner (2012) considered both rates of state change and state derivation but here only overall rates of change of characters in matrices are considered. Rate heterogeneity in molecular sequences is known to mislead phylogenetic analysis if unaccounted for, especially in simple models (Yang, 1996; Sullivan and Swofford, 2001). It is therefore important to not only understand the degree of among-character rate heterogeneity present in DMC matrices but also which discretized distribution (lognormal or gamma) is an empirically better fit to a typical matrix of characters and what effect model misspecification has on phylogenetic analysis. Furthermore, to the authors' knowledge, the behaviour of the discrete approximation to either the continuous gamma or lognormal distribution under varying numbers of rate categories has not been examined in the context of DMCs. The choice of the number of discrete rate categories is a trade-off between computational tractability and an accurate representation of the underlying continuous probability distribution (Yang, 1994). The analysis presented here has four objectives: a) a generalized empirical test of Wagner's (2012) hypothesis using an existing software package under typical Bayesian analytic conditions; b) an empirical quantification of the degree of among-character rate heterogeneity in published data matrices of DMCs; c) a quantification of the effect of distribution choice and in particular, misspecification on phylogenetic analysis and; finally, d) an analysis of the optimal number of discrete categories to approximate the continuous lognormal and gamma distributions to model ACRV in DMCs.

MATERIALS AND METHODS Data Sets A total of approximately 200 data sets of DMCs used in published, mostly peer-reviewed articles and volumes were downloaded from TreeBASE (www.treebase.org). Matrices were downloaded using an R-script subject only to

35 the requirement that they contained at least fifty characters. Because matrices of DMCs are frequently resampled or augmented by multiple authors and republished, the original data set contained redundancies. Therefore, each matrix and its associated primary literature were inspected and pruned to create as independent a set of matrices as possible, retaining the largest data sets when similar matrices were available. This resulted in a final set of 77 matrices of DMCs that were more or less independent, although some characters and taxa will inevitably have been sampled in different matrices (Appendix 2.1). The final subset consisted of representatives of plant (n = 25), invertebrate (n = 25) and vertebrate (n = 27) data matrices. This is particularly important as biologists working with different taxonomic groups may exhibit cultural biases in character choice and coding (e.g. discretization of morphoclines, treatment of multistate characters, etc.). The relative complexity of groups of organisms themselves may also affect estimates of rates (Schopf et al., 1975). The number of characters per data set ranged from 50 to 444 with an average of 138±94 (mean±SD) characters while the number of operational taxonomic units (OTUs) per matrix ranged from 9 to 246 with an average of 57±39 OTUs. The original matrices were processed to remove OTUs that consisted only of missing data; these were typically present because the original analysis was partitioned and included taxa with molecular sequence data that were not sampled for morphology. Source articles were also reviewed to determine which OTUs where designated as outgroups and these were maintained in all analyses described below. Throughout the following analyses, all multistate characters were considered as unordered because the use of ordering is variable and non-random among systematists.

Bayesian Analysis The latest version of MrBayes v3.2.1 (Ronquist et al., 2012b) was modified following Yang (1994) to implement lognormally distributed among-character rates using median (or quantile in the terminology of Pupko and Mayrose, 2010) discretization with K equiprobable rate classes. MrBayes was selected to investigate these models because it implements the Lewis (2001) MkV model for

36 morphological characters and is frequently used to estimate phylogeny of DMCs. The mean of the lognormal distribution was fixed to 1.0 such that the rate category multipliers varied only with the distribution shape parameter σ2 (Probability Density Function: Eq. 1; Yang, 1994; Johnson et al., 1994).

ln (x−μ )2 2 − 2 σ 1 σ2 (Eq.1) f (x ;σ ,μ=−( )) = e 2 2 x σ √2π Median or quantile discretization was implemented by using (Eq. 2) and the standard normal quantile function to determine the midpoint of the K rate category classes followed by rescaling the category means so that the mean of the discrete

-1 distribution is 1.0; ri is the rate for category I = {1,...,K} and Ф (p) is the standard normal quantile function. The lognormal shape parameter and model were otherwise treated identically to the gamma distribution shape parameter and model in terms of implementation.

μ +σΦ−1( ) p − 2 ( ) = e K = 2i 1 μ=−( σ ) Eq. 2 ri μ+σΦ−1( ) , p , ∑ e p 2 K 2 i Skinner (2010) used a similar discrete approximation of the lognormal distribution to integrate over rate heterogeneity across lineages but parameterized his model such that rates varied with the standard deviation of the lognormal distribution (designated as σ) rather than its shape parameter. It is also possible to parametrize the lognormal distribution based on the non-squared shape parameter σ. Here, the squared shape parameter was chosen but all three methods should produce a nearly identical discrete approximation, although the effective parameter space and posterior distribution of the distribution parameter will differ. MrBayes implements both equal site rate and discrete gamma-distributed site rate models. By default MrBayes uses mean rather than median discretization of the gamma distribution and this method was used for all analyses described here. The relationships between the lognormal and gamma distribution shape parameters (σ2 and α=β, respectively) and the category rates at K = 4 and K = 12 are graphically depicted in Figure 2.1. The modified source code of MrBayes is available from the authors upon request.

37 For each data set, MrBayes was used to determine the marginal model likelihood of the three candidate models (equal rates—EQ, gamma-distributed rates—GA and lognormally-distributed rates—LN), integrating over uncertainty in all other parameters, including phylogenetic topology by using stepping-stone sampling (Xie et al., 2011; Ronquist et al., 2012b). For each model, MrBayes was executed using four independent runs using Metropolis-coupled Markov Chain Monte Carlo sampling with three hot Markov chains and one cold chain for a total of 16 MC2 chains per model per data set. The substitution model was set to the MkV model using the variable-characters only ascertainment correction (Lewis 2001; MrBayes command: lset coding=variable). For both unequal rates models, four discrete rate categories were used. Because stepping-stone sampling also considers the prior in calculating the marginal likelihood (Xie et al., 2011), two alternative priors on the distribution shape parameters were considered: uniform and exponentially distributed. First, uniform priors on the interval [0.0001,200] were applied to the lognormal and gamma distribution shape parameters σ2 and α, respectively. This interval covers the range of effective variability for these parameters (see Fig. 2.1) and the latter prior is the default implemented in MrBayes for the gamma rates model. In addition, the analysis was repeated using exponential priors with a mean of 1.0 on the respective distribution shape parameters. Stepping-Stone sampling was used to calculate marginal model likelihood for each rate model under each set of priors, sampling from the posterior to the prior for a total of at least 15 million generations up to ~30 million generations for the largest data set, sampling every 1 000 generations (parameters for each data set are included in Appendix 2.2). In all cases, initial sampling from the posterior was conducted for at least 5 million generations for each independent run to provide posterior estimates of the topology, branch lengths and distribution parameters. Stepping-stone sampling towards the prior began after 5 million generations, discarding the first 25% of samples from each sampling step as a burnin. The default sampling scheme using 50 steps towards the prior, drawing δ values from a beta distribution with α = 0.4 was applied. All MrBayes analyses

38 were conduced using the CLUMEQ/Colosse and CLUMEQ/Guillimin high performance computing (HPC) facilities and the non-MPI version of MrBayes. Total computation time for the analyses described in this paper exceeded 1 core- year. From the initial 5 million generations sampling from the model's posterior, which were extracted from the MrBayes output using an R script (R Core Team, 2012; available from the authors), a burnin fraction of 0.25 was discarded from each of the four independent runs. Convergence was assessed by using MrBayes to calculate the final average standard deviation of split frequencies (ASDSF) and by ensuring estimated sample sizes for all parameters exceeded 200 using the R package CODA (Drummond et al., 2006; Plummer et al., 2006; Appendix 2.2). Marginal posterior estimates for the majority-rule consensus topology, branch lengths and shape parameters were summarized and annotated using the MrBayes commands sump and sumt. For each data set, the mean estimated marginal model likelihood from the four independent runs was retrieved from the MrBayes output for each model and this value was used as the estimated marginal model likelihood under each prior scheme. Bayes factors were computed as twice the log difference in marginal model likelihoods following Nylander et al. (2004) and were used to assess relative support for the models; the degree of rate heterogeneity was estimated by computing Bayes factors between the equal rates (EQ) and both unequal-rates (GA and LN) models under each distribution shape prior. Interpretation of Bayes factors is subjective, and guidelines from Kass and Raftery (1995; Nylander et al.,

2004) were initially followed: 0 < 2·ln(B10) < 2 no evidence for model 1 over model 0, 2< 2·ln(B10) < 6 positive evidence for model 1 over 0, 6 < 2·ln(B10) < 10 strong evidence for model 1 over 0, 2·ln(B10) > 10 very strong evidence for model 1 over 0. However, initial calculations determined that because of variability in estimates of marginal model likelihoods, a more conservative interpretation was adopted that considered all differences of 2·ln(B10) < 6 to be equivocal. Under each prior combination for the unequal-rates models, if rate heterogeneity was detected using a conservative threshold of 2·ln(B10) > 10 for both the GA:EQ and

39 LN:EQ comparisons, relative support between the lognormal and gamma models was examined by further calculating Bayes factors between these models. Bayes factors were also calculated between the model runs with the highest mean marginal model likelihood under each prior scheme. Pairwise Bayes Factors between the LN and GA models were plotted using histograms in R (R Core Team, 2012).

Topology Comparisons Consensus tree topologies summarizing the marginal posterior tree distribution including only nodes with posterior probability greater than 0.95 were calculated for the GA and LN models under the highest likelihood prior using SumTrees v.3.3.1 (Sukumaran and Holder, 2010). These were compared using the function dist.topo implemented in the R package APE v3.0 (Paradis et al., 2004; R Core Team, 2012). This function implements the Penny and Hendy (1985; Rzhetsky and Nei, 1992) topological distance measure, which corresponds to twice the number of internal branches subtending different partitions between the two phylogenies. This measure was then standardized by dividing by the mean number of internal branches between the compared trees. The topological distance was calculated using only the consensus trees to reduce the effect of topological uncertainty on the comparison. Furthermore, to ensure that comparisons were only made where a sufficient convergence on the models' marginal posterior distribution of topologies was obtained, pairwise comparisons were only performed if the final average standard deviation of split frequencies between the four independent runs of each model was less than 0.01 (e.g. Ronquist et al., 2012b).

Parsimony-based Estimation of Rate Distributions Although parsimony and likelihood-based methods are very different, the distribution of total changes per character inferred by parsimony offers an approximation to the distribution of character rates (e.g. Tourasse and Gouy, 1997). In order to estimate the distribution of among-character rates, maximum

40 parsimony was used to determine the total number of changes for each character across the entire phylogeny for six focal data sets. Of these matrices, two were from data sets where the EQ model was an equivocal fit to the data relative to the GA and LN models under both prior schemes, two were from data sets that showed evidence (2·ln[BLG] > 6) for the LN model over the GA model and two that showed evidence for the GA model (2·ln[BGL] > 6) under both prior schemes. For each data set examined, PAUP* v4.0b10 (Swofford, 2003) was used to map the characters under the ACCTRAN optimization onto each sampled phylogeny (>15000 trees) in the marginal posterior distribution of trees derived from the equal-rates (EQ) Bayesian analysis for each data set (details above). Distributions for each sampled phylogeny from the posterior were then combined for each data set and summarized using relative frequency histograms generated in R (R Core Team, 2012; R script to process the PAUP* output available from the authors)

Optimal Number of Discrete Rate Categories To estimate the minimum number of discrete rate categories required to accurately approximate the continuous lognormal and gamma distributions, an analysis similar to Yang's (1994: Fig. 5) was performed. The four focal data sets included in the parsimony analysis described above showing strong evidence for the unequal-rates models were used. These data sets were then reanalyzed to determine the marginal model likelihood of both unequal rates models using the same MrBayes analysis described above but modified to vary the number of discrete rate categories K across the following values {2, 3, 4, 5, 6, 8, 10, 14, 18} for both the GA and LN rate models under each prior scheme (total 32 model runs). Full stepping-stone sampling analyses, estimating marginal model likelihood, integrating over uncertainty in parameters including topology were applied for each K. Because these analyses were computationally intensive at high K, only two independent runs of four Markov chains each were calculated for each combination of rate-model, prior and K. As well, only 11 million generations per run were considered, discarding a sample from the posterior of 1 million generations but otherwise following the same procedure described above.

41 Marginal model likelihood for each K, prior and model was then plotted against K to determine the minimum number of categories sufficient to represent the continuous distribution.

Focal Phylogenetic Analysis For a focal topology, branch length and clade support comparison, one data set (TreeBaseID S12929; Sidlauskas and Vari, 2008; see below for rationale) was analyzed using MrBayes in a standard (i.e. not stepping-stone sampling) MCMC analysis. The analysis was conducted under four different among-character rate variation models: equal-rates (EQ), gamma-distributed rates (GA) and lognormally-distributed rates (LN) at K = 4 and K = 12 rate categories. MC3 sampling was conducted using four independent runs of four Markov chains running for 15 million generations, sampling every 1 000 generations for each rate model using a uniform prior, as above, on the distribution shape parameters. All other parameters were identical to the model selection analysis described above. Convergence of independent runs was assessed by calculating Potential Scale Reduction Factors (PSRF; Gelman and Rubin, 1992) for all model parameters and split frequencies as well as calculating the final average standard deviation of split frequencies (ASDSF) after combining runs and discarding a burnin fraction of 25% using the MrBayes sump command. Sufficient sampling was assessed by ensuring that estimated sample size (ESS) of all parameters was greater than 200 using Tracer v1.5 (Drummond et al., 2006; Rambaut and Drummond, 2007). Majority-rule consensus trees were calculated for each model using the MrBayes sump command, after discarding the burnin fraction. Mean marginal posterior estimates of total tree length were calculated for each rate model using Tracer, combining samples from the posterior of each run after discarding the burnin fraction. Pair-wise comparisons between clade posterior probabilities were made between nodes shared by the compared majority-rule consensus trees using the R package APE (Paradis, 2004).

42 RESULTS Equal Rates Models and Unequal Rates Models Calculated marginal model likelihoods for each data set and each rate model are recorded in Appendix 2.2 for both prior combinations. Variability in marginal model likelihoods between independent runs was low for most data sets, with average ranges of 1.2, 1.3 and 1.4 of log likelihood units between replicates for the EQ, GA and LN model runs (uniform prior model). In the most extreme example, data set S11535 (TreeBaseID), a range of 12 log likelihood units was noted between independent runs for the GA model under the uniform prior. For this reason, this data set was excluded from further analysis. Furthermore, given the observed variability of runs, any Bayes factor difference less than 6 was deemed equivocal (see also Clarke and Middleton, 2008). In 61/76 data sets, the Bayes factors comparing both unequal rates models to the equal rates models exceeded the threshold of 10 for the presence of significant ACRV using uniform priors on the distribution shape parameters (Table 2.1). Using exponential priors on the distribution shape parameters, 64/76 data sets exceed the same thresholds. The exponential prior almost always led to higher likelihoods for both rates models relative to the uniform prior, and the results of the comparisons of the prior choice yielding the highest likelihood were essentially equivalent to the exponential comparisons (Table 2.1; Appendix 2.2). Bayes factors comparing the lognormal rate model (under the prior choice of highest likelihood) to the equal rates were significantly correlated to the number of characters (Fig. 2.2a; Spearman's rank correlation test r = 0.41, p < 0.001) and much more strongly correlated to the number of OTUs present (Fig. 2.2b, r = 0.86, p < 0.001). Correlations calculated between GA:EQ comparisons and Bayes factors calculated using uniform priors or the exponential likelihood priors on distribution shape parameters were nearly identical (not shown).

Gamma and Lognormal Rate Distributions Among the 61/76 data sets with significant rate heterogeneity under uniform

43 shape distribution priors, 2/61 data sets showed evidence (2·ln(BGL ) > 6) for the GA model over the LN model and a single data set showed very strong evidence

(2·ln(BGL ) > 10) (Table 2.1; Fig. 2.3a). Conversely, 18/61 data sets showed evidence for the LN model and 9/61 showed very strong evidence. The remaining 41 data sets (67%) were equivocal. Results for Bayes factors calculated from models using exponentially distributed priors on rate parameters were similar but more equivocal with respect to model choice and generally weakened support for the LN model relative to the GA model (Table 2.1; Fig. 2.3b). A further comparison of Bayes factors calculated between data sets using the highest marginal model likelihood recovered under either prior scheme was nearly identical to the comparison using exponential priors, as these nearly always led to the highest likelihood for each model (Fig 2.3c). Under exponential priors and the prior yielding the highest likelihood, 47 out of the 64 data sets with evidence for an unequal-rates model were equivocal with respect to support for the GA or LN models. However, among data sets where Bayes factors were not equivocal, the

LN model was supported more often: 13/64 (20%) data sets 2·ln(BLG) > 6 compared to 4/64 (6%) for 2·ln(BGL). There was no obvious relationship between taxonomic group and rate model nor was standardized topological distance between consensus trees obviously related to model preference under the prior choice yielded the highest likelihood (Fig. 2.3d). Most consensus trees were identical between models (Fig. 2.3d), and inspection of several topologies that differed indicated that observed tree distance were due to nodes around the posterior probability cutoff (0.95), which were included in one rate model's consensus tree but excluded in the other's. The results of the initial Bayesian analysis were used to designate six focal data sets for further analysis: two data sets where the EQ model had equivocal support relative to the unequal-rates models under all prior schemes (TreeBase ID#: S10249 and S13029), two with evidence for the LN model (S10265 and S12929; 2·ln(BLG) = 9.6 and 11.4 under the uniform priors and 6.87 and 8.37 under the exponential) and two with evidence for the GA model (S2128, S12833; 2·log℮(BLG) = 8.9 and 13.3 under the uniform priors and 10.4 and 13.9 under the exponential priors). Bayes factors for

44 these data sets under the prior choice that yielded the highest likelihood were identical to the exponential prior scheme.

Parsimony Analysis Parsimony estimation of character rate distributions of the six focal data sets were summarized using relative frequency histograms (Fig. 2.4). The two data sets where the EQ model had equivocal support relative to the unequal-rates models had relatively few changes per character (Fig 2.4a,b). The data sets where the LN model was supported had relatively more characters with low rates and a rapidly decreasing number of characters with higher rates of change (Fig. 2.4c,d). Finally, the data sets where the GA model was supported had a relatively more uniform distribution of character rates (Fig. 2.4e,f).

Optimal Number of Discrete Rate Categories Mean marginal model likelihoods calculated for varying K across the four focal data sets and priors are recorded in Table 2.2 and plotted in Figure 2.5. Model likelihoods were always highest under the exponential prior for a given rates model. Variability within runs and between likelihoods under adjacent K values for the same data sets were relatively high but an overall trend is apparent: in the four focal data sets examined, regardless of whether the gamma or lognormal rate model was originally the best fit, four discrete rate categories appeared to be sufficient to represent the continuous gamma distribution (Table 2.2; dotted lines in Fig. 2.5). Beyond four rate categories, increasing K did not significantly increase the GA model's marginal likelihood. However, in both data sets where the lognormal model was a better fit at K = 4 in the original analysis, the marginal model likelihood under the LN model increased with K and only begins to asymptote at K > 10–12 (Fig. 2.5a,b). In data sets that preferred the gamma model, the behaviour of the LN model with respect to K was similar to the GA (Fig. 2.5c,d) but decreased at K > 4 for the S2128 data set (Fig 2.5c) indicating that the continuous distribution would be a worse fit than the discretized distribution (see Discussion)

45 Effect on Phylogenetic Topology and Branch Lengths: Sidlauskas and Vari (2008) Majority-rule consensus trees inferred by MrBayes under the equal (a), gamma, K = 4 (b), lognormal, K = 4 (c) and lognormal, K = 12 (d) ACRV models for data set S12929 for Sidlauskas and Vari (2008) are summarized in Figure 2.6. PSRFs were 1.0 and ESS was > 200 for all model parameters and split frequencies. The final ASDSF was < 0.01 for each model run, indicating that each independent run for each rate model converged to the posterior distribution. This data set, which demonstrated greatest fit to the LN model (see Fig. 2.5b), was used in a phylogenetic analysis of relationships in the Anostomidae family of South American fishes and consisted of 158 DMCs coded for 60 OTUs (Sidlauskas and Vari, 2008). The data matrix had 3% missing data and characters had an average of 2.35 states. The means of the marginal posterior distribution of total tree lengths were, respectively, 4.54, 6.65, 7.07 and 9.1 character changes per character for the EQ, GA (K = 4), LN (K = 4) and LN (K =12) ACRV models. There were several differences between the topology inferred under the EQ model and trees inferred under the GA and LN models: the EQ model resolved more nodes than the GA or LN models (Fig. 2.6a, arrows). There were one or two nodes that differed between the GA and LN models (K = 4 or K = 12; Fig. 2.6b,c,d, starred arrows). Clade posterior probabilities were highly similar between the GA and LN models (Fig. 2.7c) but quite different in the EQ model (Fig. 7a,b). The LN (K = 12) had slightly higher overall clade posterior probabilities relative to the GA model (Fig. 2.7d) and LN (K = 4) model (not shown).

DISCUSSION Rate Heterogeneity in Data Sets of Discrete Morphological Characters Several previous Bayesian studies of DMCs have tested equal rates models against gamma-distributed rates models for individual data sets (e.g. Wiens et al., 2005; Müller and Reisz, 2006; Ayache and Near, 2009; Fröbisch and Shoch, 2009;

46 Prieto-Marquez, 2010). The conclusions of those studies must be qualified by the use of the harmonic mean estimator (HME) for model likelihood which can be biased, especially towards more complex models (Xie et al., 2011). Here, using the stepping stone sampling method to estimate marginal model likelihood, which weights model likelihoods by their priors to penalize more complex models (Xie et al., 2011; Ronquist et al., 2012b), most data sets still exhibited significant rate heterogeneity by Bayes factors indicating very strong evidence for unequal-rates models over equal-rates models (Table 2.1; Fig. 2.2). Bayes factors supporting unequal-rates models were more strongly correlated to the number of OTUs compared to the number of characters (Fig. 2.2), a pattern not observed in molecular sequences (e.g. Mayrose et al., 2005: Fig. 3). This suggests that increasing data set size, especially numbers of OTUs, should increase the ability to detect rate heterogeneity. This analysis corroborates those previous studies using the harmonic mean estimator and suggests that ACRV is significant in DMC matrices.

Gamma and Lognormal Rate Distributions Felsenstein (2001) argued that for molecular sequences, either the gamma or lognormal distribution should be effective to model among-site rates simply because both are distributions on the interval [0,∞]. He further argued that it would be difficult to differentiate these distributions without large quantities of data. Felsenstein's arguments were directed towards molecular sequences and without the added effect of discretization. Here, in matrices of DMCs, it was possible to discriminate between the discretized gamma and lognormal ACRV models among some data sets of modest size. This analysis strongly concurs with Felsenstein (2001) in that most data sets show equivocal support for either the GA or LN models, at least when using a 2 ln(BF) < 6 criterion under two different prior specifications. However, among the subset of the data matrices that are not equivocal, evidence for the lognormally-distributed ACRV model was estimated more often than evidence for the gamma rates model; this offers some, albeit qualified, support to Wagner (2012)'s hypothesis. However, this conclusion must

47 be strongly qualified as clearly data set dependent: in some data sets there was strong evidence for gamma-distributed rates model, especially under an exponential prior on the rate distribution shape parameter. (Table 2.1; Fig. 2.3c). Wagner (2012) also noted support for gamma-distributed models in some data sets, especially considering character change rates. Preference for either rate model was not obviously related to major taxonomic division or data set size (see Fig. 2.3d), although a comprehensive testing of that question is beyond the sample size and scope of this analysis. Stepping-stone sampling considers the prior in the estimation of the marginal model likelihood (Xie et al., 2011). In the case of the identical uniform priors used here, because of the nature of the underlying distributions, more prior weight was given to equal-rates among sites using the gamma-model while the lognormal was the converse (see Fig. 2.1). Under the exponentially distributed prior, the opposite is true because the direction of variation (i.e. rates becoming more equal) is reversed among these two rate distributions (Fig. 2.1). Here, both priors were tested to determine the effect of the prior on model choice, which was appreciable (see. Fig. 2.3a vs. Fig. 2.3b). Even considering the prior choice that produced the highest marginal model likelihood for each model, there is still a weak preference for the lognormal model among the data sets that are not equivocal (Fig. 2.3c). However, there were also some data sets that conclusively favoured the gamma rates model under the same comparison. The influence of the prior on the marginal model likelihood was relatively strong among these data sets and may reflect a relatively large weight of the prior compared to the information content of the data. Future studies using larger, potentially more informative data sets of DMCs may also offer sufficient information to better discriminate these two models. Finally, an alternative “generalized” stepping-stone approach to calculate marginal model likelihoods that does not require sampling close to the actual prior might offer more efficient and precise model comparison (e.g. Fan et al., 2011) but this does not yet accommodate variable topologies (Baele et al., 2012). Parsimony-based reconstruction of six focal character rate distributions (Fig.

48 2.4) corroborated the Bayesian analysis and hinted why the lognormal distribution was a better fit to some data sets. In both data sets where the equal-rates model was an equally good choice relative to unequal-rates models (S10249, S13029; Fig 2.4a,b), parsimony reconstruction of character changes confirmed a largely equal distribution of rates. The two data sets where the evidence for the LN was recovered (Fig. 2.4c,d) had proportionally more characters with slow rates and a smaller number of characters with much higher rates. In contrast, the data sets where evidence for the GA model was recovered had much more even distribution of character change rates, that is, more unequal rates (Fig 2.4e,f). Category rates when K = 4 for the LN and GA models under varying distribution shape parameters (Fig. 2.1) hint at why the lognormal distribution was a better fit to data sets containing many characters with low, but not equally low rates. Compared to the gamma distribution (Fig. 2.1c,d), the lognormal distribution (Fig 2.1a,b) has several slow categories and one fast rate category while the gamma distribution's rate category means are generally more evenly spaced on the {1...K} interval except for the small interval near 0. The parsimony analysis supports the hypothesis that the choice of ACRV models in select data sets is driven by underlying patterns of character evolution. However, the complex interaction between subjective factors (character construction and coding) and objective factors (phylogenetic signal, number of OTUs, etc.) make thorough understanding of the observed patterns complex.

Optimal Number of Discrete Rate Categories Computation time increases with the number of discrete rate categories used to approximate the continuous lognormal or gamma distributions as the overall character transition probabilities are integrated over all rate categories (Yang, 1994). Because of this significant computational cost, it is important to minimize the number of rate categories while maintaining an accurate approximation of the continuous distribution. For molecular sequences, Yang (1994) determined that four rate categories were sufficient to represent the gamma distribution when using mean discretization of rate categories. Mayrose et al. (2005) argued that

49 larger data sets may require a greater number of categories or a gamma-mixture model. The choice of four categories is common in evolutionary studies and is usually the default setting in software packages (e.g. RAxML v7.2.8; Stamaskis, 2006). Because DMCs are not nucleotides nor amino acids, the choice of four may not be appropriate. Here, given the high computational cost at high numbers of rate categories (K), only four focal data sets were assessed: two matrices where the LN model was supported (S12929, S10265) and two matrices where GA model was supported (S2128, S12833). The analysis of marginal model likelihood under each rate model for the four focal data sets revealed divergent behavior between the GA and LN models. For matrices where the LN model was supported, marginal model likelihood under the LN model increased with K and only began to asymptote at K > 10–12 (Fig. 2.5a,b). Interestingly this pattern is similar to that observed by Skinner (2010: Fig. 1a) when he discretized the lognormal distribution to model among- lineage rate heterogeneity. This is in contrast to the GA model, where marginal model likelihood reached an asymptote at K ≈ 4 in data sets for which it was and was not supported (Fig. 2.5). It is possible that discretization of the lognormal distribution using the mean rather than the median of the respective rate categories might improve the model's performance with respect to K, as this pattern is observed in molecular data (e.g. see Yang, 1994). It is not immediately clear why the marginal model likelihood for the LN model in the S2128 data set actually decreased at K > 4; a possible explanation is that the continuous lognormal distribution would have been a poorer fit to the data and that discretization, which at lower K is less representative of the continuous distribution, buffered the effect of the model misspecification. Although it is possible that this analysis underestimated variability in marginal likelihoods by using only two runs per K per data set, these general trends appear robust. Further interpretation of smaller- scale trends would require a greater number of independent runs for each K. Because the main analysis presented above used K = 4 rate categories, it may underestimate the support for the LN over the GA model in some data sets or overestimate it in others. Investigators are therefore encouraged to use as many

50 rate categories as feasible to test LN and GA models. Increasing the number of discrete rate categories may impose non-trivial computational cost. However, phylogenetic analyses under the MkV model (Lewis, 2001) have a fixed instantaneous rate matrix and few variable parameters: typically only the among-character distribution shape parameter, branch lengths and topology. Comparatively few free parameters and the small size of most character matrices compared to molecular analyses mean that the increase in computational resources required to at least double the number of rate categories to eight might not be onerous except for large data sets. This is particularly true given that multi-core personal computers are now commonplace and that HPC resources coupled with parallelized phylogenetic packages are increasingly available for phylogenetic analysis of DMCs (e.g. BEAGLE; Ayres et al., 2012). The larger number of rate categories required to sufficiently approximate the lognormal distribution (Fig. 2.5a,b) and the computational cost associated argue that the discretization method may not be the optimal approach to approximate the underlying continuous lognormal distribution for DMCs. Alternatives to approximate the distribution without the need for discretization (sensu Yang, 1994) may significantly increase model fit without excessive computational cost, particularly if compared to discrete models using high values for K. The Laguerre quadrature (Felsenstein, 2001; Mayrose et al., 2005) has been used with success to approximate the continuous gamma distribution and similar alternatives may exist for the lognormal distribution.

Effects on Estimations of Phylogenetic Topology Phylogenies recovered under the GA and LN models rarely differed in the phylogenetic topologies recovered (Fig 2.3d). However, this comparison was based on summaries of the marginal posteriors on topology using a consensus tree containing only very strongly supported clades and so may be overly conservative with respect to topological differences that are poorly supported. The degree of topological incongruence between the LN and GA models was not obviously related to the degree of preference (or alternatively, the degree of

51 misspecification) of models and was highest among data sets with equivocal evidence for each model (Fig 2.3d). The observed topological differences between ACRV models is likely complex and possibly related to the degree of phylogenetic signal, degree of ACRV, and stochastic variability from the MC3 sampling, etc. Because the recovered topology did sometimes depend on the ACRV model, it seems prudent to suggest that researchers should test and apply the model of highest marginal likelihood for the data set under investigation. The focal analysis of the S12929 data set recovered the greatest topological differences between the EQ model and both unequal-rates models (Fig. 2.6: arrows). Clade posterior probabilities were very different between the EQ model and both unequal-rates models (Fig. 2.7a,b), a fact observed in other studies comparing equal rates to gamma-distributed rates but possibly also related to coding of autapomorphies (Müller and Reisz, 2006). Interestingly, the EQ model resolved more nodes in the majority-rule tree than did the GA or LN models, which is likely symptomatic of the overestimation of Bayesian clade posterior probabilities when the model is overly simplified (e.g. Alfaro and Holder, 2006). Topologies inferred under the unequal-rates models were very similar: both LN models resolved an additional node at the base of the tree (Fig. 2.6c, starred arrow) while the GA resolved an additional node in the Lepornius genus (Fig. 2.6b, starred arrow). Finally, the LN (K = 12) resolved a different node among Lepornius relative to the LN (K = 4) or GA models (Fig. 2.6d, starred arrow). Functionally, the topologies recovered under the GA and LN models were nearly identical and clade posterior probabilities were also very similar for clades with PP > 0.5, although slightly higher for the LN (K = 12) model (Fig. 2.7d). Overall tree length was higher for the LN (K = 12) model and indeed, tree lengths appeared to be high for all rate models. This may be related to the recent observation that MrBayes' default independent exponential priors on branch lengths can lead to overestimates of total tree length in some data sets, especially if a gamma-distributed rate model is used (Brown et al., 2010; Marshall, 2010; Rannala et al., 2012; Zhang et al., 2012). It would have been interesting to examine the relationship between discretization, rate model choice and overall

52 tree length using the modified version of MrBayes created by Zhang et al. (2012) that resolves this issue by using compound Dirichlet priors on branch lengths. However, when this was attempted, the likelihoods estimated by this version of the MrBayes program using stepping-stone sampling were highly variable (occasionally > 200 log likelihood units between independent runs of the same model) for both equal-rates and gamma-rates models. It was not possible to determine the source of this variability. Implementation issues aside, the relationship between the use of DMCs and interactions between among-character rate variation models, overall branch lengths and the Lewis (2001) ascertainment correction clearly requires further consideration. In summary, although topological differences were observed between unequal rate models, these appear to be generally minor relative to not accounting for rate heterogeneity among characters.

Alternative Approaches to Model ACRV The results presented here must be accompanied by a strong caveat: as with any model selection study, the results are conditioned on the choice of candidate models (Posada and Buckley, 2004). Here, only three models were evaluated and this analysis does not preclude alternative models that may demonstrate significantly better fit to the data. Approaches to ACRV can be divided by analogy into “random-effects” models and “fixed-effects”-type statistical models (Yang, 2006; Yang and Rannala, 2013). The approaches here are random-effect models because they integrate across rate heterogeneity without expressing assigning characters to a priori rate categories. Alternative random-effects approaches include Dirichlet-process-prior based models, where the number of rate categories and assignments are estimated from the data (e.g. Huelsenbeck and Suchard, 2007). In this context, fixed-effect type models include a priori partitioning of sets of characters and estimating rates within each partition. Partitioned Bayesian analysis of molecular sequences has been demonstrated to significantly increase model fit (e.g. Brandley et al., 2005; Brown and Lemmon, 2007) and is an alternative and complimentary approach to dealing with heterogeneity across

53 sites/characters, especially when biologically informed (e.g. partitioning by codon position: Shapiro et al., 2006). These approaches are now commonplace, especially for phylogenomic analyses, and extensive tools are now available to facilitate partition selection (e.g. PartitionFinder; Lanfear et al., 2012). Clarke and Middleton (2008) found that partitioning DMCs anatomically and allowing branch lengths to vary across partitions led to higher marginal model likelihoods over an unpartitioned gamma-distributed rate model in a fossil data set. However, the results from that study are qualified by Clarke and Middleton's use of the Harmonic Mean Estimator (HME) for marginal model likelihood, which may overestimate likelihoods of complex models (e.g. Xie et al., 2011; Baele et al., 2012; this is not meant as a criticism as stepping-stone sampling has only recently become available in MrBayes and Clarke and Middleton expressly qualified their study on the appropriateness of the HME). Partitioned models may be particularly appropriate for DMCs because character integration and selection affecting multiple characters may lead to correlated evolutionary changes, and therefore rates (Wagner, 2012). In the absence of easily available implementations of more complicated strategies for partitioning DMCs (e.g. Lanfear et al., 2012), investigators are encouraged to perform model testing using biologically informative a priori partitions and both among-character rate distributions using as large a number of discrete rate categories as possible.

CONCLUSIONS Rate heterogeneity was widespread in this diverse data set of discrete morphological character matrices as indicated by strong preference for unequal rates models. Where significant rate heterogeneity was present, Bayesian model selection suggested that most data sets were equivocal in support for the gamma- distributed and lognormally-distributed rates models. However, where model testing was not equivocal, the lognormal model was more frequently supported relative to a gamma-distributed rates model. However, the reverse was also true for several data sets. Overall, the results suggest weak evidence for Wagner's

54 (2012) hypothesis with strong qualifications. Prior choice was found to affect marginal model likelihood estimation but the observed patterns hold after comparison of model likelihoods under the prior choice yielding the highest marginal model likelihood for each model were compared. Parsimony analysis of estimated character distributions in four focal data sets suggests that the underlying character rate distribution may drive observed patterns in some data sets. Although rate distribution choice sometimes affected topological reconstruction, this pattern was difficult to characterize and did not appear to be related to the degree of model preference as measured by pairwise Bayes factor. Lognormally-distributed rate models may require a larger number (K > 8) of discrete rate categories to approximate the continuous distribution, in contrast to gamma-distributed rate models where K = 4 rate categories appeared sufficient. Researchers should therefore use as many discrete rate categories as computationally feasible to ensure an accurate representation of the underlying continuous distribution when using the lognormal distribution. Although the data were not conclusive, the rate distribution choice was observed to have a minor effect on the posterior distribution of topologies and branch lengths, researchers are suggested to test which among-character rate model is best supported by their data and explore alternative character partitioning strategies using Bayesian marginal model likelihood estimation and Bayes factors.

55 Category Rate [r ] Category Rate [r ] a) i c) i K =4 K =4 Shape Parameter (α=β Parameter Shape Shape Parameter (σ Parameter Shape 2 ) ) 56

Category Rate [ri] d) Category Rate [ri] b) K =12 K =12 Shape Parameter (α=β Parameter Shape Shape Parameter (σ Parameter Shape 2 ) ) Figure 2.1. Plot of discrete category rates against distribution shape parameters for the discretized lognormal distribution at K = 4 (a) and K = 12 (b) and the gamma distribution at K = 4 (c) and K = 12 (d) discrete rate classes.

57 ) LE Pairwise 2 ln ( B

Number of Characters ) LE Pairwise 2 ln ( B

Number of OTUs

Plants Vertebrates Invertebrates

58 Figure 2.2. Plot of data set size in terms of a) number of discrete morphological characters and b) number of operational taxonomic unit (OTUs) against Bayes factors calculated using twice the log difference between the mean marginal model likelihood for the equal and lognormally distributed among-character rate models. The marginal likelihood of the lognormal model was estimated using the distribution shape prior that yielded the highest model likelihood (see text). Positive Bayes factors represent support for the lognormal model over the equal rates model. Plant data sets are indicated by green circles, vertebrate data sets by red squares and invertebrate data sets by blue diamonds.

59 a) Frequency

b) Frequency

c) Frequency

d) Pairwise Tree Distance Tree Pairwise

-30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30

Pairwise Bayes Factor 2 ln(BLG)

Plants Vertebrates Invertebrates

60 Figure 2.3. Bayesian comparison of the discrete gamma and lognormally- distributed among-character rates models. Histograms summarize the Bayes factors 2 ln(BLG) comparing the gamma and lognormal models of among- character rate variation for each data set under a) uniform shape distribution priors, b) exponential shape distribution priors and c) highest likelihood under either prior. Negative 2 ln(BLG) values indicate support for the gamma model over the lognormal model while positive values indicate support for the lognormal model. Vertical lines indicate 2 ln(BLG) = {-6,6}, and 2 ln(BLG) = {-10,10}, where there is respectively, strong evidence and very strong support for one model over the other. Standardized pairwise topology distance d) between consensus trees (posterior probabilities > 0.95) estimated under the gamma and lognormal models plotted against 2 ln(BLG). Note that the number of pairwise comparisons is less in d) because comparisons were only made between analyses where the average standard deviation of split frequencies was less than 0.01 (see text). Data points in are scaled to relative data set size. Plant data sets are indicated by green circles, vertebrate data sets by red squares and invertebrate data sets by blue diamonds.

61 a) b) S10249 (n=137) S13029 (n=128) Relative Frequency Relative Frequency

c) Total Changes / Character d) Total Changes / Character

S10265 (n=165) S12929 (n=158) Relative Frequency Relative Frequency

e) Total Changes / Character f) Total Changes / Character

S2128 (n=308) S12833 (n=308) Relative Frequency Relative Frequency

Total Changes / Character Total Changes / Character

62 Figure 2.4. Parsimony reconstruction of the distribution of total character changes per character for six focal data sets using PAUP*. These relative frequency histograms summarize total character changes per character pooled across a sample of topologies (> 10 000) from the marginal posterior of topologies from the equal-rates Bayesian analysis (see text); n is the number of discrete morphological characters present in each data set. The equal rates model was at least an equivocal fit to data sets a) S10249 and b) S10329 relative to the unequal- rates models; the lognormal model was preferred for data sets c) S10265 and d) S12929 and the gamma model was preferred for data sets e) S2128 and f) S12833.

63 a) b)

S10265 S12929 Marginal Model Log Likelihood Marginal Model Log Likelihood

c) Number of Discrete Rate Categories (K) d) Number of Discrete Rate Categories (K)

S2128 S12833 Marginal Model Log Likelihood Marginal Model Log Likelihood

Number of Discrete Rate Categories (K) Number of Discrete Rate Categories (K)

GA, Exponential Prior GA, Uniform Prior LN, Exponential Prior LN, Uniform Prior

64 Figure 2.5. Relationship between mean marginal model log likelihood for the discrete gamma (GA) and lognormal (LN) models of among-character rate variation and the number of discrete rate classes (K) under uniform and exponential shape distribution prior choices. LN model likelihoods are by solid lines while the GA model likelihoods are dotted lines. Marginal model likelihoods calculated under the exponential model are red diamonds, while those calculated under the uniform prior model are blue circles. The lognormal rates model was a better fit to data sets S10265 and S12929 (a,b) while the gamma rate model was a better fit to S2128 and S12833 (c,d). While the gamma distribution is accurately represented using K = 4 rate classes, the continuous lognormal distribution is more better represented with greater number of discrete rate categories (K > 8) but only data sets where it has the best fit (a,b). Error bars represent the range of marginal model likelihoods from the two independent stepping-stone sampling runs (see text).

65 a) b)

*

c) d)

*

*

0.5 changes/character

66 Figure 2.6. Majority-rule consensus summarizing the posterior distribution of phylogenies inferred under the a) equal-rates model, the b) gamma-rates model (K = 4, and the lognormal rates model (K = 4, c; K = 12, d). Clade posterior probabilities are indicated by colored dots: yellow > 0.5 and green > 0.95. Branch lengths were annotated using the mean of the marginal posterior distribution of the respective branch lengths using MrBayes. Means of the marginal posterior distribution of total tree lengths were, respectively, 4.54, 6.65 7.07 and 9.1 character changes per character for the EQ, GA (K = 4), LN (K = 4) and LN (K = 12) models. Arrows indicate topological differences (see text for details).

67 b) a) Clade Posterior Probability (GA) Clade Posterior Probability (LN, K =4)

Clade Posterior Probability (EQ) Clade Posterior Probability (EQ) c) d) Clade Posterior Probability (GA) Clade Posterior Probability (LN, K =12) Clade Posterior Probability (LN, K=4) Clade Posterior Probability (GA)

68 Figure 2.7. Comparison of clade posterior probabilities for the S12929 data set under different among-character rate heterogeneity models. a) Equal-rates model vs. gamma-distributed rates (K = 4) model, b) equal-rates vs. lognormal- distributed (K = 4) rates model, c) gamma-distributed rates (K = 4) vs. lognormal- distributed rates model (K = 4) and d) gamma-distributed rates (K = 4) vs. lognormal distributed rate model (K = 12)

69 Table 2.1. Results of the Bayesian analysis of the three models of among-character rate variation: equal rates, gamma- distributed rates, and lognormally-distributed rates Uniform Distribution Shape Priors GA/Equal 2 LN/Equal 2 GA/LN 2 ln(B ) LN/GA 2 ln(B ) ln(B ) ln(B ) GL LG GE LE

2 ln(B ) 2 ln(B ) Data Sets 2 ln(B ) 2 ln(B ) 2 ln(B ) 2 ln(B ) Group # Data Sets GE LE with Unequal GL GL LG LG > 10 > 10 Rates* > 6 > 10 > 6 > 10 Vertebrates 27 21 (78%) 21 (78%) 21 (78%) 0 (0%) 0 (0%) 7 (33%) 4 (19%) Invertebrates 24 22 (%) 22 (%) 22 (92%) 2 (9%) 1 (5%) 8 (36%) 3 (14%) Plants 25 18 (72%) 19 (76%) 18 (72%) 0 (0%) 0 (0%) 3 (16%) 2 (11%)

70 Total 76 61 (80%) 62 (82%) 61 (80%) 2 (3%) 1 (2%) 18 (30%) 9 (%)

Exponential Distribution Shape Priors GA/Equal 2 LN/Equal 2 GA/LN 2 ln(B ) LN/GA 2 ln(B ) ln(B ) ln(B ) GL LG GE LE 2 ln(B ) 2 ln(B ) Data Sets 2 ln(B ) 2 ln(B ) 2 ln(B ) 2 ln(B ) Group # Data Sets GE LE with Unequal GL GL LG LG > 10 > 10 Rates* > 6 > 10 > 6 > 10 Vertebrates 27 22 (81%) 22 (81%) 22 (81%) 1 (5%) 0 (0%) 5 (23%) 3 (14%) Invertebrates 24 23 (96%) 23 (96%) 23 (96%) 2 (9%) 2 (9%) 5 (22%) 2 (9%) Plants 25 19 (76%) 19 (76%) 19 (76%) 1 (5%) 0 (0%) 3 (12%) 1 (5%) Total 76 64 (88%) 64 (88%) 64 (88%) 4 (6%) 2 (9%) 13 (20%) 6 (9%) Table 2.1. Continued. Highest Marginal Model Likelihood (Exponential or Uniform Prior) GA/Equal 2 LN/Equal 2 GA/LN 2 ln(B ) LN/GA 2 ln(B ) ln(B ) ln(B ) GL LG GE LE 2 ln(B ) 2 ln(B ) Data Sets 2 ln(B ) 2 ln(B ) 2 ln(B ) 2 ln(B ) Group # Data Sets GE LE with Unequal GL GL LG LG > 10 > 10 Rates* > 6 > 10 > 6 > 10 Vertebrates 27 22 (81%) 22 (81%) 22 (81%) 1 (5%) 0 (0%) 5 (23%) 3 (14%) 71 Invertebrates 24 23 (96%) 23 (96%) 23 (96%) 2 (9%) 2 (9%) 5 (22%) 2 (9%) Plants 25 19 (76%) 19 (76%) 19 (76%) 1 (5%) 0 (0%) 3 (12%) 1 (5%) Total 76 64 (88%) 64 (88%) 64 (88%) 4 (6%) 2 (9%) 13 (20%) 6 (9%) Table 2.2. Results of the Bayesian analysis of the discrete approximation of the gamma and lognormal distributions of ACRV Uniform Rate Distribution Priors Unequal Rates: Mean Model Marginal Model Likelihood EQ Data # # Model Rate Set OTUs Chars. ln(L) Model K = 2 K = 3 K = 4 K = 5 K = 6 K = 8 K = 10 K = 14 K = 18 LN -5299 -5283 -5281 -5280 -5276 -5271 -5268 -5267 -5267 S10265 111 176 -5582 GA -5299 -5286 -5286 -5286 -5287 -5284 -5283 -5282 -5282 LN -2576 -2573 -2572 -2570 -2568 -2566 -2565 -2562 -2561 S12929 60 158 -2660 GA -2576 -2578 -2577 -2577 -2578 -2578 -2576 -2579 -2577 LN -11227 -11181 -11176 -11177 -11175 -11177 -11176 -11179 -11178 S12833 78 207 -11714 GA -11226 -11174 -11169 -11170 -11169 -11169 -11167 -11166 -11167

72 LN -9231 -9220 -9220 -9222 -9224 -9223 -9225 -9224 -9226 S2128 56 308 -9478 GA -9229 -9216 -9216 -9216 -9215 -9216 -9216 -9215 -9216 Exponential Rate Distribution Priors Unequal Rates: Mean Model Marginal Model Likelihood EQ Data # # Model Rate Set OTUs Chars. ln(L) Model K = 2 K = 3 K = 4 K = 5 K = 6 K = 8 K = 10 K = 14 K = 18 LN -5293 -5282 -5278 -5276 -5273 -5270 -5267 -5266 -5264 S10265 111 176 -5582 GA -5295 -5280 -5281 -5282 -5284 -5281 -5279 -5279 -5277 LN -2574 -2570 -2569 -2567 -2565 -2563 -2562 -2561 -2560 S12929 60 158 -2660 GA -2572 -2574 -2573 -2573 -2573 -2573 -2572 -2572 -2573 LN -11222 -11177 -11173 -11175 -11174 -11175 -11173 -11176 -11177 S12833 78 207 -11714 GA -11224 -11168 -11165 -11164 -11163 -11165 -11165 -11164 -11163 LN -9226 -9217 -9216 -9218 -9218 -9219 -9222 -9220 -9221 S2128 56 308 -9478 GA -9227 -9212 -9212 -9213 -9212 -9213 -9214 -9212 -9212 Appendix 2.1. List of data sets and TreeBase Accession Numbers

Treebase Number Number of Taxonomic StudyID of OTUs Chars. Group Reference S10029 37 123 Vertebrates O'Leary (1999) S10045 192 105 Plants Baker et al. (2009) S10081 31 80 Vertebrates Fröbisch and Schoch (2009) S10108 27 202 Vertebrates Kenaley (2010) Werenkraut and Ramierz S10135 101 444 Invertebrates (2009) S10146 33 92 Plants Almeida et al. (2009) S10204 33 54 Invertebrates Pitz and Sierwald (2010) S1023 44 77 Invertebrates Smith et al. (2005) S10249 31 137 Vertebrates Holland et al. (2010) S10252 50 52 Plants Chemisquy et al. (2010) S10265 111 176 Invertebrates Frick et al. (2010) S10466 23 100 Invertebrates Beutel et al. (2010) S1050 95 55 Plants Hughes et al. (2004) S10512 78 54 Invertebrates De Wit et al. (2011) S10667 129 109 Plants Sundue (2010) S107 26 93 Plants Rodman (1991) S108 19 55 Plants Simpson (1990) S10868 78 50 Plants Huang et al. (2010) S10905 40 148 Vertebrates Lin and Hastings (2011) S10908 64 363 Vertebrates Wiens et al. (2010) S11114 49 161 Vertebrates Pyron (2011) S1116 59 80 Plants Malécot et al. (2004) S1117 46 67 Invertebrates Percy (2003) S112 13 63 Plants Eriksson (1994) S1121 46 147 Invertebrates Wappler (2004) S1125 16 55 Invertebrates Banks and Paterson (2004) S11258 149 69 Invertebrates Syme and Oakley (2011) S11368 12 51 Vertebrates Agnolin (2007) S11387 246 76 Plants Simmons et al. (2012) S114 56 88 Plants Karis (1989) S11535 105 392 Invertebrates Sharkey et al. (2012) S11810 53 117 Vertebrates Peach and Rouse (2004) S11957 100 359 Vertebrates Seiffert et al. (2005) S11965 54 219 Vertebrates Anderson et al. (2008) S12060 32 88 Invertebrates DaCosta et al. (2006) S12319 58 236 Vertebrates Makovicky et al. (2005) S12447 28 52 Vertebrates Bibi et al. (2012) S12832 71 245 Vertebrates Ksepka et al. (2012)

73 Appendix 2.1. Continued. Treebase Number Number of Taxonomic StudyID of OTUs Chars. Group Reference S12833 78 207 Invertebrates Lambkin and Bartlett (2011) S12850 44 249 Vertebrates Albert (2001) S12853 68 83 Vertebrates Britto (2003) S12870 76 64 Invertebrates Damgaard (2008) S12929 60 158 Vertebrates Sidlauskas and Vari (2008) S12933 66 157 Vertebrates Smith (1992) S13029 21 128 Vertebrates Bourdon (2011) S13043 43 339 Vertebrates Ezcurra and Brusatte (2011) S1330 44 349 Vertebrates Gaubert et al. (2005) S1346 14 72 Plants Penneys and Judd (2005) S1398 13 89 Vertebrates Vieira et al. (2005) S1512 25 132 Vertebrates Müller and Reisz (2006) S1516 23 126 Vertebrates Asher and Hofreiter (2005) S1518 64 242 Invertebrates Agnarsson (2006) S1580 73 68 Invertebrates Marek and Bond (2006) S1666 37 105 Plants Wortley et al. (2007) S1782 23 61 Plants Sousa et al. (2007) S1816 48 137 Invertebrates Just and Wilson (2007) S1820 61 62 Plants Roalson et al. (2008) S1875 42 65 Invertebrates Corona and Morrone (2007) S1884 26 61 Plants Ruiz-Sánchez et al. (2008) S1986 44 57 Invertebrates Weiblen (2001) S1988 26 105 Plants Gernandt et al. (2008) S2128 56 308 Invertebrates Liljeblad et al. (2008) S2132 48 64 Invertebrates Maruyama et al. (2008) S2158 51 129 Vertebrates Voss and Jansa (2009) S2179 71 113 Plants Vega et al. (2009) S2229 94 81 Plants Clement and Weiblen (2009) S856 28 176 Vertebrates Kearney (2003) S860 49 97 Invertebrates Cotton (2001) S874 45 147 Invertebrates Marshall (2003) Wilson and Edgecombe S920 61 165 Invertebrates (2003) S95 28 54 Plants Donoghue and Doyle (1989) S953 70 110 Plants Farmer and Schilling (2002) S9941 35 136 Plants Schneider et al. (2009) S9957 88 220 Invertebrates Dikow (2009) S997 65 117 Plants Schulman and Hyvnen (2003) S9993 82 127 Plants Struwe et al. (2009) Car2012 61 351 Vertebrates Carrano et al. (2012)

74 Appendix 2.2. Results of the Bayesian analysis Equal Rates Treebase # Post. StudyID # Gen.1 Gen.2 ASDSF3 Run1 Run2 Run3 Run4 Mean S10029 1.5e+07 4.95E+06 0.007 -1721 -1721 -1721 -1721 -1721 S10045 2.88E+07 9.50E+06 0.034 -6838 -6834 -6833 -6836 -6834 S10081 1.5e+07 4.95E+06 0.006 -999 -999 -999 -999 -999 S10108 1.5e+07 4.95E+06 0.004 -2099 -2099 -2099 -2099 -2099 S10135 1.52E+07 5.00E+06 0.008 -7426 -7426 -7423 -7427 -7425 S10146 1.5e+07 4.95E+06 0.007 -2111 -2111 -2111 -2111 -2111 S10204 1.5e+07 4.95E+06 0.006 -733 -733 -733 -733 -733 S1023 1.5e+07 4.95E+06 0.007 -893 -893 -893 -893 -893 S10249 1.5e+07 4.95E+06 0.004 -1096 -1096 -1096 -1096 -1096 S10252 1.5e+07 4.95E+06 0.010 -1592 -1592 -1592 -1592 -1592 S10265 1.67E+07 5.49E+06 0.021 -5581 -5582 -5583 -5581 -5582 S10466 1.5e+07 4.95E+06 0.005 -819 -819 -819 -819 -819 S1050 1.5e+07 4.95E+06 0.009 -1550 -1552 -1551 -1551 -1551 S10512 1.5e+07 4.95E+06 0.011 -1782 -1783 -1784 -1784 -1783 S10667 1.94E+07 6.39E+06 0.014 -4362 -4365 -4363 -4363 -4363 S107 1.5e+07 4.95E+06 0.006 -1333 -1333 -1333 -1333 -1333 S108 1.5e+07 4.95E+06 0.006 -393 -393 -393 -393 -393 S10868 1.5e+07 4.95E+06 0.009 -1067 -1065 -1066 -1066 -1066 S10905 1.5e+07 4.95E+06 0.004 -1673 -1673 -1673 -1673 -1673 S10908 1.5e+07 4.95E+06 0.006 -7230 -7227 -7229 -7227 -7228 S11114 1.5e+07 4.95E+06 0.006 -2949 -2950 -2950 -2950 -2950 S1116 1.5e+07 4.95E+06 0.007 -1977 -1977 -1977 -1977 -1977 S1117 1.5e+07 4.95E+06 0.010 -1888 -1887 -1888 -1887 -1887 S112 1.5e+07 4.95E+06 0.007 -366 -366 -366 -366 -366 S1121 1.5e+07 4.95E+06 0.034 -2801 -2800 -2801 -2800 -2800 S1125 1.5e+07 4.95E+06 0.007 -413 -413 -413 -413 -413 S11258 2.24E+07 7.38E+06 0.015 -2640 -2641 -2638 -2640 -2639 S11368 1.5e+07 4.95E+06 0.005 -227 -227 -227 -227 -227 S11387 3.69E+07 1.22E+07 0.014 -2780 -2778 -2773 -2782 -2774 S114 1.5e+07 4.95E+06 0.008 -1336 -1336 -1336 -1336 -1336 S11810 1.5e+07 4.95E+06 0.006 -1634 -1633 -1633 -1633 -1633 S11957 1.5e+07 4.95E+06 0.029 -15903 -15906 -15905 -15902 -15903 S11965 1.5e+07 4.95E+06 0.012 -4581 -4581 -4582 -4580 -4580 S12060 1.5e+07 4.95E+06 0.007 -1108 -1108 -1108 -1108 -1108 S12319 1.5e+07 4.95E+06 0.008 -2967 -2966 -2967 -2967 -2967 1Total number of generations for posterior and stepping-stone sampling per run 2Number of generations of MC3 sampling from the posterior per run 3Final average standard deviation of split frequencies between runs

75 Appendix 2.2. Continued. Equal Rates Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S12447 1.5e+07 4.95E+06 0.006 -730 -730 -729 -729 -730 S12832 1.5e+07 4.95E+06 0.009 -2905 -2906 -2905 -2905 -2905 S12833 1.5e+07 4.95E+06 0.034 -11714 -11717 -11714 -11714 -11714 S12850 1.5e+07 4.95E+06 0.005 -2733 -2734 -2735 -2734 -2734 S12853 1.5e+07 4.95E+06 0.011 -1774 -1775 -1775 -1774 -1774 S12870 1.5e+07 4.95E+06 0.006 -933 -935 -934 -933 -933 S12929 1.5e+07 4.95E+06 0.007 -2662 -2659 -2661 -2660 -2660 S12933 1.5e+07 4.95E+06 0.008 -3743 -3743 -3743 -3743 -3743 S13029 1.5e+07 4.95E+06 0.005 -897 -897 -897 -897 -897 S13043 1.5e+07 4.95E+06 0.006 -4154 -4153 -4154 -4154 -4154 S1330 1.5e+07 4.95E+06 0.007 -7094 -7095 -7095 -7094 -7095 S1346 1.5e+07 4.95E+06 0.007 -637 -637 -637 -637 -637 S1398 1.5e+07 4.95E+06 0.006 -510 -510 -510 -510 -510 S1512 1.5e+07 4.95E+06 0.005 -1198 -1198 -1197 -1198 -1198 S1516 1.5e+07 4.95E+06 0.006 -1264 -1264 -1264 -1264 -1264 S1518 1.5e+07 4.95E+06 0.003 -3822 -3823 -3822 -3823 -3822 S1580 1.5e+07 4.95E+06 0.017 -2370 -2370 -2370 -2369 -2370 S1666 1.5e+07 4.95E+06 0.007 -1910 -1910 -1910 -1910 -1910 S1782 1.5e+07 4.95E+06 0.006 -881 -881 -882 -881 -881 S1816 1.5e+07 4.95E+06 0.006 -2607 -2607 -2607 -2607 -2607 S1820 1.5e+07 4.95E+06 0.008 -1667 -1666 -1666 -1667 -1667 S1875 1.5e+07 4.95E+06 0.007 -1157 -1157 -1157 -1157 -1157 S1884 1.5e+07 4.95E+06 0.007 -869 -869 -869 -869 -869 S1986 1.5e+07 4.95E+06 0.006 -1466 -1466 -1466 -1465 -1466 S1988 1.5e+07 4.95E+06 0.006 -1216 -1216 -1216 -1216 -1216 S2128 1.5e+07 4.95E+06 0.015 -9477 -9479 -9479 -9478 -9478 S2132 1.5e+07 4.95E+06 0.006 -822 -821 -822 -822 -822 S2158 1.5e+07 4.95E+06 0.007 -1697 -1696 -1696 -1697 -1696 S2179 1.5e+07 4.95E+06 0.013 -3805 -3804 -3805 -3806 -3805 S2229 1.5e+07 4.95E+06 0.012 -2995 -2995 -2996 -2995 -2995 S856 1.5e+07 4.95E+06 0.005 -2122 -2122 -2122 -2121 -2122 S860 1.5e+07 4.95E+06 0.011 -1814 -1814 -1814 -1814 -1814 S874 1.5e+07 4.95E+06 0.018 -2812 -2812 -2811 -2811 -2811 S920 1.5e+07 4.95E+06 0.008 -3759 -3760 -3761 -3760 -3760 S95 1.5e+07 4.95E+06 0.006 -771 -771 -771 -771 -771 S953 1.5e+07 4.95E+06 0.008 -3958 -3957 -3957 -3957 -3957 S9941 1.5e+07 4.95E+06 0.005 -1646 -1646 -1646 -1646 -1646 S9957 1.5e+07 4.95E+06 0.033 -7731 -7733 -7729 -7731 -7730

76 Appendix 2.2. Continued. Equal Rates Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S997 1.5e+07 4.95E+06 0.013 -3330 -3330 -3331 -3330 -3330 S9993 1.5e+07 4.95E+06 0.010 -3478 -3479 -3480 -3480 -3479 Car2012 1.5e+07 4.95E+06 0.008 -4396 -4397 -4397 -4397 -4397

