Ecological and Evolutionary Drivers of Biodiversity and Extinction Risks in Amphibians and Squamates

by João Filipe Riva Tonini

Licentiate in Biological Sciences, October 2008, Universidade Federal do Espírito Santo B.S. in Biological Sciences, October 2009, Universidade Federal do Espírito Santo M.S. in Biological Sciences, February 2011, Universidade Federal do Espírito Santo

A Dissertation submitted to

The Faculty of The Columbian College of Arts and Sciences of The George Washington University in partial fulfillment of the requirements for the degree of Doctor of Philosophy

August 31, 2017

Dissertation directed by

R. Alexander Pyron Robert F. Griggs Assistant Professor of Biology The Columbian College of Arts and Sciences of The George Washington University certifies that João Filipe Riva Tonini has passed the Final Examination for the degree of

Doctor of Philosophy as of July 18, 2017. This is the final and approved form of the dissertation.

Ecological and Evolutionary Drivers of Biodiversity and Extinction Risks in Amphibians and Squamates

João Filipe Riva Tonini

Dissertation Research Committee:

R. Alexander Pyron, Robert F. Griggs Assistant Professor of Biology,

Dissertation Director

Amy Zanne, Associate Professor of Biology, Committee Member

Rayna C. Bell, Curator of Amphibians and Reptiles, Committee Member

© Copyright 2017 by João Filipe Riva Tonini All rights reserved

Dedication

To Lari and Caú.

Acknowledgements

I am immensely grateful to the Brazilian Government for the financial support of my studies and research through undergraduate, masters, and doctorate degrees (since 2004 to 2017). Additional support during the doctoral program was provided by The George

Washington University through a graduate teaching assistantship and Harlan fellowships.

My special thanks to my dearest mentor Dr. Alex Pyron, for sharing his lab, resources, and vast areas of expertise. Alex have been always supportive and keen to help me thinking on meaningful research questions. He motivated and pushed to pursue my own research interest and supported my choices. Thank you for the friendship and great lunch times in fancy places!

I am thankful to current and past members of the Pyron Lab, in special to Cat

Hendry. Cat helped in every single thing that I ever had to do in the graduate program, we had great times over beers, and she is the greatest! Thanks as well to Sara Green, Sara

Rhodig, and Tim Colston.

Thanks to all members of my graduate (g) and dissertation (d) committee: Amy

Zanne (g), Guillermo Orti (g), Kelly Zamudio (g), and Rayna Bell (d). Their guidance and wiliness to help improved the quality of the dissertation and manuscripts. My special thanks to Amy and Guillermo for encouraging on my research interests, sharing their lab infrastructure, and for great moments hiking and enjoying a good meal with caipirinhas.

Thank you to Hartmut Doebel, Robert Donaldson, and Tara Scully for guidance during the teaching assistantships.

I am indebted to colleagues, and their institutions, for sharing resources and expertise: Roy McDiarmid, Kevin de Queiroz, Addison Winn, and Jeremy Jacobs

(Smithisonian National Museum of Natural History); Felipe Toledo and Sandra Goutte

(Universidade de Campinas); Frank Burbrink and David Kizirian (American Museum of

Natural History); Karen Beard (Utah State University); Walter Jetz (Yale University). I am also grateful to Adam Wong and Glenn MacLachlan in the high-performance computer cluster of GWU (Colonial One) for helping to troubleshoot the analyses of my doctorate research.

I wish to thank Rafael de Sá (University of Richmond) for opening the doors of his lab and support me to be able to compete for a position in a doctoral program in the United

States. Thanks Rafael for sharing the excitement about frogs, supporting my research interests, for the friendship, and great trips to amazing places!

I was fortunate to enjoy the time during the graduate program with many great friends whom Lari and I shared great moments, laughs, and delicious food: Rodrigo

Ferreira and Cecília Weichert, David Santana and Janine Ziermann, Pedro Peloso and

Silvia Pavan, Dan Mulcahy and Bonnie Blaimer, Chuy and Belén Chávez, Thiago Moreira and Sandra Lara, Drew and Jackie Thompson, Nick and Thayna Joice, Lenice and Oscar

Shibatta, Ligia Benavides, Maddison Anderson, Lily Hughes, Joe Stiegler, Bob and Laura

Kallal, Drew Moore, David Stern, Amy Millo, Ricardo Betancur and Diana Arcila, Karen

Poole, Dominic White, Aidan, Jimmy Munoz, Joana Mooney, Carol Perez, Tiffini Smith,

Amulya Yarpala. My special thanks to Rodrigo for the friendship and collaboration of many years, it has been reinvigorating enjoy times with you dear friend.

At last, I thank to my parents (Eduardo and Neide), my brother (Luiz Guillerme), and my beloved Lari whom is my greatest motivation to pursue my dreams. Thank you for our beautiful son, Caú.

Abstract of Dissertation

Ecological and Evolutionary Drivers of Biodiversity and Extinction Risks in Amphibians and Squamates

Amphibians and squamates are together the most diverse clade of terrestrial vertebrates, and their biodiversity is thought to reflect many of the most important biogeographic and ecological forces that have generated species richness through time. To understand patterns of relatedness and the drivers of diversification, phylogenetic trees are instrumental to estimate the relative contribution of evolutionary and ecological processes.

My dissertation comprises four chapters related to phylogenetic patterns of biodiversity in amphibians and squamates. First, I tested across the frog Tree of Life whether species could escape the constraint imposed by body size on sound frequency and evolve new types of calls. The results show that frogs have multiple shifts in body-size allometry for calls.

These shifts comprise species endemic to hyper-diverse regions such as Africa, Australia,

New Guinea, Southeast Asia, and the Neotropics. Those shifts seem to reflect biogeographic invasions and instances of ecomorphological escape. Second, I asked what drives species co-occurrence and community assembly in Neotropical frogs? I find that the composition of most communities has been generated by stochastic variation of speciation, local extinction, and colonization rates. Thus, Neotropical ecoregions comprise distinct assemblages of frogs, as demonstrated by several regionalization studies, but community assemblages are a random sample of the regional pool. Third, I asked how many times has phytotelm-breeding evolved across Neotropical frogs, and do these lineages ever revert to

pond- or stream-breeding? If not, is phytotelm-breeding an evolutionary dead end, preventing diversification and raising extinction rates? I find that the history of phytotelm- breeding is labile, with support for at least 68 potential origins and 107 reversals. There is some support for state-dependent extinction (higher for phytotelm-breeding lineages), but

I cannot uniformly reject state-independent models. Fourth, I asked whether extinction risk in squamates is clustered evolutionarily and whether high-risk species represent a disproportionate amount of total evolutionary history. I found currently assessed threat status to be phylogenetically clustered at broad level in Squamata, suggesting it is critical to assess extinction risks for close relatives of threatened lineages. There is no association between threat and distinctiveness, suggesting that extinctions may not result in a disproportionate loss of evolutionary history. The results show that immediate efforts should focus on geckos, iguanas, and chameleons, in Amazon, Southeast Asia, and New

Guinea.

Table of Contents

Dedication iv

Acknowledgments v

Abstract of Dissertation vii

List of Figures xii

List of Tables xvi

Chapter 1: Evolutionary variation in allometric constraints on call evolution in frogs suggests ecomorphological escape 1

Abstract ...... 1

Introduction ...... 2

Material and Methods ...... 6

Phylogenetic and comparative data ...... 6

Statistical modeling ...... 9

Results ...... 11

Diversity and sound frequency and body size across frogs ...... 11

Regime shifts in acoustic allometric scaling ...... 12

Discussion ...... 19

Future directions ...... 22

Chapter 2: Transitions to phytotelm-breeding in Neotropical frogs are common, but may increase extinction 25

Abstract ...... 25

Introduction ...... 26

Material and Methods ...... 31

Data sampling and ancestral state reconstruction ...... 31

Macroevolutionary models ...... 32

Summary phylogenetic metrics ...... 34

Results ...... 36

Ancestral states reconstruction ...... 36

Macroevolutionary models ...... 39

Summary phylogenetic metrics ...... 41

Discussion ...... 42

Chapter 3: Neutral biogeographic processes primarily drive phylogenetic structure of Neotropical frog communities 47

Abstract ...... 47

Introduction ...... 48

Neotropical frog communities ...... 50

Material and Methods ...... 53

Phylogenetic and community data ...... 53

Null assembly models ...... 58

Sensitivity analysis ...... 61

Results ...... 62

Community dataset & Assembly models ...... 62

Sensitivity analysis ...... 66

Discussion ...... 68

Future directions ...... 71

Chapter 4: Fully sampled phylogenies of squamates reveal evolutionary patterns in threat status 73

Abstract ...... 73

Introduction ...... 74

Material and Methods ...... 76

Threatened vs non-threatened species ...... 76

Fully-sampled phylogeny of Squamates ...... 78

Phylogenetic measures of biodiversity ...... 80

ED and extinction risks ...... 82

Results ...... 83

Phylogenetic patterns of threat status ...... 83

ED and extinction risks ...... 88

Discussion ...... 89

References 93

List of Figures

1 Regression of body size and Dominant frequency across 12

2 Multiple regime shifts in acoustic allometry across the frog Tree of Life 14

3 Variation in Dominant frequency and body size across frog families and

species in regime shifts 17

4 Densities of intercept and slope estimates in the acoustic allometric scaling 18

5 Character reconstruction under the best fit macroevolutionary model (Mk2) 37

6 Stochastic mapping showing uncertainty and ambiguous reconstruction of

deep branches 38

7 Phylogeny of Neotropical frogs with tip states of endemic species color by

ecoregion and map of 555 local communities across ecoregions 55

8 Density plots representing PSV distribution of the 50 communities analyzed

under the DAMOCLES model 63

9 Phylogenetic structure of the 50 most diverse communities of Neotropical

frogs 64

10 Variation across Neotropical ecoregions of z-scores, local extinction (µ),

colonization (γ0), colonization decline (γ1) rates estimated under the

dynamic model of community assembly 65

11 Regressions of z-scores and standardized log(species richness) 67

12 Fully-sampled phylogeny of 9,755 Lepidosauria species with branches color

by ED values 86

13 Fully-sampled phylogeny trimmed to 3,296 Lepidosauria species assessed

by IUCN with branches color by ED values 87

14 Fully-sampled phylogeny trimmed to 783 Lepidosauria species classified

as threatened with branches color by ED values 88

15 Variation in ED values (log[Ma]) across 9,755 Lepidosauria species 89

List of Tables

1 Summary of macroevolutionary models and model selection 40

2 Speciation, extinction, and state transition rate parameters estimated under

the seven macroevolutionary models tested 40

3 Significance of summary statistic metrics comparing observed data with

simulated tip states under the seven macroevolutionary models test 42

4 Neotropical ecoregions used to delimit the regional species pool of frogs 56

5 Summary of the IUCN data and reference taxonomy of the Reptile Database 78

6 Phylogenetic Species Variability (PSV) results 84

7 Phylogenetic Species Cluster (PSC) results 85

Chapter 1: Evolutionary variation in allometric constraints on call evolution in

frogs suggests ecomorphological escape

Abstract. In sexually reproducing animals, species-specific acoustic signals for mating represent pre-zygotic boundaries that mediate species recognition and speciation. Across animals that use acoustic signal for communication, body size explains a large proportion of the sound-frequency diversity, and most groups tend to share an allometric scaling relationship. In general, larger animals tend to produce lower-pitched sounds and smaller animals tend to produce higher-pitch sounds. Here, we used frogs as a model system, as they emit vocalizations for mating that are species-specific, innate, and conspicuous, to attract females. Thus, frequency of advertisement calls has high phylogenetic signal. In contrast, sexual and natural selection might counteract the constraint impose by body size on sound frequency. We tested whether species could escape the general constraint imposed by body size on sound frequency, and evolve to new phenotypic optima different from the one shared by most species of frogs. Our goal is to identify the location, number, and magnitude of regime shifts in acoustic allometry scaling in frogs. Our dataset comprises information on size, frequency, and calling site for 2,176 species from 293 genera, and 42 families distributed worldwide. We find that frogs underwent multiple shifts in intercept and slope of the acoustic allometric scaling between body size and sound frequency. Shifts occur at deep phylogenetic scales and apparently related to major biogeographic invasions such as poison frogs in the Neotropics and Myobatrachridae in

Australasia, and ecomorphological transitions such as acoustic adaptation to noisy environments and transitions from terrestrial to arboreal habitats. Most shifts represent

diverse clades endemic to tropical areas such as Africa, Australia, New Guinea, Southeast

Asia, and South America. Though colonization of new areas and changes in ecomorphology seem to drive these adaptations, it is unlikely that a single process explains all shifts. Our results provide a platform for studies aiming to test specific hypotheses about the role of eco-evolutionary drivers of the complex relationship between physiology, sexual signaling, ecomorphology, biogeography, and community assembly.

Introduction

In sexually reproducing animals, species-specific acoustic signals for mating represent pre-zygotic boundaries that mediate species recognition and speciation. Body size represents a major constraint on sound production in vertebrates that use vocalizations for communication (Fletcher 2004). Larger animals typically have larger larynxes or other sound-producing organs, longer vocal folds that oscillate at lower frequencies, and longer vocal tracts that produce lower resonances (Martin 1972; Nevo and Schneider 1976; Ryan

1988). Thus, sound frequency is usually inversely proportional to body size, i.e. the larger the individual the lower the pitch or frequency of the sound, which is known as acoustic allometry (Stebbins 1983; Bradbury and Vehrenkamp 1998; Fletcher 2004). Typically, the constants of these relationships (i.e., the slope and intercept) are relatively conserved evolutionarily and shared across a wide range of species (Gringas et al. 2013a; Rodríguez et al. 2015a).

Changes in biological scaling are typically associated with major evolutionary transitions (e.g., Uyeda et al. 2017). In animals that use sounds for communication, these might represent new physiological vocalization mechanisms (e.g., syrinx vs. larynx),

occupancy of new habitats (e.g., Ryan and Brenowitz 1985), or behavioral changes related to sexual selection (Charlton and Reby 2016). Within lineages, we might expect finer-scale variation in allometric scaling relationships associated with these factors (Ohmer et al.

2009), but such studies have rarely been carried out in a phylogenetic context across a diverse group. Here, we are interested in how the scaling relationship between body size and sound frequency in frog calls is related to their evolutionary history, and when and where shifts may have occurred. These changes in acoustic scaling could be interpreted as discrete shifts among Simpsonian adaptive zones, in which rates of trait divergence are predicted to decrease through time as lineages accumulate and niches associated with these

“adaptive zones” become saturated (Simpson 1944, 1953).

Frogs are known for their extraordinary diversity in vocalizations for mating and body sizes, among other traits (Narins et al. 2007; Wells 2007). Frog advertisement calls have been a model system to study species communication, given their biological significance and relative simplicity (Blair 1964; Ryan 1980; Gerhardt and Huber 2002;

Hoskin et al 2009). While many frogs can emit several types of sound, in most species males emit vocalizations that are species-specific, innate, and conspicuous, to attract females during the breeding season (Wilczynski and Ryan 1988). Furthermore, the female tympanum is tuned to recognize species-specific advertisement calls (Capranica 1965).

Thus, frog advertisement calls have high phylogenetic signal (Gringas et al. 2013b) relatively to birds and mammals that have higher learning capability and can often produce multiple sounds that might not reflect long-term evolutionary relationships (Raposo and

Höfling 2003). In frogs, frequency of advertisement calls is correlated to body size, as expected (Gringas et al. 2013a).

Numerous factors might affect how vocalizations vary with body size in frogs.

Background noise affects both sound transmission and its content integrity, which hinders sound detection and consequently the receivers’ ability to decode and react to the signal

(Morton 1975). For instance, some frog species breeding in noisy environments such as fast-flowing streams and torrents have adapted to produce ultrasonic (e.g., Huia cavitympanum; Arch et al. 2008) or multimodal (e.g., Hylodes asper; Haddad and Giaretta

1999) signals for communication. Furthermore, frogs living in close forest and open habitats have been shown to have distinct call characteristics (Schiøtz 1967). Moreover, in hyper-diverse communities with mixed-species choruses, differences in calls can facilitate avoidance of acoustic interference by using different frequency bands (Straughan 1973).

Then, acoustic segmentation would allow species to coexist in hyper-diverse communities

(e.g., Amézquita et al. 2011). Alternatively, sounds of sympatric species may cause masking interference in one another and could drive changes in sound frequency as well

(e.g., Amézquita et al. 2006).

Shifts in acoustic allometric scaling may have occurred deep in the evolutionary history of frogs related to the invasion of new geographic areas (e.g., Neotropics, Australia,

New Guinea) or the evolution of new ecomorphologies affecting body size, adaptive shifts which are well-known in frogs (Van Bocxlaer et al. 2010; Moen et al. 2013; Vidal-García et al. 2014). Changes from aquatic to arboreal or arboreal to terrestrial habit have been observed in many frogs (e.g., Ranidae and Hylidae) impacting the evolution of body size and affecting diversification rates (Moen et al. 2013; Moen et al. 2017). Moreover, turnover in net diversification (speciation minus extinction) across frogs (Feng et al. 2017) might be related to phenotypic disparity in characters such as body size and sound frequency given

that these traits are related to many physiological and ecological aspects of species niche

(Wells 2007). Finally, shifts may be rapid and drastic, with short-term changes related to microhabitat usage, strong sexual selection, or simply high rates of evolution in labile behavioral traits (e.g., Blomberg et al. 2003), which erase phylogenetic signal (but see

Goicoechea et al. 2010; Gringas et al. 2013b).

Based on these earlier work, we might expect the physical structure of environments, ambient noise, species interactions (e.g., masking interference, reproductive character displacement), changes in ecomorphology, and turnover in diversification rates to exert pressure on phenotypic optima of sound frequency. In addition, these forces might have concomitant influence on body size selecting species to stay in the same acoustic allometric scaling as their close relatives, or select species to change to a new scaling relationship different from the one shared by most frogs. However, it is unclear how tightly coupled variation of body size and dominant frequency are across the frog Tree of Life.

Frogs worldwide might share one phenotypic optimum (or adaptive regime) for the intercept and slope of the correlation. Alternatively, ecological and evolutionary processes might have driven frogs to call using different frequency bands than closely related lineages or having a distinct scaling relationship with body size.

Thus, we can test in a phylogenetic framework (Uyeda et al. 2017) when and where lineages shift to a new intercept of call frequency, slope, or both, for the evolutionary correlation between body size and sound frequency. Here, we use the frog Tree of Life to ask whether lineages have shifted to acoustic allometric scaling regimes distinct from the one shared by most species included in the present study. Since changes in ecomorphology could also have an effect on both call frequency and body size, we included data on calling

site as well. Multiple ecological and evolutionary mechanisms are expected to generate shifts in acoustic allometric scaling and shifts can happen at any phylogenetic scale, thus we used an approach does not make a priori assumption on taxonomic level or ecological aspect that should lead to shifts in the evolutionary correlation of body size and sound frequency. We used data for 2,176 frog species, comprising 293 genera and 42 families distributed worldwide. Our goal is to identify the location, the number, and magnitude of shifts in regimes of acoustic allometry scaling in frogs.

Our results show significant regime shifts in acoustic allometric scaling in intercept, in slope, and in both, in lineages geographically distributed in hyper-diverse Tropical regions such as Africa, Australia, New Guinea, Southeast Asia, and the Neotropics.

However, these are relatively deep in the evolutionary history of frogs, and appear to be related to major biogeographic and ecomorphological transitions. Within lineages, most species share a common scaling relationship. Our findings provide the framework to focus on frog lineages that have had unique evolution of acoustic allometry scaling given the phylogenetic history. Future studies should focus on local community processes, physiological and morphological aspects to disentangle the contribution of ecological processes to the pattern of body size and frequency of acoustic signals in extant frogs.

Material and Methods

Phylogenetic and comparative data

Advertisement calls are among the most energetically expensive activities in vertebrates (Wells 2001). During the breeding season, calling males increase rates of oxygen consumption and may experience energetic costs 10 to 25 times higher than those

individuals at rest (Pough et al. 1992; Wells 2001). Moreover, larger males are known to have higher survivorship and fertilization rates, thus frequency of advertisement calls transmits information on fitness benefits as well (Wells 2007). The dominant frequency of these vocalizations represents the highest peak in energy (or amplitude), and matches the peak sensitivities in the ear of conspecific females (Capranica 1965; Nevo and Capranica

1985). Dominant frequency in particular is known to scale with size, as expected, but with some apparent variation across clades (Gringas et al. 2013a). Dominant frequency is an ideal variable for large-scale comparative acoustic analyses because in most frogs it has greater variation between species than within species, in fact it can vary over one order of magnitude among species at the level of anuran families and, unlike temporal variables, is not affected by ambient temperature (Köhler et al. 2017). Furthermore, it is a static, easy and uncontroversial call-characteristic to measure (Köhler et al. 2017).

