The evolution of ‘bizarre structures’ in dinosaurs: biomechanics, sexual selection, social selection or species recognition?

Authors


  • Editor: Steven Le Comber

Correspondence
Kevin Padian, Department of Integrative Biology and Museum of Paleontology, University of California, Berkeley, CA, USA. Email: kpadian@berkeley.edu

Abstract

‘Bizarre structures’ in dinosaurs have four main traditional explanations: mechanical function, sexual selection, social selection and species recognition. Any of these can be plausible for individual species, but they fail to be persuasive when other lines of evidence cannot adequately test them. The first three also fail as general propositions when phylogenetic analyses based on other characters do not support scenarios of selective improvement of such functions in their clade (or the explanation simply does not apply to any other species in the clade). Moreover, the hypothesis of sexual selection requires significant sexual dimorphism, which has never been conclusively established in dinosaurs.

We propose instead that species recognition may have been a more general force that drove the evolution of bizarre structures in dinosaurs. That is, the bizarre structures communicate to other individuals a variety of possible associational cues, including species identification, potential protection and social habits and the appropriateness of potential mates. In other words, bizarre structures amount to an advertisement for positive association. Neither species recognition nor any other hypothesis should be a ‘default’ explanation. Although direct observation is impossible, we propose two tests. First, contrary to adaptive, social or sexual selection, under the species recognition model morphology should be expected to evolve without obvious directional trends, because the only objective is to differ from one's relatives. Hence, patterns of evolution of bizarre structures should be relatively proliferative and non-directional. Second, several contemporaneous species should overlap in geographic range (sympatric, parapatric, peripatric). Fossil species often show evidence of this pattern in the past by ‘ghost ranges’ of related taxa. These tests together could reinforce or weaken an argument for species recognition.

Introduction

‘Bizarre structures’ in dinosaurs and other extinct animals (e.g. Gould, 1974) are of perennial interest to paleontologists and have become a staple of textbooks on evolution because they raise perennial questions. What did these structures do? How did they evolve? If they were so useful, how did they contribute to their bearers' evolutionary success? If their bearers are extinct, did they become a liability at some point?

In this paper, we explore the principal explanations for the evolution of ‘bizarre structures.’ The kinds of explanations we discuss include the teleology of what they were for and how they evolved. We recast these explanations using current methods of comparative biology. Our goal is less to argue for a particular theory that explains everything than to suggest how these kinds of evolutionary problems should be addressed, and to suggest some criteria for testing them. Our hope is that others will both improve on our suggestions and bring new data to the questions.

By ‘bizarre structures’ we mean features that are unusual enough, to the trained eyes of paleobiologists, to invite explanations beyond the basic functions of feeding, locomotion, respiration and so on (Farlow & Dodson, 1974; Gould, 1974; Molnar, 1977; Main et al., 2005). In many respects these structures are similar (but not necessarily analogous) to certain structures in living animals. They include the frills and horns of ceratopsians, the domes of pachycephalosaurs, the crests of lambeosaurine hadrosaurs, the scute complexes of ankylosaurs and the plates and spikes of stegosaurs. We discuss four general types of explanations: mechanical function, sexual selection, social selection and species recognition. The first two of these are pre-eminent in paleobiological explanation (e.g. Galton, 1970; Farlow & Dodson, 1974; Dodson, 1975; Hopson, 1975; Farlow, Thompson & Rosner, 1976; Molnar, 1977; Buffrenil, Farlow & de Ricqlès, 1986; etc.). The third has been advocated most recently and thoroughly by Hieronymus et al. (2009). The fourth has not been extensively considered by any authors, although it has been frequently acknowledged in functional and behavioral considerations (e.g. Farlow & Dodson, 1974; Hopson, 1975; Molnar, 1977; Sampson, 1999; Hieronymus et al., 2009). There has been an historical predilection to attempt first to explain a bizarre structure in mechanical terms; if this explanation appears weak or is contraindicated, it has been traditional to attribute the feature to ‘sexual display’ by virtue of its apparent uselessness for mechanical function. In this way, sexual display has often become a ‘default’ explanation that was seldom explicitly tested or questioned.

