Evolutionary mode routinely varies among morphological traits within fossil species lineages

Melanie J. Hopkinsa,b,1 and Scott Lidgarda

aDepartment of Geology, Field Museum of Natural History, Chicago, IL 60605; and bMuseum für Naturkunde, Leibniz Institute for Research on Evolution and Biodiversity at the Humboldt University Berlin, 10115 Berlin, Germany

Edited by Neil H. Shubin, The University of Chicago, Chicago, IL, and approved October 19, 2012 (received for review June 11, 2012) Recent studies have revitalized interest in methods for detecting seen in the overall species morphology captured in discriminant evolutionary modes in both fossil sequences and phylogenies. analysis of the same set of traits. It is notable that these traits Most of these studies examine single size or shape traits, often were relatively unimportant in distinguishing species, particularly implicitly assuming that single phenotypic traits are adequate ancestors and descendents (21). Despite thousands of papers on representations of species-level change. We test the validity of this punctuated equilibrium and stasis, the question that Cheetham and assumption by tallying the frequency with which traits vary in others posed has remained unanswered: are (or similarly, when are) evolutionary mode within fossil species lineages. After fitting single traits adequate representations of species-level change? models of directional change, unbiased random walk, and stasis to We investigate this query further by posing a more specific a dataset of 635 traits across 153 species lineages, we find that empirical question: How often do single traits show conflicting within the majority of lineages, evolutionary mode varies across patterns in the same sequence? We define a sequence as a tem- traits and the likelihood of conflicting within-lineage patterns poral series of fossil samples belonging to a species lineage and increases with the number of traits analyzed. In addition, single apply Hunt’s (16) Akaike Information Criterion (AIC) based traits may show variation in evolutionary mode even in situations method for determining whether each sequence is best charac- where the overall morphological evolution of the lineage is terized by directional change (modeled as a generalized random dominated by one type of mode. These quantified, stratigraphi- walk with a non-zero mean step size), unbiased random walk cally based findings validate the idea that morphological patterns (modeled as a generalized random walk with a zero mean step of mosaic evolution are pervasive across groups of organisms size), or stasis (modeled following ref. 23). In addition, using ’ throughout Earth s history. parameters suitably scaled from an actual lineage (24), we sim- ulate a sequence in which a clear overall directional trend is integration | modularity | punctuated equilibrium | rate of evolution | evident, compile an exhaustive set of length:length shape traits, trends and fit the three modes of evolution to each trait. Finally, we compare results based on single traits with results based on uch of the research on stasis and punctuated equilibrium has multivariate traits describing the same species lineages. Mfocused on processes that could generate or influence pat- terns of morphological evolution, including stabilizing selection (1, Results and Discussion 2), metapopulation dynamics (3), environmental stability, habitat Within the full dataset of fossil sequences (n = 635, including tracking, and stress (3–6). Recently, however, renewed discussions traits derived from multivariate analyses), the relative frequency have highlighted the patterns themselves. These investigations of sequences best characterized by directional change, unbiased largely fall into one of two categories, each focusing on a different random walk, and stasis agree with Hunt (17): just over half show aspect of the theory of punctuated equilibrium (7, 8). The first an unbiased random walk, slightly fewer show stasis and very few considers whether morphological evolution is concentrated at spe- show directional change (Table 1). This result also holds for the ciation events or occurs gradually along branches of a phylogenetic subset of single (univariate) traits. However, among the subset tree. Here, methods applied to trees of extant taxa test whether the containing only strongly supported results (AICc weight is 2.7 variance in phenotypes increases as a function of the number of times higher than the next best supported model; Materials and speciation events (inferring a punctuational mode) or of total Methods), slightly more sequences show stasis versus an unbiased branch length, i.e., time (inferring a gradualist mode) (9–13). The random walk (Table 1). In general, static trends are more likely second category distinguishes morphological evolutionary patterns to be strongly supported than random walks (of weakly sup- within sequences of populations in the fossil record, particularly to ported sequences, 65% show an unbiased random walk, 22% determine the relative frequency of stasis compared with other show stasis, and 13% show directional change; G = 72.488, P < modes of change. Recent methods either expand on earlier work in 0.0001). This is unsurprising given that sequences either show treating unbiased random walks as null hypotheses [e.g., Hurst very high or very low support for stasis (Fig. 1) and that the mean measure (14), see ref. 15 for earlier work] or treat an unbiased AICc weight for sequences showing an unbiased random walk is random walk as a model of evolutionary change to be judged 0.706, as opposed to 0.868 for sequences showing stasis (Dataset alongside other models using model selection criteria (16–18). S1). In contrast to Hunt (17), we find less support that shape What studies in both categories have in common is that the traits are more likely than size traits to experience stasis (full quantitative assessment of morphological change is mostly based on dataset: G = 6.446, P = 0.040; single traits only: G = 3.549, P = single traits, either size or shape. In our dataset of 635 sequences 0.170; and strongly supported sequences: G = 4.234, P = 0.120). compiled from literature on the fossil record, only 17% were de- rived using multivariate analysis of several traits (Dataset S1). The remaining sequences are comprised primarily of trait lengths and Author contributions: M.J.H. and S.L. designed research; M.J.H. performed research; M.J.H. trait length:length ratios (Dataset S1). In many cases, more than and S.L. analyzed data; and M.J.H. and S.L. wrote the paper. one trait was measured from a sequence, but each was treated The authors declare no conflict of interest. separately. Potential problems with this approach have been noted This article is a PNAS Direct Submission. before (19–22). In 1987, Cheetham pointed out that among 46 1To whom correspondence should be addressed. E-mail: [email protected]. Metrar- single traits measured across sequences of nine species of This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. abdotos (bryozoan), a few traits departed from the static pattern 1073/pnas.1209901109/-/DCSupplemental.