Gamma Rates, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S10029 1.5e+07 4.95E+06 0.005 -1692 -1691 -1692 -1691 -1692 S10045 2.88E+07 9.50E+06 0.016 -6404 -6406 -6403 -6407 -6404 S10081 1.5e+07 4.95E+06 0.006 -995 -995 -995 -995 -995 S10108 1.5e+07 4.95E+06 0.004 -2090 -2090 -2090 -2090 -2090 S10135 1.52E+07 5.00E+06 0.009 -7115 -7115 -7115 -7116 -7115 S10146 1.5e+07 4.95E+06 0.006 -2089 -2089 -2089 -2089 -2089 S10204 1.5e+07 4.95E+06 0.006 -709 -709 -709 -709 -709 S1023 1.5e+07 4.95E+06 0.006 -872 -872 -872 -872 -872 S10249 1.5e+07 4.95E+06 0.004 -1096 -1096 -1096 -1096 -1096 S10252 1.5e+07 4.95E+06 0.009 -1553 -1553 -1553 -1553 -1553 S10265 1.67E+07 5.49E+06 0.013 -5286 -5286 -5285 -5286 -5285 S10466 1.5e+07 4.95E+06 0.005 -815 -815 -815 -815 -815 S1050 1.5e+07 4.95E+06 0.007 -1490 -1489 -1490 -1491 -1490 S10512 1.5e+07 4.95E+06 0.013 -1698 -1698 -1698 -1697 -1698 S10667 1.94E+07 6.39E+06 0.012 -4156 -4159 -4157 -4158 -4157 S107 1.5e+07 4.95E+06 0.006 -1319 -1319 -1319 -1319 -1319 S108 1.5e+07 4.95E+06 0.005 -393 -393 -393 -393 -393 S10868 1.5e+07 4.95E+06 0.008 -1001 -1000 -1000 -999 -1000 S10905 1.5e+07 4.95E+06 0.004 -1651 -1651 -1651 -1651 -1651 S10908 1.5e+07 4.95E+06 0.008 -7073 -7070 -7070 -7073 -7071 S11114 1.5e+07 4.95E+06 0.007 -2888 -2887 -2886 -2888 -2887 S1116 1.5e+07 4.95E+06 0.007 -1943 -1943 -1943 -1944 -1943 S1117 1.5e+07 4.95E+06 0.009 -1862 -1862 -1861 -1862 -1862 S112 1.5e+07 4.95E+06 0.004 -366 -366 -366 -366 -366 S1121 1.5e+07 4.95E+06 0.028 -2684 -2684 -2685 -2684 -2684 S1125 1.5e+07 4.95E+06 0.007 -413 -413 -413 -413 -413 S11258 2.24E+07 7.38E+06 0.012 -2390 -2390 -2390 -2392 -2390 S11368 1.5e+07 4.95E+06 0.005 -227 -227 -227 -227 -227 S11387 3.69E+07 1.22E+07 0.014 -2411 -2411 -2412 -2410 -2411 S114 1.5e+07 4.95E+06 0.008 -1283 -1283 -1284 -1284 -1283

77 Appendix 2.2. Continued. Gamma Rates, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S11810 1.5e+07 4.95E+06 0.007 -1625 -1626 -1626 -1627 -1626 S11957 1.5e+07 4.95E+06 0.032 -15400 -15402 -15403 -15404 -15401 S11965 1.5e+07 4.95E+06 0.005 -4488 -4488 -4489 -4488 -4488 S12060 1.5e+07 4.95E+06 0.006 -1086 -1085 -1085 -1085 -1085 S12319 1.5e+07 4.95E+06 0.008 -2955 -2954 -2953 -2954 -2954 S12447 1.5e+07 4.95E+06 0.007 -721 -720 -721 -721 -721 S12832 1.5e+07 4.95E+06 0.012 -2837 -2837 -2836 -2837 -2837 S12833 1.5e+07 4.95E+06 0.013 -11169 -11170 -11168 -11170 -11169 S12850 1.5e+07 4.95E+06 0.003 -2677 -2678 -2678 -2677 -2678 S12853 1.5e+07 4.95E+06 0.010 -1692 -1692 -1693 -1692 -1692 S12870 1.5e+07 4.95E+06 0.006 -868 -867 -868 -867 -868 S12929 1.5e+07 4.95E+06 0.006 -2578 -2577 -2578 -2576 -2577 S12933 1.5e+07 4.95E+06 0.007 -3668 -3670 -3669 -3668 -3669 S13029 1.5e+07 4.95E+06 0.005 -897 -897 -897 -897 -897 S13043 1.5e+07 4.95E+06 0.004 -4110 -4111 -4111 -4110 -4110 S1330 1.5e+07 4.95E+06 0.024 -6952 -6954 -6954 -6952 -6953 S1346 1.5e+07 4.95E+06 0.007 -632 -632 -632 -632 -632 S1398 1.5e+07 4.95E+06 0.006 -510 -510 -510 -510 -510 S1512 1.5e+07 4.95E+06 0.005 -1172 -1172 -1172 -1172 -1172 S1516 1.5e+07 4.95E+06 0.006 -1264 -1264 -1264 -1264 -1264 S1518 1.5e+07 4.95E+06 0.004 -3742 -3744 -3743 -3741 -3742 S1580 1.5e+07 4.95E+06 0.010 -2285 -2286 -2286 -2286 -2286 S1666 1.5e+07 4.95E+06 0.006 -1889 -1890 -1889 -1889 -1889 S1782 1.5e+07 4.95E+06 0.006 -880 -880 -880 -880 -880 S1816 1.5e+07 4.95E+06 0.009 -2512 -2513 -2513 -2513 -2513 S1820 1.5e+07 4.95E+06 0.008 -1616 -1616 -1617 -1617 -1616 S1875 1.5e+07 4.95E+06 0.006 -1100 -1101 -1101 -1101 -1101 S1884 1.5e+07 4.95E+06 0.007 -869 -869 -869 -869 -869 S1986 1.5e+07 4.95E+06 0.006 -1402 -1401 -1402 -1402 -1402 S1988 1.5e+07 4.95E+06 0.006 -1215 -1215 -1215 -1215 -1215 S2128 1.5e+07 4.95E+06 0.007 -9216 -9216 -9216 -9216 -9216 S2132 1.5e+07 4.95E+06 0.006 -811 -811 -810 -811 -811 S2158 1.5e+07 4.95E+06 0.006 -1670 -1669 -1669 -1669 -1669 S2179 1.5e+07 4.95E+06 0.009 -3649 -3649 -3650 -3649 -3649 S2229 1.5e+07 4.95E+06 0.012 -2933 -2933 -2933 -2932 -2933 S856 1.5e+07 4.95E+06 0.006 -2111 -2111 -2110 -2111 -2111 S860 1.5e+07 4.95E+06 0.008 -1777 -1776 -1776 -1776 -1776 S874 1.5e+07 4.95E+06 0.019 -2692 -2691 -2691 -2691 -2691

78 Appendix 2.2. Continued. Gamma Rates, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S920 1.5e+07 4.95E+06 0.009 -3661 -3661 -3662 -3662 -3662 S95 1.5e+07 4.95E+06 0.007 -769.6 -769.2 -769.2 -769.4 -769.3 S953 1.5e+07 4.95E+06 0.007 -3808 -3807 -3808 -3808 -3807 S9941 1.5e+07 4.95E+06 0.005 -1630 -1630 -1630 -1630 -1630 S9957 1.5e+07 4.95E+06 0.009 -7303 -7306 -7303 -7308 -7304 S997 1.5e+07 4.95E+06 0.013 -3134 -3134 -3133 -3134 -3134 S9993 1.5e+07 4.95E+06 0.009 -3387 -3388 -3389 -3388 -3388 Car2012 1.5e+07 4.95E+06 0.009 -4350 -4349 -4348 -4348 -4348

Lognormal Rates Model, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S10029 1.5e+07 4.95E+06 0.006 -1692 -1691 -1691 -1691 -1691 S10045 2.88E+07 9.50E+06 0.024 -6408 -6406 -6406 -6408 -6407 S10081 1.5e+07 4.95E+06 0.007 -996 -996 -996 -996 -996 S10108 1.5e+07 4.95E+06 0.003 -2090 -2090 -2090 -2090 -2090 S10135 1.52E+07 5.00E+06 0.013 -7112 -7111 -7110 -7107 -7108 S10146 1.5e+07 4.95E+06 0.006 -2091 -2090 -2091 -2090 -2090 S10204 1.5e+07 4.95E+06 0.006 -705 -704 -704 -704 -704 S1023 1.5e+07 4.95E+06 0.006 -870 -870 -870 -870 -870 S10249 1.5e+07 4.95E+06 0.004 -1103 -1103 -1103 -1103 -1103 S10252 1.5e+07 4.95E+06 0.009 -1554 -1554 -1554 -1554 -1554 S10265 1.67E+07 5.49E+06 0.015 -5281 -5281 -5281 -5283 -5281 S10466 1.5e+07 4.95E+06 0.005 -814 -814 -814 -814 -814 S1050 1.5e+07 4.95E+06 0.008 -1486 -1487 -1485 -1485 -1485 S10512 1.5e+07 4.95E+06 0.014 -1697 -1697 -1697 -1697 -1697 S10667 1.94E+07 6.39E+06 0.012 -4157 -4160 -4159 -4160 -4159 S107 1.5e+07 4.95E+06 0.006 -1320 -1320 -1320 -1320 -1320 S108 1.5e+07 4.95E+06 0.006 -398 -398 -398 -398 -398 S10868 1.5e+07 4.95E+06 0.008 -996 -996 -995 -995 -995 S10905 1.5e+07 4.95E+06 0.004 -1647 -1647 -1646 -1647 -1647 S10908 1.5e+07 4.95E+06 0.018 -7071 -7074 -7075 -7073 -7073 S11114 1.5e+07 4.95E+06 0.006 -2885 -2885 -2885 -2886 -2885 S1116 1.5e+07 4.95E+06 0.007 -1944 -1944 -1944 -1944 -1944 S1117 1.5e+07 4.95E+06 0.009 -1863 -1863 -1863 -1863 -1863 S112 1.5e+07 4.95E+06 0.004 -370 -370 -370 -370 -370 S1121 1.5e+07 4.95E+06 0.037 -2682 -2682 -2683 -2682 -2682

79 Appendix 2.2. Continued. Lognormal Rates Model, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S1125 1.5e+07 4.95E+06 0.006 -418 -418 -418 -418 -418 S11258 2.24E+07 7.38E+06 0.012 -2386 -2382 -2385 -2378 -2380 S11368 1.5e+07 4.95E+06 0.004 -232 -232 -232 -232 -232 S11387 3.69E+07 1.22E+07 0.015 -2407 -2401 -2402 -2402 -2402 S114 1.5e+07 4.95E+06 0.008 -1276 -1277 -1276 -1276 -1276 S11810 1.5e+07 4.95E+06 0.006 -1620 -1619 -1619 -1619 -1619 S11957 1.5e+07 4.95E+06 0.034 -15404 -15407 -15403 -15405 -15404 S11965 1.5e+07 4.95E+06 0.005 -4487 -4486 -4487 -4487 -4486 S12060 1.5e+07 4.95E+06 0.007 -1085 -1085 -1086 -1085 -1085 S12319 1.5e+07 4.95E+06 0.009 -2955 -2954 -2955 -2955 -2955 S12447 1.5e+07 4.95E+06 0.006 -720 -720 -720 -720 -720 S12832 1.5e+07 4.95E+06 0.008 -2829 -2828 -2828 -2828 -2828 S12833 1.5e+07 4.95E+06 0.013 -11176 -11176 -11175 -11177 -11176 S12850 1.5e+07 4.95E+06 0.003 -2669 -2669 -2670 -2670 -2669 S12853 1.5e+07 4.95E+06 0.012 -1689 -1689 -1689 -1688 -1689 S12870 1.5e+07 4.95E+06 0.005 -853 -854 -852 -853 -853 S12929 1.5e+07 4.95E+06 0.005 -2571 -2573 -2574 -2571 -2571 S12933 1.5e+07 4.95E+06 0.009 -3671 -3670 -3671 -3670 -3670 S13029 1.5e+07 4.95E+06 0.006 -902 -902 -902 -902 -902 S13043 1.5e+07 4.95E+06 0.004 -4109 -4109 -4109 -4109 -4109 S1330 1.5e+07 4.95E+06 0.013 -6953 -6955 -6954 -6954 -6954 S1346 1.5e+07 4.95E+06 0.006 -630 -630 -630 -630 -630 S1398 1.5e+07 4.95E+06 0.005 -516 -516 -516 -516 -516 S1512 1.5e+07 4.95E+06 0.005 -1167 -1167 -1167 -1167 -1167 S1516 1.5e+07 4.95E+06 0.007 -1268 -1269 -1269 -1269 -1269 S1518 1.5e+07 4.95E+06 0.004 -3741 -3739 -3740 -3737 -3738 S1580 1.5e+07 4.95E+06 0.014 -2286 -2285 -2285 -2286 -2285 S1666 1.5e+07 4.95E+06 0.009 -1889 -1889 -1890 -1890 -1889 S1782 1.5e+07 4.95E+06 0.007 -883 -883 -883 -883 -883 S1816 1.5e+07 4.95E+06 0.006 -2512 -2512 -2512 -2513 -2512 S1820 1.5e+07 4.95E+06 0.008 -1615 -1615 -1616 -1615 -1615 S1875 1.5e+07 4.95E+06 0.006 -1096 -1097 -1097 -1097 -1097 S1884 1.5e+07 4.95E+06 0.009 -874 -874 -874 -874 -874 S1986 1.5e+07 4.95E+06 0.005 -1400 -1401 -1400 -1400 -1400 S1988 1.5e+07 4.95E+06 0.006 -1218 -1218 -1219 -1218 -1218 S2128 1.5e+07 4.95E+06 0.008 -9220 -9220 -9220 -9222 -9220 S2132 1.5e+07 4.95E+06 0.006 -807 -807 -807 -807 -807 S2158 1.5e+07 4.95E+06 0.007 -1666 -1666 -1666 -1667 -1666

80 Appendix 2.2. Continued. Lognormal Rates Model, Uniform Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S2179 1.5e+07 4.95E+06 0.009 -3647 -3648 -3648 -3648 -3648 S2229 1.5e+07 4.95E+06 0.012 -2934 -2935 -2937 -2935 -2935 S856 1.5e+07 4.95E+06 0.005 -2112 -2112 -2112 -2112 -2112 S860 1.5e+07 4.95E+06 0.010 -1775 -1775 -1774 -1775 -1775 S874 1.5e+07 4.95E+06 0.019 -2690 -2690 -2689 -2690 -2690 S920 1.5e+07 4.95E+06 0.012 -3660 -3661 -3661 -3662 -3661 S95 1.5e+07 4.95E+06 0.006 -771 -771 -771 -771 -771 S953 1.5e+07 4.95E+06 0.007 -3808 -3809 -3809 -3808 -3808 S9941 1.5e+07 4.95E+06 0.004 -1628 -1629 -1628 -1629 -1629 S9957 1.5e+07 4.95E+06 0.010 -7306 -7309 -7302 -7308 -7304 S997 1.5e+07 4.95E+06 0.013 -3131 -3132 -3131 -3132 -3132 S9993 1.5e+07 4.95E+06 0.010 -3388 -3388 -3387 -3390 -3388 Car2012 1.5e+07 4.95E+06 0.008 -4347 -4348 -4348 -4350 -4348

Gamma Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S10029 1.5e+07 4.95E+06 0.006 -1687 -1688 -1687 -1687 -1687 S10045 2.88E+07 9.50E+06 0.019 -6406 -6399 -6405 -6404 -6400 S10081 1.5e+07 4.95E+06 0.006 -992 -992 -992 -992 -992 S10108 1.5e+07 4.95E+06 0.003 -2087 -2087 -2087 -2087 -2087 S10135 1.52E+07 5.00E+06 0.011 -7106 -7111 -7112 -7114 -7108 S10146 1.5e+07 4.95E+06 0.006 -2086 -2086 -2086 -2086 -2086 S10204 1.5e+07 4.95E+06 0.006 -704 -704 -704 -704 -704 S1023 1.5e+07 4.95E+06 0.006 -868 -868 -868 -867 -868 S10249 1.5e+07 4.95E+06 0.005 -1101 -1101 -1101 -1100 -1101 S10252 1.5e+07 4.95E+06 0.010 -1549 -1549 -1550 -1550 -1549 S10265 1.67E+07 5.49E+06 0.011 -5281 -5281 -5281 -5283 -5281 S10466 1.5e+07 4.95E+06 0.005 -811 -811 -812 -812 -811 S1050 1.5e+07 4.95E+06 0.007 -1485 -1486 -1488 -1486 -1486 S10512 1.5e+07 4.95E+06 0.011 -1694 -1693 -1694 -1693 -1694 S10667 1.94E+07 6.39E+06 0.011 -4152 -4151 -4152 -4155 -4152 S107 1.5e+07 4.95E+06 0.005 -1315 -1315 -1315 -1315 -1315 S108 1.5e+07 4.95E+06 0.006 -394 -394 -394 -395 -394 S10868 1.5e+07 4.95E+06 0.008 -995 -997 -995 -996 -996 S10905 1.5e+07 4.95E+06 0.004 -1647 -1647 -1647 -1647 -1647 S10908 1.5e+07 4.95E+06 0.012 -7067 -7069 -7067 -7066 -7067

81 Appendix 2.2. Continued. Gamma Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S11114 1.5e+07 4.95E+06 0.006 -2884 -2884 -2884 -2883 -2884 S1116 1.5e+07 4.95E+06 0.007 -1940 -1940 -1939 -1940 -1939 S1117 1.5e+07 4.95E+06 0.008 -1858 -1859 -1859 -1859 -1859 S112 1.5e+07 4.95E+06 0.006 -366 -366 -366 -366 -366 S1121 1.5e+07 4.95E+06 0.021 -2680 -2679 -2680 -2680 -2680 S1125 1.5e+07 4.95E+06 0.006 -414 -414 -414 -414 -414 S11258 2.24E+07 7.38E+06 0.014 -2384 -2388 -2386 -2386 -2385 S11368 1.5e+07 4.95E+06 0.004 -228 -228 -228 -228 -228 S11387 3.69E+07 1.22E+07 0.013 -2403 -2396 -2407 -2402 -2397 S114 1.5e+07 4.95E+06 0.008 -1278 -1278 -1279 -1279 -1278 S11810 1.5e+07 4.95E+06 0.006 -1622 -1622 -1623 -1624 -1622 S11957 1.5e+07 4.95E+06 0.049 -15399 -15396 -15399 -15397 -15397 S11965 1.5e+07 4.95E+06 0.005 -4486 -4485 -4484 -4485 -4485 S12060 1.5e+07 4.95E+06 0.007 -1081 -1081 -1081 -1081 -1081 S12319 1.5e+07 4.95E+06 0.008 -2952 -2953 -2952 -2951 -2952 S12447 1.5e+07 4.95E+06 0.005 -716 -716 -717 -716 -716 S12832 1.5e+07 4.95E+06 0.010 -2833 -2830 -2831 -2831 -2831 S12833 1.5e+07 4.95E+06 0.013 -11166 -11165 -11164 -11164 -11164 S12850 1.5e+07 4.95E+06 0.004 -2672 -2674 -2674 -2673 -2673 S12853 1.5e+07 4.95E+06 0.011 -1688 -1685 -1687 -1686 -1686 S12870 1.5e+07 4.95E+06 0.005 -861 -862 -862 -863 -862 S12929 1.5e+07 4.95E+06 0.006 -2573 -2573 -2574 -2573 -2573 S12933 1.5e+07 4.95E+06 0.007 -3668 -3663 -3665 -3666 -3664 S13029 1.5e+07 4.95E+06 0.005 -899 -898 -899 -898 -899 S13043 1.5e+07 4.95E+06 0.004 -4107 -4108 -4106 -4107 -4107 S1330 1.5e+07 4.95E+06 0.009 -6948 -6950 -6949 -6949 -6949 S1346 1.5e+07 4.95E+06 0.007 -627 -627 -627 -627 -627 S1398 1.5e+07 4.95E+06 0.006 -512 -512 -512 -512 -512 S1512 1.5e+07 4.95E+06 0.005 -1168 -1167 -1168 -1168 -1168 S1516 1.5e+07 4.95E+06 0.007 -1265 -1265 -1265 -1265 -1265 S1518 1.5e+07 4.95E+06 0.004 -3738 -3738 -3738 -3739 -3738 S1580 1.5e+07 4.95E+06 0.012 -2282 -2282 -2284 -2282 -2282 S1666 1.5e+07 4.95E+06 0.007 -1886 -1886 -1886 -1886 -1886 S1782 1.5e+07 4.95E+06 0.007 -878 -878 -878 -878 -878 S1816 1.5e+07 4.95E+06 0.007 -2508 -2509 -2509 -2508 -2508 S1820 1.5e+07 4.95E+06 0.009 -1613 -1612 -1611 -1613 -1612 S1875 1.5e+07 4.95E+06 0.006 -1096 -1096 -1096 -1096 -1096 S1884 1.5e+07 4.95E+06 0.007 -871 -871 -871 -870 -871

82 Appendix 2.2. Continued. Gamma Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S1986 1.5e+07 4.95E+06 0.006 -1397 -1397 -1396 -1397 -1397 S1988 1.5e+07 4.95E+06 0.007 -1214 -1214 -1214 -1214 -1214 S2128 1.5e+07 4.95E+06 0.008 -9213 -9211 -9210 -9211 -9211 S2132 1.5e+07 4.95E+06 0.007 -807 -807 -807 -808 -807 S2158 1.5e+07 4.95E+06 0.006 -1664 -1665 -1665 -1665 -1665 S2179 1.5e+07 4.95E+06 0.010 -3645 -3646 -3645 -3644 -3645 S2229 1.5e+07 4.95E+06 0.010 -2929 -2930 -2929 -2931 -2929 S856 1.5e+07 4.95E+06 0.005 -2108 -2108 -2108 -2108 -2108 S860 1.5e+07 4.95E+06 0.007 -1772 -1772 -1772 -1772 -1772 S874 1.5e+07 4.95E+06 0.014 -2687 -2687 -2687 -2687 -2687 S920 1.5e+07 4.95E+06 0.010 -3657 -3659 -3657 -3658 -3657 S95 1.5e+07 4.95E+06 0.006 -767 -767 -767 -767 -767 S953 1.5e+07 4.95E+06 0.007 -3803 -3803 -3802 -3803 -3803 S9941 1.5e+07 4.95E+06 0.005 -1626 -1626 -1626 -1626 -1626 S9957 1.5e+07 4.95E+06 0.009 -7303 -7305 -7305 -7303 -7304 S997 1.5e+07 4.95E+06 0.014 -3129 -3129 -3129 -3129 -3129 S9993 1.5e+07 4.95E+06 0.008 -3385 -3389 -3386 -3386 -3386 Car2012 1.5e+07 4.95E+06 0.008 -4346 -4346 -4345 -4344 -4345

Lognormal Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S10029 1.5e+07 4.95E+06 0.005 -1688 -1688 -1687 -1688 -1688 S10045 2.88E+07 9.50E+06 0.024 -6405 -6399 -6403 -6404 -6400 S10081 1.5e+07 4.95E+06 0.007 -992 -992 -991 -991 -991 S10108 1.5e+07 4.95E+06 0.003 -2085 -2086 -2086 -2086 -2086 S10135 1.52E+07 5.00E+06 0.011 -7105 -7113 -7105 -7107 -7106 S10146 1.5e+07 4.95E+06 0.006 -2086 -2086 -2086 -2085 -2086 S10204 1.5e+07 4.95E+06 0.006 -702 -702 -702 -702 -702 S1023 1.5e+07 4.95E+06 0.006 -867 -867 -867 -867 -867 S10249 1.5e+07 4.95E+06 0.004 -1098 -1098 -1098 -1098 -1098 S10252 1.5e+07 4.95E+06 0.009 -1550 -1549 -1549 -1549 -1550 S10265 1.67E+07 5.49E+06 0.014 -5279 -5278 -5277 -5277 -5278 S10466 1.5e+07 4.95E+06 0.006 -810 -810 -810 -810 -810 S1050 1.5e+07 4.95E+06 0.009 -1483 -1483 -1483 -1484 -1483 S10512 1.5e+07 4.95E+06 0.014 -1695 -1695 -1694 -1694 -1694 S10667 1.94E+07 6.39E+06 0.013 -4158 -4158 -4156 -4157 -4157

83 Appendix 2.2. Continued. Lognormal Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S107 1.5e+07 4.95E+06 0.006 -1316 -1316 -1316 -1316 -1316 S108 1.5e+07 4.95E+06 0.006 -394 -393 -393 -393 -393 S10868 1.5e+07 4.95E+06 0.007 -993 -993 -994 -994 -993 S10905 1.5e+07 4.95E+06 0.005 -1644 -1644 -1644 -1644 -1644 S10908 1.5e+07 4.95E+06 0.007 -7069 -7069 -7069 -7070 -7069 S11114 1.5e+07 4.95E+06 0.006 -2881 -2880 -2881 -2881 -2881 S1116 1.5e+07 4.95E+06 0.007 -1940 -1940 -1939 -1938 -1939 S1117 1.5e+07 4.95E+06 0.009 -1858 -1859 -1858 -1858 -1858 S112 1.5e+07 4.95E+06 0.006 -365 -365 -365 -365 -365 S1121 1.5e+07 4.95E+06 0.022 -2679 -2679 -2679 -2679 -2679 S1125 1.5e+07 4.95E+06 0.006 -413 -413 -413 -413 -413 S11258 2.24E+07 7.38E+06 0.012 -2379 -2381 -2384 -2379 -2379 S11368 1.5e+07 4.95E+06 0.006 -227 -227 -227 -227 -227 S11387 3.69E+07 1.22E+07 0.013 -2396 -2390 -2405 -2400 -2392 S114 1.5e+07 4.95E+06 0.008 -1274 -1274 -1275 -1274 -1274 S11810 1.5e+07 4.95E+06 0.006 -1617 -1618 -1616 -1618 -1617 S11957 1.5e+07 4.95E+06 0.053 -15403 -15409 -15400 -15402 -15402 S11965 1.5e+07 4.95E+06 0.005 -4481 -4482 -4482 -4483 -4482 S12060 1.5e+07 4.95E+06 0.007 -1082 -1082 -1082 -1081 -1082 S12319 1.5e+07 4.95E+06 0.008 -2951 -2951 -2951 -2951 -2951 S12447 1.5e+07 4.95E+06 0.006 -716 -716 -716 -716 -716 S12832 1.5e+07 4.95E+06 0.011 -2825 -2825 -2825 -2825 -2825 S12833 1.5e+07 4.95E+06 0.012 -11172 -11173 -11174 -11170 -11171 S12850 1.5e+07 4.95E+06 0.004 -2667 -2666 -2666 -2665 -2666 S12853 1.5e+07 4.95E+06 0.012 -1687 -1685 -1686 -1686 -1686 S12870 1.5e+07 4.95E+06 0.005 -856 -855 -857 -856 -856 S12929 1.5e+07 4.95E+06 0.006 -2569 -2569 -2569 -2568 -2569 S12933 1.5e+07 4.95E+06 0.008 -3664 -3666 -3666 -3664 -3665 S13029 1.5e+07 4.95E+06 0.004 -897 -897 -897 -897 -897 S13043 1.5e+07 4.95E+06 0.004 -4104 -4105 -4105 -4105 -4105 S1330 1.5e+07 4.95E+06 0.014 -6950 -6950 -6949 -6950 -6950 S1346 1.5e+07 4.95E+06 0.006 -627 -627 -627 -627 -627 S1398 1.5e+07 4.95E+06 0.005 -511 -511 -511 -511 -511 S1512 1.5e+07 4.95E+06 0.006 -1165 -1165 -1165 -1165 -1165 S1516 1.5e+07 4.95E+06 0.006 -1263 -1263 -1263 -1264 -1263 S1518 1.5e+07 4.95E+06 0.004 -3737 -3734 -3735 -3736 -3735 S1580 1.5e+07 4.95E+06 0.014 -2281 -2281 -2283 -2282 -2282 S1666 1.5e+07 4.95E+06 0.007 -1885 -1885 -1885 -1885 -1885

84 Appendix 2.2. Continued. Lognormal Rates, Exponential Distribution Shape Prior Treebase # Post. StudyID # Gen. Gen. ASDSF Run1 Run2 Run3 Run4 Mean S1782 1.5e+07 4.95E+06 0.006 -878 -878 -878 -878 -878 S1816 1.5e+07 4.95E+06 0.006 -2510 -2509 -2509 -2509 -2509 S1820 1.5e+07 4.95E+06 0.009 -1611 -1612 -1612 -1611 -1611 S1875 1.5e+07 4.95E+06 0.007 -1095 -1095 -1095 -1094 -1095 S1884 1.5e+07 4.95E+06 0.007 -869 -869 -869 -869 -869 S1986 1.5e+07 4.95E+06 0.006 -1397 -1397 -1397 -1398 -1397 S1988 1.5e+07 4.95E+06 0.007 -1214 -1214 -1214 -1213 -1214 S2128 1.5e+07 4.95E+06 0.008 -9215 -9217 -9216 -9217 -9216 S2132 1.5e+07 4.95E+06 0.006 -805 -804 -804 -803 -804 S2158 1.5e+07 4.95E+06 0.006 -1662 -1662 -1663 -1663 -1663 S2179 1.5e+07 4.95E+06 0.009 -3644 -3644 -3644 -3645 -3644 S2229 1.5e+07 4.95E+06 0.015 -2933 -2932 -2930 -2931 -2931 S856 1.5e+07 4.95E+06 0.005 -2107 -2107 -2107 -2107 -2107 S860 1.5e+07 4.95E+06 0.010 -1771 -1770 -1770 -1771 -1770 S874 1.5e+07 4.95E+06 0.012 -2686 -2686 -2687 -2687 -2687 S920 1.5e+07 4.95E+06 0.008 -3657 -3657 -3656 -3657 -3657 S95 1.5e+07 4.95E+06 0.007 -767 -767 -767 -767 -767 S953 1.5e+07 4.95E+06 0.008 -3803 -3805 -3805 -3807 -3805 S9941 1.5e+07 4.95E+06 0.004 -1625 -1625 -1625 -1625 -1625 S9957 1.5e+07 4.95E+06 0.018 -7304 -7305 -7308 -7305 -7305 S997 1.5e+07 4.95E+06 0.011 -3128 -3128 -3128 -3127 -3128 S9993 1.5e+07 4.95E+06 0.010 -3385 -3384 -3385 -3382 -3383 Car2012 1.5e+07 4.95E+06 0.007 -4343 -4344 -4343 -4344 -4344

85 CONNECTING TEXT

In the second chapter of this thesis, I used a diverse sample of matrices of discrete morphological characters to examine the prevalence of rate heterogeneity among characters in phylogenetic analysis. I found evidence for significant among-character rate heterogeneity in most data sets and also found weak support for Wagner's (2012) hypothesis that the lognormal distribution is more appropriate to model rate variation. In the balance of this thesis, I examined absolute rates of evolution of discrete morphological characters in a phylogenetic context and extended current methods to better account for phylogenetic and divergence time uncertainty. I integrated the results of the second chapter by testing and applying models of among-character rate-heterogeneity in these absolute rate calculations. I selected modern birds as a case study because there is a large and complete data set of discrete morphological characters available (Livezey and Zusi, 2006; 2007) and because the phylogeny and timescale of avian evolution is uncertain and can not be reduced to a single dated phylogeny with any confidence. A key requirement to estimate absolute rates of evolution is a dated phylogeny with estimates of branch durations so that the quantity of morphological evolution, the product of rate and time on each branch, can be deconfounded. In the following chapter, I used Bayesian divergence time dating to estimate a distribution of 10 000 dated phylogenies for a broad sample of extant birds, including those in the Livezey and Zusi (2006; 2007) data set to provide a temporal and phylogenetic framework for rates analysis.

86 CHAPTER 3

EMBRACING UNCERTAINTY: A DISTRIBUTION OF EVOLUTIONARY

TIMESCALES FOR MODERN BIRD LINEAGES

The phylogeny and evolutionary timescale of extant birds (crown group Neornithes), and the interrelationships of neoavian orders in particular, have been controversial (Sibley and Alquist, 1990; Poe and Chubb, 2004; Livezey and Zusi, 2007; Hackett et al., 2008; Brown and van Tuinen, 2011; Livezey, 2011). There is significant discordance between interrelationships estimated using morphological characters (e.g. Livezey and Zusi, 2007) and those derived from molecular sequences (e.g. Ericson et al., 2006; Hackett et al., 2008; McCormack et al., 2013). Although some authors have argued that the basal divergences of Neoaves represent a hard polytomy that may never be resolved (e.g. Poe and Chubb, 2004), progress due to extensive sequencing of independent nuclear loci (e.g. Hackett et al., 2008; Wang et al., 2012; Smith et al., 2013; McCormack et al., 2013) has been made. However, uncertainty with respect to many neoavian relationships remains among molecular studies. This is likely because the radiation at the base of Neoaves was short, probably within 10 Ma (e.g. Ericson et al., 2006) with some branches possibly less than than 1 Ma (Braun et al., 2011), which left little time for molecular synapomorphies to accrue; this is reflected in very short internodal distance at the base of Neoaves in phylograms (e.g. Hackett et al., 2008). This is compounded by putatively ancient lineages with relatively small crown groups (e.g. Coliiformes, Trogoniformes, Musophagiformes, Phaethontidae, Pteroclididae; all clade names after Hackett et al., 2008 unless otherwise specified) and enigmatic monotypic taxa whose relationships resist definitive classification (e.g. Opisthocomus, Leptosomus). Using a mixture of fast evolving introns and more slowly evolving exons and a steadily increasing sample of independent loci, certain higher-level relationships are emerging as tentatively

87 supported, such as for example: Passeriformes+Psittaciformes (Hackett et al., 2008; Suh et al., 2011; McCormack et al., 2013), Podicipediformes+Phoenicopteriformes (e.g. Chubb, 2004; Mayr, 2004; Ericson et al., 2006; Hackett et al., 2008) and the paraphyly of traditional ratites (Harshman et al., 2008; Hackett et al., 2008; Phillips et al., 2010; Haddrath and Baker, 2012; Smith et al., 2013), among others. However, other relationships resist classification and argue that divergence time and evolutionary comparative studies must consider a range of hypotheses (Jetz et al., 2012). Penny and Phillips (2004) proposed five models for the divergence of birds and its relationship to the Cretaceous-Paleogene (K–Pg) mass extinction: model A) modern birds are descendant from a single lineage of birds that crosses the K–Pg boundary (e.g. Fedducia's [2003] 'big bang'); B) modern birds are descendant from a handful of lineages that cross the K–Pg and all diversification occurs in the Paleogene; C) many ecologically diverse stem lineages cross the K–Pg but diversification is strictly Paleogene; D) most lineages cross the K–Pg, significant ecological diversification in the Cretaceous but crown groups diversify in the Paleogene and E) all orders and most crown groups originate and diversify in the Cretaceous and cross the K–Pg boundary. Although the presence of crown-group neornithean taxa in the Cretaceous is beyond reasonable doubt and consequently, several crown group lineages must have been present in at least the latest Cretaceous, only the single taxon Vegavis iiai is definitively considered to be a representative of an extant crown group (Clarke et al., 2005). Among avian divergence time studies, most support models C, D or E and differ on whether there was extensive divergence of orders before (e.g. Brown et al., 2008) or after (e.g. Ericson et al., 2006) the K–Pg boundary. An empirical study of the Cretaceous fossil record of birds concluded that its fragmentary nature means that the evolution of multiple groups of modern birds in the Cretaceous is possible (Brocklehurst et al., 2012), although the current fossil record is more consistent with a mostly but not exclusively Paleogene origin of neoavian orders (e.g. Dyke and Gardiner, 2011). Thus, significant uncertainty remains in the phylogeny and timescale of neornithean evolution.

88 Combined “total-evidence” morphological and molecular approaches to phylogeny (e.g. Nylander et al., 2004), and more recently divergence time estimation (Pyron et al., 2011; Ronquist et al., 2012a; Wood et al., 2013) offer a new approach to the traditionally separate molecular and morphological analyses of avian evolution (but see e.g. Ksepka and Thomas [2012] for a smaller scale combined phylogenetic analysis). Avian morphology based on the anatomical framework created by Baumel (1993) was exhaustively characterized by Livezey and Zusi (2006; 2007) for ~150 neornithean genera and nearly 2954 discrete morphological characters. This provides an ideal framework where fossil taxa might be coded for all possible characters and included as tip-dated terminal taxa in a “total-evidence” divergence dating analysis for modern birds (sensu Ronquist et al., 2012a; see Laurin, 2012 for a recent review of similar approaches). However, these analyses and also the use of this rich morphological resource to explore macroevolutionary patterns in avian evolution have until very recently (i.e. Jetz et al., 2012), been hampered by the lack of molecular phylogeny and timescale for these genera and by patchy genetic sampling among the species examined by Livezey and Zusi (2006; 2007). A distribution of chronograms for crown group Neornithes has recently been published using a novel phylogenetic methodology to combine multiple sub- analyses into a composite distribution of chronograms (Jetz et al., 2012). The Jetz et al. method to assemble the final distribution of chronograms is complex, and Pagel (2012) criticized Jetz et al.'s reliance on fixing most of the structure of the backbone phylogeny a priori using phylogenetic constraints derived from Ericson et al. (2006) or Hackett et al. (2008)'s phylogenetic analyses. Although the Jetz et al. analysis is an innovative approach with great value in its unprecedented coverage and systematic approach, any phylogenetic analysis attempting to estimate relationships and divergence times for ~10 000 species, including ~3 000 without molecular sequence data will necessarily sacrifice some accuracy for completeness. Here, an alternative approach with a focus on estimating a plausible distribution of divergence times using a more traditional supermatrix analysis of existing molecular sequence data was adopted. Combining two major analyses of

89 extant avian phylogeny (e.g. Ericson et al., 2006 and Hackett et al., 2008), augmenting these using sequences retrieved from Genbank and fixing fewer interrelationships a priori, a Bayesian divergence time dating analysis was conducted to generate a distribution of 10 000 chronograms representing a distribution of phylogenetic hypotheses that incorporate uncertainty in both crown avian relationships and in divergence times. The goal of this analysis was not to resolve long standing questions about inter- ordinal relationships, which probably require more extensive gene sampling and consideration of rare genomic changes (Braun et al., 2011; Haddrath and Baker, 2012; McCormack et al., 2013; Yuri et al., 2013). Similarly, accurate and precise dating of the timescale of neornithean evolution may also require such molecular sampling to obtain sufficient accuracy and precision to definitively close the "rocks vs. clocks" debate (e.g. dos Reis et al., 2012), if such a debate is even relevant (see Brown et al., 2008). Rather, the objective of this analysis is to provide a plausible, although potentially imprecise, distribution of relationships and divergence times for a sample of extant species within a single phlyogenetic analysis (contra Jetz et al., 2012). The taxa sampled were also chosen specifically to provide generic-level coverage of the comprehensive Livezey and Zusi (2006; 2007) data set of avian morphology to facilitate both future analyses of morphological evolution and exploration of “total-evidence” partitioned phylogenetic topology and divergence time analyses for Crown Neornithes.

MATERIALS AND METHODS A number of recent studies have shown that supermatrices, even those with a large proportion of missing data, can still provide phylogenetic evidence (Wiens and Morill, 2011; Pyron and Wiens, 2011; Hinchliff and Roalson, 2013 but see also Lemmon, et al. 2009 for a counter-argument). Alternatively, other studies have argued for the use of composite taxa as an alternative approach to a species level supermatrix, which simultaneously eases computation burden and reduces the quantity of missing data (Campbell and LaPointe, 2009; Campbell and

90 LaPointe, 2011; Pyron et al., 2013 but see Malia et al., 2003 for a counter- argument). Composite taxa were also used in the avian divergence time analyses of Ericson et al. (2006) and Haddrath and Baker (2012). Here, both approaches were applied. First, a species-level supermatrix was created and phylogeny was estimated using maximum likelihood. This analysis was then used to collapse confirmed monophyletic genera into a supermatrix of composite taxa for further phylogenetic and divergence time analysis.

Sequence Data A supermatrix was assembled for 447 extant neornithean species and three crocodylian outgroups (see Supplementary Methodology for full details). Species were selected based on four criteria: a) inclusion in the phylogenetic analyses of Ericson et al. (2006) and Hackett et al. (2008), b) available mitochondrial genome, c) congeneric inclusion in Livezey and Zusi's (2006; 2007) morphological analysis and/or d) other either phylogenetically or historically important taxa. An operational taxonomy of the selected species was assembled and species names were harmonized between all data sources (Appendix 3.1; not intended as a comprehensive review of the taxonomy of these species). To construct the supermatrix, the Ericson et al. (2006) and Hackett et al. (2008) alignments were first combined for the loci that overlapped and then these and other nuclear alignments from the Ericson et al. (2006) were then augmented using all sequences available on GenBank for the taxa of interest (full details in Supplementary Methodology and accession numbers in Appendix 3.2). Finally, NCBI GenBank was queried to collect sequence data for the following mitochondrial genes: 12S rRNA, CYTB, COX1 and ND2. All protein coding alignments were adjusted to be inframe to permit the use of codon position models (see Supplementary Methodology). The final species level supermatrix consisted of approximately 39 Kb, with an average of 68% missing data per taxon, although some taxa had > 95% missing data. A generic-level supermatrix was also generated by collapsing species into genus level composite (or chimeric) taxa. Only genera that were confirmed to be monophyletic in the species level 91 phylogenetic analysis were combined, although in five cases species putatively assigned to different genera were combined to form monophyletic chimeric taxa (see details and list in Supplementary Methodology and Appendix 3.3). To increase the conservativeness of this analysis, when composite taxa were created, any overlapping gene sequences were combined and conflicting alignment positions were represented using nucleotide ambiguity codes. The generic-level supermatrix consisted of 314 genera and had a mean of 59% missing data and 0.2% ambiguous data per taxon. Alignments are available from the authors upon request.

Phylogenetic Analysis Several partitioning schemes were investigated for the full species-level supermatrix using maximum likelihood analysis with RAxML and compared using Bayesian Information Criterion (BIC) scores (Schwarz, 1978; Stamatakis, 2006; Table 3.1; full details in Supplementary Methodology). The optimal scheme considered was to partition all concatenated exons from Hackett et al., RAG1 and the concatenated protein-coding mtDNA into all three codon positions each, and to consider all introns and the 12S loci as separate partitions. Although other schemes that further partitioned exons by loci were possible, this would create many very small partitions from the partial exons included in the Hackett et al., (2008) data set that might have led to unstable model parameter estimates. RAxML v7.2.8 was used to estimate phylogeny using 512 independent tree searches and 512 bootstrap replicates using the GTRCAT model, followed by optimization under the GTRGAMMA model for the species-level matrix (Stamatakis, 2006). For the generic-level supermatrix, 1024 bootstrap replicates and independent searches for the generic level matrix were performed (full details in Supplementary Methodology). The analysis was conducted using the CLUMEQ/Colosse and CLUMEQ/Guillimin computer clusters (total ~5 core- years of computation). Recovered relationships were summarized using the maximum likelihood estimates of topology (TMLE), annotated with bootstrap support values for each matrix plotted using FigTree v1.3.1 (Rambaut, 2008).

92 Divergence Time Analysis Bayesian divergence time dating is computationally intensive and attempts to use the complete species or genus-level supermatrix were intractable, even using significant high-performance computing (HPC) resources. To increase computation tractability, divergence time dating was conducted with a subset of the generic-level supermatrix consisting of loci chosen to minimize missing data: 12S, CYTB, COX1, RAG1, MYC, MB, ODC1, and FGB (full details in Supplementary Methodology). Excluding the crocodylian outgroup and three fossil taxa, the reduced generic-level supermatrix consisted of 310 taxa and ~10 Kb, with an average 42% missing data. Divergence time dating was conducted using BEAST v1.7.5 using an uncorrelated relaxed, lognormally-distributed molecular clock, with no rate autocorrelation (Drummond et al., 2006; Drummond et al., 2012). BEAST was selected over other divergence time dating software packages because it simultaneously estimates divergence times and topologies and allows flexible model and prior specification: these are essential in the case of modern birds whose phylogeny is not fully resolved. Furthermore, recent divergence time studies of modern birds have also employed BEAST (e.g. Brown et al., 2008; Haddrath and Baker, 2012; Jetz et al., 2012). A total of 23 fossil calibrations were used to provide hard minimum and soft maximum constraints for the divergence time analysis (Fig. 3.1; Table 3.2; see Supplementary Methodology for fossil choice and justification). It is now recognized that fossil calibration densities, however carefully chosen (e.g. Ho, 2007; Benton et al., 2009; Ho and Phillips, 2009) are inputs to divergence time dating software which are then combined with an overall tree prior (a Yule process in this case) to construct marginal priors on node divergence times. This may lead to mismatch between calibration densities and actual marginal priors (Heled and Drummond, 2011), particularly in software that uses multiplicative prior construction (e.g. BEAST; Heled and Drummond, 2011; dos Reis et al., 2012). Therefore, marginal priors were estimated by running the analysis without the data, and in the case of the root calibration, this led to the adjustment of the root

93 calibration density (see Supplementary Methodology). BEAST XML control files are available from the authors upon request. Nodes receiving greater than 95% bootstrap support from the genus-level phylogenetic analysis were constrained (see Fig. 3.1) to reduce the size of topology space to be explored. Eight independent runs of at least 100 million generations, sampling every 2000 generations, were performed using the CLUMEQ/Guillimin HPC facility which required approximately three weeks and 1.5 core-years of computation (full details in Supplementary Methodology). Convergence and mixing of runs was examined visually using Tracer v1.5; results from the eight runs were then combined after discarding a burnin of 10 million generations from each run and sufficient sampling was assessed by ensuring that parameters had effective sample sizes (ESS) greater than 200 using Tracer v1.5, which was the case for all parameters (Drummond et al., 2006; Rambaut and Drummond, 2007). Samples from the combined posterior distribution of chronograms were then thinned to produce a computationally tractable sample from the posterior distribution of 10 000 dated trees for further analysis. The marginal posterior distribution of trees was summarized using the maximum clade credibility tree with node heights placed at the mean of the respective marginal posterior distribution of node divergence times using treeannotater from the BEAST package. The MCC tree was plotted using the phyloch package in R (Heibl, 2008; R Core Team, 2012).

Comparisons with the Jetz et al. (2012) Distribution of Chronograms Jetz et al. (2012) produced a distribution of 10 000 chronograms which, although not extensively discussed in their analysis, constitutes a hypothesis of divergence times and relationships for extant birds. To compare with the present study, the full distribution of chronograms based on the Hackett et al., (2008) topological constraints was downloaded from www.birdtree.org. This tree distribution was then summarized using treeannotator to determine the maximum clade credibility tree on the full chronogram (9993 species) tree distribution. This phylogeny was then annotated with clade posterior probabilities, mean node

94 heights and 95% highest probability density (HPD) intervals for each node. To compare divergence dates between a sample of lower level divergences, both the Jetz et al. (2012) and this analysis' chronogram distributions were then pruned to the taxonomic sampling of the Livezey and Zusi's (2007) morphological data set and compared for reciprocally monophyletic clades using a modified version of an R script published by Ronquist et al. (2012a; available upon request).

RESULTS AND DISCUSSION Phylogenetic Analysis The results of the maximum likelihood analysis of the generic-level supermatrix are shown in Figure 3.1. Results from the species-level analysis are included as Supplementary Figure 3.1. In both cases, the maximum likelihood estimate of topology (TMLE) was very similar in overall structure to the topology recovered by Hackett et al. (2008) but with several minor differences, notably among clades that received very poor bootstrap support in both analyses (Fig. 3.1; Suppl. Fig. 3.1). Interestingly, the first basal divergence in Neoaves was different between Hackett et al. (2008) and this analysis, although this area of the tree has low bootstrap support in both analyses (< 0.5; Fig. 3.1: arrow 1). The topologies recovered by the species-level and generic-level analyses were nearly identical in terms of major groupings and differed only on the arrangement of some nodes with poor bootstrap support (e.g. < 0.75). Similarity to the Hackett et al. (2008) topology was not surprising as that data set's gene sampling dominated the supermatrix used here. Analysis of 1024 bootstrap replicates for the generic-level supermatrix revealed poor support for most higher-level relationships and a similar level of support relative to the Hackett et al. (2008) analysis. Noticeably, although the recovered topology was nearly identical, support for the clade uniting Cuculiformes+”Gruiformes”+Otididae with the Musophagiformes+Water Birds (Fig. 3.1: arrow 2; clade names were used operationally and defined in Fig. 3.2; clade J and in Hackett et al., 2008) was considerably weaker in the species-level analysis relative to Hackett et al. (.81 vs. .45 bootstrap support). However, within

95 this group, the sister relationship between Musophagiformes and the water bird clade was stronger in this analysis (Fig. 3.1: arrow 3; 0.73 vs. < 0.5 in Hackett et al., 2008). This analysis also recovered weaker support for the clade uniting Charadriiformes and “Land Birds” (Fig. 3.1: arrow 4, clade G in Hackett et al., 2008). Comprehensive empirical assessment of relative clade supports was tentative given the different methods used (GARLI vs. RAxML) and was beyond the scope of this paper. In summary, the phylogeny presented here placed approximately ~130 genera into the framework presented by Hackett et al. (2008) but did not significantly increase or decrease resolution with several exceptions (see above); readers are referred to Hackett et al. (2008) for a more detailed discussion of of the estimated relationships and implications for avian taxonomy.

Divergence Time Analysis The mean of the posterior distribution of lognormal branch rate distribution's shape parameter (σ) was 0.499 (0.457—0.543 95% HPD), which indicates significant among-lineage rate heterogeneity and rejected the molecular clock (Brown and Yang, 2011), consistent with other dating analyses in birds (e.g. Pereira and Baker, 2006; Brown et al., 2008; Haddrath and Baker, 2012). The posterior mean of covariance between ancestor and descendant branches was 0.155 (0.007—0.24; 95% HPD) which indicated the presence of weak rate autocorrelation between ancestors and descendant branch rates (Drummond et al., 2006). Rate autocorrelation has not been observed in other relaxed clock studies of birds (e.g. Brown et al., 2008; Brown and van Tuinen, 2011) suggesting that as taxonomic sampling increases, the signature of heritable molecular rates becomes more recoverable (e.g. Drummond et al., 2006). Further, this suggests that future large-scale studies of avian divergence times might consider using autocorrelated models of rate evolution (e.g. MCMCTree: Yang, 2007), particularly if an implementation sampling phylogenies becomes available..

The maximum clade credibility chronogram (TMCC) annotated with mean node ages and 95% credibility intervals for node age summarizes a sample of 10 000 chronograms drawn from the posterior of the Bayesian divergence time analysis

96 is included as Figure 3.2. TMCC (Fig. 3.2) differed slightly from the TMLE (Fig. 3.1) estimated by the maximum likelihood analysis: the order of basal-most divergence in TMCC reverted to the relationship recovered by Hackett et al. (2008). Although other major clades were nearly identical in topology, some differences were observed among clades with weak bootstrap support and posterior probabilities (e.g. < 0.70; < 0.95, respectively). In particular, the arrangement of basal-most divergences among the sister groups to the waterbird clade was different (see Fig. 3.1 vs. Fig. 3.2, position of “Gruiformes”, Otididae and Cuculiformes). Opisthocomus was placed closest to the clade including Columbiformes and Apodiformes rather than the water birds and allies and Cariama was recovered as sister to Falconidae rather than Falconidae+Passeriformes+Psittaciformes, a position also weakly recovered by a recent study using 30 independent nuclear loci (Wang et al., 2012). Although many nodes were fixed (see Fig. 3.2), topological differences were not surprising given that the divergence time analysis used an alignment of only ~25% of the original generic-level alignment length and which contained a greater relative contribution of mtDNA, the use of which has recovered highly incongruent phylogenetic hypotheses (e.g. Brown et al., 2008; Pacheco et al., 2011). However, it is also possible that these discrepancies were due to model differences. BEAST allowed the flexible specification of models including, saliently, relative rate parameters to account for among- partition rate heterogeneity, a feature not implemented in RAxML. Nevertheless, the recovered tree distribution, rather than TMCC, which summarized it, maintains the uncertainty inherent in the data for all further analyses (see below). The disparity in topologies recovered underscores potential discordance between the underlying species tree and gene trees, which may be particularly problematic in birds where many divergences occurred over a short time interval (Haddrath and Baker, 2012). This may in turn affect divergence time analysis (McCormack et al., 2011). Approaches exist to estimate and date species trees (e.g. *BEAST; Heled and Drummond, 2010) and dating of an estimated species tree was performed by Haddrath and Baker (2012). Although not attempted here for reasons of computational tractability, the use of species-tree methods may be preferable for

97 future phylogenomic analyses of avian divergence times. Marginal priors, calibrations densities and marginal posteriors on divergence times were compared in Figure 3.3. Credibility intervals (95% highest posterior density—HPD) on the divergence times were quite broad, which reflects significant uncertainty in estimates: this was likely due to a combination of uncertainty in the topology, ambiguous and missing data in the generic-level supermatrix, uncertainty stemming from the variability of rates and uncertainty from the conservative priors used to calibrate the analysis. Examination of marginal priors revealed several cases of mismatch between the specified calibration densities and the actual marginal priors, likely due to the multiplicative construction of marginal priors in BEAST (Heled and Drummond, 2011). This was particularly apparent for the root node (Figs. 3.2, 3.3: node 1) but also for Stem Pici (Fig. 3.3: stem lineage H) and Stem Psittaciformes (Figs. 3.2, 3.3: stem lineage N). To determine if the relatively ancient prior on the root and basal nodes affected the analysis, an identical analysis placing a hard maximum on the root age at 130.2 Ma ago (similar to Haddrath and Baker, 2012; age of the Jehol Biota, see Supplementary Methodology) was performed and yielded nearly identical, and surprisingly, slightly older divergence times on the root and basal nodes (Suppl. Fig. 3.2; Pearson's r = 0.9997). In several cases, fossil calibrations clearly underestimated clade age, for example: the calibration on stem Pteroclididae at 24.52 Ma ago (Figs. 3.2, 3.3: stem lineage R) was quite young given this clade's relatively basal position (e.g. Livezey and Zusi, 2007; Hackett et al., 2008). As with other studies, the age of Neornithes predated the earliest definitive fossil evidence by at least 50–60 Ma (e.g. Ericson et al., 2006; Pereira and Baker, 2006; Brown et al., 2008; Haddrath and Baker, 2012; Jetz et al., 2012). It has been argued that Neornithes may have originally evolved in the southern hemisphere, and that the poorer fossil record from this region may explain the large gap between the putative origin of crown birds and the first definitive fossil occurrence (Cracraft, 2001; Haddrath and Baker, 2012; Brocklehurst et al., 2012). Due to the computational burden of performing each analysis, further testing of the sensitivity of the estimated divergence times to prior choice was not feasible.

98 Comparisons to Other Divergence Time Studies – Higher-Level Divergences Broadly interpreted, the divergence times estimated here were distributed between the older times estimated by Brown et al. (2008) and the younger ages estimated by Ericson et al. (2006). Higher-level divergence time estimates, where comparable clades defined by equivalent taxonomic sampling existed, were quite similar to those estimated by Jetz et al. (2012), Haddrath and Baker (2012), and Pacheco et al. (2011) with most credibility intervals overlapping (Table 3.3; Fig. 3.4). The divergence dates reported here appear to be generally congruent to an analysis of the Hackett et al. data set figured but unpublished by Brown and van Tuinen (2011). The mtDNA-based analysis conducted by Brown et al. (2008) recovered divergence times older than the other recent analyses using mixed or nuclear DNA data sets, including the present study (Fig. 3.4c) but these are generally consistent with ages from earlier mtDNA studies (e.g. Pereira and Baker, 2006). Even considering the relatively ancient marginal priors on the age of Neornithes, Neognathae and Neoaves used here, the marginal posterior distributions of divergence dates for these nodes are younger than those estimated by Brown et al. (2008). Determining whether these differences are due to Brown et al. (2008)'s use of quickly evolving mtDNA (e.g. Zheng et al., 2011), due to differences in the phylogenetic topology recovered (e.g. Brown et al. [2008] recovered Passeriformes as the basal-most neoavian clade) or due to differences in fossil calibrations or, especially differences in marginal priors on divergence times, is beyond the scope of this paper but merits comprehensive consideration. The greatest variability between this study and that of Jetz et al. (2012) for higher-level divergences was in the divergence date estimates for crown group paleognaths and tinamous (Fig. 3.4a). The date estimate recovered by Jetz et al. (2012) for tinamous is much older relative to other studies. This may be because Jetz et al. recovered tinamous as the earliest diverging paleognath in contrast to this study and other recent studies that recovered a more derived position for tinamous within Paleognathae (e.g. Harshman et al., 2008; Hackett et al., 2008; Phillips et al., 2010; Haddrath and Baker, 2012; Smith et al., 2013). The

99 divergence estimate for crown paleognaths had the highest credibility interval size in this study and, as there were no internal fossil calibrations within paleognaths, the slightly older estimate obtained by Haddrath and Baker (2012) using paleognath-focused taxon sampling and an internal fossil calibration may be more accurate.

Comparison to Jetz et al. (2012) – Lower Level Divergences The Jetz et al. (2012) study used novel methodology to combine a “backbone” analysis of higher-level divergences with clade-level chronograms. Although a comprehensive assessment of the Jetz et al. method and composite “pseudo- posterior” chronogram distribution is beyond the scope of this paper, an empirical comparison of recovered divergence times was performed (Fig. 3.4d). Detailed comparisons and interpretation of the differences between the divergence time resulting from this study and those recovered by Jetz et al., (2012) were complicated because marginal priors on divergence times in Jetz et al.'s “backbone” analysis were not reported and the behavior of the tree distribution assembly algorithm has not been carefully studied. Empirically, mean divergence times for major shared divergences between this study and that of Jetz et al. (as defined here by the broad taxonomic sampling of Livezey and Zusi, 2007) were strongly correlated (Pearson's r = 0.88, n = 109; Fig. 3.4d). Interestingly, variability in divergence time estimates between the studies was greatest at lower taxonomic levels. This is surprising because uncertainty in divergence time estimates is generally expected to increase with divergence age (dos Reis and Yang, 2013). This may be because although this study was largely congruent with the “backbone” analysis conducted by Jetz et al. (Fig. 3.4a), the multiple clade- level divergence time analyses employed by Jetz et al. and then grafted to their backbone analysis produced quantitatively different divergence times. Jetz et al. (2012) also employed fossil calibrations on crown group Charadriiformes and crown group Psittaciformes based on fossils that did not meet the criteria for inclusion here (see Supplementary Methodology). No strongly consistent directional bias (early or late) relative to this study was observed, although Jetz et

100 al.'s divergences are on average ~1.5 Ma older, the large and frequently overlapping credibility intervals in both studies suggested caution in interpreting mean differences. Further detailed study at the clade level will likely be necessary to determine the relative validity and accuracy of tipwards divergence times estimated by this and Jetz et al.'s studies. However, there are three arguments why the results of this analysis may be more appropriate relative to Jetz et al. (2012): a) divergence dates estimated here were sampled from the Bayesian analysis' posterior, not a pseudo-posterior distribution of grafted chronograms produced from multiple analyses, b) more numerous and carefully chosen fossil calibrations were included here and c) no taxa without molecular sequence data were included in this analysis.

Implications for Avian Evolution Most inter-ordinal level divergences were recovered from the Cretaceous although 95% HPD credibility intervals sometimes extended either to near or across the K–Pg boundary into the Paleogene (Fig. 3.2). In aggregate, the divergence times estimated here concur with previous studies (e.g. Slack et al., 2006; Brown et al., 2008; Pacheco et al., 2011; Jetz et al., 2012) that many lineages probably survived the K–Pg extinction event and that the basal radiation and origin of neoavian orders occurred mostly before but also across the K–Pg boundary. This implies that significant fossil discoveries remain to be made from the Cretaceous of taxa on the stem lineages of modern orders. The most important differences between studies concerned the degree to which within order diversification of crown groups and family-level divergences occurred in the Cretaceous. Here, the timing of these divergence and diversification was nearly exclusively Paleogene in age (Fig. 3.2) in contrast to Brown et al. (2008) who estimated that relatively more divergences occurred in the Cretaceous. This analysis was most consistent with Penny and Phillips' (2004) models C and D but given the relatively large credibility intervals on divergence dates estimated here, further arbitration between these models is not possible. Future studies of large phylogenomic data sets (e.g. dos Reis et al., 2012), coupled with more

101 sophisticated model-based treatment of the fossil record (e.g. Wilkinson et al., 2011; Ronquist et al., 2012a) may provide the accuracy and precision required to fully explore the nature of ecological diversification of Neoaves and its relationship to the K–Pg extinction event. The results of this study also supported the contention that the radiation at the base of Neoaves was indeed very rapid, probably within approximately 10–15 Ma (Fig. 3.2) contrasting with the longer time period inferred by Brown et al. (2008: Fig. 4). Credibility intervals for the base of neoavian radiation were strongly overlapping on TMCC and clade posterior probabilities were low, suggesting very rapid and uncertain divergences. Surveying the 10 000 chronogram sample from the posterior distribution of chronograms, 5669 chronograms contained branch lengths less than 0.1 Ma and there were an average of 7 branches per chronogram with durations less than 0.5 Ma and 19 branches less than 1 Ma. These results further emphasize the need for future studies to address gene tree discordance given very short times between deep divergences inferred here.

CONCLUSIONS The analysis of neornithean phylogeny and divergence times presented here, which is among the largest single Bayesian divergence time dating analysis of crown group birds, did not conclusively resolve either the phylogenetic interrelationships of major clades nor did it comprehensively test the hypothesis of whether there was extensive ecological diversification of Neornithes in the Cretaceous. Indeed, The conclusions presented here were clearly tempered by the high proportion of missing data, poor node support at basal nodes and the width of divergence time credibility intervals. However, despite uncertainty, this analysis did recover support for a Cretaceous age for basal divergences in Neoaves and the origin of most modern orders while within-order diversification of crown groups was confined to the Cenozoic. This study further provided two important outputs. First, a distribution of 10 000 chronograms was estimated from a single, consistent Bayesian analysis that, in aggregate, constituted a plausible set of

102 relationships and timescales for the evolution of major lineages of crown group birds. This provides a useful comparison to Jetz et al.'s (2012) much larger composite analysis which was found to be broadly consistent in terms of ages of higher-level divergences, although estimates of lower-level divergences were more variable. Second, the distribution of chronograms and supermatrices generated here were designed to provide generic-level coverage of the comprehensive Livezey and Zusi (2006; 2007) morphological data set; this will facilitate both morphological hypothesis testing and comparative analysis as well as the use of “total-evidence” approaches integrating molecules and morphology to both date divergences and estimate phylogeny. Indeed, the seemingly intractable problem of comprehensively estimating deep avian phylogeny and reconciling “rocks and clocks” may in the future be resolved by combined “total-evidence” phylogenetic analysis of truly phylogenomic molecular data sets (e.g. dos Reis et al., 2012) combined with large morphological data sets that incorporate fossil taxa into the Livezey and Zusi (2007) framework using emerging methods (e.g. Pyron, 2011; Ronquist et al., 2012a; Laurin, 2012; Wood et al., 2013).