We acknowledge several classes of facts. First, some structures may have served more than one function. For example, ankylosaur armor may have been defensive but also distinctive enough to have served a role in species recognition. After all, exaptation is a pre-eminent factor in macroevolutionary change. Second, because many soft part features and also nearly all behaviors are not preserved in the fossil record, affairs may have been far more complex than paleontologists can detect. We also allow that not enough is known to determine the origin of some features; for example, there are too few known specimens of cranially adorned theropod taxa such as Dilophosaurus, Cryolophosaurus and Carnotaurus to permit a test of evolutionary explanations. We see no reason to be dogmatic about particular hypotheses, and no reason not to be pluralistic about explanations when appropriate. Our goal is to propose a set of explicit tests of mechanical and behavioral hypotheses that we hope will set up discriminatory criteria for these kinds of explanations.

Methods

Although there are many approaches to explaining morphology in extinct organisms (Hickman, 1980), inferences about function and behavior are based on two general models: homology and analogy (essentially, historical and ahistorical explanations: Weishampel, 1997). The accepted approach to evaluating homology of function and behavior in extinct animals is Witmer's (1995) extant phylogenetic bracket (EPB). For this purpose, a phylogeny of living and related fossil forms is required. The degree to which a condition can be inferred reliably as present in an extinct taxon is related to its position among living forms that are known to share the function or behavior (Fig. 1). Because crocodiles and birds, the two extant brackets of extinct dinosaurs, share none of the bizarre structures of extinct dinosaurs, the EPB cannot provide much direct guidance on these problems. There are simply no available homologous structures, with the possible exceptions of the cranial crests of lambeosaurine hadrosaurs and cassowaries, and the scutes of crocodiles and thyreophorans (which, being absent in their respective common ancestors, must be regarded as parallelisms, despite an obvious homological basis in bone histology: Scheyer & Sander, 2004; Main et al., 2005).

Figure 1.

 The Extant Phylogenetic Bracket (Witmer, 1995) of extinct dinosaur clades that characteristically and independently evolved ‘bizarre structures:’ stegosaurs and ankylosaurs (parallel elaboration of homologously shared dermal scutes), lambeosaurine hadrosaurs, neoceratopsians, pachycephalosaurs and ceratosaurian theropods. The bizarre morphologies of these dinosaurian groups are not shared by their living bracket, crocodiles and birds; crocodiles have scutes, but they are never elaborated like those of thyreophoran dinosaurs.

Analogy to living forms is the approach that remains when arguments of homology cannot be made, and it is even more problematic. The quality of an explanation depends in part on the precision of definition of the features that are compared, and the separation of those features (and functions) from ancillary or irrelevant ones (Whewell, 1859; Padian, 1995; Wilson, 1998).

Classes of explanation of bizarre structures

The two general classes of explanation of bizarre structures in dinosaurs relate to function and display (including sexual selection, social selection and species recognition). Each kind of explanation has a long history in the literature, including discussions of dinosaur behavior (Horner & Gorman, 1988; Carpenter, Hirsch & Horner, 1994; Currie & Padian, 1997; Farlow & Brett-Surman, 1997; Horner & Dobb, 1997; Carpenter, 1999; Weishampel, Dodson & Osmolska, 2004; Hieronymus et al., 2009). We summarize these classes of explanation in Table 1.

Table 1.   Some proposed functions of ‘bizarre’ structures
  1. Note that functions of species recognition encompass interactions both between species (discourage association of non-conspecifics) and within species (‘encourage association of conspecifics,’ both of the same sex and of different sexes).

Mechanical
 Defense
 Communication
 Thermoregulation
 Sensory
 Procuring food
Display
 Interspecific
  Deter predators
  Discourage association of non-conspecifics
 Intraspecific
  Intrasexual
   Establish territory
   Ward off rivals for resources (including mates)
   Encourage association of conspecifics
   Compete for resources
  Intersexual
   Attract mates
   Encourage association of conspecifics

It is important that we define our terms. Mechanical function refers to a specific adaptation such as feeding, locomotion, insulation or communication. Sexual selection is the advantage gained to access to mates when one sex possesses a specific feature that the other does not, and uses it to attract mates or repel rivals for mates (Darwin, 1859, 1871). We want to emphasize here the importance of a discrete structure, function or behavior present in one sex but not the other, that is used for these two purposes. We also emphasize that this true sexual dimorphism is different from a simple ‘sexual difference’ in which one sex is slightly larger or more robust than the other, but possesses no particular structures for these purposes. (We recognize that there is debate about this among behavioral ecologists, and we discuss it elsewhere.) Social selection refers to features that individuals in a species use to improve their competitive advantage for resources. Species recognition refers to features that allow others of the same species to recognize each other for various social purposes. Mate recognition is not the same thing, but it is a subset because it is important for individuals to mate with others in the same species. We want to state emphatically that we do not reject the possible operation of any and all of these processes in extinct dinosaurs in principle. We ask how well established any and all of these are in specific cases.