20520–20525 | PNAS | December 11, 2012 | vol. 109 | no. 50 www.pnas.org/cgi/doi/10.1073/pnas.1209901109 Downloaded by guest on September 28, 2021 Table 1. Relative frequency of directional change (GRW), A unbiased random walk (URW), and stasis 300 Dataset GRW URW Stasis N

Full dataset 36 (5.7) 333 (52.4) 266 (41.9) 635 Single traits 34 (6.4) 283 (53.5) 212 (40.1) 529 Frequency SS traits 8 (1.9) 192 (45.9) 218 (52.2) 418 Simulation 12 (13.3) 35 (38.9) 43 (47.8) 90 0 100 200 Simulation, SS 4 (7.5) 13 (24.5) 36 (67.9) 53 0.0 0.2 0.4 0.6 0.8 1.0 AICc weights, GRW Total number followed by percentage of total in parentheses. SS, strongly supported.

Focusing next on evolutionary modes of just single traits within a species lineage (n = 529), two strong patterns emerge. First, the majority of species lineages show conflicting results among traits; in other words, within most lineages, different Frequency evolutionary modes characterize different traits. Second, even 0 50 100 though there are fewer studies where many traits were measured, 0.0 0.2 0.4 0.6 0.8 1.0 it is evident that a conflicting pattern among traits is more likely AICc weights, URW as the total number of traits analyzed increases (Fig. 2A, G test = 20.838, P = 0.035). These findings hold for subsets limited to size traits (Fig. 2C, G test = 17.513, P = 0.041), shape and meristic traits (Fig. 2D, G test = 21.141, P = 0.004), but not strongly supported traits (Fig. 2B, G test = 4.114, P = 0.767) where the signal is overwhelmed by the tendency for strongly supported 100 200 traits to show stasis. Within lineages that show conflict, traits may Frequency still be dominated by a particular mode (Fig. 3). For example, 0 among 10 length measurements taken from a sequence of sam- 0.0 0.2 0.4 0.6 0.8 1.0 Mandarina chichijimana ples of (land snail; ref. 25), 9 show an AICc weights, stasis unbiased random walk, and only 1 shows stasis. However, none of our lineages are comprised of traits showing only directional B Stasis change (Fig. 3). These patterns emerge despite known correla- tions among particular traits in some sequences that should bias us against finding variation in evolutionary modes.