We collated data from published manuscripts and books, commercial resources, and museum collections to gather data on mean dominant frequency of advertisement calls

(DF; the frequency with higher amplitude) and male snout-vent length (i.e., a measure hereafter referred as body size) for 2,176 species (32% of the total species-level frog diversity), in 293 genera (65% of frog genus-level diversity) and 42 families (77% of frog family-level diversity) distributed worldwide (AmphibiaWeb 2007). All data sources are provided in Appendix 1 at https://drive.google.com/file/d/0B-k1-

50WDeJZR1NEbWVfTTJWWnM/view?usp=sharing.

Related to the calls themselves, microhabitat and calling site in particular is a trait likely correlated with body size and DF, due to high competition for physical space and frequency bands for acoustic communication, and is typically well characterized for most

species (Salthe and Duellman 1973; Crump 1974; Jungfer and Weygoldt 1999). We classified frog species according to their preferred calling site to include in the macroevolutionary analyses. Frogs calling while floating or submerged were classified as

1; frogs perching on vertical surfaces, trees, and herbaceous vegetation were classified as

2; and frogs calling from the ground, leaf-litter, sitting on shallow pools, side of streams, rocks on streams were classified as 3 (e.g., Haddad et al. 2013; Rodríguez et al. 2015b).

Numerous studies have published data on advertisement calls and for the large majority data on DF were informed, we collected the data on mean DF from these publications (Appendix 1). However, for descriptions of species advertisement calls in which authors did not provide the value for mean DF, we requested from authors and natural history collections audio files of advertisement calls for those species and used it to extract the mean DF (Appendix 1). We used Audacity® v2.1.3 (Audacity Team 2017) to edit audio files, reduce effects of background noise, and sample the mean DF (e.g., Stucky

2013; Ferreira et al. 2015). Audio files were re-sampled at 44,100 Hz and 32-bit resolution.

We selected a portion of the audio without the advertisement call and obtained the background noise profile using the function Noise Reduction in Audacity. We tuned the settings of this function after preliminarily analysis of a large sample of audio files looking for values that would have minimal effect on the call amplitude by comparing the original with the edited audio file. The final parameter settings of the Noise Reduction function for extracting mean DF were noise reduction to 26 dB, sensitivity to 13, and frequency smoothing to 3 bands. We estimated the power spectrum of the advertisement calls using

Fast Fourier Transform and Hanning window function of size 512 samples, and the mean

peak frequency with higher amplitude of calls (e.g. dominant frequency) was extracted

(Appendix 1).

We used the phylogenetic dataset of Jetz & Pyron (in press), which contains 7,238 species of amphibian, taken from the 19 February 2014 edition of AmphibiaWeb. In all the analyses below, we used a representative tree from the PASTIS distribution of 10,000 fully sampled (7,238 species) trees, and the summary data-only (4,061 species) tree (annotated code as Supporting Information). We used the fully sampled time-calibrated phylogeny of amphibians pruned to include the 2,176 frog species with information on body size, sound frequency, and calling site and the data-only including 1,610 species. Major taxonomic differences are in Leptodactylidae (41% less species in the data-only compared to the

PASTIS), Microhylidae (39% less species in the data-only), Hyperoliidae (38% less species in the data-only), Myobatrachidae (37% less species in the data-only), Hylidae

(31.5% less in the data-only), and Bufonidae (21% less species in the data-only).

Statistical modeling

We used a Bayesian phylogenetic framework implemented in the R (R Core Team

2016) package bayou 2.0 (Uyeda and Harmon 2014; Uyeda et al. 2017) to detect shifts in scaling regimes of the acoustic allometry relationship between DF, body size, and calling site across frogs. The bayou model uses a reversible-jump (rjMCMC) algorithm that jointly estimates the location in the phylogeny, number, and magnitude of shifts in adaptive optima

(Uyeda and Harmon 2014). bayou does not make a priori assumption on the number of shifts and the location in the phylogeny where adaptive shifts are expected to occur. This method has several advantages over similar approaches. Importantly, the reversible-jump

framework produces a full posterior of credible models and parameter values, and, therefore, incorporates uncertainty in the number of shifts, placement in the phylogeny, and parameter estimates (Uyeda and Harmon 2014).

We log-transformed values of DF and body size to achieve normality and fit a linear model (e.g., Uyeda et al. 2017). The regression models explored were 1) θDF ~ βsvl and 2)

θDF ~ βsvl + βsit + βsvl*sit. In bayou, we estimated the intercept (θ; the optimum value of DF in the Ornstein-Uhlenbeck process), slope (β) of the linear relationship between sound frequency (θDF), body size (βsvl), and calling site (βsit), per-unit-time magnitude of the uncorrelated diffusion (σ2), the rate of adaptation (α), and the number of selective optima

(K).

For the prior on the slope, we set the mean to zero (i.e., a flat prior), following a normal distribution with 0.2 standard deviation. For the prior on the intercept, we used the mean ln(DF) and 1.5 times the standard deviation of the data on ln(DF). The proposal schemes used in bayou for identifying regime shifts are derived from Uyeda et al. (2017).

We ran two independent MCMC chains of 13,333,334 generations with different starting random seeds, sampling every 1,000th generation, which after a burn in of 25% yielded a distribution of ten thousand samples of the posterior probability. We checked for convergence between the two chains using R Gelman (Gelman and Rubin 1992) by comparing the posterior probabilities of branches. Effective Sample Size (ESS) higher than

100 were considered suitable (see Supplemental Information).

Shifts with posterior probabilities higher than 0.95 were considered adaptive shifts in acoustic allometry regime (e.g., Uyeda et al. 2017). Although this is a conservative threshold compared to previous examples using the bayou model (e.g., Uyeda and Harmon

2014; Cuff et al. 2015; Uyeda et al. 2017), it has been a standard threshold in Bayesian statistics when considering results as having high support. We believe this is an important value given the potential issues of power for comparative methods in general using these models (Ho and Ané 2014; Cooper et al. 2016), as well as the PASTIS trees in particular

(Rabosky 2015; Title and Rabosky 2016), and so regard this high cutoff as more conservative that what has previously been employed. In Bayesian rjMCMC Ornstein-

Uhlenbeck models, the priors can have a large influence on the results (e.g., Ho and Ané

2014). Therefore, we ran the Bayesian estimation procedure with no data to check the mean number of shifts a priori. We provided an annotated code, convergence plots, and model comparisons as Supporting Information.

Results

Diversity of sound frequency and body size across frogs

Across the thousands of frogs included in this study, body size explains a large proportion of the frequency diversity in advertisement call of extant species (R2 = 0.52;

Fig. 1). The mean Dominant frequency of species sampled ranges from 200 Hz (e.g.,

Chiromantis kelleri and Heleioporus barycragus) to 16,590 Hz (e.g., Huia cavitympanum), and body size ranges from 8 mm (e.g., Brachycephalus hermogenesi) to 165 mm (e.g.,

Pyxicephalus adspersus). The species Brachycephalus hermogenesi has the greatest ratio of Dominant frequency to body size (850 Hz:1 mm) while Pyxicephalus adspersus represents the smallest one (1.35 Hz:1 mm).

Figure 1. Linear regression of log body size and log Dominant frequency (in red) and

Phylogenetic Generalized Least Squares (dotted gray) across 2,176 frog species included in the present study.

Regimes shifts in acoustic allometric scaling

In general, the results using the fully-sampled and data-only phylogenies agree with each other. The differences are that Southeast Asian microhylids and Neotropical hylodids were recovered as regime shifts in the PASTIS but not in the analyses using data-only phylogeny. These groups have fewer species represented in the data-only phylogeny relatively to the PASTIS, they were excluded in the data-only analyses because we lack molecular data for them. Since the results of the PASTIS and data-only largely agree, and in the PASTIS analyses, we can use all trait data collected, we present the results of the fully sampled phylogenies.

The model including only DF and body size had stronger support (βsvl; marginal log likelihood = -779.46) over the model including calling site and interaction between body size and calling site (βsvl + βsit + βsvl*sit; marginal log likelihood = -845.97). All shifts identified in the model βsvl were also present in the βsvl + βsit + βsvl*sit model, however in the latter we did not recover the Southeast Asian bufonids and within Australiasian myobatrachids only Heleiopours were recovered as part of the shift.

Our findings of the best-fit model (θDF ~ βsvl) show that frogs exhibit ten distinct acoustic allometric scaling regimes. Out of those, one regime shift from the background correlation corresponds to a singleton species imputed to the phylogeny (Chiromantis kelleri). We conservatively interpret the results to support eight major shifts in the acoustic scaling of sound frequency and body size across the frog Tree of Life, that are distinct from the background evolutionary correlation shared by most species (Fig. 2).

Figure 2. Multiple regime shifts in acoustic allometry across the frog Tree of Life. a) Values on the y-axis represents intercepts of ln Dominant frequency (Hz) and x-axis represents male ln Snout-Vent length in mm (used here as a synonym of body size). Colors represent distinct acoustic allometry regimes identified by the bayou model with posterior probabilities higher than 0.95. b) regime shifts in acoustic allometry mapped to the frog phylogeny. In gray, the background correlation of body size and sound frequency shared by most species; red, Papua New Guinean microhylids, (3) Xenorhina oxycephala; pink,

Southeast Asian ranids, (5) Huia cavitympanum (Ranidae); yellow, Australasian

myobatrachids, (6) Limnodynastes terraeginea; cyan, Neotropical hylodids, (8) Hylodes lateristrigatus; green, Southeast Asian bufonids, (2) Peltophryne misera; blue, Neotropical odontophrynids, (1) Proceratophrys laticeps; orange, Neotropical dendrobatids and aromobatids, (7) Epipedobates tricolor; purple, African pyxicephalids, (4) Pyxicephalus edulis.

The majority of species (1,928 species out of the 2,176 included in this study) share the same slope and intercept of acoustic allometric scaling (θroot = 8.8, βroot = -0.31), as shown in gray in Fig. 2a. However, frog lineages endemic to hyper-diverse regions such as

Africa, Southeast Asia, Australia, New Guinea, and the Neotropics were estimated to represent regime shifts in acoustic allometry scaling. Specifically, frogs with similar slope as root regime but with lower intercepts indicate that species produce lower frequency sounds for a given body size, and they are found in the African genera Pyxicephalus and

Aubria (Pyxicephalidae; Fig. 2 in purple), in the Australasian Myobatrachidae

(Limnodynastes, Notaden, Philoria, Neobatrachus, Platyplectrum, Heleioporus; Fig. 2 in yellow), in the New Guinean Microhylidae (Xenorhina and Asterophrys; Fig. 2 in red), and in the Neotropical Odontophrynidae (Proceratophrys, Odontophrynus,

Macrogenioglottus; Fig. 2 in blue). In contrast, frogs with similar slope as the root regime but higher intercepts indicate that species produce higher frequency sounds for a given body size such as in the Neotropical Dendrobatidae and Aromobatidae (Hyloxalus,

Ranitomeya, Andinobates, Excidobates, Oophaga, Dendrobates, Adelphobates,

Phyllobates, Ameerega, Colostethus, Silverstoneia, Epipedobates, Mannophryne,

Aromobates, Anomaloglossus, Rhaeobates, Allobates; Fig. 2 in orange).

Furthermore, we recovered species that have distinct slopes than the root regime in the acoustic allometric relationship. Southeast Asian frogs in the genera Meristogenys and

Huia (Ranidae; Fig. 2 in pink), and in Ansonia and Pelophryne (Bufonidae; Fig. 2 in green), as well as in the Neotropical Hylodidae (Hylodes and Crossodactylodes; Fig. 2 in orange) have shallower slopes than the root regime and higher intercepts. This indicates that these species have higher frequency sound for a given body size at larger body sizes (Fig. 2a), as these species are around values of 3.5 in the x-axis of Fig. 2a.

Species of microhylids, myobatrachids, pyxicephalids in the regime shifts tend to have larger body size and lower frequencies than remaining species in those respective families (Fig. 3). In contrast, ranids and bufonids species in the regime shifts tend to have smaller body size and high frequency calls (Fig. 3). In figure 3, the boxplots of body size and sound frequency for hylodids, odontophrynids, and dendrobatids overlap completely because the families as a whole represent regime shifts (Fig. 3).

Figure 3. Variation in dominant frequency and body size across frog families and species in regime shifts. Boxplots in colors represent trait values for species in the regime shifts and each gray box on the left of each color box are the values for species in the families sharing the background regime. In red, Papua New Guinean microhylids; purple, Sub-

Saharan pyxicephalids; pink, Southeast Asian ranids; yellow, Australian myobatrachidae; cyan, Neotropical hylodids; green, Southeast Asian bufonids; blue, Neotropical odontophrynids; orange, Neotropical dendrobatids.

The root regime, dendrobatids, and odontophrynids show great variation in intercept and slope, which are not explained by body size or calling site (Fig. 4). This may represent additional variation which our analyses were too insensitive to recover at present.

Figure 4. Densities of intercept and slope estimates in the acoustic allometric scaling. In gray is the correlation of body size and sound frequency shared by most frog species. In red, Papua New Guinean microhylids; pink, Southeast Asian ranids; yellow, Australian myobatrachidae; cyan, Neotropical hylodids; green, Southeast Asian bufonids; blue,

Neotropical odontophrynida; orange, Neotropical dendrobatids. Frog images: s,

Asterophrys turpicola; t, Pyxicephalus edulis; u, Meristogenys kinabaluensis; v,

Limnodynastes peronii; w, Crossodactylus timbuhy; x, Ansonia spinulifer; y,

Proceratophrys paviotti; z, Oophaga pumillio.

Discussion

This is the first large-scale phylogenetic study of call evolution across the frog Tree of Life, and as such, offers an important but only preliminary view on pattern, and to some extent process. Overall, the vast majority of frog species appear to share a single scaling regime for acoustic allometry, at least as far as dominant frequency (the most energetically expensive portion of the call). However, at least eight major shifts, representing potentially distinct ecomorphologies or biogeographic invasions, have occurred in the frog Tree of

Life. The reasons for these shifts appear to be complex, but may be related to biogeographic events and ecomorphological shifts in broad-scale habitat usage. Although calling site in particular do not describe the general pattern, at least as classified here, it could have driven individual shifts (e.g., ranids and bofunids).

In the present study, we identified as regime shifts frog clades already known to have unique properties of body size and sound frequency, such as species of Huia that produce ultrasonic calls (Arch et al. 2008). Our results show hylodids are also associated with a different acoustic allometric scaling relationship, in addition produce multimodal behavioral signals including foot flagging (Haddad and Giaretta 1999; Narvaes and

Rodrigues 2005; Caldart et al. 2014; de Sá et al. 2016; Malagoli et al. 2017) and facultative inflation of vocal sacs (de Sá et al. 2016). Species from both these regime shifts have been model organisms for studies investigating the effects of abiotic ambient noise on species communication (e.g., Goutte et al. 2016; Röhr et al. 2014). Thus, they can be seen as good

checkpoints to show that the bayou model is identifying clades that we already know a priori to have unique relationships between body size and dominant frequency. For instance, one could hypothesize that frog species calling in ultrasound but not included in the present study would represent a regime shift in acoustic allometric scaling as well.

Several of the regime shifts represent subsets of much larger radiations (e.g., Feng et al. 2017), the remainder of which share the background regime in acoustic allometry scaling. For instance, Ranidae, Bufonidae, and Microhylidae contain a great number of species distributed worldwide, while Myobatrachidae and Pyxicephalidae have a wide distribution and large diversity restricted to Australasia and Sub-Saharan Africa, respectively. Independent invasions of the same biogeographic areas (see Pyron 2014), and potentially ecomorphological shifts (e.g., Moen et al. 2013), are associated with multiple independent regime shifts across the frog Tree of Life. These range in age from 32.7 Ma

(e.g., Southeast Asian microhylids) to 102.7 Ma (e.g., Neotropical dendrobatids) (mean =

51.3 Ma, standard deviation = 22.31 Ma), suggesting that episodes of ancient vicariance, dispersal, and diversification (see San Mauro et al. 2005; Roelants et al. 2007; Pyron 2014) appear to represent major episodes for the evolution of new adaptive regimes in vocalization. In particular, no young or recently derived, diverse clades appear to represent significant shifts. However, it is unlikely that a single process is responsible for driving all shifts identified, especially as we recovered differences in slopes for some shifts (e.g., ranids, hylodids, and bufonids). Selection on either body size or DF may also have independent effects that are difficult to disentangle in different environments.

For at least three shifts, in Neotropical Hylodidae, some Southeast Asian Ranidae and Bufonidae, the change in allometric scaling seems to result clearly from specialized

adaptations to higher-frequency calls in noisy environments. Although our results show that the model including calling site was not selected, these results suggest that calling site could be a factor in driving allometry shifts in some clades as being an important determinant of shifts. We note qualitatively that ecomorphological differences in different biogeographic regions drive the shifts at least in part, such as change from terrestrial to arboreal habit in ranids and bufonids; clades with generally larger body size, terrestrial species. Alternatively, the shifts in regime may not be related directly to biogeographic or ecomorphological changes, but to idiosyncratic physiological or morphological adaptations that have subtle origins which are difficult to detect in a large-scale analysis such as this. For instance, all shifts are in tropical clades, despite the presence of highly specialized, ancient lineages such as Ascaphidae in temperate areas, which also breed in noisy environments. Whether or not similar, as-yet undetected historical scenarios have affected the other five shifts remains to be seen. Historical contingency in the microhabitat or general calling environment of myobatrachids, odontophrynids, dendrobatids, microhylids, and pyxicephalids may have driven similar adaptations. As noted in

Dendrobatidae, both the slope and intercept appear to be multimodal, and this group represents species distributed widely across biomes and microhabitats in Central and South

America, each of which may be driving local regime shifts.

All of the shifts represent clades that are unique in at least some respects from their close relatives. However, across the frog Tree of Life, ecomorphological diversity is extremely high, and there are numerous instances of both habitat and biogeographic transitions that are not apparently related to regime shifts in acoustic allometry scaling. For instance, Mantellidae, Craugastoridae, and Rhacophoridae represent large radiation in

distinct continents with great diversity of ecomorphology but were not recovered as regime shifts in acoustic allometric scaling. Part of this uncertainty in not recovering major clades known to represent biogeographic invasions and/or changes in ecomorphology may also be related to sampling. While we have included a large proportion of families and genera, we have still only sampled about a third of the extant diversity of frogs. Indeed, both the background regime and the regime shift of dendrobatids appear to show multimodality in both slope and intercept (Fig. 4). Thus, increasing sampling may result in more highly supported regime shifts corresponding to these multiple modes. However, we have included many of the most ancient and distinctive frog lineages, such as Alytidae, Pipidae,

Rhinophrynidae, Scaphiopodidae, Megophryidae, Pelobatidae, Heleophrynidae, and

Calyptocephalidae as well as a large proportion of both temperate and tropical species from all currently-inhabited biogeographic realms (Appendix 1). We acknowledge the absence of some highly distinctive lineages in terms of biogeography and ecomorphology, such as

Sooglossidae and Nasikabatrachidae, and whether those lineages represent shifts in acoustic allometry scaling remains to be tested.

Future directions

We can hypothesize at least one future test of the specific hypothesis that convergence in ecomorphology for broad-scale habitat use affects allometric scaling, and potentially explains the observed acoustic regime shifts. The Southeast Asian bufonids

Ansonia and Pelophryne represent a significant shift within Bufonidae. In addition to breed in fast-flowing streams, these species are smaller and more arboreal than the majority of the broader radiation of toads, which tend to be larger, terrestrial, and robust (Pramuk et al.

2008; Van Bocxlaer et al. 2010). Highly similar ecomorphs are observed in the Neotropical toads Dendrophryniscus and Amazophrynella, i.e., small size and mostly arboreal habits.

However, out of seven species of Amazophrynella we only had call data for one, and acoustic signals of Dendrophryniscus have not been described yet. As noted above, similar ecomorphs in disjunct regions been recovered as regime shifts (e.g. hylodids and ranids that breed in noisy environments were recovered as regime shifts), and we suggest that future studies examine the Dendrophryniscus/Amazonphrynella clade specifically, to test if it represents a similar regime shift to Ansonia/Pelophryne. If so, this would indicate another instance of apparent ecomorphological changes driving acoustic allometry regimes in separate continental radiations.

There are a number of future avenues that should be investigated in future studies.

These include other aspects of call spectra, such as fundamental frequency, and temporal variables such as call, pulse and note duration, interval between calls, number of pulses per note, number of notes per call. Furthermore, physiological and morphological studies to understand sound production may help elucidate the origin of shifts. The size, shape, and presumably function of vocal sacs varies significantly in frogs, and may have a strong effect. Additionally, increasing sampling to include more complete representation of species may increase the power, precision, and accuracy of these comparative analyses.

Finally, given the dominant role of sexual signaling and call evolution in speciation in frogs

(Narins et al. 2007), call evolution may have played a significant role in the explosive diversification of anurans worldwide. Future studies might examine whether shifts in call variables such as dominant frequency are associated with shifts in diversification rates through time, and across lineages. This could potentially explain at least part the drastic

disparities in observed species richness and estimated diversification regimes across the frog Tree of Life (Roelants et al. 2007; Pyron and Wiens 2013).