103 Musophaga A Colius+Urocolius Tauraco Tyto Corythaeola Phodilus Corythaixoides+Crinifer B Athene Gavia Otus 3 Jabiru Asio Mycteria Strix Ciconia Ninox Platalea Harpactes Nipponia D Trogon Plegadis Pharomachrus C Theristicus Malacoptila Eudocimus Galbula Balaeniceps Nystalus Scopus Monasa E Pelecanus Bucco Cochlearius Trachyphonus Trigisoma Capito Botaurus Pteroglossus Ardea Selenidera Nycticorax Andigena Nyctanassa Aulacorhynchus Egretta Ramphastos Sula Semnornis Morus Pogoniulus F Phalacrocorax Megalaima Anhinga H Picumnus G Fregata Colaptes Pterodroma Dryocopus Calonectris+Puffinus Picoides Pelecanoides Jynx Fulmarus Indicator Pachyptila Momotus Oceanodroma Todiramphus Hydrobates Megaceryle Phoebetria Ispidina Thalassarche Alcedo Phoebastria Todus Oceanties Brachypteracias Spheniscus I Coracias Eudyptula Eurystomus J Eudyptes Merops Aptenodytes Phoeniculus Eupodotis+Afrotis K Upupa Ardeotis Rhinopomastus Phaenicophaeus Bucorvus Coccyzus Bycanistes Cuculus Aceros Eudynamys Tockus Coua Leptosomus Centropus Cathartes Guira Sarcoramphus 2 Crotophaga Coragyps Neomorphus Leptodon Geococcyx Gyps Laterallus Misaetus Aramides Accipiter Rallus Buteogallus Himantornis Buteo Porphyrio Gampsonyx Fulica L Pandion Sarothrura Sagittarius Heliornis Herpetotheres Podica Micrastur Aptornis Falco Aramus Daptrius 1 M Grus Caracara Psophia 4 Strigops Opisthocomus Forpus Pyrrhura Aratinga * Nandayus Amazona Brotogeris Steaornis N Psittacus Nyctibius Agapornis Batrachostomus Neophema Podargus Chalcopsitta Podager Trichoglossus Chordeiles Melopsittacus Caprimulgus Platycercus Eurostopodus Psittacula Metallura Alisterus Archilochus Micropsitta Hylocharis Cacatua Colibri Nymphicus Phaethornis Turdus Glaucis Catharus O Aerodramus Parus Apus Bombycilla P Aeronautes Hypsipetes Cypseloides Sylvia Streptoprocne Zosterops Hemiprocne Phylloscopus Aegotheles Acrocephalus Eurypyga Regulus Rhynochetos Fringilla Podiceps Hemispingus Tachybaptus * Motacilla Podilymbus Passer Phoenicopterus Vidua Monias Taeniopygia Mesitornis Ploceus Turtur Picathartes Chalcophaps Corvus Treron Aphelocoma Columba Lanius Zenaida Malurus Leptotila Climacteris Geotrygon Ptilonorhynchus Didunculus Menura Goura Pitta Raphus Smithornis Pezophaps Sapayoa Otidiphaps Todirostrum Ducula Minoectes Gallicolumba Cnemotriccus Geopelia Sayornis Columbina Pitangus Q Phaethon Tyrannus R Pteroclididae Lepidothrix Anseranas Dendrocolaptes Dendrocygna Lepidocolaptes Anas Cranioleuca S Aythya Scytalopus Malacorhynchus Grallaria Oxyura Thamnophilus Branta Acanthisitta Anser T Cariama Biziura Thinocorus Chauna Attagis Ortalis Pedionomus Crax Rostratula Meleagris U Jacana Dendragapus Arenaria Falcipennis Tringa Coturnix Phalaropus Francolinus Scolopax Alectoris V Turnix Gallus Uria Rollulus W Synthliboramphus Callipepla Cepphus Colinus Stercorarius Numida Rynchops Acryllium Rissa Alectura Larus Megapodius Chlidonias Apteryx Dromas Casuarius Glareola Dromaius Cursorius Eudromia Phegornis+Charadrius Nothoprocta Haematopus Crypturellus Himantopus Tinamus Cladorhynchus Anomalopteryx Inidorhyncha Emeus Pluvialis Dinornis Burhinus Rhea Chionis Pterocnemia Struthio

104 Figure 3.1. Maximum likelihood estimate of topology from the analysis RAxML analysis of the genus-level supermatrix. The highest likelihood topology out of 1024 independent tree searches is annotated with bootstrap support values from 1024 bootstrap replicates. Coloured dots at the node indicate proportional bootstrap support (BS): red: BS < 0.70; yellow: 0.70 ≤ BS < 0.95; and green: BS ≥ 0.95. Capital letters indicate the location of fossil calibrations in Table 3.2 for the divergence time analysis (see also Supplementary Methodology). All nodes coloured green and marked with asterisks were constrained in the divergence time analysis. The tree was rooted with a composite crocodylian outgroup (not shown) and figured branch lengths are arbitrarily drawn for readability.

105 Malacorhynchus Anseriformes Oxyura Aythya Anas S Branta Anser Biziura Dendrocygna Anseranas Chauna Coturnix Francolinus Alectoris

Gallus Galliformes Meleagris Falcipennis Dendragapus Rollulus Callipepla Colinus Numida Acryllium Crax Ortalis Alectura Megapodius Grus

Aramus "Gruiformes" M Psophia Heliornis Podica Sarothrura Porphyrio Himantornis Rallus Laterallus Aramides Fulica E Pelecanus Threskiornitidae Balaeniceps Balaenicipitidae

Scopus Pelecanidae Scopidae

Tigrisoma Ardeidae Cochlearius Egretta Botaurus Nycticorax Ardea Nyctanassa Plegadis Nipponia

Patalea "Suliformes" C Eudocimus Theristicus Anhinga Phalacrocorax F Sula Morus G Fregata Ciconia Mycteria Ciconidae Jabiru Procellariiformes Thalassarche Phoebetria Phoebastria Oceanites Pachyptila Pelecanoides Calonectris+Puffinus Fulmarus Pterodroma Hydrobates Oceanodroma Aptenodytes J Eudyptes Eudyptula Sphenisciformes Spheniscus Gavia Gaviiformes Tauraco Musophaga Musophagiformes Corythaixoides+Crinifer

Corythaeola Cuculiformes Guira Crotophaga Neomorphus Geococcyx Cuculus Eudynamys Phaenicophaeus Coccyzus Coua Centropus Ardeotis Eupodotis+Afrotis Otididae Forpus Nandayus Aratinga

Pyrrhura Psittaciformes Brotogeris Amazona Psittacus Psittacula Alisterus Micropsitta Melopsittacus Trichoglossus Chalcopsitta Agapornis Neophema N 9 Platycercus Cacatua Nymphicus Strigops Parus 4 Turdus Catharus Bombycilla Hemispingus Fringilla Passer Motacilla Vidua Ploceus Taeniopygia Phylloscopus Acrocephalus Hypsipetes Sylvia

Zosterops Passeriformes Regulus Picathartes Lanius Corvus Aphelocoma Malurus Ptilonorhynchus Climacteris Menura Pitangus Tyrannus Sayornis Cnemotriccus Todirostrum Mionectes Lepidothrix Grallaria Dendrocolaptes Lepidocolaptes 7 Cranioleuca Scytalopus Thamnophilus Sapayoa Smithornis

Pitta Falconidae Acanthisitta Cariama T Caracara Daptrius Falco Micrastur Herpetotheres Cretaceous Paleogene Neogene Early Late Paleoc. Eocene Oligoc. Miocene P.

120 100 80 60 40 20 0

106 Buteo Accipitridae Buteogallus Accipiter Nisaetus Gyps Leptodon Gampsonyx Pandion L Sagittarius Leptosomus Tockus Aceros Bycanistes Bucerotidae Bucorvus Phoeniculus K Upupa Upupiformes Rhinopomastus "Coraciiformes"* Momotus Alcedo Ispidina Megaceryle Todiramphus Todus Merops Coracias Eurystomus I Brachypteracias Galbula Nystalus Bucco Monasa Malacoptila Picumnus

Picoides Piciformes Colaptes Dryocopus Jynx Indicator Ramphastos Aulacorhynchus Andigena H Selenidera Pteroglossus Semnornis Capito Pogoniulus Trachyphonus Megalaima Pharomachrus D Trogon Trogoniformes 1 Harpactes 5 Ninox Asio Otus Strix Strigiformes B 6 Athene Phodilus Tyto Colius+Urocolius Coliiformes A Cathartes Sarcoramphus Cathartidae Coragyps Arenaria Phalaropus Tringa Scolopax Rostratula Jacana U Attagis Thinocorus

Pedionomus Charadriiformes Larus Rissa Chliodonias Rynchops Stercorarius Cepphus W Syhthlioboramphus Uria Glareola Cursorius 8 2 Dromas Turnix V Phegornis+Charadrius Cladorhynchus Himantopus Haematopus Ibidorhyncha Pluvialis "Caprimulgiformes" Chionis Burhinus Steatornis Nyctibius Eurostopodus Caprimulgus Chordeiles Podager Podargus Batrachostomus Aegotheles Phaethornis

Glaucis Apodiformes Colibri Archilochus Hylocharis Metallura O Cypseloides Streptoprocne P Aeronautes Aerodramus Apus Hemiprocne Rhynochetos Eurypyga Eurypygiformes Phaethon Phaethontidae Q Monias Mesitornis Mesitornithidae Podilymbus Podiceps Podicipediformes Tachybaptus Phoenicopterus Phoenicopteriformes R Pteroclididae Pteroclididae Treron Turtur Columbiformes Chalcophaps Leptotila Zenaida Geotrygon Columba Columbina Gallicolumba Geopelia Ducula Didunculus Goura Otidiphaps Opisthocomus Rhea

Pterocnemia Paleognathae Casuarius Dromaius Emeus Anomalopteryx Dinornis Nothoprocta Eudromia 3 Tinamus Crypturellus Apteryx Struthio Cretaceous Paleogene Neogene Early Late Paleoc. Eocene Oligoc. Miocene P.

120 100 80 60 40 20 0

107 Figure 3.2. (Two Pages). Maximum clade credibility chronogram summarizing the sample from the posterior distribution of chronograms estimated by the divergence time analysis. Node ages are drawn at the mean of the respective marginal posterior distributions. Group names mostly follow Hackett et al. (2008) and used operationally here and were not intended as a systematic review; * “Coraciiformes” includes the following families: Todidae, Momotidae, Alcedinidae, Halcyonidae, Cerylidae, Meropidae, Coraciidae and Brachypteraciidae. The Cretaceous–Paleogene (K–Pg) boundary is marked by the red vertical line. Clade posterior probabilities are indicated by coloured dots: red: PP < 0.95; yellow: 0.95 ≤ PP < 0.99; and green: PP ≥ 0.99. Nodes crossed with a solid line were constrained a priori (see Supplementary Methodology). Blue bars represent the 95% highest probability density credibility interval for node ages. Fossil calibrations on stem lineages are indicated by capital letters arbitrary drawn along the lineage and also correspond to stem lineages included in Figure 3.3. Numbered nodes and stem lineages correspond to node densities included in Figure 3.3.

108 Neornithes (1) Crown Charadriiformes (2)

Crown Paleognathae (3) Stem Turnicidae (V)

Neognathae (4) Stem Jacanidae (U)

Neoaves (5) Stem Alcidae (W)

Stem Anatidae (S) Stem Coliiformes (A)

Stem Apodiformes (O) Stem Strigiformes (B)

Stem Apodidae (P) Crown Strigiformes (6)

Stem Phaethontidae (Q) Stem Trogoniformes (D)

Stem Pteroclididae (R) Stem Pici (H)

Stem Threskiornithidae (C) Stem Coraciidae+Brachypteraciidae (I)

Stem Pelecanidae (E) Stem Upupiformes (K)

Stem Phalacrocoracidae (F) Stem Pandionidae (L)

Stem Fregatidae (G) Stem Cariamiformes (T)

Stem Spenisciformes (J) Stem Psittaciformes (N)/ Stem Passeriformes

Stem Gruoidea (M) Crown Passeriformes (7)

Stem Charadriiformes (8) Crown Psittaciformes (9)

120140 80100 60 40 20 0 120140 80100 60 40 20 0

Empirical Prior Density Prior Calibration Density Posterior Density

109 Figure 3.3. Fossil calibration densities (blue dotted lines), marginal priors (red dotted lines) and posterior densities (solid black lines) for selected nodes and stem lineages. The Cretaceous–Paleogene (K–Pg) boundary is indicated by the red vertical lines. All densities were scaled so that their maximum was one (Dos Reis et al., 2012). Stem-lineages densities do not necessarily correspond to a single node in Figure 3.2 as topology was variable (see Supplementary Methodology). Node and stem-lineage numbers and letters correspond to the corresponding nodes and lineages identified in Figure 3.2. Numbered nodes or lineages (1,2,3,4,5,6,7,8,9) did not have fossil calibrations and priors were determined by the overall tree prior. Densities indicated by capital letters included fossil calibrations, which were then combined with the overall tree prior by BEAST to generate the marginal prior, which does not necessarily match the calibration density (see text). Mismatch between the calibration densities and marginal priors are particularly apparent for Neornithes (Node 1), stem Pici (stem lineage H) and stem Psittaciformes (stem lineage N).

110 a) b)

c) d)

111 Figure 3.4. Comparison between the divergence times estimated by this study and other recent analyses. Higher-level divergences are compared between this study and a) Jetz et al. (2012), b) Haddrath and Baker (2012) and c) Brown et al. (2008). Error bars from this analysis are 95% HPD credibility intervals. Error bars from the Jetz et al. (2012) analysis a) are 95% HPD intervals extracted from the final tree distribution (see text). Error bars from Haddrath and Baker (2012) b) are HPD intervals reported by the authors from their BEAST analysis. Error bars from Brown et al. (2008) are ±1 standard deviation from the mean (normal approximation), as reported for their BEAST analysis. Comparison of divergence times of all reciprocally monophyletic clades between this analysis and the Jetz et al. (2012) analysis d) for clades defined by the taxonomic sampling of the Livezey and Zusi (2007) data set. Divergence times are colour coded by clade posterior probabilities (PP) recovered by this analysis (red: PP < 0.95; yellow: 0.95 ≤ PP < 0.99; and green: PP ≥ 0.99). Clade abbreviations in a), b) and c) correspond to the clades listed in Table 4: No–Neornithes, Ng–Neognathae, Na–Neoaves, Pg– Crown Paleognathae, Pa–Crown Passeriformes, Ps–Crown Psittaciformes, PP– Passeriformes/Psittaciformes, Ch–Crown Charadriiformes, Gl–Crown Galliformes, An–Crown Anseriformes, GA–Galloanserae, Ap–Crown Apodiformes, Mi–Podicipediformes/Phoenicopteriformes, St–Crown Strigiformes, Cu–Crown Cuculiformes, Pi–Crown Piciformes, Ti–Crown Tinamiformes.

112 Table 3.1. Partition scheme testing for the species level supermatrix using RAxML Partition Scheme Partitions Parameters1 Ln(L) BIC2 ΔBIC Nuclear exons concatenated by codon position (1,2,3), RAG1 and mtDNA seperated by loci and codon position 33 1198 -1390705 2794087 - (1,2,3), all introns and 12S seperated and all UTRs contcatenated Nuclear exons concatenated by codon position (1+2,3), RAG1 and mtDNA seperated by loci and codon position 28 1153 -1391977 2796155 2068.4 (1+2,3), all introns and 12S seperated and all UTRs contcatenated Nuclear exons concatenated by codon position (1,2,3), all mtDNA exons concatenated by codon position 24 1117 -1392290 2796399 2312.7 (1,2,3), all introns and 12S seperate and all UTRs

113 contcatenated Nuclear exons concatenated by codon position (1,2,3), all mtDNA exons concatenated by codon position 9 982 -1394956 2800303 6216.6 (1,2,3), 12S and all introns concatenated and all UTRs contcatenated All loci split into exons, introns and UTRs 42 1279 -1407033 2827600 33513.6 Nuclear exons concatenated, all mtDNA concatenated, 4 937 -1411633 2833180 39093.5 all introns concatenated, all UTRs concatentated All alignments concatenated by locus 25 1126 -1413699 2839313 45226.5 1Includes branch lengths 2Bayesian Information Criterion (Schwarz, 1978) Table 3.2. Fossil calibrations and hard miminum ages used to calibrate the divergence dating analysis Group Calibrated Fossil Taxon Age (Ma ago)

A Stem Coliiformes Sandcoleus copiosus 56.20

B Stem Strigiformes Ogygoptynx wetmorei 56.99 Stem Group C Threskiornithidae Rhynchaeites sp 54.00

D Stem Trogoniformes Septentrogon madseni 54.00

E Stem Pelecanidae Pelecanus sp. 28.10

F Stem Phalacrocoracidae ?Borvocarbo stoeffelensi 24.52

G Stem Fregatidae Limnofregata azygosternon 51.56

H Stem Pici Rupelramphastoides knopf 28.10 Stem Coraciidae+ I Brachypteraciidae Primobucco mcgrewi 51.56

J Stem Sphenisciformes Waimanu manneringi 60.50 Stem group K Upupidae+Phoeniculidae Messelirrisor grandis sp. 47.50

L Crown Pandionidae “Palaeocircus ciivieri” 33.90

M Stem Gruoidea Parvigrus pohli 28.10

N Stem Psittaciformes Cyrilavis colburnorum 51.56

Stem Apodiformes (Apodidae, Hemiprocnidae, O Trochilidae) Eocypselus vincenti 54.00

P Stem Apodidae Scaniacypselus szarskii 47.50

Q Stem Phaethontidae Lithoptila abdounensis 56.00

R Crown group Pteroclididae Leptoganga sepultus 24.52

114 Table 3.2. Continued

Group Calibrated Fossil Taxon Age (Ma ago)

S Stem Anatidae Vegavis iaai 66.00 Paleopsilopterus T Stem Cariamiformes itaboraiensis 50.20

U Stem Jacanidae Nupharanassa tolutaria 33.00

V Stem Group Turnici Turnipax oehslerorum 28.10

W Stem group Alcidae Alicdae incertae sedis 34.20

115 Table 3.3. Gene sequence sampling and partition scheme for the divergence time analysis Alignment Codon Substitution Partition Sequences Length Position Model (bp) 1 nucDNA Exons (MYC, FGB, 1+2 920 GTR+I+Γ 2 ODC1, MB) 3 460 GTR+Γ 3 1 958 GTR+I+Γ Recombination Activation 4 2 958 GTR+I+Γ Gene 1 (RAG1) Exon 5 3 958 SYM+Γ 6 mtDNA Exons (COX1, CYTB. 1 1245 GTR+I+Γ 7 ND2) 2 1245 GTR+I+Γ 8 12S rRNA - 782 GTR+I+Γ Fibrinogen Beta Chain (FGB) 9 - 1270 GTR+Γ Introns 4,5,6,7 10 Myoglobin (MB) Intron 2 - 717 GTR+Γ Orthinurine Decarboxylase 1 11 - 505 GTR+Γ (OCD1) Introns 6, 7 v-myc myelocytomatosis viral 12 oncogene homolog (MYC) - 347 GTR+Γ IntronB

116 Table 3.4. Divergence time estimates and comparisons to recent avian divergence time analyses This Analysis Node Age Lower 95% Upper Clade (Mean) HPD1 95% HPD Neornithes 111.7 94.9 130.0 (Paleognathae/Neognathae) 103.7 90.4 118.1 Neognathae (Neoaves/Galloanserae) Neoaves 82.0 74.6 90.0 Paleognathae 73.8 56.5 92.2 Crown Passeriformes 67.1 59.6 74.8 Crown Psittaciformes 52.3 42.0 62.5 Passeriformes/Psittaciformes 74.9 67.9 82.6 Crown Charadriiformes 65.2 56.2 74.1 Crown Galliformes 72.1 61.1 83.3 Crown Anseriformes 76.8 69.4 85.4 Galloanserae 86.4 76.3 97.1 Crown Apodiformes 61.9 55.1 69.4 Podicipediformes/ 57.9 42.7 70.9 Phoenicopteriformes Crown Strigiformes 58.4 47.6 68.8 Crown Cuculiformes 55.1 45.6 64.2 Crown Piciformes 63.0 56.7 70.2 Crown Tinamiformes 35.8 23.8 47.6 1Highest posterior density

117 Table 3.4. Continued. Jetz et al., 2012 Node Age Lower 95% Upper Clade (Mean) HPD 95% HPD Neornithes 108.4 91.1 130.2 (Paleognathae/Neognathae) 100.5 87.3 115.1 Neognathae (Neoaves/Galloanserae) Neoaves 83.3 75.6 91.8 Paleognathae 82.9 59.6 106.3 Crown Passeriformes 70.3 62.6 77.6 Crown Psittaciformes 56.9 54.3 60.5 Passeriformes+Psittaciformes 77.3 70.0 85.4 Crown Charadriiformes 65.3 56.6 75.6 Crown Galliformes 60.2 46.7 73.3 Crown Anseriformes 71.1 66.3 78.3 Galloanserae 78.7 70.5 88.8 Crown Apodiformes 62.3 55.5 70.2 Podicipediformes/ 51.9 27.2 72.5 Phoenicopteriformes Crown Strigiformes 64.4 49.0 77.8 Crown Cuculiformes 67.4 57.4 77.8 Crown Piciformes 60.8 51.0 70.6 Crown Tinamiformes 71.4 50.7 94.4

118 Table 3.4. Continued. Haddrath and Baker, 2012 (BEAST) Node Age Lower 95% Upper Clade (Mean) HPD 95% HPD Neornithes 127.0 114.0 133.0 (Paleognathae/Neognathae)

Neognathae (Neoaves/Galloanserae) 106.0 91.0 121.0 Neoaves 83.0 71.0 98.0 Paleognathae 97.0 73.0 119.0 Crown Passeriformes 49.0 26.0 69.0 Crown Psittaciformes -- - Passeriformes+Psittaciformes -- - Crown Charadriiformes 63.0 46.0 80.0 Crown Galliformes -- - Crown Anseriformes -- - Galloanserae 83 71 98 Crown Apodiformes 50 47 53 Podicipediformes/ Phoenicopteriformes -- - Crown Strigiformes -- - Crown Cuculiformes -- - Crown Piciformes -- - Crown Tinamiformes 41.0 24.0 61.0

119 Table 3.4. Continued. Pacheco et al., 2011 (BEAST) Node Age Lower 95% Upper Clade (Mean) HPD 95% HPD Neornithes -- - (Paleognathae/Neognathae) -- - Neognathae (Neoaves/Galloanserae) Neoaves -- - Paleognathae -- - Crown Passeriformes 78.5 71.8 85.7 Crown Psittaciformes 61.4 54.0 69.7 Passeriformes/Psittaciformes -- - Crown Charadriiformes 64.6 55.3 80.4 Crown Galliformes -- - Crown Anseriformes -- - Galloanserae -- - Crown Apodiformes 58.34 46.48 69.87 Podicipediformes/ 67.15 - - Phoenicopteriformes Crown Strigiformes 75.95 64.64 87.07 Crown Cuculiformes 59.76 48.27 71.62 Crown Piciformes -- - Crown Tinamiformes -- -

120 Table 3.4. Continued. Brown et al., 2008 (BEAST) Node Age Standard Clade (Mean) Deviation2 Neornithes 133.2 8.1 (Paleognathae/Neognathae)

Neognathae (Neoaves/Galloanserae) 126.0 7.1 Neoaves 118.5 6.8 Paleognathae 105.9 11.7 Crown Passeriformes 106.6 7.2 Crown Psittaciformes -- Passeriformes/Psittaciformes -- Crown Charadriiformes 81.7 6.3 Crown Galliformes 99.0 8.4 Crown Anseriformes 100.5 8.3 Galloanserae 110.4 7.8 Crown Apodiformes 80.5 9.9 Podicipediformes/ -- Phoenicopteriformes Crown Strigiformes 84.2 9.1 Crown Cuculiformes 74.1 8.6 Crown Piciformes 93.6 6.8 Crown Tinamiformes --

2Brown et al., (2008: Table 3) used a normal approximation to calculate standard deviation of age estimates from the recovered (and unreported) credibility intervals

121 CHAPTER 3: SUPPLEMENTARY METHODOLOGY

Extant birds have been the subject of intensive phylogenetic investigation. However, most analyses (with several notable exceptions, below) focus on within- order or family level relationships. To combine the information available from these sub-ordinal studies, a supermatrix was created by combining the existing phylogenetic analysis of Hackett et al. (2008) and Ericson et al. (2006) and then sampling further existing sequence data for focal avian taxa (see main text for focal taxon choice). The final species supermatrix of species (n = 450, including outgroups) includes representatives of extant avian orders and most non-passerine families. Because species nomenclature occasionally differed between studies, an operational set of species names and associated GenBank taxonomic IDs (Appendix 3.1) was created to facilitate the analysis described here and was not intended as a comprehensive assessment or systematic review of generic assignments. To reduce the quantity of missing data, the final species-level supermatrix was further collapsed into a generic-level supermatrix consisting of composite taxa (see details below).

SPECIES- LEVEL SUPERMATRIX The alignments used by Ericson et al. (2006) and Hackett et al. (2008) were first combined (details below). Existing analyses and NCBI Genbank were then surveyed to determine a set of loci that were reasonably well sampled across the focal taxa (Appendix Table 3.2). NCBI Genbank was comprehensively queried and all available sequences for the following mitochondrial loci were retrieved: 12S small-subunit rRNA (12S), cytochrome-b (CYTB), NADH Dehydrogenase Subunit 2 (ND2), Cytochrome C Oxidase, subunit 1 (COX1). Both text searches for loci names and BLAST searchs with homologous sequences were used to retrieve sequences. The complete list of taxa, including common names, GenBank taxonomic IDs, and (non-exhaustive) common synonyms are provided in Appendix Table 3.1.

122 Merging the Hackett et al. (2008) and Ericson et al. (2006) Data Sets The multiple sequence alignment for the nuclear loci used in Hackett et al.’s (2008; H2008 hereafter) phylogenetic analysis were downloaded from TreeBase (http://www.treebase.org/). Alignments for each locus were extracted from the concatenated data set using the locus boundaries provided with the alignment files. Each locus was further split into its component alignments of individual introns, exons and UTRs using the published boundaries. Hackett et al. included 19 genes and of these, three were at least partially sampled by Ericson et al's (2006; E2006 hereafter) analysis: Fibrinogen Beta Chain (FGB), V-Myc Myelocytomatosis Viral Oncogene Homolog (avian) (MYC), and Myoglobin (MB). These constitute three out of five loci used in the E2006 analysis. The alignment used by Ericson et al. (2006) was downloaded from the Naturhistorika riksmuseet website (www.nrm.se). Ericson et al. (2006) included composite taxa that consisted of sequences from multiple species; these were decomposed by assigning each loci to the species from which it was sampled using Ericson et al. (2006: Table ESM-1). Sequences for the preceding three loci were then extracted from the species-level E2006 data set. Of these sequences, those already sampled and not reported as missing data in the H2008 matrix were removed. Pruned of duplicate taxa, the E2006 data set includes further taxonomic sampling for FGBi7, MYC, and MB of 56, 46, and 56 species respectively. NCBI Genbank was then queried for all other available sequences for the MYC, FGB (intron 5 and intron 7) and MB loci for all species in the supermatrix. GenBank searches yielded an additional 19, 81, 32 and 33 at least partial sequences for the FGB intron 5, FGB intron 7, MYC and MB, respectively (GenBank Accession numbers in Appendix 3.2). The E2006 and GenBank sequences were then aligned to the original H2008 alignment using using ClustalX (Larkin et al., 2007) and manually checked and adjusted as necessary for each loci individually. In addition to the three loci used to augment the H2008 data set, Ericson et al. (2006) also reported sequence data for ornithine decarboxylase 1 (ODC1) for 88

123 species. The multiple sequence alignment of this gene was extracted from the E2006 multiple sequence alignment using the boundaries provided with the sequences. GenBank was then queried for all available sequences for ODC1 for all species included in the supermatrix and a total of 56 additional sequences were sampled (Accession numbers in Appendix 3.2). These sequences were then aligned against the original E2006 ODC1 alignment using ClustalX (Larkin et al., 2007) and manually checked. Ericson et al. (2006) also reported sequence data for the frequently sampled recombination activating gene 1 (RAG1). Because the length of RAG1 fragments sequenced varies and longer variant sequences may have been reported for the same taxa after Ericson et al. (2006)'s publication, NCBI GenBank was used to download the longest RAG1 sequence available for each species in the supermatrix. A total of 223 sequences, including corresponding translations were retrieved, including those originally used by Ericson et al. (2006). The RAG1 sequences were then aligned in protein space using ClustalOmega (Sievers et al., 2011), manually adjusted as necessary and then backtranslated to nucleotides using the GenBank cDNAs with T-Coffee (Notredame et al., 2000) while maintaining the open reading frame.

Mitochondrial loci (COX1, CYTB, ND2, 12S rRNA) For the three protein mitochondrial coding genes COX1, CYTB and ND2, NCBI GenBank was manually queried, and nucleotide and protein sequences were downloaded for each gene. For species where the mitochondrial genome was available in the curated NCBI RefSeq database (Pruitt et al., 2007), the RefSeq annotations were used to extract the mitochondrial loci. Where mitochondrial genomes were available but not accessioned in the RefSeq database, the entire genome was downloaded and a homologous sequence from a closely related taxon was used to confirm the loci annotations present. NCBI searches resulted in a total of 349, 397 and 328 at least partial sequences for each gene, respectively (Accession Numbers in Appendix 3.2). These three loci were aligned individually in protein space using ClustalOmega (Sievers et al., 2011), manually inspected

124 and adjusted as necessary and then backtranslated to nucleotides using T-Coffee (Notredame et al., 2000) and the GenBank cDNA data, maintaining the open reading frame. For the 12S rRNA sequences, NCBI GenBank was manually queried for the species in the supermatrix and 318 sequences were downloaded, using mitochondrial genomes, where available, as above. An existing alignment of bird 12S sequence based on secondary structure (Fain et al., 2007) was used to construct a profile and then the 12S sequences were aligned to this profile using ClustalX (Larkin et al., 2011). The RM-Coffee package, which uses multiple secondary structure-based rRNA alignment algorithms was also used to estimate the 12S alignment which was similar to that obtained by alignment to the Fain et al. (2007) profile and the latter was used for all further analysis. The final alignment was further processed using Gblocks v0.91b (Castresana, 2000) to remove poorly aligned stretches with the following settings: B1=158, B2=212, B3=10, B4=10, B5=”Half” .

Further Alignment Processing Following Hackett et al. (2008), for each intron sub-alignment, long apomorphic insertions shared by fewer than four taxa were removed using T- Coffee v9.03 (Notredame et al., 2000). H2008 includes significant exonic sequence but these were not aligned inframe in the original analysis. Following Lanfear et al. (2010), H2008 exonic sequences were first concatenated by gene (where more than one consecutive exon was included) and the start and ends of the sub-alignments were trimmed such that the exonic region was inframe. The open reading frame was confirmed by conceptually translating the resulting nucleotide alignments using T-Coffee v9.03 (Notredame et al., 2000) and both ensuring that there were no inframe stop codons and by using BlastP (Altschul et al., 1990) searches of GenBank to compare each conceptual translation to the annotated translations available on GenBank.

Gene Tree Estimation

125 Trees were estimated for each sub loci included in the supermatrix (e.g. separate tree for each intronic and exonic region per loci). For each final sub- alignment, RAxML v7.2.8 was used to estimate a maximum likelihood phylogeny under the GTRCAT model followed by final optimization under the GTR+GAMMA model (Stamatakis, 2006). The gene trees were inspected to ensure there were no anomalously long branches that might indicate an alignment error, or that a pseudogene had been included in the original alignments. For each gene tree, there were no such branches, although several groups (e.g. Upupiformes; Turnix spp.) had generally longer branches across most trees. This appears to reflect underlying heterogeneity in evolutionary processes for these groups (e.g. see also Hackett et al., [2008]: Fig. 3).

Concatenation of Alignments The alignments of each locus were then concatenated into a single supermatrix for all of the species using T-Coffee v9.03 (Notredame et al., 2000) and the boundaries of each partition were recorded. The final species-level supermatrix consisted of 450 species, with ~39Kb of sequence data. The overall proportion of missing or gapped alignment positions was 0.68±0.26 (mean±sd) and the percentage of these positions per taxon ranged from a maximum of 0.986 for Aptornis defossor (New Zealand Adzebill, an extinct taxon) to a minimum of 0.194 for Cathartes aura (Turkey Vulture). The final species-level alignment is available from the authors upon request.

Generic-level Supermatrix The assembled species supermatrix contains numerous examples of genera wherein each data set (e.g. Hackett et al., 2008; RAG1) are sampled for different species within the same genus. To reduce the quantity of missing data, the species- level supermatrix was collapsed into a generic-level supermatrix of composite taxonomic units. Where a genus was represented by a single species, the tip was renamed to the corresponding genus. Where a genus was represented by more than one species, the monophyly of the genus relative to the other taxa included

126 was first confirmed by examining the results of the phylogenetic analysis of the species-level supermatrix (see below). If monophyly was confirmed, species were merged by combining the species-level alignments as follows: where alignment positions were represented by a single sampled species, that species' sequence was used; where alignment positions were represented by multiple species, nucleotide polymorphism coding was used to conservatively estimate a consensus sequence. If the genus was not monophyletic, which was the case for several genera, the species-level alignments were merged into higher taxonomic units. using the same methodology described above (Appendix 3.3). In one case, for the genus Caprimulgus, which was not recovered as monophyletic, only the better sampled Caprimulgus longirostris was retained to represent the genus. The outgroup was collapsed into a single crocodylian taxon using the same methodology. Using this procedure, a generic-level supermatrix consisting of 314 OTUs was produced. All alignment processing was conducted using an R script and some functions from the APE package (Paradis et al., 2004; scripts available from the authors upon request). The use of composite taxa decreased the overall supermatrix-wide proportion of missing data to 0.59±0.27 (mean±SD) although for some genera, the decrease was substantial. The proportion of sites represented by an ambiguity code excluding total ambiguity (e.g. N or ?) was 0.002±0.006 (mean±SD). For certain taxa, however, ambiguity codes were more common. The Urocolius+Colius composite taxon had the highest proportion of ambiguity codes at 0.045 and when corrected for totally ambiguous sites, the taxon had a proportion of 0.061 not totally ambiguous sites represented by nucleotide ambiguity codes. The final genus-level alignment is available from the authors upon request.

MAXIMUM LIKELIHOOD PHYLOGENETIC ANALYSIS Maximum Likelihood phylogenetic analyses were performed on both the species and generic-level supermatrices using RAxML v7.2.8 (Stamatakis, 2006). Given the size of these supermatrices, RAxML's speed and ability to use HPC

127 MPI-based computing clusters was indispensable and the same software package was one of the methods used by Hackett et al. (2008) in their original partitioned analysis. A Bayesian analyses of phylogeny using MrBayes v3.2.1 (Ronquist et al., 2012b) proved computationally intractable even using extensive HPC resources.

Partition Scheme The phylogenetic analyses of both supermatrices were conducted using partitioned models. Hackett et al. (2008) partitioned their alignment primarily by locus (concatenating introns, exons and UTRs). Here, several partition schemes were tested using RAxML v7.2.8 (Stamatakis, 2006; Table 3.1). For each potential partition scheme described in Table 3.1, a RAxML tree search was conducted using the species-level supermatrix with 64 independent tree searches per partition scheme under the GTRCAT model with final model optimization under the GTRGAMMA model. The independent tree search that yielded the highest likelihood was used to calculate the Bayesian Information Criterion (BIC; Schwarz, 1978) score. The partition scheme with the lowest BIC score was used for all subsequent phylogenetic analyses. The partition schemes tested and model likelihoods are described in Table 3.1, ranked by BIC.

Phylogenetic Topology Searches RAxML v7.2.8 (Stamatakis, 2006) was used to search for the topology with the best fit to the species-level and genus-level supermatrices under the optimal partitioning scheme (Table 3.1). RAxML was executed using 1024 cores at the CLUMEQ/Colosse HPC supercomputing cluster at the Université de Laval, Québec, Canada. Because the RAxML hill-climbing optimization algorithm can become stuck in islands of sub-optimal phylogenies, a total of 512 and 1024 independent RAxML tree searches were performed under the GTRCAT model followed by final optimization under the GTRGAMMA model for both the species and genus-level supermatrices. The crocodylian ougroup was specified

128 using RAxML command “-o” and a different random seed was given to each run. The results from the tree search yielding the highest likelihood were retained. Non-parametric rapid bootstrapping was performed using RAxML (-x option) to estimate the robustness of the recovered relationships using 512 bootstrap replicates for the species-level matrix and 1024 bootstrap replicates for the generic-level analysis. The species and generic-level analyses each required approximately 2.5 core-years of computation. The maximum likelihood estimate of tree topology and branch lengths (TMLE) was annotated with bootstrap support values using RAxML.

DIVERGENCE TIME ANALYSIS To estimate divergence times, a partitioned Bayesian analysis was carried out using BEAST v1.7.5 (Drummond et al., 2012) under a relaxed, lognormally- distributed molecular clock (Drummond et al., 2006). Attempts to use the full species or generic-level supermatrix for divergence time dating were not computationally tractable using BEAST, even using extensive HPC resources. Although an approximate likelihood approach implemented in the MCMCTree package may have been tractable (Yang, 2007; dos Reis and Yang, 2011; dos Reis et al., 2012), this would require fixing phylogenetic topology because MCMCTree does not sample phylogenies (Yang, 2007), which is difficult to justify in light of the uncertainty in neoavian phylogeny. Furthermore, the approximate likelihood methodology used by MCMCTree may not be accurate when the gene trees of individual partitions have near zero length branches, as was occasionally the case in this supermatrix (dos Reis and Yang, 2011). Therefore, it was necessary to sub- sample loci from the full genus-level supermatrix to make the computation of divergence times with BEAST tractable. By selecting loci that were relatively well sampled across all genera, this also had the effect of significantly reducing the proportion of ambiguous alignment positions.

Sequence Data

129 The reduced generic-level supermatrix was created by sub-sampling the exonic and intronic regions of FGB, MYC, MB, ODC1, RAG1, COX1, CYTB, ND2 and 12S loci from the generic-level supermatrix. These genes provided the best coverage across all genera and consisted of a mixture of more quickly evolving nuclear intronic regions and mtDNA and more slowly evolving nuclear exonic regions. The third codon position from the three protein coding mitochondrial loci was removed as saturation would be expected to eliminate any informative variation at the timescales considered here (Ho and Philips, 2009; Brown and van Tuinen, 2011). The phylogenetically distant crocodylian outgroup was removed because test runs indicated the long branch duration leading to the neornithian ingroup strongly affected estimation of the relaxed clock and tree priors. Three sequences from fossil taxa with poor coverage were also removed (Aptonris, Pezophaps and Raphus). To further increase computational tractability, the intronic alignment for FGB was further processed using GBlocks v.0.91b to more conservatively remove weakly aligned regions (Castresana, 2000; GBlocks parameters: -B1 158 -B2 158 -B3 8 -B4 10 -B5 “all”) and all intronic alignments were further processed to remove autapomorphic insertions shared by fewer than 10 taxa using T-Coffee. The final reduced generic-level supermatrix used for divergence-time dating consisted of 310 OTUs and ~10.4Kb of sequence data with overall proportion of totally ambiguous alignment positions (missing or gapped) of 0.42±0.23. Per taxon, the proportion of missing or gapped alignment positions ranged from 0.016 (Nyctibius) to 0.95 (Monasa). For all analyses, the very short MB exonic region (19bps) was concatenated inframe to the MYC exonic alignment.

Partition, Model Choice and Model Priors Initially, a partition scheme where each intron was partitioned separately and each protein coding locus was split into codon position 1, 2 and 3 was considered (e.g. Shapiro et al., 2006). However, test runs of the divergence time analysis revealed poor convergence and multiple sub-optimal runs (stable, lower posterior) which appeared to be stuck in a lower likelihood arrangement of partition

130 multipliers. This may be due to the small number of patterns in some of the highly conserved nuclear exon partitions (ODC1, MYC and FGB) and was also reflected by unstable substitution model parameters; therefore, a partition scheme was adopted where the small nuclear exons were concatenated together for codon positions and 1+2 and 3. This partition scheme had higher likelihood relative to an alternative scheme separating the short nuclear exons by loci instead of by codon position (RAxML optimization onto the topology from the generic-level supermatrix, ΔBIC = 674.72). The mitochondrial genes were partitioned into codon positions 1 and 2 (3 was not considered) and the longer RAG1,was also partitioned into codon positions 1, 2 and 3 while all introns were separated. For each partition, MrModelTest v2.3 (Nylander, 2004) was used to determine the substitution model of best fit for each partition (Table 3.3) using the phylogeny that resulted from the genus-level RAxML supermatrix analysis. MrModelTest recommended the use of an invariant site model for several partitions where the proportion of invariant sites was less then 0.05 (model averaged output from MrModelTest). Because of concerns about the simultaneous optimization of this parameter and the gamma shape parameter (e.g. Yang, 2006), an invariant site model was only used for those partitions with > 0.05 estimated invariant sites (Table 3.3). Jeffreys priors were used for all instantaneous rates for each GTR substitution model (see Zwickl and Holder, 2004). Priors on partition relative rates were truncated (at 0 and 1x10100) normal distributions with a mean of one and standard deviation of one. All other substitution model priors were left at the BEAST default values.

Fossil Calibrations The choice of fossil calibrations is important in divergence time analysis (Benton et al., 2009; Warnock et al., 2012). Misapplied fossil calibrations can affect the posterior distribution of divergence times (Parham et al., 2012). Fossil calibrations may be misapplied if they are applied to an incorrect node through error in the interpretation of the affinity of the fossil (Ksepka et al., 2011a). Saliently, whether a fossil is attributed to an extant group's crown or its stem-

131 lineage can dramatically change which node is calibrated, particularly in early diverging groups. In a recent article, Parham et al. (2012) argued for a specimen- based approach to the choice and specification of fossil calibrations. Justifications for the fossil calibrations used in this study are provided below; not all fossils have been tested in a phylogenetic framework, but a rationale for the choice was provided for each fossil. In all cases below, the available avian fossil record can only reliably provide hard minimum dates.

Calibration Densities Although the hard minimum calibrations described below could be considered in several cases quite conservative, it is important to emphasize that these are minima only below which our prior knowledge is that these groups must with 100% certainty have diverged. These minima will necessarily predate true divergences, potentially by a long time (see Brown et al., 2008 for a discussion). Lognormal distributions with shape parameter (σ) set to 1.0 were used to create soft calibrations in the form of a prior probability distribution for the calibrated nodes. The shape of the lognormal function at σ = 1.0 leads to the highest prior probability of divergence to be older than the hard minima, with a long tail. Because it is difficult to specify a justified hard maximum for any calibration, a soft maximum was adopted (Warnock et al., 2012). An approach similar to that of Jetz et al. (2012) was adopted; soft maxima were set such that the 99% quantile of the lognormal calibration density for all calibrations was set to approximately 130.2 Ma ago by empirically selecting a scale parameter for the lognormal distribution (this is very close to setting the 97.5% quantile to 110 Ma ago as in Jetz et al., 2012) . This age of 130.2 Ma ago is the upper bound on the age of the Jehol Biota in China, that although contains an extensive fossil record of avian taxa, does not contain any taxa attributable to crown Neornithes (Chang et al., 2009; Brocklehurst et al., 2012). Although the parameters for the lognormal calibration densities and the chosen soft maxima were somewhat arbitrary, this remains a problem with all avian divergence time studies (Brown and van Tuinen, 2011). Here, calibration

132 parameters are reported and marginal priors were calculated and reported so that marginal posterior estimates of divergences can be evaluated against actual marginal priors (Fig. 3.2). In the future, it is hoped that “total-evidence” approaches treating fossil taxa as terminal taxa (Pyron, 2011; Ronquist et al., 2012a; Wood et al., 2013) or other models explicitly modeling the fossil record will remove this subjective aspect of prior specification (Wilkinson et al., 2011; Warnock et al., 2012).

Root Calibration and Actual Marginal Priors A lognormal prior was placed on the root using the oldest well characterized neoavian taxon, Vegavis iaai (see below for details on the fossil) at 66 Ma ago. Initially, similar to Jetz et al. (2012), a lognormal prior with a hard minimum of 66.0 Ma ago, an σ of 0.5 (to place the mean further from the calibration point) and mean (in logspace) of 3.1 was used; this yielded a calibration density with a 95% quantile range of [74.33, 125.1] Ma ago. However, when empirical priors were calculated by running the analysis without data (see below for details), this led to a 95% prior highest probability density (HPD) range on the root of [110.6, 187.21] Ma ago. This was probably due to both the multiplicative construction of marginal prior in BEAST which can yield substantial mismatch between calibration densities and marginal priors (Heled and Drummond, 2011) and because the younger fossil calibration densities with long tails interact with the root prior. An attempt was made to use the calibrated yule prior in BEAST, which should correct the tree prior to input calibrations (Heled and Drummond, 2011) but this yielded very young marginal priors on the root (mean ~70 Ma ago) with narrow credibility intervals that did not match the input calibration. To obtain a more reasonable marginal prior, an additional uniform prior on the interval [66, 160] Ma ago was added to the root calibration and the mean of the lognormal prior was reduced to 2.5; this yielded a more reasonable, although still ancient marginal prior on the root with a mean of 132 Ma ago and HPD interval [109.1, 159.8]. To assess the effect of the antiquity of these priors, the analysis was also repeated with a hard maximum bound on the root set at 130.2 Ma ago, the age of

133 the Jehol Biota; all other parameters were identical.

Stem Lineages In all cases below, fossil calibrations were applied to the stem of an extant clade or taxon and the sister taxon was not necessarily constrained. BEAST v1.7.5 allows for stem lineages to be calibrated (the includeStem=”true” option) without a priori specification of which sister taxon and which node this calibrated; these were updated as tree operators modified the tree topology during MCMC sampling.

Stem Coliiformes (Fig. 3.1A) Taxon: Sandcoleus copiosus (Houde and Olson, 1992) Specimen: USNM 433973-434025, National Museum of Natural History, Smithsonian Institution, Washington D.C., USA Input Calibration Density: lognormal: mean = 2.03, offset = 56.2 Houde and Olson (1992) described the earliest representative of a stem coliiform fossil from the early Eocene Willwood formation of the United States. Several cladistic analyses have placed Sandcoleus on the Coliiform stem-lineage position along with other members of the extinct group Sandcoleidae (Mayr and Mourer-Chauvir, 2004; Zelenkov and Dyke, 2008; Ksepka and Clarke, 2009; 2010a). Sandcoleus has been used to calibrate the age of stem Coliiformes in several divergence time analyses (e.g. Ericson et al., 2006; Brown et al., 2008). The age of this fossil has been estimated to approximately 56.2–56.6 Ma ago (Secord et al., 2006; Ksepka and Clarke, 2009); therefore, a hard minimum of 56.2 Ma ago was applied to the divergence of stem Coliiformes.

Stem Strigiformes (Fig. 3.1B) Taxon: Ogygoptynx wetmorei (Rich and Bohaska, 1976) Specimen: AMNH 2653, American Museum of Natural History, New York, NY, USA Input Calibration Density: lognormal: mean = 2.02, offset = 56.99

134 Rich and Bohaska (1976) formerly described the oldest representative of Strigiiformes from the late Paleocene “Mason Pocket” of Colorado. Although no cladistic analysis of fossil Strigiformes has yet been performed (Kurochkin and Dyke, 2011), Mayr (2009a) argued that Ogygoptynx is likely more closely related to crown owls than younger European fossil taxa but that it was likely still a stem- lineage representative. This fossil has been used to provide a minimum divergence for stem Strigiformes in several divergence time studies (e.g. Ericson et al., 2006; Brown et al., 2008). Because it was not possible to locate exact geochronological estimates for the age of Mason Pocket, the upper boundary of the Tiffanian was adopted at 56.99 Ma ago. This may be overly conservative as the Mason Pocket may in fact be older (Secord et al., 2006: 233). Therefore a conservative hard minimum of 56.99 Ma ago was applied to the stem of Strigiformes.

Stem Group Threskiornithidae (Fig. 3.1C) Taxon: Rhynchaeites sp. (Mayr and Bertelli, 2011) Specimen: MGUH 20288, Geological Museum of the University of Copenhagen, Copenhagen, Denmark Input Calibration Density: lognormal: mean = 2.07, offset = 54.0 Mayr and Bertelli (2011) described a species of stem Threskiornithidae that they assign to Rhynchaeites based on shared derived characteristics with modern taxa. This genus is known from Messel (Mayr, 2009a) and this specimen was described from the Eocene Fur Formation of Denmark. The Eocene Fur formation has been constrained to 54 Ma ago (see below; Chambers et al., 2003). Therefore, this fossil of Rhynchaeites was used to justify a hard minimum divergence of 54 Ma ago for the stem divergence of extant Threskiornithidae.

Stem Trogoniformes (Fig. 3.1D) Taxon: Septentrogon madseni (Kristoffersen, 2002) Specimen: MGUH VP 3496, Geological Museum, University of Copenhagen, Copenhagen, Denmark Input Calibration Density: lognormal: mean = 2.07, offset = 54.0

135 Trogoniformes are a well established avian order with a single family Trogonidae. The earliest Trogoniformes were formally described by Kristoffersen (2002) from the early Eocene Fur Formation of Denmark. Although as far as can be ascertained, Septentrogon has not been subjected to a cladistic analysis its Trogoniform affinity was supported by several derived cranial morphologies it shares with crown-group Trogons (Kristoffersen, 2002; Mayr 2005a; 2009a; 2009b). This fossil has been used to calibrate stem Trogoniformes, whose monophyly has been confirmed by all major phylogenetic analysis (e.g. Hackett et al., 2008). This fossil has been used in several divergence time analyses (e.g. Ericson et al., 2006; Brown et al., 2008). The Fur formation has been recently been determined to be of Ypresian age, with two ash layers within the formation dated to 54.5 and 54 Ma ago (Chambers et al., 2003). This fossil of Septentrogon was therefore used to specify a hard minimum constraint on the divergence of stem Trogoniformes at 54 Ma ago.

Stem Group Pelecanidae (Fig. 3.1E) Taxon: Pelecanus sp. (Louchart et al., 2011) Specimen: NT-LBR-039, Private Collection of N. Tourment, Marseille, France Input Calibration Density: lognormal: mean = 2.45, offset = 28.1 Louchart et al. (2011) described a nearly complete skull of Pelecanus sp. From the Rupelian of France. They further argued that, although the fossil assignable to the extant genus Pelecanus due to highly derived beak morphology, it is likely outside the extant species in this genus. Because exact geochronology was not available for this site, the uppermost boundary of the Rueplian at 28.1 Ma ago was applied (Gradstein et al., 2012). Because Pelecanidae were represented by the single genus Pelecanus in this analysis, this fossil was used to justify a calibration on the minimum divergence for the stem of Pelecanus at 28.1 Ma ago.

Stem Group Phalacrocoracidae (Fig. 3.1F) Taxon: ?Burvocarbo soeffelensis (Mayr, 2007b) Specimen: PW 2005/5022-LS, Landesamt fur Denkmalpflege Rheinland-Pfalz,

136 Mainz, Germany Input Calibration Density: lognormal: mean = 2.49, offset = 24.52 Mayr (2007b) described a fossil ?Burvocarbo which he assigned tentatively to the Stem of Phalacrocraoidea (Anhingidae+Phalacrocaravidae, Livzey and Zusi, 2007) but possibly also on the stem of crown Phalacrocoracidae. The monophyly of Anhingidae+Phalacrocracidae is now strongly corroborated by most sources (e.g. Livezey and Zusi, 2007; Hackett et al., 2008; Mayr 2009a; this study). Since the fossil's formal description, two studies (Smith, 2010; Worthy, 2011), including one explicit cladistic analysis (Smith, 2010), have argued that ?Burvocarbo is indeed a representative of stem Phalacrocoracidae. This fossil was used to calibrate the stem of Phalacrocoracidae by Ericson et al., (2006). This is also congruent with stem representatives of Anhingidae, the sister taxon, from a similar age (Worthy, 2012). The locality is numerically constrained to MP28, whose lower age was estimated at 24.56±0.04 Ma ago (Mertz et al., 2007). This fossil of ?Burvocarbo was therefore used to justify a hard minimum age of 24.52 Ma ago for stem Phalacrocoracidae.

Stem Fregatidae (Fig. 3.1G) Taxon: Limnofregata azygosternon (Olson, 1977) Specimen: USNM 22753, National Museum of Natural History, Smithsonian Institution, Washington, D.C., USA Input Calibration Density: lognormal: mean = 2.11, offset = 51.56 Olson (1977) described the earliest known frigate bird relative from the early Eocene Green River formation. Further specimens and a new species of Limnofregata were described by Olson and Matsuoka (2005) and there was strong support for a stem lineage position for this taxon in a recent cladistic analysis (Mayr, 2009a; Smith, 2010). This calibration has been used in several divergence time studies (e.g. Ericson et al., 2006; Brown et al., 2008; Jetz et al., 2012). The bottom of the Green River Formation has been dated to 51.57±0.09 Ma ago (Smith et al., 2008) and, therefore, this fossil was used to justify a hard minimum bound on the stem lineage of Fregatidae at 51.56 Ma ago.

137 Stem Pici (Fig. 3.1H) Taxon: Rupelramphastoides knopf (Mayr, 2005a) Specimen: SMF Av 500a+b, Forschungsinstitut Senckenberg, Frankfurt, Germany Input Calibration Density: lognormal: mean = 2.45, offset = 28.1 Mayr (2005b) described a fossil Piciform from the Ruepelian of Germany. In the original description and subsequent work. Mayr (2005b; 2009a) and Mayr and Gregorová (2012) argued that although Rupelramphastoides may be a crown group representative of the Pici, it could not be conclusively demonstrated. Ericson et al., (2006) and Brown et al., (2008) applied this fossil to provide a minimum age for the divergence of crown Pici, while Jetz et al., (2012) interpreted the fossil as a stem-lineage representative. Here, the more conservative stem-lineage position was adopted and because exact geochronology is not available for the site (e.g. Mayr, 2005b), a hard minimum on the fossil's age was placed at the Rupelian/Chattian boundary at 28.1 Ma ago (Gradstein et al., 2012).

Stem Coraciidae+Brachypteraciidae (Fig. 3.1I) Taxon: Primobucco mcgrewi (Ksepka and Clarke, 2010b) Specimen: UWGM3299, University of Wyoming Geological Museum, Laramie, WY, USA Input Calibration Density: lognormal: mean = 2.11, offset = 51.56 Rollers are an extant avian group formed by the traditional avian families Coraciidae and Barchypteraciidae. Clarke et al. (2009) and Ksepka and Clarke (2010b) described new fossils from the early Eocene Green River Formation of the United States and performed a cladistic analysis placing the fossil taxon Primobucco on the roller stem lineage. Although this taxon has not, as far as could be ascertained, been used as a fossil calibration, the younger and closely related taxon Eocoracias from the middle Eocene Messel of Germany has been used in several studies (e.g. Ericson et al., 2006; Brown et al., 2008; Jetz et al., 2012). The bottom of the Green River Formation has been dated to 51.57± 0.09 Ma ago (Smith et al., 2008).The species-level RAxML analysis recovered the monophyly

138 of both Coraciidae and Barchypteraciidae as well as the clade uniting these groups (Fig. 3.1). Equally constrained phylogenetically to the same lineage as Eocoracias,this older fossil was therefore adopted and a hard minimum calibration was applied to the stem of the Craociide+Brachypteraciide clade at 51.56 Ma ago.

Stem Sphenisciformes (Fig. 3.1J) Taxon: Waimanu manneringi (Slack et al., 2006) Specimen: CMzfa35, Canterbury Museum, Canterbury, New Zealand Input Calibration Density: lognormal: mean = 1.94, offset = 60.5 Slack et al. (2006) described a fossil penguin and perform a cladistic analysis which places this taxon, Waimanu on stem of modern of Sphenisciformes. Further cladistic analysis by Clarke et al. (2007) and Kespka et al. (2012) corroborated the placement of Waimanu at the base of Sphensiciformes. This calibration has been used in most avian divergence studies (e.g. Brown et al., 2008; Haddrath and Baker, 2012; Jetz et al., 2012; etc.). Slack et al. (2006: Supplementary Materials) provided a justification for their estimate of the age of Waimanu manneringi and determined a range of 60.5-61.6 Ma ago. Therefore, this fossil was used to justify a hard minimum bound on the stem divergence of Sphenisciformes at 60.5 Ma ago.

Stem Upupiformes (Fig. 3.1K) Taxon: Messelirrisor grandis sp. Specimen: SMNK PAL 3803, Staatliches Museum fur Naturkunde, Karlsruhe, Germany Input Calibration Density: lognormal: mean = 2.174, offset = 47.5 Mayr (2000, 2006a, 2009a) described the extinct family Messelirrisoridae from the Middle Ecoene Messel deposits of Germany. Mayr (2006a) provided a cladistic analysis that places this family on the stem of extant Upupiformes, an extant order formed by the families Upupidae and Phoeniculidae (e.g. Hackett et al., 2008). This fossil calibration has been used by most divergence avian

139 divergence time studies (e.g. Ericson et al., 2006; Brown et al., 2008; Jetz et al., 2012). The lower boundary on the age of the Messel deposits has recently been dated to 47.5-49 Ma ago (Franzen, 2005) and therefore 47.5 Ma ago was adopted as a hard minimum calibration for stem Upupiformes (Upupidae+Phoeniculidae) based on this fossil.

Stem Pandionidae (Fig. 3.1L) Taxon: “Palaeocircus ciivieri” (Harrison and Walker, 1976) Specimen: BMNH A2338, British Museum of Natural History, London, United Kingdom Input Calibration Density: lognormal: mean = 2.37, offset = 33.9 Harrison and Walker (1976) described the earliest known occurrence of a member of the Crown group Pandionidae which is represented by a single extant taxon Pandion. Although Mayr (2006b; 2009a) disagreed with assignment of this fossil to Paleocircus viiveri, the ungule phalanx is diagnostic of Pandionidae (Mayr, 2006b). This calibration was used by several divergence time studies to calibrate stem Pandionidae as there is a only a single taxon Pandion in the crown group on most analyses (Ericson et al., 2006; Brown et al., 2008). The exact geochronology of the specimen was not available, but a conservative lower bound on the age of the fossil was provided by the Oligocene/Eocene boundary at 33.9 Ma ago (Gradstein et al., 2012). Because Pandion was represented by a single taxon in this analysis, the fossil provided a hard minimum divergence of 33.9 Ma for the stem of Pandionidae rather than the crown group which was not defined.

Stem Gruoidea (Fig. 3.1M) Taxon: Parvigrus pohli (Mayr et al., 2005c) Specimen: WDC-CCF 02, Wyoming Dinosaur Center, Thermopolis, WY, USA. Input Calibration Density: lognormal: mean = 2.45, offset = 28.1 Mayr et al. (2005c) described a nearly complete fossil relative of extant Grues. Although in the original description, Mayr (2005c) considered Parvigrus sister to the extant clade Gruidae+Aramidae, subsequent revision and cladistic analysis

140 (Mayr 2009a; 2013) has placed the taxon as sister to the clade Gruoidea (Psophidae+Gruidae+Aramidae).This fossil was used to calibrate Stem Gruidae+Aramidae by Ericson et al. (2006), Brown et al. (2008) and Jetz et al., (2012). Therefore, this fossil provides a hard minimum divergence for the stem of extant Gruoidea, a clade that was recovered in the genus-level RAxML analysis (Fig. 3.1). Because exact geochronology was not available for the site, the upper boundary at 28.1 Ma ago of the Ruepelian (Gradstein et al., 2012) was used as a conservative estimate of the fossil's age and was adopted as a hard minimum age for the divergence of stem Gruoidea.

Stem Psittaciformes (Fig. 3.1N) Taxon: Cyrilavis colburnorum (Ksepka et al., 2011b) Specimen: FMNH PA 754 Department of Geology, Field Museum of Natural History, Chicago, IL, USA Input Calibration Density: lognormal: mean = 2.11, offset = 51.56 The calibration of the Psittaciformes has been contentious. Several divergence time studies (e.g. Wright et al., 2008; Kundu et al., 2012; Pacheco et al., 2011; Jetz et al., 2012) have applied a minimum divergence time of approximately 50– 54 Ma ago based on the early Eocene putative crown group taxon Mopsitta (Waterhouse et al., 2008). However, the assignment of this taxon, represented by a single humerus, to the crown of Psittaciformes has been controversial and has not been supported by any cladistic analysis nor has any Eocene fossil been recovered as a member of crown Psittaciformes in any cladistic analysis (Mayr 2009a; Ksepka et al., 2011b; Mayr and Bertelli, 2011; Ksepka and Clarke, 2012). Although additional analysis or material may confirm Mopsitta or another fossil taxon as a crown Psittaciform, it was not possible to apply a crown calibration given the lack of consensus in the literature. The stem lineage of Psittaciformes has been comparatively well sampled from the early-late Eocene and stem affinities of several taxa have been corroborated by cladistic analysis (Ksepka et al., 2011b; Ksepka and Clarke, 2012). Several possibly slightly older taxa have been described from the London Clay (Mayr, 2009a) but are not as tightly

141 constrained temporally. Therefore, this fossil of Cyrilavis from the temporally well calibrated (Smith et al., 2008) Fossil Butte of the Green River formation was used to place a hard minimum divergence on the stem of Psittaciformes at 51.56 Ma ago.

Stem Apodiformes (Apodidae, Hemiprocnidae, Trochilidae; Fig. 3.1O) Taxon: Eocypselus vincenti (Harrison, 1984) Specimen: MGUH 26730 and MGUH 26729, Geological Museum of the University of Copenhagen, Copenhagen, Denmark Input Calibration Density: lognormal: mean = 2.07, offset = 54.0 Originally described by Harrison (1984) from the London Clay (Dyke et al., 2004), additional material was described from the Eocene Fur Formation by Dyke et al. (2004) and Mayr (2010). Although always considered an Apodiform (Dyke et al., 2004), the exact position of Eocypselus has been debated (Mayr, 2009a). Mayr (2009a; 2010) argued that Eocypselus is an apodiform stem lineage fossil and because this is more conservative with respect to setting a minimum divergence time than the more inclusive placement by Dyke et al. (2004), this was adopted here. As described above, the Fur formation has been dated to 54.5 and 54 Ma ago (Chambers et al., 2003). This fossil of Eocypselus was therefore used to justify a minimum divergence for stem Apodiformes at 54 Ma ago.