Even in lineages where the trend in overall morphology is 0.8 0.2 dominated by a certain mode, some traits will show other modes of change. Almost half of the variation in the simulation based on an actual trend in trilobite cranidial shape (24) is summarized 0.6 0.4 by the first principal component (PC), and directional change of PC 1 scores is strongly supported (Fig. 4B). However, most of the possible length:length ratios (90 in total) are better characterized by an unbiased random walk or stasis, and this characterization 0.4 0.6 remains true for the subset of strongly supported results (Fig. 4C and Table 1). Thus, even in this case where directional change seems obvious, there are ample traits that do not change con- 0.2 0.8 gruently with the directional mode. Further, randomly chosen samples of length:length ratios are more likely to show variation GRW in mode of evolution among traits, especially as sample size increases; at n > 9, no samples are entirely consistent in mode 0.8 0.6 0.4 0.2 (Fig. 4D). The pattern of conflicting within-lineage patterns URW fi supports similar ndings in empirical fossil sequences (Fig. 2). Fig. 1. Distribution of AICc weights. (A) Frequency of AICc weights for di- Another insightful example is based on data from a study of rectional change (GRW), unbiased random walk (URW), and stasis. (B) Ter- lacustrine bivalves (26). The relative beak height and the relative nary diagram showing AICc weights for each model for each analyzed trait. anterior length of the valve are best characterized by directional Strongly supported traits shown in gray. change. However, PC analysis of this and nine additional traits reveals that none of the first four principal components are best fi characterized by directional change. Rather, beak height and the organism change signi cantly. However, there are also clear valve length load heavily on the two principal components that examples where the mode of evolution varies across traits within EVOLUTION show only moderate support for directional change. This sug- species lineages that appear static overall. For example, morpholog- gests that while some traits in this lineage do show some di- ical change in the Globorotalia truncatulinoides (foraminiferan) from rectional change, they do not account for a large proportion of Deep Sea Drilling Project site 591 is strongly characterized by stasis if variation within the traits chosen to represent the lineage. the three evolutionary modes are fit to the discriminant function Because our ability to recognize directional change in a trait scores extracted from an eigenanalysis of 34 size and shape traits (27). depends on that change being greater than change in other aspects of If the 34 traits are analyzed separately, however, over a third of them size or shape, it may be inevitable that some traits remain compar- are best characterized by an unbiased random walk (Dataset S1). atively static even as others show a directional shift. An increase in Although our results show that single measured traits may EARTH, ATMOSPHERIC,

horn size, for example, is no less notable because no other parts of not fully represent species-level change, determining the general AND PLANETARY SCIENCES

Hopkins and Lidgard PNAS | December 11, 2012 | vol. 109 | no. 50 | 20521 Downloaded by guest on September 28, 2021 A B 20 15 10 Frequency Frequency 5 0 0 5 10 15 20 25

0 5 10 15 0 5 10 15 No. of traits measured per species lineage No. of traits measured per species lineage

C 20 D 15 Frequency Frequency 510 0 0 5 10 15 20

0 5 10 15 0 5 10 15 No. of traits measured per species lineage No. of traits measured per species lineage

Fig. 2. Stacked histograms showing distribution of species lineages where all measured traits show same evolutionary mode (black) and where evolutionary mode varies across measured traits (gray). (A) All single traits, (B) size traits only, (C) shape traits only, and (D) strongly supported traits only.

tendency of the overall variation within the species lineage is hardly modes of species lineages. For paleobiological studies, this is an simple. In addition to the multivariate statistical methods— inevitable consequence of the nature of the fossil record. Typi- ordination, geometric morphometrics—used in the examples cally, only skeletal elements are preserved and both preservation above, one might initially compare results with the distribution and sampling vary in uniformity and resolution. Our simulation of modes among separate individual traits to extract comple- data are a patent example: with rare exception, only trilobite mentary perspectives of species change. However, scientists exoskeletons are preserved in the fossil record and the high fre- will always be extrapolating from a small sample of available mor- quency of disarticulation among skeletal elements limited the study phological (or genetic, or other) traits to characterize evolutionary to just one part of the skeleton. Moreover, species themselves are all determined by subsets of possible traits and may be individuated using a variety of different protocols (28). Among paleontologists, species deter- 100 minations are based on morphological variation. However, concepts differ in how variation is used for delimiting species, depending on the goal of the study. For example, studies based in 80 cladistic parsimony look for changes in otherwise invariant traits, studies using numerical phenetics may claim measurable traits

60 are all equivalent, and biostratigraphic studies may seek out (directionally) variable traits for finer temporal resolution. Here, we have accepted the species delimitations in the original studies, which in total may comprise different species concepts. Nor were the traits analyzed necessarily selected randomly in the original studies. Some may have been chosen because they permitted

% traits showing mode showing % traits consistent measurement or had prior taxonomic importance, but also because they seemed to show change or lack of change.