Mechanical explanations

Explanations of individual taxa

Many possible mechanical explanations have been proposed and tested for various bizarre skeletal features of individual dinosaur species (Weishampel, 1981, 1997; Farke, 2004; Farke, Wolff & Tanke, 2009; Hieronymus et al., 2009). In our view, Weishampel's (1981) classic study of the crest of the hadrosaur Parasaurolophus is a model for examining functional inferences in extinct individual taxa. Weishampel first divided all proposed hypotheses into testable and untestable, and then proceeded to see if the testable ones could be falsified or supported by other lines of evidence. He found that most hypotheses of display and behavior could not be explicitly tested, but some mechanical functions, such as snorkeling, head-butting and air storage, could be tested and rejected. Weishampel tested the proposed function of a resonance chamber by building a model of the nasal passages and diverticula, and passing a spectrum of oscillating frequencies through them. Certain frequencies, as expected, resonated better than others, and Weishampel independently tested this outcome by determining whether the auditory organs were well attuned to those frequencies by studying the size and morphology of the stapedial region. Whereas this study did not ‘prove’ any particular function, and could not logically rule out several weakly supported or untestable explanations (see Weishampel, 1997), it is a model study for testing functional hypotheses of individual organisms in paleobiology.

But Weishampel's approach, thorough as it was, did not account for all aspects of the problem, as he recognized. He noted one: the characters related to vocalizing and hearing reflect different (if not contradictory) respective phylogenetic brackets (Weishampel, 1997). In other words, other character state distributions do not match, so they did not apparently evolve in step. This is a specific problem for that case. Disjunct sets of character distributions cannot support a unified functional hypothesis that purports to explain the evolution of an adaptation (although in this case an exaptation may be possible).

One shortcoming of most functional explanations for bizarre structures in extinct dinosaurs is that the evolution of these features and functions in a clade is very seldom considered. Without doing so, there is no evidence that the function (in the sense of an adaptation) evolved at all, and therefore the hypothesized function itself must be considered in doubt, unless there is good independent evidence of it. The demonstration of its evolution requires a phylogenetic component.

Phylogenetic dissection of adaptation (PDA)

When paleobiologists discuss functions of bizarre structures, they are generally discussing adaptations. It is a truism of evolutionary biology that adaptations are shaped by natural selection (Williams, 1992). Paleobiologists cannot assess selection in populations through generations, as microevolutionists can (e.g. Endler, 1986; Brandon, 1996). But they can assess natural selection at a more general hierarchical level in lineages, living and extinct, by mapping the elaboration of structures and the improvement of proposed functions upon phylogenies based on other characters (e.g. Padian, 2001; Padian & Horner, 2002, 2004).

In order for an adaptation to be assessed (Padian, 1982, 1987), its necessary components must be identified and separated from non-essential ones. By plotting these character states on a phylogeny built from other characters, the assembly of the adaptation can be traced. Even after the basic adaptation is assembled, further modifications can be tracked in the same way (Padian, 2001). This method of PDA can be formalized in the following way (modified from Padian, 1982, 1987, 1995, 2001):

  • 1Identify the adaptation, its diagnostic (vs. merely associated) features and the groups that possess it.
  • 2Perform phylogenetic analyses of the groups, including closest sister taxa, using all available character taxa.
  • 3Identify the phylogenetic sequence of acquisition of each diagnostic feature of the adaptation.
  • 4Analyze the apparent roles, if any, of diagnostic characters at each successive stage before the adaptation is assembled, using functional, physical, ecological, genetic and other lines of evidence.

The implication of this method for the assessment of bizarre structures in dinosaurs is that, if such explanations are to move beyond the ad hoc, they must be able to explain the evolution of these features, the assembly of their characters and functions. In other words, at successive nodes along the spine of the cladogram, one should be able to point to specific characters diagnostic of the proposed adaptation, and assess their function with respect to the organism as a whole. Such assessments need to take into account the roles of other features in the functional complex in order to provide an adequate cross-test (Padian, 2001). Moving to successive nodes along the spine of the cladogram, the evolution of the features from stage to stage should emerge. If there is no evidence for the improvement of a function or the assembly of a new one, the adaptive hypothesis fails. Therefore, functional explanations that are not tested phylogenetically have no demonstrated evolutionary basis and are of limited value (Fig. 2; Weishampel, 1997).