Stem Apodidae (Fig. 3.1P) Taxon: Scaniacypselus szarskii (Mayr and Peters, 1999) Specimen: SMF-ME 3576, Forschungsinstitut Senckenberg, Frankfurt am Main, Germany Input Calibration Density: lognormal: mean = 2.174, offset = 47.5 Originally referred to the Hemiprocidae in the original description by Peters (1985), Mayr and Peters (1999) referred Scaniacypselus to the stem-lineage of Apodidae (Mayr, 2003). This position was corroborated by cladistic analysis (Mayr, 2003). This calibration was used by Jetz et al., (2012). As described above, the Messel of Germany has been dated to 47.5 Ma ago (Franzen, 2005).

142 Therefore, this fossil of Scaniacypselus was used to provide a hard minimum divergence for the stem lineage of Apodidae.

Stem Phaethontidae (Fig. 3.1Q) Taxon: Lithoptila abdounensis (Bourdon et al., 2005) Specimen: OCP.DEK/GE 1087, Office Chérifien des Phosphates, Direction des Exploitations de Khouribga, Severice de Géologie, Khouribga, Morocco. Input Calibration Density: lognormal: mean = 2.03, offset = 55.8 Bourdon et al. (2005) described a species of an extinct stem lineage of the tropicbirds, the Prophaethonitidae (Harrison and Walker, 1976). The sister-group relationship between tropicbirds and Prophaethontidae has been corroborated by two cladistic analyses (Bourdon et al., 2005; Smith, 2010). This fossil was used to calibrate stem Phaethontidae by Ericson et al. (2006) and Brown et al., (2008). The age of the fossil locality was reported as upper Paleocene (Thanetian) by Boudon et al. (2005) and because precise geochronology was not available for this site, the Paleocene/Eocene boundary was selected as conservative estimate for the age of the fossil (56.0 Ma ago; Gradstein et al., 2012), and therefore, this fossil of Lithoptila was used to justify a hard minimum estimate for the divergence of stem Phathonidae at 56.0 Ma ago.

Stem Group Pteroclididae (Fig. 3.1R) Taxon: Leptoganga sepultus, (Mourer-Chauviré, 1993) Specimen: PFY 11283, USTL Montpellier, France Input Calibration Density: lognormal: mean = 2.49, offset = 24.52 Mayr-Chauviré (1993) described new collections of the fossil Leptoganga from the Late Oligocene of Quercy, France. Mayr (2009a) further argued that Leptoganga is a member of the stem of extant Pteroclididae. This fossil was used as a minimum calibration on Pteroclididae by Ericson et al. (2006) and Brown et al. (2008). Recovered from MP28 (Mourer-Chauviré, 1993), the age of this fossil is numerically constrained to 24.52 Ma ago (see above; Mertz et al., 2007). Mourer-Chauviré (1993) notes other fossils assigned to Pteroclididae in the genus

143 Archaeoganga that may be slightly older (~28.1 Ma ago) but these are from older collections that lack stratigraphic attribution. Therefore, this fossil of Leptoganga was used to justify a hard minimum bound on the divergence of the stem of Pteroclididae of 24.52 Ma ago.

Stem Anatidae (Fig. 3.1S) Taxon: Vegavis iaai (Clarke et al., 2005) Specimen: MLP 93-I-3-I, Museo de La Plata, La Plata, Argentina Input Calibration Density: lognormal: mean = 1.83, offset = 66.0 Clarke et al (2005) described and performed a cladistic analysis of a Cretaceous member of Crown Anseriformes. Based on three successive phylogenetic analyses, Clarke et al. placed Vegavis on the stem of the anseriform family Anatidae that includes all extant Anseriformes with the exception of Anhimidae (Chauna and Anhima) and the Magpie goose (Anseranas). The cladistic analysis was equivocal as to whether Vegavis was the sister taxon of crown Anatidae, or more closely related to an other fossil taxon Presbyornis. This calibration has been used extensively by other studies including by Brown et al. (2007; 2008), Pacheco et al. (2011), Haddrath and Baker (2012) and Jetz et al. (2012). No subsequent study has revised Vegavis' position on the stem of Anatidae. The species and genus-level phylogenetic analyses conducted here supported the same phylogenetic arrangement of Chauna, Anseranas and Anatidae as Clarke et al. (2005). Vegavis was reported from the Maastrictian sediments of Vega island, Antarctica. A confidence range for the age of the specimen was provided between 66 and 68 Ma ago, and therefore, the earlier value of 66.0 Ma ago was applied as a hard minimum age to Stem Anatidae based on this fossil.

Stem Cariamiformes (Fig. 3.1T) Taxon: Paleopsilopterus itaboraiensis (Alvarenga, 1985) Specimen: MNRJ-4040-V, Museu Nacional da Universidade Federal do Rio de Janeiro, Brazil

144 Input Calibration Density: lognormal: mean = 2.13, offset = 50.2 Although they are represented by two extant genera Cariama and Chunga, the Cariamiformes (or Cariamidae) have an extensive Paleogene fossil record (Mayr 2009a). The oldest formerly described stem lineage fossil attributed to Cariamiformes is the Phorusrhacid Paleopsilopterus (Alvarenga, 1985). This taxon is recognized as the oldest Phorursracid (e.g. Alvarenga and Holfling, 2003), an extinct lineage on the stem of Cariamiformes known from North and South America, Europe and Africa (Alvarenga and Hofling, 2003; Mayr, 2009a; Mourer-Chauviré et al., 2011; Alvarenga et al., 2011). Paleopsilopterus was described from the late Paleocene, potentially early Eocene of Brazil; a lava flow above the filling was dated at 52.6±2.4 Ma ago (Oliveira and Goin, 2011) and therefore a conservative hard minimum for the divergence of stem Cariamiformes was adopted at 50.2 Ma ago based on this fossil.

Stem Jacanidae (Fig. 3.1U) Taxon: Nupharanassa tolutaria (Rasmussen et al., 1987) Specimen: DPC2580, Duke University Primate Center, NC, USA Input Calibration Density: lognormal: mean = 2.39, offset = 33.0 Rasumessen et al. (1987) described a stem representative of the Jacanidae, a family of birds in the Charadriiformes. This fossil, along with two others found in the same formation shared diagnostic characteristics with extant Jacanidae (Rasmussen et al., 1987; Mayr, 2009a). This fossil has been used as a calibration in several divergence time studies (e.g. Ericson et al., 2006; Brown et al., 2008; Haddarth and Baker, 2012). Although originally thought to be of early Oligocene in age, Seiffert (2006) revised the age of Quarry E (the locality of Nupharamassa tolutaria) to be approximately 33 Ma ago. Therefore, this fossil was used to justify a hard minimum of 33 Ma ago on the divergence of the stem of Jacanidae.

Stem Group Turnici (Fig. 3.1V) Taxon: Turnipax oechslerorum (Mayr and Knopf, 2007) Specimen: SMF Av 506a+b:, Forschungsinstitut Senckenberg, Frankfurt am Main,

145 Germany Input Calibration Density: lognormal: mean = 2.45, offset = 28.1 Mayr and Knopf (2007) described a fossil buttonquail from the Lower Oligocene Rupelian of Germany. Although traditionally related to the quails, the buttonquails are now considered derived charadriiforms (e.g. Fain and Houde, 2004; Ericson et al., 2006; Hackett et al., 2010). Mayr and Knopf (2007) and Mayr (2009a) argued for a phylogenetic position on the stem of Turnici based on several synapomorphies and also noted a mosaic of charadriiform characters. Because the exact geochronology for the site was not available, the uppermost boundary of the Rupelian at 28.1 Ma ago (Gradstein et al., 2012) was used as a hard minimum bound on the divergence of stem Turnici based on this fossil (see above).

Stem Group Alicdae (Fig. 3.1W) Taxon: Alicdae incertae sedis (Chandler and Parmley, 2002) Specimen: GCVP 5690, Georgia College Vertebrate Paleontology Collection, Milledgeville, GA, USA Input Calibration Density: lognormal: mean = 2.37, offset = 34.2 Chandler and Parmley (2002) reported a distal humerus attributable to Alicdae from the Late Eocene Clinchfield Formation of Georgia (Smith and Clarke, 2011). Mayr (2009a; 2011b) and Smith and Clarke (2011) argued for a position on the stem lineage of Alicdae which was supported by the conclusions of a cladistic analysis reported by Smith (2011). The Clinchfield Formation has been dated to 34.2-36 Ma ago and therefor the upper limit of 34.2 Ma ago was used to specify a hard bound on the divergence of stem Alcidae (Smith and Clarke, 2011) based on this fossil.

Stem/Crown Charadriiformes Jetz et al. (2012: Supplementary Materials) included a hard minimum calibration on the divergence of crown Charadriiformes based on the Eocene Fur formation fossil Morsoravis sedilis (Bertelli et al., 2010). Jetz et al. (2012) may

146 have assigned this fossil to the crown of Charadriiformes based on the cladogram figured in Dyke and van Tuinen (2004: Fig. 7), which was cited. However, in the formal description, Bertelli et al. (2010) performed a cladistic analysis that recovered Morsoravis as the sister taxon to the two extant charadriiform groups included in the analysis (Recurvirostridae and Burhinidae). This placement can only justify assignment of Morsoarvis to the stem of Charadriiformes. Mayr (2011a) argued that Morsoarvis may not be a charadriiform and reported results of a cladistic analysis using similar taxaonomic and character sampling as Bertelli et al. (2010) but including the Messel taxon Pumiliornis which he recovered as sister taxon to Morsoarvis. Mayr (2011a) also included only the same two charadriiform taxa. Given the completeness of the specimen, the authors of this study believe that the affinities of Morsoarvis might be elucidated with an updated cladistic analysis including denser sampling from extant Charadriiformes. Until such time as the phylogenetic position of Morsoarvis is examined further, it was not possible to use this fossil with confidence to provide a minimum divergence time for stem Charadriiformes.

Relaxed Clock Priors An uncorrelated, relaxed lognormally-distributed molecular clock was used to model rate variation across the phylogeny (Drummond et al., 2006). A diffuse informative prior for the mean of the lognormal distribution was estimated by first conducting a strict clock-based analysis using the same data, priors and substitution models. This analysis was conducted using two independent BEAST runs for 20 x 106 generations until the posterior and clock rate parameters reached effective sample sizes > 200 after runs were combined and a burnin of 5 x 106 generations was discarded (e.g. Drummond et al., 2006). These analyses were conducted twice (total four runs) with both uniform [0,1] and exponential (mean = 0.1) priors on the clock rate. Posterior estimates from each set of runs were nearly identical and were averaged to produce a mean substitution estimate of 8.59E-3 substitutions per site per Ma. A diffuse prior was placed on the mean of the lognormal branch rate distribution by using a lognormal distribution with a mean

147 of 8.59E-3 and a scale parameter of 2.5. The default BEAST prior was applied to the standard deviation of the lognormal branch rate distribution.

Divergence Time Runs The divergence time analysis was executed using the CLUMEQ/Colosse and CLUMEQ/Guillimin supercomputing clusters. BEAST v1.7.5 (Drummond et al., 2012) was executed using 8 independent runs using the BEAGLE SSE option (Ayres et al., 2012) and 6 or 8 cores per run. To aid convergence in this large data set, the starting topology was set to the TMLE from the full genus-level supermatrix phylogenetic analysis (Fig. 3.1). Furthermore, nodes from that analysis receiving greater than 0.95 bootstrap support proportions (see green nodes in Fig. 3.1) were used to define monophyly constraints along with the nodes Neoaves (e.g. Ericson et al., 2006; Hackett et al., 2008) and Passeriformes+Psittaciformes (e.g. Hackett et al., 2008; Suh et al., 2011; McCormack et al., 2013). The analysis was first executed without data to generate samples from the prior and determine marginal priors on node ages (see Fig. 3.2). The final BEAST MCMC analysis was executed for at least 100 x 106 generations for each chain, sampling every 2000 generations. The results of each independent run were then combined after discarding a burnin fraction of 10 x 106 generations. Convergence was assessed visually using Tracer v1.5 for each independent run to ensure each had converged to a nearly identical posterior distribution of likelihoods and divergence times (Rambaut and Drummond, 2007). Sufficient sampling of the posterior was assessed by ensuring that all parameters had effective sample sizes (ESS) greater than 200 (Drummond et al., 2006). Most parameters had ESS > 10000 and most divergence times had ESS > 300, while some divergence times with wide credibility intervals had 200 < ESS < 300. The posterior distribution of chronograms was summarized by first combining the trees from each independent run after discarding a burnin fraction of 10 x 106 generations. The tree sample was then thinned using logcombiner from the BEAST package to produce a posterior sample of 10 000 chronograms. This sample was then summarized using the treeannotator program from the BEAST

148 package to calculate the maximum clade credibility chronogram (TMCC) with nodes placed at the mean height of the marginal posterior on node ages (-heights mean option). Treeannotator was used to further annotate TMCC with clade posterior probabilities and 95% highest posterior densities (HPD credibility intervals) for node ages. BEAST XML controls files for analyses described above are available from the authors upon request.

149 Megapodius_eremita 1 1 Megapodius_freycinet Alectura_lathami Numida_meleagris 1 Acryllium_vulturinum Callipepla_gambelii 1 1 Colinus_cristatus 1 0.99 Colinus_virginianus 1 Rollulus_rouloul Dendragapus_obscurus 1 1 0.99 Falcipennis_canadensis Meleagris_gallopavo 1 Alectoris_graeca 1 0.8 Francolinus_capensis 0.47 Coturnix_chinensis 0.97 0.47 1 Coturnix_coturnix 1 Coturnix_japonica Gallus_gallus Ortalis_canicollis 1 Crax_rubra 1 Crax_alector Chauna_torquata Anseranas_semipalmata 0.98 Dendrocygna_arcuata 0.99 Biziura_lobata Anas_crecca 0.55 1 Anas_platyrhynchos 1 1 0.2 Aythya_americana Malacorhynchus_membranaceus 0.62 Branta_canadensis 0.97 0.81 1 Anser_albifrons 1 Anser_anser 0.97 Anser_erythropus Oxyura_jamaicensis

1

Struthio_camelus Pterocnemia_pennata 1 Rhea_americana 0.99 Dinornis_giganteus 1 Anomalopteryx_didiformis 1 Emeus_crassus 0.99 0.99 Eudromia_elegans 0.95 Nothoprocta_perdicaria 1 Crypturellus_undulatus 0.99 Crypturellus_soui 0.99 0.61 Tinamus_major 0.98 Tinamus_guttatus Dromaius_novaehollandiae 1 Casuarius_casuarius 0.88 Apteryx_haastii 1 Apteryx_australis

150 Opisthocomus_hoazin Guira_guira 1 Crotophaga_ani 1 0.99 Crotophaga_sulcirostris Neomorphus_geoffroyi 0.99 Geococcyx_californianus Phaenicophaeus_curvirostris 1 1 Coccyzus_erythropthalmus 1 1 Coccyzus_americanus Eudynamys_taitensis 1 0.99 0.5 Cuculus_canorus Coua_cristata 0.96 Centropus_viridis Ardeotis_kori Afrotis_afra 1 1 0.93 Eupodotis_senegalensis Eupodotis_ruficrista 0.39 Aptornis_defossor Podica_senegalensis 1 0.48 0.88 Heliornis_fulica Sarothrura_elegans 0.77 Fulica_americana Himantornis_haematopus 0.24 0.99 0.62 Rallus_limicola 0.55 0.67 Porphyrio_martinica 0.48 Aramides_ypecaha 0.68 Laterallus_melanophaius 0.99 Laterallus_albigularis Aramus_guarauna 1 Grus_canadensis 1 1 Grus_grus Psophia_viridis 1 Psophia_crepitans 0.52 Psophia_leucoptera Gavia_stellata 1 Gavia_arctica 1 Gavia_pacifica 0.95 Gavia_adamsii 1 Gavia_immer Oceanites_oceanicus Phoebastria_nigripes 1 Thalassarche_melanophrys 1 Phoebetria_palpebrata 1 Pterodroma_hasitata 0.31 Pterodroma_brevirostris Puffinus_griseus 0.45 1 Puffinus_tenuirostris 0.57 0.17 0.49 Puffinus_pacificus 0.91 0.68 0.39 Calonectris_diomedea 0.84 Puffinus_lherminieri 0.99 1 0.57 Pelecanoides_urinatrix 1 Fulmarus_glacialis Pachyptila_desolata Oceanodroma_tethys 1 Hydrobates_pelagicus Aptenodytes_patagonicus 1 Eudyptes_chrysocome Spheniscus_magellanicus 1 1 0.92 Spheniscus_humboldti Eudyptula_minor Ciconia_ciconia 1 0.86 Ciconia_boyciana 1 Mycteria_americana 0.88 Jabiru_mycteria Theristicus_caerulescens 1 1 Theristicus_caudatus Eudocimus_albus 1 Plegadis_falcinellus Platalea_minor 0.96 0.99 0.99 Platalea_ajaja Nipponia_nippon Pelecanus_thagus 1 0.87 1 1 Pelecanus_occidentalis 1 Pelecanus_erythrorhynchos Pelecanus_conspicillatus 0.86 0.73 1 Pelecanus_onocrotalus Balaeniceps_rex 0.8 Scopus_umbretta 0.81 Cochlearius_cochlearius 0.63 Tigrisoma_lineatum 1 Tigrisoma_fasciatum 1 Botaurus_lentiginosus Egretta_eulophotes 0.86 1 1 Egretta_novaehollandiae 0.79 Ardea_alba 1 Ardea_herodias 1 0.26 Ardea_cocoi Nyctanassa_violacea 0.51 Nycticorax_nycticorax Fregata_minor 0.99 Fregata_aquila 0.99 Fregata_magnificens Sula_dactylatra 0.98 1 Morus_bassanus Phalacrocorax_auritus 0.99 1 Phalacrocorax_carbo 0.93 Phalacrocorax_urile 0.99 1 Phalacrocorax_pelagicus Anhinga_anhinga Musophaga_violacea 1 Tauraco_erythrolophus Corythaixoides_leucogaster 1 0.99 Crinifer_piscator 0.63 Corythaixoides_concolor 1 0.82 Corythaixoides_personata Corythaeola_cristata Eurypyga_helias 1 0.76 Rhynochetos_jubatus Nyctibius_bracteatus 0.74 Nyctibius_grandis 0.9 1 Nyctibius_aethereus 0.7 Nyctibius_maculosus 0.9 0.35 Nyctibius_griseus Steatornis_caripensis Podargus_strigoides 1 0.99 Batrachostomus_septimus Eurostopodus_macrotis 1 Caprimulgus_longirostris 0.5 1 Caprimulgus_europaeus 0.74 Podager_nacunda 1 Chordeiles_minor Aegotheles_insignis 0.65 1 Aegotheles_albertisi 0.85 Aegotheles_cristatus Colibri_coruscans Hylocharis_chrysura 1 0.94 1 0.99 Archilochus_colubris 1 Metallura_eupogon Phaethornis_griseogularis 0.99 Phaethornis_syrmatophorus 1 Glaucis_hirsuta 1 1 Glaucis_aeneus Hemiprocne_comata 1 Hemiprocne_mystacea 0.54 Hemiprocne_longipennis 1 Cypseloides_niger 1 Streptoprocne_zonaris 1 Aeronautes_saxatalis Apus_pallidus 0.39 1 1 0.36 Apus_apus Aerodramus_vanikorensis Monias_benschi 1 Mesitornis_unicolor Chalcophaps_indica 1 Turtur_tympanistria Columbina_squammata 0.23 1 Columbina_passerina Columba_leucocephala 0.24 1 Columba_livia Geotrygon_violacea 0.7 1 1 Geotrygon_montana 1 Zenaida_auriculata 0.12 1 0.78 Zenaida_macroura Leptotila_verreauxi Ducula_aenea 0.37 Otidiphaps_nobilis 0.44 Didunculus_strigirostris 1 0.6 Raphus_cucullatus 0.2 0.51 0.98 0.56 Pezophaps_solitaria Goura_cristata Gallicolumba_luzonica 0.69 Geopelia_striata Treron_sieboldii 1 Treron_vernans Phaethon_rubricauda 1 Phaethon_lepturus 0.12 0.42 Pterocles_bicinctus 1 Pterocles_namaqua 0.85 Pterocles_gutturalis 0.71 Pterocles_coronatus 0.62 Syrrhaptes_paradoxus Tachybaptus_novaehollandiae Podiceps_cristatus 1 0.81 0.16 1 Podiceps_auritus 1 Podiceps_nigricollis 1 Podilymbus_podiceps Phoenicopterus_chilensis 1 Phoenicopterus_ruber Scolopax_minor 1 Arenaria_interpres 0.82 Phalaropus_tricolor 0.99 Tringa_incanus 1 Rostratula_benghalensis 1 Jacana_spinosa 1 1 Jacana_jacana Pedionomus_torquatus 1 Attagis_gayi 1 Thinocorus_orbignyianus 1 0.99 Thinocorus_rumicivorus Turnix_suscitator 1 Turnix_sylvatica 0.6 Turnix_pyrrhothorax 0.85 Turnix_varia Glareola_nordmanni 1 Glareola_pratincola 0.99 0.92 Cursorius_temminckii Chlidonias_niger Rissa_tridactyla 0.99 1 Larus_atricilla 0.98 Larus_dominicanus 0.92 0.5 1 0.98 Larus_marinus 0.99 Rynchops_niger Stercorarius_maccormicki 1 Stercorarius_skua 0.7 0.96 Cepphus_columba 1 Uria_lomvia 0.53 Synthliboramphus_antiquus Dromas_ardeola Burhinus_senegalensis 0.73 Burhinus_bistriatus 0.76 0.95 Burhinus_oedicnemus Chionis_alba 0.98 Pluvialis_dominica Himantopus_mexicanus 0.99 Cladorhynchus_leucocephalus 0.73 0.99 Haematopus_ostralegus 0.7 0.97 0.95 Haematopus_ater 0.78 Ibidorhyncha_struthersii Charadrius_collaris Charadrius_semipalmatus 0.91 1 0.96 Charadrius_vociferus Phegornis_mitchelli

151 Cariama_cristata Micrastur_semitorquatus 1 1 Micrastur_gilvicollis Herpetotheres_cachinnans Caracara_plancus 1 1 Daptrius_ater 1 Falco_tinnunculus Falco_sparverius 1 0.46 Falco_rufigularis 0.75 Falco_peregrinus 0.89 0.39 Falco_mexicanus Falco_subbuteo Picathartes_gymnocephalus Bombycilla_cedrorum 1 Bombycilla_garrulus Passer_domesticus 1 0.25 Passer_montanus 1 Fringilla_montifringilla 0.86 0.77 0.54 Hemispingus_frontalis 1 0.99 Motacilla_alba Ploceus_cucullatus 0.44 1 Taeniopygia_guttata 0.75 Vidua_chalybeata Regulus_calendula Zosterops_japonica 0.97 Sylvia_atricapilla 0.52 1 Sylvia_crassirostris 0.7 1 0.99 Sylvia_nana 0.99 0.94 Hypsipetes_amaurotis Phylloscopus_occipitalis 0.77 Acrocephalus_scirpaceus Parus_atricapillus 1 Parus_major 0.78 Catharus_guttatus 0.99 1 0.71 Turdus_falcklandii Corvus_corone 1 0.99 Corvus_frugilegus 0.99 1 Aphelocoma_ultramarina Lanius_collurio Malurus_melanocephalus 0.99 Ptilonorhynchus_violaceus 0.9 Climacteris_erythrops Menura_novaehollandiae Sapayoa_aenigma Pitta_erythrogaster 0.99 1 1 Pitta_guajana 0.78 Smithornis_sharpei 1 Smithornis_rufolateralis Lepidothrix_coronata 0.99 Todirostrum_cinereum 1 1 Mionectes_macconnelli 1 Sayornis_phoebe 0.7 Cnemotriccus_fuscatus 1 Pitangus_sulphuratus 0.99 0.99 0.73 Tyrannus_tyrannus Thamnophilus_nigrocinereus Scytalopus_magellanicus 0.99 Lepidocolaptes_wagleri 0.97 1 0.99 Dendrocolaptes_certhia 0.99 Cranioleuca_baroni Grallaria_varia 1 Grallaria_squamigera Acanthisitta_chloris Nymphicus_hollandicus 1 Cacatua_galerita 0.82 0.86 1 Cacatua_sulphurea Nandayus_nenday 1 Pyrrhura_frontalis 0.71 0.46 Aratinga_pertinax 1 Amazona_autumnalis 0.84 1 Brotogeris_cyanoptera 0.96 Forpus_modestus Psittacus_erithacus Psittacula_alexandri 1 0.93 0.97 Alisterus_scapularis 0.99 Micropsitta_finschii 0.99 Platycercus_elegans Melopsittacus_undulatus 0.8 1 Trichoglossus_haematodus 1 0.57 Chalcopsitta_cardinalis Neophema_elegans 0.6 Agapornis_roseicollis Strigops_habroptilus Coragyps_atratus 1 Cathartes_aura 0.64 Sarcoramphus_papa 0.92 Sagittarius_serpentarius 0.97 Pandion_haliaetus Gampsonyx_swainsonii 0.99 Leptodon_cayanensis 1 Buteogallus_meridionalis 1 Buteo_buteo 1 1 1 Buteo_jamaicensis Accipiter_gentilis 0.99 0.41 Accipiter_striatus Nisaetus_alboniger 1 1 Nisaetus_nipalensis Gyps_africanus Leptosomus_discolor Pharomachrus_auriceps 1 Pharomachrus_mocinno 1 Harpactes_ardens 0.67 Trogon_melanurus 1 Trogon_curucui 1 1 Trogon_viridis 0.79 Trogon_personatus Bucorvus_abyssinicus 1 Bucorvus_leadbeateri Tockus_flavirostris 1 1 0.99 Tockus_erythrorhynchus 1 Tockus_camurus Aceros_corrugatus 0.66 0.88 0.96 Bycanistes_brevis 0.66 Rhinopomastus_cyanomelas 0.77 1 Phoeniculus_purpureus Upupa_epops Malacoptila_semicincta Galbula_pastazae 0.85 1 Galbula_ruficauda 0.99 Galbula_cyanescens 0.95 0.7 Galbula_albirostris Monasa_nigrifrons 0.95 Nystalus_maculatus 0.49 Bucco_macrodactylus 0.79 Indicator_minor 0.87 1 Indicator_variegatus 0.99 Indicator_maculatus 1 Jynx_torquilla 1 Picoides_villosus 1 Dryocopus_pileatus 1 0.97 Colaptes_auratus 1 Picumnus_cirratus Megalaima_haemacephala 1 Megalaima_virens Trachyphonus_darnaudii 1 Capito_niger 0.95 Semnornis_ramphastinus 0.77 0.92 0.94 Ramphastos_vitellinus 0.75 Pteroglossus_castanotis 0.97 1 Pteroglossus_azara 0.97 0.93 Andigena_cucullata 0.99 0.95 Selenidera_reinwardtii Aulacorhynchus_prasinus Pogoniulus_pusillus Todus_angustirostris 1 Todus_mexicanus 0.99 Todus_subulatus 1 Momotus_momota Ispidina_picta 1 1 Alcedo_leucogaster 0.7 1 Alcedo_atthis Megaceryle_alcyon 0.84 0.99 Todiramphus_sanctus Merops_apiaster 1 Merops_nubicus 0.62 Merops_viridis Coracias_spatulata 0.6 1 Coracias_garrulus 0.99 1 Coracias_caudata 0.99 Eurystomus_orientalis Brachypteracias_squamigera 0.99 Brachypteracias_leptosomus Phodilus_badius 1 Tyto_alba 1 Ninox_novaeseelandiae 1 Athene_cunicularia 0.92 Otus_scops Strix_occidentalis 0.98 0.99 0.71 Strix_varia 0.2 Asio_otus 1 Asio_flammeus Urocolius_macrourus Colius_striatus 1 1 0.99 Colius_colius Urocolius_indicus

152 Supplementary Figure 3.1. (Three Pages). Maximum likelihood estimate of topology from the RAxML analysis of the 450 taxon species-level supermatrix. The highest likelihood topology (ln(L) = -1390705) out of 512 independent tree searches is annotated with bootstrap support values from 512 bootstrap replicates (proportions). The tree was rooted with a crocodylian outgroup (not shown) and figured branch lengths are drawn with arbitrary lengths to increase readability.

153 Divergence Times (160 Ma Ago Hard Maximum) (160 Ma Ago Hard Times Divergence

Divergence Times (130.2 Ma Ago Hard Maximum)

154 Supplementary Figure 3.2. Sensitivity of the divergence time analysis to hard maximums on tree age. Comparison of mean divergence times between equivalent nodes among the maximum clade credibility trees of the main divergence time analysis (hard maximum tree age of 160 Ma ago) and an alternative analysis with a hard maximum of 130.2 Ma ago on the root age. Dates are highly correlated: Pearson's r = 0.9997. The eldest divergences are actually slightly older when a hard maximum is placed on the root at 130.2 Ma ago.

155 Appendix 3.1. Operational taxonomy and Genbank taxonomic IDs NCBI Species Synonym Common Name H20081 TaxID Acanthisitta chloris 57068 Rifleman Y Accipiter gentilis 8957 Goshawk N Sharp-shinned Accipiter striatus 56330 N Hawk Aceros corrugatus 175825 Wrinkled hornbill N Eurasian reed Acrocephalus scirpaceus 48156 N warbler Vulturine Acryllium vulturinum 8992 N guineafowl Mountain Owlet- Aegotheles albertisi 48277 N nightjar Australian Owlet- Aegotheles cristatus 48279 N nightjar Euaegotheles Feline Owlet- Aegotheles insignis 671282 Y insignis nightjar Aerodramus vanikorensis 243317 Uniform Swiftlet Y White-throated Aeronautes saxatalis 190674 N Swift Southern Black Afrotis afra Eupodotis afra 156154 N Korhaan Peach-faced Agapornis roseicollis 60468 N Lovebird Common Alcedo atthis 36245 N Kingfisher White-bellied Alcedo leucogaster 325337 Y Kingfisher Alectoris graeca 40178 Rock Partridge N Australian Brush- Alectura lathami 81907 Y Turkey Australian King Alisterus scapularis 458117 Y Parrot American Alligator mississipiensis 8496 Y Alligator 1Species included in the Hackett et al. (2008) analysis

156 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Amazona autumnalis 151759 Red-lored Amazon N Anas crecca 75839 Common Teal N Anas platyrhynchos 8839 Mallard Y Andigena Hooded Mountain- Andigena cucullata 240742 N cuculata Toucan Anhinga anhinga 56067 American Darter Y Anomalopteryx didiformis 8811 Little Bush Moa N White-fronted Anser albifrons 50365 N Goose Anser anser 8843 Domestic Goose N Lesser White- Anser erythropus 132586 Y fronted Goose Anseranas semipalmata 8851 Magpie Goose Y Aphelocoma ultramarina 55973 Mexican Jay N

Aptenodytes patagonicus 9234 King Penguin N Apteryx australis 8822 Brown Kiwi Y Apteryx haastii 8823 Great Spotted Kiwi N South Island Aptornis defossor 54366 N Adzebill Apus apus 8895 Common Swift N Apus pallidus 1160816 Pallid Swift N Aramides ypecaha 296133 Giant Wood-Rail N Aramus guarauna 54356 Limpkin Y Brown-Throated Aratinga pertinax 492513 N Parakeet Ruby-Troated Archilochus colubris 190676 N Hummingbird

157 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Ardea alba Egretta alba 110620 Great White Egret N Ardea cocoi 399589 Cocoi Heron N Ardea herodias 56072 Great Blue Heron Y Ardeotis kori Choriotis kori 89386 Kori Bustard Y Arenaria interpres 54971 Ruddy Turnstone Y Asio flammeus 56267 Short-eared Owl N Asio otus 111810 Long-eared Owl N Speotyto Athene cunicularia 194338 Burrowing Owl Y cunicularia Rufous-bellied Attagis gayi 227233 N Seedsnipe Aulacorhynchus prasinus 57399 Emerald Toucanet N Aythya americana 30385 Redhead Y Balaeniceps rex 33584 Shoebill Y Philippine Batrachostomus septimus 382305 Y Frogmouth Biziura lobata 45648 Musk Duck Y Bombycilla cedrorum 161648 Cedar Waxwing N Bohemian Bombycilla garrulus 125297 Y Waxwing Botaurus lentiginosus 110661 American Bittern N Brachypteracias Short-legged 135165 N leptosomus Ground-roller Brachypteracias Scaly Ground 188337 Y squamigera Roller Branta canadensis 8853 Canada Goose N Cobalt-winged Brotogeris cyanoptera 671079 N Parakeet Chestnut-capped Bucco macrodactylus 458186 Y Puffbird

158 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Abyssinian Bucorvus abyssinicus 153643 Y Ground-hornbill Southern Ground Bucorvus leadbeateri Bucorvus cafer 155086 N Hornbill Double-striped Burhinus bistriatus 240201 Y Thick-knee Eurasian Thick- Burhinus oedicnemus 85105 N knee Senegal Thick- Burhinus senegalensis 228417 N knee Buteo buteo 30397 Common Buzzard N Buteo jamaicensis 56263 Red-tailed Hawk Y Heterospizias Buteogallus meridionalis 223496 Savanna Hawk N meridionalis Silvery-cheeked Bycanistes brevis 175843 N Hornbill Sulfur-Crested Cacatua galerita 141274 N Cockatoo Yellow-crested Cacatua sulphurea 141271 Y Cockatoo Caiman crocodilus 8499 Spectacled Caiman N Lophortyx Callipepla gambelii 67773 Gambel's Quail N gambelii Calonectris diomedea 52122 Cory's Shearwater N Black-spotted Capito niger 91789 Y Barbet Caprimulgus europaeus 111811 Eurasian Nightjar N Band-winged Caprimulgus longirostris 48285 Y Nightjar Polyborus Caracara plancus 8951 Crested Caracara N plancus

159 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Red-legged Cariama cristata 54380 Y seriema Southern Casuarius casuarius 8787 Y Cassowary Cathartes aura 43455 Turkey Vulture Y Catharus guttatus 9185 Hermit Thrush N Centropus viridis 121388 Philippine Coucal Y Cepphus columba 28696 Pigeon Guillemot N Common Emerald Chalcophaps indica 187108 N Dove Chalcopsitta cardinalis 176062 Cardinal Lory Y Charadrius collaris 50395 Collared Plover N Semipalmated Charadrius semipalmatus 50400 N Plover Charadrius vociferus 50402 Killdeer Y Chauna torquata 30388 Southern Screamer Y Chionis alba 240203 Snowy Sheathbill N Chlidonias niger 279945 Black Tern N Caprimulgus Common Chordeiles minor 48398 N minor Nighthawk Ciconia boyciana 52775 Oriental Stork N Ciconia ciconia 8928 White Stork Y Cladorhynchus 425628 Banded Stilt N leucocephalus Red-browed Climacteris erythrops 254447 Y Treecreeper Cnemotriccus fuscatus 183540 Fuscous Flycatcher N Yellow-billed Coccyzus americanus 33603 Y Cuckoo Coccyzus Black-billed 33604 N erythropthalmus Cuckoo Cochlearius cochlearius 110676 Boat-billed Heron Y

160 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Colaptes auratus 51355 Northern Flicker N Sparkling Violet- Colibri coruscans 214663 Y ear Colinus cristatus 114915 Crested Bobwhite Y Colinus virginianus 9014 Northern Bobwhite N White-backed Colius colius 57410 Y Mousebird Speckled Colius striatus 57412 N Mousebird White-crowned Columba leucocephala 372311 N Pigeon Columba livia 8932 Rock Pigeon Y Common Ground- Columbina passerina 111974 Y dove Scardafella Columbina squammata 115699 Scaled Dove N squammata Lilac-breasted Coracias caudata 56292 Y Roller Coracias garrulus 188338 European Roller N Racket-tailed Coracias spatulata 81902 N Roller Coragyps atratus 33614 Black Vulture N Corvus corone 30422 Carrion Crow Y Corvus frugilegus 75140 Rook N Corythaeola cristata 103954 Great Blue Turaco Y Grey Go-away- Corythaixoides concolor 103956 N bird Corythaixoides Criniferoides White-bellied Go- 121526 N leucogaster leucogaster away bird Bare-faced Go- Corythaixoides personata 119405 N away bird Coturnix chinensis 46218 King Quail N Coturnix coturnix 9091 Common Quail Y Coturnix japonica 93934 Japanese Quail N

161 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Coua cristata 121393 Crested Coua Y Cranioleuca baroni 86269 Baron's Spinetail N Crax alector 84985 Black Curassow Y Crax rubra 84990 Great Curassow N Western Grey Crinifer piscator 119408 N Plantain-eater Crotophaga ani 103947 Smooth-billed Ani N Crotophaga sulcirostris 33598 Groove-billed Ani Y Crypturellus soui 458187 Little Tinamou Y Undulated Crypturellus undulatus 48396 N Tinamou Cuculus canorus 55661 Common Cuckoo Y Temminck’s Cursorius temminckii 227177 N Courser Cypseloides niger 46500 Black Swift N Daptrius ater 56348 Black Caracara Y Dendragapus obscurus 90755 Dusky Grouse N Amazonian Barred Dendrocolaptes certhia 190289 Y Woodcreeper Wandering Dendrocygna arcuata 8872 N Whistling-duck Tooth-billed Didunculus strigirostris 187111 N Pigeon Dinornis Dinornis giganteus 147464 Giant Moa N robustus Dromaius 8790 Emu Y novaehollandiae Dromas ardeola 458190 Crab-plover Y Pileated Dryocopus pileatus 51359 Y Woodpecker Green Imperial Ducula aenea 187114 N Pigeon

162 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Egretta eulophotes 458089 Chinese Egret N Ardea Egretta novaehollandiae novaehollandia 390973 White-faced Heron N e Emeus crassus 147466 Eastern Moa N American White Eudocimus albus 371913 Y Ibis Elegant Crested Eudromia elegans 8805 Y Tinamou Eudynamys taitensis 497866 Longtailed Cuckoo N Eudyptes Rockhopper Eudyptes chrysocome 79626 N crestatus Penguin Little Blue Eudyptula minor 37083 Y Penguin Lophotis Red-crested Eupodotis ruficrista 172689 Y ruficrista Korhaan White-bellied Eupodotis senegalensis 89385 N Bustard Great Eared- Eurostopodus macrotis 135169 Y nightjar Eurypyga helias 54383 Sunbittern Y Eurystomus orientalis 188341 Oriental Dollarbird N Canachites Falcipennis canadensis 109674 Spruce Grouse N canadensis Falco mexicanus 279964 Prairie Falcon Y Falco peregrinus 8954 Peregrine Falcon N Falco rufigularis 399590 Bat Falcon N Cerchneis Falco sparverius 56350 American Kestrel N sparveria Falco subbuteo 344233 Eurasian Hobby N Falco tinnunculus 100819 Common Kestrel N Dusky-billed Forpus modestus 870969 N Parrotlet

163 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Pternistis Francolinus capensis 57755 Cape Spurfowl N capensis Ascension Fregata aquila 244445 N frigatebird Magnificent Fregata magnificens 37042 Y Frigatebird Fregata minor 57241 Great Frigatebird N Fringilla montifringilla 36255 Brambling Y Fulica americana 81903 American Coot N Fulmarus glacialis 30455 Northern Fulmar N Yellow-billed Galbula albirostris 458192 Y Jacamar Bluish-fronted Galbula cyanescens 135170 N Jacamar Coppery-chested Galbula pastazae 118187 N Jacamar Rufous-tailed Galbula ruficauda 176937 N Jacamar Luzon Bleeding- Gallicolumba luzonica 187121 N heart Gallus gallus 9031 Chicken Y Gampsonyx swainsonii 56285 Pearl Kite Y Yellow-billed Gavia adamsii 372293 N Loon Black-throated Gavia arctica 57069 N Loon Great Northern Gavia immer 37039 Y (Common) Loon Gavia pacifica 85097 Pacific loon N Gavia stellata 37040 Red-throated Loon N Gavialis gangeticus 94835 Gharial Y Geococcyx californianus 8947 Roadrunner Y

164 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Geopelia striata 444140 Zebra Dove N Geotrygon montana 115651 Ruddy Quail-dove Y Violaceous Quail- Geotrygon violacea 869458 N dove Black-winged Glareola nordmanni 670345 N Pratincole Glareola pratincola 43316 Collared Pratincole N Glaucis aeneus 472812 Bronzy Hermit N Rufous-breasted Glaucis hirsuta 190457 N Hermit Western Crowned- Goura cristata 115654 N pigeon Grallaria squamigera 117164 Undulated Antpitta N Variegated Grallaria varia 117165 Y Antpitta Grus canadensis 40820 Sandhill Crane Y Grus grus 40816 Eurasian Crane N Guira guira 30392 Guira Cuckoo N White-backed Gyps africanus 43490 N Vulture Blackish Haematopus ater 185689 N Oystercatcher Eurasian Haematopus ostralegus 31908 Y Oystercatcher Harpactes ardens 59408 Philippine Trogon N Heliornis fulica 54369 Sungrebe Y Whiskered Hemiprocne comata 243314 N Treeswift Grey-rumped Hemiprocne longipennis 135173 N Treeswift Moustached Hemiprocne mystacea 46510 Y Treeswift

165 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Oleaginous Hemispingus frontalis 85537 N Hemispingus Herpetotheres cachinnans 56343 Laughing Falcon Y Himantopus mexicanus 227231 Black-necked Stilt N Himantornis haematopus 458194 Nkulengu Rail Y European Storm Hydrobates pelagicus 79651 N Petrel Gilded Hylocharis chrysura 135175 N Hummingbird Brown-eared Hypsipetes amaurotis Ixos amaurotis 36296 N Bulbul Ibidorhyncha struthersii 425643 Ibisbill N Spotted Indicator maculatus 545262 Y Honeyguide Indicator minor 219522 Lesser Honeyguide N Scaly-throated Indicator variegatus 189529 N Honeyguide African Pygmy Ispidina picta Ceyx pictus 570428 N Kingfisher Jabiru mycteria 33591 Jabiru Stork N Jacana jacana 54508 Wattled Jacana Y Jacana spinosa 118786 Northern Jacana N Jynx torquilla 189526 Eurasian Wryneck N Lanius collurio 56324 Red-backed Shrike N Leucophaeus Larus atricilla 126679 Laughing Gull N atricilla Larus dominicanus 37036 Kelp Gull N Great Black- Larus marinus 8912 Y backed Gull

166 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID White-throated Laterallus albigularis 189532 N Crake Rufous-sided Laterallus melanophaius 54363 N Crake Lepidocolaptes Wagler's Lepidocolaptes wagleri 75979 N squamatus Woodcreeper Blue-crowned Lepidothrix coronata Pipra coronata 321398 Y Manakin Leptodon Leptodon cayanensis 321080 Gray-headed Kite N cayanesis Leptosomus discolor 188344 Cuckoo Roller Y Leptotila verreauxi 135631 White-tipped Dove N Semicollared Malacoptila semicincta 488312 N Puffbird Malacorhynchus 45646 Pink-eared Duck Y membranaceus Red-backed Fairy Malurus melanocephalus 175006 Y Wren Megaceryle alcyon Ceryle alcyon 488309 Belted Kingfisher N Megalaima Coppersmith 240716 N haemacephala Barbet Megalaima virens 219518 Great Barbet Y Melanesian Megapodius eremita 81904 Y Megapode Common Megapodius freycinet 8979 N Megapode Meleagris gallopavo 9103 Common Turkey N Melopsittacus undulatus 13146 Budgerigar N Menura novaehollandiae 47692 Superb Lyrebird Y European Bee- Merops apiaster 265611 N eater Northern Carmine Merops nubicus 57421 Y Bee-eater

167 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Blue-throated Bee- Merops viridis 135176 N eater Brown roatelo Mesitornis unicolor 54374 Y (Mesite) Fiery-throated Metallura eupogon 66407 N Metaltail Lined Forest Micrastur gilvicollis 56335 N Falcon Collared Forest Micrastur semitorquatus 56334 Y Falcon Finsch's Pygmy Micropsitta finschii 57423 Y Parrot McConnell's Mionectes macconnelli 254557 Y flycatcher Blue-crowned Momotus momota 57426 Y Motmot Black-fronted Monasa nigrifrons 882749 N Nunbird Monias benschi 399593 Subdesert Mesite Y Morus bassanus Sula bassana 37578 Northern Gannet Y Motacilla alba 45807 White Wagtail N Musophaga violacea 103959 Violet Turaco N Mycteria americana 33587 Wood Stork N Nandayus nenday 51908 Nanday Parakeet N Rufous-vented Neomorphus geoffroyi 78208 N Ground Cuckoo Neophema elegans 85098 Elegant Parrot N Ninox novaeseelandiae 57242 Southern Boobook N Nipponia nippon 128390 Crested Ibis N Spizaetus Blyth's Hawk- Nisaetus alboniger 307441 N alboniger eagle Mountain Hawk- Nisaetus nipalensis 214438 N eagle

168 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Nothoprocta perdicaria 30464 Chilean Tinamou Y Helmeted Numida meleagris 8996 Y Guineafowl Yellow-crowned Nyctanassa violacea 56294 N Night Heron Nyctibius aethereus 48425 Long-tailed Potoo N Nyctibius bracteatus 48426 Rufous Potoo Y Nyctibius grandis 48427 Great Potoo Y Nyctibius griseus 48428 Common Potoo N Nyctibius maculosus 48430 Andean Potoo N Black-crowned Nycticorax nycticorax 8901 N Night Heron Nymphicus hollandicus 13180 Cockatiel N Spot-backed Nystalus maculatus 135178 N Puffbird Procellaria Wilson's Storm- Oceanites oceanicus 79653 Y oceanica petrel Wedge-rumped Oceanodroma tethys 79633 Y Storm Petrel Opisthocomus hoazin 30419 Hoatzin Y Ortalis canicollis 125068 Chaco Chachalaca N Otidiphaps nobilis 187130 Pheasant Pigeon Y Eurasian Scops- Otus scops 126827 N owl Oxyura jamaicensis 8884 Ruddy Duck Y Pachyptila desolata 79635 Antarctic Prion N Pandion haliaetus 56262 Osprey Y Black-capped Parus atricapillus 48891 N Chickadee Parus major 9157 Great Tit N Passer domesticus 48849 House Sparrow N Eurasian tree Passer montanus 9160 Y sparrow

169 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Pedionomus torquatus 227192 Plains-wanderer Y Common Diving Pelecanoides urinatrix 37079 Y Petrel Pelecanus conspicillatus 317791 Australian Pelican N Pelecanus American White 33618 N erythrorhynchos Pelican Pelecanus occidentalis 37043 Brown Pelican Y Great White Pelecanus onocrotalus 36301 N Pelican Pelecanus Pelecanus thagus occidentalis 1108811 Peruvian Pelican N thagus Pezophaps solitaria 187133 Rodrigues Solitaire N Phaenicophaeus Chestnut-breasted 33595 Y curvirostris Malkoha White-tailed Phaethon lepturus 97097 Y Tropicbird Red-tailed Phaethon rubricauda 57073 Y Tropicbird Grey-chinned Phaethornis griseogularis 382311 Y Hermit Phaethornis Phaethornis Tawny-bellied syrmathophoru 472907 N syrmatophorus Hermit s Double-crested Phalacrocorax auritus 56069 N Cormorant Phalacrocorax carbo 9209 Great Cormorant Y Phalacrocorax pelagicus 56070 Pelagic Cormorant N Red-faced Phalacrocorax urile 56071 N Cormorant Steganopus Phalaropus tricolor 227175 Wilson's Phalarope N tricolor

170 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Golden-headed Pharomachrus auriceps 59410 Y Quetzal Resplendent Pharomachrus mocinno 289186 N Quetzal Diademed Phegornis mitchelli 227179 Y Sandpiper-plover Phodilus badius 111818 Oriental Bay Owl Y Diomedea Hawaiian Black- Phoebastria nigripes 54021 Y nigripes footed Albatross Light-mantled Phoebetria palpebrata 46545 N Albatross Phoenicopterus Phoenicopterus chilensis 117000 Chilean Flamingo Y ruber chilensis Phoenicopterus ruber 9217 American flamingo N Green Phoeniculus purpureus 113189 Y Woodhoopoe Western Crowned- Phylloscopus occipitalis 47980 N warbler Picathartes White-necked 175131 Y gymnocephalus Rockfowl Picoides villosus 51356 Hairy Woodpecker N White-barred Picumnus cirratus 137535 N Piculet Pitangus Pitangus sulphuratus 371930 Great Kiskadee N sulfuratus Pitta erythrogaster 363773 Red-bellied Pitta N Pitta guajana Pitta guajara 265622 Banded Pitta Y Platalea ajaja Ajaia ajaja 371920 Roseate Spoonbill N Black-faced Platalea minor 259913 N Spoonbill

171 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Platycercus elegans 269190 Crimson Rosella Y Plegadis falcinellus 52788 Glossy Ibis N Oriolus Ploceus cucullatus 135448 Village Weaver Y cucullatus American Golden- Pluvialis dominica 240205 N plover Nacunda Podager nacunda 135181 N Nighthawk Podargus strigoides 8905 Tawny Frogmouth Y Podica senegalensis 54371 African Finfoot N Podiceps auritus 37050 Horned Grebe Y Great Crested Podiceps cristatus 345573 N Grebe Podiceps Black-necked Podiceps nigricollis 85099 N caspicus (Eared) Grebe Podilymbus podiceps 9252 Pied-billed Grebe N Pogoniulus Red-fronted Pogoniulus pusillus 488313 N pusilus Tinkerbird Gallinula martinica; Porphyrio martinica 54567 Purple Gallinule N Porphyrula martinica Red-breasted Psittacula alexandri 232635 Y Parakeet Psittacus erithacus 57247 Grey Parrot Y Common Psophia crepitans 54359 Y Trumpeter Pale-winged Psophia leucoptera 399595 N Trumpeter Dark-winged Psophia viridis 302536 N Trumpeter Double-banded Pterocles bicinctus 302525 N Sandgrouse Crowned Pterocles coronatus 56297 N Sandgrouse

172 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Yellow-throated Pterocles gutturalis 240206 N Sandgrouse Namaqua Pterocles namaqua 216825 Y Sandgrouse Darwin's (Lesser) Pterocnemia pennata Rhea pennata 8795 N Rhea Pterodroma brevirostris 37064 Kerguelen Petrel N Black-capped Pterodroma hasitata 53671 N Petrel 390974;3 Ivory-billed Pteroglossus azara N 90975 Aracari Chestnut-eared Pteroglossus castanotis 91779 N Aracari Ptilonorhynchus 28724 Satin Bowerbird N violaceus Puffinus griseus 37052 Sooty Shearwater Y Audubon's Puffinus lherminieri 47976 N Shearwater Wedge-tailed Puffinus pacificus 48683 N Shearwater Short-tailed Puffinus tenuirostris 48684 N Shearwater Maroon-bellied Pyrrhura frontalis 211511 N Parakeet Rallus limicola 156759 Virginia Rail Y 91793; Channel-billed Ramphastos vitellinus N 322700 Toucan Raphus cucullatus 187135 Dodo N Ruby-crowned Regulus calendula 73321 Y Kinglet Rhea americana 8797 Greater Rhea Y Rhinopomastus Common Scimitar- 113115 N cyanomelas bill Rhynochetus Rhynochetos jubatus 54386 Kagu Y jubatus

173 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Black-legged Rissa tridactyla 75485 N Kittiwake Rollulus rouloul 30405 Crested Partridge Y Greater Painted- Rostratula benghalensis 118793 Y snipe Rynchops niger 227184 Black Skimmer N Sagittarius serpentarius 56258 Secretary Bird Y Broad-billed Sapayoa aenigma 239371 Y Sapayoa Sarcoramphus papa 43583 King Vulture Y Buff-spotted Sarothrura elegans 156766 Y Flufftail Sayornis phoebe 56315 Eastern Phoebe N American Scolopax minor 56299 N Woodcock Scopus umbretta 33581 Hamerkop Y Scytalopus magellanicus 9169 Andean tapaculo Y Golden-collared Selenidera reinwardtii 240743 N Toucanet Semnornis Semnornis ramphastinus 91785 Toucan Barbet N rhamphastinus Rufous-sided Smithornis rufolateralis 137762 Y Broadbill Grey-headed Smithornis sharpei 81930 N Broadbill Humboldt's Spheniscus humboldti 9240 N Penguin Magellanic Spheniscus magellanicus 37081 N Penguin Steatornis caripensis 48435 Oilbird Y Catharacta Stercorarius maccormicki 395889 South Polar Skua N maccormicki Catharacta Stercorarius skua 54056 Great Skua N skua

174 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID White-collared Streptoprocne zonaris 46505 Y Swift Strigops habroptilus 57251 Kakapo N Syrnium occidentale; Strix occidentalis 201991 Spotted Owl Y Ciccaba occidentalis Strix varia 57075 Barred Owl N Struthio camelus 8801 Ostrich Y Sula dactylatra 56068 Masked Booby N Sylvia atricapilla 48155 Blackcap N Sylvia Eastern Orphean Sylvia crassirostris 216226 N hortensis Warbler Sylvia nana 175014 Desert Warbler Y Synthliboramphus 28708 Ancient Murrelet N antiquus Pallas's Syrrhaptes paradoxus 302527 Y Sandgrouse Tachybaptus 342690 Australasian Grebe N novaehollandiae Taeniopygia guttata 59729 Zebra Finch N Red-crested Tauraco erythrolophus 121530 Y Turaco Thalassarche Diomedea Black-browed 54026 N melanophrys melanophris Albatross Thamnophilus Blackish-grey 175015 Y nigrocinereus Antshrike Harpiprion 161738; Theristicus caerulescens Plumbeous Ibis N caerulescens 1118847 Theristicus caudatus 399597 Buff-necked Ibis N Grey-breasted Thinocorus orbignyianus 161742 Y Seedsnipe Thinocorus rumicivorus 391620 Least Seedsnipe N Fasciated Tiger Tigrisoma fasciatum 488314 N Heron

175 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Rufescent Tiger Tigrisoma lineatum 110694 N Heron White-throated Tinamus guttatus 94827 Y Tinamou Tinamus major 30468 Great Tinamou N Red-billed Dwarf Tockus camurus 458195 Y Hornbill Red-billed Tockus erythrorhynchus 81911 N Hornbill Eastern Yellow- Tockus flavirostris 302528 N billed Hornbill Todiramphus sanctus Halcyon sancta 342380 Sacred Kingfisher N Common Tody Todirostrum cinereum 196049 N Flycatcher Narrow-billed Todus angustirostris 176936 Y Tody Todus mexicanus 135184 Puerto Rican Tody N Todus subulatus 176934 Broad-billed Tody N Trachyphonus Trachyphonus darnaudii 135185 Usambiro Barbet N usambiro White-bellied Treron sieboldii 228418 N Green Pigeon Pink-necked Treron vernans 115702 Y Green-Pigeon Trichoglossus 176049 Rainbow Lorikeet N haematodus Heteroscelus Tringa incanus 507599 Wandering Tattler N incanus Blue-crowned Trogon curucui 59414 N Trogon Black-tailed Trogon melanurus 56311 N Trogon Trogon personatus 57437 Masked Trogon Y

176 Appendix 3.1. Continued. NCBI Species Synonym Common Name H2008 TaxID Green-backed Trogon viridis 59419 N Trogon Turdus Turdus falcklandii 411538 Austral Thrush Y falklandii Red-chested Turnix pyrrhothorax 292130 N Buttonquail Turnix suscitator 292133 Barred Buttonquail N Turnix Common Turnix sylvatica 9248 Y sylvaticus buttonquail Painted Turnix varia Turnix varius 56307 N Buttonquail Turtur tympanistria 870970 Tambourine Dove N Tyrannus tyrannus 43165 Eastern Kingbird Y Tyto alba Strix alba 56313 Barn Owl Y Upupa epops 57439 Hoopoe Y Uria lomvia 28711 Thick-billed Murre N Red-faced Urocolius indicus Colius indicus 458196 Y Mousebird Blue-naped Urocolius macrourus 85101 N Mousebird Vidua chalybeata 81927 Village Indigobird Y Zenaida auriculata 115703 Eared Dove N Zenaida macroura 47245 Mourning Dove N Japanese White- Zosterops japonica 36299 N eye

177 Appendix 3.2. Genbank/Ensembl accession numbers Species 12S CYTB COX1 ND2 Acanthisitta chloris AY325307 AY325307 AY325307 AY325307 Accipiter gentilis NC011818 NC011818 NC011818 NC011818 Accipiter striatus - U83305 HM033202 - Aceros corrugatus HM755883 HM755883 HM755883 HM755883 Acrocephalus scirpaceus NC010227 NC010227 NC010227 NC010227 Acryllium vulturinum NC014180 NC014180 NC014180 NC014180 Aegotheles albertisi - AY090666 - - Aegotheles cristatus NC011718 NC011718 NC011718 NC011718 Aegotheles insignis - FJ588456 - EU042514

Aerodramus vanikorensis - FJ588453 JQ173912 AY294502 Aeronautes saxatalis EU167032 EU166978 DQ433288 EU166921 Afrotis afra DQ674591 - - - Agapornis roseicollis NC011708 NC011708 NC011708 NC011708 Alcedo atthis - D38329 GU571232 EF585373 Alcedo leucogaster - - JQ173937 DQ111833 Alectoris graeca - Z48772 - - Alectura lathami NC007227 NC007227 NC007227 NC007227 Alisterus scapularis EU197063 FJ499014 JN801395 - Alligator mississipiensis NC001922 NC001922 NC001922 NC001922 Amazona autumnalis AY301338 AY283455 AY194379 AY194446 Anas crecca FN675616 AF059064 GU571238 EU585670 Anas platyrhynchos NC009684 NC009684 NC009684 NC009684 Andigena cucullata EU167076 EU167021 AY959801 EU166970 Anhinga anhinga AY941811 U83155 FJ027122 - Anomalopteryx didiformis NC002779 NC002779 NC002779 NC002779 Anser albifrons NC004539 NC004539 NC004539 NC004539 Anser anser NC011196 NC011196 NC011196 NC011196 Anser erythropus - EU585617 GU571729 EU585680 Anseranas semipalmata NC005933 NC005933 NC005933 NC005933 Aphelocoma ultramarina - HQ123827 DQ432740 HQ123747

Aptenodytes patagonicus AY139621 AY139623 EU525303 AY139622 Apteryx australis GU071057 U76050 AY016010 -

178 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Apteryx haastii NC002782 NC002782 NC002782 NC002782 Aptornis defossor U76019 - - - Apus apus NC008540 NC008540 NC008540 NC008540 Apus pallidus JQ520042 JQ353946 - - Aramides ypecaha - - FJ027147 - Aramus guarauna DQ485816 DQ485899 FJ027151 - Aratinga pertinax NC015197 NC015197 NC015197 NC015197 Archilochus colubris NC010094 NC010094 NC010094 NC010094 Ardea alba - AF193822 FJ027162 GU346989 Ardea cocoi - - FJ027163 - Ardea herodias - AF193821 DQ433329 - Ardeotis kori DQ674590 AJ511472 - - Arenaria interpres NC003712 NC003712 NC003712 NC003712 Asio flammeus U83758 U89171 U83781 - Asio otus AY274022 AF082067 GU571271 AY274069 Athene cunicularia AF231330 EU348965 DQ433340 - Attagis gayi EU167081 EU167025 DQ385173 EU166934 Aulacorhynchus prasinus U89207 U89191 JQ174123 JF424373 Aythya americana NC000877 NC000877 NC000877 NC000877 Balaeniceps rex GU071053 U08937 GU071053 GU071053 Batrachostomus septimus - EF100673 - - Biziura lobata - EU585627 - EU585690 Bombycilla cedrorum - AF285786 EF484234 FJ177330 Bombycilla garrulus AB042364 AY329449 GU571276 DQ466855 Botaurus lentiginosus DQ485799 DQ485897 DQ433353 - Brachypteracias leptosomus AF407429 AF407445 AF407481 AF407463 Brachypteracias squamigera AF407430 AF407447 AF407482 AF407464 Branta canadensis NC007011 NC007011 NC007011 NC007011 Brotogeris cyanoptera NC015530 NC015530 NC015530 NC015530 Bucco macrodactylus ---- Bucorvus abyssinicus - AF346908 - - Bucorvus leadbeateri NC015199 NC015199 NC015199 NC015199 Burhinus bistriatus DQ674587 - JQ174203 - Burhinus oedicnemus - - GQ481405 - Burhinus senegalensis AY274007 AY274043 - AY274073

179 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Buteo buteo NC003128 NC003128 NC003128 NC003128 Buteo jamaicensis U83720 AY274044 DQ434504 AY274074

Buteogallus meridionalis GQ264628 GQ264793 FJ027237 GQ264879 Bycanistes brevis NC015201 NC015201 NC015201 NC015201 Cacatua galerita - AB177977 JF414289 JF414343 Cacatua sulphurea - AB177976 JF414291 EU327605 Caiman crocodilus NC002744 NC002744 NC002744 NC002744 Callipepla gambelii DQ485791 DQ485889 DQ433416 AF028764 Calonectris diomedea AY139624 AY139626 AY567883 AY139625 Capito niger AY940747 AY897060 AY940778 JQ445228 Caprimulgus europaeus - AJ004074 GQ481445 -

Caprimulgus longirostris - GU586641 JN801536 - Caracara plancus - U83313 FJ027297 JN614684 Cariama cristata U76024 - JQ627334 JF916443 Casuarius casuarius NC002778 NC002778 NC002778 NC002778 Cathartes aura NC007628 NC007628 NC007628 NC007628 Catharus guttatus EU372672 X74261 DQ433455 EU372684 Centropus viridis - AF204992 - AF205021 Cepphus columba EU372666 U37293 EF380325 EU372680 Chalcophaps indica HM746789 HM746789 HM746789 HM746789 Chalcopsitta cardinalis - AF346321 HQ629760 - Charadrius collaris - AH005179 JN801556 - Charadrius semipalmatus EU167040 EU166986 DQ433493 EU166929 Charadrius vociferus DQ485792 DQ485890 DQ385167 DQ385082 Chauna torquata AY140700 AY274030 AY140730 AY274053 Chionis alba ---- Chlidonias niger AY631325 AY631289 EU525341 AY631361 Chordeiles minor EU167037 EU166983 AF168060 - Ciconia boyciana NC002196 NC002196 NC002196 NC002196 Ciconia ciconia NC002197 NC002197 NC002197 NC002197 Cladorhynchus leucocephalus EF373074 EF373125 - EF373232 Climacteris erythrops ---- Cnemotriccus fuscatus NC007975 NC007975 NC007975 NC007975