02040 Because observed evolutionary patterns may include con- tributions from immigration and emigration on ecological time- GRW URW Stasis scales, and climatic, habitat, or facies shifts on much longer Fig. 3. Boxplots showing proportion of traits within species lineages that geological timescales, it is also possible that different modes show directional change (GRW), unbiased random walk (URW), and stasis. characterize the same trait, contingent upon the particular interval Tallied for sequences where at least four traits were measured. of time or particular locality being considered. Changes in

20522 | www.pnas.org/cgi/doi/10.1073/pnas.1209901109 Hopkins and Lidgard Downloaded by guest on September 28, 2021 A B PC 1 (45.0%)

15 GRW: 0.984 URW: 0.016 Stasis: 0.000 PC 2 (10.8%) GRW: 0.094 10 URW: 0.492 Stasis: 0.413 PC 3 (8.1%) GRW: 0.009 URW: 0.047 Stasis: 0.944

PC 4 (7.8%) GRW: 0.010

Standardized trait mean URW: 0.053 Stasis: 0.938 05 1 mm PC 24 (5e-4%) GRW: 0.012 URW: 0.067 Oldest Youngest Stasis: 0.921 C D 10 10 50 50

All directional (GRW) All random walk (URW) −10 −5 −10 −5 10 10 Frequency 50 50 Standardized trait mean

All stasis Strongly supported 0 200 400 600 800 1000 −10 −5 −10 −5 2345678910 Oldest Youngest Oldest Youngest No. of traits per sample

Fig. 4. Results from simulated trilobite cranidial shape data. (A) Landmarks representing overall shape of cranidium (required only from half because of bilateral symmetry) and all possible length measurements between landmarks. (B) Time series of selected PC axes scores. (Upper Left) Image shows example of typical morphology from oldest sample; (Upper Right) image shows typical morphology from stratigraphically youngest sample (scale bars = 1 mm). Change in morphology is dominated by an expansion and rotation of palpebral lobes relative to rest of the cranidium. (C) Time series of individual length:length measurements, divided up into four panels showing those that were best characterized by directional change (Upper Left) (black), unbiased random walk (Upper Right) (gray), stasis (Lower Left) (white), and strongly supported trends of all types (Lower Right). Bars on right show range of means for each mode of evolution at the end of the sequence. (D) Stacked histograms showing distribution of samples of randomly selected length:length measurements where all traits in the sample show the same evolutionary mode (black) and where evolutionary mode varies across traits (gray). Number of samples taken at each sample size = 1,000.

geographically patchy or clinal morphological variation likely rate models of character change among nearly all cases in 115 contribute to all fine-scale temporal trends in the fossil record, al- fossil invertebrate clades (32). Second, our simulation results though their general significance is disputed; in the few studies involved random resampling, which by virtue of subsetting,

examining trends in replicate sections, both congruent and in- accommodates both the possibilities of temporal and geographic EVOLUTION congruent patterns are reported (29–31). Simulation or mod- variation. Finally, the size and scope of our study lends assurance eling approaches merit further investigation, but necessarily to at least the broad patterns it reveals. The dataset includes involve numerous interacting parameters that resist precise representatives from across eukaryotes and throughout the estimation, and are expected to differ among taxa, localities, fossil record, compares many more quantified sequences than and time intervals. any other we are aware of, and does so using a consistent None of these contributions negate our overall findings, which analytical protocol. Our results are so striking that they compel are supported by several lines of evidence. First, our findings us to think differently about the whole question of single traits are reinforced by an independent study of character rate dis- as proxies for species-level change: the apparent ubiquity of EARTH, ATMOSPHERIC,

tributions using phylogenetic compatibility, which rejected single- mosaic patterns of morphological evolution. AND PLANETARY SCIENCES