Chapter 2: Transitions to phytotelm-breeding in Neotropical frogs are common, but

may increase extinction

Abstract. Of the numerous life-history modes observed in frogs, few are as consequential or striking as the transition from pond- or stream-breeding to the deposition of eggs or larvae in phytotelmata, i.e. bodies of water contained in plants such as bromeliad tanks, tree holes, vine holes, leaf axils, husks of fruits, and bamboo internodes. Phytotelmata might increase offspring survivorship due to reduced mortality from competition, desiccation, and predation. These are relatively safe sheltered habitats for frog embryos, larvae, and adults, compared to other water bodies where competitors and predators are often abundant. In the Neotropics, over 150 species from 38 genera and ten families have specialized to reproduce in phytotelmata. However, few of these lineages appear to represent diverse adaptive radiations resulting from what otherwise might be assumed to represent a significant source of ecological opportunity. Here, we use a fully-sampled phylogeny of frogs and breeding-site data for 3,105 species of Neotropical anurans to answer three main questions. First, how many times has phytotelm-breeding evolved, and do these lineages ever revert to pond- or stream-breeding? Second, is there any support for state-dependent diversification models, suggesting that breeding site affects diversification rates? Third, is it possible that phytotelm-breeding represents an evolutionary ‘dead-end,’ which might be optimal in the short-term for some species, but increase extinction rates failing to promote long-term success. We find that the character history of phytotelm- breeding is ambiguous, with support for at least 68 potential origins and 107 reversals.

While macroevolutionary model fitting can be challenging, we find some support for the

model in which non-phytotelm-breeding lineages have a greater rate of transition to phytotelm-breeding. Furthermore, phytotelm-breeding lineages have higher extinction rates, but not speciation rate than non-phytotelm-breeding. However, we cannot uniformly reject state-independent models. This is corroborated by phylogenetic summary statistics designed to detect evolutionary ‘dead-ends,’ which yield weak but inconsistent support among state-dependent models overall. Thus, we conclude that breeding site is a highly labile character in frogs that changes frequently, but that the apparently significant transition from terrestrial breeding sites to phytotelm-breeding does not exert an unambiguous effect on diversification rates. However, the weak signal for the ‘Suicide’ model of phytotelm-breeding as evolutionary dead-end fits well with existing ecological theories, and deserves additional consideration.

Introduction

Adaptive radiations often result from new sources of ecological opportunity, such as key innovations, colonization of new areas, or extinction of competitors (Glor 2010).

Habitat selection, and habitats used for reproduction in particular, potentially represent both “new areas” and “key innovations” together, as maximizing mating success and offspring survival is a key component of individual fitness (Fretwell and Lucas 1970). In frogs, breeding habitat is a critical but limited resource due to high competition for space, frequency bands for acoustic communication, and risk of predation (Salthe and Duellman

1973; Crump 1974; Jungfer and Weygoldt 1999). Although most frog species breed and deposit clutches in standing water (e.g., ponds), streams, or on the ground, some frogs have

evolved alternative reproductive modes finding independence from these breeding habitats

(Magnusson and Hero 1991; Jungfer and Weygoldt 1999).

Frogs worldwide both having a larval stage (i.e., tadpoles) that goes through metamorphosis to assume the adult morphology and direct development (i.e., do not have a larval stage) have specialized to deposit clutches in water bodies accumulated in plants

(known as phytotelmata) such as bromeliad tanks, tree holes, vine holes, leaf axils, husks of fruits, and bamboo internodes (Lehtinen et al. 2004). Phytotelm-breeding might increase offspring survivorship due to reduced costs of trade-offs such as competition, desiccation, and predation. For instance, phytotelmata might represent relatively safe sheltered habitats for frog embryos, larvae, and adults, compared to other water bodies (ponds, streams, and ground) where competitors and predators are often abundant (e.g. Lehtinen et al. 2004;

Buxton and Sperry 2017). However, phytotelmata have a different physical structure than ponds and streams, lower nutrient levels, and are not as available across a landscape as these other breeding habitats.

To adjust to these constraints, obligate phytotelm-breeding frogs minimize competition by producing extremely small egg clutches and cannibalistic larvae with the effect that only one larva per phytotelmata survives (Weygoldt 1981; Giaretta 1996). Other species produce non-feeding larvae with enough yolk stored in their intestines to allow metamorphosis (Blommers-Schlösser 1975; Duellman and Maness 1980; Weygoldt and

Potsch de Carvalho e Silva 1991; Krügel 1993; Glaw and Vences 1994). Furthermore, the larvae of some frogs are known to feed on conspecific eggs, and several species have evolved parental care; their tadpoles are fed on eggs provided by the mother (Weygoldt

1980; Zimmermann and Zimmermann 1981; Jungfer 1985, 1987, 1996; Ueda, 1986; Meyer

1992; Brust 1993; Jungfer et al. 1996; Thompson 1996).

Phytotelm-breeding may give rise to relatively diverse clades of descendants that share this trait such as Andinobates (all 13 species), Oophaga (all 8 species), Ranitomeya

(all 16 species), Crossodactylodes (all 5 species), Dendrophryniscus (6 out of 10 species), and Phyllodytes (all 12 species). However, given the constraints in phytotelmata and high level of specialization shown by obligate phytotelm-breeding frogs to succeed in these habitats, species might colonize and specialize to breed in phytotelmata but most often do not give rise to large clades sharing this trait. This would suggest a very labile trait with higher state transition rates from phytotelm- to non-phytotelm-breeding (e.g., Bennet et al.

2013), but with potentially lower subsequent diversification rates.

Furthermore, this specialization of mating and larval or egg development in phytotelmata may limit phytotelm-breeding frogs’ capacity to adapt to a changing environment. For instance, local extinction of a given phytotelm resource used as breeding site could disrupt population dynamics of phytotelm-breeding species due to reduce ability of specialized species to use other habitats (e.g., Futuyma and Moreno 1988), which would result in reproductive failure and over time could lead to extinction. Thus, the shorter-term microevolutionary benefits of phytotelm-breeding may come at the cost of longer-term macroevolutionary success (e.g., Day et al. 2016; Bromham et al. 2016). Therefore, the evolutionary history of phytotelm-breeding might show high trait loss, higher rates of extinction, or lower rates of speciation than non-phytotelm-breeding frogs. As trait- associated changes in diversification rates can leave their mark on phylogenies we can test these hypotheses in a phylogenetic framework (e.g., Maddison et al. 2007).

If specialization to phytotelm-breeding leads to lower speciation rates, this could result in longer branches connecting terminal taxa (Day et al. 2016). In addition, if specialization of phytotelm-breeding causes high rates of extinction or frequent loss of this trait, these lineages are unlikely to persist for long enough to give rise to large clades of descendants (Bromham et al. 2016). Thus, we might expect phytotelm-breeding frogs to be scattered across the tips of the phylogeny, in shallow clades with recent origins, and the observed number of phytotelm-breeding would be relatively lower than the number of reconstructed origins of the trait (Bromham et al. 2016). We might also expect if this signature is present, it will lead to strong support for state-dependent extinction models

(e.g., Maddison et al. 2007; Bromham et al. 2016), though these must be interpreted with caution (Rabosky and Goldberg 2015).

Does phytotelm-breeding arise frequently, or only a few times? Can phytotelm- breeding lineages transition back to ponds or streams, or is it an evolutionary ‘dead end?’

Numerous putative ‘key innovations’ and ‘new areas’ have been suggested to alter diversification dynamics, such as parity mode in squamates (Pyron and Burbrink 2014) and toe pads in geckos (Gamble et al. 2012). The number of origins and potential for reversion in these types of traits is crucial to understanding their impact on evolutionary histories, but they are controversial given uncertainty associated to ancestral reconstruction models and low power to distinguish between hypotheses when using data on extant species alone

(King and Lee 2016; Harrington and Reeder 2017). If the extant diversity of phytotelm- breeders is explained by only a few origins accounting for most species, it would suggest that phytotelmata represent a significant source of ecological opportunity that has promoted adaptive radiation such as in Andinobates, Oophaga, Ranitomeya, Crossodactylodes,

Dendrophryniscus, and Phyllodytes. If phytotelm breeding arises about as many times as expected under a null model and also reverses frequently, this might indicate that breeding site does not exert a strong influence on diversification dynamics. If phytotelm-breeding arises frequently, rarely reverses, and rarely results in diverse radiations, this would suggest that strong links to water-holding plants for reproduction is potentially an evolutionary

‘dead-end.’ We can evaluate these hypotheses using phylogenetic comparative methods

(e.g., Maddison et al. 2007; Bromham et al. 2016).

Here, we use the distribution of phytotelm-breeding across the phylogeny of

Neotropical frogs (3,105 species) to investigate the directionality and frequency of transitions from other breeding microhabitats to phytotelmata, and vice-versa. We also tested whether phytotelm-breeding frogs have distinct diversification rates, and whether this specialization corresponds to an evolutionary ‘dead-end.’ Our results show that there are numerous transitions to phytotelm-breeding in Neotropical frogs but reversals rate is higher. We found similar support for three macro-evolutionary models, one those predicts phytotelm-breeding lineages having higher extinction rates. Although we find some support for the number of observed phytotelm-breeding frogs to be lower than the reconstructed number of origins under state-dependent extinction model, distribution of tip lengths of phytotelm- and non-phytotelm-breeding frogs are difficult to distinguish between of competing generative models.

Materials and Methods

Data sampling and ancestral states

The phylogenetic dataset of Jetz & Pyron (in press) contains 7,238 species of amphibian, taken from the 19 February 2014 edition of AmphibiaWeb. From these, we gathered a list of Neotropical frogs, which comprise 3,105 species. These species are distributed across Neotropical ecoregion including Mexican province, Tropical Middle

America, Antilles, Amazon, Cerrado, Chaco, and Atlantic Forest (Morrone 2014). We used scientific literature published in peer-review journals, books, AmphibiaWeb database, and the IUCN database to classify phytotelm-breeding as a binary trait. Species constructing nests, depositing eggs (e.g., species with direct development) or tadpoles in water-holding plants were classified as phytotelm-breeding (1) and species using other resources for nursery were classified as non-phytotelm-breeding (0) (Appendix 1 at https://drive.google.com/file/d/0B-k1-50WDeJZVklxTnQ3UjlwNjg/view?usp=sharing).

This follows numerous recent authors in their classification of phytotelm- vs. non- phytotelm-breeding behavior (e.g. Peixoto 1995; Lehtinen et al. 2004).

In all the analyses below, we used a representative tree from the PASTIS distribution of 10,000 fully sampled (7,238 species) amphibian phylogenies, and the summary data-only (4,061 species) phylogenies. We trimmed the PATIS and the data-only phylogeny to include only Neotropical frog species. In the PASTIS, we have represented all 3,105 species that we have trait data but in the data-only we had to exclude 1,526 species that we have trait data but are missing molecular data, therefore are not represent in the data-only phylogeny. Simulations for both the PASTIS and data-only phylogenies ran with

100 replicates to generate summary distributions of statistics. To explore the frequency of

transition rates between states and model uncertainty in character reconstruction, we used stochastic mapping (Huelsenbeck et al. 2003). We ran this analysis using the function make.simmap in the package phytools (Revell 2012). We simulated one hundred stochastic maps with all transition-rates different. This approach samples possible character-histories from the posterior distribution, and allows us to estimate both the number of origins and reversals, as well as the potential uncertainty in these estimates. An infinitely large number of potential histories could explain the observed distribution of tip states, thus sampling these in proportion to their likelihood shows the potential range of historical outcomes.

These could bear heavily on our interpretation of the historical impact of the trait on diversification.

Macroevolutionary models

Our first hypothesis was that phytotelm-breeding exerts a strong influence on diversification rates. We tested whether the observed data are consistent with a model where phytotelm-breeding behavior increases the extinction rate, lowers the speciation rate and/or influences trait-transition rates. Alternative macroevolutionary models can be generated and tested against the observed phylogenetic distribution of phytotelm-breeding tip states, using the parameter space (speciation, extinction, and trait transition rates) of the observed phylogeny to generate expected values of summary metrics under different macroevolutionary scenarios. To develop these alternative models, we used the fully- sampled phylogenies of amphibians (and the data-only phylogeny; Supplement

Information) trimmed to the pool of Neotropical frogs. We used the BiSSE model

(Maddison et al. 2007) to estimate maximum-likelihood parameter values using the packages diversitree (FitzJohn 2012).

We used the following seven macroevolutionary models (e.g. Bromham et al.

2016):

1) Full; phytotelm-breeding affects speciation (λ0 ≠ λ1) and extinction (µ0 ≠ µ1), and transition (q01 ≠ q10) rates of gain (0 to 1) and loss (1 to 0) are unequal.

2) Baseline; phytotelm-breeding does not affect the speciation (λ0 = λ1) or extinction rates (µ0 = µ1), and rates of trait gain and loss are equal (q01 = q10) between the states (i.e., Mk1).

3) Dead-end; phytotelm-breeding increases the extinction rate (µ0 < µ1), but the speciation and trait transition rates are equal between phytotelm-breeding and non- phytotelm-breeding lineages (λ0 = λ1, q0 = q1).

4) Suicide; non-phytotelm-breeding lineages have a greater rate of transition to phytotelm-breeding (q01 > q10). Furthermore, phytotelm-breeding lineages have higher extinction rates (µ0 < µ1), but not speciation rate (λ0 = λ1);

5) Lonely; phytotelm-breeding reduces speciation (λ0 < λ1), but extinction and trait transition rates are equal in phytotelm- and non-phytotelm-breeding frogs (µ0 = µ1, q01 = q10);

6) Irreversible; the transition from phytotelm-breeding to non-phytotelm-breeding is zero (q10 = 0), speciation and extinction rates are equal in phytotelm- and non-phytotelm- breeding frogs (λ0 = λ1, µ0 = µ1);

7) Character-independent diversification (e.g. Mk2), phytotelm- and non- phytotelm-breeding lineages have equal speciation and extinction rates (λ0 = λ1, µ0 = µ1) but transition rates between states are asymmetric (q01 ≠ q10).

In addition to those models, Bromham et al. (2016) proposed the Labile model, in which transition rates are equal (q01 = q10) but higher than in the Baseline model. This resulted from their simulation regime, where these rates were fixed at high or low values.

Moreover, in the Labile model the trait does not affect speciation and extinction rates, which is similar to the Character-independent diversification model (Mk2). The Labile model is identical to the Baseline model, and overlaps with the Mk2 model; thus, we did not include it in the analyses. We used the AIC values to estimate Akaike weights for model comparison.

Summary phylogenetic metrics

Our second, related hypothesis is that phytotelm-breeding represents an evolutionary ‘dead-end,’ which would show a signal in the branch-length distributions and observed number of tips per origins for the trait. We used the tip age rank sum metric

(TARS) to test whether phytotelm-breeding frogs have longer tip lengths than non- phytotelm-breeding frogs (Bromham et al. 2016). In addition, we used the number of tips per origin (NoTO) test whether the number of observed lineages of phytotelm-breeding frogs is lower than the number of reconstructed origins (Bromham et al. 2016).

Rabosky and Goldberg (2015) recommended using power simulations to test for model adequacy, i.e. whether we would expect each of the seven evolutionary models to produce the observed distribution of tip states. Thus, we used ‘backward simulation’ to

generate phylogenies with the same number of phytotelm- and non-phytotelm-breeding frogs as in the observed data, taking into account the sampling fraction (completeness) of the phylogeny (e.g. Hua and Bromham 2016). We used a Brownian motion model of trait evolution to generate a null model of the expected distribution of tip sates according to

TARS and NoTO across simulated phylogenies under the seven macroevolutionary models mentioned above. Then, we estimated TARS and NoTO for the simulated tip states on the phylogenies under the models, and tested for differences with the observed data. In TARS, our expectation is that phytotelm-breeding frogs would have a longer tip length than non- phytotelm-breeding. Thus, we accept a given macroevolutionary model for TARS if p ≥

0.95, it means that phytotelm-breeding have longer tip lengths than the 95% the distribution of values generated under the null model (Bromham et al. 2016). In NoTO, our expectation is that the number of phytotelm-breeding lineages is lower than then number of reconstructed origins. Thus, we accept a given macroevolutionary model for NoTO if p ≤

0.05, it means that the number of phytotelm-breeding lineages is lower than the 95% distribution of values generated under the null model (Bromham et al. 2016).

These phylogenetic metrics (e.g., tip age rank sum and number of tips per origin) and simulation procedures were implemented in the R (R Core Team 2016) package phylometrics (Hua and Bromham 2016) with significance determined by the nonparametric

Wilcoxon rank-sum test (Bromham et al. 2016).

Results

Ancestral States

The results using the fully sampled and data-only phylogenies agree with each other. The macroevolutionary models with better fit to the data estimated using both the

PASTIS and data-only are the same (see below), the summary phylogenetic statistics resulted in similar values, and the stochastic character mapping estimated similar proportion of state transitions (Supplemental Information). Thus, we present in the text results of the analyses using the fully sampled phylogenies and include the results of the data-only tree as Supplemental Information.

The character-state transition estimated using the Mk2 model show 70 unambiguous changes from non-phytotelm-breeding to phytotelm-breeding (Fig. 1;

Supplement Information). This is similar to the stochastic mapping estimates from simmap

(Fig. 2). However, the stochastic model shows the uncertainty in character state reconstruction and model of character transition (Fig. 2), including occasional estimation of deeper nodes in the phylogeny as phytotelm-breeding. This suggests that the distribution of phytotelm-breeding among the tips does not unambiguously support a single preferred character-history, either early origin with frequent reversals, or multiple recent origins.

Overall, the simmap results estimate 175 changes between states, 68 origins and 107 reversals (data-only phylogeny: 99 changes between states, 31 origins and 68 reversals).

The diversity of phytotelm-breeding frogs in the Neotropics is distributed across

Amazon, Atlantic Forest, West Indies, and Tropical Middle America but mainly concentrated in the Amazon and Atlantic Forest (Appendix 1). Our results explain the distribution of 150 phytotelm-breeding species in ten families occurring in the Neotropics

(Aromobatidae, Bufonidae, Centrolenidae, Craugastoridae, Dendrobatidae,

Eleutherodactylidae, Hemiphractidae, Hylidae, Leptodactylidae, and Microhylidae).

Qualitatively, we observe that many of the reversals occur in clades (e.g.,

Eleutherodactylus) where species co-opt easily available phytotelmata such as open pools in terrestrial plants, such that transitions to and from phytotelm-breeding represent only small shifts in behavior and the location of egg or larvae deposition. In contrast, reversals in clades with strong specialization, such as to breed in bromeliads tanks (e.g., Syncope,

Scinax, Osteopilus, Dendrophryniscus), represent fewer reversals to terrestrial breeding.

Figure 1. Character reconstruction under the best fit macroevolutionary model (Mk2). Red branches indicate Neotropical phytotelm-breeding frogs and blue non-phytotelm-breeding frogs. In pink are branches with ambiguous estimation.

Figure 2. Stochastic mapping showing uncertainty and ambiguous reconstruction of deep branches. These samples show the highly variable nature of likely character-histories. Red indicates phytothelm-breeding while blue indicates non-phytothelm-breeding.

Macroevolutionary Models

Out of the seven macroevolutionary models tested for phytotelm-breeding affecting speciation, extinction, and state transition rates, the Full, Suicide, and Mk2 models have higher Akaike weights than the others (Table 1). However, we did not find strong support to distinguish these three models from each other. The Full model assumes that phytotelm- and non-phytotelm-breeding frogs have distinct rates of speciation, extinction, and transition between states. In contrast, the Suicide assumes that phytotelm- and non- phytotelm-breeding frogs have similar speciation rates but phytotelm-breeding have higher extinction rates than non-phytotelm-breeding frogs. Furthermore, state transition from non- phytotelm- to phytotelm-breeding is higher than the reverse. The Mk2 model assumes asymmetric transition between states but speciation and extinction rates are similar in phytotelm- and non-phytotelm-breeding frogs.

The Mk2 had highest Akaike weights and the lowest number of parameters (Table

1), then we conservatively consider the Mk2 the best fit to the data. Phytotelm breeding arises about as many times as expected under a stochastic model and also reverses frequently, as result state transition rates are higher from phytotelm- (q10) to non- phytotelm-breeding (q01) than the reverse. This indicates that phytotelm-breeding is a labile trait that does not exert a strong influence on diversification dynamics (Table 2).

Nonetheless, we do note some support for an evolutionary model for phytotelm-breeding increasing extinction rates (e.g., Suicide; Table 1).

Table 1. Summary of macroevolutionary models and model selection. The best fit model is in bold. Df, degrees of freedom in terms of number of fixed parameters in the models.

Df lnLik AIC ChiSq Pr(>|Chi|) Akaike weights

Full 6 -12656 25325 0.199

Baseline 3 -12680 25366 46.955 3.55E-10 *** 0

Dead-end 4 -12662 25332 11.211 0.003 ** 0.005

Suicide 5 -12658 25325 1.916 0.166 0.207

Lonely 4 -12678 25364 43.197 4.17E-10 *** 0

Irreversible 3 -12711 25428 109.046 < 2.2e-16 *** 0

Mk2 4 -12657 25323 1.827 0.401 0.589

Table 2. Speciation, extinction, and state transition rate parameters estimated under the seven macroevolutionary models tested.