180 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Coccyzus americanus AF173582 U09265 JN801580 EU327609 Coccyzus erythropthalmus AY274015 AF082048 AY666445 AY274062 Cochlearius cochlearius - AF193830 JQ174484 - Colaptes auratus EU372668 EU372676 AF272582 EU372687 Colibri coruscans - AY150651 FJ027415 AY830476 Colinus cristatus AF222575 - JQ174493 AF222544 Colinus virginianus EU167061 EU372675 DQ433524 EU166949 Colius colius U89216 U89174 - EU327610 Colius striatus AY274011 AY274032 AF168061 AY274058 Columba leucocephala AY274023 AY274041 JQ175690 AY274070 Columba livia NC013978 NC013978 NC013978 NC013978 Columbina passerina - AF182686 JN801583 EU327611 Columbina squammata EF373296 AF483347 EF373368 EF373330 Coracias caudata U89225 U89184 JQ174537 - Coracias garrulus AF407431 AF407448 GQ481619 AF407465 Coracias spatulata AY274010 AF082060 - AY274057 Coragyps atratus AY426746 U08946 JN801602 - Corvus corone AF386463 HE805743 GQ481632 JQ023936 Corvus frugilegus NC002069 NC002069 - NC002069 Corythaeola cristata - AF168117 AF168069 - Corythaixoides concolor - AF168118 AF168070 - Corythaixoides leucogaster - AF102099 - - Corythaixoides personata - AF102089 - - Coturnix chinensis NC004575 NC004575 NC004575 NC004575 Coturnix coturnix AM902516 L08377 GQ481649 X57246 Coturnix japonica NC003408 NC003408 NC003408 NC003408 Coua cristata - AF204986 - AF205015 Cranioleuca baroni EU167045 EU166991 - EU166958 Crax alector - AY141921 AY141911 AY141931 Crax rubra AY274003 AY274029 AY141915 AY274050 Crinifer piscator AY274021 AY274040 - AY274068 Crotophaga ani AY274016 AY274035 AF168065 AY274063 Crotophaga sulcirostris - U09260 JN801304 -

181 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Crypturellus soui - FJ899151 JN801607 - Crypturellus undulatus AY139627 AY139629 - AY139628 Cuculus canorus AY274013 AY274034 GU571356 AY274060 Cursorius temminckii DQ385277 DQ385226 DQ385175 DQ385090 Cypseloides niger - AY294480 - AY294542 Daptrius ater - U83309 JQ174641 - Dendragapus obscurus AF222580 AF230178 DQ433564 AF222549 Dendrocolaptes certhia - AY089817 JN801633 AY089856 Dendrocygna arcuata AF536740 AF082061 - U97735 Didunculus strigirostris AF483306 AF483343 - - Dinornis giganteus NC002672 NC002672 NC002672 NC002672 Dromaius novaehollandiae NC002784 NC002784 NC002784 NC002784 Dromas ardeola HM369462 HM369461 - HM369460 Dryocopus pileatus NC008546 NC008546 NC008546 DQ479187 Ducula aenea AF483294 AF483331 JQ174727 - Egretta eulophotes NC009736 NC009736 NC009736 NC009736 Egretta novaehollandiae NC008551 NC008551 NC008551 NC008551 Emeus crassus NC002673 NC002673 NC002673 NC002673 Eudocimus albus EU167082 EU167026 DQ433629 EU166938 Eudromia elegans NC002772 NC002772 NC002772 NC002772 Eudynamys taitensis NC011709 NC011709 NC011709 NC011709 Eudyptes chrysocome NC008138 NC008138 NC008138 NC008138 Eudyptula minor NC004538 NC004538 NC004538 NC004538 Eupodotis ruficrista - AJ511465 - - Eupodotis senegalensis EU167062 EU167007 - EU166955 Eurostopodus macrotis EU167043 FJ588447 JQ174827 EU166925 Eurypyga helias U76025 - JQ174830 - Eurystomus orientalis NC011716 NC011716 - NC011716 Falcipennis canadensis AF222577 AF170992 DQ432923 AF222546 Falco mexicanus - EU233076 AY666553 - Falco peregrinus AF090338 U83307 AF090338 JF909877 Falco rufigularis - - JQ174838 - Falco sparverius NC008547 NC008547 NC008547 NC008547 Falco subbuteo GU390892 - GU571392 GU816825 Falco tinnunculus NC011307 NC011307 NC011307 NC011307

182 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Forpus modestus HM755882 HM755882 HM755882 HM755882 Francolinus capensis DQ832112 AM236909 - DQ768282 Fregata aquila EU167044 EU166990 AY369050 EU166963 Fregata magnificens AF173576 FR691093 AY369052 FR691093 Fregata minor AY369043 FR691320 JF498856 FR691320 Fringilla montifringilla EF027309 DQ192024 EU847708 GU816851 Fulica americana EU167074 EU167019 DQ434598 EU166956 Fulmarus glacialis - AJ004178 GU571407 - Galbula albirostris - - JX487425 JQ445396 Galbula cyanescens ---- Galbula pastazae EU167046 EU166992 - EU166968 Galbula ruficauda - AF441634 JN801678 - Gallicolumba luzonica HM746790 HM746790 HM746790 HM746790 Gallus gallus NC001323 NC001323 NC001323 NC001323 Gampsonyx swainsonii U83723 - JN801680 - Gavia adamsii EU167048 EU166994 GU571416 EU166951 Gavia arctica AY139633 AY139635 AY567889 AY139634 Gavia immer EU167047 EU166993 AY567890 EU166950 Gavia pacifica NC008139 NC008139 NC008139 NC008139 Gavia stellata NC007007 NC007007 NC007007 NC007007 Gavialis gangeticus NC008241 NC008241 NC008241 NC008241 Geococcyx californianus NC011711 NC011711 NC011711 NC011711 Geopelia striata HM746791 HM746791 HM746791 HM746791 Geotrygon montana EF373301 AF182696 JX487436 FJ175719 Geotrygon violacea NC015207 NC015207 NC015207 NC015207 Glareola nordmanni - - GQ481967 - Glareola pratincola EU372667 EU372674 GQ481970 EU372681 Glaucis aeneus - FJ588451 JQ174947 EU042554 Glaucis hirsuta - - JQ174948 AY830486 Goura cristata EF373302 AF182709 EF373374 EF373336 Grallaria squamigera AY139636 AY139638 - AY139637 Grallaria varia - AY370541 JQ174959 AF127204 Grus canadensis EU167051 EU166997 FJ769855 EU166954 Grus grus DQ485809 U27546 FJ769849 FJ769849 Guira guira - AF168113 AF168066 AF205027 Gyps africanus - X86748 JQ174983 AY987082

183 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Haematopus ater NC003713 NC003713 NC003713 NC003713 Haematopus ostralegus EU167052 EU166998 GU571427 EU166930 Harpactes ardens U94810 U94796 JQ174999 EU372688 Heliornis fulica DQ485819 DQ485902 JQ175018 - Hemiprocne comata - FJ588455 JQ175023 AY294545 Hemiprocne longipennis ---- Hemiprocne mystacea - AY150655 - EU042517 Hemispingus frontalis AY139639 AF383020 AF383098 AY139640 Herpetotheres cachinnans - U83319 JQ175038 JF909875 Himantopus mexicanus EU167077 EU167022 DQ385166 EU166932

Himantornis haematopus ---- Hydrobates pelagicus - AF076059 AY567885 - Hylocharis chrysura - - FJ027661 GU167236 Hypsipetes amaurotis EU167068 EU167013 EF515799 EU166962 Ibidorhyncha struthersii EF373086 EF373136 - EF373244 Indicator maculatus - - GU566418 GU566533 Indicator minor - AY279266 HQ997965 DQ188158 Indicator variegatus AY940753 AY940802 AY940781 - Ispidina picta EU167031 EU166977 JQ175171 EU166940 Jabiru mycteria - U72770 JQ175174 - Jacana jacana EU167053 EU166999 DQ385171 EU166935 Jacana spinosa DQ485796 DQ485894 DQ433701 - Jynx torquilla AY940754 AY940803 AY940782 DQ479151 Lanius collurio EU167054 EU167000 EF621613 EU166959 Larus atricilla EU167055 EU167001 DQ432969 EU166931 Larus dominicanus NC007006 NC007006 NC007006 NC007006 Larus marinus EF373088 AJ508140 GU571453 EF373246 Laterallus albigularis - - JQ175221 -

Laterallus melanophaius U76018 DQ485906 - - Lepidocolaptes wagleri EU167041 EU166987 - EU166957 Lepidothrix coronata - FJ899382 EF111043 GU985508 Leptodon cayanensis - AY987240 - AY987059 Leptosomus discolor AF407432 AF407449 AF407484 AF407466 Leptotila verreauxi NC015190 NC015190 NC015190 NC015190

184 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Malacoptila semicincta EU167035 EU166981 - EU166966 Malacorhynchus membranaceus - EU585651 - EU585714

Malurus melanocephalus - - GU825758 GU825874 Megaceryle alcyon EU167039 EU166985 DQ433478 EU166945 Megalaima haemacephala AY940758 AY279274 AY940785 HQ874471 Megalaima virens - AY279271 - HQ874498 Megapodius eremita AY274005 AF082065 - AY274052 Megapodius freycinet - AM236880 - AF394631 Meleagris gallopavo NC010195 NC010195 NC010195 NC010195 Melopsittacus undulatus NC009134 NC009134 NC009134 NC009134

Menura novaehollandiae NC007883 NC007883 NC007883 NC007883 Merops apiaster - - GQ482160 EU021520 Merops nubicus U89224 U89185 JQ175353 EU021529 Merops viridis EU167057 EU167003 - EU166941 Mesitornis unicolor U76022 - - - Metallura eupogon EU167083 EU167027 - EU166922 Micrastur gilvicollis NC008548 NC008548 NC008548 NC008548 Micrastur semitorquatus - U83314 JN801804 JF909871 Micropsitta finschii U89232 U89176 JQ175373 EU327634 Mionectes macconnelli - EF110846 JX487614 EF110710 Momotus momota EU167058 EU167004 JX487648 EU166942 Monasa nigrifrons - - JN801825 - Monias benschi ---- Morus bassanus EF101669 AY567921 AY567893 - Motacilla alba EU167059 EU167005 GU571487 EU166960 Musophaga violacea AY274020 AY274039 AF168068 AY274067 Mycteria americana U83712 AF082066 FJ027865 AY274076 Nandayus nenday U70746 AY274038 EU621632 AY274066 Neomorphus geoffroyi AY274017 AY274036 AF168067 AY274064 Neophema elegans AY274018 AY274037 JQ175544 AY274065 Ninox novaeseelandiae NC005932 NC005932 NC005932 NC005932 Nipponia nippon NC008132 NC008132 NC008132 NC008132

185 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Nisaetus alboniger NC007599 NC007599 NC007599 NC007599 Nisaetus nipalensis NC007598 NC007598 NC007598 NC007598 Nothoprocta perdicaria U76046 U76053 U76060 - Numida meleagris NC006382 NC006382 NC006382 NC006382 Nyctanassa violacea EU167033 EU166979 JQ175585 EU166936 Nyctibius aethereus ---- Nyctibius bracteatus - GU586675 - EU042512 Nyctibius grandis EU344977 EU344977 EU344977 EU344977 Nyctibius griseus HM746792 HM746792 HM746792 HM746792 Nyctibius maculosus EU167060 EU167006 - EU166926 Nycticorax nycticorax NC015807 NC015807 - NC015807 Nymphicus hollandicus NC015192 NC015192 NC015192 NC015192 Nystalus maculatus - - JN801874 - Oceanites oceanicus - AF076062 DQ433048 - Oceanodroma tethys - AF076066 JQ175598 - Opisthocomus hoazin U83748 AY274048 AF168071 AF076363 Ortalis canicollis AF165448 AF165472 AF165496 AY140746 Otidiphaps nobilis EF373312 AF483352 EF373384 EF373346 Otus scops - AJ004037 EF515810 EU601039 Oxyura jamaicensis AY747700 EU585658 AY666448 EU585721 Pachyptila desolata - AF076068 - - Pandion haliaetus NC008550 NC008550 NC008550 NC008550 Parus atricapillus - AF347937 HM033704 - Parus major EU167064 EU167009 HM185332 EU166961 Passer domesticus AF447245 AY495393 FJ027964 AY030143 Passer montanus AB042360 AY030118 GQ482320 GU816845 Pedionomus torquatus DQ674586 DQ385225 DQ385174 DQ385089 Pelecanoides urinatrix X82518 AF076076 - - Pelecanus conspicillatus DQ780883 DQ780883 DQ780883 DQ780883 Pelecanus erythrorhynchos EF101668 U08938 EF101673 JX683963 Pelecanus occidentalis L33382 AY567920 AY567892 JX683965 Pelecanus onocrotalus JX683981 U83158 JX683930 JX683967 Pelecanus thagus JX683987 - JX683936 JX683973 Pezophaps solitaria AF483300 AF483337 - - Phaenicophaeus curvirostris - U09264 - -

186 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Phaethon lepturus AY369045 AF158252 JN801349 - Phaethon rubricauda EU167065 EU167010 NC007979 EU166964 Phaethornis griseogularis - - - EU042578 Phaethornis syrmatophorus EU167084 EU167028 - EU166923 Phalacrocorax auritus AF373588 AF373590 DQ433896 - Phalacrocorax carbo AY009323 U83164 GU571539 -

Phalacrocorax pelagicus EU167066 EU167011 JN801351 EU166965 Phalacrocorax urile AY009340 HQ379740 JN801355 - Phalaropus tricolor DQ674581 AY894240 AY666195 AY894189 Pharomachrus auriceps U94813 U94799 - EU603916 Pharomachrus mocinno EF622093 EF622094 - EU603918 Phegornis mitchelli EF373099 EF373149 - EF373257 Phodilus badius - AJ004042 - - Phoebastria nigripes EU167042 EU166988 DQ433935 EU166974 Phoebetria palpebrata - U48943 - - Phoenicopterus chilensis - - FJ028026 - Phoenicopterus ruber U83714 AY274045 AY567894 AY274075 Phoeniculus purpureus EU167067 EU167012 AF221663 EU166943 Phylloscopus occipitalis EU372671 EU372678 - EU372683 Picathartes gymnocephalus - JN614899 - GU816831 Picoides villosus AY940768 AY942890 AY942877 DQ361290 Picumnus cirratus AY940763 DQ479243 AY940788 DQ479153 Pitangus sulphuratus - - JN801937 - Pitta erythrogaster - JN614860 EU541462 JN614687 Pitta guajana ---- Platalea ajaja DQ674560 GU346974 FJ028111 GU346988 Platalea minor NC010962 NC010962 NC010962 NC010962 Platycercus elegans - DQ467900 JQ175888 DQ105446 Plegadis falcinellus U88020 U72786 GU572048 - Ploceus cucullatus - AF290141 HQ998035 AF290104 Pluvialis dominica DQ674562 - DQ433962 -

187 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Podager nacunda - GU586685 JQ175916 - Podargus strigoides EU167069 EU167014 JQ175917 EU166927 Podica senegalensis DQ485821 DQ485904 HQ998023 - Podiceps auritus L33388 - GU571583 DQ157675 Podiceps cristatus NC008140 NC008140 NC008140 NC008140 Podiceps nigricollis EU167071 EU167016 GQ482520 EU166973 Podilymbus podiceps EU167070 EU167015 DQ433969 EU166972 Pogoniulus pusillus EU167056 EU167002 - EU166971 Porphyrio martinica U77160 U77166 JQ175969 - Psittacula alexandri - AB177958 - - Psittacus erithacus U88022 AY082076 EU621657 GU816826 Psophia crepitans DQ485817 DQ485900 JQ176018 - Psophia leucoptera - - JN801956 - Psophia viridis DQ485818 DQ485901 - - Pterocles bicinctus DQ674558 - - - Pterocles coronatus EU167073 EU167018 - EU166939 Pterocles gutturalis ---- Pterocles namaqua DQ385267 DQ385216 DQ385165 DQ385080 Pterocnemia pennata NC002783 NC002783 NC002783 NC002783 Pterodroma brevirostris NC007174 NC007174 NC007174 NC007174 Pterodroma hasitata EU167072 EU167017 DQ434001 EU166975 Pteroglossus azara NC008549 NC008549 NC008549 NC008549 Pteroglossus castanotis - AF123520 FJ028177 - Ptilonorhynchus violaceus GU393986 X74256 AF197833 JN614690 Puffinus griseus X82533 U74353 JN801370 - Puffinus lherminieri - AF076085 DQ434015 - Puffinus pacificus U88025 AF076088 DQ434019 - Puffinus tenuirostris - U74352 DQ434025 - Pyrrhura frontalis - AY751643 - JN614685 Rallus limicola JQ360508 JQ348020 GU097263 - Ramphastos vitellinus GQ412318 AY959835 JN801966 AY959863 Raphus cucullatus AF483301 AF483338 - - Regulus calendula - AY894885 HM033742 AY329435 Rhea americana NC000846 NC000846 NC000846 NC000846 Rhinopomastus cyanomelas - - AF221666 - Rhynochetos jubatus NC010091 NC010091 NC010091 NC010091

188 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Rissa tridactyla DQ385280 DQ385229 DQ385178 AY590389 Rollulus rouloul - EF571185 - - Rostratula benghalensis DQ674583 FJ205692 EF632098 EF373265 Rynchops niger DQ674567 DQ385230 DQ385179 DQ385094 Sagittarius serpentarius EU167078 AJ604483 U83776 EU166948 Sapayoa aenigma ---- Sarcoramphus papa - X86760 JN801979 - Sarothrura elegans ---- Sayornis phoebe U83765 AF536744 DQ434060 AF536747 Scolopax minor U83744 AF082068 DQ434062 AY274072 Scopus umbretta AF339360 U08936 - EU372682 Scytalopus magellanicus - X60945 FJ028252 EU479679 Selenidera reinwardtii EU167075 EU167020 AY959802 EU166969

Semnornis ramphastinus - GQ458001 GQ457984 GQ458015 Smithornis rufolateralis - AY065727 HQ998107 - Smithornis sharpei NC000879 NC000879 NC000879 NC000879 Spheniscus humboldti DQ137201 AY567916 AY567888 GQ354793

Spheniscus magellanicus DQ137200 DQ137218 EU525505 GQ354791 Steatornis caripensis EU167079 EU167023 - EU166928 Stercorarius maccormicki U76773 U76799 - - Stercorarius skua EU167080 EU167024 DQ385176 EU166933 Streptoprocne zonaris AY275845 AY294481 JQ176293 AY294543 Strigops habroptilus NC005931 NC005931 NC005931 NC005931 Strix occidentalis - - DQ434178 GU784959 Strix varia U88028 AF448260 DQ434180 - Struthio camelus NC002785 NC002785 NC002785 NC002785 Sula dactylatra EU372670 EU372677 JQ176311 EU372686 Sylvia atricapilla NC010228 NC010228 NC010228 NC010228 Sylvia crassirostris NC010229 NC010229 NC010229 NC010229 Sylvia nana - AJ534537 GQ482732 JF502343 Synthliboramphus antiquus NC007978 NC007978 NC007978 NC007978

189 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Syrrhaptes paradoxus DQ674559 - GQ482739 - Tachybaptus novaehollandiae NC010095 NC010095 NC010095 NC010095 Taeniopygia guttata NC007897 NC007897 NC007897 NC007897 Tauraco erythrolophus - AF102084 - - Thalassarche melanophrys NC007172 NC007172 NC007172 NC007172 Thamnophilus nigrocinereus - EF030331 - EF030300 Theristicus caerulescens - - JQ176448 - Theristicus caudatus - - JQ176452 -

Thinocorus orbignyianus DQ674585 JQ963056 FJ028422 JQ963040 Thinocorus rumicivorus EF373112 EF373160 - EF373270 Tigrisoma fasciatum EU167034 EU166980 JN802047 EU166937 Tigrisoma lineatum - AF193831 - - Tinamus guttatus ---- Tinamus major NC002781 NC002781 NC002781 NC002781 Tockus camurus ---- Tockus erythrorhynchus AY274008 AF082071 - AY274055 Tockus flavirostris - - EU621665 EU327669 Todiramphus sanctus NC011712 NC011712 NC011712 NC011712 Todirostrum cinereum - AF453809 JQ176509 EF501890 Todus angustirostris - AF441624 - DQ111829 Todus mexicanus - AF441628 - - Todus subulatus - AF441618 - JN568807 Trachyphonus darnaudii EU167036 EU166982 - EU166967 Treron sieboldii AY274024 AY274042 - AY274071 Treron vernans AF483284 AF483321 - - Trichoglossus haematodus - AB177942 JN801464 EU327671 Tringa incanus AY894145 AY894230 EU525563 HM640882 Trogon curucui AY274009 AY274031 JN802059 AY274056 Trogon melanurus U83763 U94805 JN802063 AY625200 Trogon personatus U89238 U89201 JQ176546 AY625198

190 Appendix 3.2. Continued. Species 12S CYTB COX1 ND2 Trogon viridis NC011714 NC011714 NC011714 NC011714 Turdus falcklandii EU154511 DQ910943 JN802070 FJ177345 Turnix pyrrhothorax - AY703820 - - Turnix suscitator - AY703822 - - Turnix sylvatica DQ385283 DQ385232 DQ385181 DQ385096 Turnix varia U83743 AF168104 AF168057 - Turtur tympanistria HM746793 HM746793 HM746793 HM746793 Tyrannus tyrannus AF173600 AY509645 JN801392 - Tyto alba EU167085 EU167029 EU410491 EU166976 Upupa epops EU167086 EU167030 GQ482885 EU166944 Uria lomvia AJ242687 U37308 GU571682 JQ435063 Urocolius indicus ---- Urocolius macrourus AY274012 AY274033 - AY274059 Vidua chalybeata NC000880 NC000880 NC000880 NC000880 Zenaida auriculata NC015203 NC015203 NC015203 NC015203 Zenaida macroura EF373325 AF182703 DQ434832 EF373359 Zosterops japonica AY136569 DQ837523 HQ608875 EU372685

Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Acanthisitta chloris EU726220 AY056975 H2008 EU739911 Accipiter gentilis E2006 DQ881796 GU189472 E2006 Accipiter striatus - - GU189478 - Aceros corrugatus ---- Acrocephalus scirpaceus FJ883145 - - FJ883111 Acryllium vulturinum ---- Aegotheles albertisi E2006 AY233362 FJ588484 E2006 Aegotheles cristatus - - FJ588483 - Aegotheles insignis GU166932 DQ482636 H2008 H2008

Aerodramus vanikorensis GU166928 - H2008 H2008 Aeronautes saxatalis GU166927 - - - Afrotis afra E2006 AY339100 E2006 E2006 Agapornis roseicollis GQ395349 GQ395351 GQ395363 - Alcedo atthis E2006 DQ111792 E2006 E2006 Alcedo leucogaster GQ861192 DQ111794 H2008 H2008 Alectoris graeca ---- Alectura lathami E2006 AF294687 H2008 H2008

191 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Alisterus scapularis - GQ505198 H2008 H2008 Alligator mississipiensis - AF143724 H2008 - Amazona autumnalis ---- Anas crecca JX137987 - - - Anas platyrhynchos EU580305 HQ902532 H2008 H2008 Andigena cucullata ---- Anhinga anhinga E2006 - H2008 H2008 Anomalopteryx didiformis - - JX533005 JX533003 Anser albifrons - DQ137227 - - Anser anser ---- Anser erythropus - - H2008 H2008 Anseranas semipalmata JQ255523 - H2008 H2008

Aphelocoma ultramarina HQ123909 - - -

Aptenodytes patagonicus - DQ137247 - - Apteryx australis E2006 - H2008 H2008 Apteryx haastii ---- Aptornis defossor ---- Apus apus E2006 AF294664 - E2006 Apus pallidus - - - JQ520113 Aramides ypecaha E2006 AY756084 - E2006 Aramus guarauna E2006 DQ881798 H2008 H2008 Aratinga pertinax JX877009 - - - Archilochus colubris ---- Ardea alba ---- Ardea cocoi E2006 DQ881799 E2006 E2006 Ardea herodias - - H2008 H2008 Ardeotis kori - - H2008 H2008 Arenaria interpres - AY228792 H2008 H2008 Asio flammeus E2006 EU348868 E2006 E2006 Asio otus - EU348876 - EU601097 Athene cunicularia - EU348871 H2008 H2008 Attagis gayi - EF373170 - - Aulacorhynchus prasinus ----

192 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Aythya americana - - H2008 H2008 Balaeniceps rex E2006 DQ881800 H2008 H2008

Batrachostomus septimus - DQ482613 H2008 H2008 Biziura lobata JQ255526 - H2008 H2008 Bombycilla cedrorum - FJ177356 - - Bombycilla garrulus EU680709 AY056981 H2008 H2008 Botaurus lentiginosus ---- Brachypteracias leptosomus E2006 AF294676 E2006 E2006 Brachypteracias squamigera - - H2008 H2008 Branta canadensis - - AY034414 - Brotogeris cyanoptera ---- Bucco macrodactylus - - H2008 H2008 Bucorvus abyssinicus - - - H2008 Bucorvus leadbeateri ---- Burhinus bistriatus - AY339103 H2008 H2008 Burhinus oedicnemus EF552714 - - - Burhinus senegalensis ---- Buteo buteo - EU345528 - - Buteo jamaicensis - EF078718 H2008 H2008 Buteogallus meridionalis E2006 AY233359 GU189505 E2006 Bycanistes brevis ---- Cacatua galerita - GQ505231 - - Cacatua sulphurea E2006 DQ881801 H2008 H2008 Caiman crocodilus - AY239166 EF646345 - Callipepla gambelii ---- Calonectris diomedea ---- Capito niger - - H2008 - Caprimulgus europaeus - DQ482634 GU586575 - Caprimulgus longirostris GU166933 DQ482621 H2008 H2008 Caracara plancus E2006 AY461409 E2006 E2006 Cariama cristata E2006 DQ881802 H2008 H2008 Casuarius casuarius - - H2008 H2008 Cathartes aura E2006 AY461395 H2008 H2008

193 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Catharus guttatus EU680712 AY307184 EF568205 DQ466820 Centropus viridis - - H2008 H2008 Cepphus columba - EF373173 - - Chalcophaps indica ---- Chalcopsitta cardinalis - - H2008 H2008 Charadrius collaris E2006 AY339106 E2006 E2006 Charadrius semipalmatus ---- Charadrius vociferus - AF143736 H2008 H2008 Chauna torquata E2006 AF143728 H2008 H2008 Chionis alba - AY339107 - AY339080 Chlidonias niger ---- Chordeiles minor - - GU586602 - Ciconia boyciana ---- Ciconia ciconia EF552716 - H2008 H2008 Cladorhynchus leucocephalus - EF373176 - - Climacteris erythrops - AY443268 H2008 H2008 Cnemotriccus fuscatus EU231928 FJ501615 - EU231825 Coccyzus americanus - DQ482640 H2008 H2008 Coccyzus erythropthalmus ---- Cochlearius cochlearius - - H2008 H2008 Colaptes auratus - - - DQ188152 Colibri coruscans GU166983 DQ482639 H2008 H2008 Colinus cristatus - - H2008 H2008 Colinus virginianus ---- Colius colius - - H2008 H2008 Colius striatus E2006 AF294669 E2006 E2006 Columba leucocephala ---- Columba livia - AY228768 H2008 H2008 Columbina passerina - - H2008 H2008 Columbina squammata E2006 EF373501 E2006 E2006 Coracias caudata E2006 AF143737 H2008 H2008 Coracias garrulus ---- Coracias spatulata ---- Coragyps atratus E2006 DQ881804 E2006 E2006

194 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Corvus corone FJ358080 AY056989 H2008 H2008 Corvus frugilegus - - - AY395583 Corythaeola cristata - - H2008 - Corythaixoides concolor ---- Corythaixoides leucogaster E2006 AF294654 E2006 E2006 Corythaixoides personata ---- Coturnix chinensis ---- Coturnix coturnix - JN979998 H2008 - Coturnix japonica ---- Coua cristata - - H2008 H2008 Cranioleuca baroni ---- Crax alector - - H2008 H2008 Crax rubra ---- Crinifer piscator --- Crotophaga ani ---- Crotophaga sulcirostris - - - H2008 Crypturellus soui - - H2008 H2008 Crypturellus undulatus ---- Cuculus canorus E2006 AF294655 E2006 H2008 Cursorius temminckii - AY228780 - AY339081 Cypseloides niger ---- Daptrius ater - AY461397 H2008 H2008 Dendragapus obscurus ---- Dendrocolaptes certhia GQ140023 FJ461166 H2008 H2008 Dendrocygna arcuata ---- Didunculus strigirostris ---- Dinornis giganteus ---- Dromaius novaehollandiae - - H2008 H2008 Dromas ardeola - HM369459 H2008 H2008 Dryocopus pileatus - - H2008 H2008 Ducula aenea ---- Egretta eulophotes ----

195 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Egretta novaehollandiae ---- Emeus crassus ---- Eudocimus albus - - H2008 H2008 Eudromia elegans - - H2008 H2008 Eudynamys taitensis ---- Eudyptes chrysocome - DQ137233 - - Eudyptula minor - DQ137235 H2008 H2008 Eupodotis ruficrista - - H2008 H2008 Eupodotis senegalensis ---- Eurostopodus macrotis - AF294661 H2008 H2008 Eurypyga helias E2006 DQ881806 H2008 H2008 Eurystomus orientalis ---- Falcipennis canadensis ---- Falco mexicanus - EF078767 H2008 H2008 Falco peregrinus - AY461399 - JF909610 Falco rufigularis - DQ881807 E2006 E2006 Falco sparverius - AY461400 - - Falco subbuteo E2006 - - E2006 Falco tinnunculus - EU233251 - - Forpus modestus JX877049 - - - Francolinus capensis ---- Fregata aquila ---- Fregata magnificens E2006 DQ881808 H2008 H2008 Fregata minor EF552722 - - - Fringilla montifringilla GU816920 AY056994 H2008 H2008 Fulica americana ---- Fulmarus glacialis E2006 DQ881809 E2006 E2006 Galbula albirostris - - H2008 H2008 Galbula cyanescens E2006 AF294682 E2006 E2006 Galbula pastazae ---- Galbula ruficauda ---- Gallicolumba luzonica ---- Gallus gallus EF552724 AF143730 H2008 H2008 Gampsonyx swainsonii - EF078725 H2008 H2008 Gavia adamsii ---- Gavia arctica E2006 - E2006 E2006

196 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Gavia immer - AF143733 H2008 H2008 Gavia pacifica ---- Gavia stellata ---- Gavialis gangeticus - AF143725 AY277490 - Geococcyx californianus - - H2008 H2008 Geopelia striata - EF373504 - - Geotrygon montana - EF373506 H2008 H2008 Geotrygon violacea ---- Glareola nordmanni ---- Glareola pratincola ---- Glaucis aeneus GU167006 - FJ588478 - Glaucis hirsuta GU167007 - - - Goura cristata - EF373507 - - Grallaria squamigera GQ140073 AY065749 AY065693 AY065778 Grallaria varia - - H2008 H2008 Grus canadensis - AF143732 H2008 H2008 Grus grus E2006 - - - Guira guira E2006 AY165799 - E2006 Gyps africanus - EF078730 - - Haematopus ater - AY228794 - - Haematopus ostralegus - AY339111 H2008 H2008 Harpactes ardens - AY625239 - AY600499 Heliornis fulica - - H2008 H2008 Hemiprocne comata - - FJ588481 - Hemiprocne longipennis E2006 AF294665 E2006 E2006 Hemiprocne mystacea GU166936 DQ482637 H2008 H2008 Hemispingus frontalis ---- Herpetotheres cachinnans - AY461402 H2008 JF909622 Himantopus mexicanus - AY228795 - -

Himantornis haematopus - - H2008 H2008 Hydrobates pelagicus E2006 DQ881811 E2006 E2006 Hylocharis chrysura E2006 AF294667 E2006 E2006 Hypsipetes amaurotis ---- Ibidorhyncha struthersii - EF373188 - -

197 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Indicator maculatus - - H2008 H2008 Indicator minor - AY165794 E2006 E2006 Indicator variegatus ---- Ispidina picta - DQ111800 AY165831 AY165813 Jabiru mycteria E2006 DQ881812 E2006 E2006 Jacana jacana E2006 AY228776 H2008 H2008 Jacana spinosa ---- Jynx torquilla - - - DQ188146 Lanius collurio E2006 AY228042 E2006 E2006 Larus atricilla ---- Larus dominicanus FJ668838 FJ012986 - - Larus marinus - AY228799 H2008 H2008 Laterallus albigularis E2006 DQ881813 E2006 E2006

Laterallus melanophaius ---- Lepidocolaptes wagleri ---- Lepidothrix coronata EU231846 AY057020 H2008 H2008 Leptodon cayanensis E2006 EF078742 GU189425 E2006 Leptosomus discolor E2006 AY233361 H2008 H2008 Leptotila verreauxi ---- Malacoptila semicincta ---- Malacorhynchus membranaceus JQ255524 - H2008 H2008

Malurus melanocephalus - AY057001 H2008 H2008 Megaceryle alcyon - DQ111803 - E2006 Megalaima haemacephala ---- Megalaima virens - AY165793 H2008 H2008 Megapodius eremita - - H2008 H2008 Megapodius freycinet - AF143731 - - XM003206 Meleagris gallopavo - 378 -- Melopsittacus undulatus - DQ143354 - -

Menura novaehollandiae EF441242 AY057004 H2008 H2008 Merops apiaster ---- Merops nubicus - - H2008 H2008

198 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Merops viridis E2006 AF294675 E2006 E2006 Mesitornis unicolor E2006 AY756082 H2008 H2008 Metallura eupogon ---- Micrastur gilvicollis - AY461403 - JF909627 Micrastur semitorquatus - AY461404 H2008 H2008 Micropsitta finschii - GQ505240 H2008 H2008 Mionectes macconnelli - AY443302 H2008 H2008 Momotus momota E2006 AF295192 H2008 H2008 Monasa nigrifrons ---- Monias benschi E2006 DQ881815 H2008 H2008 Morus bassanus E2006 DQ881831 H2008 H2008 Motacilla alba EU680699 AY228023 AY227989 AY228307 Musophaga violacea - - AY277500 - Mycteria americana E2006 DQ881816 E2006 E2006 Nandayus nenday - DQ143326 - - Neomorphus geoffroyi ---- Neophema elegans - HQ316836 - - Ninox novaeseelandiae ---- Nipponia nippon ---- Nisaetus alboniger ---- Nisaetus nipalensis - EF078759 GU189464 - Nothoprocta perdicaria - - H2008 H2008 Numida meleagris - - H2008 H2008 Nyctanassa violacea ---- Nyctibius aethereus E2006 AF294659 FJ588473 E2006 Nyctibius bracteatus GU166930 - H2008 H2008 Nyctibius grandis - DQ482612 H2008 H2008 Nyctibius griseus - - FJ588472 - Nyctibius maculosus - - FJ588470 - Nycticorax nycticorax ---- Nymphicus hollandicus ---- Nystalus maculatus E2006 AF294680 E2006 E2006 Oceanites oceanicus - - H2008 H2008 Oceanodroma tethys - - H2008 H2008 Opisthocomus hoazin E2006 AY233357 H2008 H2008 Ortalis canicollis - AY140774 AY034417 -

199 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Otidiphaps nobilis - EF373517 H2008 H2008 Otus scops - EU348925 - EU601079 Oxyura jamaicensis JQ255525 - H2008 H2008 Pachyptila desolata ---- Pandion haliaetus E2006 AJ601459 H2008 H2008 Parus atricapillus ---- Parus major EU680749 AY443314 AF377263 AY228310 Passer domesticus - EF568263 EF568232 EF449710 Passer montanus EF625341 AF143738 H2008 H2008 Pedionomus torquatus - AY228789 H2008 H2008 Pelecanoides urinatrix E2006 DQ881818 H2008 H2008 Pelecanus conspicillatus ---- Pelecanus erythrorhynchos ---- Pelecanus occidentalis - - H2008 H2008 Pelecanus onocrotalus E2006 DQ881819 E2006 E2006 Pelecanus thagus ---- Pezophaps solitaria ---- Phaenicophaeus curvirostris - - H2008 H2008 Phaethon lepturus - - H2008 H2008 Phaethon rubricauda E2006 DQ881820 H2008 H2008 Phaethornis griseogularis GU167058 DQ482638 H2008 H2008 Phaethornis syrmatophorus GU167069 - - - Phalacrocorax auritus ---- Phalacrocorax carbo E2006 DQ881821 H2008 H2008

Phalacrocorax pelagicus ---- Phalacrocorax urile HQ379809 - - - Phalaropus tricolor - AY228778 - - Pharomachrus auriceps - AY625248 H2008 H2008 Pharomachrus mocinno ---- Phegornis mitchelli - AY228781 H2008 H2008 Phodilus badius - EU348928 H2008 H2008

200 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Phoebastria nigripes E2006 DQ881805 H2008 H2008 Phoebetria palpebrata E2006 DQ881822 E2006 E2006 Phoenicopterus chilensis E2006 DQ881823 H2008 H2008 Phoenicopterus ruber ---- Phoeniculus purpureus - - H2008 H2008 Phylloscopus occipitalis - - - AY887703 Picathartes gymnocephalus GU816900 AY057019 H2008 H2008 Picoides villosus - - - DQ352431 Picumnus cirratus - AF295195 E2006 E2006 Pitangus sulphuratus EU231908 FJ501719 - EU231805 Pitta erythrogaster DQ785948 DQ320616 - DQ785984 Pitta guajana DQ785950 AY057021 H2008 H2008 Platalea ajaja ---- Platalea minor ---- Platycercus elegans - - H2008 H2008 Plegadis falcinellus ---- Ploceus cucullatus - AY057022 H2008 H2008 Pluvialis dominica - AY339115 - AY339088 Podager nacunda E2006 DQ482630 GU586624 E2006 Podargus strigoides E2006 DQ482614 H2008 H2008 Podica senegalensis E2006 DQ881824 E2006 E2006 Podiceps auritus - - H2008 H2008 Podiceps cristatus E2006 DQ881825 E2006 E2006 Podiceps nigricollis ---- Podilymbus podiceps ---- Pogoniulus pusillus ---- Porphyrio martinica ---- Psittacula alexandri - - H2008 H2008 Psittacus erithacus GU816898 EF517674 H2008 H2008 Psophia crepitans - - H2008 H2008 Psophia leucoptera E2006 DQ881826 E2006 E2006 Psophia viridis ---- Pterocles bicinctus ---- Pterocles coronatus ---- Pterocles gutturalis E2006 AY339116 E2006 E2006 Pterocles namaqua - - H2008 H2008

201 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Pterocnemia pennata ---- Pterodroma brevirostris ---- Pterodroma hasitata ---- Pteroglossus azara ---- Pteroglossus castanotis E2006 AF294686 E2006 E2006 Ptilonorhynchus violaceus JN614590 AY057026 AY064289 AY064742 Puffinus griseus - - H2008 H2008 Puffinus lherminieri E2006 DQ881827 E2006 E2006 Puffinus pacificus ---- Puffinus tenuirostris ---- Pyrrhura frontalis E2006 AY233360 E2006 E2006 Rallus limicola - - H2008 H2008 Ramphastos vitellinus ---- Raphus cucullatus ---- Regulus calendula EU680760 AY057028 H2008 H2008 Rhea americana E2006 DQ881836 H2008 H2008 Rhinopomastus cyanomelas E2006 AF294677 E2006 - Rhynochetos jubatus E2006 DQ881828 H2008 H2008 Rissa tridactyla - AY228785 - - Rollulus rouloul - - H2008 H2008 Rostratula benghalensis - AY228801 H2008 H2008 Rynchops niger E2006 AY228784 E2006 E2006 Sagittarius serpentarius E2006 EF078705 H2008 H2008 Sapayoa aenigma DQ785969 DQ320609 H2008 H2008 Sarcoramphus papa - - H2008 H2008 Sarothrura elegans - - H2008 H2008 Sayornis phoebe ---- Scolopax minor - EF373207 - - Scopus umbretta E2006 DQ881830 H2008 H2008 Scytalopus magellanicus - AY443331 H2008 H2008 Selenidera reinwardtii ---- Semnornis ramphastinus ----

202 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Smithornis rufolateralis DQ785972 AY057031 H2008 H2008 Smithornis sharpei ---- Spheniscus humboldti E2006 AF143734 - E2006

Spheniscus magellanicus - DQ137239 - - Steatornis caripensis E2006 DQ482611 H2008 H2008 Stercorarius maccormicki ---- Stercorarius skua - AY228783 - - Streptoprocne zonaris GU166929 - H2008 H2008 Strigops habroptilus ---- Strix occidentalis - DQ482641 H2008 H2008 Strix varia ---- Struthio camelus EF552743 AF143727 H2008 H2008 Sula dactylatra HQ379822 - - - Sylvia atricapilla EU680770 EF568261 AY227998 AY887727 Sylvia crassirostris ---- Sylvia nana - AY057033 H2008 H2008 Synthliboramphus antiquus - EF373212 - - Syrrhaptes paradoxus - - H2008 H2008 Tachybaptus novaehollandiae ---- ENSTGUG Taeniopygia guttata 0000001014 - 7 -- Tauraco erythrolophus - DQ482643 H2008 H2008 Thalassarche melanophrys ---- Thamnophilus nigrocinereus - AY057034 H2008 H2008 Theristicus caerulescens E2006 DQ881810 E2006 E2006 Theristicus caudatus E2006 DQ881832 E2006 E2006

Thinocorus orbignyianus - AY228803 H2008 H2008 Thinocorus rumicivorus - EF373213 - - Tigrisoma fasciatum ---- Tigrisoma lineatum E2006 DQ881833 E2006 E2006

203 Appendix 3.2. Continued. Species ODC1 RAG1 MYC MB Tinamus guttatus - AF143726 H2008 H2008 Tinamus major - - AY034421 - Tockus camurus - - H2008 H2008 Tockus erythrorhynchus E2006 AF294679 E2006 E2006 Tockus flavirostris ---- Todiramphus sanctus - DQ111817 - - Todirostrum cinereum E2006 FJ501755 E2006 E2006 Todus angustirostris - DQ111790 H2008 H2008 Todus mexicanus E2006 AF294673 E2006 E2006 Todus subulatus ---- Trachyphonus darnaudii - AF294683 AF295156 AY165825 Treron sieboldii ---- Treron vernans - - H2008 - Trichoglossus haematodus ---- Tringa incanus - AY894213 - - Trogon curucui - AY625233 - AY600506 Trogon melanurus E2006 AY625229 E2006 E2006 Trogon personatus - AY625227 H2008 H2008 Trogon viridis - AY625236 - - Turdus falcklandii - AY057039 H2008 H2008 Turnix pyrrhothorax E2006 - E2006 - Turnix suscitator ---- Turnix sylvatica - EF380262 H2008 H2008 Turnix varia - AY756083 - E2006 Turtur tympanistria ---- Tyrannus tyrannus - AF143739 H2008 - Tyto alba E2006 DQ881834 H2008 H2008 Upupa epops E2006 AF294678 H2008 H2008 Uria lomvia - EF373216 - - Urocolius indicus - - H2008 H2008 Urocolius macrourus ---- Vidua chalybeata - - H2008 H2008 Zenaida auriculata ---- Zenaida macroura - EF373530 - - Zosterops japonica FJ358079 FJ358145 - FJ357979

204 Appendix 3.2. Continued. Species FGBI5 FGBI7 Acanthisitta chloris H2008 H2008 Accipiter gentilis - E2006 Accipiter striatus -- Aceros corrugatus -- Acrocephalus scirpaceus -- Acryllium vulturinum -- Aegotheles albertisi -- Aegotheles cristatus - E2006 Aegotheles insignis H2008 H2008

Aerodramus vanikorensis H2008 H2008 Aeronautes saxatalis -- Afrotis afra - E2006 Agapornis roseicollis GQ395347 GQ395348 Alcedo atthis -- Alcedo leucogaster H2008 H2008 Alectoris graeca -- Alectura lathami H2008 H2008 Alisterus scapularis H2008 H2008 Alligator mississipiensis -- Amazona autumnalis - AY301482 Anas crecca - JX137885 Anas platyrhynchos H2008 H2008 Andigena cucullata - HQ424094 Anhinga anhinga H2008 H2008 Anomalopteryx didiformis - JX532999 Anser albifrons -- Anser anser -- Anser erythropus H2008 H2008 Anseranas semipalmata H2008 H2008 Aphelocoma ultramarina --

Aptenodytes patagonicus -- Apteryx australis H2008 H2008

205 Appendix 3.2. Continued. Species FGBI5 FGBI7 Apteryx haastii -- Aptornis defossor -- Apus apus - E2006 Apus pallidus - JQ520071 Aramides ypecaha - E2006 Aramus guarauna H2008 H2008 Aratinga pertinax JX877116 - Archilochus colubris - AY830616 Ardea alba - E2006 Ardea cocoi - EF552752 Ardea herodias H2008 H2008 Ardeotis kori H2008 H2008 Arenaria interpres H2008 H2008 Asio flammeus - E2006 Asio otus -- Athene cunicularia H2008 H2008 Attagis gayi - AY695175 Aulacorhynchus prasinus - JF424520 Aythya americana H2008 H2008 Balaeniceps rex H2008 H2008 Batrachostomus septimus H2008 H2008 Biziura lobata H2008 H2008 Bombycilla cedrorum EF468341 EF471862 Bombycilla garrulus H2008 H2008 Botaurus lentiginosus - AY695229 Brachypteracias leptosomus - E2006 Brachypteracias squamigera H2008 H2008 Branta canadensis -- Brotogeris cyanoptera - FJ667202 Bucco macrodactylus H2008 H2008 Bucorvus abyssinicus H2008 H2008 Bucorvus leadbeateri -- Burhinus bistriatus H2008 H2008 Burhinus oedicnemus - EF552756 Burhinus senegalensis --

206 Appendix 3.2. Continued. Species FGBI5 FGBI7 Buteo buteo -- Buteo jamaicensis H2008 H2008

Buteogallus meridionalis GQ265033 E2006 Bycanistes brevis -- Cacatua galerita -- Cacatua sulphurea H2008 H2008 Caiman crocodilus -- Callipepla gambelii - DQ494145 Calonectris diomedea -- Capito niger H2008 H2008 Caprimulgus europaeus --

Caprimulgus longirostris H2008 H2008 Caracara plancus JN650307 E2006 Cariama cristata H2008 H2008 Casuarius casuarius H2008 H2008 Cathartes aura H2008 H2008 Catharus guttatus EF468342 EF471863 Centropus viridis H2008 H2008 Cepphus columba - AY695193 Chalcophaps indica - AY443694 Chalcopsitta cardinalis H2008 H2008 Charadrius collaris -- Charadrius semipalmatus -- Charadrius vociferus H2008 H2008 Chauna torquata H2008 H2008 Chionis alba -- Chlidonias niger -- Chordeiles minor - AY695137 Ciconia boyciana -- Ciconia ciconia H2008 H2008 Cladorhynchus leucocephalus -- Climacteris erythrops H2008 H2008 Cnemotriccus fuscatus EF501887 -

207 Appendix 3.2. Continued. Species FGBI5 FGBI7 Coccyzus americanus H2008 H2008 Coccyzus erythropthalmus -- Cochlearius cochlearius H2008 H2008 Colaptes auratus - AY082398 Colibri coruscans - H2008 Colinus cristatus H2008 H2008 Colinus virginianus - AY952654 Colius colius H2008 H2008 Colius striatus - E2006 Columba leucocephala - AF182656 Columba livia H2008 H2008 Columbina passerina H2008 H2008 Columbina squammata - E2006 Coracias caudata H2008 H2008 Coracias garrulus -- Coracias spatulata -- Coragyps atratus - E2006 Corvus corone H2008 H2008 Corvus frugilegus - AY667448 Corythaeola cristata H2008 H2008 Corythaixoides concolor -- Corythaixoides leucogaster -- Corythaixoides personata - E2006 Coturnix chinensis -- Coturnix coturnix H2008 H2008 Coturnix japonica - AY952657 Coua cristata H2008 H2008 Cranioleuca baroni -- Crax alector H2008 H2008 Crax rubra -- Crinifer piscator -- Crotophaga ani -- Crotophaga sulcirostris H2008 H2008

208 Appendix 3.2. Continued. Species FGBI5 FGBI7 Crypturellus soui H2008 H2008 Crypturellus undulatus -- Cuculus canorus H2008 H2008 Cursorius temminckii -- Cypseloides niger -- Daptrius ater H2008 H2008 Dendragapus obscurus -- Dendrocolaptes certhia H2008 H2008 Dendrocygna arcuata -- Didunculus strigirostris -- Dinornis giganteus -- Dromaius novaehollandiae H2008 H2008 Dromas ardeola H2008 H2008 Dryocopus pileatus H2008 H2008 Ducula aenea -- Egretta eulophotes -- Egretta novaehollandiae -- Emeus crassus -- Eudocimus albus H2008 H2008 Eudromia elegans H2008 H2008 Eudynamys taitensis -- Eudyptes chrysocome -- Eudyptula minor H2008 H2008 Eupodotis ruficrista H2008 H2008 Eupodotis senegalensis - AY695152 Eurostopodus macrotis H2008 H2008 Eurypyga helias H2008 H2008 Eurystomus orientalis - AY695156 Falcipennis canadensis - DQ306966 Falco mexicanus H2008 H2008 Falco peregrinus JF899883 - Falco rufigularis -- Falco sparverius - AY695162 Falco subbuteo - E2006 Falco tinnunculus --

209 Appendix 3.2. Continued. Species FGBI5 FGBI7 Forpus modestus JX877140 - Francolinus capensis -- Fregata aquila -- Fregata magnificens H2008 H2008 Fregata minor - E2006 Fringilla montifringilla H2008 AY494568 Fulica americana - AY695244 Fulmarus glacialis - E2006 Galbula albirostris H2008 H2008 Galbula cyanescens -- Galbula pastazae -- Galbula ruficauda - AY695154 Gallicolumba luzonica -- Gallus gallus H2008 H2008 Gampsonyx swainsonii H2008 H2008 Gavia adamsii -- Gavia arctica - EF552767 Gavia immer H2008 H2008 Gavia pacifica -- Gavia stellata -- Gavialis gangeticus -- Geococcyx californianus H2008 H2008 Geopelia striata - EF373481 Geotrygon montana H2008 H2008 Geotrygon violacea - HQ993565 Glareola nordmanni -- Glareola pratincola -- Glaucis aeneus - EU042390 Glaucis hirsuta - AY830637 Goura cristata - AF182676 Grallaria squamigera GQ140165 - Grallaria varia H2008 H2008 Grus canadensis H2008 H2008 Grus grus - E2006 Guira guira - E2006 Gyps africanus - AY987182

210 Appendix 3.2. Continued. Species FGBI5 FGBI7 Haematopus ater -- Haematopus ostralegus H2008 H2008 Harpactes ardens - AY600470 Heliornis fulica H2008 H2008 Hemiprocne comata -- Hemiprocne longipennis - E2006 Hemiprocne mystacea H2008 H2008 Hemispingus frontalis -- Herpetotheres cachinnans H2008 H2008 Himantopus mexicanus - AY695203

Himantornis haematopus H2008 H2008 Hydrobates pelagicus - E2006 Hylocharis chrysura - E2006 Hypsipetes amaurotis -- Ibidorhyncha struthersii -- Indicator maculatus H2008 H2008 Indicator minor - E2006 Indicator variegatus - AY082401 Ispidina picta -- Jabiru mycteria - E2006 Jacana jacana H2008 H2008 Jacana spinosa - E2006 Jynx torquilla - AY082400 Lanius collurio - E2006 Larus atricilla - AY695186 Larus dominicanus - FJ668935 Larus marinus H2008 H2008 Laterallus albigularis - AY082411

Laterallus melanophaius - E2006 Lepidocolaptes wagleri -- Lepidothrix coronata H2008 H2008 Leptodon cayanensis - E2006 Leptosomus discolor H2008 H2008 Leptotila verreauxi - HQ993559

211 Appendix 3.2. Continued. Species FGBI5 FGBI7 Malacoptila semicincta -- Malacorhynchus membranaceus H2008 H2008

Malurus melanocephalus H2008 H2008 Megaceryle alcyon - E2006 Megalaima haemacephala - HQ874514 Megalaima virens H2008 H2008 Megapodius eremita H2008 H2008 Megapodius freycinet -- Meleagris gallopavo - AY952660 Melopsittacus undulatus --

Menura novaehollandiae H2008 H2008 Merops apiaster EU021582 - Merops nubicus EU021591 AY695157 Merops viridis - E2006 Mesitornis unicolor H2008 H2008 Metallura eupogon - GU166842 Micrastur gilvicollis JF899902 - Micrastur semitorquatus H2008 H2008 Micropsitta finschii H2008 H2008 Mionectes macconnelli H2008 H2008 Momotus momota H2008 H2008 Monasa nigrifrons -- Monias benschi H2008 H2008 Morus bassanus H2008 H2008 Motacilla alba EU680664 - Musophaga violacea -- Mycteria americana - E2006 Nandayus nenday -- Neomorphus geoffroyi -- Neophema elegans -- Ninox novaeseelandiae -- Nipponia nippon --

212 Appendix 3.2. Continued. Species FGBI5 FGBI7 Nisaetus alboniger -- Nisaetus nipalensis - AY987189 Nothoprocta perdicaria H2008 H2008 Numida meleagris H2008 H2008 Nyctanassa violacea - AY695231 Nyctibius aethereus - E2006 Nyctibius bracteatus H2008 H2008 Nyctibius grandis H2008 H2008 Nyctibius griseus -- Nyctibius maculosus -- Nycticorax nycticorax -- Nymphicus hollandicus -- Nystalus maculatus - E2006 Oceanites oceanicus H2008 H2008 Oceanodroma tethys H2008 H2008 Opisthocomus hoazin H2008 H2008 Ortalis canicollis - AY140710 Otidiphaps nobilis H2008 H2008 Otus scops -- Oxyura jamaicensis H2008 H2008 Pachyptila desolata -- Pandion haliaetus H2008 H2008 Parus atricapillus - HQ417160 Parus major DQ320586 HQ417161 Passer domesticus - AY494571 Passer montanus H2008 H2008 Pedionomus torquatus H2008 H2008 Pelecanoides urinatrix H2008 H2008 Pelecanus conspicillatus - JX683941 Pelecanus erythrorhynchos - E2006 Pelecanus occidentalis H2008 H2008 Pelecanus onocrotalus - JX683948 Pelecanus thagus - JX683955 Pezophaps solitaria -- Phaenicophaeus curvirostris H2008 H2008

213 Appendix 3.2. Continued. Species FGBI5 FGBI7 Phaethon lepturus H2008 H2008 Phaethon rubricauda H2008 H2008 Phaethornis griseogularis H2008 H2008 Phaethornis syrmatophorus - EU042418 Phalacrocorax auritus - AY695211 Phalacrocorax carbo H2008 EF552777

Phalacrocorax pelagicus -- Phalacrocorax urile -- Phalaropus tricolor - AY695184 Pharomachrus auriceps H2008 H2008 Pharomachrus mocinno -- Phegornis mitchelli H2008 H2008 Phodilus badius H2008 H2008 Phoebastria nigripes H2008 H2008 Phoebetria palpebrata - E2006 Phoenicopterus chilensis H2008 H2008 Phoenicopterus ruber - E2006 Phoeniculus purpureus H2008 - Phylloscopus occipitalis -- Picathartes gymnocephalus H2008 H2008 Picoides villosus - U67904 Picumnus cirratus - E2006 Pitangus sulphuratus -- Pitta erythrogaster DQ320592 - Pitta guajana H2008 H2008 Platalea ajaja - AY695214 Platalea minor -- Platycercus elegans H2008 H2008 Plegadis falcinellus -- Ploceus cucullatus H2008 H2008 Pluvialis dominica - AY695201

214 Appendix 3.2. Continued. Species FGBI5 FGBI7 Podager nacunda - E2006 Podargus strigoides H2008 H2008 Podica senegalensis - E2006 Podiceps auritus H2008 H2008 Podiceps cristatus - E2006 Podiceps nigricollis -- Podilymbus podiceps - AY695145 Pogoniulus pusillus -- Porphyrio martinica -- Psittacula alexandri H2008 H2008 Psittacus erithacus H2008 H2008 Psophia crepitans H2008 H2008 Psophia leucoptera - DQ881988 Psophia viridis - E2006 Pterocles bicinctus - E2006 Pterocles coronatus -- Pterocles gutturalis -- Pterocles namaqua H2008 H2008 Pterocnemia pennata -- Pterodroma brevirostris -- Pterodroma hasitata -- Pteroglossus azara - HQ424102 Pteroglossus castanotis - E2006 Ptilonorhynchus violaceus - EU341403 Puffinus griseus H2008 H2008 Puffinus lherminieri - DQ881991 Puffinus pacificus -- Puffinus tenuirostris - E2006 Pyrrhura frontalis - E2006 Rallus limicola H2008 H2008 Ramphastos vitellinus - GQ423011 Raphus cucullatus -- Regulus calendula H2008 H2008 Rhea americana H2008 H2008 Rhinopomastus cyanomelas - E2006 Rhynochetos jubatus H2008 H2008

215 Appendix 3.2. Continued. Species FGBI5 FGBI7 Rissa tridactyla -- Rollulus rouloul H2008 H2008 Rostratula benghalensis H2008 H2008 Rynchops niger - E2006 Sagittarius serpentarius H2008 H2008 Sapayoa aenigma H2008 H2008 Sarcoramphus papa H2008 H2008 Sarothrura elegans H2008 H2008 Sayornis phoebe -- Scolopax minor -- Scopus umbretta H2008 H2008 Scytalopus magellanicus H2008 H2008 Selenidera reinwardtii - HQ424122

Semnornis ramphastinus - HQ424125 Smithornis rufolateralis H2008 H2008 Smithornis sharpei -- Spheniscus humboldti - E2006

Spheniscus magellanicus -- Steatornis caripensis H2008 H2008 Stercorarius maccormicki -- Stercorarius skua -- Streptoprocne zonaris H2008 H2008 Strigops habroptilus -- Strix occidentalis H2008 H2008 Strix varia -- Struthio camelus H2008 H2008 Sula dactylatra - AY695212 Sylvia atricapilla EU680691 - Sylvia crassirostris -- Sylvia nana H2008 H2008 Synthliboramphus antiquus --

216 Appendix 3.2. Continued. Species FGBI5 FGBI7 Syrrhaptes paradoxus H2008 H2008 Tachybaptus novaehollandiae -- Taeniopygia guttata -- Tauraco erythrolophus H2008 H2008 Thalassarche melanophrys -- Thamnophilus nigrocinereus H2008 H2008 Theristicus caerulescens - E2006 Theristicus caudatus - E2006

Thinocorus orbignyianus H2008 AY695176 Thinocorus rumicivorus -- Tigrisoma fasciatum -- Tigrisoma lineatum - E2006 Tinamus guttatus H2008 H2008 Tinamus major -- Tockus camurus H2008 H2008 Tockus erythrorhynchus - DQ882000 Tockus flavirostris - E2006 Todiramphus sanctus -- Todirostrum cinereum EF501858 E2006 Todus angustirostris H2008 H2008 Todus mexicanus - E2006 Todus subulatus HQ530138 - Trachyphonus darnaudii -- Treron sieboldii -- Treron vernans H2008 H2008 Trichoglossus haematodus -- Tringa incanus -- Trogon curucui - AY600477 Trogon melanurus - E2006

217 Appendix 3.2. Continued. Species FGBI5 FGBI7 Trogon personatus H2008 H2008 Trogon viridis - AY695161 Turdus falcklandii H2008 H2008 Turnix pyrrhothorax -- Turnix suscitator -- Turnix sylvatica H2008 H2008 Turnix varia - E2006 Turtur tympanistria -- Tyrannus tyrannus H2008 H2008 Tyto alba H2008 H2008 Upupa epops H2008 - Uria lomvia -- Urocolius indicus H2008 H2008 Urocolius macrourus -- Vidua chalybeata H2008 H2008 Zenaida auriculata - AF182667 Zenaida macroura - AY082416 Zosterops japonica --

218 Appendix 3.3: Composite taxa with multiple genera

Composite Taxon Species Included

Puffinus griseus, Puffinus tenuirostris, Puffinus Puffinus+Calonectris pacificus, Calonectris diomedea, Puffinus lherminieri

Corythaixoides leucogaster, Crinifer piscator, Corythaixoides+Crinifer Corythaixoides concolor, Corythaixoides personata

Pterocles bicinctus, Pterocles namaqua, Pterocles Pteroclididae gutturalis, Pterocles coronatus, Syrrhaptes paradoxus

Charadrius collaris, Charadrius semipalmatus, Charadrius+Phegornis Charadrius vociferus, Phegornis mitchelli

Urocolius macrourus, Colius striatus, Colius colius, Colius+Urocolius Urocolius indicus

219 CONNECTING TEXT

To estimate the absolute rate of evolution using discrete morphological characters, two quantities are necessary: the total quantity of morphological evolution (morphological change on branches of a phylogenetic tree) and the time duration of the corresponding branches. Focusing on a case study using modern birds, in the preceding chapter I estimated a posterior sample of 10 000 dated phylogenies that provides a plausible set of phylogenetic relationships and branch durations for major lineages of modern birds. In the following chapter, I used this distribution and combined it with an exceptionally large and complete data set of discrete morphological characters to estimate absolute rates of anatomical evolution in modern bird lineages through time in a phylogenetic context. I built on previous methods and, using the distribution of phylogenies generated in the third chapter, I extend these methods to incorporate phylogenetic and divergence time uncertainty. I further incorporated the results of the second chapter by testing the discrete gamma and lognormally-distributed models for among-character rate heterogeneity and by applying the preferred model to the estimation of morphological branch lengths using likelihood-based methods.