Hopkins and Lidgard PNAS | December 11, 2012 | vol. 109 | no. 50 | 20523 Downloaded by guest on September 28, 2021 Mosaic evolution is one among a historical welter of ideas were largely collected from figures using WinDig (56); additional data were relating to the organism as a combination of ancient and more obtained directly from authors (Acknowledgments). Sequences were in- derived morphological parts, or to these parts evolving at dif- cluded if there were at least six temporal samples with five or more speci- ferent rates within and among lineages (33, 34). The central idea, mens; of these, sequences were not included if the analysis was at the genus if not the term itself (35), extends back to the nineteenth century, level or higher (e.g., refs. 57, 58) or if one or more of the above four parameters went unreported (e.g., refs. 59, 60), including situations where to Louis Dollo’s principle of “crossing of specializations” (36, 37) fl only the order of samples was reported (e.g., ref. 61). We refer to sequences and to con icting rates of trait change in precise stratigraphic as species lineages because they may comprise series of samples belonging studies of fossil populations (38, 39). In 1944, G. G. Simpson to two or more chronospecies as well as single species. The dataset is com- “inextricably bound the tempo of evolution with the mode, or prised of both micro- and macrofossils from marine shelf, deep sea, open pattern, of evolution within the matrix of organism-environment ocean (pelagic), terrestrial, and lacustrine environments, including brachio- interactions” (40, 41); indeed the measurement of rates is not pods, bryozoans, echinoderms, mammals, fish, mollusks, conodonts, ostra- separable from assumptions about the mode of evolution (42, cods, trilobites, planktonic and benthic foraminiferans, diatoms, and 43). At the same time, Simpson primarily related varying rates radiolarians. Sequences range from 0.01 to 35.6 million years long with and patterns of change in single characters or character com- a median duration of 2.8 million years. Finally, all of the traits are quanti- plexes—not the aggregate of all characters evident within line- tative and measurable throughout each sequence analyzed and thus do not document the appearance of new traits within sequences. ages—to environmental circumstances and adaptation (41). The generalized random walk, unbiased random walk, and stasis models Thus, Simpson may also have inadvertently codified an analytical were fit by maximum likelihood to each sequence using the paleoTS package frame for morphological rates that persisted for decades, despite for R, in accord with the protocols and model assumptions previously de- widespread acceptance of mosaic patterns by paleontologists, scribed by Hunt (16, 18). Model support was assessed using the bias-corrected including himself. Akaike Information Criterion; the model with the lowest AICc is the best Given that mosaic evolution has long been acknowledged by supported. Akaike weights, which are a transformation of the AICc values evolutionary biologists, our results may not appear surprising. that sum to one, are then interpreted as probabilities that each of the The question “What alternative exists?” is for us less relevant models considered respectively is the best supported model (62). To compare than “How does the explicit recognition of mosaic evolution help our results with those of Hunt (17), we used G tests to assess whether counts us to develop a more nuanced understanding?” (35, 44, 45). A of best supported modes were independent of categorical variables such as critical problem in this vein is deducing independence of traits trait type. However, instead of using Wilcoxen rank-sum tests to test for while inferring the overall mode of evolution. Patterns of trait differences in median AIC weights for particular models across groups, we chose to evaluate separately the subset of sequences for which the best- independence versus integration affect all types of multivariate supported evolutionary mode was also strongly supported. Results were approaches. Thus, the use of multiple traits in studies of evolu- retained if the best supported mode had an AICc weight 2.7 times greater tionary mode in fossil lineages can be an opportunity to in- than the second best-supported evolutionary mode. This criterion corre- corporate complementary insights from studies of morphological sponds to the “rule of 2,” where a model is considered well supported rel- integration and modularity in living organisms (46, 47). For ex- ative to other models under evaluation if the difference in likelihood values ample, quantitative assessments of modularity may identify traits is greater than 2 (62). However, this