λ0 λ1 µ0 µ1 q01 q10

Full 0.053 0.045 0 0 0.001 0.026

Baseline 0.053 0.053 0 0 0.001 0.001

Dead-end 0.054 0.054 0 0.05 0.002 0.002

Suicide 0.054 0.054 0 0.032 0.002 0.012

Lonely 0.053 0.042 0 0 0.001 0.001

Irreversible 0.053 0.053 0 0 0.002 0

Mk2 0.053 0.053 0 0 0.001 0.028

Summary phylogenetic metrics

We observed in the trait simulations under the generative models compared to the observed distribution of phytotelm-breeding frogs relatively concordant results with the macroevolutionary models. The TARS metric show that the distribution of tip lengths of phytotelm-breeding cannot be distinguished between the seven models. Thus, no model is supported over the others, and we conservatively continue to interpret this as support for the simplest, state-independent Mk2 model. Furthermore, our results from the TARS metric show that phytotelm-breeding species do not have longer tip lengths relatively to non-phytotelm-breeding (Table 3).

In contrast, the specialization to phytotelm-breeding resulted in a lower number of expected tips per origin than the number of origins observed with the trait, according to the

NoTO metric. The NoTO rejected all models except the Dead-end (using the PASTIS phylogeny), and Suicide and Full (using the data-only phylogeny) models. These results suggest that models with higher extinction offer a better fit than other models with state- independent to explain the number of phytotelm-breeding frogs across the phylogeny. This mirrors a weak signal for the Suicide model in the BiSSE results (Table 1).

Table 3. Significance of summary statistic metrics comparing observed data with simulated tip states under the seven macroevolutionary models test. In NoTO p ≤ 0.05 and in TARS p ≥ 0.95 indicates the model could not be recovered from the data if it had generated the pattern.

PASTIS Data-only

NoTO TARS NoTO TARS

Full 0 0.21 0.01 0.1

Baseline 0 0.19 0.06 0.39

Dead-end 0.04 0.36 0.12 0.07

Suicide 0.16 0.37 0 0.49

Lonely 0 0.24 0.1 0.38

Irreversible 0 0.21 0.1 0.41

Mk2 0 0.18 0.09 0.48

Discussion

We did not see strong concordance across analyses for a single generative model, though both macroevolutionary model-fitting and phylogenetic summary metrics show some support for state-dependent extinction (e.g., Suicide and Dead-end models).

However, this result may indicate the presence of a Type-I error due to phylogenetic pseudoreplication (Rabosky and Goldberg 2015), which requires further study for corroboration (Maddison and FitzJohn 2015). We continue to accept the state-independent

Mk2 model as the most conservative explanation for the data, but suggest that additional

study is needed to firmly settle the potential for state-dependent diversification rates on phytotelm-breeding (e.g., higher extinction and evolutionary 'dead-end').

Our ancestral-state estimates are highly ambiguous, indicating that phytotelm- breeding may have evolved early in Neotropical frogs’ evolutionary history as a strategy to overcome competition and predation in pond and stream environments. Alternatively, there may have been as many as 68 recent origins of the trait in lineages across the

Neotropical frog Tree of Life. A lack of clear signal for state-dependent speciation and extinction models, from either macroevolutionary model-fitting or phylogenetic summary statistics, suggests that colonization of phytotelmata alone as a breeding site in Neotropical frogs did not represent a significant source of ecological opportunity resulting in adaptive radiation. It is also possible that phytotelm-breeding is a proxy for other, more important factors affecting diversification rates of Neotropical frogs, such as direct development, parental care, and related traits such as calling site (see Pyron and Wiens 2013; Castroviejo-

Fisher et al. 2014; Grant et al. 2006).

Alternatively, the signal for the Suicide model may reflect other traits shared wholly or in part by phytotelm-breeding lineages that raise their extinction rates due to other ecological or evolutionary processes (e.g., Beaulieu and O’Meara 2016). For instance, it is still unclear how rapidly non-phytotelm-breeding species can specialize to deposit eggs in phytotelmata, which could have resulted in the weak signal for the Suicide and Dead-end models (Table 3). If transitions to phytotelm-breeding occur rapidly, but specialization occurs more slowly than the background turnover-rate, then this might lead to higher extinction (e.g., Suicide model) in phytotelm-breeding lineages. We recover a weak signal for such state-dependent extinction model, but caution that numerous other factors may be

confounding these results (see Rabosky and Goldberg 2015). Thus, additional targeted studies and simulations will be needed to corroborate or refute this hypothesis, possibly specific to individual phytotelm-breeding lineages.

In addition, it is unclear what degree of obligate specialization or potential commensalism or other symbiotic relationships (e.g., utilization by larvae of chemicals secreted by the plants) may exist between frogs and phytotelmata to promote or retard further specialization or diversification, or even promote extinction. Natural history observations show that species specialized to deposit eggs and tadpoles in bromeliad tanks, tree holes, husks of fruits, and bamboos internodes rarely use more than one of these resources (Lehtinen et al. 2004). This suggests high specificity for adaptations to a given phytotelmata resource. Thus, future studies might partition phytotelm-breeding behavior to a lower level of granularity (e.g., bromeliad tanks versus bamboo internodes), and include phytotelm-breeding as a multistate character (e.g., opportunistic vs. obligate) to investigate the role of potential intermediate stage. Currently, detailed natural history information is available for a small set of phytotelm-breeding species or in travel logs from field trips housed in natural history collections that have not been digitize or may not be trivial to find the information. Related behaviors such as nest-building in Neotropical lineages like Leptodactylidae and Old World lineages like Rhacophoridae may also be tested for a similar signal.

Sources of phytotelmata considered here are variable and may have been unequally available through time. For instance, tree holes, vine holes, leaf axils, husks of fruits, and bamboo internodes are phytotelmata opportunities available to frogs for over 140 Ma

(Magallón et al. 2015). In contrast, bromeliad tank is a key innovation in Bromeliaceae

lineages that evolved ~10 Ma (Silvestro et al. 2013). The average diversification time of phytotelm-breeding is 50 Ma (Jetz and Pyron in press). Thus, it is intriguing that species within genera such as in Andinobates, Oophaga, Ranitomeya, Crossodactylodes, and

Phyllodytes specialize on bromeliad tanks, and show no reversals back to breed in ponds, streams, or on the ground like their close relatives. Further extensions of existing macroevolutionary models might incorporate more ecological and evolutionary realism by including time-constrained state-dependent transitions.

If phytotelm-breeding affects long-term survivorship and restricts species from returning to explore other breeding habitats, we would expect species with this trait to have longer tip lengths than non-phytotelm-breeding. Although we did not find support for differences in the TARS metric, qualitatively we can identify instances that lineages represent deeper nodes. For instance, Dendropsophus bromeliaceus is the first obligate bromeliad-dwelling species in a genus with 102 species, and represents the deepest node in the phylogeny of the genus (Ferreira et al. 2015). Interestingly, the sister genus to

Dendropsophus is also restricted to reproduce in bromeliads (e.g. Xenohyla) and has only

2 species (Frost 2017). Another example is Melanophryniscus setiba which clusters at the base of the genus, lives in sand dunes (Peloso et al. 2014), and most likely uses bromeliad tanks for reproduction since these are reliable sources of water in environments that water do not accumulate easily. These are examples of phytotelm-breeders having longer tip lengths than close relatives, non-phytotelm-breeders. Using more detailed classification of phytotelm-breeding might increase support for our expectation of phytotelm-breeding frogs having longer tip lengths due to high extinction rates and lower reversals, specifically in bromeliad specialists. Finally, fieldwork and increase digitalization of information

available in natural history collections is crucial to yield new natural history discoveries and improve our understanding of the phytotelm-breeding specialization to partition phytotelm-breeding more finely.

Chapter 3: Neutral biogeographic processes primarily drive phylogenetic structure

of Neotropical frog communities

Abstract What determines species co-occurrence and community assembly in an ecosystem? Answering this question requires null expectations for how local species composition changes through time under stochastic rates of fundamental processes known to ultimately influences local species pools (e.g., speciation, local extinction, and colonization). Combining data on evolutionary relationships and community composition in high-turnover ecosystems, such as the Neotropics, can help disentangle the roles of ecological processes in defining the phylogenetic structure of communities. Given that stochastic and deterministic evolutionary and ecological processes can both affect the resulting phylogenetic pattern in present day communities, it is thus first necessary to attempt to reject a null biogeographic model of community assembly. Our goal is to test whether communities of Neotropical frogs assembled via deterministic or stochastic routes of evolutionary processes including colonization, speciation, and local extinction. We use a fully-sampled phylogeny of all 3,100 species of Neotropical frogs distributed across 26 families, 186 genera, and we summarized community matrices including ~46% of the continental diversity to fit a dynamic null model of community assembly. Our results show that most communities are neutral. Despite similar species richness between Amazon and

Atlantic Forest, our results show distinct phylogenetic structure of local communities.

Communities with high species richness such as Santa Teresa in the Atlantic Forest and

Barro Colorado in Tropical Middle America were recovered as neutral. However, phylogenetically clustered assemblages are present in the Chaco and western Amazon,

meaning that those communities are composed of more closely related species than would be expected under the neutral model. Clustering in the western Amazon is a potential result of higher diversification within the ecoregion of a few hyper diverse-endemic lineages such as in poison frogs, glass frogs, and tree frogs (e.g. biogeographic overrepresentation).

Clustering in the Chaco derives from habitat filtering in colder, drier climates. However, after we accounted for species occurring in surrounding areas of the local communities but not reported during field surveys, these communities were less clustered phylogenetically or neutral. Moreover, species richness is slightly correlated to phylogenetic clustering but this relationship vanishes when we included missing species in the community surveys but known to occur nearby. Therefore, the rare signal of clustering may be artifactual in most or all of our clustered communities. Thus, Neotropical ecoregions comprise distinct assemblages of frogs as demonstrated by several biogeography studies, but the phylogenetic distribution of community assemblages is a random sample of the regional pool.

Introduction

The pattern of relatedness among species in a community is determined by the interaction of ecological and evolutionary processes (reviewed in Weber et al. 2017).

Communities of co-existing species assemble via colonization (and successful establishment of a lineage in a new region or habitat), speciation, and local extinction (Pigot and Etienne 2015). In turn, these evolutionary processes are expected to leave visible signatures in the phylogenetic structure and species richness of present-day communities

(Pigot and Etienne 2015; Burbrink et al. 2015; Weber et al. 2017).

Furthermore, ecological processes such as habitat filtering (i.e., a process through which species fail to establish in a community due to incompatibility with relevant environmental factors; Keddy 1992) and species interactions (e.g., competitive exclusion between species due to limiting similarity; MacArthur and Levins 1963) are also expected to leave a signature in the phylogenetic structure of communities (Webb et al. 2002). For instance, ecological processes would drive communities to be phylogenetically clustered

(i.e., where species are more closely related than expected under a null model) through conservatism in certain ecological traits shared across closely related species, and through the exclusion of distant related species ecologically similar but with low competitive abilities (Webb et al. 2002; Mayfield and Levine 2010; Cavender-Bares et al. 2009). In contrast, communities might have a signature of phylogenetic overdispersion (i.e., where species are more distantly related than expected) through the extinction of closely related, functionally similar species (Purschke et al. 2013), or through habitat filtering of convergent traits and colonization of species distantly related, functionally dissimilar to residents (Liu et al. 2015).

In examining the impact of processes known to influence phylogenetic distribution of communities, the choice of null models for testing our expectations of relatedness is a point of fundamental importance. Early investigations of phylogenetic community assembly offered relatively simple models, suggesting that habitat filtering of conserved ecological traits would be the primary mechanism resulting in clustering (Webb et al.

2002). Conversely, habitat filtering of convergent traits in distantly related species would result in overdispersion, as would strong competitive interactions among closely related species sharing similar ecological traits, resulting in local exclusion. However, recent

theoretical and empirical work has challenged these views showing that the interplay among traits, biotic and abiotic factors can result in numerous, subtle patterns, including clustering driven by competition (Mayfield and Levine 2010).

In addition to problems of interpreting process from pattern due to ecological drivers, these early models neglect the historical role of speciation and extinction within the regional species pool to generate patterns of overdispersion and clustering in communities that arise via primarily neutral evolutionary processes (Hubbell 2001;

Cavender-Bares et al. 2009; Pigot and Etienne 2015). These early models assume a static regional pool that is sub-sampled randomly by filtering or competition (Webb et al. 2002).

Random instances of speciation, local extinction, and colonization within the regional species pool can generate local communities of co-occurring species that appear to be clustered or dispersed simply by chance (Pigot and Etienne 2015). Thus, the relative influence of ecological and evolutionary patterns and processes structuring diverse communities across landscapes are still unknown for most groups. This is particularly true for diverse continental radiations, especially in tropical regions.

Neotropical frog communities

In frogs, high species richness in the tropics is explained by long-term presence, high speciation and low extinction rates, and limited dispersal to temperate areas (Pyron and Wiens 2013). Furthermore, the high species richness of one of the most speciose family of frogs, the Hylidae commonly known as treefrogs, in Amazonian communities is explained by biogeographic processes where lineages that colonized the region early diversified rapidly resulting in more species (e.g., effect of time-for-speciation; Wiens and

Donoghue 2004), as well as long-term sympatry of multiple treefrog clades (Smith et al.

2007; Wiens et al. 2006, 2011).

In general, Neotropical frogs occur in diverse communities including dozens or hundreds of species across dozens of families spanning hundreds of millions of years of evolutionary history (Pyron 2014). Moreover, the Neotropics comprises more than half of the worldwide diversity of frogs (~3,100 species) and multiple distinct ecoregions with varying diversity, geological history, climatic conditions, and forest composition and structure (e.g., Olson et al. 2001; Morrone 2014). This diverse history has resulted in geographically distinct regional and local species pools that have experienced different ecological pressures over geological timescales. Combining data on evolutionary relationships and community composition in systems with high turnover in beta diversity, such as the Neotropical ecoregions, can help disentangle the roles of ecological and evolutionary processes on community assembly (e.g., Weber et al. 2017).

It is mostly unknown whether phylogenetic structure of present-day communities of Neotropical frogs primarily reflect the influence evolutionary processes of assembly, ecological processes such as filtering and competition, or a mixture of both. Given the possibility that both stochastic variation of evolutionary processes (e.g., speciation, extinction, and colonization) and non-neutral ecological processes (e.g., filtering and competition) can result in clustering or dispersion (see Mayfield and Levine 2010; Pigot and Etienne 2015), it is necessary to disentangle the two scales by first testing the biogeographic model of community assembly. The DAMOCLES model (Pigot and Etienne

2015) allows for this by fitting a Maximum Likelihood (ML) model to a time-calibrated phylogeny and community matrices to estimate parameter values of local extinction,

colonization, and colonization decline (e.g., given to community pool saturation), and then simulates null communities under these parameters. The observed values of a given phylogenetic community-assembly summary statistics can then be compared to the null communities, to test for significance. In addition, ML estimates from the data can be used in regressions to investigate the relative influence of local extinction, colonization, and colonization decline in generating the observed phylogenetic pattern in communities.

In the present study, we ask whether local species pools of Neotropical frogs assembled via deterministic or stochastic variation in colonization, local extinction, and speciation. We used the phylogeny of amphibians (Jetz and Pyron in press), trimmed to the

3,100 species of Neotropical frogs and a matrix of species composition by local communities and ecoregions to simulate assemblages under the dynamic null model of community assembly (Pigot and Etienne 2015). Then, we used a summary phylogenetic metric that is independent of species richness (Phylogenetic Species Variability - PSV;

Helmus et al. 2007) to examine the phylogenetic distribution of the empirical data relative to the simulated communities using the dynamic null model. Overall, our results show that the phylogenetic structure of most communities across the Neotropics are not distinguishable from neutral models, and do not appear to reflect clustering or dispersion resulting from filtering or competition, nor from deterministic variation of evolutionary processes affecting community assembly. The phylogenetic diversity in Neotropical communities of frogs assembled under neutral variation of evolutionary processes and represent a random sample of the regional pool.

Material and Methods

Phylogenetic and Community Data

Amphibian distributions have been used as benchmarks for new methods aiming to identify biogeography regionalization (Olson et al. 2001; Holt et al. 2013; Vilhena and

Antonelli 2015; Elder et al. 2016). The ecoregions of Neotropical amphibians estimated using different methods largely agree (Elder et al. 2016) and match with regional estimates using data from geology, climatic gradients, and forest structure and composition (e.g.

Olson et al. 2001; Morrone 2014). We used Neotropical ecoregions as the spatial definition of regional pools (e.g., Olson et al. 2001, Vilhena and Antonelli 2014, Edler et al. 2016,

Morrone 2014) (Fig. 1).

In this study, for the Neotropics, the northernmost limit represents the Mexican transition zone, an area of overlap between the Nearctic and Neotropical regions, while the southernmost limit represents the South American transition zone area of overlap between the Neotropical and Andean regions (e.g. Morrone 2014). Several studies focusing on biogeographic regionalization have shown that the southernmost portion and the Andean area of South America have closest links with other Austral areas, mainly Australia,

Tasmania, New Guinea, New Zealand, and South Africa (reviewed in Morrone 2014).

Moreover, the northern Mexico has been assigned to the Nearctic region (Holarctic realm).

Thus, we did not include communities from the Mexican transition zone and South

American transition zone (Morrone 2014).

We next classified the geographic distribution of the 3,100 species of Neotropical frogs based on expert knowledge, museum collections, species lists, and range maps to generate regional species pools (e.g., available on AmphibiaWeb, IUCN, GBIF) (Appendix

1 available at the online repository https://drive.google.com/drive/folders/0B-k1-

50WDeJZa2dnSlN0SGRGdU0?usp=sharing).

In addition, we assembled a dataset of 555 local communities of Neotropical anurans distributed across Neotropical ecoregions (Fig. 1, Table 1, Appendix 2 available at the online repository https://drive.google.com/drive/folders/0B-k1-

50WDeJZa2dnSlN0SGRGdU0?usp=sharing). We searched scientific articles published in peer-review journals, government data, and gray literature of Master and Doctoral thesis emphasizing detailed studies of amphibian fauna on local sites, as well as museum records

(Appendix 2). Communities are defined as co-occurring populations of different species that evolved under similar environmental selection pressures over time-scales and space

(HilleRisLambers et al. 2012). Unfortunately, the fossil data for frogs is extremely limited and sparse to allow detail time-series of community assemblage changes (but see Silvestro et al. 2015; Kemp and Hadly 2016), which makes it impractical to use this definition. Thus, we used a more operational definition of communities as unique survey sites defined during species surveys (e.g., Wiens et al. 2011; Tucker et al. 2016). Sites were typically several km2 in size but represented a single biome (e.g. lowland rainforest, open savanna) with multiple habitats (e.g. forest, pond, and stream). The definition of sites used here is standard in field surveys focusing on local amphibian diversity and in studies of tropical amphibians

(Duellman 2005; Wiens et al. 2011). Furthermore, the summary metrics of choice is independent of species richness which minimizes potential bias add by the species richness-area relationship (Rosenzweig 1995).

Figure 1. Phylogeny of Neotropical frogs with tip states of endemic species color by ecoregion and map of 555 local communities across ecoregions. Tip states in gray indicate species that occur in two or more ecoregions. Colors correspond to Mexican province (light green), Antilles (pink), Tropical Middle America (yellow), Amazon (green), Cerrado-

Caatinga (brown), Atlantic Forest (cyan), and Chaco (light brown).

Table 1. Neotropical ecoregions used to delimit the regional species pool of frogs.

Ecoregions Abbreviations Communities Species richness

Atlantic Forest AF 78 583

Cerrado-Caatinga CEC 97 278

Chacoan CHA 21 169

Tropical Middle America TMA 67 1197

Antilles ANT 32 189

Amazon AMZ 179 613

Mexican province MEX 81 307

Communities are defined as unique survey sites (see Material and Methods).

The final dataset includes 12,426 records of 1,445 species representing Neotropical local species pools of anurans, out of the 3,100 species comprising Neotropical anuran richness (Appendix 2). Ideally, the entire dataset of 555 communities and 1,445 species would be analyzed, giving a fine-scale geographic mapping of assembly processes.

However, several factors led us to limit the scale of the analyses. Most importantly, high- quality presence-absence data is needed for robust conclusions about community assembly, and a great deal of care is needed to verify the accuracy of species identification in community surveys (see Gerhold et al. 2008; Moen et al. 2009; Prinzing et al. 2008; Wiens et al. 2011; Burbrink et al. 2015). The communities we examined varied widely in the geographic area and the search effort, and we thus expect high rates of Type II error

(species omission) in many of them. Furthermore, differences in ecology (e.g., fossorial, terrestrial, or arboreal) and natural abundance (e.g., rare or common species) might affect

species detection as well. Second, there is likely a subtle but insidious effect of spatial autocorrelation that is not adequately addressed by current community assembly models.