220 CHAPTER 4

ESTIMATING ABSOLUTE RATES OF EVOLUTION USING DISCRETE

MORPHOLOGICAL CHARACTERS ACROSS AN UNCERTAIN

PHYLOGENY: RATES OF ANATOMICAL EVOLUTION IN MODERN

BIRDS

Rates of evolution are at the centre of fundamental questions in evolutionary biology. Paleontologists have examined rates of phenotypic evolution through time, although not always in an explicitly phylogenetic framework (e.g. Simpson, 1944; Haldane, 1949) and there has also been considerable interest in the study of rates of phenotypic evolution on ecological timescales (e.g. Hendry and Kinnison, 1999). Recently, there has been renewed focus on estimating rates of phenotypic evolution from comparative data in a phylogenetic framework (e.g. Harmon et al., 2010; Venditti et al., 2011). Reconstructing rates of phenotypic evolution on phylogenies may offer insights into whether rates of evolution are punctuated through time and related to speciation and adaptive radiation (e.g. Gould and Eldredge, 1977; Adams et al., 2009; Rabosky and Adams, 2012) or to molecular evolutionary rates (e.g. Davies and Savolainen, 2006), among other questions (Lloyd et al., 2012). Comparatively little attention has been devoted to the use of discrete morphological characters (DMCs) to estimate absolute evolutionary rates (Lloyd et al., 2012). This is reflected in a diversity of methods to reconstruct rates of evolution of continuous traits on phylogenies and advanced likelihood-based approaches to test models of evolution (e.g. O'Meara, 2006; Eastman et al., 2011; Revell et al., 2012) but fe w, mostly maximum parsimony-based methods for discrete characters (e.g. Ruta et al., 2006, Lloyd et al., 2012; Chapter 1, this volume). This is surprising for four reasons. First, DMCs are w idely used for systematic study and extensive data is available in curated databases (e.g. Morphobank; O'Leary and Kaufman, 2011). Second, the use of DMCs to

221 reconstruct absolute rates of evolution has a long history of use in mostly paleontological studies (e.g. Westlock, 1949; Derstler, 1982; Forrey, 1988; Cloutier, 1991; Wagner, 1997; Ruta et al., 2006; Brusatte et al., 2008; Roelants et al., 2011; Lloyd et al., 2012). Third, where the evolution of continuous traits is usually investigated at the level of one, two or tens of traits, hundreds (e.g. Luo et al., 2007; Carrano et al., 2012) and recently, thousands (e.g. Livezey and Zusi, 2006; 2007; Naish et al., 2012; O'Leary et al., 2013) of DMCs coded for hundreds of taxa are already available. This rich source of morphological data potentially offers more information to parameterize phenotypic rates at the organismal level. Fourth, although issues remain (e.g. Goloboff and Pol., 2006; Sterli et al., in press), DMCs can be modeled in likelihood-based analysis using continuous time Markov chains. Although these models are simpler than most molecular sequence models (e.g. Lewis, 2001), the underlying approach is similar. This offers the possibility that developments for modeling the rate of molecular evolution might be applied to matrices of DMCs. Most current methods to reconstruct absolute rates of morphological evolution using DMCs share certain common features (Chapter 1, this volume; see also Brusatte's [2011] review). Using a phylogeny and a matrix of DMCs, morphological evolution in terms of number of character changes per branch is estimated using maximum parsimony (e.g. Brussatte et al., 2008; Lloyd et al., 2012) and/or a likelihood-based method estimating branch lengths in units of character changes per character (e.g. Roelants et al., 2011). Then, changes on a single branch or a group of branches are divided by an estimate of the corresponding temporal duration (Brusatte, 2011). These durations have typically been estimated directly from the fossil record (see e.g. Brusatte, 2011), although Roelants et al. (2011) used Bayesian divergence time dating in extant taxa. Most previous studies generally relied on two assumptions: a single or handful of reconstructed phylogenies that were frequently estimated from the same DMC matrix and usually fixed temporal durations (but see Roelants et al., 2011 and Lloyd et al., 2012 for attempts to accommodate uncertainty). These assumptions are problematic as parsimony optimization of character changes can be sensitive to taxonomic sampling, topology and choice of optimization scheme

222 (ACCTRAN/DELTRAN) for ambiguous reconstructions (Dececchi and Larsson, 2009; Brusatte, 2011; Lloyd et al., 2012). Furthermore, point estimates of branch durations clearly underestimate the uncertainty in major timing events, especially if the underlying topology is variable. Finally, maximum parsimony may underestimate changes on long branches because it cannot account for multiple changes in the same character along a branch (e.g. Bromham et al., 2002). In this paper, previous methods are extended to relax phylogenetic and divergence time assumptions by using empirical distributions of dated phylogenies derived from independent data. Likelihood-based methods are also used to reconstruct morphological evolution, testing explicit models of rate heterogeneity among characters. This approach is demonstrated using an iconic vertebrate radiation with uncertainty in both phylogeny and divergence times, coupled with one of the largest available data sets of DMCs. Extant birds have been the subject of intensive phylogenetic study (e.g. Sibley and Ahlquist, 1990; Fain and Houde, 2004; Ericson et al., 2006; Livezey and Zusi, 2007; Hackett et al., 2008; Pratt et al., 2009; Braun et al., 2011; McCormack et al., 2013). Despite this, there is still uncertainty among several deep divergences and about the timescale of neornithean evolution (Ericson et al., 2006; Brown et al., 2007; 2008; Hackett et al., 2008; Brown and van Tuinen, 2011; Livezey, 2011), which may require genome-level sequence sampling and careful attention to gene-tree discordance to resolve (Braun et al., 2011; Haddrath and Baker, 2012; McCormack et al., 2013). Despite the lack of a fully resolved phylogeny and precise divergence times, two recent studies provide distributions of dated phylogenies that allow these uncertainties to be incorporated (see below). Coupled with these tree distributions, there is a wealth of morphological data available for representatives of major extant lineages of birds. Livezy and Zusi (2006; 2007), using the anatomical framework created by Baumel (1993), discretized the avian anatomical phenotype into 2954 DMCs scored for ~150 neornithean species, with coverage of most non-passerine groups. Even when compared with a recent collaborative approach that yielded ~4500 characters for mammals (O'Leary et al., 2013), the Livezey and Zusi (2006; 2007) data set is

223 unique in its completeness (< 2% missing data for extant species) despite minor reported errors (e.g. Mayr, 2007a). The radiation of Neoaves has been described as rapid (e.g. Ericson et al., 2006; Braun et al., 2011; Chapter 3, this volume) but the tempo of aggregate phenotypic evolution across these taxa has not been explored using DMCs. A prediction of the theory of adaptive radiation is that phenotypic evolutionary rates should be at first rapid as new adaptive zones are occupied, followed by rate decreases as niches are occupied and/or constraints are encountered (e.g. Schulter 2000; the “early-burst” model sensu Harmon et al., 2010). The Livezey and Zusi (2006; 2007) data set, hereafter referred to as LZ2007, offers a rich sample of DMCs to explore the tempo of morphological change of the whole avian anatomical phenotype across this iconic group's early evolution and through time. Initially conceived to reconstruct avian phylogeny, LZ2007's utility for this purpose has been criticized in that, because of its exhaustive approach, it includes characters with a high degree of homoplasy and convergence that drown any phylogenetic signal (Mayr, 2007a). By using an independently-derived phylogeny and timescale, it is possible to sidestep this issue entirely and to estimate rates of evolution directly. Livezey and Zusi (2007) also retained autapomorphic characters in their matrix both to facilitate character coding at lower taxonomic levels and expressly to allow parameterization of evolutionary rates at the tips of the phylogeny (Livezey and Zusi, 2007), which would ordinarily be problematic in most data sets (Bromham et al., 2002; Müller and Reisz, 2006; Seligmann, 2010). Matrices of DMCs have also been used to address the question of mosaic evolution of morphology in early birds (Clarke and Middleton, 2008) and amphibians (Roelants et al., 2011). Matrices of DMCs are easily partitioned, and rates can be modeled separately to test hypotheses of mosaic evolution. In a Bayesian model testing framework, Clarke and Middleton (2008) reported evidence for mosaic evolution on the extinct stem lineage of modern birds. The objective of this study is to demonstrate how rates of phenotypic evolution can be estimated using DMCs, integrating across phylogenetic and divergence time uncertainty by using empirical tree distributions. The large

224 LZ2007 data set is used to demonstrate the approach and to estimate absolute aggregate rates of morphological evolution of avian anatomy across major clades of modern birds and through time. Using these estimated rates, the hypothesis that the aggregate rates of phenotypic evolution were highest early in the evolution of birds will be examined. Quantitative differences in evolutionary rates between major clades sampled by LZ2007 will also be examined. Compared to previous studies, this study more carefully controls for phylogenetic and divergence uncertainty and also models among-character rates using explicitly tested models. Further, a simple anatomical partitioning scheme for LZ2007 is examined to test for evidence of mosaic evolution among gross anatomical partitions in a Bayesian model testing framework.

MATERIALS AND METHODS The approach applied here to calculate absolute rates of morphological evolution using DMCs requires (see also Brusatte [2011] for a review focused on paleontological data sets): 1) a data set of discrete morphological characters, 2) a chronogram (or chronograms) with branch durations in units of time and 3) a method to estimate the quantity of morphological evolution (i.e. character change / divergence) on each branch of the phylogeny. Finally, these are combined to estimate absolute rates of evolution by 4) dividing the quantity of morphological evolution by the corresponding branch's temporal duration.

1) Morphological Data LZ2007 was retrieved from the electronic supplementary material provided with the publication of Livezey and Zusi (2006). First, all extinct taxa were removed as were characters that had the same state in all remaining taxa. All characters coded as 'x' or noncomparable were set as missing. The remaining variable characters were recoded so that states were numbered beginning with zero. Character deletions and recoding were accomplished using an R script (R Core Team, 2012; available from the authors). Recoding states is important as many software packages determine state space by highest state observed. This

225 resulted in a data matrix with 145 genera coded for 2237 variable characters, with a total of 2% missing data or 6981 data cells out of 324 365. In all analyses, characters were considered unweighted and unordered. The data set was further partitioned into five anatomical divisions following the hierarchy created by Baumel (1993; character assignments in Appendix 4.1): soft tissue and general myology (MY, 230 characters), cranial (AC, 505 characters), axial/vertebral (AV, 211 characters), forelimbs (FL, 683 characters) and hindlimbs (HL, 608 characters). Because of the completeness of this data set, corrections for incomparable characters (e.g. patristic dissimilarity sensu Wagner, 1997) were not applied. The partitioned data matrix is available from the authors upon request.

2) Chronogram Distributions Calculation of absolute rates of evolution of discrete morphological characters requires an explicit timescale. Here, two separate distributions of chronograms, Harrison and Larsson (Chapter 3, this volume; hereafter HL) and Jetz et al. (2012; hereafter JZ2012) spanning the taxonomic sampling of LZ2007 were used to provide two alternate hypotheses of the timescale and phylogeny of modern bird relationships. Both distributions of chronograms were pruned to match the taxonomic sampling of LZ2007 using R and the APE package (Paradis, 2004; R Core Team, 2012; see Supplementary Methods for details). Maximum clade credibility (MCC) trees summarizing the distributions were calculated using treeannotator (Drummond et al., 2012; details in Supplementary Methods). Both MCC trees were plotted using the R package phyloch (Heibl, 2008): HL MCC chronogram is included as Figure 4.1 and the JZ2012 MCC chronogram as Supplementary Fig. 4.1.

3) Reconstruction of Morphological Evolution Three methods were applied to reconstruct morphological evolution on the branches of a given topology using LZ2007: maximum parsimony (PAUP* v4.0b10 using ACCTRAN and DELTRAN optimizations; Swofford, 2003), maximum-likelihood (RAxML v7.2.8; Stamatakis, 2006) and Bayesian inference

226 (MrBayes v3.2; Ronquist et al., 2012b). Likelihood-based methods estimate branch lengths, while the maximum parsimony methods optimize character changes onto branches; although not the same quantity (see e.g. Felsenstein, 2004), for the purposes of this study, both are referred to as “morphological branch lengths”. Per branch parsimony character changes were scaled by dividing by the total number of characters to yield estimates of changes per character for comparison with likelihood-based methods. Several modifications were made to RAxML and MrBayes to model the evolution of discrete morphological characters (see Supplementary Methods for modifications and details on branch length estimation for both methods) including implementation of the discrete lognormal distribution to model among-character rate heterogeneity (Wagner, 2012; Harrison and Larsson, Chapter 2, this volume). To determine the most appropriate model for among-character rate heterogeneity for RAxML and MrBayes, a Bayesian model testing approach was followed using MrBayes (see details in Supplementary Methodology). The anatomical partitioning schemes described in 1) were also tested using the same approach (see details in Supplementary Methodology). Two partitioned models were tested: one model where relative rates of evolution were varied between character partitions but branch lengths were jointly estimated and a second where branch lengths were unlinked by partition. The former model allows partitions to have higher or lower rates overall but the same pattern of rates across branches, while the latter model effectively allows each partition to have different rates on each branch (see Clarke and Middleton, 2008 for more details).

4) Estimating Rates of Evolution: MCC Chronograms For each MCC topology, evolutionary rates were estimated for each branch, using morphological branch lengths optimized under Parsimony (ACCTRAN, DELTRAN), maximum likelihood (RAxML) and Bayesian inference (MrBayes). Both likelihood-based methods applied lognormally distributed rates among characters as this model was selected by the Bayesian model selection (see Results and Table 4.1). Morphological branch lengths were then divided by the

227 corresponding branch durations of the MCC chronogram. Branch lengths estimated using MrBayes incorporated the partitioning model of best fit (relative rates between partitions, exponential prior on lognormal shape parameters; see Table 4.1). Rates and cumulative morphological branch lengths were visualized following the method used by Roelants et al. (2011): each MCC tree was plotted by drawing each path length from Neognathae to each tip, placing nodes at the corresponding node ages and the cumulative morphological branch lengths under each optimization. The slope of branches therefore represents the estimated rate of evolution of that branch. Significance of high rates under the RAxML optimization was tested by using a simulation procedure similar to that of Roelants et al, (2011; details in Supplementary Methodology). Estimated branch rates were compared against a distribution of corresponding branch rates estimated by RAxML for each branch on 1 000 stochastically simulated data sets on the MCC topologies under a constant-rates model with a one-tailed comparison (i.e. branches were considered significantly high if the observed rate was > 9 500 simulated rates for that branch). Branch lengths estimated from the simulated data sets were standardized so that the tree length equaled that of the LZ2007 data (see Supplementary Methods). All simulations and plots were constructed in R using the APE package with scripts available from the authors upon request (Paradis, 2004; R Core Team, 2012).

4) Estimating Rates of Evolution: Chronogram Distributions To incorporate phylogenetic and branch duration uncertainty, rates were estimated from LZ2007 on all 10 000 chronograms in each distribution using both parsimony (PAUP*: ACCTRAN and DELTRAN) and maximum-likelihood (RAxML, lognormal rates) optimization of morphological branch lengths. The above procedure described for the MCC trees was repeated for each of the 10 000 individual chronograms in each distribution. This yielded a set of 10 000 chronograms for each distribution, with estimated morphological branch lengths and rates of evolution assigned to each branch on each chronogram. An R script using the APE package was used to control the computations by calling RAxML

228 and PAUP* and processing the output trees to divide morphological branch lengths by the corresponding branch durations (script available from the authors). It was not computationally feasible to perform the MrBayes analyses across the distribution of chronograms, as each tree required > 4 days of analysis. These analyses were conducted on the CLUMEQ/Guillimin computing cluster as the computation on a personal computer would have been impractical (~ 1 core- month total). An assumption of uniform rates of change through time has been adopted as a null model to interpret the recovered rates of evolution for discrete morphological characters (e.g. Roelants et al., 2011; Lloyd et al., 2012) and has theoretical justification (Larsson et al., 2012). Null rate distributions were estimated using two different simulation procedures, both assuming a constant rate of evolution: the branch randomization method of Lloyd et al. (2012) and the stochastic simulation approach of Roelants et al., (2011; full details in Supplementary Methodology). The stochastic simulation was primarily considered because it incorporated among-character rate-heterogeneity and produced simulated data sets that could be analyzed using each optimization method. Use of the branch randomization method was confined to the parsimony methods to compare with the stochastic simulations (see results). The stochastic simulations were computationally intensive and only a single stochastically simulated data set was generated on each of the 20 000 total chronograms. Null rate distributions were generated by repeating the analysis described above using the simulated data sets and by standardizing recovered branch lengths to the total tree length recovered using the LZ2007 data set for each chronogram and morphological branch length optimization method (see Supplementary Methodology).

Rates of Evolution of Monophyletic Clades Because it was not possible to visualize the reconstructed rates using the Roelants et al. (2011) method for a distribution of chronograms, rates of evolution were compared between selected clades that were monophyletic in all

229 chronograms (> 4 genera: defined in Fig. 4.1; Supp. Fig. 4.1 and Table 4.2). To compare rates at the base of the neoavian radiation, a set of branches was chosen for each of the 10 000 chronogram by selecting branches up to three nodes distant from the node Neoaves, not including any branches to tips (Roelants et al., 2011 also defined similar groups of branches on a single tree). This provided an objective definition of the basal branches across the variable topologies and divergence times (see Fig. 4.1 and Suppl. Fig. 4.1 for this definition on the MCC trees). Rates were estimated by dividing the sum of morphological branch lengths by the corresponding total temporal duration of all branches within the group (Roelants et al., 2011 and Lloyd et al., 2012 used a similar method). This procedure was repeated for each of the 10 000 chronograms in each distribution using the RAxML morphological branch length estimates derived above. Both chronogram distributions included branch lengths that were very short (e.g. < 100 years), and as these may have been artifacts of the divergence dating, all branches with durations less than 100 000 years were excluded. This pruning is further justified by the fact that most geological ages are constrained to accuracies of only approximately 500 000 years, and it seems unlikely that divergence dates, based in part on fossil calibrations, could be more accurate than geological dating methods. The distribution of estimated clade rates was then log10 transformed and visualized using boxplots. A log10 scale for rates was used because observed branch rates differed by orders of magnitude (rates were also log transformed by Ruta et al., 2006). The stochastically simulated data sets and estimated null rates were analyzed in the same manner to provide expectations of clade rates under a constant rate of evolution. Previous studies have compared rates of evolution between binned branch rates using Mann-Whitney tests (e.g. Brusatte et al., 2008), the extension of these tests across a 10 000 chronogram distribution coupled with multiple comparisons between clades is not clear. Therefore patterns were interpreted graphically using the 95% (2.5% to 97.5%) quantile ranges of clade rate estimates across the tree distributions, with reference to rates derived from the simulated data sets.

230 Visualizing and Testing Absolute Rates of Evolution Through Time Drawing on the molecular clock literature, a rates through time heatmap (see Drummond and Suchard, 2010 for an application to influenza evolution) was generated to visualize the complex distribution of the 10 000 chronograms with reconstructed rates through time. To generate the heatmap, the time duration from the upper 95% credibility interval of the age of Neornithes for the Harrison and Larsson chronogram distribution (130.2 Ma ago) to the present was split into 300 equally spaced time bins. For each time bin, the chronogram distribution was iterated across and for all branches appearing within the time interval of the bin, the corresponding branch rates were log10 transformed and binned into 500 rate bins (chosen to encompass the range of observed branch rates across all chronograms) for that time interval. Frequencies were then standardized for each time bin individually to produce a relative frequency histogram of rates in each time bin. The two deepest branches leading to the outgroup (Paleognathae) and ingroup (Neognathae) and branches of duration less than 100 000 years were ignored (see above). Heatmaps were generated using R and the APE package (scripts available from the authors). The heatmap approach captures uncertainty in divergence times and rates, and does not require the arbitrary choice of a point estimate of branch age (e.g. branch end point: Brown et al., 2008 or a fossil age based procedure: Lloyd et al., 2012). Null model expectations were calculated by generating a heatmap for rates estimated from the stochastically simulated data sets and recording the 95% (2.5% to 97.5%) quantile range of values recorded in each time bin: these quantile ranges were then plotted onto the heatmaps for each chronogram distribution estimated from LZ2007. Heatmaps and null distributions were generated using the RAxML, ACCTRAN and DELTRAN rate estimates for each chronogram distribution derived above. To test the hypothesis that rates of evolution decreased through time (i.e. consistent with an “early-burst” hypothesis), Kendall's rank correlation test was performed between estimated branch rates and ages (Ruta et al., 2006). This non- parametric test does not assume an underlying distribution for the data. Point estimates of branch ages were approximated by the temporal mid point of the

231 branch (Ruta et al., 2006 employed stratigraphic information in their fossil data set). All branches of duration less than 0.1 Ma and the branches from the root to Neognathae and Paleognathae were excluded, as above. The null hypothesis was that branch rates are not correlated to branch age. If the hypothesis of higher rates early in modern bird evolution is correct, a significant negative correlation should exist between branch rates and ages (measured towards the present). The test was first conducted on the MCC trees under all optimizations (ACCTRAN, DELTRAN, RAxML and MrBayes) using Kendall's one-tailed correlation test implemented in R (R Core Team, 2012). Then, to incorporate phylogenetic and divergence time uncertainty, this procedure was repeated for each chronogram in each tree distribution to derive a distribution correlation coefficients. This approach is similar to the method of Martins (1996) and Houseworth and Martins (2001), but these authors used linear regression and parametric correlations, where an estimate of variance from the model fit for each tree is available and could be pooled. In this analysis, that variance was ignored and the results were interpreted with respect to correlation coefficients calculated from the null rate distribution derived from the stochastic simulations. R was used to perform the calculations using the APE package and the cor function (Paradis, 2004; R Core Team, 2012) and, the tests were conducted twice with branches leading to terminal taxa either included or excluded to mitigate the possible confounding factor that terminal branch rates might be systematically underestimated (see discussion; R scripts available from the authors).

Aggregate Morphological and Molecular Evolutionary Rates Through Time Statistical testing of the relationship between morphological and molecular evolutionary rates requires careful design (see Davies and Savolainen, 2006; Seligmann, 2010). Such tests are beyond the scope of this study but a simple visual comparison of aggregate rates of evolution through time was made by comparing the heatmap generated using the RAxML optimization of morphological rates (as above) with a heatmap generated by the same methods using the rates of molecular evolution recovered by the HL divergence time

232 analysis (Chapter 3, this volume). In that analysis, branch rates of molecular evolution were estimated using the uncorrelated lognormal relaxed molecular clock (see full details in Chapter 3, this volume) and were treated the same way as morphological rates for heatmap construction. Because the molecular analysis included approximately twice the number of genera, this comparison is meant to be general and exploratory and is not a statistical test of correlation between these rates.

RESULTS Among-character Rate Heterogeneity and Partition Testing Results from the Bayesian model selection analysis are presented in Table 4.1. Bayes factors for the comparison between the gamma-distributed and lognormal-distributed rates models were approximately 60 across all prior combinations and on both MCC topologies, indicating that there was very strong evidence for the lognormal model relative to the gamma model (Kass and Raftery, 1995; Nylander et al., 2004). The lognormal among-character rate variation model was therefore used for all likelihood-based analyses. Further model testing was conducted to determine if partitioning characters by anatomical regions increased model fit. The model with the highest marginal likelihood was to allow rates to vary proportionally among partitions relative to a single set of branch lengths (Table 4.2; BF > 100 relative to the next model under either exponential or uniform priors). The model with unlinked branches rates was not supported relative to unpartitioned models and the proportional rates model. The posterior mean estimates and 95% highest posterior density (HPD) credibility intervals for the relative rate parameters were (all ESS > 200): myology and soft- tissue characters: 0.68 [0.61, 0.75], cranial characters 0.88 [0.82, 0.94], vertebral characters: 1.36 [1.26, 1.46], forelimb characters: 1.01 [0.96, 1.06] and hindlimb characters 1.08 [1.023, 1.15].

Rates of Evolutionary Change on MCC Trees

233 Results from the RAxML optimization of morphological branch lengths are included as Figure 4.2 for the HL MCC and JZ2012 MCC trees, highlighting four focal clades: Neoaves, Charadriiformes, Columbiformes and Passeriformes (RAxML, ACCTRAN and DELTRAN optimizations for the HL tree for all focal clades in Table 4.2 are included as Supp. Figs. 4.4, 4.6 and 4.7, respectively and RAxML optimization for the JZ2012 tree for all focal clades is included as Supp. Fig. 4.5). Branches highlighted in red exceeded the 95% quantile of estimated rates for that branch based on 1 000 simulated data sets under a constant rate model on each MCC chronogram. Highest rates of evolution were concentrated among many branches at the base of Neoaves and especially among stem lineages of crown groups both before and across the K–Pg boundary (greatest slopes in Fig. 4.2). Some slowly evolving branches were also observed at the base of neoavian radiation (short, flat branches in Fig. 4.2: Neoaves). Most but not all clades were characterized by higher rates of evolution along their stem lineages and slower rates in the crown groups (Fig. 4.2, Suppl. Figs. 4.4, 4.5). Passeriformes and Columbiformes, represented by 8 and 6 genera respectively, displayed a pattern of relatively lower accumulation of changes within clades (Fig. 4.2). Charadriiformes, represented by 22 genera, exhibited higher rates of evolution within the clade than expected under a constant-rates model for seven lineages, including for example, the lineage to the highly derived Turnix (Fig. 4.2: Charadriiformes, arrows). The MrBayes and RAxML optimizations were very similar (see Suppl. Fig. 4.3 relative to Fig. 4.2) and suggested that the anatomical partition scheme did not strongly influence relative branch lengths. ACCTRAN and DELTRAN optimizations differed from the RAxML optimizations in that ACCTRAN concentrated more changes on early branches before and around the K–Pg boundary and DELTRAN optimized more changes towards the tips (Suppl. Figs. 4.6 and 4.7, respectively). However, in all cases, the general pattern of highest rates of evolution in many branches at the base of neoavian radiation and along stem lineages was recovered, although DELTRAN optimization recovered the weakest such pattern (Suppl. Fig. 4.7).

234 Rates of Evolution of Focal Clades Rates of evolution estimated for focal clades using RAxML branch length optimization are recorded in Table 4.2 and displayed graphically in Figure 4.3 for the HL and JZ2012 tree distributions. The group of branches designed as the basal neoavian radiation had the highest median evolutionary rate across the 10 000 chronograms, and the 95% (2.5% to 97.5%) quantile range of rates did not overlap with the 95% quantile range of rates for the same branches estimated from 10 000 data sets simulated under a constant-rates model (Fig. 4.3, Basal Radiation). Variability among basal branch rates between chronograms was very high (see 95% quantile range, Fig 4.3 a,b: Basal Radiation). Rates of evolution of focal clades were generally much lower than expected under a constant-rates model derived from the stochastically simulated data. Clade rates were not obviously related to the number of genera present for either tree distribution, although Charadriiformes had among the highest rates of evolution and was the best sampled clade (Fig. 4.3). In both tree distributions, Galloanserae (Galliformes+Anseriformes) had lower than expected rates of evolution and lower rates of evolution relative to Neoaves as a whole, but was similar to focal clades within Neoaves. This was especially true in the HL tree distribution, likely because Harrison and Larsson (Chapter 3, this volume) recovered an older age for the Galliformes/Anseriformes split than did JZ2012 (see Fig. 4.1 and Supp. Fig. 4.1). Although clade rates estimated under each tree distribution were generally congruent, several differences emerged. For example, Ardeidae, the heron family, was estimated to have higher evolutionary rates in HL relative to JZ2012. Inspection of the MCC trees and Figure 4.2 suggested that this difference was driven by topology and divergence times. The mean age of the most recent common ancestor for this group in the HL distribution was 30.6 Ma ago with a 95% credibility interval of [21.2–40.7] Ma ago. In JZ2012 distribution, the same node had a mean age of 55.6 Ma ago and a credibility interval of [46.0–64.9] Ma ago. Additionally, Jetz et al. (2012) fixed the sister taxon of this node to Threskiornithidae (see Jetz et al., 2012: Suppl. Fig. 1) whereas the HL analysis was unfixed and recovered a weak alliance between Ardeidae and the

235 Pelecanus+Scopus+Balaeniceps group (posterior probability = 0.53). The much younger age explained the higher rates observed in the HL distribution. Procellariiformes also differed in rates for similar reasons: mean ages for the most recent common ancestors were 48.1 and 55.7 Ma ago, respectively, for HL and JZ2012. In the case of Galliformes, the reverse pattern was recovered: higher rates in JZ2012 are likely due to a younger estimate of clade age (60.2 vs. 72.1 Ma ago). In summary, although rate estimates for some clades were different, likely due to differences in clade age and to a lesser extent, topology, overall patterns were congruent between the two chronogram distributions and suggested both a) rate heterogeneity between extant clades and that b) highest rates were concentrated early in avian evolution and were lower than expected in crown clades (although see discussion regarding tip rate estimates, below).

Absolute Rates of Evolutionary Change Through Time Heatmaps visualizing rates of evolution through time for HL and JZ2012 under the RAxML optimization are included as Figure 4.4. The heatmaps revealed a relatively diffuse, strongly multimodal distribution of branch rates through time, but with a general declining trend towards the present. The median of the distribution in each time bin was plotted and corroborated the trend of higher rates in the Late Cretaceous and early Paleogene, declining towards the present (solid lines, Fig. 4.4). Variability in rates was very high in the Late Cretaceous, with the highest rates, as well as near zero rates of morphological evolution (Fig 4.4, time slice i., 80.290–79.856 Ma ago). Rates were slightly less variable and remained higher at the K–Pg boundary (Fig. 4.4, time slice ii., 66.402–65.968 Ma ago) and then generally declined towards the present, but retained a multimodal distribution with numerous sub-peaks at time slices iii. (40.362–39.928 Ma ago) and iv. (20.398–9.964 Ma ago). Branch rate expectations based on 10 000 simulated data sets are indicated by the dashed lines in Figure 4.4 which represent the 2.5% to 97.5% quantile range of standardized branch rates per time bin estimated by RAxML from the stochastically simulated data sets. The median branch rate fell within this range in the Cretaceous but dropped below constant rate expectations

236 after approximately 40 Ma before present. However, in the Late Cretaceous there was significant rate density that fell above and below this range. High rate variability appeared to be driven especially by very short branch lengths at the base of Neoaves, even after branches < 0.1 Ma in duration were removed (see Fig. 4.1 and Suppl. Fig. 4.1). This was reflected in the stochastically simulated data sets by greatly increased variability, as short branches either randomly accumulated greater or fewer character changes. Although similar in overall trend, there were qualitative differences between the pattern recovered in HL (Fig. 4.4a) and JZ2012 (Fig. 4.4b): the stochastic simulation revealed greater variability in JZ2012 in the Late Cretaceous, probably associated with greater numbers of very short branches: across 10 000 chronograms, JZ2012 had 84621 branches of time duration between 0.1 and 0.5 Ma while HL and 42573. The branch rate density around -3.7 log10 (changes/character/Ma) in the JZ2012 heatmap (Fig. 4.4b) in the Cretaceous corresponded to the long branch leading to Eudromia: this branch was characterized by very low rates and the wide credibility interval in JZ2012 for the split between Eudromia and other paleognaths [59.6– 106.3] Ma ago meant that this branch's rate was recorded throughout the Cretaceous. HL recovered Eudromia in a derived position and therefore did not recover the same pattern. Finally, very low rates at approximately -8 log10 (changes/character/Ma) corresponded to morphological branch lengths that were estimated to be near zero by RAxML. These are probably explained by a general lack of synapomorphies uniting some basal divergences in these topologies, especially when quickly evolving characters were taken into account by the among-character rate heterogeneity model in RAxML (see below and discussion). ACCTRAN and DELTRAN optimizations of morphological branch lengths yielded a similar pattern of higher rates in the Late Cretaceous (HL: Suppl. Fig. 4.8; JZ2012 similar, not shown), especially under the ACCTRAN optimization, as this polarized ambiguous changes towards the root of the phylogeny. When character changes were simulated using Lloyd et al.'s (2012) branch randomization method, which assumed a single uniform rate of character change, higher variability was found in the Late Cretaceous, probably associated with

237 short branches (Suppl. Fig. 4.8, dotted lines). However, branch rates through time estimated using the parsimony methods on the stochastically simulated data sets revealed a surprising pattern: although the highest and lowest rates of evolution in the Late Cretaceous were outside the 95% quantile range of rates estimated from the stochastically simulated data (see dashed lines in Supp. Fig. 4.8), these quantile ranges, based on data sets simulated under a constant rate of evolution, were surprisingly similar in overall pattern to the rates estimated from LZ2007, especially under ACCTRAN optimization. Further, rates reconstructed from the stochastic simulated data sets were not constant through time: they were actually higher in the Late Cretaceous and Paleocene, not simply more variable as recovered by the RAxML analysis (Supp. Fig. 4.8). The simulated data sets expressly included the degree of rate heterogeneity present in LZ2007 (which was high, with lognormal shape parameter σ2 ≈ 1.7 on the HL MCC topology, see also Chapter 2: Fig. 2.1, this volume). A possible but speculative explanation for the observed patterns is that very short branches preferentially accrued changes from quickly evolving characters in the stochastic simulation. In the maximum- likelihood analysis, quickly evolving characters would be given less weight in branch length estimation because rate-heterogeneity among characters was explicitly modeled (Lee and Worthy, 2012). Parsimony, at least in the unweighted form used here, considers all character changes equiprobable across all branches and characters (Felsenstein, 2004; Lee and Worthy, 2012), and possibly reconstructed changes of these quickly evolving characters on most branches: therefore leading to relatively increased rates because the shortest branches were concentrated at the base of most chronograms (see Fig. 4.1), especially when those changes are ambiguously reconstructed and optimized under ACCTRAN. Quickly evolving characters in LZ2007 may also explain the lack of near zero branch rates recorded by the ACCTRAN and DELTRAN optimizations relative to RAxML. Correlation Test for the Early-Burst Hypothesis Results from Kendall's correlation test generally supported the hypothesis that the rate of evolution decreased towards the present, but with strong

238 qualifications (Table 4.3; Fig. 4.5). Excluding tips and the branches from the root, this pattern was consistently significant across all analyses using the HL MCC and JZ2012 MCC trees, except RAxML on JZ2012 where the correlation was not significant. Including tips, correlations were always significant across all methods and trees (Table 4.3). Correlations were most significant on MCC trees for the parsimony-based methods, including DELTRAN, which would be expected to optimize ambiguous changes towards the tips. The RAxML optimization always estimated weaker correlations, probably because, as observed with in the heatmap analysis, many branches were estimated with near-zero rates in the Late Cretaceous. MrBayes optimization on the MCC trees did not recover these near zero branch rates, possibly because the (potentially inappropriate, see Zhang et al., 2012) independent priors on branch lengths made near zero branch lengths unlikely. The distribution of correlation coefficients estimated on the distributions of chronograms under RAxML optimization corroborated the correlation tests on the MCC trees (Fig. 4.5). The distribution of correlation coefficients and its 95% (2.5% to 97.5%) quantile range overlapped with zero only for JZ2012, excluding tips (Fig. 4.5c). The distribution of the correlation coefficients estimated from the null rate distributions overlapped slightly with the observed data but the 2.5% to 97.5% intervals did not overlap in all other plots (Fig. 4.5a,b,d). In summary, the correlations tests generally supported the hypothesis that rates declined through time although this was qualified by weaker support in the Jetz et al. chronogram distribution and the RAxML optimization in particular.

DISCUSSION Modular Evolution in the Livezey and Zusi (2007) Data Set A model partitioning characters into anatomical divisions was supported when the entire phylogenetic breadth of crown birds was considered. However, only relative rates differed between partitions and there was no evidence for a model that allowed branch lengths (and therefore branch rates) to vary individually between partitions. It would have been ideal to examine other

239 partition schemes of varying complexity and test these against randomized partition assignments (e.g. as in Clarke and Middleton, 2008). However, model likelihood estimation using stepping-stone sampling (see Supplementary Methodology) on this large data set was computationally intensive and each analysis required approximately two core-weeks of computation, which precluded such a comprehensive analysis. Vertebral characters were found to be the fastest evolving in general and myological characters the slowest. Further characterization of the relative rates of partitions was beyond the scope of this study but emphasizes the potential utility of discrete morphological characters to investigate mosaic evolution in a phylogenetic context (Clarke and Middleton, 2008; Brusatte et al., 2008; Dececchi and Larsson, 2009; Roelants et al., 2011). This result differed from the similar Bayesian analysis of Clarke and Middleton (2008) who recovered strong evidence for partitioned models with unlinked branches in a data set of DMCs for stem birds. The most likely explanation for this difference was methodological: Clarke and Middleton (2008) used the harmonic mean estimator (HME) to estimate model marginal model likelihoods. The HME is known to be biased and to consistently overestimate likelihoods for more complex models relative to more appropriate path sampling approaches like stepping-stone sampling used here (Fan et al., 2011; Xie et al., 2011). Indeed, unlinking branch lengths across partitions greatly increased model complexity: here, the unlinked-branch length model had 4.8 times the number of parameters relative to the linked-rates model (Table 4.1). Because stepping-stone sampling incorporates the prior in the estimate of marginal likelihood, this highly dimensional model was penalized (Xie et al., 2011). However, it is not possible to exclude the possibility that Clarke and Middleton's (2008) data set of characters exhibited greater rate heterogeneity among partitions and branches without reanalysis of that data set using stepping-stone sampling. These results suggest that future Bayesian model selection studies investigating the partitioning of DMCs are warranted and should use stepping-stone sampling or an equivalent approach to accurately estimate model likelihoods.

240 Estimating Rates of Evolution of Discrete Morphological Characters The estimation of absolute rates of evolution using discrete morphological characters was extended here across an empirical distribution of chronograms. The results of this analysis were complex and it is clear that relaxing assumptions of fixed phylogenetic topology and divergence times engendered high variability in the recovered rate estimates. RAxML analyses performed well on the stochastically simulated data sets and recovered approximately constant rates of evolution among characters simulated under that model, although much greater variability was associated with the short deep branches at the base of Neoaves in the Late Cretaceous (Fig. 4.4). Simulations under a constant-rates stochastic model also revealed a possible caveat of using maximum parsimony-based methods. As implemented here, parsimony methods treated all characters changes equiprobably across characters and branches (Lee and Worthy, 2012); it is possible that characters simulated at higher rate categories were so variable that they were reconstructed on most internal branches and because of the uneven distribution of short branches, led to higher reconstructed evolutionary rates in the Late Cretaceous (Suppl. Fig. 4.8). Further simulation studies are required to understand the specific relationships between rate heterogeneity, maximum parsimony and estimated morphological branch lengths to fully understand these patterns. Relative to a single measure like the mean, the heatmaps used here permitted the visualization of the complex, multimodal distribution of branch rates active on the phylogeny through time. Calculating a single measure of central tendency per time slice and clade (e.g. as in Venditti et al. 2011) may have masked the complex and highly variable underlying distribution of branch rates in this data set. Binning branch rates according to time period may be more appropriate given the temporally staggered sampling in paleontological data sets (e.g. Brusatte et al., 2008) but not for extant taxa. Here, there was also evidence for the difficulties of accurately parameterizing rates across very short branches, both in terms of branch durations and in terms of morphological branch lengths. Because these are estimated independently in this method, the quantities appeared to interact and produce very high variability which complicated the analysis (e.g.

241 Fig. 4.4, time slice i). Fully model-based approaches that treat branch durations as inputs may alleviate the problem by estimating rates as directly related to branch durations through an explicit model of evolution (see Future Research, below).

Implications for the Evolution of Modern Birds The results of this analysis were complex and in some cases ambiguous, but some consistent patterns could be distinguished both through time and across clades and lineages. First, rates of anatomical evolution through time were highly variable in the Late Cretaceous (Fig. 4.4), coincident with the base of the neoavian radiation (Fig. 4.1; Suppl. Fig. 4.1). Indeed, rates were both highest and lowest during this time period, outside of expected stochastic variation (Fig. 4.4, i). Despite these low branch rates observed in the Late Cretaceous using the RAxML analysis, rates of evolution were generally higher before and across the K–Pg and then declined towards the present (Fig. 4.4, Suppl. Fig. 4.8), but median rates were within the constant rate expectations until approximately 40 Ma before present, when they dropped below expectations. Reconstructions of cumulative morphological change on the MCC trees (Fig. 4.2; Suppl. Figs. 4.4–4.7) and rates in the clade level analysis (Fig. 4.3) suggested this pattern was driven by higher rates of evolution in some branches at the base of the neoavian radiation, and especially along stem lineages to the modern orders in the Late Cretaceous and early Paleogene, and declined towards the present in the focal clades. The temporal resolution of this analysis did not permit a careful study of the relationship of rates of phenotypic evolution to the K–Pg boundary and possible ecological implications of the mass extinction (e.g. Penny and Phillips, 2004; Lindow, 2011), but higher rates of morphological evolution (Fig. 4.2, 4.4) and basal divergences (Fig. 4.1) are concentrated before, across and after the boundary. The very high and near zero rates of evolution observed in the Late Cretaceous probably reflected two trends: although many branches in the early evolution of Neoaves in the Late Cretaceous accumulated many changes over a short period of time (Fig. 4.2), other short branches accumulated very little

242 change, especially when accounting for quickly evolving characters (e.g. see Fig. 4.2 vs. Suppl. Fig. 4.6); this may have reflected a lack of morphological synapomorphies on some branches and may be related to poorly supported nodes uniting some of the diverse clades and enigmatic taxa at the base of the neoavian radiation (e.g. Fig. 4.1: Opisthocomus, Leptosomus, Phaethontidae, etc). Basal relationships were also poorly supported in Livezey and Zusi's (2007) phylogenetic analysis of this data set. Correlation tests supported the hypothesis of an “early-burst” of phenotypic evolution across the HL distribution of chronograms, but depended on tips being included for JZ2012 under the RAxML optimization. In all other optimizations, the correlation was significant (Table 4.3). This corroborates the generally declining pattern of rates through time suggested by the heatmaps and MCC plots. Inclusion of the tips in correlation analyses always lead to significantly decreasing rates through time. This is because rates of character change were slower within the focal clades examined here (Fig. 4.3), although it is possible that underscoring of autapomorphies and insufficient taxonomic sampling may also explain some of this trend (see limitations, below). In summary, given all the evidence, the hypothesis of declining rates through time was tentatively supported, with the qualifications discussed above. Although results from a recent analysis argued early bursts of morphological evolution in comparative data are relatively rare (Harmon et al., 2010), the same study found evidence of an early-burst of body size evolution across birds but not individual clades. Studies using discrete morphological characters have generally also recovered this pattern (e.g. Ruta et al., 2006; Roelants et al., 2011; Lloyd et al., 2012). These results are therefore consistent with the hypothesis that the early evolution of modern birds, especially neoavians, was characterized by an adaptive radiation with higher rates of phenotypic evolution, here estimated using whole anatomical phenotype. Ideally, the rates of phenotypic evolution estimated here would also be examined in concert with measures of disparity and diversification to further characterize the relationship between rates, diversification, and disparity through time in order to better characterize the tempo and mode of modern bird evolution (e.g. Roelants et

243 al., 2011; Rabosky and Adams, 2012). As well, the early-burst hypothesis would ideally be further examined in this data set by testing an explicit model of decreasing rates of evolution (e.g. as in Harmon et al., 2010), but this may require fully model-based approaches (see Future Directions, below). Fossil taxa would ideally have been included in this analysis, to simultaneously break stem branches and further localize changes in characters by better estimating ancestral states (e.g. Finarelli and Flynn, 2006). However, because the fossil record of modern birds is sparse in the Cretaceous (e.g. Lindow, 2011; Brocklehurst et al., 2012), few informative fossils exist. In fact, although there are probable candidates, only one or two undisputed crown bird fossils were present in the Cretaceous. For example, Vegavis, an anseriform, is known from Maastrichtian rocks dated at approximately 66 Ma ago from Vega Island near Antarctica (Clarke et al., 2005). Some well characterized fossils from the Paleocene/Eocene would provide better resolution to the temporal pattern of character evolution (e.g. stem Psittaciformes fossils: Ksepka et al., 2011b; Ksepka and Clarke, 2012 and stem Coliiformes: Ksepka and Clarke, 2010a). These fossils were used to calibrate the chronograms used here (see Chapter 3) but were not included in LZ2007. The exhaustive set of characters in LZ2007 lays the groundwork for future studies to incorporate fossil taxa, provided they can be placed in the chronograms, which may be possible with emerging “total- evidence” methods facilitated by the Harrison and Larsson supermatrix (e.g. Ronquist et al., 2012a, Wood et al., 2013; Chapter 3, this volume).

Limitations of the Present Analysis There are several possible confounding factors affecting the conclusions of this analysis. The node density effect (NDE) and underestimation of morphological evolution on long branches are related. The node density effect is a positive relationship between the number of nodes through which a lineage passes and the total quantity of evolutionary change estimated (Fitch and Beintema, 1990; Venditti et al., 2006). This is because longer unbroken branches may underestimate change through hidden homoplasy which may be revealed

244 with increased taxon sampling (Sanderson, 1990). Parsimony-based methods are considered to be particularly susceptible to the NDE as they cannot account for multiple substitutions in a single character (e.g. Felsenstein, 2004). In the case of this analysis, the NDE would be expected to bias results by relatively underestimating the rate of evolution on longer branches in the reconstruction of morphological branch lengths. However, likelihood-based methods may be less susceptible to this bias because they can account for multiple substitutions per character per lineage (e.g. Bromham et al., 2002 but see Hugall and Lee, 2007). Because the patterns estimated in this analysis were generally consistent across reconstruction methods it is less likely that the NDE has severely biased the conclusions here. Another potential confounding factor in this analysis is underestimation of branch rates on lineages towards the tips of a phylogeny. Matrices of DMCs are typically constructed to resolve the phylogenetic relationships of the taxa of interest and therefore frequently exclude autapomorphies (Lewis, 2001; Müller and Reisz, 2006; Seligmann, 2010). This may lead to underestimates of morphological branch lengths and therefore rates on the terminal branches of the phylogeny (Bromham et al., 2002). LZ2007 should be comparatively robust to this effect as Livezey and Zusi (2007) expressly included autapomorphies for this reason and the data set is relatively comprehensive. However, rates are generally lower on tip lineages here, although not universally so (e.g. Fig. 4.2). Furthermore, Seligmann (2010) argued that although terminal branches may be underestimated, they are parameterized using actual data rather than nodal reconstructions, which introduce a further confounding effect. In any case, the correlation analyses excluding tips still recovered a signal of declining rates through time for the HL tree distribution, although the JZ2012 distribution was equivocal. Further taxonomic sampling of focal clades and careful attention to the inclusion of autapomorphies would be ideal but the effort associated with adding a single taxon to this data set (e.g. coding 2954 character states) remains an impediment.

245 Broad-scale Patterns of Molecular and Morphological Evolutionary Rates Examination of the aggregate behaviour of branch rates of morphological and molecular evolution of time (Fig. 4.6) revealed what appear to be broadly different aggregate patterns through time. However, because the original Harrison and Larsson divergence time analysis included approximately twice the number of genera, this plot was exploratory and was intended only to comparatively visualize the aggregate relationships between rates and time derived from each analysis. Rates did not appear to be correlated in a general sense, with molecular rates lowest in the Late Cretaceous and highest towards the present and on the lineages to Neoaves and Galloanserae (oldest branches considered here), while the morphological rates had generally the opposite aggregate pattern (see above). However, these aggregate patterns may mask underlying positive correlations between rates at the branch or clade levels (e.g. Davies and Savolainen, 2006; Seligmann, 2010) and the relationships between the two rates should not be over- interpreted, especially as taxonomic sampling differed greatly between the underlying analyses and many confounding factors may have been present. In future analyses, time-series tests may be used to explore these types of patterns more carefully, including against other factors such as environmental variables, though they require careful attention to taxonomic sampling, phylogeny and statistical design.

FUTURE RESEARCH

The approach presented above extended previous work on estimating evolutionary rates using DMCs to better incorporate divergence time and phylogenetic uncertainty. However, relative to the Bayesian approaches available for continuously-valued characters (e.g. Eastman et al., 2011; Vendetti et al., 2011; Revell et al., 2012), it is still early in the evolution of rate estimation using DMCs. In particular, the analysis presented above did not expressly model the rate of evolution of the rate of evolution and only estimated its outcome by using

246 independently-derived branch durations to deconfound rate from time. An ideal modeling approach to reconstructing absolute rates of change of discrete morphological characters should have the following properties (see also Chapter 1, this volume): a) be a Bayesian approach that samples dated chronograms as empirical priors from the posterior distribution of a divergence time analysis directly, perhaps simultaneously with that analysis (e.g. Ronquist et al., 2012a, Wood et al., 2013) b) make explicit assumptions to model the rate of evolution of the rate of evolution or its outcome, building on existing molecular clock models (e.g. random local clocks: Drummond and Suchard, 2010, lognormally-distributed uncorrelated clocks: Drummond et al., 2006 or autocorrelated models: Aris- Brosou and Yang, 2002). Such a method is within reach: the software package BEAST contains MCMC operators to iterate over empirical tree distributions, flexible clock models and the ability to model DMCs using the Lewis (2001) Mk model (e.g. Pyron, 2011; Wood et al., 2013). The authors of the present study further modified BEASTv1.8.0pre (Drummond et al., 2012) to implement the ascertainment-corrected MkV model (Lewis, 2001) and lognormally-distributed rates across characters. Early experiments using this modified version of BEAST and modeling rate evolution using random local clocks by sampling the Harrison and Larsson chronogram distribution appeared congruent with the above analysis. However, there are significant concerns over the sufficiency of existing clock operators when applied to DMCs, and MCMC mixing was poor. As well, chronograms sampled from the Harrison and Larsson (Chapter 3) distribution were found to be quite discretely separated in the model's likelihood space. This led the MCMC sampler to select a single chronogram from the empirical prior distribution and to not explore others. These problems are not likely to be intractable, however, and it is the authors' hope that a fully Bayesian approach to modeling absolute morphological rates from DMCs across empirical prior chronogram distributions will soon be possible.

247 CONCLUSIONS

This study extended current methodological approaches to the estimation and visualization of rates of evolution of discrete morphological characters to integrate across uncertain phylogeny and divergence times. Using Bayesian model testing, the discrete lognormal model of among-character rate heterogeneity was supported over the gamma-distributed rates model, and both were supported over an equal-rates model for this data set. Contrary to a previous study, there was no evidence that unlinking branch lengths between anatomical partitions of characters led to better model fit, although a relative-rates model using anatomical partitions was supported for LZ2007. Using maximum-likelihood and parsimony- based methods on two distributions of 10 000 chronograms, estimates of absolute branch rates of anatomical evolution from the Livezey and Zusi (2007) data set were found to be highly variable in general and especially during the Late Cretaceous, coincident with the base of the neoavian radiation. Very high rates of evolution and very low rates were active in the Late Cretaceous. However, in aggregate, the relative density of higher rates was greater in the Late Cretaceous and early Paleogene and declined towards the present. Analysis of rates of evolution by pooling branches in the basal neoavian radiation and by examining lineage plots on MCC trees supported the hypothesis that rates of evolution were generally higher along some basal branches of Neoaves and stem ordinal lineages and were lower in the focal clades examined here. Excluding the more slowly evolving terminal lineages, correlation between rates and time before present was always significantly negative in the Harrison and Larsson chronogram distribution but depended on rate optimization method for the Jetz et al. (2012) distribution. However, when tip lineages were included, both distributions had significant negative correlations between branch rates and time before the present. The results of this study are qualified by high variability in the data, patchy taxonomic sampling of crown clades and possible underestimation rates on terminal branches. However, in aggregate, across rate measures used here, the results were

248 generally consistent with an early-burst of evolutionary change and concurred with a previous study that supported the hypothesis that modern birds may constitute an example of an adaptive radiation with high early rates of phenotypic evolution.

249 K-Pg

Anser Anas Anseriformes Anseranas Chauna Meleagris Dendragapus Gallus Fancolinus Alectoris Galliformes Callipepepla Numida Ortalis Megapodius Amazona Melopsittacus Psittaciformes Trichoglossus Cacatua Parus Passer Bombycilla Aphelocoma Ptilonorhynchus Passeriformes Menura Pitangus Pitta Caracara BR Falco Cariama Cathartes Accipiter Gyps Accipitridae Pandion Sagittarius Pharomachrus Trogon Tockus Phoeniculus Upupa Merops Coracias Brachypteracias Momotus "Coraciiformes" Todiramphus Alcedo Todus Monasa Galbula BR Picoides Jynx Indicator Piciformes Ramphastos Aulacorhynchus Semnornis Megalaima Strix Otus Strigiformes Phodilus Tyto Colius Leptosomus Phalaropus Tringa Rostratula Jacana Thinocorus Pedionomus Turnix Rissa BR BR Chlidonias Rhynchops Stercorarius Charadriiformes Uria Glareola Cursorius Dromas Cladorhynchus Himantopus Haematopus Ibidorhyncha Pluvialis Chionis Burhinus Afrotis Crotophaga Geococcyx Cuculus Cuculiformes Coccyzus Centropus BR Corythaixoides Plegadis Platalea Botaurus Nycticorax Ardea Ardeidae Cochlearius Tigrisoma Pelecanus Balaeniceps BR Scopus Anhinga Phalacrocorax Sula "Suliformes" Fregata Oceanties Phoebastria Pachyptila Procellariformes Puffinus Pelecanoides Speniscus Gavia Ciconia Grus Aramus Psophia "Gruiformes" Heliornis Porphyrio Rhynochetos Eurypyga Mesitornis BR Podiceps Phoenicopterus Phaethon Geopelia Ducula Goura Columbiformes Didunculus BR Zenaida Columba Pterocles Aegotheles Glaucis BR Cypseloides Apus Apodiformes Hemiprocne Caprimulgus Podargus Nyctibius Steatornis BR Opisthocomus Apteryx Casuarius Dromaius Eudromia Rhea Struthio Cretaceous Paleogene Neogene Early Late Pal. Eocene Oligo. Miocene P. 120 100 80 60 40 20 0 MYa

250 Figure 4.1. Maximum clade credibility chronogram summarizing the Harrison and Larsson (Chapter 3, this volume) chronogram distribution, reduced to the sampling of the Livezey and Zusi (2007) data set. Nodes were placed at the means of the respective marginal posterior distributions of node heights. Clade posterior probabilities are indicated by coloured dots: red PP < 0.95; yellow: 0.95 ≤ PP < 0.99; and green: PP ≥ 0.99. Blue bars represent the 95% highest posterior density credibility interval for node ages. Boxes labeled “BR” indicate branches included in the operational definition of the basal neoavian radiation on this topology (see text). Clade names indicated by bars and text on the right margin are used operationally to identify monophyletic clades in this analysis (see also Table 4.2 for assignments of genera to clades). These generally correspond to group names used by Hackett et al. (2008). The solid red line indicates the location of Cretaceous–Paleogene (K–Pg) boundary. Geological time scale follows Gradstein et al. (2012).

251 Neoaves Neoaves RAxML RAxML HL JZ2012 Changes/Character

Columbiformes Columbiformes RAxML RAxML HL JZ2012 Changes/Character

Charadriiformes Charadriiformes RAxML RAxML HL JZ2012 Changes/Character

Passeriformes Passeriformes RAxML RAxML HL JZ2012 Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

252 Figure 4.2. Plots relating morphological branch change to branch duration across four focal clades for the Harrison and Larsson (Chapter 3, this volume; left) and Jetz et al. (2012; right) maximum clade credibility chronograms. The paleognath outgroup is not figured. All root to tip paths are plotted using cumulative morphological branch lengths estimated using RAxML and the lognormal distribution of among-character rates (see text). Branch slope represents branch rates, and those branches coloured in red exceeded the rates observed for the corresponding branch in 950/1000 stochastically simulated data sets under an constant rate model (see text). The vertical red line indicates the Cretaceous– Paleogene (K–Pg) boundary. Note how the difference in phylogenetic topology between the two MCC chronograms changes the distribution of evolutionary changes and rates on lineage to Columbiformes.

253 a) # Genera [changes/character/Ma] 10 log

GalloanseraeNeoaves*Basal RadiationAccipitridaeAnseriformesApodiformesArdeidaeCharadriiformesColumbiformes"Coraciifomres"CucliformesGalliformes"Gruiformes"PasseriformesPiciformesProcellariiformesPsittacifomresStrigiformes"Suliformes"

b) # Genera [changes/character/Ma] 10 log

GalloanseraeNeoaves*Basal RadiationAccipitridaeAnseriformesApodiformesArdeidaeCharadriiformesColumbiformes"Coraciifomres"CucliformesGalliformes"Gruiformes"PasseriformesPiciformesProcellariiformesPsittacifomresStrigiformes"Suliformes"

 Figure 4.3. Distribution of rates of evolution for focal monophyletic clades (see Fig. 4.1 and Table 4.2 for definitions) across the 10 000 chronogram distributions. Mean rates were calculated for each chronogram by summing total morphological branch lengths estimated by RAxML for all branches within the clade and dividing by the clade's total branch duration. The results for the 10 000 chronograms were then log10 transformed and plotted using boxplots for the a) Harrison and Larsson and b) Jetz et al. (2012) chronogram distributions. Distributions of null rates for clades based on 10 000 simulated data sets (1 per chronogram, see text) are included as paired box plots to the right of the observed data, coloured grey. The body of the boxplots indicates the interquartile range, the solid line the median, the whiskers are placed at the 2.5 and 97.5% quantiles of the distribution, and values beyond that range are not shown. The box plots are split into two subplots of the two deepest branching neornithean clades: Neoaves and Galloanserae (Galliformes+Anseriformes). The basal radiation of Neoaves, as defined by all branches up to three nodes distant from Neoaves but not including tips was treated as above to calculate mean evolutionary rates (see text). The mean rate distribution figured for Neoaves* does not include these branches. The second subplot shows rates for focal neoavian clades monophyletic across all trees. The graph immediately above the box plots shows the number of genera in each clade. The basal radiation group of branches had the highest median rate across both chronogram distributions.

255 256 Figure 4.4. Heatmap visualization of the complex distribution of branch rates through time across 10 000 chronograms using the RAxML estimations of morphological branch lengths. Heatmaps for a) Harrison and Larsson (Chapter 3, this volume) and b) Jetz et al. (2012) are drawn from the 118.1 Ma ago (the age of the upper 95% credibility interval for Neognathae in the Harrison and Larsson chronogram distribution) to the present. Plotted from this age, there are 274 equally spaced time bins and 500 equally spaced rate bins. These heatmaps consist of 274 relative frequency histograms of branch rates active on the phylogeny during each time bin. Relative frequency histograms are shown for four time slices to the right of each heatmap (i., ii., iii. and iv.), these correspond to time bins of 80.290–79.856, 66.402–65.968, 40.362–39.928 and 20.398–9.964 Ma before present, respectively. Time bin ii. spans the Cretaceous–Paleogene (K– Pg) boundary and is indicated by the red vertical line. The four relative frequency histograms figured to the right of each heatmap have an x-axis that corresponds to the y-axis of the heatmap and all histograms have an identical range on the y-axis of [0–0.03]. The solid black lines in the main plot represent the median rate in each time bin. Dashed lines represent the 2.5% to 97.5% quantile range of rates for each time bin estimated using RAxML from 10 000 stochastically simulated data sets, one per chronogram, under an a constant rate model (see Supplementary Methodology for details). Note the increased variability in both rate estimates from the Livezey and Zusi (2007) data and from stochastic simulations near time slice i., probably related to very short branches (< 1 Ma) and lack of synapomorphies near the base of Neoaves (see Fig. 1; discussion). Rates are multimodal across the entire time range and are both highest and lowest near time slices i. and ii. which approximately correspond to the base of the neoavian radiation in the Late Cretaceous and many interordinal divergences at the K–Pg boundary (see Fig 1, Supp. Fig. 1).

257 a) Harrison and Larsson b) Harrison and Larsson Tips Excluded Tips Included Frequency

c) Jetz et al., (2012) d) Jetz et al., (2012) Tips Excluded Tips Included Frequency

Kendall's Tau (τ) Kendall's Tau (τ)

258 Figure 4.5. Results of the non-parametric correlation test between branch rates and branch age (measured towards the present) across 10 000 chronograms. The test was repeated for each chronogram distribution a–b) Harrison and Larsson (Chapter 3, this volume) and c–d) Jetz et al. (2012). Kendall's tau (τ) coefficients were calculated twice: first excluding branches to terminal taxa a), c) and including these branches b), d). In all cases, the temporal midpoint of the branch was used to estimate its age. For each plot, τ was calculated from the Livezey and Zusi (2007) data set for each of the 10 000 chronograms (blue distribution) and from 10 000 simulated data sets under a constant rate of evolution model (one/chronogram/distribution, see Supplementary Methodology for details). A negative correlation indicates branch rates are lower towards the present which supports the “early-burst” model of high rates of morphological evolution (see text). Although no specific hypothesis test was applied, the 2.5% to 97.5% quantile intervals (values beyond those quantiles are coloured red) for the actual data do not overlap those of the simulated data sets in a), b) and d) and were always less than zero but strongly overlap and included 0 in c). Conservatively interpreted, there is evidence for a decrease in rates through time except for the Jetz et al. (2012) data set, with tips excluded.

259 260 Figure 4.6. Heatmap visualization of the complex distribution of morphological branch rates a) through time across 10 000 chronograms using the Livezey and Zusi (2007) data set and the Harrison and Larsson (Chapter 3, this volume) chronogram distribution and b) molecular branch rates derived from the posterior distribution of 10 000 chronograms estimated by the Harrison and Larsson's (Chapter 3, this volume) Bayesian divergence time analysis. Heatmaps are drawn as in Fig. 4.4. Solid lines indicate the median rate of evolution in each time bin while dashed lines in a) indicate the 2.5% to 95% quantile range of rates estimated from 10 000 simulated data sets.