is an arbitrary cutoff and should not be that can more appropriately be treated as independent in phy- interpreted as a level of statistical significance (62). We chose not to use the logenetic and morphological rate studies. stricter criterion of an AICc weight of 0.89 (the likelihood criterion of Our analysis only considers mosaic evolution as pattern (48). rejecting a hypothesis when an outcome is eight times [or more] less prob- However, because different forms of integration (e.g., functional able [63]) because this reduced the dataset by over 70%. It is notable, integration of parts [49, 50], and developmental modularity [51]) however, that applying this criterion reduces the dataset to 179 sequences, 91% of which are best characterized by stasis (see also Fig. 1B). can influence the variation, direction, rate, and magnitude of ’ We also applied the same analysis to subsets of data that included only the lineages morphological changes at different temporal scales single traits. Finally, within the subset of sequences where single traits were (e.g., ref. 52), they are potential keys to process explanations measured, we tallied the relative frequencies of each evolutionary mode of evolutionary modes. Variation in evolutionary mode among within species lineages from unique localities. traits within species may occur because different traits vary in Using geometric morphometrics, Hopkins and Webster (24) documented their response to the same selection pressure, are subjected to a directional shift in the shape of the cranidium from a series of five suc- different selection pressures at different levels of organization, cessive samples of the trilobite Zacanthopsis levis-palmeri. To simulate or are more constrained or more evolvable due to degrees of a larger number of samples that mathematically represented this trend, we trait integration and functional and developmental modularity. randomly sampled 25 specimens from a multivariate normal distribution A promising way forward will take the pervasiveness of mosaic defined by 1 of 10 evenly spaced landmark configurations and the co- evolution into account, not just whether single traits adequately variance matrix of one of the original samples (UCR 10097, Klondike Gap, = represent species’ evolutionary tempos and modes. A more nu- Nevada, n 32) using the mtvnorm package for R (64). This yielded 10 anced understanding invokes convergence of evidential infer- samples of 25 specimens where the mean of each was represented by 1 of 10 successive mean landmark configurations representing the overall di- ences (53): paleontological approaches that compare canonical rectional trend. We extracted all of the between-landmark lengths and concepts of evolutionary mode (43) and their further devel- found the ratio between each length and the length of the cranidium opment to accommodate more holistic, multivariate patterns of (defined by two landmarks), for a total of 90 length:length ratios (derived trait covariance (“overall morphology”); measuring congruence from 14 landmarks) (Fig. 4A). We performed PC analysis of the warp scores of these results with individual trait patterns within lineages; derived from a thin-plate spline decomposition of the landmark coor- and developing ways of studying mosaic evolution and morpho- dinates (65, 66). We then fit the three evolutionary modes to each of the 90 logical integration that are shared by paleontology and neontology length:length ratios as well as the sample mean scores from the PC analysis. (45, 54, 55). Finally, we randomly sampled the 90 length:length ratios (1,000 iterations without replacement, starting at n = 2) and fit the three evolutionary Materials and Methods models to each. We began with the results from a previous study investigating the relative frequency of directional change, unbiased random walk, and stasis in the ACKNOWLEDGMENTS. We thank G. Hunt for making his 2007 AICc results available and for his help and advice using the paleoTS package. We also fossil record (17). This dataset consisted of AICc results for 251 sequences thank D. Lazarus, K. Kim, M. Kucera, B. Lauridsen, W. Theriot, E. Erba, and from 32 references. We excluded nine sequences because they did not meet D. Anderson for providing additional data; and A. Haber for advice on sam- our selection criteria. We added an additional 393 sequences from 61 ref- pling from multivariate normal distributions. This project was supported by erences (Datasets S1 and S2). Model selection (16) requires trait means, a John Caldwell Meeker Postdoctoral Fellowship (Department of Geology, variances, sample sizes, and relative stratigraphic position or age. New data Field Museum of Natural History, Chicago) and by the VolkswagenStiftung.