Figure 2.

 Phylogeny as a test of the assembly of an adaptation. Consider an adaptation defined by five necessary features (1–5) following a PDA analysis (Padian, 1995, 2001; see text). Phylogenetic analysis reveals the clade that shares this function (and thereby shares all five features); it also reveals the order of assembly of those five features on nodes along the spine of a more inclusive cladogram. At each of the nodes along this spine (which have successive subsets of those five features), a different function can be proposed and tested; new secondary (exaptive) functions and improvements in existing functions may also be proposed and tested.

Testing mechanical hypotheses

We divide these into four general (and not mutually exclusive) classes: defense, communication, thermoregulation and sensory function.

Defense

These features can be attributed to repulsion of predators and to conspecifics of the same sex in agonistic behaviors (non-exclusively). Notable examples are the horns and frills of ceratopsians, the plates and spikes of stegosaurs, the scutes and tail club of ankylosaurs and the domes of pachycephalosaurs. Weishampel (1981) tested the possibility of a defensive function of lambeosaurine crests and concluded that the bone was too thin to have been of any use in this regard.

Ankylosaurs would seem to pose the least controversial example of a defensive function for bizarre structures, in this case the dermal scutes (traditionally and tellingly called ‘armor’) and tail ‘club’ (in ankylosaurids only: Carpenter, 1997, 2001; Vickaryous, Maryanska & Weishampel, 2004). Scutes cover the skull, the neck, the back, and much of the tail, but there is great variety in their size, form and extent among ankylosaurs (Carpenter, 1997). This suggests that there was no ‘optimal’ pattern of scute form and distribution, and therefore it is difficult to propose that a defensive function was successively ‘improved’ in ankylosaurs. However, consideration of their outgroups shows that ankylosaurs had more extensive dermal ossifications than the basal thyreophorans Scutellosaurus and Scelidosaurus (the latter often considered an ankylosaur), not to mention the stegosaurs, which lost all but the parasagittal rows (Main et al., 2005). This pattern points to defense as a plausible basal function of ankylosaur scutes, and suggests that whatever the variations in scute form and distribution, they were ‘good enough’ to serve an adequate defensive function. Yet, as Carpenter (1997: p. 315, fig. 22.6) notes, the variation in scute form, and notably in the more conspicuous long neck spikes, suggests no obvious defensive strategy (see also Scheyer & Sander, 2004), and may instead be primarily related to display. Sexual dimorphism has not been established, so sexual selection has no support, but social selection (Hieronymus et al., 2009) could be investigated further. Several evolutionary strategies may have been involved here. The enlarged and fused scutes at the end of the ankylosaurid tail, preceded by a series of fused caudal vertebrae, have often been invoked as a weapon, and this seems to be supported by the enlarged areas of muscle attachment on the pelvis, hindlimbs and transverse processes of the anterior caudal vertebrae, despite some limits in vertical mobility (Vickaryous et al., 2004).

Most attributions of defense to the frills of neoceratopsians have focused on Triceratops (Fig. 3). This is apparently because Triceratops has prominent orbital horns as well as a solid frill, so its function in ‘jousting’ is easily visualized (e.g. Farke, 2004). However, Triceratops is virtually (along with Avaceratops) the only neoceratopsian with a solid frill, which is also the shortest among large neoceratopsians (Fig. 3). Other large neoceratopsians have substantial openings in their frills, which would have been of little use in defense. It now turns out that the adult Triceratops is in fact what has been called Torosaurus, and its frill is not only fenestrated but also quite thin, as in other neoceratopsians (Scannella & Horner, in press). We hypothesize that, because it is so similar to young Triceratops, the adult form of Avaceratops may turn out to have been fenestrated as well.

Figure 3.

 Cladogram of advanced ceratopsians, after Dodson et al. (2004). No specific functional or behavioral hypotheses are currently proposed (or have been tested) to account for evolutionary trends in the features related to the bizarre cranial characters of these taxa.

And horns vary widely; chasmosaurines had orbital horns of various sizes and orientations, but most centrosaurines had small orbital horns, and nasal horns of variable size that show no obvious function in combat (Farke et al., 2009). Farke (2004) used restored scale models of Triceratops to determine how individuals might have fought each other, interlocking horns, and Farke et al. (2009) showed that injuries occurred significantly more often on skull bones that would have been expected according to his predictions. However, even if this function is plausible, it has not been proposed and tested for other chasmosaurines, although it was absent in centrosaurines (Farke et al., 2009). The most recent published phylogenies of neoceratopsians (Xu et al., 2002; Dodson, Forster & Sampson, 2004; Fig. 3) show no directional pattern of improvement of either brow horns or nose horns. Hence there is no evidence for adaptation to a particular function, and other hypotheses also need to be considered as a general explanation for the evolution of horns and frills.

