Ecology, Evolution and Organismal Biology Ecology, Evolution and Organismal Biology Publications
1-2016 Phylogenies, the Comparative Method, and the Conflation of Tempo and Mode Antigoni Kaliontzopoulou CIBIO/InBio, Vairão, Portugal
Dean C. Adams Iowa State University, [email protected]
Follow this and additional works at: http://lib.dr.iastate.edu/eeob_ag_pubs Part of the Evolution Commons, and the Statistical Models Commons The ompc lete bibliographic information for this item can be found at http://lib.dr.iastate.edu/ eeob_ag_pubs/208. For information on how to cite this item, please visit http://lib.dr.iastate.edu/ howtocite.html.
This Article is brought to you for free and open access by the Ecology, Evolution and Organismal Biology at Iowa State University Digital Repository. It has been accepted for inclusion in Ecology, Evolution and Organismal Biology Publications by an authorized administrator of Iowa State University Digital Repository. For more information, please contact [email protected]. Page 1 of 52 Systematic Biology
1 2 3 1 Running head: CONFLATION OF TEMPO AND MODE 4 5 6 7 2 8 9 10 3 Phylogenies, the Comparative Method, and the Conflation of Tempo and Mode 11 12 13 4 14 15 *,1,2 2,3 16 5 ANTIGONI KALIONTZOPOULOU AND DEAN C. ADAMS 17 18 19 20 6 21 22 1 23 7 CIBIO/InBio, Centro de Investigação em Biodiversidade e Recursos Genéticos, Campus 24 25 8 Agrario de Vairão, 4485 661 Vairão, Portugal 26 27 28 9 2 Department of Ecology, Evolution, and Organismal Biology, Iowa State University, Ames, Iowa 29 30 31 10 50011, USA 32 33 3 34 11 Department of Statistics, Iowa State University, Ames, Iowa 50011, USA 35 36 37 12 *Correspondence to be sent to: CIBIO/InBio, Centro de Investigação em Biodiversidade e 38 39 40 13 Recursos Genéticos, Campus Agrario de Vairão, 4485 661 Vairão, Portugal; Email: 41 42 14 [email protected] 43 44 45 15 46 47 48 49 50 51 This is a pre-copyedited, author-produced version of an article accepted for publication in 52 Systematic Biology following peer review. The version of record Antigoni Kaliontzopoulou, 53 Dean C. Adams; Phylogenies, the Comparative Method, and the Conflation of Tempo and 54 Mode. Syst Biol 2016; 65 (1): 1-15 is available online at doi: 10.1093/sysbio/syv079 55 56 57 58 59 60 1 http://mc.manuscriptcentral.com/systbiol Systematic Biology Page 2 of 52
1 2 3 16 Abstract 4 5 6 7 17 The comparison of mathematical models that represent alternative hypotheses about the tempo 8 9 18 and mode of evolutionary change is a common approach for assessing the evolutionary processes 10 11 19 underlying phenotypic diversification. However, because model parameters are estimated 12 13 14 20 simultaneously, they are inextricably linked, such that changes in tempo, the pace of evolution, 15 16 21 and mode, the manner in which evolution occurs, may be difficult to assess separately. This may 17 18 22 potentially complicate biological interpretation, but the extent to which this occurs has not yet 19 20 21 23 been determined. In this study, we examined 160 phylogeny × trait empirical datasets, and 22 23 24 conducted extensive numerical phylogenetic simulations, to investigate the efficacy of 24 25 26 25 phylogenetic comparative methods to distinguish between models that represent different 27 28 26 evolutionary processes in a phylogenetic context. We observed that, in some circumstances, a 29 30 27 high uncertainty exists when attempting to distinguish between alternative evolutionary scenarios 31 32 33 28 underlying phenotypic variation. When examining datasets simulated under known conditions, 34 35 29 we found that evolutionary inference is straightforward when phenotypic patterns are generated 36 37 30 by simple evolutionary processes that are represented by modifying a single model parameter at 38 39 40 31 a time. However, inferring the exact nature of the evolutionary process that has yielded 41 42 32 phenotypic variation when facing complex, potentially more realistic, mechanisms is more 43 44 problematic. A detailed investigation of the influence of different model parameters showed that 45 33 46 47 34 changes in evolutionary rates, marked changes in phylogenetic means, or the existence of a 48 49 35 strong selective pull on the data, are all readily recovered by phenotypic model comparison. 