20524 | www.pnas.org/cgi/doi/10.1073/pnas.1209901109 Hopkins and Lidgard Downloaded by guest on September 28, 2021 1. Hansen TF, Houle D (2004) Evolvability, stabilizing selection, and the problem of 32. Wagner PJ (2012) Modelling rate distributions using character compatibility: im- stasis. Phenotypic Integration: Studying the Ecology and Evolution of Complex Phe- plications for morphological evolution among fossil invertebrates. Biol Lett 8(1): notypes, eds Pigliucci M, Preston K (Oxford Univ Press, Oxford, UK), pp 130–150. 143–146. 2. Estes S, Arnold SJ (2007) Resolving the paradox of stasis: Models with stabilizing se- 33. Takhtazhian A (1959) Essays on the Evolutionary Morphology of Plants (American lection explain evolutionary divergence on all timescales. Am Nat 169(2):227–244. Institute of Biological Sciences, Washington, DC). 3. Eldredge N, et al. (2005) The dynamics of evolutionary stasis. Paleobiology 31: 34. Nelson G (2004) Cladistics: Its arrested development. Milestones in Systematics, eds 133–145. Williams DM, Forey PL (CRC Press, Boca Raton, FL). 4. Parsons PA (2005) Environments and evolution: Interactions between stress, resource 35. Stebbins GL (1983) Mosaic evolution: An integrating principle for the modern syn- inadequacy and energetic efficiency. Biol Rev Camb Philos Soc 80(4):589–610. thesis. Experientia 39(8):823–834. 5. Sheldon PR (1996) Plus ça change—A model for stasis and evolution in different en- 36. Gould SJ (1970) Dollo on Dollo’s law: Irreversibility and the status of evolutionary vironments. Palaeogeogr Palaeoclimatol Palaeoecol 127:209–227. laws. J Hist Biol 3:189–212. 6. Badyaev AV (2005) Role of stress in evolution: From individual adaptability to evo- 37. Rieppel O (2012) Othenio Abel (1875-1946) and the ’phylogeny of the parts’. Cladis- lutionary adaptation. Variation: A Central Concept in Biology, eds Hallgrimsson B, tics, 10.1111/j.1096-0031.2012.00428.x. Hall BK (Elsevier Academic Press, New York), pp 277–302. 38. Hyatt A (1880) The Genesis of the Tertiary species of Planorbis at Steinheim (Anni- 7. Eldredge N, Gould SJ (1972) Punctuated equilibria: An alternative to phyletic grad- versary Memoir of Boston Society of Natural History, Boston, MA). ualism. Models in Paleobiology, ed Schopf TJM (Freeman, Cooper, & Co., San Fran- 39. Brice W (2004) Henry Shaler Williams (1847-1918) and punctuated equilibria. Earth Sci cisco), pp 82–115. Hist 23:32–40. 8. Gould SJ, Eldredge N (1977) Punctuated equilibria: The tempo and mode of evolution 40. Simpson GG (1944) Tempo and Mode in Evolution (Columbia Univ Press, New York). reconsidered. Paleobiology 3:115–151. 41. Laporte LF (2000) George Gaylord Simpson: Paleontologist and Evolutionist (Colum-  9. Laurin M, Gussekloo SWS, Marjanovic D, Legendre L, Cubo J (2012) Testing gradual bia Univ Press, New York). and speciational models of evolution in extant taxa: The example of ratites. J Evol Biol 42. Roopnarine PD (2003) Analysis of rates of morphologic evolution. Annu Rev Ecol Evol – 25(2):293 303. Syst 34:605–632. 10. Bokma F (2002) Detection of punctuated equilibrium from molecular phylogenies. J 43. Hunt G (2012) Measuring rates of phenotypic evolution and the inseparability of – Evol Biol 15:1048 1056. tempo and mode. Paleobiology 38:351–373. “ ” 11. Bokma F (2008) Detection of punctuated equilibrium by Bayesian estimation of 44. West-Eberhard MJ (2003) Developmental Plasticity and Evolution (Oxford Univ Press, speciation and extinction rates, ancestral character states, and rates of anagenetic Oxford, UK). – and cladogenetic evolution on a molecular phylogeny. Evolution 62(11):2718 2726. 45. Klingenberg CP (2010) Evolution and development of shape: Integrating quantitative fi 12. Rabosky DL (2012) Positive correlation between diversi cation rates and phenotypic approaches. Nat Rev Genet 11(9):623–635. evolvability can mimic punctuated equilibrium on molecular phylogenies. Evolution 46. Rollian C, Willmore K (2009) Morphological integration at 50: Patterns and processes – 66(8):2622 2627. of integration in biological anthropology. Evol Biol 36:1–4. 13. Pagel M (1999) Inferring the historical patterns of biological evolution. Nature 401 47. Murren CJ (2012) The integrated phenotype. Integr Comp Biol 52(1):64–76. – (6756):877 884. 48. Takhtazhian A (1991) Evolutionary Trends in Flowering Plants (Columbia Univ Press, 14. Roopnarine PD (2001) The description and classification of evolutionary mode: A New York). computational approach. Paleobiology 27:446–465. 49. Barel CDN (1983) Form-relations in the context of constructional morphology: The eye 15. Bookstein FL (1988) Random walk and the biometrics of morphological characters. and suspensorium of lacustrine cichlidae (Pisces, Teleostei). With a discussion on the Evol Biol 23:369–398. implications of phylogenetic and allometric form interpretation. Neth J Zool 34: 16. Hunt G (2006) Fitting and comparing models of phyletic evolution: Random walks and 439–502. beyond. Paleobiology 32:578–602. 