For stegosaurs, as Main et al. (2005) have shown, the elaboration of plates and spikes shows no phylogenetic trends in adaptation to proposed functions of thermoregulation (Galton & Upchurch, 2004b). The possible function of defense has been rejected by several authors (Buffrenil et al., 1986; Main et al., 2005): the plates consist of a thin layer of compact bone surrounding a central core of well-vascularized, lattice-like (spongy) trabecular bone that would be crushed easily by the teeth of any large theropod. A possible function in deterring predators by making the animal appear larger has been suggested, but again it would not explain why Stegosaurus has large plates and those of the contemporaneous Kentrurosaurus and others are much smaller.

Pachycephalosaur domes have been assumed to have been used in head-butting, ever since Colbert's (1955) casual suggestion (review in Maryanska, Chapman & Weishampel, 2004). However, histological studies have shown that the columnar cell structure of these domes would not have deflected the forces incurred in battering, as reasonably proposed by Sues (1978) on the basis of biomechanical models of gross anatomy. The spongy bone that was thought to be protective of the brain during head-butting, by analogy to similar bone in bighorn sheep (Galton, 1970), is actually characteristic of juvenile skulls; the skulls of adults, in which most head-butting would have been expected to occur, have compact bone in their external cortices (Goodwin & Horner, 2004). Moreover, the spongy bone of juvenile skulls is organized in such a way that reflects radial growth of the bone, which indicates rapid growth (Francillon-Vieillot et al., 1990: p. 512). Rather than deflecting concussive forces from the brain cavity, this radial organization would have more likely directed them into the brain cavity (Goodwin & Horner, 2004).

Maryanska et al. (2004) recently renewed the argument for mechanical agonistic behavior, but their analysis had no control for ontogeny or sexual dimorphism, so there is no support for assigning male status to larger and thicker domes as they did. Moreover, the knobs and spikes that ornamented some pachycephalosaur skulls (such as Stygimoloch) would not have been visible until the heads were lowered, and in any case could not have been involved in combat (Goodwin, Rosner & Johnson, 1998). For these reasons a function in combat for both the domes and ornamentation is implausible. Moreover, there is now evidence that Stygimoloch was a subadult form of Pachycephalosaurus, which has somewhat less extreme spikes than Stygimoloch, thus casting doubt on the functional interpretation (Horner & Goodwin, 2009).

Communication

Weishampel's (1981, 1997) study of Parasaurolophus described above was the first example of an explicit test of a hypothesis that a particular structure functioned in communication. As noted, this function may apply to this genus, but it has not been proposed and tested for other lambeosaurines until recently, when Evans, Ridgely & Witmer (2009) examined Lambeosaurus, Corythosaurus and Hypacrosaurus. They showed, as Weishampel (1981) had done using Lophorhothon, that the ear region was capable of hearing the low-frequency sounds that Weishampel calculated might have been produced by the resonating crests of these hadrosaurs. However, phylogenetic analysis of lambeosaurines (Horner, Weishampel & Forster, 2004) shows no apparent trends in selection for improvement of the features related to this function (Fig. 4), and Evans et al. (2009) noted that Hypacrosaurus altispinus had a particularly derived and convoluted nasal chamber.

Figure 4.

 Cladogram of lambeosaurine hadrosaurs, after Horner et al. (2004). No specific functional or behavioral hypotheses are currently proposed (or have been tested) to account for evolutionary trends in the features related to the bizarre cranial characters of these taxa.

Thermoregulation

Only two kinds of dinosaurian structures have been proposed as thermoregulatory structures. The first is the plates of stegosaurs. Main et al. (2005) showed that the explanation hypothesized for stegosaurs (Buffrenil et al., 1986) could not be completely eliminated for Stegosaurus itself but was unlikely to apply to related taxa, so there was no evidence of the evolution of a functional adaptation in the group. The other example is the frills of ceratopsians; like the plates of stegosaurs, these structures bear numerous superficial vascular grooves that could be interpreted as conductors of blood vessels that could modify body core temperatures (Rigby Jr, 1990). However, this hypothesis has never been rigorously tested, despite some intriguing evidence (Barrick et al., 1998), and it is more conservative to suppose that the blood vessels nourished the rapid growth of frills and plates, which seem to have become more elaborated at the sub-adult stage (Horner & Marshall, 2002; Dodson et al., 2004; Main et al., 2005).