Nearby communities will share many wide-ranging species, which may have disproportionate impacts on gross measures of clustering or dispersion. This may have the dual effect of i) inflating prevalence estimates for patterns such as clustering due to pseudo replication (as many communities may share the same pattern of clustering due to sharing the same taxa), and ii) hiding the effects of filtering or competition for species which are not shared. Most importantly, the most diverse communities generally have the most saturated species-accumulation curves for their search effort.

Thus, we limited our analyses to the seven most diverse communities from each of the seven ecoregions, for a total of 49 communities. We consider this an optimal strategy for several reasons. First, the most diverse communities tend to be the most well studied, with the greatest amount of search effort over longer periods of time. Since species presence is a positive phenomenon (i.e., it generally cannot be observed in error), more diverse community lists from greater search effort will likely tend to be more accurate. A less diverse community list may also be highly accurate, but could also simply reflect a failure to observe species, which are present, a Type II error. Second, at least qualitatively, the most diverse communities in each ecoregion are fairly dispersed geographically within ecoregions. Finally, the most diverse communities have the largest amount of data to estimate model parameters, as well as the highest likelihood of strong interspecific interactions, in both cases due to the larger number of observed species. For most phylogenetic comparative methods, power increases with sample size, so these communities should offer the best opportunity to recover significant signals of clustering

or overdispersion. For these most-diverse communities, we performed a secondary check of the literature and examined the range maps of related taxa, to verify that we had an accurate picture of the total diversity of species present at each site.

Because Amazonian diversity is concentrated in the western portion of the basin, near the Andean slopes, this left a vast swath of the interior continent without any representation. Therefore, we rounded our final list of communities to 50 by including the anuran list for the Reserva Adolpho Ducke, near Manaus, Brazil (Appendix 3 available at the online repository https://drive.google.com/drive/folders/0B-k1-

50WDeJZa2dnSlN0SGRGdU0?usp=sharing). This is a well-studied forest plot near the center of the continent, and should provide at least a preliminary indication of any differences in community assembly across the broad expanse of Amazonia.

Null Assembly Models

We used a dynamic null model of community assembly based on the fundamental processes of colonization, local extinction, and speciation, which is implemented in the R

(R Core Team 2016) package DAMOCLES (Pigot and Etienne 2015). In this model at any given time, species can be locally present or absent. Species are added to the local community via colonization γ and lost from the local community through local extinction

µ. This model assumes that per lineage rates of colonization and local extinction are equal across species (i.e., the model is neutral at the species level; MacArthur and Wilson 1967;

Hubbell 2001), and the species pool is dynamic according to the speciation or global extinction of species within the clade. The model also assumes that speciation λ occurs in allopatry and that the area of the local community is sufficiently small that in-situ

speciation is negligible (Hubbell 2001). As a result, when a species in the local community speciates, only one daughter species will be locally present. In contrast, when a locally absent species undergoes speciation both daughters will also be locally absent (Pigot and

Etienne 2015). Simulated local species pool not necessarily will have the same species richness as the observed community used to extract parameters of the DAMOCLES model.

Thus, the model simulates a dynamic local species pool.

We used the fully sampled time-calibrated phylogeny of amphibians (Jetz and

Pyron in press). This dataset was constructed using a sparse supermatrix for 4,061 species, with the 3,177-remaining species imputed under a birth-death model for a total of 7,238 described, extant amphibians. Based on the taxonomic and geographic reference of the

AmphibiaWeb (http://amphibia-web.org/) we pulled the list of Neotropical species of frogs, leaving us 3,100 species (Fig. 1). Then, we used the phylogeny of Neotropical frogs trimmed to the regional species pool of the ecoregions and the community matrices of local species pools to estimate values for the parameters in the DAMOCLES model (i.e., µ, γ0, and γ1). For each local community, we used the estimates to simulate five hundred null communities under stochastic variation of these parameters. We estimated PSV using the package picante (Kembel et al. 2010) for the simulated communities to create a null distribution and we compared the PSV of simulated communities with observed values of

Neotropical communities.

The PSV index uses information of a single community, thus provides information on the α-diversity of the species pool (Tucker et al. 2016). To estimate species phylogenetic variability from a community phylogeny the method measures the evolution of a neutral trait under Brownian motion evolving independently along the branches of a time-

calibrated phylogeny (Helmus et al. 2007). For a given species the value of this neutral trait in any point in time will be normally distributed and variance of the trait is proportional to the distance from the root node to the branch of the species (Felsenstein 1985; Garland et al. 1993; Helmus et al. 2007). Thus, the community phylogeny defines a multivariate normal distribution with a covariance matrix whose elements give expected evolutionary covariance between species. PSV has the assumptions that 1) local pools are random draws of regional pools, 2) it considers that at the beginning of a radiation lineages are all equally related to each other, and 3) over time variation in speciation and extinction rates make species more closely related changing the shape of the phylogeny. Thus, PSV quantifies how phylogenetic relatedness decrease the variance of this hypothetical unselected trait shared by all species in a community (Helmus et al. 2007). In the context of present study, this neutral trait would be species co-occurrence in communities (e.g. geography). PSV values closer to 0 indicate clustering and values closer to 1 indicate overdispersion.

PSV, Mean Phylogenetic Distance (MPD), and Mean Nearest Taxon Distance

(MNTD) estimate the divergence patterns in a group of species (e.g., α-diversity). PSV and

MPD are mathematically similar and measured phylogenetic distance using all pairwise distances for a given group of taxa, whereas the MNTD metric compares the shortest pairwise distances (Helmus et al. 2007; Webb et al. 2002; Tucker et al. 2016). In contrast to MPD, PSV is independent of the number of species. Additionally, it is standardized over a star phylogeny and does not need an extra round to standardize over the number of species across communities (Helmus et al. 2007). We used z-scores to standardize PSV values across simulations; z-scores less than -1.95 indicates phylogenetic clustering and z-scores higher than 1.95 indicates phylogenetic overdispersion, thus it is significant at p < 0.05.

Then, we standardized the species richness, µ, γ0, and γ1 to use in linear regressions to investigate the relationship between the parameters in the model generating the observed phylogenetic structure. We ran a regression model using z-scores as dependent variable and species richness, µ, γ0, and γ1 as explanatory variables. We ran Bayesian Variable selection analyses to choose among models using the package BayesVarSel (Garcia-

Donato and Forte 2016).

To test the effect of phylogenetic uncertainty in generating model parameters and resulting phylogenetic structure, we used a sample of five PASTIS trees (Jetz and Pyron in press). Tests across all phylogenies consistently converged to the same results.

Sensitivity analyses

Species detectability is a challenge for most field survey studies. Although we took actions to reduce the Type II error caused by species omission, certainly the studies listed here representing the fifty most diverse communities are not free of this potential bias. To test the consistency of our overall results and how sensitive they are, for the clustered or overdispersed communities, we examined whether species known to occur in the surrounding areas were not listed on the species surveys. For instance, within the Amazon communities Lynch (2005) listed several species known to occur around Leticia, Colombia that were not detected during his field surveys. The author listed 24 additional species from numerous genera likely to be present, which were not found during the surveys. We ran an additional test of this community by including these species, raising the total species richness from 89 to 113 (Appendix 1).

In addition, Núñez et al. (2004) did not detect in El Espinillar, Departamento Salto,

Uruguay some species known to occur in the Departamento. For instance, Odontophrynus americanus and Elachistocleis ovalis are widespread species in the Chaco and, as few other species in Salto, most likely occur in El Espinillar. In fact, range maps of both species show that their ranges extend over this community. Although E. ovalis is considered of doubtful identity and needs further investigation (Caramaschi 2010), including these two genera not yet represented in the El Espinillar community would give a more comprehensive picture of the true local diversity, and potentially impact the finding of clustering or overdispersion. As with Leticia, we reran the analyses of this community including all species found in Departamento Salto, raising the species richness from 20 to 26 (Appendix

1).

Finally, the PN Serra do Divisor (PNSD) and RE Alto Juruá (REAJ) are geographically close to each other and share many Amazonian species (Souza 2005). Since all species in the former locality are present in the latter, we removed REAJ and reran the regressions. Furthermore, we also tested whether phylogenetically clustered or overdispersed communities mostly drive the results by removing those and rerunning the regressions as well.

Results

Community Dataset & Assembly Models

The fifty most diverse communities of Neotropical frogs comprise 797 species, which corresponds to ~26% of the continental species pool. The results from DAMOCLES suggest that most Neotropical communities of frogs are not different from the null model

(Figure 2). Although communities across ecoregions might have similar species richness such as communities in Amazon and in the Atlantic Forest have both high species richness

(ranging from 75–111 and 51–77 species, respectively), only communities in western

Amazon, Tropical Middle America, and Chaco were recovered as clustered (Figure 3).

Phylogenetic clustering indicates that communities have more closely related species than expected under a neutral process of biogeographic assembly. We did not recover communities as overdispersed.

Figure 2. Density plots representing PSV distribution of the 50 communities analyzed under the DAMOCLES model. Background color represents ecoregions: Mexican province

(light green), Antilles (pink), Tropical Middle America (yellow), Amazon (green),

Cerrado-Caatinga (brown), Atlantic Forest (cyan), and Chaco (light brown). In gray is

shown the distribution of PSV of the simulated communities under the DAMOCLES model.

Density plots in black indicate phylogenetically clustered communities. Red bars represent observed PSV values for the respective community. Clustered communities have z-score ≤

-1.95 and overdispersed ≥ 1.95. No communities were significantly overdispersed.

Figure 3. Phylogenetic structure of the 50 most diverse communities of Neotropical frogs.

Size of the circles is scaled by the local species richness and the colors (scale blue-red) correspond to the z-scores. Clustered communities are represented in red (z-score ≤ -1.95) and overdispersed in blue (z ≥ 1.95).

The z-scores, and local extinction (µ), colonization (γ0) and colonization decline

(γ1) rates estimated for communities across ecoregions largely overlap (Figure 4).

However, the Cerrado-Caatinga and Amazon have higher µ than the Mexican province,

Chacoan, and Tropical Middle America (Figure 4). Colonization rates are higher as well in the Cerrado-Caatinga than in Tropical Middle America and Mexico (Figure 4).

Figure 4. Variation across Neotropical ecoregions of z-scores, local extinction (µ), colonization (γ0), colonization decline (γ1) rates estimated under the dynamic model of community assembly. Mexican province (light green), Antilles (pink), Tropical Middle

America (yellow), Amazon (green), Cerrado-Caatinga (brown), Atlantic Forest (cyan), and

Chaco (light brown). Clustered communities have z-score ≤ -1.95 (dotted-line) and overdispersed ≥ 1.95.

Sensitivity Analysis

The model with the highest Bayes Factor and posterior probability was z-scores ~ species richness (Supplementing Information), suggesting a weak but direct relationship between species richness and phylogenetic clustering. In general, the higher the species richness in local communities the more likely that those species are phylogenetically closely related (Figure 4). However, if we remove the six clustered communities this relationship vanishes. In addition, as we get closer to the true local species richness by adding undetected species known to occur in the surrounding areas, communities become less-clustered phylogenetically (e.g., Letícia) or neutral (e.g., El Espinillar).

Overall, we suggest that these results indicate a potential artifact in the signal of clustering in the six significant communities. It is possible, perhaps likely, that additional field studies would recover additional species from those communities, erasing the signal of clustering which arises solely from Type II omission of phylogenetic lineages. Thus, we tentatively conclude that none of the communities we examined are likely to be significantly clustered by non-neutral ecological processes such as habitat filtering or competition, and that most, if not all, community phylogenetic structure in Neotropical anuran assemblages reflects random subsampling of the regional pool.

Figure 5. Regressions of z-scores and standardized log(species richness). Clustered communities have z-score ≤ -1.95 and overdispersed ≥ 1.95. Gray points represent the six phylogenetically clustered communities. Red line (adjusted R2 = 0.2, p = 0.006) indicates regression using species reported in community’s surveys, gray line (adjusted R2 = 0.2, p

= 0.007) correspond to regression of removing REAJ and adding species occurring in the surrounding areas of Leticia and El Espinillar, and gray dotted-line (adjusted R2 = 0.07, p

= 0.136) represents the regression removing the six clustered communities. Open dots indicate the new z-scores of Letícia and El Espinillar after adding missing species.

Discussion

Evolutionary and ecological processes affect the phylogenetic structure of communities (Weber et al. 2017). If the resulting phylogenetic pattern is clustered compared to the neutral model, this might be due to high in situ speciation at the regional pool (e.g., where species from a single lineage are over-represented) followed by colonization of communities by closely related species, or exclusion of distant related species ecologically similar but with low competitive abilities. Alternatively, habitat filtering for a given trait shared among closely related taxa may explain clustering. Finally, competitive exclusion may result in clustering if species differ greatly in traits correlated with phylogenetic distance that determines competitive dominance (Mayfield and Levine

2010), but this prediction lacks evidence in natural systems (e.g., Bennett et al. 2013).

In contrast, phylogenetic overdispersion may originate from competition if closely related taxa overlap greatly in traits that are positively related to phylogenetic distance

(Mayfield and Levine 2010), or from habitat filtering if colonization is dominated by distantly related taxa with converging traits to survive under a given set of environmental factors. Colonizers would be functionally dissimilar to residents (e.g., diet guild, microhabitat). Alternatively, as shown in our results, most or all observed phylogenetic structure of communities might simply reflect the long-term operation of biogeographic processes such as speciation, extinction, and geographic movement (i.e., colonization).

Our results show that the phylogenetic pattern of most communities of Neotropical frogs is driven by stochastic variation in local extinction, colonization, and colonization decline. In addition, we identified six communities in the western Amazon and Chaco that are likely to have more phylogenetically closely related species. However, clustering might

reflect species omission in surveys given that most highly diverse and well-sampled communities of Neotropical frogs are neutral. Furthermore, adding misrepresented species in communities that were recovered as phylogenetically clustered under the dynamic model of community assembly makes those communities less clustered or neutral. The regional pools, in turn, are structured by a host of ecological and evolutionary factors (Wiens and

Donoghue 2004; Allen et al. 2006; Buckley and Jetz 2008; Buckley et al. 2010; Wiens et al. 2011; Davies and Buckley 2012; Jetz et al. 2012; Pyron and Wiens 2013; Pyron 2014;

Jetz and Pyron in press).

In our study, undersampling bias is a likely explanation for communities being recovered as clustered. For instance, in El Espinillar, E. ovalis and O. americanus occur very nearby; their presence in other communities render those communities to be recovered as neutral. Similarly, in the western Amazon, undersampling of some lineages may explain strong patterns of clustering (Figure 3). Compare Iquitos with PNSD, both with 111 species recorded (Appendix 3). The few genera present in Iquitos and absent in PNSD may simply be a sampling bias. The nearby communities of REAJ and RE Riozinho da Liberdade

(RERL) present similar patterns. In a quantitative sense, Lynch (2005) demonstrated that the frog fauna of Letícia was heavily undersampled before his fieldwork; while we record

89 valid species from his analysis (up from 40 species in Lynch 2000), the author suggests more species may be present based on nearby localities, and that this shortfall is due mostly to poor sampling methodologies (Lynch 2005).

An accurate census of the local community at localities such as Leticia, PNSD,

REAJ, and RERL seems likely to yield a less-clustered result, as seen from many of the most diverse, long-term, high-effort areas such as Iquitos, EB Santa Lúcia, and La Selva.

Recent taxonomic studies have described several new species of Amazonian frogs (e.g.

Funk et al. 2012; Carminer et al. 2016; Peloso et al. 2014). We suggest that the true local species pool of these high-diversity western Amazon communities is much different than currently believed, and that they will likely tend towards a neutral estimate of assembly processes as their diversity becomes better characterized. The most-accurately characterized local faunas across the warm tropical ecoregions of Central and South

America uniformly demonstrate a lack of significant signal for non-neutral assembly processes. Thus, local species pools of frogs across the Neotropical ecoregions are more likely to represent a random sample of the regional pool.

Contrastingly, we entertain several possible hypotheses for the observation of clustering (assuming the pattern is genuine) restricted to a few communities in the western

Amazon and southern Chaco. First, Neotropical ecoregions notably comprise geographically localized and distinct regional species pools of frogs (Olson et al. 2011;

Vilhena and Antonelli 2015; Elder et al. 2016), which are likely to have been sorted out into ecoregions through processes such as environmental filtering from the continental pool by climate and geological events (e.g. biogeographic filtering) leading to the observed biogeographic overrepresentation (e.g. Olson et al. 2001). Potentially, hyper-diverse communities such as in the western Amazon are localized in terms of occurrence (i.e., widespread species truly are present/absent in nearby communities). This might be due to high in situ speciation within the Amazon relative to other ecoregions. For instance, treefrogs (Hylidae), poison frogs (Dendrobatidae and Aromobatidae), glass frogs

(Centrolenidae), and Terraranan frogs represented a large proportion of those high- diversity Amazonian communities. Thus, over-representation of local micro-endemics

from regionally hyper-diverse lineages given different ecological opportunities for species

(e.g., Moen and Wiens 2017) may lead to clustering. However, this hypothesis remains to be tested.

Second, there may be a similar or related pattern of biogeographic regionalization of hyper-diverse communities. For instance, the community at Mindo, Ecuador with 53 species is highly clustered. It is possible that, given long-term areas of refugia and endemism (e.g., high phylogeographic endemism in certain areas of the Atlantic Forest;

Carnaval et al. 2014), unique geographic areas have narrowly endemic, hyper-diverse segments of the regional pool, containing closely related species that are not widespread throughout the ecoregion. Thus, the local-to-regional co-occurrence of these species may not reflect the influence of processes such as habitat filtering or competition, but rather the historical signature of climatic shifts or similar biogeographic processes. Thus, a community sampled from such an area may be "clustered" through artifacts of biogeography that are not accounted for by the DAMOCLES model, as it still considers the entire regional pool without accounting for potential regionalization within that pool.

Different segments of the tropical Andes might thus have very different local assemblages that have evolved in situ, which might all be recovered as "clustered," though not due to non-neutral ecological processes (sensu Webb et al. 2002).

Future Directions

Although we recovered the phylogenetic distribution of most communities of

Neotropical frogs represent a random sample of the regional pool, trait variation within those communities still might be driven by deterministic variation of ecological processes

such as habitat filtering for certain traits and competition. In addition, those processes might affect differently more labile traits such as behavior (e.g., frog advertisement calls) and more conserved traits such as body shape (e.g., Blomberg et al. 2003) to structure community trait variation. Thus, examining the distribution of functional traits may reveal stronger patterns than phylogenetic structure and species richness.

New macroevolutionary models allow for testing the relative effect of species interaction on trait evolution while reconstructing the pool of interacting species that should over time affect trait variation (Nuismer and Harmon 2014; Drury et al. 2016;

Manceau et al. 2017). Using such an approach, it is possible to compare whether the expected phylogenetic structure under a scenario in which colonization and local extinction rates are independent of species traits to a scenario in which colonization depends on the distribution of species traits and their variation through time.

Here we have corroborated previous studies showing that Amazonian communities are among the most diverse across the Neotropics (Wiens et al. 2011), in this case for frogs.

These communities are unique as they have lower z-scores (suggesting phylogenetic clustering) than other high diversity communities (e.g., Atlantic Forest). However, doubt remains as to whether this pattern is an artifact caused of undersampling, and most

Neotropical communities do not deviate from the null model. As we add data on these frog communities through both novel field surveys of Amazonian communities and digitizing existing but currently inaccessible data in the print literature, we will better understand the ecology and evolution of these diverse frog assemblages.

Chapter 4: Fully sampled phylogenies of squamates reveal evolutionary patterns in

threat status

Abstract Macroevolutionary rates of diversification and anthropogenic extinction risk differ vastly throughout the Tree of Life. This results in a highly heterogeneous distribution of evolutionary distinctiveness (ED) and threat status among species. We examine the phylogenetic distribution of ED and threat status for squamates (amphisbaenas, lizards, and snakes) using fully-sampled phylogenies containing 9,574 species and expert-based estimates of threat status for ~4,000 species. We ask whether threatened species are more closely related than would be expected by chance and whether high-risk species represent a disproportionate amount of total evolutionary history. We found currently-assessed threat status to be phylogenetically clustered at broad level in Squamata, suggesting it is critical to assess extinction risks for close relatives of threatened lineages. Our findings show no association between threat status and ED, suggesting that future extinctions may not result in a disproportionate loss of evolutionary history. Lizards in degraded tropical regions (e.g.,

Madagascar, India, Australia, and the West Indies) seem to be at particular risk. A low number of threatened high-ED species in areas like the Amazon, Borneo, and Papua New

Guinea may be due to a dearth of adequate risk assessments. It seems we have not yet reached a tipping point of extinction risk affecting a majority of species; 63% of the assessed species are not threatened and 56% are Least Concern. Nonetheless, our results show that immediate efforts should focus on geckos, iguanas, and chamaeleons, representing 67% of high-ED threatened species and 57% of unassessed high-ED lineages.