50. Schwenk K, Wagner G (2001) Function and the evolution of phenotypic stability: 17. Hunt G (2007) The relative importance of directional change, random walks, and Connecting pattern to process. Am Zool 41:552–563. stasis in the evolution of fossil lineages. Proc Natl Acad Sci USA 104(47):18404–18408. 51. Klingenberg CP (2008) Morphological integration and developmental modularity. 18. Hunt G (2008) Gradual or pulsed evolution: When should punctuational explanations Annu Rev Ecol Evol Syst 39:115–132. be preferred? Paleobiology 34:360–377. 52. Monteiro LR, Nogueira MR (2010) Adaptive radiations, ecological specialization, and 19. Raup DM, Crick RE (1981) Evolution of single characters in the Jurassic ammonite the evolutionary integration of complex morphological structures. Evolution 64(3): Kosmoceras. Paleobiology 7:200–215. 724–744. 20. Levinton JS (1983) Stasis in progress: The empirical basis of macroevolution. Annu Rev 53. Forber P, Griffith E (2011) Historical reconstruction: Gaining epistemic access to the Ecol Syst 14:103–137. deep past. Philos Theory Biol 3:1–19. 21. Cheetham AH (1987) Tempo of evolution in a Neogene bryozoan: Are trends in single 54. Gerber S, Hopkins MJ (2011) Mosaic heterochrony and evolutionary modularity: The morphological characters misleading? Paleobiology 13:286–296. trilobite genus Zacanthopsis as a case study. Evolution 65(11):3241–3252. 22. Anderson DK (1993) A method for recognizing morphological stasis. Morphological 55. Webster M, Zelditch ML (2011) Modularity of a Cambrian ptychoparioid trilobite Change in Quaternary Mammals of North America, eds Martin RA, Barnosky AD – (Cambridge Univ Press, Cambridge, UK), pp 13–23. cranidium. Evol Dev 13(1):96 109. fi 23. Sheets HD, Mitchell CE (2001) Why the null matters: Statistical tests, random walks 56. Lovy D (1996) WinDig, v. 2.5. Available at http://www.unige.ch/sciences/chi /cpb/ and evolution: Genetica 112-113:105–125. windig.html. 24. Hopkins MJ, Webster M (2008) Morphological and ontogenetic change in the “Early” 57. Sheldon PR (1987) Parallel gradualistic evolution of Ordovician trilobites. Nature 330 – Cambrian trilobite Zacanthopsis during an interval of environmental change. Advances (6148):561 563. in Trilobite Research, eds. Rabano, I, Gozolo, R & Garcia-Bellido, DC (Instituto Geologico 58. Grey M, Haggart JW, Smith PL (2008) Variation in evolutionary patterns across the – y Minero de España, Cuadernos del Museo Geominero 9, Madrid), pp. 185–187. geographic range of a fossil bivalve. Science 322(5905):1238 1241. 25. Chiba S (1996) A 40,000-year record of discontinuous evolution of island snails. Pa- 59. Cheetham AH (1986) Tempo of evolution in a Neogene bryozoan: Rates of mor- – leobiology 22:177–188. phologic change within and across species boundaries. Paleobiology 12:190 202. 26. Geary DH, Hunt G, Magyar I, Schreiber H (2010) The paradox of gradualism: Phyletic 60. Sorhannus U, Fenster EJ, Burckle LH, Hoffman A (1988) Cladogenetic and anagenetic evolution in two lineages of lymnocardiid bivalves (Lake Pannon, central Europe). changes in the morphology of Rhizosolenia praebergonii Mukhina. Hist Biol 1: Paleobiology 36:592–614. 185–206. 27. Lazarus D, Hilbrecht H, Spencer-Cervato C, Thierstein H (1995) Sympatric speciation 61. Reyment RA, Hayami I, Carbonnel G (1977) Variation of discrete morphological and phyletic change in Globorotalia truncatulinoides. Paleobiology 21:28–51. characters in Cytheridea (Crustacea: Ostracoda). Bull Geol Inst Univ Uppsala, 7: 23–36. 28. Wheeler QD, Meier R (2000) Species Concepts and Phylogenetic Theory (Columbia 62. Burnham KP, Anderson DR (2002) Model Selection and Multimodel Inference Univ Press, New York). (Springer, New York), 2nd Ed. 29. Bralower TJ, Parrow M (1996) Morphometrics of the Paleocene coccolith genera 63. Wagner PJ, Kosnik MA, Lidgard S (2006) Abundance distributions imply elevated Cruciplacolithus, Chiasmolithus, and Sullivania: A complex evolutionary history. Pa- complexity of post-Paleozoic marine ecosystems. Science 314(5803):1289–1292. leobiology 22:352–385. 64. Genz A, et al. (2011) mvtnorm: Multivariate Normal and t Distributions. v. 0.9-9991. 30. Renaud S, Schmidt DN (2003) Habitat tracking as a response of the planktic fora- Available at http://CRAN.R-project.org/package=mvtnorm. minifer Globorotalia truncatulinoides to environmental fluctuations during the last 65. Rohlf FJ (1990) Rotational fit (Procrustes) methods. Proceedings of the Michigan 140 kyr. Mar Micropaleontol 49:97–122. Morphometrics Workshop, eds. Rohlf FJ, Bookstein FL (University of Michigan Mu- –

31. Kim K, Sheets HD, Mitchell CE (2009) Geographic and stratigraphic change in the seum of Zoology Special Publication), pp. 227 236. EVOLUTION morphology of Triarthrus beckii (Green) (Trilobita): A test of the Plus ça change model 66. Zelditch ML, Swiderski DL, Sheets HD, Fink WL (2004) Geometric Morphometrics for of evolution. Lethaia 42:108–125. Biologists (Elsevier Academic Press, New York). EARTH, ATMOSPHERIC, AND PLANETARY SCIENCES

Hopkins and Lidgard PNAS | December 11, 2012 | vol. 109 | no. 50 | 20525 Downloaded by guest on September 28, 2021