Sensory

Ostrom (1961, 1962) proposed that the crest of Parasaurolophus-enhanced olfaction: that is, an extended nasal epithelium with sensory cells may have improved the animal's ability to smell. However, as Hopson (1975) noted, lambeosaurine crest variability is too great to be explained simply by selection for olfaction. Moreover, lambeosaurines had no particularly specialized or enlarged olfactory lobes in the brain, compared with other dinosaurs (Ostrom, 1961; Evans et al., 2009).

Procuring food

Bizarre structures such as tusks are used by some animals to procure food, but to our knowledge no such function has been seriously proposed or tested for dinosaurs.

Display

Display functions can be divided broadly into antagonistic versus attractive: the repulsion of various threats versus the attraction of potential mates (Table 1). But sometimes, as in many mammals and some birds, these functions are related (Darwin, 1871). Attraction only applies to the other sex of the same species, but not all structures involved here fall into the category of sexual selection.

Interspecific

Hypotheses about structures that may play a role in repelling potential predators are difficult to test. Buffrenil et al. (1986) determined that the plates of stegosaurs were not well constructed to resist the bites of predators such as Allosaurus. The plates may have made the animals look larger, and this function may also be attributed to most bizarre cranial structures of dinosaurs, as well as to the plates of ankylosaurs (Carpenter, 1997). However, it is difficult to know how to test this hypothesis. Moreover, the evolutionary literature suggests that structures hypothesized to repel predators in living forms, whether by aposematic mimicry or agonistic display, do not appear to enjoy long-term success unless the threat they promise can be fulfilled (Futuyma, 2009).

Intraspecific

(i) Intrasexual: Females seldom contest each other, except to establish social hierarchies (as in some mammals that travel in social groups or herds), but males commonly contest males, among both invertebrates (notably arthropods) and vertebrates (Darwin, 1871). In general, territory and resources form the basis of male competition in mammals and in birds. Possession of resources is usually linked to competitive superiority among males, and this advantage in turn makes males more able to secure females, or more attractive to females, because females are thought to perceive greater advantage in mating with these males. (Some birds short-circuit the process or use a proxy to attract females through colorful feathers or eloquent songs [Darwin, 1871; Andersson, 1994].) Some bizarre structures in extinct dinosaurs may have threatened rivals, but this is difficult to test without direct knowledge of behaviors that are not preserved in the fossil record.

(ii) Intersexual: The principal means of intersexual display is display for mates, traditionally called sexual display. Sexual display usually implies sexual selection, and explanations of sexual selection must be evaluated much like those for mechanical adaptations. In contemporary populations, sexual selection often acts on minor features and elaborates them (Mendelson & Shaw, 2005); intense sexual selection can result in runaway selection (Futuyma, 2009) and (or) divergent selection (Kroodsma et al., 1985; Price, 1998). Evolutionary theory holds that this kind of divergence can result in speciation (Futuyma, 2009), and that like natural selection, sexual selection can be responsible for patterns of sorting in clades (Vrba, 1984; Sampson, 1999). This could be shown if the characters subject to sexual selection show non-random trends in clades (though the variation of the trends themselves does not have to be directional or trendlike).

A problem with invoking sexual display as the explanation of bizarre structures can be traced to Darwin's (1871) original formation of the problem of sexual selection. Darwin emphasized that sexual selection could only apply when one sex bears structures used in intersexual display (or agonistic behavior in intrasexual interaction). In other words, sexual selection cannot be invoked without discrete, qualitative features of sexual dimorphism. (We acknowledge that many neobiologists [apparently originating with West-Eberhard's, 1983 conflation of the concepts] feel that sexual dimorphism is not necessary for sexual selection, but Darwin defined the concept in this way and by definition he cannot be wrong. This does not deny that various other phenomena associated with competition for mates and reproductive success are interesting and important; but they are not strictly part of sexual selection.) Unfortunately, this degree of sexual dimorphism, typical of birds and some mammals, has not been sufficiently established for dinosaurs.