50 51 52 36 However, under evolutionary processes with a milder restraining pull acting on trait values, 53 54 37 alternative models representing very different evolutionary processes may exhibit similar 55 56 38 goodness of fit to the data, potentially leading to the conflation of interpretations that emphasize 57 58 59 60 2 http://mc.manuscriptcentral.com/systbiol Page 3 of 52 Systematic Biology
1 2 3 39 tempo and mode during empirical evolutionary inference. This is a mathematical and conceptual 4 5 6 40 property of the considered models that, while not prohibitive for studying phenotypic evolution, 7 8 41 should be taken into account and addressed when appropriate. 9 10 11 42 12 13 14 15 43 Keywords: phylogeny, comparative method, tempo, mode, phenotypic evolution, model fit 16 17 18 44 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 3 http://mc.manuscriptcentral.com/systbiol Systematic Biology Page 4 of 52
1 2 3 45 The phylogenetic comparative method, where species trait values are examined in light of 4 5 6 46 the phylogeny of the group to infer the evolutionary processes that have shaped phenotypic 7 8 47 diversity, is a major framework in evolutionary biology (Harvey and Pagel 1991). In recent 9 10 11 48 years, remarkable advances have been made by the development of new tools for investigating 12 13 49 macroevolutionary phenotypic patterns and testing hypothesis about the biological mechanisms 14 15 50 that shape them. Rooted in the approaches of phylogenetic independent contrasts (Felsenstein 16 17 18 51 1985, 1988) and phylogenetic generalized least squares (PGLS: Grafen 1989; Rohlf 2001), 19 20 52 numerous methods have been developed to investigate how phenotypes diversify over 21 22 53 evolutionary time. Testing for diversifying selection and adaptation (Butler and King 2004) or 23 24 25 54 for adaptive radiation (Harvey and Rambaut 2000; Glor 2010; Harmon et al. 2010); 26 27 55 understanding whether morphological disparity is coupled to cladogenesis (Harmon et al. 2003; 28 29 56 Ricklefs 2004; Rabosky and Adams 2012) or species diversification (Bokma 2002; Adams et al. 30 31 32 57 2009; Rabosky and Adams 2012); identifying phenotypic convergence and parallelism (Harmon 33 34 58 et al. 2005; Stayton 2006; Revell et al. 2007; Adams 2010); and examining the correlation 35 36 37 59 among traits through evolutionary history (Martins and Garland 1991; Pagel 1998; Revell and 38 39 60 Collar 2009) are only some examples of how the study of phenotypic traits on phylogenies have 40 41 61 aided biologists in understanding the processes driving diversification. 42 43 44 Common to all these approaches is the use of mathematical models that aim at 45 62 46 47 63 approximating the tempo and mode of evolutionary change (Simpson 1944; Fitch and Ayala 48 49 64 1994). These models are rooted in similar methods first developed in paleontology to explore 50 51 52 65 how phenotypes evolve. Researchers in this field have long been concerned with evolutionary 53 54 66 tempo and mode, which they study by using data from the fossil record to infer these 55 56 67 evolutionary parameters (Gingerich 1976; Gould and Eldredge 1977; Gould 1980; Fitch and 57 58 59 60 4 http://mc.manuscriptcentral.com/systbiol Page 5 of 52 Systematic Biology