Introduction

Human activities are the primary driver of the ongoing sixth mass-extinction event

(Ceballos et al. 2015). The list of extinct taxa has increased significantly within the past 20 years and extinction risks are spread across the Tree of Life, including groups such as plants

(Vamosi and Wilson 2008), marine invertebrates (Solan et al. 2004), terrestrial arthropods

(den Boer 1990), and many vertebrate lineages (Hoffmann et al. 2010). Among vertebrates, the extinction risks of extant squamates (lizards, snakes, and amphisbaenas) are not well explored compared to birds, mammals, and amphibians (Isaac et al. 2007, 2012; Fritz and

Purvis 2010; Jetz et al. 2014; Meiri and Chapple 2016). This is despite the fact that squamates are one of the most diverse and widespread lineages of terrestrial vertebrate, with >9,500 extant species.

A recent global analysis of extinction risks in squamates assessed only 15% (1450) of the >9,500 squamate species (Böhm et al. 2013). As of 2015, the IUCN Red List has evaluated the threat status of about 4,000 squamates, which includes approximately 70%

(753 of 1,073) of squamate genera. These recent assessments of extinction risk and methodological advances in phylogenetic estimation allow us to explore the phylogenetic pattern of threat status in squamates, and their degree of unique Evolutionary

Distinctiveness (ED). The ED metric measures the proportion of phylogenetic diversity

(total evolutionary history, measured as the sum of branch lengths in millions of years weighted by the number of tips sharing that branch) represented by each individual species

(Isaac et al. 2007).

First, we ask if species with high extinction-risk are more closely related than would be expected by chance. For example, certain lineages might be more prone to extinction

due to specific traits (Purvis et al. 2000; Cardillo et al. 2008; Fritz et al. 2009; Lee and Jetz

2010). This may be due to extinction risks associated with ecology and life-history factors such as body size, generation time, and population growth-rate, which are likely to be conserved among closely related species. In addition, extinction risk might correlate with environmental heterogeneity, geographical range, and human populations, which are expected to affect closely related species if they are endemic to regions highly impacted by climate change or habitat modification. In contrast, extinction risk might be phylogenetically overdispersed if threats are concentrated in specific geographic areas that contain a large number of distantly related lineages (such as Madagascar and Australia), but are relatively low elsewhere. A lack of phylogenetic clustering or overdispersion might indicate that systemic threats are affecting a wide cross-section of species across the globe

(Sinervo et al. 2010; Böhm et al. 2013).

Second, we ask if species at high risk of extinction represent a disproportionate amount of the total evolutionary history of squamates; if threatened lineages have higher

ED. It has been shown that ED varies across lineages. For example, median ED is 6.2Ma for all extant birds (Jetz et al. 2014), 7.9Ma for mammals (Isaac et al. 2007), and 12.5Ma for amphibians (Isaac et al. 2012). However, ED for squamates is unknown. Given limited resources for conservation, and often forcing triage in management decisions, threatened lineages with higher ED may be "worth more" than others (Bottrill et al. 2008; Isaac et al.

2012).

A species on a long branch with few relatives, such as the genus Sphenodon

(tuataras), has very high ED, and represents a large proportion of the extant evolutionary diversity. From an evolutionary perspective, this would be a massive loss if they became

extinct. Conversely, in a large radiation of relatively young species, such as the South

American genus Liolaemus (tree iguanas), the loss of any single species may not have a huge impact on the amount of ED remaining in the group. Identifying whether extinction- prone species have particularly high ED, and which threatened species have the highest distinctiveness, allows more targeted allocation of funds and effort in the difficult field of conservation triage (Bottrill et al. 2008).

At least 929 of the 1,073 squamate genera have species represented by DNA- sequence data (e.g. in GenBank), totaling ~5,500 of the ~9,500 species (~60%). Thus, we can create robust fully-sampled phylogenies containing all species, based on large-scale phylogenetic inference and taxonomic imputation (Jetz et al. 2012). Our overall goal is to identify lineages and geographic areas of highest conservation priority. We restrict this study to Lepidosauria (Squamata [lizards, snakes, and amphisbaenas] and

Rhynchocephalia [the tuatara]), to investigate a single well-known clade. Future studies might extend this research to crocodilians and turtles. We suggest future avenues for research, highlight unassessed taxa with the highest ED that merit immediate assessment, and identify the threatened taxa with the highest ED representing high priorities for current conservation efforts.

Material and Methods

Threatened vs. non-threatened species

The most recent edition of the IUCN Red List of Threatened Species includes 4,169 species of squamates, plus the one species of tuatara (total of 4,170 species; IUCN 2015).

Based on the IUCN Red List Categories and Criteria, we defined extinction risk as a binary

trait: threatened or non-threatened. As per IUCN definitions, species classified as Critically

Endangered (CR: 133 species), Endangered (EN: 310 species), and Vulnerable (VU: 326 species) were considered ‘threatened.’ The 14 species classified as Extinct (EX) were also considered in our ‘threatened’ category because we were interested in species that have gone or are close to extinction. Species listed as Near Threatened (NT: 268 species), and

Least Concern (LC: 2,245 species) were considered ‘non-threatened’. Data Deficient (DD:

775 species) and Unassessed (UA: 5,684 species) species were not included in the analyses.

The species-level taxonomy and geographic distributions (limited to countries or other large, sub-national regions) follow The Reptile Database (Uetz and Hošek 2015).

Species-level distribution maps are not yet available for squamates. One hundred species in the IUCN dataset were removed because they did not match valid species listed in the

Reptile Database (Appendix 1). Thus, the final data matrix comprises 3,296 squamate species and Sphenodon (i.e., out of the total 4,170 species, 100 were removed and 775 DD were not included), with 783 classified as threatened and 2,513 species as non-threatened

(Appendix 2). Despite relative incompleteness at the species level, all major lineages and geographic areas are represented by assessments of at least a few species (IUCN 2015), with 34% (3,295) of the 9,754 squamate species in the reference taxonomy from the Reptile

Database, and 70% (751) of 1,073 genera (see below).

We analyze squamates as a whole and the following clades separately (Table 1):

Gekkota (geckos and relatives), Scincoidea (skinks and relatives), Lacertoidea (teiids, lacertids, amphisbaenians, and relatives), Anguimorpha (glass lizards, monitors, and relatives), Iguania (iguanas, chameleons, and relatives), and Serpentes (snakes). This infraorder classification follows previous authors (Jones et al. 2013; Pyron et al. 2013). We

omitted dibamids from clade-specific analyses because only two have been assessed, both non-threatened.

Table 1. Summary of the IUCN data and reference taxonomy of the Reptile Database used in the present study. Dibamidae and Sphenodon were omitted.

Infraorder Total genera IUCN genera Total species IUCN species

Anguimorpha 20 15 (75%) 221 96 (43%)

Gekkota 122 91 (75%) 1602 473 (30%)

Iguania 115 84 (73%) 1759 586 (33%)

Lacertoidea 119 72 (61%) 892 270 (30%)

Scincoidea 172 117 (68%) 1721 510 (30%)

Serpentes 509 371 (73%) 3491 1358 (39%)

Fully-sampled phylogeny of Squamata

We followed recent protocols, using a combination of phylogenetic inference and taxonomic assignment, to generate a posterior distribution of fully-sampled phylogenies for Squamata using the Phylogenetic Assembly with Soft Taxonomic Inferences (PASTIS) approach (Jetz et al. 2012, Thomas et al. 2013). The data and full methods are presented in detail in the supporting information (data available in Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.db005), and briefly outlined here. Using the March 2015 update of the Reptile Database, we created a baseline taxonomy consisting of 9,754 currently recognized squamates, plus the tuatara. We revised an existing molecular supermatrix (Pyron et al. 2013) to include all available sequence data for 17 genes, 7

mitochondrial and 10 nuclear, for 5,415 squamates plus the tuatara. Using ExaML/RAxML

(Stamatakis 2006), we estimated the Maximum-Likelihood (ML) topology for these species. This topology was enforced as a constraint for all subsequent analyses, for those species with data (Appendix Fig. 1). Topological variability is visualized as the coefficient of variation in estimated ED values across the 100 trees (Appendix Fig. 2).

We identified 175 subclades which accounted for all 9,754 species. We then extracted a subset of the matrix containing 175 species representing each subclades, which we dated using MrBayes 3.2 (Ronquist et al. 2012) under a relaxed-clock model, with node- age calibrations taken from a recent integrative stratigraphic and molecular assessment of squamate divergence times (Jones et al. 2013). For each the 175 subclades, we estimated trees scaled to relative time under the same relaxed-clock model. In these analyses, the topology of species with DNA-sequence data was fixed, and the remaining unsampled species were assigned randomly within their genus or higher-level clade. From these, one subclade tree from each subclade was grafted onto its parent lineage on the backbone tree, with the root age then re-scaled to absolute time.

Overall, this yields a distribution of 10,000 trees containing 9,754 species (data available in Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.db005). In each of these trees, the ML topology for 5,415 species is constant, and the placement of the unsampled species is drawn from the posterior distribution of their possible locations within each genus or higher-level clade. Thus, the 10,000 trees are limited to a smaller, more probable region of tree-space that accounts for the known phylogenetic relationships of sampled species, and the known taxonomic classification of unsampled species. For most analyses, we used the all-compatible consensus of these 10,000 trees, which integrates

over the phylogenetic uncertainty of the missing species by collapsing poorly known clades into polytomies. For others (described below), we used also a sample of 100 trees from the posterior to calculate a range of values.

This and similar approaches have shown good results for applications involving evolutionary rates and distinctiveness of threatened species, even for unsampled taxa (Isaac et al. 2012, Jetz et al. 2012, 2014). While these trees may not be suitable for estimating quantities such as rates of continuous-character evolution, simulations show them to be adequate for assessing branch-length related measures such as diversification rate, and by extension, ED (Rabosky 2015). Thus, they should be useful for creating null models of the distribution of threat status that are conservative with respect to remaining phylogenetic uncertainty.

Phylogenetic measures of biodiversity

The measures we used to identify evolutionary patterns in threat status (see Helmus et al. 2007) are Phylogenetic Species Variability (PSV), which investigates average phylogenetic distance among all threatened species (hereafter ‘average relatedness’), and

Phylogenetic Species Clustering (PSC), which investigates average phylogenetic distance of threatened species to its non-threatened closest relative (hereafter ‘nearest-NT- neighbor’). Values for both metrics range from 0 to 1. The PSV metric measures a hypothetical, neutral, continuously-valued trait shared by all species, which evolves randomly and independently along the phylogeny. When PSV is 1, the species phylogeny is a star (e.g. polytomy), indicating that the sampled species are maximally unrelated

(overdispersed). The interaction between speciation and extinction through time drives

diversification, increases relatedness, and consequently decreases PSV, indicating reduced variability and maximum relatedness (clustering) among species as PSV goes to 0. For PSC values approaching 1, species are maximally unrelated at the tips of the phylogeny

(overdispersed), whereas values of 0 indicate strong clustering.

The PSV and PSC metrics are comparable to Net Relatedness Index (NRI) and

Nearest Taxon Index (NTI), respectively (Helmus et al. 2007). However, NRI and NTI require an extra step to standardize their variance across communities and center their means at 0 using random selection from the species pool, while PSV and PSC measure phylogenetic signal that is not confounded with species richness. In addition, they are standardized against a hypothetical community of species that are maximally unrelated (i.e. a star phylogeny).

We tested for nonrandom phylogenetic patterns by randomly shuffling cells within rows (Kembel and Hubbell 2006; Helmus et al. 2007; see also Webb et al. 2002). The null hypothesis assumes that if the expected phylogenetic variance of threatened species is equal to the variance of the entire squamate species pool, then the resulting phylogenetic pattern is random. In contrast, if the pool of threatened species does not represent a random sample from the total species pool, then the phylogenetic pattern of extinction risk is either

‘clustered’ or ‘overdispersed.’ If current extinction risk is more likely to affect a group of closely related species the pattern is ‘clustered,’ whereas if extinction risk is more likely to affect a group of more distantly related species, the pattern is ‘overdispersed.’

For the null hypotheses, we generated 100 random sets of threatened species out of the total species pool and estimated PSV and PSC for each set, using the consensus tree

(Null 1). In addition, we tested whether PSV of the whole species pool differs from the

threatened species by sampling 100 trees from the posterior, and estimating PSV for each tree to create a distribution (Null 2). For PSV, both the random sets of threatened species with estimations using the consensus tree, and the comparison between the whole species pool and the threatened species across a sample of 100 posterior trees are valid nulls.

For PSC, the null hypothesis using the consensus tree and random sets of threatened species is ideal (e.g. Null 1), because polytomies make estimates more conservative in terms of tip clustering. Mean observed values within 2.5% and 97.5% quantiles of the null distribution were considered ‘random,' values under the 2.5% quantile were considered

‘clustered,’ and values above the range of 97.5% quantile were considered ‘overdispersed.’

We ran analyses in R version 3.2.2 (Core Team, 2015), using the package ‘picante’

(Kembel et al. 2010) and ‘caper’ (Orme et al. 2015).

The lack of assessment for the remaining 5,684 squamate species prevents us from testing hypotheses correlating causes of threat with species-level phylogenetic patterns (as highlighted in Meiri and Chapple 2016). For instance, if the result is clustered, we might identify which clusters of closely related species significantly drive the pattern, and test whether similar underlying threats affects distinct clusters of lineages. Therefore, we only highlight preliminary patterns. Below, we discuss potential future studies to address root causes at the species level.

ED and extinction risk

We sampled 100 trees from the posterior and calculated ED values for all 9,755 species using the R package 'caper,' taking the median across each species (Figs. 1–3). For the 3,296 species with a threat status (Fig. 4), we performed a simple t-test in R version

3.2.2 (R Core Team 2016) to determine if median ED differed between threatened and non- threatened species. Given that ED is expected to co-vary strongly with phylogenetic relatedness, we then performed a phylogenetic ANOVA in the R package 'geiger,' simulating 1000 null F-statistics to create a test distribution for comparison with the observed value.

Finally, we identified the upper 25% quartile of highest-ED Unassessed species.

These are the taxa which should be the focus of immediate assessment efforts, to plan future conservation strategies interested in maintaining maximum diversity of evolutionary history. We also identified the upper 25% quartile of highest-ED threatened lineages, which would represent the largest loss in evolutionary history if they were to go extinct, and thus should have conservation efforts increased. From these, we summarized the infraorders representing a majority of species, which indicate the broad-scale direction of priorities. We also present the top-50 species from each list to illustrate the geographic and taxonomic patterns. Appendix 4 and 5 are available at https://drive.google.com/open?id=0B-k1-50WDeJZdHZacWpMR0RoWGs. Additional supplemental files and further methodological information are deposit at in Dryad Digital

Repository: http://dx.doi.org/10.5061/dryad.db005.

Results

Phylogenetic patterns of threat status

There are strong phylogenetic patterns in threat for both squamates as a whole (Figs.

1-3) and the individual subclades (Tables 2 and 3). The permutation tests of PSV showed phylogenetic clustering, indicating that in terms of average relatedness, extinction

prone squamates are more closely related than expected by chance (Table 2). By analyzing each infraorder of squamates separately, threatened species of Anguimorpha, Lacertoidea, and Serpentes showed phylogenetic clustering as well (Table 2). In contrast, threats in

Gekkota were phylogenetically overdispersed, whereas Iguania and Scincoidea did not differ from random (Table 2).

For PSC, the phylogenetic pattern of threatened squamate species is overdispersed

(Table 3). This shows that the nearest-NT-neighbors of extinction-prone species are more distantly related than expected at random. All major clades of squamate showed phylogenetic overdispersion, except Anguimorpha, which did not differ from random

(Table 3).

Table 2. Phylogenetic Species Variability (PSV) results. Values in the columns Null 1 and

Null 2 correspond to the 5% quantile of PSV.

mean PSV Null 1 Null 2 Phylostructure

Squamates 0.601 0.616-0.628 0.571-0.615 Clustered/random

Anguimorpha 0.424 0.589-0.682 0.608-0.706 Clustered

Gekkota 0.852 0.803-0.844 0.815-0.865 Overdispersed

Iguania 0.747 0.717-0.763 0.724-0.771 Random

Lacertoidea 0.707 0.762-0.791 0.763-0.793 Clustered

Scincoidea 0.553 0.498-0.577 0.502-0.588 Random

Serpentes 0.587 0.618-0.679 0.592-0.684 Clustered

Table 3. Phylogenetic Species Cluster (PSC) results. Values in the column Null 1 correspond to the 5% quantile of PSC.

Mean PSC Null 1 Phylostructure

Squamates 0.926 0.883-0.893 Overdispersed

Anguimorpha 0.873 0.807-0.886 Random

Gekkota 0.701 0.562-0.633 Overdispersed

Iguania 0.872 0.787-0.814 Overdispersed

Lacertoidea 0.884 0.764-0.812 Overdispersed

Scincoidea 0.886 0.813-0.834 Overdispersed

Serpentes 0.870 0.795-0.821 Overdispersed

Figure 1. Fully sampled phylogeny of 9,755 Lepidosauria species with branches color by

ED values. Warmer and colder colors represent high and low ED, respectively. In the inset, the y-axis represents proportion of species and the x-axis ED in millions of years, on a log scale.

Figure 2. Fully sampled phylogeny trimmed to 3,296 Lepidosauria species assessed by

IUCN with branches color by ED values. Warmer and colder colors represent high and low

ED, respectively. In the inset, the y-axis represents proportion of species and the x-axis ED in millions of years, on a log scale. In light gray is the distribution of ED across all lepidosaurs (Fig. 1), and dark gray is the distribution of ED of the assessed lineages in the tree.

Figure 3. Fully sampled phylogeny trimmed to 783 Lepidosauria species classified as threatened with branches color by ED values. Warmer and colder colors represent high and low ED, respectively. In the inset, the y-axis represents proportion of species and the x- axis ED in millions of years. In light gray is the distribution of ED across all lepidosaurs

(Fig. 1), and dark gray is the distribution of ED of the assessed lineages in the tree.

Figure 4. Variation in ED values (log[Ma]) across 9,755 Lepidosauria species. IUCN categories of threat status are Least Concern (LC), Near Threatened (NT), Vulnerable

(VU), Endangered (EN), Critically Endangered (CR), Data Deficient (DD), Extinct (EX), and Unassessed (UA).

ED and extinction risk

Squamates listed in IUCN threat categories do not have significant differences in median ED (Fig. 4). Futhermore, neither the standard nor the phylogenetically corrected t- tests showed significant difference of ED between threatened and non-threatened taxa.

Species in each group have a median ~12Ma of ED. Geckos and iguanians represent 57%

(595 species and 288, respectively) of high-ED linages with threat status unassessed

(Appendix 3). Among top-50 highest-ED Unassessed taxa, median ED ranges from

40.63Ma to 92.13Ma (Table 3). Areas such as the Amazon, Borneo, New Guinea, Africa, and southeast Asia have numerous high-ED species Unassessed by the IUCN (Table 3).

Among high-ED squamates classified as threatened, 90 species are listed as

Vulnerable, 84 as Endangered, 26 as Critically Endangered, and six are putatively Extinct

(Appendix 3). The top-25 highest-ED threatened taxa reveal urgent need to increase conservation priorities, again particularly in geckos and iguanians (Appendix 4). As might be expected, high-ED threatened taxa are also concentrated in areas of high anthropogenic pressures, such as Madagascar, Australia, the West Indies, and other tropical islands (e.g.,

New Caledonia). Gekkonids in particular represent a very large number of high-ED threatened species in these regions (Appendix 5).

Discussion

Overall, our results indicate that current extinction risk is typically clustered at a broad level across the Squamata Tree of Life, concentrated particularly in geckos and iguanians. This result differs from birds, in which imperiled species are more likely to be overdispersed (Jetz et al. 2014). Within infraorders of squamates, extinction risk is either clustered or overdispersed (see also Böhm et al. 2013), which apparently results from the co-occurrence of multiple lineages in regions of high anthropogenic threat. Median ED for squamates is similar to amphibians, around 12 Ma (Isaac et al. 2012). As in birds and mammals, there is little variation in ED between IUCN Red List categories (Arregoitia et al. 2013; Jetz et al. 2014).

Unsurprisingly, Madagascar tops the list of the top-25 highest ED threatened squamates (52% of species in Appendix 4) with a hugely diverse, high-ED, highly endemic

fauna that is heavily endangered (Andreone et al. 2008; Hannah et al. 2008; Jenkins et al.

2014). However, high-ED species also occur in areas such as the Amazon, Borneo, New

Guinea, Africa, and southeast Asia, as well as lineages that are less affected by climate change (e.g., snakes), but these high-ED species have not been assessed (Appendix 4).

Assessments of the remaining high-ED species (e.g., Table 3) will be crucial for complete conservation-management strategies worldwide, though many will likely be Data

Deficient, with geographical range and population trends difficult to ascertain. Predicting threat status from trait data may assist in this effort (see below).

The low number of high-ED threatened species in areas like the Amazon, Borneo, and Papua New Guinea may thus represent a dearth of adequate risk assessments, rather than a true lack of human impact or extinction risk (Curran et al. 2004; Malhi et al. 2007;

Shearman and Bryan 2010; Natusch and Lyons 2012). These regions are known to have many diverse and evolutionarily distinct lineages of squamates (da Silva and Sites 1995;

Das 2006; Allison 2007), and in fact several of those are represented in Appendix 4. For instance, the Bornean earless monitor Lanthanotus is a monotypic lineage with high ED

(61 Ma), and known from very few specimens. This lizard is likely to be highly threatened, but it has not been assessed by the IUCN. We suggest that threat assessments of squamates from regions of high biodiversity, such as Amazonia and the mainland and archipelagos of southeast Asia, are a crucial next step toward effective conservation planning.