261 Table 4.1. Results of Bayesian model selection analysis

Among- Rate Number of Mean Marginal Topology Character Distribution Partitioning Bayes Factor1 Parameters Model Likelihood Rates Shape Prior HLMCC Equal - - 287 -82716.57 16240.7 HLMCC Gamma Uniform - 288 -74704.08 215.72 HLMCC Gamma Exponential - 288 -74697.62 202.8 HLMCC Lognormal Uniform - 288 -74669.33 146.22 HLMCC Lognormal Exponential - 288 -74667.68 142.92

262 Anatomical, Linked HLMCC Lognormal Uniform 298 -74613.94 35.44 Branch Lengths

Anatomical, Linked HLMCC Lognormal Exponential 298 -74596.22 * Branch Lengths Anatomical, HLMCC Lognormal Uniform Unlinked Branch 1440 -75072.88 953.32 Lengths Anatomical, HLMCC Lognormal Exponential Unlinked Branch 1440 -75058.91 925.38 Lengths 1Bayes Factors were calculated as twice the difference in marginal model likelihoods between the best model (*) and all other candidate models Table 4.1. Continued. Among- Rate Number of Mean Marginal Topology Character Distribution Partitioning Bayes Factor Parameters Model Likelihood Rates Shape Prior JETZMCC Equal - - 287 -82465.83 16454.3 JETZMCC Gamma Uniform - 288 -74335.51 193.66 JETZMCC Gamma Exponential - 288 -74332.61 187.86 JETZMCC Lognormal Uniform - 288 -74308.64 139.92 JETZMCC Lognormal Exponential - 288 -74306.22 135.08 Anatomical, Linked JETZMCC Lognormal Uniform 298 -74255.01 32.66 Branch Lengths 263 Anatomical, Linked JETZMCC Lognormal Expoential 298 -74238.68 * Branch Lengths Anatomical, JETZMCC Lognormal Expoential Unlinked Branch 1440 -74671.01 864.66 Lengths Anatomical, JETZMCC Lognormal Expoential Unlinked Branch 1440 -74655.48 833.6 Lengths Table 4.2. Definition and evolutionary rates of focal clades Harrison and Larsson Jetz et al. (2012) Median Clade Median Clade Mean Mean Number Evolution Rate Evolution Rate Clade Clade of and 95% and 95% Age (Ma Quantiles Age (Ma Quantiles Genera 1 ago) (log )2 ago) (log ) Clade Name Genera 10 10 -2.72 -2.86 Accipitridae Accipiter, Gyps, Pandion,Sagittarius 4 54.5 [-2.81,-2.64] 62.8 [-2.93, -2.77] -3.06 -3.01 Anseriformes Anas, Anser, Anseranas, Chauna 4 76.8 [-3.11, -3.03] 71.1 [-3.08, -2.97] -3.20 -3.21

264 Apodiformes Apus, Cypseloides, Glaucis, Hemiprocne 4 61.9 [-3.24, -3.16] 62.3 [-3.43, 3.18] Ardea, Chochlearius, Botaurus, -3.09 -3.35 Ardeidae Nyctocorax, Tigrisoma 5 30.6 [-3.22, -2.94] 55.6 [-3.43, -3.28] Chionis, Burhinus, Chlidonias, Cladorhynchus, Cursorius, Dromas, Glareola, Haematopus, Himantopus Ibidorhyncha, Jacana, Pedionomus, Phalaropus, Pluvialis, Rissa, Rostratula, Rynchops, Stercorarius, Thinocorus, -2.81 -2.79 Charadriiformes Tringa, Turnix, Uria 22 65.2 [-2.85, -2.76] 65.3 [-2.88, 2.71] 1Clade age does not necessarily correspond to age of total crown group of these clades because Livezey and Zusi (2007) did not always include a representative of the most basal divergence (e.g. No Strigopoidea in Psittaciformes: see Chapter 2, this volume: Fig. 3.2) 2This measure is the median of 10 000 calculations of clade mean rate, one per chronogram (see text) and the 95% (2.5% to 97.5%) quantile range of those Table 4.2. Continued. Harrison and Larsson Jetz et al. (2012) Median Clade Median Clade Mean Mean Number Evolution Rate Evolution Rate Clade Clade of and 95% and 95% Age (Ma Quantiles Age (Ma Quantiles Genera 1 ago) (log )2 ago) (log ) Clade Name Genera 10 10 Columba, Didunculus, Ducula, -2.98 -3.05 Columbiformes Geopelia,Goura, Zenaida, 6 31.7 [-3.08, -2.88] 33.3 [-3.14, -2.96] Alcedo, Brachypteracias, Coracias, -3.10 -3.10 “Coraciiformes” Merops, Momotus, Todiramphus, Todus 7 61.7 [-3.17, -3.06] 61.7 [-3.17, 3.02] Centropus, Coccyzus, Crotophaga, -3.19 -3.29

265 Cuculiformes Cuculus, Geococcyx 5 50.1 [-3.26, -3.11] 67.4 [-3.36, -3.22] Alectoris, Callipepla, Numida, Dendragapus, Francolinus, Gallus, -3.24 -3.12 Galliformes Megapodius, Meleagris, Ortalis 9 72.1 [-3.29, -3.14] 60.2 [-3.22, -2.95] Aramus, Grus, Heliornis, Porphyrio, -2.87 -2.82 “Gruiformes” Psophia 5 61.4 [-2.94, -2.80] 51.7 [-2.92, -2.70]

Aphelocoma, Bombycilla, Menura, Parus, -3.26 -3.31 Passeriformes Passer, Pitangus, Pitta, Ptilonorhynchus 8 64.3 [-3.31, -3.21] 67.0 [-3.36, -3.26] Table 4.2. Continued. Harrison and Larsson Jetz et al. (2012) Median Clade Median Clade Mean Mean Number Evolution Rate Evolution Rate Clade Clade of and 95% and 95% Age (Ma Quantiles Age (Ma Quantiles Genera 1 ago) (log )2 ago) (log ) Clade Name Genera 10 10

Aulacorhynchus, Galbula, Indicator, Jynx, Megalaima, Monasa, Picoides, -3.05 -2.97 Piciformes Ramphastos, Semnornis 9 63.0 [-3.10, -3.00] 60.8 [-3.06, -2.87]

266 Oceanites, Pachyptila, Pelecanoides, -2.88 -3.01 Procellariiformes Phoebastria, Puffinus 5 48.1 [-2.98, -2.77] 55.7 [-3.08, -2.94] Amazona, Cacatua, Melopsittacus, -3.40 -3.40 Psittaciformes Trichoglossus 4 43.6 [-3.48, -3.31] 45.1 [-3.48, -3.25] -3.23 -3.33 Strigiformes Otus, Phodilus, Strix, Tyto 4 58.4 [-3.31, -3.14] 64.4 [-3.48,-3.24] -2.67 -2.68 “Suliformes” Anhinga, Phalacrocorax, Fregata, Sula 4 56.5 [-2.73, -2.62] 58.8 [-2.76, -2.62] Table 4.3. Kendall's correlation tests for decreasing rates of evolution through time Harrison and Larsson Tree Distribution MCC Chronogram 10 000 Chronogram Distribution Tips Excluded Tips Included Tips Excluded Tips Included Kendall's Kendall's 2.5% and 97.5% 2.5% and 97.5% Rate p (τ < 0) p (τ < 0) Mean τ Mean τ Optimization τ τ Quantiles Quantiles RAxML -0.184 5.97E-04 -0.185 1.53E-06 -0.181 [-0.233, -0.127] -0.185 [-0.225, -0.145] MrBayes -0.260 2.40E-06 -0.265 1.29E-11 - - - - ACCTRAN -0.461 < 2.2E-16 -0.436 < 2.2E-16 -0.415 [-0.468, -0.357] -0.412 [-0.449, -0.375] DELTRAN -0.386 5.70E-12 -0.393 < 2.2E-16 -0.354 [-0.414, -0.291] -0.368 [-0.411, -0.324] 267

Jetz et al. (2012) Tree Distribution MCC Chronogram 10 000 Chronogram Distribution Tips Excluded Tips Included Tips Excluded Tips Included Kendall's Kendall's 2.5% and 97.5% 2.5% and 97.5% p p Mean τ Mean τ Rate τ τ Quantiles Quantiles Optimization RAxML -0.014 4.04E-01 -0.150 8.20E-05 -0.034 [-0.095, 0.0238] -0.148 [-0.196, -0.097] MrBayes -0.130 1.21E-02 -0.271 4.18E-12 - - - - ACCTRAN -0.351 3.40E-10 -0.447 < 2.2E-16 -0.322 [-0.381, -0.263] -0.414 [-0.459, -0.365] DELTRAN -0.309 2.66E-08 -0.403 < 2.2E-16 -0.271 [-0.329, -0.211] -0.369 [-0.418, -0.317] CHAPTER 4: SUPPLEMENTARY METHODOLOGY

Chronogram Distributions: Jetz et al., (2012) The chronogram distribution generated by Jetz et al. (2012) was downloaded from the www.birdtree.org website created by those authors. The complete 9993 species distribution of 10 000 chronograms, including taxa without molecular sequence data based on the Hackett et al. (2008) phylogenetic constraints, was retrieved (“Hackett Stage 2 Trees”; Jetz et al., 2012). Taxon names were harmonized between the Jetz et al. and LZ2007 data sets by assigning the genera in Livezey and Zusi (2007) to species based on the primary species considered (in Livezey and Zusi, 2006: Appendix 1). The APE R package v3 (Paradis, 2004) was used to process Jetz et al's (2012) distribution of chronograms to remove taxa not present in LZ2007 data set while maintaining ultrametricity (R script available from the authors). The chronogram distribution was summarized by using treenannotator from the BEAST v1.7.5 package (Drummond et al., 2012) to first calculate the maximum clade credibility (MCC) tree of the chronogram distribution, placing node heights at the respective means of the marginal posterior distributions of node ages. However, this led to a negative branch length and to avoid arbitrarily correcting this, the MCC topology was calculated from the full 9993 species chronogram distribution and then pruned to the LZ2007 taxonomic sampling using APE. Treeannotater was then used to annotate this topology with mean node heights, clade posterior probabilities and 95% highest posterior density (HPD) intervals for node ages based on the pruned chronogram distribution, which yielded no negative branch lengths. The MCC tree is included as Supplementary Figure 4.1 and is available from the authors upon request.

Chronogram Distributions: Harrison and Larsson (Chapter 3, this volume) The chronogram distribution of Harrison and Larsson (Chapter 3, this volume) was explicitly constructed to provide generic-level coverage of LZ2007.

268 APE was used to remove the genera not present in LZ2007 while maintaining ultrametricity. As with the Jetz et al. (2012) chronogram distribution, the tree distribution was then summarized using treeannotator, placing node heights at the mean of their respective marginal posterior distributions of node ages. The MCC tree is included as Figure 4.1.

Modifications and use of RAxML to Estimate Branch Length using Discrete Morphological Characters. RAxML v.7.2.8 (Stamaskis, 2006) is a fast maximum-likelihood phylogeny estimation software package capable of employing Lewis (2001)'s Mk model of evolution for discrete morphological characters and has been used to estimate phylogeny from discrete morphological characters (e.g. Lee and Worthy, 2012; Syme and Oakley, 2012; O'Leary et al., 2013). RAxML was modified to implement the discrete lognormal distribution model for among-character rate variation following the methods of Yang (1994). Recent evidence suggests that the discrete lognormal distribution is sometimes more appropriate to model among- character rate heterogeneity relative to the discrete gamma distribution (Wagner, 2012; Harrison and Larsson, Chapter 2, this volume). The discrete lognormal distribution was implemented using median discretization after Yang (1994) with

2 equation 1 where ri is the category rate, σ is the shape parameter of the lognormal distribution and Φ-1 is the standard normal quantile function.

μ+σΦ−1( ) p − 2 ( ) = e K = 2i 1 μ=−( σ ) Eq.1 r i μ +σΦ−1( ) , p , ∑ e p 2 K 2 i RAxML is hardcoded to use four equiprobable discrete rate categories when a discrete distribution of rates across characters is used. This was maintained as it is difficult to modify, although Harrison and Larsson (Chapter 2, this volume) noted that more rate categories led to higher likelihoods with the lognormal distribution in some data sets. To estimate morphological branch lengths on a given phylogenetic topology,

269 RAxML was executed using the -f e option and by partitioning morphological characters by state space (e.g. all binary characters, all three state characters, etc.) which instructed RAxML to assign each partition the corresponding Mk model for that state space (RAxML -q option). If RAxML is provided with a concatenated matrix of discrete morphological characters, it will model each character under an Mk model with state space equal to the highest observed state across all characters. The source code of RAxML was further modified to link the among- character rate heterogeneity distribution shape parameter under either the discrete gamma or lognormal among-character rate distribution models across partitions, rather than estimating an independent shape parameter per partition. The modified source code for RAxML is available from the authors upon request. RAxML does not implement the Lewis (2001) MkV model, which corrects the likelihood calculation for the fact that constant characters are not included in discrete morphological character matrices (Lewis, 2001; see also Allman et al., 2010). However, this did not appear to significantly effect estimates of morphological branch lengths (see below) in this study, probably because across this deep phylogeny, the probability of a character remaining constant across the tree is likely to be low, which led to a small correction. To verify that the use of the Mk rather than the MkV model did not bias this study, RAxML was modified to estimate equal rates among-characters and was then used to reconstruct branch lengths using the Livezey and Zusi (2007) data set on the two MCC trees (Jetz et al. and Harrison and Larsson). The GARLI maximum likelihood package (Zwickl, 2006), which implements the MkV model but cannot model rate heterogeneity across characters in DMC data sets, was then used to estimate branch lengths on the same topologies. The resulting ingroup (rooting differences meant the basal-most branches differed) branch lengths were very similar (Suppl. Fig. 4.2; Spearman's r >0.99 on the Harrison and Larsson MCC topology; Spearman's r > 0.99 on the Jetz et al. MCC topology) and suggested that the use of RAxML on this data set was appropriate, although caution is urged

270 in reconstructing morphological branch lengths on data sets with fewer taxa where not correcting for the lack of constant characters may bias branch lengths (Lewis, 2001). Because rate heterogeneity was found to be present in this data set (see Table 4.1, main text), this is likely a more important factor than not accounting for the ascertainment correction: therefore RAxML was used for all analyses rather than GARLI, although a maximum likelihood implementation with both the MkV model and among-character rate heterogeneity models would have been preferable.

Modifications to MrBayes to Estimate Branch Lengths of Discrete Morphological Characters The latest version of MrBayes v3.2.1 (Ronquist et al., 2012b) was modified to implement the lognormal distribution of among-character rates. Equation 1 was used to implement quantile discretization; although the number of rate categories can be specified in the range of {2,20}, for all analyses described below, four rate categories were used to both increase computational tractability and to be readily comparable to the RAxML analyses. There have been recent concerns about the validity of branch length estimates from MrBayes on some data sets (Marshall, 2010; Rannala et al., 2012; Zhang et al., 2012). These studies suggested that MrBayes may sometimes overestimate branch lengths, therefore here, rates of evolution, because of the influence of the default independent exponential branch length priors (Rannala et al., 2012; Zhang et al. 2012). An attempt was made to use the modified version of MrBayes provided by Zhang et al. (2012) with a compound gamma-dirichlet branch length prior that addressed this issue, but this led to highly variable marginal model likelihoods between independent runs using stepping-stone sampling under both equal and gamma-distributed rates. Therefore, the standard version of MrBayes was used, modified with lognormally distributed rates. The modified source code of MrBayes is available from the authors upon request.

271 Bayesian Model Testing Models of Among-Character Rate Heterogeneity and Anatomical Partition Schemes The equal rates among characters and the discrete gamma and lognormal models of among-character rate heterogeneity were tested using MrBayes (see Table 4.1). The stepping-stone sampling marginal model likelihood estimator implemented in MrBayes (Xie et al., 2011; Ronquist et al., 2012b) was used and the topology was fixed to the MCC trees for each tree distribution. The marginal model likelihood was estimated under the equal, discrete gamma and discrete lognormal among-character rate models using two independent runs of four Markov chains (three heated, one cold) of 10 million generations each. The model tests were repeated using a uniform [0.0001–200] or exponential (mean = 1.0) prior on the distribution shape parameters (see Harrison and Larsson, Chapter 2, this volume). All other priors were left at MrBayes default values. The default sampling scheme using 50 steps, sampling from the posterior towards the prior, drawing δ values from a beta distribution with α = 0.4 was applied. A burnin fraction of 1 million generations from the posterior was discarded and burnin fractions of 25% were applied at each sampling step. All MrBayes analyses were conducted using the CLUMEQ/Colosse and CLUMEQ/Guillimin high performance computing (HPC) facilities using the non-MPI version of MrBayes. Bayes factors were used to compare alternate models and were calculated and interpreted following guidelines in Nylander et al. (2004). In all cases, the lognormal distribution of among-character rate heterogeneity was supported (see Table 1) and was used for all subsequent partitioning analyses. To determine if anatomical partition of the data led to better model fit, as observed in a previous study (Clarke and Middleton, 2008), Bayesian model testing was conducted (see character partitions in Suppl. Table 4.1). Partition testing was conducted using the lognormal model of among-character rate heterogeneity and on the maximum clade credibility trees from each distribution

272 of chronograms. MrBayes was used to estimate marginal model likelihoods under two partitioned models using stepping-stone sampling. The first model estimated separate lognormal shape parameters and per-partition rate multipliers for all five partitions: this allowed rates of evolution to vary proportionally to one set of branch lengths. The second partitioned model estimated separate per-partition lognormal shape parameters and separate branch lengths for each partition which allows rates to vary non-proportionally (see Clarke and Middleton, 2008 for a detailed discussion). Stepping-stone sampling was conducted as described above for these models using the CLUMEQ/Guillimin computing cluster.

MrBayes analysis of Branch Lengths on the MCC Topologies MrBayes was used to estimate branch lengths on each MCC topology (Harrison and Larsson, Chapter 3, this volume; Jetz et al., 2012) under the model of evolution with the strongest support: five anatomical partitions with linked branch lengths and five independently estimated lognormal distribution shape parameters with exponential priors of mean 1.0 (Table 4.1). Branch length estimates were obtained by performing a MrBayes MCMC analysis on the fixed MCC topology using four independent runs of four Markov chains each (three heated, one cold) for 10 million generations each, sampling every 1 000 generations. Convergence of branch length estimates from each independent run was assessed using MrBayes' convergence diagnostics reported by the sump command after discarding a burnin fraction of 25% of samples. All branch length estimates had Potential Scale Reduction Factors (PSRF; Gelman and Rubin, 1992) of 1.0 and all model parameters had effective sample sizes (ESS > 200; Drummond et al., 2006) indicating convergence to and sufficient sampling of the model's posterior.

Stochastic Character Simulations and the Lloyd et al. (2012) “Branch Randomization” Procedure

273 The stochastic character simulation approach described by Roelants et al. (2011) was also used to generate simulated data sets under the assumption of equal rates of evolution through time. For a single chronogram, the rTraitDisc function in the R package APE (Paradis, 2004; R Core Team, 2012) was used to simulate the evolution of a data set of characters. The simulation used the Mk (equal rates) model, and the overall rate of evolution was parameterized by dividing the total tree length estimated by the RAxML-lognormal for LZ2007 on that chronogram by the total time duration of the chronogram. Characters were simulated separately by state space (e.g. 2 state, 3 states, etc.) to match the empirical distribution of states in LZ2007. Characters were split evenly and simulated using four discrete rate categories drawn from the discretized lognormal distribution (Eq. 1), which was parameterized using the estimated lognormal shape parameter (σ2 = 1.7009) from the RAxML analysis of LZ2007 on the Harrison and Larsson MCC topology. In cases where the number of characters per state space was less than 16 in LZ2007, 16 simulated characters were generated (four per rate class) and then randomly sampled without replacement to match the number of characters in that state space for LZ2007. Throughout, characters where simulated until the number of variable characters (by state space) matched the empirical distribution of the Livezey and Zusi (2007) data set. In this manner, no constant characters were included in the simulated data sets (following Lewis, 2001). These simulations were computationally intensive and only a single simulated data set was generated for each of the 10 000 chronograms in each chronogram distribution (total 20 000 simulated data sets, > 1 core-month of computation). A further set of 1000 simulated data sets was generated based on the topology and branch lengths of each of the MCC chronograms to specifically test branch rates on those trees. Null branch rate distributions were then estimated by repeating the rates analyses described in the main text using the corresponding simulated data set instead of the real data for each chrornogram. To control for total tree length differences from the simulations, following Roelants et al. (2011)

274 who noted the same issue, branch lengths were standardized: here, the estimated branch lengths from the simulated data sets were multiplied by a single constant factor such that the total estimated tree length matched that estimated from LZ2007 on the corresponding chronogram and estimation method. The simulated data sets and R scripts used to run the simulations are available from the authors. To compare with the stochastic simulations on the parsimony-based methods, the “branch randomization” approach of Lloyd et al. (2012) was also used. On a single chronogram, the simulation proceeds by linearizing all branch durations end-to-end and then standardizing total length to a [0–1] interval. Then, random numbers on that interval are drawn from the uniform distribution up to the total observed number of character changes in the data set (as determined by maximum parsimony tree length estimated from LZ2007 on that chronogram). The branch duration endpoints were then used to bin these random numbers and assign them to a branch as a character change. This procedure was then repeated once for each of the 10 000 chronograms to derive a null distribution of character changes that were then divided by the corresponding branch durations to yield a null distribution of rate estimates. This simulation was thus independent of the reconstruction methods except for the total number of changes, unlike the stochastic simulations which produced simulated data sets rather than changes directly. These calculations were performed in R using the APE package (Paradis, 2004) and scripts are available from the authors.

275 BR K-Pg BR Apodiformes BR

BR Columbiformes

BR "Suliformes" BR

Ardeidae

Procellariformes BR

Cuculiformes

"Gruiformes"

Accipitridae BR

Piciformes

"Coraciiformes"

BR

Strigiformes

Passeriformes BR

Psittaciformes

BR

Charadriiformes

Anseriformes

Galliformes

Cretaceous Paleogene Neogene Early Late Pal. Eocene Oligo. Miocene P.

120 100 80 60 40 20 0 MYa

276 Supplementary Figure 4.1. Maximum clade credibility chronogram summarizing the Jetz et al. (2012) chronogram distribution, pruned to the sampling of the Livezey and Zusi (2007) data set. Nodes are placed at the means of the marginal posterior distribution on node height. Clade posterior probabilities are indicated by coloured dots: red PP < 0.95; yellow: 0.95 ≤ PP < 0.99; and green: PP ≥ 0.99. Blue bars represent the 95% credibility interval for node ages. Boxes labeled “BR” indicate branches included in the operational definition of the basal neoavian radiation on this topology (see text). Clade names indicated by bars and text on the right margin are used operationally to identify monophyletic clades in this analysis (see also Table 4.2 for assignments of genera to clades). These generally correspond to group names used by Hackett et al. (2008). The solid red line indicates the location of the Cretaceous–Paleogene (K–Pg) boundary. Geological time scale follows Gradstein et al. (2012).

277 Harrison and Larsson MCC Topology Jetz et al. (2012) MCC Topology GARLI Branch Length (Equal-Rates MkV Model) GARLI Branch Length (Equal-Rates MkV Model)

RAxML Branch Length (Equal-Rates Mk Model) RAxML Branch Length (Equal-Rates Mk Model)

278 Supplementary Figure 4.2. Comparison of morphological branch lengths estimated by RAxML, modified to implement an equal-rates among sites model and branch lengths estimated by GARLI, an alternative software package for maximum-likelihood calculations that only implements equal rates among characters. Branch lengths were compared as estimated on the Harrison and Larsson maximum MCC topology (left; Chapter 3, this volume) and the Jetz et al. (2012) MCC topology (right). RAxML implements the Mk model while GARLI implements the ascertainment-corrected MkV model. Only ingroup branches were compared (see Supplementary Methodology). For this data set, there does not appear to be a strong effect of the ascertainment correction on ingroup branch length estimates: Spearman's r > 0.99 for both Harrison and Larsson and Jetz et al. (2012) chronogram distributions.

279 a) b) Neoaves Columbiformes MrBayes MrBayes HL HL Changes/Character Changes/Character

c) Time Before Present [Ma] d) Time Before Present [Ma]

Charadriiformes Passeriformes MrBayes MrBayes HL HL Changes/Character Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

280 Supplementary Figure 4.3. Plots relating morphological branch lengths estimated by MrBayes to branch durations across four focal clades for the Harrison and Larsson MCC chronogram (Chapter 3, this volume; left). The paleognath outgroup is not figured. All root to tip paths are plotted using cumulative morphological branch lengths estimated using the mean branch lengths from the posterior distribution of the Bayesian analysis using five anatomical partitions with relative rates and five separately estimated lognormal distribution shape parameters. Branch slopes represent branch rates. The vertical red line indicates the Cretaceous–Paleogene (K–Pg) boundary. Note the similarity between these plots and Fig. 4.2 for the plot using morphological branch lengths estimated by RAxML.

281 Neognathae Accipitridae RAxML RAxML HL HL Changes/Character

Anseriformes Apodiformes RAxML RAxML HL HL Changes/Character

Ardeidae "Coraciiformes" RAxML RAxML HL HL Changes/Character

Cuculiformes Galliformes RAxML RAxML HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

282 Changes/Character Changes/Character Changes/Character HL RAxML Procellariformes HL RAxML "Gruiformes" HL RAxML Strigiformes Time Before Present [Ma] 283 HL RAxML Psittaciformes HL RAxML Piciformes HL RAxML "Suliformes" Time Before Present [Ma] Supplementary Figure 4.4. (Two Pages). Plots relating morphological branch lengths to branch duration across all focal clades for the Harrison and Larsson (Chapter 3, this volume) maximum clade credibility chronogram, not including those in Figure 4.2. The paleognath outgroup is not figured. Clade names and genera assigned to each are listed in Table 4.2 and figured in Figure 4.1. All root to tip paths are plotted using cumulative morphological branch lengths estimated using RAxML and the lognormal distribution of among-character rates (see text). Branch slope represents branch rates and those branches coloured in red exceeded the rates observed for the corresponding branch in 950/1000 stochastically simulated data sets under an equal-rates model (see text). The vertical red line indicates the Cretaceous–Paleogene (K–Pg) boundary. Note the overall pattern of higher, but also low rate branches at the base of Neoaves and high rates along many, but not all stem lineages to the clades figured in each subplot. Some clades, e.g. “Suliformes”, have significantly high rates of evolution within the group while others e.g. Psittaciformes, have lower rates of evolution on branches within the clade.

284 Neognathae Accipitridae RAxML RAxML JZ2012 JZ2012 Changes/Character

Anseriformes Apodiformes RAxML RAxML JZ2012 JZ2012 Changes/Character

Ardeidae "Coraciiformes" RAxML RAxML JZ2012 JZ2012 Changes/Character

Cuculiformes Galliformes RAxML RAxML JZ2012 JZ2012 Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

285 Changes/Character Changes/Character Changes/Character JZ2012 RAxML Procellariformes JZ2012 RAxML "Gruiformes" JZ2012 RAxML Strigiformes Time Before Present [Ma] 286 JZ2012 RAxML Psittaciformes JZ2012 RAxML Piciformes JZ2012 RAxML "Suliformes" Time Before Present [Ma] Supplementary Figure 4.5. (Two Pages). Plots relating morphological branch lengths estimated by RAxML to branch durations across all focal clades for the Jetz et al. (2012) maximum clade credibility chronogram, not including those in Figure 4.2. Please refer to Supplementary Figure 4.4 for a full description of these plots. Compared with plots created using the Harrison and Larsson MCC tree, recovered patterns very similar. The greatest differences are concentrated in the basal arrangement of some clades (e.g. see the branches leading to Cuculiformes; see also Fig. 4.1 vs. Supp. Fig. 4.1 for the relative positions of this clade) and the older age of the most common ancestors of some clades (e.g. Ardeidae), which leads to lower estimated rates of evolution within the group (see also text for more detailed discussion).

287 Neognathae Neoaves ACCTRAN ACCTRAN HL HL Changes/Character

Accipitridae Anseriformes ACCTRAN ACCTRAN HL HL Changes/Character

Apodiformes Ardeidae ACCTRAN ACCTRAN HL HL Changes/Character

Charadriiformes Columbiformes ACCTRAN ACCTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

288 "Coraciiformes" Cuculiformes ACCTRAN ACCTRAN HL HL Changes/Character

Galliformes "Gruiformes" ACCTRAN ACCTRAN HL HL Changes/Character

Passeriformes Piciformes ACCTRAN ACCTRAN HL HL Changes/Character

Procellariformes Psittaciformes ACCTRAN ACCTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

289 Strigiformes "Suliformes" ACCTRAN ACCTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

290 Supplementary Figure 4.6. (Three Pages). Plots relating morphological branch lengths to branch duration across all focal clades for the Harrison and Larsson (Chapter 3, this volume) maximum clade credibility chronogram. All root to tip paths are plotted using cumulative morphological changes per character estimated using maximum parsimony under ACCTRAN optimization of ambiguous character changes. The paleognath outgroup is not figured. Clade names and genera assigned to each are listed in Table 4.2. Relative to the RAxML analysis for the Harrison and Larsson MCC tree (Fig. 4.2, Suppl. Fig. 4.4), ACCTRAN reconstructs more evolutionary change on the basal branches of the phylogeny and fewer changes on the lineages towards the tips. This is the expected behavior of the ACCTRAN as it will optimize ambiguous changes at the earliest possible point in the phylogeny. Note how under ACCTRAN optimization, many clades have very low rates of evolution.

291 Neognathae Neoaves DELTRAN DELTRAN HL HL Changes/Character

Accipitridae Anseriformes DELTRAN DELTRAN HL HL Changes/Character

Apodiformes Ardeidae DELTRAN DELTRAN HL HL Changes/Character

Charadriiformes Columbiformes DELTRAN DELTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

292 "Coraciiformes" Cuculiformes DELTRAN DELTRAN HL HL Changes/Character

Galliformes "Gruiformes" DELTRAN DELTRAN HL HL Changes/Character

Passeriformes Piciformes DELTRAN DELTRAN HL HL Changes/Character

Procellariformes Psittaciformes DELTRAN DELTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

293 Strigiformes "Suliformes" DELTRAN DELTRAN HL HL Changes/Character

Time Before Present [Ma] Time Before Present [Ma]

294 Supplementary Figure 4.7. (Three Pages). Plots relating morphological branch lengths to branch durations across all focal clades for the Harrison and Larsson (Chapter 3, this volume) maximum clade credibility chronogram. All root to tip paths are plotted using cumulative morphological changes per character estimated using maximum parsimony under DELTRAN optimization of ambiguous character changes. The paleognath outgroup is not figured. Clade names and genera assigned to each are listed in Table 4.2. DELTRAN optimization of character changes is more similar to the RAxML analysis for the Harrison and Larsson MCC tree (Fig. 4.2, Suppl. Fig. 4.4) compared to the ACCTRAN optimizations (Suppl. Fig. 4.6). However, relative to both ACCTRAN and RAxML, DELTRAN places more character changes towards the tips of the phylogeny.

295 296 Supplementary Figure 4.8. Heatmap visualization of the complex distribution of branch rates through time across 10 000 chronograms using maximum parsimony estimation of branch lengths under the a) ACCTRAN and b) DELTRAN optimizations of ambiguous character changes. Heatmaps were constructed and drawn as in Fig. 4.4. The solid lines indicate the median rate in each time bin. Dashed lines indicate the 2.5% to 97.5% quantile range of estimate rates of evolution of the 10 000 stochastically simulated data sets under a constant rates model of evolution estimated using the corresponding character change optimization. The dotted lines represent rates estimated using the “branch randomization” approach of Lloyd et al. (2012) to simulate character evolution under a constant rate, uniform process (see Supplementary Methodology). The DELTRAN data set had rates of 0 in some chronograms and these were arbitrary set to the lowest rate bin (-8 log10(rate)). The overall pattern is generally congruent with the RAxML analysis (Fig. 4.4) in that rates are highest in the Late Cretaceous and across the K–Pg and lower towards the present but there is no cluster of very low rates in the Late Cretaceous. Parsimony-based estimates of rates of evolution of the stochastically simulated data sets (dashed lines) are not constant and appear to have a similar pattern as the LZ2007 data. See text for a discussion of a possible explanation for this pattern, related to rate heterogeneity among characters.

297 Appendix 4.1: Anatomical partitioning of the Livezey and Zusi (2006; 2007) data set

**All Character numbers follow Livezey and Zusi (2006)

MY: All histological, general myological and visceral characters

2,4,5,2708,2709,2710,2711,2712,2713,2714,2715,2716,2717,2718,2720,2721,272 2,2723,2724,2725,2726,2727,2728,2729,2730,2731,2732,2733,2734,2735,2736,2 737,2738,2739,2740,2741,2742,2743,2744,2745,2746,2747,2749,2750,2751,2752 ,2753,2754,2755,2756,2757,2758,2759,2760,2762,2763,2764,2765,2766,2767,27 68,2769,2770,2771,2772,2773,2774,2775,2776,2777,2778,2779,2780,2781,2782, 2783,2784,2785,2786,2787,2788,2789,2790,2791,2792,2793,2794,2795,2796,279 7,2798,2799,2800,2801,2802,2805,2806,2807,2808,2809,2810,2811,2812,2813,2 815,2816,2817,2818,2819,2820,2821,2822,2823,2824,2825,2827,2828,2829,2830 ,2831,2832,2833,2834,2835,2836,2837,2838,2839,2840,2841,2842,2843,2844,28 45,2846,2847,2848,2849,2850,2851,2852,2853,2854,2855,2856,2857,2858,2859, 2860,2861,2862,2863,2864,2865,2866,2867,2868,2869,2870,2871,2872,2873,287 4,2875,2876,2877,2878,2879,2880,2881,2882,2883,2884,2885,2886,2888,2889,2 890,2891,2892,2893,2894,2895,2896,2897,2898,2899,2900,2901,2902,2903,2904 ,2905,2906,2907,2908,2909,2910,2911,2912,2913,2914,2915,2916,2917,2919,29 20,2921,2922,2923,2925,2926,2929,2930,2931,2932,2933,2934,2935,2941,2942, 2943,2945,2946,2949,2950,2951,2952,2953,2954

AC: All characters associated with the cranium

6,9,13,14,16,19,20,21,23,24,25,26,27,31,34,35,36,38,39,40,41,42,43,44,45,47,48, 49,50,51,52,53,56,57,58,59,60,61,62,70,74,75,76,77,78,79,91,92,93,94,95,96,97,9 8,99,100,101,103,106,107,108,109,112,113,116,117,118,120,121,122,123,124,12 5,126,128,129,130,132,134,143,145,146,147,148,149,150,151,154,155,156,157,1 58,159,160,166,169,170,172,174,175,176,188,189,191,193,194,195,196,197,198,

298 200,201,202,204,205,206,211,213,215,216,219,220,221,223,224,225,228,229,230 ,231,232,238,239,240,241,244,246,247,248,249,250,251,253,254,255,257,258,26 0,262,265,266,268,269,270,271,272,273,275,276,277,278,280,281,282,283,284,2 85,286,287,288,289,290,291,292,293,294,297,298,299,300,301,302,303,304,306, 331,333,335,336,337,340,341,343,344,345,346,347,349,351,352,353,354,355,372 ,374,383,385,386,389,390,391,392,398,405,408,409,410,412,413,414,415,417,42 0,421,422,427,429,430,431,432,433,434,435,436,437,438,439,440,441,442,443,4 44,445,446,447,448,449,450,451,452,453,454,455,456,457,458,459,460,462,463, 464,465,467,468,470,471,473,474,475,476,477,479,480,481,482,483,484,485,499 ,500,504,506,507,508,509,510,511,512,513,514,516,517,518,519,521,523,524,52 5,526,528,529,530,531,532,533,534,535,536,538,539,540,541,542,545,546,547,5 48,549,550,551,552,553,554,555,556,558,561,562,564,566,568,569,570,576,577, 578,579,580,581,582,583,584,593,594,598,599,600,601,602,603,604,605,620,622 ,623,624,625,626,631,632,633,634,638,642,643,645,647,648,649,650,652,653,65 4,655,656,657,658,659,661,662,663,666,667,668,669,673,674,675,676,677,678,6 79,680,681,682,683,684,689,693,695,696,698,700,701,702,703,709,710,711,712, 713,714,715,716,717,718,719,720,721,722,723,724,725,731,733,734,736,737,738 ,739,740,741,742,743,744,745,746,747,748,749,750,751,752,753,754,755,756,75 7,758,759,760,761,762,763,764,765,2452,2453,2454,2455,2456,2457,2458,2459, 2460,2461,2462,2463,2464,2465,2466,2467,2468,2469,2470,2471,2472,2473,247 4,2475,2476,2477,2478,2479,2480,2481,2482,2483,2484,2485,2486,2487,2488,2 489,2490,2491,2492,2493,2494,2495,2496,2497,2498,2499,2500

AV: All characters associated with the vertebral column and ribs

767,769,771,772,774,776,780,782,783,787,788,791,792,794,795,796,798,799,805 ,806,809,816,820,822,823,824,825,826,827,829,830,831,832,836,843,848,850,85 6,857,858,860,861,863,866,867,869,870,871,875,878,879,880,885,886,888,889,8 90,892,893,895,896,897,898,900,901,902,903,904,905,906,907,908,909,910,911, 912,913,914,915,916,917,918,920,921,922,923,924,926,927,928,929,930,932,933 ,934,935,936,937,939,940,941,944,945,946,947,948,949,950,952,953,954,955,95

299 6,957,958,959,960,961,962,963,964,966,967,968,969,970,977,980,983,989,990,9 94,997,998,999,1000,1001,1002,1003,1004,1005,1007,1008,1018,1019,1020,102 1,1022,1023,1024,1025,1026,1027,1028,1029,1030,1031,1032,1034,1035,1037,1 038,1039,1041,1042,1043,1044,1046,1047,1048,1049,1050,1051,1052,1053,1054 ,1055,1056,1057,1059,1060,1062,1063,1064,1065,1066,1067,1070,1071,1072,10 73,1074,1075,1076,1077,1078,1079,1080,1081,1083,1085,1087,1092,1093,1094, 1095,1096,2501,2502,2503,2504

HL: All characters associated with the hindlimbs

1758,1759,1761,1764,1766,1767,1769,1771,1773,1775,1777,1778,1779,1780,178 1,1782,1783,1784,1785,1787,1789,1790,1791,1792,1793,1794,1795,1796,1798,1 799,1800,1801,1802,1803,1804,1805,1806,1807,1808,1809,1810,1812,1813,1814 ,1815,1816,1817,1818,1820,1821,1822,1823,1824,1825,1827,1828,1829,1830,18 31,1832,1837,1841,1842,1843,1844,1845,1851,1852,1853,1854,1855,1857,1858, 1859,1860,1861,1862,1863,1864,1865,1866,1867,1868,1869,1870,1871,1872,187 3,1874,1875,1876,1877,1880,1881,1882,1883,1884,1888,1890,1891,1892,1893,1 894,1895,1896,1898,1899,1901,1903,1904,1905,1906,1907,1909,1913,1914,1915 ,1917,1918,1919,1920,1922,1923,1924,1925,1927,1928,1930,1931,1932,1933,19 37,1938,1940,1945,1946,1947,1948,1949,1950,1952,1953,1956,1957,1958,1959, 1961,1963,1966,1967,1968,1972,1973,1974,1975,1976,1978,1979,1980,1981,199 4,1996,1997,1998,1999,2000,2002,2003,2004,2007,2010,2011,2012,2014,2015,2 017,2019,2022,2023,2024,2026,2027,2028,2029,2030,2032,2033,2034,2035,2036 ,2039,2040,2041,2042,2043,2045,2046,2047,2048,2051,2052,2053,2054,2055,20 56,2059,2060,2061,2062,2063,2064,2065,2066,2067,2068,2069,2071,2072,2073, 2074,2075,2076,2077,2078,2079,2080,2081,2082,2083,2085,2086,2087,2088,208 9,2090,2091,2092,2094,2095,2096,2097,2098,2099,2100,2101,2102,2103,2104,2 105,2106,2107,2108,2109,2110,2111,2112,2113,2114,2115,2117,2118,2119,2120, 2121,2122,2124,2126,2127,2128,2129,2130,2131,2133,2134,2135,2136,2137,213 8,2141,2142,2144,2145,2146,2147,2148,2149,2150,2151,2152,2153,2154,2155,2 156,2157,2158,2159,2161,2162,2163,2165,2166,2167,2168,2169,2170,2171,2172

300 ,2173,2174,2175,2176,2177,2178,2179,2180,2181,2182,2183,2184,2186,2187,21 88,2189,2191,2192,2194,2195,2196,2197,2198,2200,2201,2203,2209,2211,2215, 2216,2217,2220,2221,2224,2225,2229,2230,2231,2232,2233,2235,2236,2246,224 7,2248,2249,2250,2251,2252,2253,2254,2255,2256,2257,2258,2259,2260,2261,2 262,2264,2265,2266,2267,2268,2269,2270,2271,2272,2273,2274,2275,2276,2277 ,2278,2279,2280,2281,2282,2283,2284,2286,2287,2288,2289,2290,2291,2292,22 93,2294,2295,2296,2297,2298,2299,2300,2301,2302,2303,2304,2305,2306,2307, 2308,2309,2310,2312,2313,2314,2315,2316,2317,2318,2319,2320,2321,2322,232 3,2324,2325,2326,2327,2328,2329,2330,2331,2332,2333,2334,2335,2336,2337,2 338,2339,2340,2341,2342,2343,2344,2345,2346,2347,2348,2349,2350,2351,2352 ,2354,2355,2356,2357,2358,2359,2360,2361,2362,2363,2364,2365,2366,2367,23 68,2369,2370,2371,2372,2373,2374,2375,2376,2377,2378,2379,2380,2381,2382, 2383,2385,2386,2388,2389,2390,2391,2392,2394,2395,2396,2397,2398,2399,240 0,2402,2403,2404,2405,2406,2407,2408,2410,2411,2412,2413,2414,2415,2416,2 417,2418,2419,2420,2421,2422,2423,2425,2426,2427,2428,2429,2430,2431,2432 ,2433,2434,2435,2436,2442,2445,2446,2449,2450,2451,2636,2637,2638,2639,26 40,2641,2642,2643,2644,2645,2646,2647,2648,2649,2650,2651,2652,2653,2654, 2655,2656,2657,2658,2659,2660,2661,2662,2663,2664,2665,2666,2667,2668,266 9,2670,2671,2672,2673,2674,2675,2676,2677,2678,2679,2680,2681,2682,2683,2 684,2685,2686,2687,2688,2689,2690,2691,2692,2693,2694,2695,2696,2697,2698 ,2699,2700,2701,2702,2703,2704,2705,2706,2707

FL: All characters associated with the forelimbs

1098,1099,1100,1101,1102,1103,1104,1105,1106,1107,1108,1109,1110,1111,1112 ,1113,1114,1115,1117,1118,1119,1120,1121,1122,1123,1124,1125,1126,1127,1128 ,1129,1131,1133,1134,1135,1136,1137,1138,1139,1140,1141,1142,1143,1144,114 5,1146,1147,1148,1149,1150,1151,1152,1153,1154,1155,1156,1157,1158,1159,11 60,1161,1162,1163,1164,1165,1166,1167,1168,1169,1170,1171,1172,1173,1174,1 175,1176,1177,1178,1179,1180,1182,1183,1184,1185,1186,1187,1188,1189,1190, 1191,1192,1193,1195,1196,1197,1198,1199,1200,1201,1202,1203,1204,1205,120

301 6,1207,1208,1209,1210,1211,1212,1213,1214,1215,1216,1218,1221,1222,1223,1 224,1225,1226,1227,1228,1229,1230,1231,1232,1233,1234,1235,1236,1237,1238 ,1239,1240,1241,1242,1243,1244,1245,1246,1247,1248,1249,1250,1251,1252,12 53,1254,1255,1256,1257,1258,1259,1260,1261,1262,1263,1264,1265,1266,1267, 1268,1270,1271,1272,1273,1274,1275,1276,1277,1280,1281,1282,1283,1284,128 5,1286,1287,1288,1289,1292,1293,1294,1295,1296,1298,1299,1300,1301,1302,1 303,1304,1305,1306,1307,1308,1309,1310,1311,1313,1314,1315,1316,1317,1318 ,1319,1320,1322,1323,1324,1326,1328,1329,1330,1331,1333,1334,1335,1336,13 37,1338,1339,1340,1341,1342,1343,1344,1346,1347,1348,1349,1350,1352,1353, 1354,1355,1356,1357,1358,1359,1360,1361,1362,1363,1364,1365,1366,1367,136 8,1369,1370,1371,1373,1374,1375,1376,1377,1378,1379,1380,1381,1382,1383,1 384,1385,1386,1387,1388,1389,1390,1391,1392,1393,1395,1396,1397,1398,1399 ,1400,1401,1402,1403,1404,1405,1406,1407,1408,1409,1410,1411,1412,1413,14 14,1415,1416,1417,1418,1419,1420,1421,1422,1423,1424,1425,1426,1427,1428, 1429,1430,1431,1432,1433,1434,1436,1437,1439,1440,1441,1442,1443,1444,144 6,1447,1448,1450,1451,1453,1454,1455,1456,1457,1458,1459,1460,1461,1463,1 464,1465,1466,1467,1468,1469,1470,1471,1472,1474,1475,1476,1477,1479,1481 ,1482,1483,1484,1485,1486,1487,1488,1489,1490,1491,1492,1493,1495,1496,14 97,1498,1499,1500,1502,1503,1504,1505,1506,1508,1509,1511,1512,1513,1514, 1515,1516,1517,1519,1521,1522,1523,1524,1525,1526,1527,1528,1529,1530,153 1,1532,1533,1534,1535,1536,1537,1538,1540,1541,1542,1543,1544,1545,1546,1 547,1548,1549,1550,1551,1554,1555,1556,1557,1558,1559,1560,1561,1562,1563 ,1564,1565,1566,1567,1568,1569,1570,1571,1572,1573,1574,1578,1580,1581,15 82,1590,1591,1597,1601,1602,1603,1604,1606,1607,1609,1610,1611,1613,1614, 1619,1621,1622,1623,1624,1625,1626,1627,1628,1629,1630,1632,1633,1634,163 5,1636,1637,1638,1639,1640,1642,1643,1644,1645,1646,1647,1648,1649,1651,1 652,1653,1654,1655,1656,1657,1658,1659,1660,1661,1662,1663,1664,1665,1666 ,1667,1668,1669,1670,1671,1672,1673,1674,1675,1677,1678,1679,1682,1693,16 94,1695,1702,1707,1708,1709,1710,1711,1712,1713,1714,1715,1717,1718,1719, 1720,1721,1722,1723,1724,1725,1726,1729,1730,1731,1733,1734,1735,1736,173 9,1741,1743,1744,1747,1749,1750,1751,1752,1753,1754,1755,1756,2505,2506,2

302 507,2509,2510,2511,2512,2513,2514,2515,2516,2517,2518,2519,2520,2521,2522 ,2523,2524,2525,2526,2527,2528,2529,2530,2531,2532,2533,2534,2535,2536,25 37,2538,2539,2540,2541,2542,2543,2544,2545,2546,2547,2548,2549,2550,2551, 2552,2553,2554,2555,2556,2557,2558,2559,2560,2561,2562,2563,2564,2565,256 6,2567,2568,2569,2570,2571,2572,2573,2574,2575,2576,2577,2578,2579,2580,2 581,2583,2584,2585,2586,2587,2588,2589,2590,2591,2592,2593,2594,2595,2596 ,2597,2598,2599,2600,2601,2602,2603,2604,2605,2606,2607,2608,2609,2610,26 11,2612,2613,2614,2615,2616,2617,2618,2619,2620,2621,2622,2623,2624,2625, 2626,2627,2628,2629,2630,2631,2632,2633,2634,2635

303 SUMMARY AND CONCLUSIONS

The characterization of the tempo and mode of phenotypic evolution is central to testing hypotheses concerning some of the fundamental questions in evolutionary biology (Lloyd et al., 2012). Rates of phenotypic evolution have been investigated using continuously valued traits and characters both in the fossil record (e.g. Haldane, 1949) and using comparative data in extant groups (e.g. Ackerly, 2009; Harmon et al., 2010). Discrete morphological characters remain a relatively underutilized resource for the study of evolutionary rates despite large quantities of data and a long history of use in systematics (Lloyd et al., 2012). This thesis explored the characterization of rates of evolution both between characters in matrices of discrete morphological characters and through time and across clades using modern birds as a case study. In the first chapter, I reviewed these concepts and summarized salient studies that used discrete morphological characters to estimate absolute rates of phenotypic evolution in a phylogenetic context. A common feature of these studies was the use of a fixed or small subset of phylogenetic hypotheses, frequently derived from the same characters used to estimate rates of evolution. A major objective of this thesis was to relax these assumptions and to estimate rates of evolution using discrete morphological characters, integrating across phylogenetic and divergence time uncertainty by using an independently derived distribution of dated phylogenies. In the second chapter of this thesis, I examined the prevalence of rate heterogeneity between discrete morphological characters in phylogenetic analysis by comparing support for models assuming equal and unequal rates of evolution among characters. I used Bayesian model selection techniques to integrate across phylogenetic and other parameter uncertainty and estimate marginal model likelihoods. Rate heterogeneity between characters was estimated to be a common property of the data sets examined, as indicated by strong support for unequal- rates models over equal-rates models in most data sets. I further tested and found some weak support for Wagner's (2012) recent hypothesis that a lognormally-

304 distributed rates model might be a better fit to model among-character rates in matrices of discrete morphological characters relative to a gamma-distributed rates model. However, most data sets were equivocal and the degree of support was sensitive to prior specification. To perform these tests, I implemented the lognormal rates model in the popular Bayesian phylogenetic software package MrBayes. Approximation of the underlying distribution of character rates using maximum parsimony for four data sets suggested that the lognormal distribution may be more appropriate when there are many slowly evolving characters and fewer quickly evolving characters. Although the effect on estimation of topology was difficult to assess across all data sets, it appeared relatively minor between unequal-rates models for the one data set carefully examined. However, with increasing data set sizes, discrimination of these models and exploration of further strategies including character partitioning to account for rate heterogeneity among discrete characters should be possible (e.g. Clarke and Middleton, 2008). This chapter contributes to the growing literature on likelihood-based phylogenetic analysis of discrete morphological characters (e.g. Lewis, 2001; Nylander et al., 2004; Müller and Reisz, 2006; Clarke and Middleton, 2008; Lee and Worthy, 2012) but it is still early in the development of biologically realistic likelihood- based models to capture heterogeneity in evolutionary processes in these data sets. In the third and fourth chapters of this thesis, I used discrete morphological characters to estimate absolute rates of phenotypic evolution. Such estimates require an external time scale to deconfound branch estimates of morphological change into its components of rate and time. Existing methods typically assumed fixed topologies and divergence times or a handful of alternate hypotheses. In this thesis, I extended current methods to estimate rates to integrate across large distributions of dated phylogenies. To illustrate the approach, I adopted the evolution of modern birds as a case study, using the Livezey and Zusi (2006; 2007) data set of discrete morphological characters, one of the largest ever assembled. In the third chapter of this thesis, in order to provide a distribution of dated phylogenies to estimate rates for the Livezey and Zusi (2007) data set, I used

305 Bayesian divergence time dating and a comprehensive sample of 310 modern bird genera. At the outset of this project, such a distribution of dated trees for modern birds did not exist; one has since been published (Jetz et al., 2012) using novel methods to construct a composite distribution of time calibrated trees for all birds from clade-level analyses. However, there are concerns about that analysis (Pagel, 2012), which for example, included ~3000 species for which there was no sequence data available. In the analysis conducted here, a single multigene supermatrix of existing sequence data was analyzed in a single, consistent Bayesian analysis. I combined the molecular sampling of two major phylogenetic analyses of modern birds (Ericson et al., 2006 and Hackett et al., 2008) and augmented these sequences with extensive data from GenBank for a total of 447 species of birds to create a supermatrix which was then collapsed to the generic level. I used a computationally tractable subset of my supermatrix for divergence time dating and applied a relaxed molecular clock dating approach simultaneously sampling phylogenetic topologies in BEAST (Drummond et al., 2012). I used a total of 23 fossil-based soft calibrations and executed this computationally challenging analysis using a supercomputing cluster. This analysis did not resolve basal avian relationships and did not recover precise divergence times. However, qualified by a high proportion of missing data, the resulting distribution of dated phylogenies, in aggregate, constituted a broad range of plausible relationships and timescales for avian evolution. The distribution of dated phylogenies estimated that the basal divergences in Neoaves occurred quickly in the Late Cretaceous, and that many lineages crossed the K–Pg boundary while diversification of crown groups within orders occurred almost exclusively in the Paleogene. This inferred timescale for modern bird evolution is largely congruent with several previous hypotheses of divergence times (e.g. Pacheco et al., 2011, Haddrath and Baker, 2012; Jetz et al., 2012) and disputes an older hypothesis of an exclusively Paleogene radiation of modern birds (e.g. Feduccia, 2003). Constructed expressly to provide taxonomic sampling of the Livezey and Zusi (2007) data set, this supermatrix will facilitate future “total-evidence” analyses of phylogeny and divergence times using newly developed methods to directly include fossils as

306 terminal taxa in divergence time analysis (e.g. Pyron, 2011; Ronquist et al., 2012a; Wood et al., 2013). Coupled with further gene sequencing, fossil discoveries and careful attention to gene tree discordance, these emerging methods may one day reach the goal of a resolved, precisely dated species tree for modern birds. In the fourth chapter of this thesis, I used the distribution of dated phylogenies from the third chapter to estimate rates of evolution of anatomical phenotype across modern birds using the Livezey and Zusi (2007) data set. I also considered the recently published Jetz et al. (2012) tree distribution. I first modified the maximum-likelihood software package RAxML to use the discrete lognormally distributed among-character rate heterogeneity model tested in the second chapter. Using Bayesian model selection techniques, I then determined that this model was supported over the gamma-distributed rates model for the Livezey and Zusi (2007) data set. I also tested and found support for a model partitioning anatomical characters and allowing relative rates to vary between partitions. Building on previous approaches (e.g. Ruta et al., 2006; Brusatte et al., 2008; Roelants et al., 2011) and using maximum parsimony and likelihood-based methods, I then estimated absolute rates of phenotypic evolution by dividing reconstructed morphological evolution on branches of each dated phylogeny in both distributions of trees by the corresponding branch durations. Adapting a visualization technique from a recent molecular clock study, aggregate rates of evolution through time were visualized across all 10 000 chronograms using heatmaps. These revealed a complex and highly variable multimodal distribution of branch rates through time. Estimated rates of evolution were most variable in the Late Cretaceous, coincident with the basal radiation of Neoaves and included both the highest and lowest rates observed. Variability appeared to be associated with the very short temporal duration of branches at the base of Neoaves. Examined in aggregate across both tree distributions and morphological branch length estimation methods, rates of anatomical evolution were generally higher in the Late Cretaceous and early Paleogene and declined towards the present. Examination of cumulative morphological evolution on branches plotted against

307 age using maximum clade credibility trees to summarize each distribution suggested rates were higher along some branches in the basal radiation of Neoaves and on the stem lineages of modern orders. Clade-level estimates of rates across the distributions of dated phylogenies also suggested higher rates at the base of the Neoaves and lower rates among the focal clades examined here. Partially corroborated by a correlation analysis of estimated branch rates and ages, the results were generally consistent with an “early-burst” of phenotypic evolution in the early evolutionary history of modern birds, as predicted by the theory of adaptive radiation. This conclusion was qualified by possible confounding factors including: uneven taxonomic sampling of clades, high branch rate variability and possible underestimation of rates along branches to terminal taxa. Accounting for phylogenetic and divergence time uncertainty led to high variability in rate estimates and suggests that consideration of these factors is important in estimating absolute rates of phenotypic evolution. Finally, I concluded by describing progress towards an ideal fully Bayesian approach to estimating evolutionary rates using discrete morphological characters. Although this thesis provided some methodological progress in the estimation of rates of evolution of discrete morphological characters by integrating across phylogenetic and divergence time uncertainty, there remains great potential for further improvement. Data sets of discrete morphological characters are becoming larger and larger and now exceed thousands of characters (e.g. Livezey and Zusi, 2007; O'Leary et al., 2013). The exploration of more complex phylogenetic modeling approaches for discrete characters is warranted and there is significant potential to develop more biologically informed models of character evolution. These may in turn inform the investigation of phenotypic rates of evolution using discrete characters to address fundamental questions in evolutionary biology. It is my hope that the relatively untapped potential of discrete morphological characters to estimate phenotypic rates of evolution will be harnessed, paleontological and neontological data sets will be merged and, that future methodological progress will mirror developments for continuously-valued traits and characters.

308 BIBLIOGRAPHY

Ackerly D (2009) Conservatism and diversification of plant functional traits: Evolutionary rates versus phylogenetic signal. Proceedings of the National Academy of Sciences of the United States of America 106 Suppl: 19699– 19706.

Adams DC, Berns CM, Kozak KH, Wiens JJ (2009) Are rates of species diversification correlated with rates of morphological evolution? Proceedings of Royal Society of London Series B 276: 2729–2738.

Agnarsson I (2006) Asymmetric female genitalia and other remarkable morphology in a new genus of cobweb spiders (Theridiidae, Araneae) from Madagascar. Biological Journal of the Linnean Society 87: 211–232.

Agnolin F (2007) Brontornis burmeisteri Moreno & Mercerat, un Anseriformes (Aves) gigante del Mioceno Medio de Patagonia, Argentina. Revista del Museo Argentino de Ciencias Naturales 9: 15–25.

Albert JS (2001) Species diversity and phylogenetic systematics of American knifefishes (Gymnotiformes, Teleostei). Miscellaneous Publications, Museum of Zoology, University of Michigan 190: 1–127.

Alfaro ME, Holder MT (2006) The posterior and the prior in Bayesian phylogenetics. Annual Review of Ecology, Evolution, and Systematics 37: 19–42.

Alfaro ME, Santini F, Brock C, Alamillo H, Dornburg A, et al. (2009) Nine exceptional radiations plus high turnover explain species diversity in jawed vertebrates. Proceedings of the National Academy of Sciences of the United States of America 106: 13410–13414.

Allman ES, Holder MT, Rhodes JA (2010) Estimating trees from filtered data: identifiability of models for morphological phylogenetics. Journal of Theoretical Biology 263: 108–119.

309 Almeida VR, de Costa AF, Mantovani A, Gonçalves-Esteves V, de Oliveira Arruda R do C, et al. (2009) Morphological phylogenetics of Quesnelia (Bromeliaceae, Bromelioideae). Systematic Botany 34: 660–672.

Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ (1990) Basic local alignment search tool. Journal of Molecular Biology 215: 403–410.

Alvarenga H (1985) Um novo Psilopteridae (Aves: Gruiformes) dos sedimentos Terciários de Itaboraí, Rio de Janeiro, Brasil. In Congresso Brasileiro de Paleontolgia, 8. Rio de Janeiro, 1983. NME-DNPM Série Geologica 27: 27–20.

Alvarenga H, Chiappe LM, Bertelli S (2011) Phorusrhacids: the terror birds. In: Dyke GJ, Kaiser G, editors. Living Dinosaurs: The Evolutionary History of Modern Birds. Oxford, UK: John Wiley & Sons, Ltd. pp. 187–208.

Alvarenga H, Höfling E (2003) Systematic revision of the Phorusrhacidae. Papéis Avulsos de Zoologia 43: 55–91.

Anderson PSL, Friedman M (2012) Using cladistic characters to predict functional variety: experiments using early gnathostomes. Journal of Vertebrate Paleontology 32: 1254-1270.

Aris-Brosou S, Yang Z (2003) Bayesian models of episodic evolution support a late precambrian explosive diversification of the Metazoa. Molecular Biology and Evolution 20: 1947–1954.

Arnold C, Matthews LJ, Nunn CL (2010) The 10kTrees website: A new online resource for primate phylogeny. Evolutionary Anthropology: Issues, News, and Reviews 19: 114–118.

Asher RJ, Hofreiter M (2006) Tenrec phylogeny and the noninvasive extraction of nuclear DNA. Systematic Biology 55: 181–194.

Ayache NC, Near TJ (2009) The utility of morphological data in resolving phylogenetic relationships of darters as exemplified with Etheostoma (Teleostei : Percidae). Bulletin of the Peabody Museum of Natural History

310 50: 327–346.

Ayres DL, Darling A, Zwickl DJ, Beerli P, Holder MT, et al. (2012) BEAGLE: An application programming interface and high-performance computing library for statistical phylogenetics. Systematic Biology 61: 170–173.

Baele G, Lemey P, Bedford T, Rambaut A, Suchard MA, et al. (2012) Improving the accuracy of demographic and molecular clock model comparison while accommodating phylogenetic uncertainty. Molecular Biology and Evolution 29: 2157–2167.

Baker WJ, Savolainen V, Asmussen-Lange CB, Chase MW, Dransfield J, et al. (2009) Complete generic-level phylogenetic analyses of palms (Arecaceae) with comparisons of supertree and supermatrix approaches. Systematic Biology 58: 240–256.

Banks JC, Paterson AM (2004) A penguin-chewing louse (Insecta : Phthiraptera) phylogeny derived from morphology. Invertebrate Systematics 18: 89–100.

Baumel JJ (1993) Handbook of Avian Anatomy: Nomina Anatomica Avium. 2nd Edition. Cambridge, MA, United States.

Benton MJ, Donoghue PCJ, Asher RJ (2009) Calibrating and constraining molecular clocks. In: Hedges SB, Kumar S, editors. The Timetree of Life. New York: Oxford University Press. pp. 35–86.

Bertelli S, Lindow BEK, Dyke GJ, Chiappe LM (2010) A well-preserved “charadriiform-like” fossil bird from the Early Eocene Fur Formation of Denmark. Palaeontology 53: 507–531.

Beutel RG, Zimmermann D, Krauß M, Randolf S, Wipfler B (2010) Head morphology of Osmylus fulvicephalus (Osmylidae, Neuroptera) and its phylogenetic implications. Organisms Diversity & Evolution 10: 311–329.

Bibi F, Vrba E, Fack F (2012) A new African fossil caprin and a combined molecular and morphological bayesian phylogenetic analysis of caprini (Mammalia: Bovidae). Journal of Evolutionary Biology 25: 1843–1854.