(iii) Social selection: This concept (West-Eberhard, 1983) was recently applied to dinosaurs by Hieronymus et al. (2009), who argued persuasively that the nasal cornifications of centrosaurine ceratopsians were progressively selected for larger size and broader display. According to them, ‘social selection occurs when there is differential success in within-species competition for any limited resource.’ Two problems with this definition, as applied to fossils, are that within-species phenomena can almost never be observed, and competition is particularly difficult to establish in extinct forms (Benton, 1996). On the other hand, it is possible to identify structures that can plausibly have functioned only in social interaction (as opposed to food gathering, thermoregulation, etc.) and that are not sexually dimorphic (so are not related to sexual selection), as Hieronymus et al. (2009) did for centrosaurine nasal horns. However, in any case social selection reduces to a kind of natural selection.

Moreover, these authors do not accurately distinguish social selection and species recognition. They state (2009: 1394) that ‘species recognition traits are under selection only in the earliest stages of courtship during mating’, following West-Eberhard (1983); but species recognition is simply a matter of possessing traits that allow an individual to recognize others of its species, for many functions besides breeding. They also state that ‘species recognition traits are only expected to occur in closely related sympatric species,’ as opposed to being able to ‘diverge in allopatric isolated populations,’ but in our view species recognition can begin at the population level and can easily diverge in populations of a single species, especially if the selective change is anagenetic.

Contrary to West-Eberhard (1983), species recognition does not entail ‘reproductive character displacement,’ or necessarily any features that relate to mating, reproduction, or competition among individuals of a species (Mayr, 1963). Those other terms are the provenance of mate recognition, social selection, and natural selection. She rightfully criticizes earlier work that attributed to species recognition many phenomena due to sexual selection or social selection (such as the hypothesis that signal distinctiveness should be reduced on islands and in isolated (allopatric) populations (West-Eberhard, 1983: 165). That was sorted out with further experimental work, but it does not nullify the concept of species recognition or imply that it is indistinguishable from these other processes. This confusion aside, it is possible to assess the predicted effects of species recognition and to separate them from those of other hypotheses.

(iv) Species recognition– Under the explanation of species recognition, bizarre structures would have no apparent mechanical function and would not specifically evolve to attract members of the opposite sex for mating (viz., Vrba, 1984; Paterson, 1993); rather, they make it easier for individuals to recognize others of the same (and different) species. That is, the bizarre structures communicate to other individuals a variety of possible associational cues, including species identification, potential protection and social habits and the appropriateness of potential mates. They are positive indicators of beneficial social affiliations. There can be a strong ontogenetic component to this process: young neoceratopsians, pachycephalosaurs and lambeosaurs lacked the extent of cranial ornaments of fully grown individuals, although they had rudimentary development, and it appears that in many cases these ornaments were rather rapidly developed at or around the attainment of adult size. Larger members of a species, whether male or female, and whether or not socially dominant, thus advertise their biological affiliation.

It is often difficult to differentiate among hypotheses of species recognition, social selection and mate recognition, even in living animals. All three are forms of intra-species recognition, but less general and also different in critical respects: it is first necessary to recognize other members of the species, and then to recognize (in the right seasonal and ontogenetic contexts, because mating in most species is not year-round and does not involve all members of the population) individuals that could serve as potential mates or rivals. This is a different process than developing gender-specific structures that assist in the specific attraction of mates, or the repulsion of intraspecific competitors for mates, which is the domain of sexual selection. Below we propose some tests of the species recognition hypothesis that distinguish it from the sexual selection hypothesis. In extinct animals only hard parts generally provide evidence, and so any evolutionary hypotheses must have an evidentiary basis in preservable structures.

Dinosaurs and sexual dimorphism

Because sexual dimorphism has been so extensively invoked to explain ‘bizarre structures’ in dinosaurs (e.g. Chapman et al., 1997), we address it in detail here.

Saurischia

Sexual dimorphism has been proposed for several theropods (mostly basal forms assigned to ‘ceratosaurs’) and ‘prosauropods’ (a paraphyletic group of basal sauropodomorphs), on the basis of an apparent difference between robust and gracile forms (Colbert (1989, 1990) on Coelophysis; Raath (1990) on Syntarsus). Differences have been noted in the relative thicknesses of bone walls, and in the morphology of trochanters. Unfortunately the statistical evidence that supports sexual dimorphism as an explanation for these differences is problematic. For example, Colbert (1990) produced considerable evidence for ontogenetic change in proportions in Coelophysis, but his inference of sexual dimorphism (widely accepted by other workers) was based on only two specimens. In Syntarsus, the difference between the ‘gracile’ and ‘robust’ morphs of the iliofemoralis trochanter is almost non-overlapping with respect to the size of the bone (represented by width of the femur head: Raath, 1990: Fig. 7.8). The size-frequency distribution of femoral ‘morphs’ is also non-overlapping with respect to the femoral head width (Raath, 1990: Fig. 7.10). Simply put, there are no small ‘robust’ morphs. Moreover, these examples are not sexual dimorphism in the sense established by Darwin (and John Hunter before him); if valid sexually, they are simply slight sexual differences, so they cannot be invoked to support sexual selection.