Species with both high and low levels of ED alike exhibit high extinction-risk, though a subset of high-ED, high-threat species present particularly urgent priorities for conservation effort (Appendix 5). Overall, lizards (e.g. geckos and iguanas) in degraded tropical regions (e.g. Madagascar, India, Australia, and the West Indies) seem to be at

particular risk (Appendix 5). This is likely due to temperature increases, caused by climate change or habitat modification, that affect heliotropic thermoconformers (see Huey et al.

2009; Sinervo et al. 2010). However, diurunal lizards are not the only species at risk. While we have not done an explicit analysis of fine-scale geographic or trait-based correlates of extinction risk, the threatened species with the highest ED span a qualitatively wide range of geographic regions, body sizes, and life histories other than basking lizards (see Meiri et al. 2012, 2013; Feldman et al. 2016).

For example, the monotypic genus Shinisaurus (Shinisauridae) is a large, semi- aquatic lizard from cool tropical forests in Southeast Asia, and represents 103 Ma of ED, which is 29Ma more than the next highest lineage (Appendices 4 and 5). In contrast, the monotypic genus Xenotyphlops (Xenotyphlopidae) is a small, fossorial blindsnake from the dry regions of northern Madagascar, but represents a proportionally large 68 Ma of ED.

Despite their differences in body size, ecological traits, and life history, these species are both high ED, and highly threatened.

The situation is also not entirely bleak. To summarize, 63% of the assessed species are not threatened, and 56% are Least Concern. If these percentages hold across the

Unassessed species, then a total of 6,145 squamates are not threatened, and 5,462 are Least

Concern. The tuatara (Sphenodon) has a median ED value of 243 Ma, and is considered

Least Concern. Additionally, the highest-ED squamate species is the amphisbaenian

Rhineura floridana, the Florida Worm Lizard (Lacertoidea), with a median ED value of

130 Ma. This taxon is considered Least Concern, and is well protected in peninsular

Florida. This is not to say that threats do not exist and that urgent conservation-management

planning is not needed, but it seems we have not yet reached a tipping point of extinction risk affecting a majority of species or lineages.

The phylogenetic clustering of threat status suggests that is crucial to assess extinction risks for close relatives of threatened lineages. Therefore, future research should be concentrated in three major areas. First would be in-depth assessments of the almost

6,000 unsampled species, for robust evaluation of threat levels. Second would be to gather relevant spatial, geographic, ecological, and life-history data that could be used to evaluate predictors of extinction risk and identify clades and regions of exceptionally high conservation priority. Related to this, a third aim is to predict threat status using those data and models for species with little currently available data on population trends (Dickinson et al. 2014; Bland et al. 2015; Jetz and Freckleton 2015; Schachat et al. 2015; Bland and

Böhm 2016). Preliminarily, our results suggest that conserving primary habitats containing diverse assemblages of species is likely to be the best overall strategy for preserving squamate diversity, difficult as this may be.

References

Allen AP, Gillooly JF, Savage VM, Brown JH (2006) Kinetic effects of temperature on

rates of genetic divergence and speciation. Proceedings of the National Academy

of Sciences USA 103: 9130–9135.

Allison A. (2007) The herpetofauna of Indonesia's Papua Province, New Guinea. In The

Ecology of Papua (AJ Marshall and BM Beehler, eds.), pp 564–616. Periplus Press,

Singapore.

Amézquita A, Hödl W, Lima AP, Castellanos L, Erdtmann L, Araújo MC (2006) Masking

interference and the evolution of the acoustic communication system in the

Amazonian dendrobatid frog Allobates femoralis. Evolution 60(9): 1874–1887.

Amézquita A, Flechas SV, Lima AP, Gasser H, Hödl W (2011) Acoustic interference and

recognition space within a complex assemblage of dendrobatid frogs. Proceedings

of the National Academy of Sciences USA 108(41): 17058–17063.

AmphibiaWeb (2017) AmphibiaWeb: information on amphibian biology and

conservation. University of California, Berkeley, CA. Available at:

http://amphibiaweb.org/.

Andreone F, Carpenter AI, Cox N, et al. (2008) The challenge of conserving amphibian

megadiversity in madagascar. PLoS Biology 6: 943–946.

Arch VS, Grafe TU, Narins PM (2008) Ultrasonic signaling by a Bornean frog. Biology

Letters 4: 19–22.

Arregoitia LDV, Blomberg SP, Fisher DO (2013) Phylogenetic correlates of extinction risk

in mammals: species in older lineages are not at greater risk. Proceedings of the

Royal Society B, Biological Sciences 280: 20131092.

Audacity Team (2017) Audacity(R): free audio editor and recorder v2.1.3

Beaulieu JM, O’Meara BC (2016) Detecting hidden diversification shifts in models of trait

dependent speciation and extinction. Systematic Biology 65(4): 583–601.

Blair WF (1964) Isolating mechanisms and interspecies interactions in anuran amphibians.

The Quarterly Review of Biology 39: 334–344.

Bland LM, Böhm M (2016) Overcoming data deficiency in reptiles. Biological

Conservation 204: 16–22.

Bland LM, Collen B, Orme CDL, Bielby J (2015) Predicting the conservation status of

data-deficient species. Conservation Biology 29: 250–259.

Blomberg SP, Garland T, Ives AR (2003) Testing for phylogenetic signal in comparative

data: behavioral traits are more labile. Evolution 57: 717–745.

Blommers-Schlösser RMA (1975) Observations on the larval development of some

Malagasy frogs, with notes on their ecology and biology (Anura: Dyscophinae,

Scaphiophrynidae and Cophylinae). Beaufortia 24: 7–26.

Böhm M, Collen B, Baillie JEM, et al. (2013) The conservation status of the world's

reptiles. Biological Conservation 157: 372–385.

Bottrill MC, Joseph LN, Carwardine J, Bode M, Cook CN, Game ET, Grantham H, Kark

S, Linke S, McDonald-Madden E, Pressey RL, Walker S, Wilson KA, Possingham

HP (2008) Is conservation triage just smart decision making? Trends in Ecology

and Evolution 23: 649–654.

Bradbury JW, Vehrencamp SL (1998) Principles of Animal Communication. Sinauer,

Sunderland, MA, 697 pp.

Bromham L, Hua X, Cardillo M (2016) Detecting macroevolutionary self-destruction from

phylogenies. Systematic Biology 65: 109–127.

Brust DG (1993) Maternal brood care by Dendrobates pumilio: A frog that feeds its young.

Journal of Herpetology 27: 96–98.

Buckley LB, Davies TJ, Ackerly DD, Kraft NJB, Harrison SP, Anacker BL, Cornell HV,

Damshen EI, Grytnes J-A, Hawkins BA, McCain C, Stephens PR, Wiens JJ (2010)

Phylogeny, niche conservatism and the latitudinal diversity gradient in mammals.

Proceedings of the Royal Society B, Biological Sciences 277: 2131–2138.

Buckley LB, Jetz W (2008) Linking global turnover of species and environments

Proceedings of the National Academy of Sciences USA 105(46): 17836–17841.

Burbrink FT, Pyron RA, McKelvy AD, Myers EA (2015) Predicting community structure

in snakes on Eastern Nearctic islands using ecological neutral theory and

phylogenetic methods. Proceedings of the Royal Society B, Biological Sciences

282(1819): 20151700.

Buxton VL, Sperry JH (2017) Reproductive decisions in anurans: a review of how

predation and competition affects the deposition of eggs and tadpoles. BioScience

67(1): 26–38.

Caldart VM, Iop S, Cechin SZ (2014) Social interactions in a Neotropical stream frog

reveal a complex repertoire of visual signals and the use of multimodal

communication. Behaviour 151: 719–739.

Capranica RR (1965) The evoked vocal response of the bullfrog: A study of

communication by sound. Research Monography 33. MIT Press, Cambridge, MA.

Caramaschi U (2010) Notes on the taxonomic status of Elachistocleis ovalis (Schneider,

1799) and description of five new species of Elachistocleis Parker, 1927

(Amphibia, Anura, Microhylidae). Boletim do Museu Nacional 527: 1–30.

Cardillo M, Mace GM, Gittleman JL, Jones KE, Bielby J, Purvis A (2008) The

predictability of extinction: biological and external correlates of decline in

mammals. Proceedings of the Royal Society B, Biological Sciences 275: 1441–

1448.

Carnaval AC, Waltari E, Rodrigues MT, Rosauer D, VanDerWal J, Damasceno R, Prates

I, Strangas M, Spanos Z, Rivera D, Pie MR, Firkowski CR, Bornschein MR,

Ribeiro LF, Moritz C (2014) Prediction of phylogeographic endemism in an

environmentally complex biome. Proceedings of the Royal Society B, Biological

Sciences 281: 20141461.

Castroviejo-Fisher S, Guayasamin JM, Gonzalez-Voyer A, Carles Vilá (2014) Neotropical

diversification seen through glassfrogs. Journal of Biogeography 41: 66–80.

Cavender-Bares J, Kozak KH, Fine PVA, Kembel SW (2009) The merging of community

ecology and phylogenetic biology. Ecology Letters 12: 1–23.

Ceballos G, Ehrlich PR, Barnosky AD, Garcia A, Pringle RM, Palmer TM (2015)

Accelerated modern human-induced species losses: entering the sixth mass

extinction. Science Advances 1: e1400253.

Charlton BD, Reby D (2016) The evolution of acoustic size exaggeration in terrestrial

mammals. Nature Communications 7: 12739.

Cooper N, Thomas GH, Venditti C, Meade A, Freckleton RP (2016) A cautionary note on

the use of Ornstein Uhlenbeck models in macroevolutionary studies. Biological

Journal of the Linnaean Society 118: 64–77.

Crump ML (1974) Reproductive strategies in a tropical anuran community. University of

Kansas Museum of Natural History Miscellaneous Publications 61: 1–68.

Cuff AR, Randau M, Head J, Hutchinson JR, Pierce SE, Goswami A (2015) Big cat, small

cat: reconstructing body size evolution in living and extinct Felidae. Journal of

Evolutionary Biology 28(8): 1516–1525.

Curran LM, Trigg SN, McDonald AK, Astiani D, Hardiono YM, Siregar P, Caniago I,

Kasischke E (2004) Lowland forest loss in protected areas of Indonesian Borneo.

Science 303: 1000–1003. da Silva Jr. NJ, Sites Jr. JW (1995) Patterns of diversity of Neotropical Squamate reptile

species with emphasis on the Brazilian Amazon and the conservation potential of

indigenous reserves. Conservation Biology 9: 873–901.

Das I (2006) Photographic guide to snakes and other reptiles of Borneo. Ralph Curtis

Books, Sanibel island, 144 pp.

Davies TJ, Buckley LB (2012) Exploring phylogenetic history of mammals species

richness. Global Ecology and Biogeography 21: 1096–1105.

Day EH, Hua X, Bromham L (2016) Is specialization an evolutionary dead end? Testing

for differences in speciation, extinction and trait transition rates across diverse

phylogenies of specialists and generalists. Journal of Evolutionary Biology 29:

1257–1267.

de Sá FP, Zina J, Haddad CFB (2016) Sophisticated Communication in the Brazilian

Torrent Frog Hylodes japi. PLoS ONE 11(1): e0145444. den Boer PJ (1990) The survival value of dispersal in terrestrial arthropods. Biological

Conservation 54: 175–192.

Dickinson MG, Orme CDL, Suttle KB, Mace GM (2014) Separating sensitivity from

exposure in assessing extinction risk from climate change. Nature Scientific

Reports 4: 6898.

Drury J, Clavel J, Manceau M, Morlon H (2016) Estimating the effect of competition on

trait evolution using Maximum Likelihood inference. Systematic Biology 65(4):

700–710.

Duellman WE, Maness SJ (1980) The reproductive behavior of some hylid marsupial

frogs. Journal of Herpetology 14: 213–222.

Duellman (2005) Cusco Amazónico: The Lives of Amphibians and Reptiles in an

Amazonian Rainforest. Comstock Publishing Associates, Cornell University Press,

Ithaca, NY, 488 pp.

Edler D, Guedes T, Zizka A, Rosvall M, Antonelli A (2016) Infomap Bioregions:

Interactive mapping of biogeographical regions from species distributions.

Systematic Biology doi: 10.1093/sysbio/syw087

Feldman A, Pyron RA, Sabath N, Mayrose I, Meiri S (2016) Body sizes and diversification

rates of lizards, snakes, amphisbaenians and the tuatara. Global Ecology and

Biogeography 25: 187–197.

Felsenstein J (1985) Phylogenies and the comparative method. American Naturalist 125:

1–15.

Feng YJ, Blackburn DC, Liang D, Hillis DM, Wake DB, Cannatella DC, Zhang P (2017)

Phylogenomics reveals rapid, simultaneous diversification of three major clades of

Gondwanan frogs at the Cretaceous-Paleogene boundary. Proceeding of the

National Academy of Sciences USA 114: E5864–E5870.

Ferreira RB, Faivovich J, Beard KH, Pombal JP Jr (2015) The First Bromeligenous Species

of Dendropsophus (Anura: Hylidae) from Brazil's Atlantic Forest. PLoS ONE

10(12): e0142893.

FitzJohn R (2012) Diversitree: Comparative phylogenetic analyses of diversification in R.

Methods in Ecology and Evolution doi:10.1111/j.2041-210X.2012.00234.x.

Fletcher NF (2004) A simple frequency-scaling rule for animal communication. The

Journal of the Acoustical Society of America. 115(5): 2334–2338.

Fretwell SD, Lucas HJ (1970) On territorial behavior and other factors influencing habitat

distributions in birds. Acta Biotheoretica 19: 16–36.

Fritz SA, Bininda-Emonds ORP, Purvis A (2009) Geographical variation in predictors of

mammalian extinction risk: big is bad, but only in the tropics. Ecology Letters 12:

538–549.

Fritz SA, Purvis A (2010) Selectivity in mammalian extinction risk and threat types: a new

measure of phylogenetic signal strength in binary traits. Conservation Biology 24:

1042–1051.

Frost DR (2017) Amphibian Species of the World: an Online Reference. Version 6.0 (Date

of access). Electronic Database accessible at

http://research.amnh.org/herpetology/amphibia/index.html. American Museum of

Natural History, New York, USA.

Funk WC, Caminer M, Ron SR (2012) High levels of cryptic species diversity uncovered

in Amazonian frogs. Proceedings of the Biological Society B, Biological Sciences

279(1734): 1806–1814.

Futuyma DJ, Moreno G (1988) The evolution of ecological specialization. Annual Review

in Ecology and Systematics 19: 207–233.

Gamble T, Greenbaum E, Jackman TR, Russell AP, Bauer AM (2012) Repeated Origin

and Loss of Adhesive Toepads in Geckos. PLoS ONE 7(6): e39429

Garcia-Donato G, Forte A (2016) BayesVarSel: Bayesian Testing, Variable Selection and

model averaging in Linear Models using R. arXiv:1611.08118

Garland T, Dickerman AW, Janis CM, Jones JA (1993) Phylogenetic analysis of

covariance by computer-simulation. Systematic Biology 42: 265–292.

Gelman A, Rubin DB (1992) Inference from iterative simulation using multiple sequences.

Statistical Science 7: 457–511.

Gerhardt HC, Huber F (2002) Acoustic communication in insects and anurans. Chicago,

University of Chicago Press.

Gerhold P, Pärtel M, Liira J, Zobel K, Prinzing A (2008) Phylogenetic structure of local

communities predicts the size of the regional species pool. Journal of Ecology 96:

709–712.

Giaretta AA (1996) Reproductive specializations of the bromeliad hylid frog Phyllodytes

luteolus. Journal of Herpetology 30: 96–97.

Glaw F, Vences M (1994) A Field Guide to the Amphibians and Reptiles of Madagascar,

2nd ed. Moos Druck, Leverkusen, Germany, 480 pp.

Glor RE (2010) Phylogenetic insights on adaptive radiation. Annual Review of Ecology,

Evolution, and Systematics 41: 251–270.

Goicoechea N, De La Riva I, Padial JM (2010) Recovering phylogenetic signal from frog

mating calls. Zoologica Scripta 39(2): 141–154.

Goutte S, Dubois A, Howard SD, Marquez R, Rowley JJL, Dehling JM, Grandcolas P,

Rongchuan X, Legendre F (2016) Environmental constraints and call evolution in

torrent-dwelling frogs. Evolution 70-4: 811–826.

Grant T, Frost DR, Caldwell JP, Gagliardo R, Haddad CFB, Kok PJR, Means DB, Noonan

BP, Schargel WE, Wheeler (2006) Phylogenetic systematics of dart-poison frogs

and their relatives (Amphibia: Athesphatanura: Dendrobatidae). Bulletin of the

American Museum of Natural History 299: 1–262

Gringas B, Boeckle M, Herbst CT, Fitch WT (2013a) Call acoustic reflect body size across

four clades of anurans. Journal of Zoology 289(2): 143–150.

Gringas B, Mohandesan E, Boko D, Fitch WT (2013b) Phylogenetic signal in the acoustic

parameters of the advertisement calls of four clades of anurans. BMC Evolutionary

Biology 13: 134.

Haddad CFB, Giaretta AA (1999) Visual and acoustic communication in the Brazilian

torrent frog, Hylodes asper (Anura: Leptodactylidae). Herpetologica 55: 324–333.

Hannah L, Dave R, Lowry PP, et al. (2008) Climate change adaptation for conservation in

madagascar. Biology Letters 4: 590–594.

Harrington S, Reeder TW (2017) Rate heterogeneity across Squamata, misleading

ancestral state reconstruction and the importance of proper null model specification.

Journal of Evolutionary Biology 30(2): 313–325.

Helmus MR, Bland TJ, Williams CK, Ives AR (2007) Phylogenetic measures of

biodiversity. The American Naturalist 169(3): E68–E83.

HilleRisLambers J, Adler PB, Harpole WS, Levine JM, Mayfield MM (2012) Rethinking

community assembly through the lens of coexistence theory. Annual Review of

Ecology, Evolution, and Systematics 43: 227–48.

Ho LST, Ané C (2014) Intrinsic inference difficulties for trait evolution with Ornstein-

Uhlenbeck models. Methods in Ecology and Evolution 5(11): 1133-1146.

Hoffmann M, Hilton-Taylor C, Angulo A, Böhm M, Brooks TM, et al. (2010) The impact

of conservation on the status of the world's vertebrates. Science 330: 1503–1509.

Holt BG, Lessard JP, Borregaard MK et al. (2013) An Update of Wallace’s Zoogeographic

Regions of the World. Science 339(6115): 74–78.

Hoskin CJ, James S, Griggs GC (2009) Ecology and taxonomy-driven deviations in the

frog call-body size relationship across the diverse Australian frog fauna. Journal of

Zoology 278: 36–41.

Hua X, Bromham L (2016) Phylometrics: an R package for detecting macroevolutionary

patterns, phylogenetic metrics and backward tree simulation under a binary state

birth-death model. Methods Ecology and Evolution doi:10.1111/2041-

210X.12531.

Hubbell SP (2001) The Unified Neutral Theory of Biodiversity and Biogeography.

Princeton University Press, Princeton.

Huelsenbeck JP, Nielsen R, Bollback JP (2003) Stochastic Mapping of Morphological

Characters. Systematic Biology 52 (2): 131–158.

Huey RB, Deutsch CA, Tewksbury JJ, Vitt LJ, Hertz PE, Álvarez Pérez HJ, Garland Jr. T

(2009) Why tropical forest lizards are vulnerable to climate warning. Proceedings

of the Royal Society B, Biological Sciences 276: 1939–1948.

Isaac NJB, Redding DW, Meredith HM, Safi K (2012) Phylogenetically-informed

priorities for amphibian conservation. PLoS ONE 7: e43912.

Isaac NJB, Turvey ST, Collen B, Waterman C, Baillie JEM (2007) Mammals on the

EDGE: Conservation priorities based on threat and phylogeny. PLoS ONE 2: e296.

IUCN (2015) The IUCN Red List of Threatened Species. Version 2015.2.

. Downloaded on 23 June 2015.

Jetz W, Thomas GH, Joy JB, Hartmann K, Mooers AO (2012) The global diversity of birds

in space and time Nature 491(7424): 444–448.

Jetz W, Thomas GH, Joy JB, Redding DW, Hartmann K, Mooers AO (2014) Global

distribution and conservation of evolutionary distinctness in birds. Current Biology

24: 919–930.

Jetz W, Freckleton RP (2015) Towards a general framework for predicting threat status of

data-deficient species from phylogenetic, spatial and environmental information.

Philosophical Transaction of the Royal Society B, Biological Sciences 370:

20140016.

Jetz W, Pyron RA (in press) Regional variation and global imperilment of evolutionary

distinctiveness in amphibians.