1 2 3 68 Ayala 1994). Paleontological studies were profoundly influenced by the hallmark contribution of 4 5 6 69 George Gaylord Simpson (1944) in which he used the word “tempo” to define the pace at which 7 8 70 phenotypic evolution proceeds. Likewise, he defined “mode” as “…the study of the way, 9 10 11 71 manner, or pattern of evolution, a study in which tempo is a basic factor…” (Simpson, 1944). In 12 13 72 his definitions, Simpson inextricably linked tempo and mode together: the self contained 14 15 73 description of how fast evolutionary changes occurs (tempo) was a basic component for 16 17 18 74 describing the way in which these changes are attained (mode). Indeed, a recent investigation of 19 20 75 the paleontological methods used to estimate and compare evolutionary rates shows that different 21 22 76 rate metrics perform better depending on the mode of evolution (Hunt, 2012). Thus, in 23 24 25 77 paleontological studies, it is clear that tempo and mode are intimately related and can often not 26 27 78 be accurately characterized independently (Hunt, 2012). This observation raises an important 28 29 79 question: is this also the case when using phylogenetic comparative approaches to assess 30 31 32 80 phenotypic evolution of extant taxa? 33 34 35 81 In modern phylogenetic comparative methods, the tempo and mode of evolution are 36 37 82 approached through mathematical models that describe extant phenotypic variation given a 38 39 40 83 phylogenetic hypothesis for the group of interest. The first breakthrough towards modeling how 41 42 84 continuous phenotypic traits evolve on phylogenies was the introduction of a random walk 43 44 model (Brownian Motion, BM; Edwards and Cavalli Sforza 1964; Felsenstein 1973, 1985, 1988; 45 85 46 47 86 Harvey and Pagel 1991). Under BM, phenotypic variation accumulates linearly over time and the 48 49 87 amount of change in the value of a phenotypic trait (X) over a small time interval (t) can be 50 51 52 88 modeled as: 53 54 89 (1) 55 = 56 57 58 59 60 5 http://mc.manuscriptcentral.com/systbiol Systematic Biology Page 6 of 52
1 2 3 90 In equation (1), dB(t) represents independent, normally distributed, random perturbations 4 5 6 91 and σ is the evolutionary rate or variance. The maximum likelihood estimator of the evolutionary 7 8 92 rate is given by: 9 10 11 12 93 (2) 13 = 14 15 94 where C is the phylogenetic variance covariance matrix, N is the number of taxa, X is the vector 16 17 95 of phenotypic trait values at the tips of the phylogeny and E(X) is the expected value of X, or the 18 19 20 96 phylogenetic mean, corresponding to the value at the root node of the phylogeny under BM 21 22 97 (O’Meara et al. 2006). The evolutionary rate σ is a central parameter of the BM model, as it 23 24 25 98 captures how fast evolution proceeds. As such, it represents Simpson’s idea of evolutionary 26 27 99 tempo. 28 29 30 100 Despite its enormous utility and wide application in evolutionary research, the BM model 31 32 33 101 is sometimes too simple to represent complex evolutionary reality (Butler and King 2004; 34 35 102 Beaulieu et al. 2012). Extensions to this model have thus been developed to allow assessing not 36 37 103 only how fast, but also how evolution has generated the phenotypic patterns observed in nature. 38 39 40 104 One family of these extended models aims at providing a solution for modeling the tempo of 41 42 105 phenotypic evolution more accurately. For instance, the pace of phenotypic evolution may vary 43 44 106 across single branches of the phylogeny (McPeek 1995; O’Meara et al. 2006; Revell 2008), 45 46 47 107 between groups of taxa on a phylogeny (Garland 1992; O’Meara et al. 2006; Thomas et al. 2006, 48 49 108 2009; Adams 2014), across evolutionary time (Pagel 1999; Blomberg et al. 2003; Harmon et al. 50 51 109 2010), or among traits (Adams 2013). Such evolutionary hypotheses are tested by fitting models 52 53 54 110 of evolution that encompass more than one evolutionary rate parameter across the phylogeny, 55 56 111 and then comparing their fit to a single rate BM. 57 58 59 60 6 http://mc.manuscriptcentral.com/systbiol Page 7 of 52 Systematic Biology