311 Bokma F (2008) Detection of “punctuated equilibrium” by bayesian estimation of speciation and extinction rates, ancestral character states, and rates of anagenetic and cladogenetic evolution on a molecular phylogeny. Evolution 62: 2718–2726.

Bourdon E (2011) The pseudo-toothed birds (Aves, Odontopterygiformes) and their bearing on the early evolution of modern birds. In: Dyke GJ, Kaiser G, editors. Living Dinosaurs: The Evolutionary History of Modern Birds. Oxford, UK: John Wiley & Sons, Ltd. pp. 209–234.

Bourdon E, Bouya B, Iarochene M (2005) Earliest African neornithine bird: A new species of prophaethontidae (Aves) from the Paleocene of Morocco. Journal of Vertebrate Paleontology 25: 157–170.

Brandley MC, Schmitz A, Reeder TW (2005) Partitioned Bayesian analyses, partition choice, and the phylogenetic relationships of scincid lizards. Systematic Biology 54: 373–390.

Braun EL, Kimball RT, Han K-L, Iuhasz-Velez NR, Bonilla AJ, et al. (2011) Homoplastic microinversions and the avian tree of life. BMC Evolutionary Biology 11: 141.

Britto MR (2003) Phylogeny of the subfamily Corydoradinae (Siluriformes: Callichthyidae), with a definition of its genera. Proceedings of the Academy of Natural Sciences of Philadelphia 153: 119–154.

Brocklehurst N, Upchurch P, Mannion PD, O’Connor J (2012) The completeness of the fossil record of Mesozoic birds: Implications for early avian evolution. PLoS ONE 7: e39056.

Bromham L, Woolfit M, Lee MS, Rambaut A (2002) Testing the relationship between morphological and molecular rates of change along phylogenies. Evolution 56: 1921–1930.

Brown JM, Hedtke SM, Lemmon AR, Lemmon EM (2010) When trees grow too long: investigating the causes of highly inaccurate bayesian branch-length

312 estimates. Systematic Biology 59: 145–161.

Brown JM, Lemmon AR (2007) The importance of data partitioning and the utility of Bayes factors in Bayesian phylogenetics. Systematic Biology 56: 643–655.

Brown JW, Payne RB, Mindell DP (2007) Nuclear DNA does not reconcile “rocks” and “clocks” in Neoaves: a comment on Ericson et al. Biology Letters 3: 257–259.

Brown JW, Rest JS, García-Moreno J, Sorenson MD, Mindell DP (2008) Strong mitochondrial DNA support for a Cretaceous origin of modern avian lineages. BMC Biology 6: 6.

Brown JW, van Tuinen M (2011) Evolving perceptions on the antiquity of the modern avian tree. In: Dyke G, Kaiser G, editors. Living Dinosaurs: The Evolutionary History of Modern Birds. Oxford, UK: John Wiley & Sons, Ltd. pp. 306–324.

Brown RP, Yang Z (2011) Rate variation and estimation of divergence times using strict and relaxed clocks. BMC Evolutionary Biology 11: 271.

Brusatte SL (2011) Calculating the tempo of morphological evolution: rates of discrete character change in a phylogenetic context. In: Elewa AMT, editor. Computational Paleontology. Berlin, Heidelberg: Springer Berlin Heidelberg. pp. 53–74.

Brusatte SL, Benton MJ, Ruta M, Lloyd GT (2008) Superiority, competition, and opportunism in the evolutionary radiation of dinosaurs. Science 321: 1485– 1488.

Burbrink FT, Chen X, Myers EA, Brandley MC, Pyron RA (2012) Evidence for determinism in species diversification and contingency in phenotypic evolution during adaptive radiation. Proceedings of Royal Society of London Series B 279: 4817–4826.

Campbell V, Lapointe F-J (2011) Retrieving a mitogenomic mammal tree using

313 composite taxa. Molecular Phylogenetics and Evolution 58: 149–156.

Campbell V, Lapointe F-J (2009) The use and validity of composite taxa in phylogenetic analysis. Systematic Biology 58: 560–572.

Carrano MT, Benson RB, Sampson SD (2012) The phylogeny of Tetanurae (Dinosauria: Theropoda). Journal of Systematic Palaeontology 10: 211– 300.

Castresana J (2000) Selection of conserved blocks from multiple alignments for their use in phylogenetic analysis. Molecular Biology and Evolution 17: 540–552.

Chambers LM, Pringle M, Fitton G, Larsen LM, Pedersen AK et al. (2003) Recalibration of the Palaeocene-Eocene boundary (P-E) using high percision U-Pb and Ar-Ar isotopic dating. Geophysical Research Abstracts 5, EGS-AGU-EUG Joint Assembly, Nice 6th-11th April 2003: 9681-9682.

Chandler R, Parmley D (2002) The earliest North American record of an auk (Aves: Alcidae) from the Late Eocene of central Georgia. Oriole 68: 7–9.

Chang S, Zhang H, Renne PR, Fang Y (2009) High-precision 40Ar/39Ar age for the Jehol Biota. Palaeogeography, Palaeoclimatology, Palaeoecology 280: 94–104.

Chemisquy MA, Giussani LM, Scataglini MA, Kellogg EA, Morrone O (2010) Phylogenetic studies favour the unification of Pennisetum, Cenchrus and Odontelytrum (Poaceae): a combined nuclear, plastid and morphological analysis, and nomenclatural combinations in Cenchrus. Annals of Botany 106: 107–130.

Chubb AL (2004) New nuclear evidence for the oldest divergence among neognath birds: the phylogenetic utility of ZENK (i). Molecular Phylogenetics and Evolution 30: 140–151.

Clarke JA, Ksepka DT, Smith NA, Norell MA (2009) Combined phylogenetic analysis of a new North American fossil species confirms widespread

314 Eocene distribution for stem rollers (Aves, Coracii). Zoological Journal of the Linnean Society 157: 586–611.

Clarke JA, Ksepka DT, Stucchi M, Urbina M, Giannini N, et al. (2007) Paleogene equatorial penguins challenge the proposed relationship between biogeography, diversity, and Cenozoic climate change. Proceedings of the National Academy of Sciences of the United States of America 104: 11545– 11550.

Clarke JA, Middleton KM (2008) Mosaicism, modules, and the evolution of birds: Results from a Bayesian approach to the study of morphological evolution using discrete character data. Systematic Biology 57: 185–201.

Clarke JA, Tambussi CP, Noriega JI, Erickson GM, Ketcham RA (2005) Definitive fossil evidence for the extant avian radiation in the Cretaceous. Nature 433: 305–308.

Clement WL, Weiblen GD (2009) Morphological evolution in the mulberry family (Moraceae). Systematic Botany 34: 530–552.

Cloutier R (1991) Patterns, trends, and rates of evolution within the Actinistia. Environmental Biology of Fishes 32: 23–58.

Corona AM, Morrone JJ (2007) Phylogenetic analysis of the subgenus Lampetis (Spinthoptera) (Coleoptera: Buprestidae) of North and Central America, and the West Indies. Journal of Natural History 41: 1035–1046.

Cotton TJ (2001) The phylogeny and systematics of blind Cambrian ptychoparioid trilobites. Palaeontology 44: 167–207.

Cracraft J (2001) Avian evolution, Gondwana biogeography and the Cretaceous- Tertiary mass extinction event. Proceedings of Royal Society of London Series B 268: 459–469.

Cunningham C, Omland K, Oakley T (1998) Reconstructing ancestral character states: a critical reappraisal. Trends in Ecology and Evolution 13: 361–366.

DaCosta MA, Larson P, Donahue JP, Weller SJ (2006) Phylogeny of milkweed

315 tussocks (Arctiidae: Arctiinae: Phaegopterini) and its implications for evolution of ultrasound communication. Annals of the Entomological Society of America 99: 723–742.

Damgaard J (2008) Phylogeny of the semiaquatic bugs (Hemiptera-Heteroptera, Gerromorpha). Insect Systematics and Evolution 39: 431–460.

Dávalos LM, Cirranello AL, Geisler JH, Simmons NB (2012) Understanding phylogenetic incongruence: lessons from phyllostomid bats. Biological Reviews 87: 991–1024.

Davies TJ, Savolainen V (2006) Neutral theory, phylogenies, and the relationship between phenotypic change and evolutionary rates. Evolution 60: 476–483. de Sousa L de OF, Wendt T, Brown GK, Tuthill DE, Evans TM (2007) Monophyly and phylogenetic relationships in Lymania (Bromeliaceae: Bromelioideae) based on morphology and chloroplast DNA sequences. Systematic Botany 32: 264–270. de Villemereuil P, Wells JA, Edwards RD, Blomberg SP (2012) Bayesian models for comparative analysis integrating phylogenetic uncertainty. BMC Evolutionary Biology 12: 102.

De Wit P, Rota E, Erséus C (2011) Phylogeny and character evolution in Grania (Annelida, Clitellata). Zoologica Scripta 40: 509–519.

Dececchi TA, Larsson HCE (2009) Patristic evolutionary rates suggest a punctuated pattern in forelimb evolution before and after the origin of birds. Paleobiology 35: 1–12.

Derstler KL (1982) Estimating the rate of morphological change in fossil groups. Proceedings of the Third North American Paleontological Convention. pp. 131–136.

Dikow T (2009) A phylogenetic hypothesis for Asilidae based on a total evidence analysis of morphological and DNA sequence data (Insecta: Diptera: Brachycera: Asiloidea). Organisms Diversity & Evolution 9: 165–188.

316 Donoghue M, Doyle J (1989) Phylogenetic analysis of angiosperms and the relationships of Hamamelidae. In: Crane P, Blackmore S, editors. Evolution, Systematics, and Fossil History of the Hamamelidae. Oxford, UK: Claredon Press. pp. 17–45. dos Reis M, Inoue J, Hasegawa M, Asher RJ, Donoghue PC, et al. (2012) Phylogenomic datasets provide both precision and accuracy in estimating the timescale of placental mammal phylogeny. Proceedings of Royal Society of London Series B 279: 3491–3500. dos Reis M, Yang Z (2011) Approximate likelihood calculation on a phylogeny for Bayesian estimation of divergence times. Molecular Biology and Evolution 28: 2161–2172. dos Reis M, Yang Z (2013) The unbearable uncertainty of Bayesian divergence time estimation. Journal of Systematics and Evolution 51: 30–43.

Drummond AJ, Ho SYW, Phillips MJ, Rambaut A (2006) Relaxed phylogenetics and dating with confidence. PLoS Biology 4: e88.

Drummond AJ, Suchard MA (2010) Bayesian random local clocks, or one rate to rule them all. BMC Biology 8: 114.

Drummond AJ, Suchard MA, Xie D, Rambaut A (2012) Bayesian phylogenetics with BEAUti and the BEAST 1.7. Molecular Biology and Evolution 29: 1969–1973.

Dyke GJ, van Tuinen M (2004) The evolutionary radiation of modern birds (Neornithes): reconciling molecules, morphology and the fossil record. Zoological Journal of the Linnean Society 141: 153–177.

Dyke GJ, Waterhouse DM, Kristoffersen A V (2004) Three new fossil landbirds from the early Paleogene of Denmark. Bulletin of the Geological Society of Denmark 51: 77–85.

Dyke G, Gardiner E (2011) The utility of fossil taxa and the evolution of modern birds: commentary and analysis. In: Dyke G, Kaiser G, editors. Living

317 Dinosaurs: The Evolutionary History of Modern Birds. Chichester, UK: John Wiley & Sons, Ltd. pp. 146–154.

Eastman JM, Alfaro ME, Joyce P, Hipp AL, Harmon LJ (2011) A novel comparative method for identifying shifts in the rate of character evolution on trees. Evolution 65: 3578–3589.

Ericson PG, Anderson CL, Britton T, Elzanowski A, Johansson US, et al. (2006) Diversification of Neoaves: integration of molecular sequence data and fossils. Biology Letters 2: 543–547.

Eriksson R (1994) Phylogeny of the Cyclanthaceae. Plant Systematics and Evolution 190: 31–47.

Ezcurra MD, Brusatte SL (2011) Taxonomic and phylogenetic reassessment of the early neotheropod dinosaur Camposaurus arizonensis from the Late Triassic of North America. Palaeontology 54: 763–772.

Fain MG, Houde P (2004) Parallel radiations in the primary clades of birds. Evolution 58: 2558–2573.

Fain MG, Krajewski C, Houde P (2007) Phylogeny of “core Gruiformes” (Aves: Grues) and resolution of the Limpkin-Sungrebe problem. Molecular Phylogenetics and Evolution 43: 515–529.

Fan Y, Wu R, Chen M-H, Kuo L, Lewis PO (2011) Choosing among partition models in Bayesian phylogenetics. Molecular Biology and Evolution 28: 523–532.

Farmer SB, Schilling EE (2002) Phylogenetic analyses of Trilliaceae based on morphological and molecular data. Systematic Botany 27: 674–692.

Farris JS (1969) A successive approximations approach to character weighting. Systematic Biology 18: 374–385.

Feduccia A (2003) “Big bang” for tertiary birds? Trends in Ecology and Evolution 18: 172–176.

318 Felsenstein J (2004) Inferring Phylogenies. Sunderland, Massachusetts: Sinauer Associates, Inc.

Felsenstein J (2001) Taking variation of evolutionary rates between sites into account in inferring phylogenies. Journal of Molecular Evolution 53: 447– 455.

Felsenstein J (2012) A comparative method for both discrete and continuous characters using the threshold model. The American Naturalist 179: 145– 156.

Finarelli JA, Flynn JJ (2006) Ancestral state reconstruction of body size in the Caniformia (Carnivora, Mammalia): the effects of incorporating data from the fossil record. Systematic Biology 55: 301–313.

Fitch WM, Beintema JJ (1990) Correcting parsimonious trees for unseen nucleotide substitutions: the effect of dense branching. Molecular Biology and Evolution 7: 438–443.

Forey PL (1988) Golden jubilee for the coelacanth Latimeria chalumnae. Nature 336: 727–732.

Franzen JL (2005) The implications of the numerical dating of the Messel fossil deposit (Eocene, Germany) for mammalian biochronology. Annales de Paléontologie 91: 329–335.

Frick H, Nentwig W, Kropf C (2010) Progress in erigonine spider phylogeny—the Savignia-group is not monophyletic (Araneae: Linyphiidae). Organisms Diversity & Evolution 10: 297–310.

Fröbisch NB, Schoch RR (2009) Testing the impact of miniaturization on phylogeny: Paleozoic dissorophoid amphibians. Systematic Biology 58: 312–327.

Garland T (1992) Rate tests for phenotypic evolution using phylogenetically independent contrasts. The American Naturalist 140: 509–519.

Gaubert P, Wozencraft WC, Cordeiro-Estrela P, Veron G (2005) Mosaics of

319 convergences and noise in morphological phylogenies: what’s in a viverrid- like carnivoran? Systematic Biology 54: 865–894.

Gavrilets S, Losos JB (2009) Adaptive radiation: contrasting theory with data. Science 323: 732–737.

Gelman A, Rubin DB (1992) Inference from iterative simulation using multiple sequences. Statistical Science 7: 457–472.

Gernandt DS, Magallón S, López GG, Flores OZ, Willyard A, et al. (2008) Use of simultaneous analyses to guide fossil‐ Based calibrations of Pinaceae phylogeny. International Journal of Plant Sciences 169: 1086–1099.

Gillooly JF, Allen AP, West GB, Brown JH (2005) The rate of DNA evolution: effects of body size and temperature on the molecular clock. Proceedings of the National Academy of Sciences of the United States of America 102: 140–145.

Gingerich PD (1993) Quantification and comparison of evolutionary rates. American Journal of Science 293-A: 433–478.

Gingerich PD (2009) Rates of evolution. Annual Review of Ecology, Evolution, and Systematics 40: 657–675.

Goldie X, Lanfear R, Bromham L (2011) Diversification and the rate of molecular evolution: no evidence of a link in mammals. BMC Evolutionary Biology 11: 286.

Goloboff PA (1997) Self-weighted optimization: tree searches and character state reconstructions under implied transformation costs. Cladistics 13: 225– 245.

Goloboff PA, Pol D (2006) Parsimony and Bayesian phylogenetics. In: Albert VA, editor. Parsimony, Phylogeny, and Genomics. Oxford, UK: Oxford University Press. pp. 148–159.

Gould SJ, Eldredge N (1977) Punctuated equilibria: the tempo and mode of evolution reconsidered. Paleobiology 3: 115–151.

320 Gradstein F, Ogg J, Schmitz M, Ogg G (2012) The Geologic Time Scale 2012. Boston, MA: Elsevier.

Gu X, Fu Y-X, Li W-H (1995) Maximum likelihood estimation of the heterogeneity of substitution rate among nucleotide sites. Molecular Biology and Evolution 12: 546–557.

Hackett SJ, Kimball RT, Reddy S, Bowie RC, Braun EL, et al. (2008) A phylogenomic study of birds reveals their evolutionary history. Science 320: 1763–1768.

Haddrath O, Baker AJ (2012) Multiple nuclear genes and retroposons support vicariance and dispersal of the palaeognaths, and an Early Cretaceous origin of modern birds. Proceedings of Royal Society of London Series B 279: 4617–4625.

Haldane J (1949) Suggestions as to quantitative measurment of rates of evolution. Evolution 3: 51–56.

Harmon LJ, Losos JB, Davies TJ, Gillespie RG, Gittleman JL, et al. (2010) Early bursts of body size and shape evolution are rare in comparative data. Evolution 64: 2385–2396.

Harmon LJ, Weir JT, Brock CD, Glor RE, Challenger W (2008) GEIGER: investigating evolutionary radiations. Bioinformatics 24: 129–131.

Harrison CJ (1984) A revision of the fossil swifts (Vertebrata, Aves, suborder Apodi), with descriptions of three new genera and two new species. Mededelingen van de Werkgroep voor Tertiaire en Kwartaire Geologie 21: 157–177.

Harrison CJ, Walker CA (1976) Birds of the British Upper Eocene. Zoological Journal of the Linnean Society 59: 323–351.

Harshman J, Braun EL, Braun MJ, Huddleston CJ, Bowie RCK, et al. (2008) Phylogenomic evidence for multiple losses of flight in ratite birds. Proceedings of the National Academy of Sciences of the United States of

321 America 105: 13462–13467.

Heibl C (2008) PHYLOCH: R language tree plotting tools and interfaces to diverse phylogenetic software packages. Available: http://www.christophheibl.de/Rpackages.html.

Heled J, Drummond AJ (2010) Bayesian inference of species trees from multilocus data. Molecular Biology and Evolution 27: 570–580.

Heled J, Drummond AJ (2011) Calibrated tree priors for relaxed phylogenetics and divergence time estimation. Systematic Biology 61: 138–149.

Hendry AP, Kinnison MT (1999) Perspective: the pace of modern life: measuring rates of contemporary microevolution. Evolution 53: 1637–1653.

Hennig W (1966) Phylogenetic systematics. Urbana, IL, USA: University of Illinois Press.

Hinchliff CE, Roalson EH (2013) Using supermatrices for phylogenetic inquiry: An example using the sedges. Systematic Biology 62: 205–219.

Ho SYW (2007) Calibrating molecular estimates of substitution rates and divergence times in birds. Journal of Avian Biology 38: 409–414.

Ho SYW, Phillips MJ (2009) Accounting for calibration uncertainty in phylogenetic estimation of evolutionary divergence times. Systematic Biology 58: 367–380.

Holland, BR, Spencer HG, Worthy TH, Kennedy M (2010) Identifying cliques of convergent characters: concerted evolution in the cormorants and shags. Systematic Biology 59: 433-445.

Houde P, Olson SL (1992) A radiation of Coly-like birds from the Eocene of North America (Aves: Sandcoleiformes New Order). Natural History Museum of Los Angeles County, Science Series 36: 137–160.

Housworth EA, Martins EP (2001) Random sampling of constrained phylogenies: conducting phylogenetic analyses when the phylogeny is partially known.

322 Systematic Biology 50: 628–639.

Huang D, Fitzhugh K, Rouse GW (2011) Inference of phylogenetic relationships within Fabriciidae (Sabellida, Annelida) using molecular and morphological data. Cladistics 27: 356–379.

Huelsenbeck JP (1997) Is the Felsenstein zone a fly trap? Systematic Biology 46: 69–74.

Huelsenbeck JP, Suchard MA (2007) A nonparametric method for accommodating and testing across-site rate variation. Systematic Biology 56: 975–987.

Hugall AF, Lee MSY (2007) The likelihood node density effect and consequences for evolutionary studies of molecular rates. Evolution 61: 2293–2307.

Hughes CE, Lewis GP, Yomona AD, Reynel C (2004) Maraniona. A new dalbergioid legume genus (Leguminosae, Papilionoideae) from Peru. Systematic Botany 29: 366–374.

Hunt G (2012) Measuring rates of phenotypic evolution and the inseparability of tempo and mode. Paleobiology 38: 351–373.

Izquierdo-Carrasco F, Smith SA, Stamatakis A (2011) Algorithms, data structures, and numerics for likelihood-based phylogenetic inference of huge trees. BMC Bioinformatics 12: 470.

Janecka J, Chowdhary B, Murphy W (2012) Exploring the correlations between sequence evolution rate and phenotypic divergence across the Mammalian tree provides insights into adaptive evolution. Journal of Biosciences 37: 897–909.

Jetz W, Thomas G, Joy J, Hartmann K, Mooers AØ (2012) The global diversity of birds in space and time. Nature 491: 444–448.

Johnson NL, Kotz S, Balakrishnan N (1994) Continuous Univariate Distributions: Volume 1. 2nd Edition. New York, NY: John Wiley & Sons, Ltd.

323 Just J, Wilson GD (2007) Revision of Austrosignum Hodgson and Munnogonium George & Strömberg (Paramunnidae) with descriptions of eight new genera and two new species, (Crustacea: Isopoda: Asellota). Zootaxa 1515: 1–29.

Karis P (1989) Systematics of the genus Metalasia. Opera Botanica 99: 1–150.

Kass RE, Raftery AE (1995) Bayes Factors. Journal of the American Statistical Association 90: 773–795.

Kearney M (2003) The phylogenetic position of Sineoamphisbaena hexatabularis reexamined. Journal of Vertebrate Paleontology 23: 394–403.

Kenaley CP (2010) Comparative innervation of cephalic photophores of the loosejaw dragonfishes (Teleostei: Stomiiformes: Stomiidae): evidence for parallel evolution of long-wave bioluminescence. Journal of Morphology 271: 418–437.

Kristoffersen AV (2002) An early Paleogene trogon (Aves: Trogoniformes) from the Fur formation, Denmark. Journal of Vertebrate Paleontology 22: 661– 666.

Ksepka DT (2009) Broken gears in the avian molecular clock: new phylogenetic analyses support stem galliform status for Gallinuloides wyomingensis and rallid affinities for Amitabha urbsinterdictensis. Cladistics 25: 173–197.

Ksepka DT, Benton MJ, Carrano MT, Gandolfo MA, Head JJ, et al. (2011a) Synthesizing and databasing fossil calibrations: divergence dating and beyond. Biology Letters 7: 801–803.

Ksepka DT, Clarke JA (2009) Affinities of Palaeospiza bella and the Phylogeny and Biogeography of Mousebirds (Coliiformes). The Auk 126: 245–259.

Ksepka DT, Clarke JA (2010a) New fossil mousebird (Aves: Coliiformes) with feather preservation provides insight into the ecological diversity of an Eocene North American avifauna. Zoological Journal of the Linnean Society 160: 685–706.

Ksepka DT, Clarke JA (2010b) Primobucco mcgrewi (Aves: Coracii) from the

324 Eocene Green River formation: New anatomical data from the earliest constrained record of stem rollers. Journal of Vertebrate Paleontology 30: 215–225.

Ksepka DT, Clarke JA (2012) A new stem parrot from the Green River Formation and the complex evolution of the grasping foot in Pan-Psittaciformes. Journal of Vertebrate Paleontology 32: 395–406.

Ksepka DT, Clarke JA, Grande L (2011b) Stem parrots (Aves, Halcyornithidae) from the Green River formation and a combined phylogeny of Pan- Psittaciformes. Journal of Paleontology 85: 835–852.

Ksepka DT, Fordyce RE, Ando T, Jones CM (2012) New fossil penguins (Aves, Sphenisciformes) from the Oligocene of New Zealand reveal the skeletal plan of stem penguins. Journal of Vertebrate Paleontology 32: 235–254.

Ksepka DT, Thomas DB (2012) Multiple cenozoic invasions of Africa by penguins (Aves, Sphenisciformes). Proceedings of Royal Society of London Series B 279: 1027–1032.

Kundu S, Jones C, Prys-Jones R, Groombridge J (2012) The evolution of the Indian Ocean parrots (Psittaciformes): extinction, adaptive radiation and eustacy. Molecular Phylogenetics and Evolution 62: 296–305.

Kurochkin EN, Dyke GJ (2011) The first fossil owls (Aves: Strigiformes) from the Paleogene of Asia and a review of the fossil record of Strigiformes. Paleontological Journal 45: 445–458.

Lambkin CL, Bartlett JS (2011) Bush Blitz aids description of three new species and a new genus of Australian beeflies (Diptera, Bombyliidae, Exoprosopini). ZooKeys 150: 231–280.

Lanfear R, Calcott B, Ho SY, Guindon S (2012) Partitionfinder: combined selection of partitioning schemes and substitution models for phylogenetic analyses. Molecular Biology and Evolution 29: 1695–1701.

Lanfear R, Ho SY, Love D, Bromham L (2010a) Mutation rate is linked to

325 diversification in birds. Proceedings of the National Academy of Sciences of the United States of America 107: 20423–20428.

Lanfear R, Welch JJ, Bromham L (2010b) Watching the clock: studying variation in rates of molecular evolution between species. Trends in Ecology and Evolution 25: 495–503.

Larkin M, Blackshields G, Brown N, Chenna R, McGettigan P, et al. (2007) Clustal W and Clustal X version 2.0. Bioinformatics 23: 2947–2948.

Larsson HC, Dececchi TA, Harrison LB (2012) Morphological largess: can morphology offer more and be modeled as a stochastic evolutionary process? In: Asher RJ, Müller J, editors. From Clone to Bone: the Synergy of Morphological and Molecular Tools in Paleobiology. Cambridge, UK: Cambridge University Press. pp. 83–116.

Laurin M (2012) Recent progress in paleontological methods for dating the Tree of Life. Frontiers in Genetics 3: 130.

Lee MS, Worthy TH (2012) Likelihood reinstates Archaeopteryx as a primitive bird. Biology Letters 8: 299–303.

Lemmon AR, Brown JM, Stanger-Hall K, Lemmon EM (2009) The effect of ambiguous data on phylogenetic estimates obtained by maximum likelihood and Bayesian inference. Systematic Biology 58: 130–145.

Lewis PO (2001) A likelihood approach to estimating phylogeny from discrete morphological character data. Systematic Biology 50: 913–925.

Lewontin R (1978) Adaptation. Scientific American 239: 156–169.

Liljeblad J, Ronquist F, Nieves-Aldrey JL, Fontal-Cazalla F, Ros-Farre P, et al. (2008) A fully web-illustrated morphological phylogenetic study of relationships among oak gall wasps and their closest relatives (Hymenoptera: Cynipidae). Zootaxa 1796: 1–73.

Lin H-C, Hastings PA (2011) Evolution of a Neotropical marine fish lineage (Subfamily Chaenopsinae, Suborder Blennioidei) based on phylogenetic

326 analysis of combined molecular and morphological data. Molecular Phylogenetics and Evolution 60: 236–248.

Lindow BEK (2011) Bird evolution across the K–Pg boundary and the basal neornithine diversification. In: Dyke G, Kaiser G, editors. Living Dinosaurs: The Evolutionary History of Modern Birds. Oxford, UK: John Wiley & Sons, Ltd. pp. 339–354.

Livezey BC (2011) Progress and obstacles in the phylogenetics of modern birds. In: Dyke G, Kaiser G, editors. Living Dinosaurs: The Evolutionary History of Modern Birds. Oxford, UK: John Wiley & Sons, Ltd. pp. 117–145.

Livezey BC, Zusi RL (2006) Phylogeny of Neornithes. Bulletin of Carnegie Museum of Natural History 37: 1–544.

Livezey BC, Zusi RL (2007) Higher-order phylogeny of modern birds (Theropoda, Aves: Neornithes) based on comparative anatomy. II. Analysis and discussion. Zoological Journal of the Linnean Society 149: 1–95.

Lloyd GT, Wang SC, Brusatte SL (2012) Identifying heterogeneity in rates of morphological evolution: discrete character change in the evolution of lungfish (Sarcopterygii; Dipnoi). Evolution 66: 330–348. doi:10.5061/dryad.pg46f.

Louchart A, Tourment N, Carrier J (2011) The earliest known pelican reveals 30 million years of evolutionary stasis in beak morphology. Journal of Ornithology 152: 15–20.

Luo Z-X, Chen P, Li G, Chen M (2007) A new eutriconodont mammal and evolutionary development in early mammals. Nature 446: 288–293.

Maddison DR (1994) Phylogenetic methods for inferring the evolutionary history and processes of change in discretely valued characters. Annual Review of Entomology 39: 267–292.

Maddison WP, Midford PE, Otto SP (2007) Estimating a binary character’s effect on speciation and extinction. Systematic Biology 56: 701–710.

327 Magnuson-Ford K, Otto SP (2012) Linking the investigations of character evolution and species diversification. The American Naturalist 180: 225– 245.

Makovicky PJ, Apesteguía S, Agnolín FL (2005) The earliest dromaeosaurid theropod from South America. Nature 437: 1007–1011.

Malécot V, Nickrent DL, Baas P, Oever L Van Den, Lobreau-Callen D (2004) A morphological cladistic analysis of Olacaceae. Systematic Botany 29: 569– 586.

Malia MJJ, Lipscomb DL, Allard MW (2003) The misleading effects of composite taxa in supermatrices. Molecular Phylogenetics and Evolution 27: 522–527.

Marek PE, Bond JE (2006) Phylogenetic systematics of the colorful, cyanide- producing millipedes of Appalachia (Polydesmida, Xystodesmidae, Apheloriini) using a total evidence Bayesian approach. Molecular Phylogenetics and Evolution 41: 704–729.

Marshall DC (2010) Cryptic failure of partitioned Bayesian phylogenetic analyses: lost in the land of long trees. Systematic Biology 59: 108–117.

Marshall IK (2003) A morphological phylogeny for four families of amblyceran lice (Phthiraptera: Amblycera: Menoponidae, Boopiidae, Laemobothriidae, Ricinidae). Zoological Journal of the Linnean Society 138: 39–82.

Martin AP, Paumbi SR (1993) Body size, metabolic rate, generation time, and the molecular clock. Proceedings of the National Academy of Sciences of the United States of America 90: 4087–4091.

Martins EP (1996) Conducting phylogenetic comparative studies when the phylogeny is not known. Evolution 50: 12–22.

Maruyama M, Steiner FM, Stauffer C, Akino T, Crozier RH, et al. (2008) A DNA and morphology based phylogenetic framework of the ant genus Lasius with hypotheses for the evolution of social parasitism and fungiculture.

328 BMC Evolutionary Biology 8: 237.

Mayr G (2000) Tiny hoopoe-like birds from the Middle Eocene of Messel (Germany). The Auk 117: 964–970.

Mayr G (2003) Phylogeny of early Tertiary swifts and hummingbirds (Aves: Apodiformes). The Auk 120: 145–151.

Mayr G (2004) Morphological evidence for sister group relationship between flamingos (Aves: Phoenicopteridae) and grebes (Podicipedidae). Zoological Journal of the Linnean Society 140: 157–169.

Mayr G (2005a) New trogons from the early Tertiary of Germany Ibis 147: 512– 518.

Mayr G (2005b) A tiny barbet-like bird from the Lower Oligocene of Germany: the smallest species and earliest substantial fossil record of the Pici (woodpeckers and allies). The Auk 122: 1055–1063.

Mayr G (2005c) A chicken-sized crane precursor from the early Oligocene of France. Die Naturwissenschaften 92: 389–393.

Mayr G (2006a) New specimens of the Eocene Messelirrisoridae (Aves: Bucerotes), with comments on the preservation of uropygial gland waxes in fossil birds from Messel and the phylogenetic affinities of Bucerotes. Palaontologische Zeitschrift 80: 390–405.

Mayr G (2006b) An osprey (Aves: Accipitridae: Pandioninae) from the early Oligocene of Germany. Senckenbergiana Lethaea 86: 93–96.

Mayr G (2007a) Avian higher-level phylogeny: well-supported clades and what we can learn from a phylogenetic analysis of 2954 morphological characters. Journal of Zoological Systematics and Evolutionary Research 46: 63–72.

Mayr G (2007b) A small representative of the Phalacrocoracoidea (Cormorants and Anhingas) from the late Oligocene of Germany. The Condor 109: 929– 942.

329 Mayr G (2009a) Paleogene Fossil Birds. Paleogene Fossil Birds. Berlin, Heidelberg: Springer.

Mayr G (2009b) A well-preserved second trogon skeleton (Aves, Trogonidae) from the middle Ecoene of Messel, Germany. Palaeobiodiversity and Palaeoenvironments 89: 1–6.

Mayr G (2010) Reappraisal of Eocypselus—a stem group apodiform from the early Eocene of Northern Europe. Palaeobiodiversity and Palaeoenvironments 90: 395–403.

Mayr G (2011a) On the osteology and phylogenetic affinities of Morsoravis sedilis (Aves) from the early Eocene Fur Formation of Denmark. Bulletin of the Geological Society of Denmark 59: 23–35.

Mayr G (2011b) The phylogeny of charadriiform birds (shorebirds and allies) – reassessng the conflict between morphology and molecules. Zoological Journal of the Linnean Society. 161: 916–934.

Mayr G (2013) Parvigruidae (Aves, core Gruiformes) from the early Oligocene of Belgium. Palaeobiodiversity and Palaeoenvironments 93: 77–89.

Mayr G, Bertelli S (2011) A record of Rhynchaeites (Aves, Threskiornithidae) from the early Eocene Fur Formation of Denmark, and the affinities of the alleged parrot Mopsitta. Palaeobiodiversity and Palaeoenvironments 91: 229–236.

Mayr G, Gregorová R (2012) A tiny stem group representative of Pici (Aves, Piciformes) from the early Oligocene of the Czech Republic. Paläontologische Zeitschrift 86: 333–343.

Mayr G, Knopf CW (2007) A stem lineage representative of buttonquails from the Lower Oligocene of Germany - fossil evidence for a charadriiform origin of the Turnicidae. Ibis 149: 774–782.

Mayr G, Mourer-Chauviré C (2004) Unusual tarsometatarsus of a mousebird from the Paleogene of France and the relationships of Selmes Peters, 1999.

330 Journal of Vertebrate Paleontology 24: 366–372.

Mayr G, Peters SD (1999) On the systematic position of the Middle Eocene swift Aegialornis szarskii Peters 1985 with description of a new swift-like bird from Messel (Aves, Apodiformes). N Jb Geol Paläont Mh 1999: 312–320.

Mayrose I, Friedman N, Pupko T (2005) A Gamma mixture model better accounts for among site rate heterogeneity. Bioinformatics 21 Suppl 2: ii151–ii158.

Mayrose I, Otto SP (2011) A likelihood method for detecting trait-dependent shifts in the rate of molecular evolution. Molecular Biology and Evolution 28: 759–770.

McCormack JE, Harvey MG, Faircloth BC, Crawford NG, Glenn TC, et al. (2013) A phylogeny of birds based on over 1,500 loci collected by target enrichment and high-throughput sequencing. PloS one 8: e54848.

McCormack JE, Heled J, Delaney KS, Peterson AT, Knowles LL (2011) Calibrating divergence times on species trees versus gene trees: implications for speciation history of Aphelocoma jays. Evolution 65: 184– 202.

Mertz DF, Renne PR, Wuttke M, Mödden C (2007) A numerically calibrated reference level (MP28) for the terrestrial mammal-based biozonation of the European Upper Oligocene. International Journal of Earth Sciences 96: 353–361.

Minin VN, Suchard MA (2008) Fast, accurate and simulation-free stochastic mapping. Philosophical Transactions of the Royal Society of London Series B 363: 3985–3895.

Mourer-Chauviré C (1993) Les gangas (Aves, Columbiformes, Pteroclidae) du Paleogene et du Miocene Inferieur de France. Palaeovertebrata 22: 73–98.

Mourer-Chauviré C, Tabuce R, Mahboubi M, Adaci M, Bensalah M (2011) A Phororhacoid bird from the Eocene of Africa. Die Naturwissenschaften 98: 815–823.

331 Müller J, Hipsley CA, Head JJ, Kardjilov N, Hilger A, et al. (2011) Eocene lizard from Germany reveals amphisbaenian origins. Nature 473: 364–367.

Müller J, Reisz RR (2006) The phylogeny of early eureptiles: comparing parsimony and Bayesian approaches in the investigation of a basal fossil clade. Systematic Biology 55: 503–511.

Nabholz B, Glémin S, Galtier N (2008) Strong variations of mitochondrial mutation rate across mammals-the longevity hypothesis. Molecular Biology and Evolution 25: 120–130.

Naish D, Dyke G, Cau A, Escuillié F, Godefroit P (2012) A gigantic bird from the Upper Cretaceous of Central Asia. Biology Letters 8: 97–100.

Neff N (1986) A rational basis for a priori character weighting. Systematic Biology 35: 110–123.

Notredame C, Higgins DG, Heringa J (2000) T-Coffee: A novel method for fast and accurate multiple sequence alignment. Journal of Molecular Biology 302: 205–217.

Nylander J (2004) MrModelTest v2. Program distributed by the author.

Nylander JA, Ronquist F, Huelsenbeck JP, Nieves-Aldrey JL (2004) Bayesian phylogenetic analysis of combined data. Systematic Biology 53: 47–67.

O’Leary MA, Kaufman S (2011) MorphoBank: phylophenomics in the “cloud”. Cladistics 27: 1–9.

O’Connor TD, Mundy NI (2009) Genotype-phenotype associations: substitution models to detect evolutionary associations between phenotypic variables and genotypic evolutionary rate. Bioinformatics 25: i94–i100.

O’Leary MA (1999) Parsimony analysis of total evidence from extinct and extant taxa and the Cetacean-Artiodactyl question (Mammalia, Ungulata). Cladistics 15: 315–330.

O’Leary MA, Bloch JI, Flynn JJ, Gaudin TJ, Giallombardo AG, et al. (2013) The

332 Placental mammal ancestor and the post-K-Pg radiation of placentals. Science 339: 662–667.

Oliveira É V, Goin FJ (2011) A reassessment of bunodont metatherians from the Paleogene of Itaboraí (Brazil): systematics and age of the Itaboraian SALMA. Revista Brasileira de Paleontologia 14: 105–136.

Olson SL (1977) A Lower Eocene Frigatebird from the Green River formation of Wyoming (Pelecaniformes: Fregatidae). Smithsonian Contributions to Paleobiology 35: 1–33.

Olson SL, Matsuoka H (2005) New specimens of the early Eocene frigatebird Limnofregata (Pelecaniformes: Fregatidae), with the description of a new species. Zootaxa 1046: 1–15.

O’Meara BC, Ané C, Sanderson MJ, Wainwright PC (2006) Testing for different rates of continuous trait evolution using likelihood. Evolution 60: 922–933.

Omland KE (1997) Correlated rates of molecular and morphological evolution. Evolution 51: 1381–1393.

Pacheco MA, Battistuzzi FU, Lentino M, Aguilar RF, Kumar S, et al. (2011) Evolution of modern birds revealed by mitogenomics: timing the radiation and origin of major orders. Molecular Biology and Evolution 28: 1927– 1942.

Pagel M (1994) Detecting correlated evolution on phylogenies: a general method for the comparative analysis of discrete characters. Proceedings of Royal Society of London Series B 255: 37–45.

Pagel M (2012) First steps for birds. Nature 491: 337.

Paradis E, Claude J, Strimmer K (2004) APE: analysis of phylogenetics and evolution in R language. Bioinformatics 20: 289–290.

Parham JF, Donoghue PC, Bell CJ, Calway TD, Head JJ, et al. (2012) Best practices for justifying fossil calibrations. Systematic Biology 61: 346–359.

333 Peach M, Rouse G (2004) Phylogenetic trends in the abundance and distribution of pit organs of elasmobranchs. Acta Zoologica 85: 233–244.

Penneys DS, Judd WS (2005) A systematic revision and cladistic analysis of Charianthus (Melastomataceae) using morphological and molecular characters. Systematic Botany 30: 559–584.

Penny D, Hendy MD (1985) The use of tree comparaison metrics. Systematic Biology 34: 75–82.

Penny D, Phillips MJ (2004) The rise of birds and mammals: are microevolutionary processes sufficient for macroevolution? Trends in Ecology and Evolution 19: 516–522.

Percy DM (2003) Radiation, diversity, and host-plant interactions among island and continental legume-feeding psyllids. Evolution 57: 2540–2456.

Pereira SL, Baker AJ (2006) A mitogenomic timescale for birds detects variable phylogenetic rates of molecular evolution and refutes the standard molecular clock. Molecular Biology and Evolution 23: 1731–1740.

Peters D (1985) Ein neuer Segler aus der Grube Messel und seine Bedeutung für den Status der Aegialornithidae (Aves: Apodiformes). Senckenbergiana lethaea 66: 143–164.

Peters SD (1998) Erstnachweis eines Seglers aus dem Geiseltal (Aves: Apodiformes). Senckenbergiana Lethaea 78: 211–212.

Phillips MJ, Gibb GC, Crimp EA, Penny D (2010) Tinamous and moa flock together: mitochondrial genome sequence analysis reveals independent losses of flight among ratites. Systematic Biology 59: 90–107.

Pitz KM, Sierwald P (2010) Phylogeny of the millipede order Spirobolida (Arthropoda: Diplopoda: Helminthomorpha). Cladistics 26: 497–525.

Plummer M, Best N, Cowles K, Vines K (2006) CODA: Convergence Diagnostics and Output Analysis for MCMC. R-News 6: 7–11.

334 Poe S, Chubb AL (2004) Birds in a bush: five genes indicate explosive evolution of avian orders. Evolution 58: 404–415.

Posada D, Buckley TR (2004) Model selection and model averaging in phylogenetics: advantages of Akaike Information Criterion and Bayesian approaches over likelihood ratio tests. Systematic Biology 53: 793–808.

Pratt RC, Gibb GC, Morgan-Richards M, Phillips MJ, Hendy MD, et al. (2009) Toward resolving deep Neoaves phylogeny: data, signal enhancement, and priors. Molecular Biology and Evolution 26: 313–326.

Prieto-Márquez A (2010) Global phylogeny of Hadrosauridae (Dinosauria: Ornithopoda) using parsimony and Bayesian methods. Zoological Journal of the Linnean Society 159: 435–502.

Pruitt KD, Tatusova T, Maglott DR (2007) NCBI reference sequences (RefSeq): a curated non-redundant sequence database of genomes, transcripts and proteins. Nucleic Acids Research 35: D61–D65.

Pupko T, Mayrose I (2010) Probabilistic methods and rate heterogeneity. In: Lodhi HM, Muggleton SH, editors. Elements of Computational Systems Biology (Volume 8). Hoboken, NJ: John Wiley & Sons, Ltd. pp. 257–280.

Pyron RA (2011) Divergence time estimation using fossils as terminal taxa and the origins of Lissamphibia. Systematic Biology 60: 466–481.

Pyron RA, Kandambi HD, Hendry CR, Pushpamal V, Burbrink FT, et al. (2013) Genus-level phylogeny of snakes reveals the origins of species richness in Sri Lanka. Molecular Phylogenetics and Evolution 66: 969–978.

Pyron RA, Wiens JJ (2011) A large-scale phylogeny of Amphibia including over 2800 species, and a revised classification of extant frogs, salamanders, and caecilians. Molecular Phylogenetics and Evolution 61: 543–583.

R Core Team (2012) R: A language and environment for statistical computing.Vienna, Austria: R Foundation for Statistical Computing.

Rabosky DL (2006) LASER : A maximum likelihood toolkit for detecting

335 temporal shifts in diversifi cation rates from molecular phylogenies. Evolutionary Bioinformatics Online 2: 247–250.

Rabosky DL, Adams DC (2012) Rates of morphological evolution are correlated with species richness in salamanders. Evolution 66: 1807–1818.

Rabosky DL, Lovette IJ (2008) Explosive evolutionary radiations: decreasing speciation or increasing extinction through time? Evolution 62: 1866–1875.

Rambaut A, Drummond AJ (2007) Tracer, version 1.5. Oxford, UK: Oxford University.

Rambaut A (2008) FigTree v1.3.1. Available from the author: http://tree.bio.ed.ac.uk/software/figtree/.

Rannala B, Zhu T, Yang Z (2012) Tail paradox, partial identifiability and influential priors in Bayesian branch length inference. Molecular Biology and Evolution 29: 325–335.

Rasmussen DT, Olson SL, Simons EL (1987) Fossil birds from the Oligocene Jebel Qatrani formation, Fayum Province, Egypt. Smithsonian Contributions to Paleobiology 62: 1–20.

Revell LJ, Mahler DL, Peres-Neto PR, Redelings BD (2012) A new phylogenetic method for identifying exceptional phenotypic diversification. Evolution 66: 135–146.

Rich PV, Bohaska DJ (1976) The world’s oldest owl: a new strigiform from the Paleocene of southwestern Colorado. Smithsonian Contributions to Paleobiology 27: 87–93.

Ricklefs RE (2004) Cladogenesis and morphological diversification in passerine birds. Nature 430: 338–341.

Roalson EH, Skog LE, Zimmer EA (2008) Untangling Gloxinieae (Gesneriaceae). II . Reconstructing biogeographic patterns and estimating divergence times among New World continental and island lineages. Zootaxa 33: 159–175.

336 Rodman JE (1991) A taxonomic analysis of glucosinolate-producing plants, part 2: cladistics. Systematic Botany 16: 619–629.

Roelants K, Haas A, Bossuyt F (2011) Anuran radiations and the evolution of tadpole morphospace. Proceedings of the National Academy of Sciences of the United States of America 108: 8731–8736.

Ronquist F (2004) Bayesian inference of character evolution. Trends in Ecology and Evolution 19: 475–481.

Ronquist F, Klopfstein S, Vilhelmsen L, Schulmeister S, Murray DL, et al. (2012a) A total-evidence approach to dating with fossils, applied to the early radiation of the Hymenoptera. Systematic Biology 61: 973–999.

Ronquist F, Teslenko M, van der Mark P, Ayres DL, Darling A, et al. (2012b) MrBayes 3.2: Efficient Bayesian phylogenetic inference and model choice across a large model space. Systematic Biology 61: 539–542.

Ruiz-Sanchez E, Sosa V, Mejía-Saules MT (2008) Phylogenetics of Otatea Inferred from Morphology and Chloroplast DNA Sequence Data, and Recircumscription of Guaduinae (Poaceae: Bambusoideae). Systematic Botany 33: 277–283.

Ruta M, Wagner PJ, Coates MI (2006) Evolutionary patterns in early tetrapods. I. Rapid initial diversification followed by decrease in rates of character change. Proceedings of Royal Society of London Series B 273: 2107–2111.

Rzhetsky A, Nei M (1992) A simple method for estimating and testing minimum- evolution trees. Molecular Biology and Evolution 9: 945–967.

Sanderson MJ (1990) Estimating rates of speciation and evolution: a bias due to homoplasy? Cladistics 6: 387–391.

Schlosser G, Wagner GP, editors (2004) Modularity in development and evolution. Chicago, IL: University of Chicago Press.

Schluter D (2000) The ecology of adaptive radiation. Oxford, UK: Oxford University Press.

337 Schneider H, Smith AR, Pryer KM (2009) Is morphology really at odds with molecules in estimating fern phylogeny? Systematic Botany 34: 455–475.

Schopf TJ, Raup DM, Gould SJ, Simberloff DS (1975) Genomic versus morphologic rates of evolution: influence of morphologic complexity. Paleobiology 1: 63–70.

Schulman L, Hyvönen J (2003) A cladistic analysis of Adelobotrys (Melastomataceae) based on morphology, with notes on generic limits within the Tribe Merianieae. Systematic Botany 28: 738–756.

Schwarz G (1978) Estimating the dimension of a model. The Annals of Statistics 6: 461–464.

Secord R, Gingerich PD, Smith ME, Clyde WC, Wilf P, et al. (2006) Geochronology and mammalian biostratigraphy of Middle and Upper Paleocene continental strata, Bighorn Basin, Wyoming. American Journal of Science 306: 211–245.

Seiffert ER (2006) Revised age estimates for the later Paleogene mammal faunas of Egypt and Oman. Proceedings of the National Academy of Sciences of the United States of America 103: 5000–5005.

Seiffert ER, Simons EL, Ryan TM, Attia Y (2005) Additional remains of Wadilemur elegans, a primitive stem galagid from the late Eocene of Egypt. Proceedings of the National Academy of Sciences of the United States of America 102: 11396–11401.

Seligmann H (2010) Positive correlations between molecular and morphological rates of evolution. Journal of Theoretical Biology 264: 799–807.

Shapiro B, Rambaut A, Drummond AJ (2006) Choosing appropriate substitution models for the phylogenetic analysis of protein-coding sequences. Molecular Biology and Evolution 23: 7–9.

Sharkey MJ, Carpenter JM, Vilhelmsen L, Heraty J, Liljeblad J, et al. (2012) Phylogenetic relationships among superfamilies of Hymenoptera.

338 Cladistics 28: 80–112.

Sibley CG, Ahlquist JE (1990) The phylogeny and classification of birds: a study in molecular evolution. New Haven, CT: Yale University Press.

Sidlauskas BL, Vari RP (2008) Phylogenetic relationships within the South American fish family Anostomidae (Teleostei, Ostariophysi, Characiformes). Zoological Journal of the Linnean Society 154: 70–210.

Sievers F, Wilm A, Dineen D, Gibson TJ, Karplus K, et al. (2011) Fast, scalable generation of high-quality protein multiple sequence alignments using Clustal Omega. Molecular Systems Biology 7: 539.

Simmons M, Bacon CD, Cappa JJ, McKenna MJ (2012) Phylogeny of Celastraceae subfamilies Cassinoideae and Tripterygioideae inferred from morphological characters and nuclear and plastid loci. Systematic Botany 37: 456–467.

Simpson GG (1949) Rates of evolution in animals. In: Jespen GL, Mayr E, Simpson GG, editors. Genetics, Paleontology, and Evolution. Princeton, New Jersey: Princeton University Press. pp. 205–228.

Simpson GG (1944) Tempo and Mode in Evolution. New York, NY: Columbia University Press.

Simpson MG (1990) Phylogeny and classification of the Haemodoraceae. Annals of the Missouri Botanical Garden 77: 722–784.

Skinner A (2010) Rate heterogeneity, ancestral character state reconstruction, and the evolution of limb morphology in Lerista (Scincidae, Squamata). Systematic Biology 59: 723–740.

Slack KE, Jones CM, Ando T, Harrison GA, Fordyce RE, et al. (2006) Early penguin fossils, plus mitochondrial genomes, calibrate avian evolution. Molecular Biology and Evolution 23: 1144–1155.

Smith G (1992) Phylogeny and biogeography of the Catostomidae, freshwater fishes of North America and Asia. In: Systematics, Historical Ecology, and

339 North American Freshwater Fishes. Stanford, CA: Stanford University Press. pp. 778–826.

Smith JJ, Jaycox M, Smith-caldas MR, Bush GL (2005) Analysis of mitochondrial DNA and morphological characters in the subtribe Carpomyina (Diptera: Tephritidae). Isreali Journal of Entomology 35: 317– 340.

Smith JV, Braun EL, Kimball RT (2013) Ratite non-monophyly: independent evidence from 40 novel loci. Systematic Biology 62: 35–49.

Smith ME, Carroll AR, Singer BS (2008) Synoptic reconstruction of a major ancient lake system: Eocene Green River formation, western United States. Geological Society of America Bulletin 120: 54–84.

Smith NA, Clarke JA (2011) An alphataxonomic revision of extinct and extant razorbills (Aves, Alcidae): a combined morphometric and phylogenetic approach. Ornithological Monographs 72: 1–61.

Smith ND (2010) Phylogenetic analysis of pelecaniformes (Aves) based on osteological data: implications for waterbird phylogeny and fossil calibration studies. PloS one 5: e13354. A

Smith NA (2011) Taxonomic revision and phylogenetic analysis of the flightless Mancallinae (Aves, Pan-Alcidae). ZooKeys 91: 1–116.

Smith SA, Donoghue MJ (2008) Rates of molecular evolution are linked to life history in flowering plants. Science 322: 86–89.

Spencer MR, Wilberg EW. In Press. Efficacy or convenience? Model-based approaches to phylogeny estimation using morphological data. Cladistics.

Stamatakis A (2006) RAxML-VI-HPC: maximum likelihood-based phylogenetic analyses with thousands of taxa and mixed models. Bioinformatics 22: 2688–2690.

Sterli J, Pol D, Laurin M. In Press. Incorporating phylogenetic uncertainty on phylogeny-based palaeontological dating and the timing of turtle

340 diversification. Cladistics

Struwe L, Albert VA, Calió MF, Frasier C, Lepis KB, et al. (2009) Evolutionary patterns in neotropical Helieae (Gentianaceae): evidence from morphology, chloroplast and nuclear DNA sequences. Taxon 58: 479–499.

Suchard MA, Rambaut A (2009) Many-core algorithms for statistical phylogenetics. Bioinformatics 25: 1370–1376.

Sugiura N (1978) Further analysis of the data by Akaike’s Information Criterion and the finite corrections. Communications in Statistics - Theory and Methods A7: 13–26.

Suh A, Paus M, Kiefmann M, Churakov G, Franke FA, et al. (2011) Mesozoic retroposons reveal parrots as the closest living relatives of passerine birds. Nature Communications 2: 443.

Sukumaran J, Holder MT (2010) DendroPy: a Python library for phylogenetic computing. Bioinformatics 26: 1569–1571.

Sullivan J, Swofford DL (2001) Should we use model-based methods for phylogenetic inference when we know that assumptions about among-site rate variation and nucleotide substitution pattern are violated? Systematic Biology 50: 723–729.

Sundue MA (2010) A morphological cladistic analysis of Terpsichore (Polypodiaceae). Systematic Botany 35: 716–729.

Swofford D (2003) PAUP*: Phylogenetic Analysis Using Parsimony (* and other methods). Version 4. Sinauer Associates, Sunderland, Massachusetts.

Syme AE, Oakley TH (2012) Dispersal between shallow and abyssal seas and evolutionary loss and regain of compound eyes in cylindroleberidid ostracods: conflicting conclusions from different comparative methods. Systematic Biology 61: 314–336.

Thomas GH, Freckleton RP, Székely T (2006) Comparative analyses of the influence of developmental mode on phenotypic diversification rates in

341 shorebirds. Proceedings of Royal Society of London Series B 273: 1619– 1624.

Tourasse NJ, Gouy M (1997) Evolutionary distances between nucleotide sequences based on the distribution of substitution rates among sites as estimated by parsimony. Molecular Biology and Evolution 14: 287–298.

Vega AS, Rua GH, Fabbri LT, de Agrasar ZER (2009) A morphology-based cladistic analysis of Digitaria (Poaceae, Panicoideae, Paniceae). Systematic Botany 34: 312–323.

Venditti C, Meade A, Pagel M (2006) Detecting the node-density artifact in phylogeny reconstruction. Systematic Biology 55: 637–643.

Venditti C, Meade A, Pagel M (2011) Multiple routes to mammalian diversity. Nature 479: 393–396.

Vieira GH, Colli GR, Báo SN (2005) Phylogenetic relationships of corytophanid lizards (Iguania, Squamata, Reptilia) based on partitioned and total evidence analyses of sperm morphology, gross morphology, and DNA data. Zoologica Scripta 34: 605–625.

Voss RS, Jansa SA (2009) Phylogenetic relationships and classification of didelphid marsupials, an extant radiation of New World metatherian mammals. Bulletin of the American Museum of Natural History 322: 1– 177.

Waddell PJ, Penny D, Moore T (1997) Hadamard conjugations and modeling sequence evolution with unequal rates across sites. Molecular Phylogenetics and Evolution 8: 33–50.

Wagner PJ (2012) Modelling rate distributions using character compatibility: implications for morphological evolution among fossil invertebrates. Biology Letters 8: 143–146.

Wagner PJ (1997) Patterns of morphologic diversification among the Rostroconchia. Paleobiology 23: 115–150.

342 Wang N, Braun EL, Kimball RT (2012) Testing hypotheses about the sister group of the passeriformes using an independent 30-locus data set. Molecular Biology and Evolution 29: 737–750.

Wappler T, Smith VS, Dalgleish RC (2004) Scratching an ancient itch: an Eocene bird louse fossil. Proceedings of Royal Society of London Series B (Suppl) 271: S255–S258.

Warnock RC, Yang Z, Donoghue PC (2012) Exploring uncertainty in the calibration of the molecular clock. Biology Letters 8: 156–159.

Waterhouse DM, Lindow BE, Zelenkov N V, Dyke GJ (2008) Two new parrots (Psittaciformes) from the Lower Eocene Fur Formation of Denmark. Palaeontology 51: 575–582.

Weiblen GD (2001) Phylogenetic relationships of fig wasps pollinating functionally dioecious Ficus based on mitochondrial DNA sequences and morphology. Systematic Biology 50: 243–267.

Werenkraut V, Ramirez MJ (2009) A revision and phylogenetic analysis of the spider genus Coptoprepes Simon (Araneae: Anyphaenidae, Amaurobioidinae). Zootaxa 2212: 1–40.

Westoll S (1949) On the evolution of the Dipnoi. In: Jepsen GL, Mayr E, Simpson GG, editors. Genetics, Paleontology, and Evolution. Princeton, New Jersey: Princeton University Press. pp. 121–184.

Wiens JJ, Bonett RM, Chippindale PT (2005) Ontogeny discombobulates phylogeny: paedomorphosis and higher-level salamander relationships. Systematic Biology 54: 91–110.

Wiens JJ, Kuczynski CA, Townsend T, Reeder TW, Mulcahy DG, et al. (2010) Combining phylogenomics and fossils in higher-level squamate reptile phylogeny: molecular data change the placement of fossil taxa. Systematic Biology 59: 674–688.

Wiens JJ, Morrill MC (2011) Missing data in phylogenetic analysis: reconciling

343 results from simulations and empirical data. Systematic Biology 60: 719– 731.

Wilkinson RD, Steiper ME, Soligo C, Martin RD, Yang Z, et al. (2011) Dating primate divergences through an integrated analysis of palaeontological and molecular data. Systematic Biology 60: 16–31.

Wilson GD, Edgecombe GD (2003) The Triassic osopod Protamphisopus wianamattensis (Chilton) and comparison with extant taxa (Crustacea, Phreatoicidea). Journal of Paleontology 77: 454–470.

Wood HM, Matzke NJ, Gillespie RG, Griswold CE (2013) Treating fossils as terminal taxa in divergence time estimation reveals ancient vicariance patterns in the palpimanoid spiders. Systematic Biology 62: 264–284.

Worthy TH (2011) Descriptions and phylogenetic relationships of a new genus and two new species of Oligo-Miocene cormorants (Aves: Phalacrocoracidae) from Australia. Zoological Journal of the Linnean Society 163: 277–314.

Worthy TH (2012) A new species of Oligo-Miocene darter (Aves: Anhingidae) from Australia. The Auk 129: 96–104.

Wortley A, Harris D, Scotland R (2007) On the taxonomy and phylogenetic position of Thomandersia. Systematic Botany 32: 415–444.

Wortley AH, Scotland RW (2006) The effect of combining molecular and morphological data in published phylogenetic analyses. Systematic Biology 55: 677–685.

Wright TF, Schirtzinger EE, Matsumoto T, Eberhard JR, Graves GR, et al. (2008) A multilocus molecular phylogeny of the parrots (Psittaciformes): support for a Gondwanan origin during the Cretaceous. Molecular Biology and Evolution 25: 2141–2156.

Xie W, Lewis PO, Fan Y, Kuo L, Chen M-H (2011) Improving marginal likelihood estimation for Bayesian phylogenetic model selection.

344 Systematic Biology 60: 150–160.

Yang Z (1996) Among-site rate variation and its impact on phylogenetic analyses. Trends in Ecology and Evolution 11: 367–372.

Yang Z (2006) Computational Molecular Evolution. Oxford, UK: Oxford University Press.

Yang Z (2007) PAML 4: Phylogenetic Analysis by Maximum Likelihood. Molecular Biology and Evolution 24: 1586–1691.

Yang Z (1994) Maximum likelihood phylogenetic estimation from DNA sequences with variable rates over sites: approximate methods. Journal of Molecular Evolution 39: 306–314.

Yang Z, Rannala B (2012) Molecular phylogenetics: principles and practice. Nature Reviews Genetics 13: 303–314.

Yuri T, Kimball RT, Harshman J, Bowie RC, Braun MJ, et al. (2013) Parsimony and model-based analyses of indels in avian nuclear genes reveal congruent and incongruent phylogenetic signals. Biology 2: 419–444.

Zelenkov N V, Dyke GJ (2008) The fossil record and evolution of mousebirds (Aves: Coliiformes). Palaeontology 51: 1403–1418.

Zhang C, Rannala B, Yang Z (2012) Robustness of compound Dirichlet priors for Bayesian inference of branch lengths. Systematic Biology 61: 779–784.

Zheng Y, Peng R, Kuro-o M, Zeng X (2011) Exploring patterns and extent of bias in estimating divergence time from mitochondrial DNA sequence data in a particular lineage: a case study of salamanders (order Caudata). Molecular Biology and Evolution 28: 2521–2535.

Zuckerkandl E, Pauling L (1965) Molecules as documents of evolutionary history. Journal of Theoretical Biology 8: 357–366.

Zwickl DJ (2006) Genetic algorithm approaches for the phylogenetic analysis of large biological sequence datasets under the maximum likelihood criterion.

345 PhD Dissertation, The University of Texas at Austin.

Zwickl DJ, Holder MT (2004) Model parameterization, prior distributions, and the general time-reversible model in Bayesian phylogenetics. Systematic Biology 53: 877–888.

346