Jones MEH, Anderson CL, Hipsley CA, Muller J, Evans SE, Schoch RR (2013) Integration

of molecules and new fossils supports a triassic origin for lepidosauria (lizards,

snakes, and tuatara). BMC Evolutionary Biology 13: 208.

Jungfer K-H (1985) Beitrag zur Kenntnis von Dendrobates speciosus O. Schmidt, 1857

(Salientia: Dendrobatidae). Salamandra 21: 263–280.

Jungfer K-H (1987) Froschlurche von Fortuna, Panama. II. Hylidae (2), Centrolenidae,

Dendrobatidae. Herpetofauna 10: 6–12.

Jungfer K-H (1996) Reproduction and parental care of the coronated treefrog, Anotheca

spinosa (Steindachner, 1864) (Anura: Hylidae). Herpetologica 52: 25–32.

Jungfer K-H, Weygoldt P (1999) Biparental care in the tadpole-feeding Amazonian

treefrog Osteocephalus oophagus. Amphibia-Reptilia 20: 235–249

Jungfer K-H, Weygoldt P, Juraske N (1996) Dendrobates vicentei, ein neuer Pfeilgiftfrosch

aus ZentralPanama. Herpetofauna 18: 17–26.

Keddy PA (1992) Assembly and response rules: two goals for predictive community

ecology. Journal of Vegetation Science 3: 157–164.

Kembel SW, Hubbell SP (2006) The phylogenetic structure of a Neotropical forest tree

community. Ecology 87: S86–S99.

Kembel SW, Cowan PD, Helmus MR, Cornwell WK, Morlon H, Ackerly DD, Blomberg

SP, Webb CO (2010) Picante: R tools for integrating phylogenies and ecology.

Bioinformatics 26(11): 1463–1464.

Kemp ME, Hadly EA (2016) Early Holocene turnover, followed by Stability, in a

Caribbean lizard assemblage. Quaternary Research 85(2): 255–261.

Krügel P (1993) Biologie und Ökologie der Bromelienfauna von Guzmania weberbaueri

im amazonischen Peru ergänzt durch eine umfassende Bibliographie der

Bromelien-Phytotelmata. Österreichischen Akademie der Wissenschaften

Biosystematics and Ecology Series 2: 1–93.

Lee TM, Jetz W (2010) Unravelling the structure of species extinction risk for predictive

conservation science. Proceedings of the Royal Society B, Biological Sciences 278:

1329–1338.

Lehtinen RM, Lannoo MJ, Assersug RJW (2004) Phytotelm-breeding anurans: past,

present and future research. Special Publications of Museum of Zoology,

University of Michigan 193: 1–9.

Lynch (2005) Discovery of the richness frog fauna in the world – An exploration of the

forests to the north of Letícia. Revista de la Academia Colombiana de Ciencias 29:

581–588.

MacArthur RH, Levins R (1967) The limiting similarity, convergence, and divergence of

coexisting species. American Naturalist 101: 377–385.

MacArthur JW, Wilson EO (1967) The Theory of Island Biogeography. Princeton

University Press, Princeton, 203 pp.

Maddison W, Midford P, Otto S (2007) Estimating a binary character’s effect on speciation

and extinction. Systematic Biology 56: 701–710.

Maddison WP, FirzJohn RG (2015) The unsolved challenge to phylogenetic correlation

tests for categorical characters. Systematic Biology 64(1): 127–136.

Magallón S, Gómez-Acevedo S, Sánches-Reyes LL, Hernández-Hernández T (2015) A

metacalibrated time-tree documents the early rise of flowering plant phylogenetic

diversity. New Phytologist 207: 437–453.

Malagoli LR, de Sá FP, Canedo C, Haddad CFB (2017) A New Species of Hylodes (Anura,

Hylodidae) from Serra do Mar, Southeastern Brazil: The Fourth with Nuptial

Thumb Tubercles. Herpetologica 73(2): 136–147.

Malhi Y, Roberts JT, Betts RA, Killeen TJ, Li W, Nobre CA (2007) Climate change,

deforestation, and the fate of the Amazon. Science 319: 169–172.

Martin WF (1972) Evolution of vocalization in the genus Bufo. In Evolution in the genus

Bufo (WF Blair ed.), pp 279–309. University of Texas Press, Austin, TX.

Mayfield MM, Levine JM (2010) Opposing effects of competitive exclusion on the

phylogenetic structure of communities. Ecology Letters 13: 1085–1093.

Meiri S, Brown JH, Sibly RM (2012) The ecology of lizard reproductive output. Global

Ecology and Biogeography 21: 592–602.

Meiri S, Bauer AM, Chirio L, Colli GR, Das I, Doan TM, Feldman A, Herrera F-C,

Novosolov M, Pafilis P, Pincheira-Donoso D, Powney G, Torres-Carvajal O, Uetz

P, Van Damme R (2013) Are lizards feeling the heat? A tale of ecology and

evolution under two temperatures. Global Ecology and Biogeography 22: 834–845.

Meiri S, Chapple DG (2016) Biases in the current knowledge of threat status in lizards.

Biological Conservation 204: 6–15.

Meyer E (1992) Erfolgreiche Nachzucht von Dendrobates granuliferus Taylor, 1958.

Herpetofauna 14: 11–21.

Moen DS, Smith SA, Wiens JJ (2009) Community assembly through evolutionary

diversification and dispersal in Middle American treefrogs. Evolution 63(12):

3228–3247.

Moen DS, Irschick DJ, WiensJJ (2013) Evolutionary conservatism and convergence both

lead to striking similarity in ecology, morphology and performance across

continents in frogs. Proceedings of Royal Society of London B, Biological Sciences

280: 20132156.

Moen DS, Wiens JJ (2017) Microhabitat and Climatic Niche Change Explain Patterns of

Diversification among Frog Families. The American Naturalist 190: 29–44.

Morrone JJ (2014) Biogeographical regionalisation of the neotropical region. Zootaxa

3782: 1–110.

Morton E (1975) Ecological sources of selection on avian sounds. American Naturalist

109: 17–34.

Narins PM, Feng AS, Fay RR (2007) Hearing and sound communication in amphibians.

Springer, New York, 380 pp.

Narvaes P, Rodrigues MT (2005) Visual communication, reproductive behavior, and home

range of Hylodes dactylocinus (Anura, Leptodactylidae). Phyllomedusa 4(2): 147–

158.

Natusch DJ, Lyons JA (2012) Exploited for pets: the harvest and trade of amphibians and

reptiles from Indonesian New Guinea. Biodiversity and Conservation 21(11):

2899–2911.

Nevo E, Capranica RR (1985) Evolutionary origin of ethological reproductive isolation in

Cricket Frogs, Acris. Evolutionary Biology 19: 147–214.

Nevo E, Schneider H (1976) Mating cell pattern of green toads in Israel and its ecological

correlate. Journal of Zoology 178: 133–145.

Núñez D, Maneyro R, de Sá RO (2004) Distribución geográfica de la fauna de anfibios del

Uruguay. Smithsonian Herpetological Information Service 134.

Ohmer ME, Robertson JM, Zamudio KR (2009) Discordance in body size, colour pattern,

and advertisement call across genetically distinct populations in a Neotropical

anuran (Dendropsophus ebraccatus). Biological Journal of the Linnean Society 97:

298–313.

Olson DM, Dinerstein E, Wikramanayake ED, Burgess ND, Powell GVN, Underwood EC,

D'Amico JA, Itoua I, Strand HE, Morrison JC, Loucks CJ, Allnutt TF, Ricketts TH,

Kura Y, Lamoreux JF, Wettengel WW, Hedao P, Kassem KR (2001) Terrestrial

Ecoregions of the World: A New Map of Life on Earth. BioScience 51: 933–938.

Orme D, Freckleton R, Thomas G, Petzoldt T, Fritz S, Isaac N, Pearse W (2015) caper:

Comparative analyses of phylogenetics and evolution in R, version 0.5.2.

Proceedings of the Royal Society B, Biological Sciences 283: 20160140.

Peixoto OL (1995) Associação de anuros e bromeliáceas na mata atlântica. Revista

Universidade Rural 17: 75–83.

Peloso PLV, Sturaro MJ, Forlani MC, Gaucher P, Motta AP, Wheeler WC (2014)

Phylogeny, taxonomic revision, and character evolution of the genera Chiasmocleis

and Syncope (Anura, Microhylidae) in Amazonia, with descriptions of three new

species. Bulletin of the American Museum of Natural History 136: 1–96

Pigot AL, Etienne RS (2015) A new dynamic null model for phylogenetic community

structure. Ecology Letters 18: 153–163.

Pough F, Magnusson W, Ryan M, Wells K, Taigen T (1992) Behavioral energetics. In

Environmental Physiology of the Amphibians (Feder M and Burggren eds.), pp.

395–436. University Chicago Press, Chicago.

Prinzing A, Reiffers R, Braakhekke WG, Hennekens SM, Tackenberg O, Ozinga WA, et

al. (2008) Less lineages – more trait variation: phylogenetically clustered plant

communities are functionally more diverse. Ecology Letters 11: 809–819.

Purvis A, Gittleman JL, Cowlishaw G, Mace GM (2000) Predicting extinction risk in

declining species. Proceedings of the Royal Society B, Biological Sciences 267:

1947–1952.

Pyron RA, Wiens JJ (2013) Large-scale phylogenetic analyses reveal the causes of high

tropical amphibian diversity. Proceedings of the Royal Society B, Biological

Sciences 280: 20131622

Pyron RA, Burbrink FT, Wiens JJ (2013) A phylogeny and revised classification of

squamata, including 4,161 species of lizards and snakes. BMC Evolutionary

Biology 13: 93.

Pyron RA (2014) Biogeographic analysis of amphibians reveals both ancient continental

vicariance and recent oceanic dispersal. Systematic Biology 63(5): 779–797.

Pyron RA, Burbrink FT (2014) Early origin of viviparity and multiple reversions to

oviparity in squamate reptiles. Ecology Letters 17(1): 13–21

Pyron RA, Wiens JJ (2013) Large-scale phylogenetic analyses reveal the causes of high

tropical amphibian diversity. Proceedings of the Royal Society of London B,

Biological Sciences 280 (1770): 20131622.

R Core Team (2016) R: A language and environment for statistical computing. R

Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-

project.org/.

Rabosky DL (2015) No substitute for real data: a cautionary note on the use of phylogenies

from birth–death polytomy resolvers for downstream comparative analyses.

Evolution 69 (12): 3207–3216.

Rabosky DL, Goldberg EE (2015) Model inadequacy and mistaken inferences of trait

dependent speciation. Systematic Biology 64: 340–355.

Raposo MA, Höfling E (2003) Overestimation of vocal characters in Suboscine taxonomy

(Aves: Passeriformes: Tyranni): causes and implications. Lundiana 4: 35–42.

Revell LJ (2012) phytools: an R package for phylogenetic comparative biology (and other

things). Methods in Ecology and Evolution 3: 217–223.

Rodríguez A, Börner M, Pabijan M, Gehara M, Haddad CFB, Vences M (2015) Genetic

divergence in tropical anurans: deeper phylogeographic structure in forest

specialists and in topographically complex regions. Evolutionary Ecology 29(5):

765–785.

Rodríguez RL, Araya-Salas M, Gray DA, Reichert MS, Symes LB, Wilkins MR, Safran

RJ, Höbel G (2015a) How acoustic signal scale with individual body size: common

trends across diverse taxa. Behavioral Ecology 26(1): 168–177.

Roelants K, Gower DJ, Wilkinson M, Loader SP, Biju SD, Guillaume K, Moriau L,

Bossuyt F (2007) Global patterns of diversification in the history of modern

amphibians. Proceeding of the National Academy of Sciences USA 104(3): 887–

892.

Röhr DL, Paterno GB, Camurugi F, Juncá FA, Garda AA (2016) Background noise as a

selective pressure: stream-breeding anurans call at higher frequencies. Organism

Diversity and Evolution 16: 269.

Ronquist F, Teslenko M, van der Mark P, Ayres DL, Darling A, Hohna S, Larget B, Liu

L, Suchard MA, Huelsenbeck JP (2012) MrBayes 3.2: Efficient Bayesian

phylogenetic inference and model choice across a large model space. Systematic

Biology 61: 539–542.

Rosenzweig ML (1995) Species Diversity in Space and Time. Cambridge University Press,

Cambridge, 460 pp.

Rosindell J, Harmon LJ, Etienne RS (2015) Unifying ecology and macroevolution with

individual-based theory. Ecology Letters 18: 472–482.

Ryan MJ (1980) Female mate choice in a Neotropical frog. Science 209: 523–525.

Ryan MJ (1988) Energy, calling, and selection. Integrative and Comparative Biology 28:

885–898.

Ryan MJ, Brenowitz EA (1985) The role of body size, phylogeny, and ambient noise in

the evolution of bird song. The American Naturalist 126: 87–100.

Safi K, Pettorelli N (2010) Phylogenetic, spatial and environmental components of

extinction risk in carnivores. Global Ecology and Biogeography 19: 352–362.

Salthe SN, Duellman WE (1973): Quantitative constraints associated with reproductive

modes in anurans. In Evolutionary Biology of the Anurans (Vial JL ed.), pp 229–

249. University of Missouri Press, Columbia.

San Mauro DS, Vences M, Alcobendas M, Zardoya R, Meyer A (2005) Initial

Diversification of Living Amphibians Predated the Breakup of Pangaea. American

Naturalist 165: 590–599.

Schachat SR, Mulcahy DG, Mendelson III JR (2016) Conservation threats and the

phylogenetic utility of IUCN Red List rankings in Incilius toads. Conservation

Biology 30: 72–81.

Schiøtz IR (1967) The treefrogs (Racophoridae) of West Africa. Spolia zoologica Musei

hauniensis 25: 1–346.

Shearman P, Bryan J (2010) A bioregional analysis of the distribution of rainforest cover,

deforestation and degradation in Papua New Guinea. Austral Ecology 36(1): 9–24.

Silvestro D, Zizka G, Schutle K (2013) Disentangling the effects of key innovations on the

diversification of Bromelioideae (Bromeliaceae). Evolution 68: 163–175.

Simpson GG (1944) Tempo and mode in evolution. Columbia University Press, New York.

Simpson GG (1953) The major features of evolution. Columbia University Press, New

York, 434 pp.

Sinervo B, Mendez-de-la-Cruz F, Miles DB, et al. (2010) Erosion of lizard diversity by

climate change and altered thermal niches. Science 328: 894–899.

Smith SA, Arif S, de Oca ANM, Wiens JJ (2007) A phylogenetic hot spot for evolutionary

novelty in Middle American treefrogs. Evolution 61(9): 2075–2085.

Solan M, Cardinale BJ, Downing AL, Engelhardt KAM, Ruesink JL, Srivastava DS (2004)

Extinction and ecosystem function in the marine benthos. Science 306(5699):

1177–1180.

Souza MB (2005) Anfíbios – Reserva Extrativista do Alto Juruá e Parque Nacional da Serra

do Divisor, Acre. Série Pesquisa e Monitoramento Participativo em Áreas de

Conservação Gerenciadas por Populações Tradicionais. Campinas, São Paulo, 77p.

Spasojevic MJ, Turner BL, Myers JA (2016) When does intraspecific trait variation

contribute to functional beta-diversity? Journal of Ecology 104: 487–496.

Stamatakis A (2006) Raxml-VI-HPC: Maximum likelihood-based phylogenetic analyses

with thousands of taxa and mixed models. Bioinformatics 22: 2688–2690.

Stebbins WC (1983) The Acoustic Sense of Animals. Harvard University Press,

Cambridge, MA, 168 pp.

Straughan IR (1973) Evolution of anuran mating calls: bioacoustical aspects. In

Evolutionary Biology of the Anurans (Vial JL ed.), pp 321–327. University of

Missouri Press, Columbia.

Stucky B (2013) Morphology, bioacoustics, and ecology of Tibicen neomexicensis sp. n.,

a new species of cicada from the Sacramento Mountains in New Mexico, U.S.A.

(Hemiptera, Cicadidae, Tibicen). ZooKeys 337: 49–71.

Thomas GH, Hartmann K, Jetz W, Joy JB, Mimoto A, Mooers AO (2013) PASTIS: An R

package to facilitate phylogenetic assembly with soft taxonomic inferences.

Methods in Ecology and Evolution 4: 1011–1017.

Thompson RL (1996) Larval habitat, ecology, and parental investment of Osteopilus

brunneus (Hylidae). In Contributions to West Indian Herpetology: a tribute to

Albert Schwartz (Powell R, Henderson RW eds.). Contributions to Herpetology 12:

259–269. Society for the Study of Amphibians and Reptiles, Ithaca, NY.

Title PO, Rabosky DL (2016) Do Macrophylogenies Yield Stable Macroevolutionary

Inferences? An Example from Squamate Reptiles. Systematic Biology syw102.

Tonini JFR, Beard KH, Ferreira RB, Jetz W, Pyron RA (2016) Fully-sampled phylogeny

of squamates reveal evolutionary patterns of threat status. Biological Conservation

204(Part A): 23–31.

Tucker CM, Cadotte MW, Carvalho SB, Davies TJ, Ferrier S, Fritz SA, Grenyer R, Helmus

MR, Jin LS, Mooers AO, Pavoine S, Purschke O, Redding DW, Rosauer DF,

Winter M, Mazel F (2016) A guide to phylogenetic metrics for conservation,

community ecology and macroecology. Biological Reviews doi:10.1111/brv.12252

Ueda H (1986) Reproduction of Chirixalus eiffingeri (Boettger). Scientific Reports of the

Laboratory in Amphibian Biology, Hiroshima University 8: 109–116.

Uetz P, Hošek J (2015) The Reptile Database, http://www.reptile-database.org, accessed

March 23, 2015.

Uyeda JC, Harmon LJ (2014) A Novel Bayesian Method for Inferring and Interpreting the

Dynamics of Adaptive Landscapes from Phylogenetic Comparative Data.

Systematic Biology 63(6): 902–918.

Uyeda JC, Pennel MW, Miller ET, Maia R, McClain CR (2017) The evolution of energetic

scaling across the vertebrate Tree of Life. American Naturalist 190(2): 000–000.

Van Bocxlaer I, Loader SP, Roelands K, Biju SD, Menegon M, Bossuyt (2010) Gradual

adaptation toward a range-expansion phenotype initiated the global radiation of

toads. Science 327(5966): 679–682.

Vidal-García M, Byrne PG, Roberts JD, Keogh JS (2014) The role of phylogeny and

ecology in shaping morphology in 21 genera and 127 species of Australo-Papuan

myobatrachid frogs. Journal of Evolutionary Biology 27(1): 181–192.

Vilhena DA, Antonelli A (2015) A network approach for identifying and delimiting

biogeographical regions. Nature Communications 6:6848.

Webb CO, Ackerly DD, McPeek MA, Donoghue MJ (2002) Phylogenies and community

ecology. Annual Review of Ecology, Evolution and Systematics 33: 475–505.

Weber MG, Wagner CE, Best RJ, Harmon LJ, Matthews B (2017) Evolution in a

Community Context: On Integrating Ecological Interactions and Macroevolution.

Trends in Ecology and Evolution http://dx.doi.org/10.1016/j.tree.2017.01.003

Wells KD (2001) The energetics of calling in frogs. In Anuran Communication (MJ Ryan

ed.), pp 45–60. Smithsonian Institution Press, Washington and London.

Wells KD (2007) The Ecology and Behavior of Amphibians. The University of Chicago

Press, Chicago, 1148 pp.

Weygoldt P (1980) Complex brood care and reproductive behavior in captive poison-arrow

frogs, Dendrobates pumilio O. Schmidt. Behavioral Ecology and Sociobiology 7:

329–332.

Weygoldt P (1981) Beobachtungen zur Fortpflanzungsbiologie von Phyllodytes luteolus

(Wied, 1824) im Terrarium. Salamandra 17: 1–11.

Weygoldt P, de Carvalho e Silva SP (1991) Observations on mating, oviposition, egg sac

formation and development in the egg-brooding frog, Fritziana goeldi. Amphihia-

Replilia 12: 67–80.

Wiens JJ, Donoghue MJ (2004) Historical biogeography, ecology and species richness.

Trends in Ecology and Evolution 19: 639–644.

Wiens JJ, Graham CH, Moen DS, Smith SA, Reeder TW (2006) Evolutionary and

ecological causes of the latitudinal diversity gradient in hylid frogs: treefrog trees

unearth the roots of high tropical diversity. American Naturalist 168: 579–596.

Wilczynski W, Ryan MJ (1988) The amphibian auditory system as a model for

neurobiology, behavior, and evolution. In The evolution of the amphibian auditory

system (B Fritzsch, MJ Ryan, W Wilczynski, TE Hetherington, W Walkowiak

eds.), pp. 3–12. Wiley, New York, NY.

Zimmermann H, Zimmermann E (1981) Sozialverhalten, Fortpflanzungsverhalten und

Zucht der Färberfrösche Dendrobates histrionicus und D. lehmanni sowie einiger

anderer Dendrobatiden. Z. Kölner Zoo 24: 83–99.