1 2 3 112 Another family of models includes an additional term, yielding an Ornstein Uhlenbeck 4 5 6 113 (OU) process, which describes an evolutionary ‘pull’ of trait mean value towards one or more 7 8 114 optima through time: 9 10 11 115 (3) 12 = + [ − ] 13 14 15 116 The first term of equation (3) corresponds to the random walk component, while the 16 17 117 second term represents the strength of selection (α) towards a phenotypic optimum (β) (Butler 18 19 118 and King 2004; Beaulieu et al. 2012). Notice that here we follow the notation of Beaulieu et al. 20 21 22 119 (2012) and represent phenotypic optima as β, to avoid confusion with the notation θ, sometimes 23 24 120 used for the relative rate parameter (i.e. Thomas et al. 2006; 2009). From the above mathematical 25 26 121 formulation, the first term of equation (3) is dominated by the evolutionary rate σ. The second 27 28 29 122 term represents a change in mean trait value, occurring towards an optimal state β under a pace 30 31 123 proportional to α (Butler and King 2004). By varying the terms α and β of equation (3), one can 32 33 34 124 represent evolutionary changes that vary in strength and direction, correspondingly (Butler and 35 36 125 King 2004; Beaulieu et al. 2012). For α=0, equation (3) collapses back to a BM process. 37 38 126 Variation in the relative influence of σ and α would then yield models that represent evolutionary 39 40 41 127 processes that lie closer or further away from the simple BM model. In contrast to the first family 42 43 128 of models, though, which focus on modifications of the speed by which evolution proceeds, 44 45 129 these models represent a shift from a random walk (BM) to an evolutionary process that also 46 47 48 130 encompasses changes in trait mean value. 49 50 51 131 Recently, more complex models have been developed in an attempt to characterize the 52 53 132 biological mechanisms underlying phenotypic evolution more accurately. For instance, this can 54 55 56 133 be done either by allowing all σ, α and β in equation (3) to vary (Beaulieu et al. 2012); or by 57 58 59 60 7 http://mc.manuscriptcentral.com/systbiol Systematic Biology Page 8 of 52
1 2 3 134 incorporating different phylogenetic means for different parts of the tree in the calculation of σ 4 5 6 135 (O’Meara et al. 2006; Thomas et al. 2006, 2009). In each case, model parameters are 7 8 136 simultaneously estimated, typically in concert with maximizing the corresponding likelihood 9 10 11 137 equation (but see also e.g. Revell et al. 2011; Eastman et al. 2011; Revell and Reynolds 2012 for 12 13 138 Bayesian implementations). Some of these parameters contribute to modeling trait variance 14 15 139 across taxa through a mean value (i.e. phylogenetic means E(X), optimal trait values β), while 16 17 18 140 others model residual variance (i.e. evolutionary rates σ, strength of selection α). Alternative 19 20 141 models are then compared by evaluating their fit to the data given the underlying phylogeny 21 22 142 through likelihood comparison methods (e.g. likelihood ratio tests, information theoretic criteria, 23 24 25 143 or Monte Carlo simulations; Boettiger et al. 2012). Through this procedure, evolutionary 26 27 144 biologists attempt to obtain a reliable model of the historical events that underlie current 28 29 145 phenotypic variation. As models become more complex, though, inference becomes more 30 31 32 146 complicated. This is because each of the mathematical parameters used to characterize 33 34 147 phenotypic evolution in a phylogenetic context is estimated with respect to the other parameters 35 36 37 148 included in the underlying model. Therefore, it is of interest to determine whether changes in 38 39 149 model parameters can be readily assessed when using phylogenies to study phenotypic evolution. 40 41 42 150 In this article we investigate the efficacy of comparative methods to distinguish between 43 44 phylogenetic comparative models that emphasize changes in different evolutionary parameters. 45 151 46 47 152 We restrict our study to those cases where evolutionary changes are found across groups on a 48 49 153 phylogeny. These encompass questions about how ecological, biogeographic, historical or other 50 51 52 154 life