An alternate possibility, that these features could be ontogenetic, is suggested by Raath's data. A broader trochanter (and possibly thicker cortex, though the correlation has not been statistically assessed) may have been acquired by both males and females as they reached sexual maturity. Sexual dimorphism has also been suggested for tyrannosaurs (Carpenter, 1990; Larson, 1997), but Carr (1999) has shown that many apparently dichotomous differences in the craniofacial skeleton, such as numbers and forms of teeth, are purely ontogenetic (as may be the case for Syntarsus), so the ‘gracile’ forms are simply juveniles. We suspect that this will hold for other dinosaurian species in which minor variations in size and structure are found, rather than the discrete structures specified by Darwin (1859, 1871) for true sexual selection.

Other bizarre structures in theropods include cranial crests (Dilophosaurus, Monolophosaurus, Cryolophosaurus) and horns (Carnotaurus and incipient frontal structures in allosaurids and tyrannosaurids); however, neither sexual dimorphism nor ontogenetic maturity can yet been examined statistically for these features.

The argument about alleged gracile and robust dimorphic adult forms follows, ceteris paribus, for the studies cited on prosauropods by Galton & Upchurch (2004a: p. 257), who provided no statistical demonstration of dimorphism, and by Weishampel & Chapman (1990), who reached inconclusive results for Plateosaurus.

Ornithischia

Sample sizes in species of stegosaurs, ankylosaurs, pachycephalosaurs and most ornithopods are too small to test the hypothesis of sexual dimorphism; it has been proposed for hadrosaurs and ceratopsians. Goodwin (1990) noted that the sample of pachycephalosaurs was too small to permit statistical evaluation of sexual dimorphism, and Goodwin & Horner (2004); Horner & Goodwin, (2009) showed that most observed variation was ontogenetic, based on independent analysis of stage of maturity using the degree of fusion of the cranial sutures and the progressive growth and reduction of specific cranial features.

Hadrosaurs

Sexual dimorphism in hadrosaurs has long been accepted by authors (e.g. Davitashvili, 1961; Hopson, 1975; Molnar, 1977; Weishampel, 1997; Carrano, Janis & Sepkoski, 1999); the supporting evidence can be traced almost entirely to Dodson's (1975) study of two genera of lambeosaurine hadrosaurs. Dodson's morphometric analysis suggested that ‘procheneosaurs’ were merely juveniles of larger species, and he reduced three genera and 12 species to two genera (Lambeosaurus and Corythosaurus) and three species. In these three species he thought he could detect sexual differences in some cranial characters, although not at all in postcrania; and no signal was found in most cranial characters. This is a problem because there is no independent means to correlate size with age, or to identify age of a specimen on the basis of other evidence. Evans & Reisz (2007) have shown that this variation is ontogenetic or characterizes chronospecies that do not overlap with each other temporally. And moreover, these are only slight proportional differences, not discrete structural ones.

Ceratopsians

As You & Dodson (2004) note, presumed sexual differences have been postulated in Protoceratops andrewsi (Dodson, 1976) and (less independently) Protoceratops hellenikorhinus for both cranial and post-cranial features, and some features of the frill are dimorphic. However, as noted above, the extent of variation in the supposedly dimorphic features was statistical (as opposed to presence/absence features of true dimorphism), and although they may have supported more conspicuous sexually dimorphic features in soft part anatomy that is not preserved, the statistical argument on the basis of hard parts is insufficient. The kind of variation appears much more akin to the sort of differences that characterize male and female crocodiles, which differ from each other mainly at adult size, where it is mostly a matter of relative robusticity (Webb et al., 1978; Chabreck & Joanen, 1979).

If dimorphism were important in small basal ceratopsians, it should be emphasized or at least detectable in larger, more derived forms, but this does not seem to be the case. Lehman (1990) suggested a pattern of sexual dimorphism in Chasmosaurus and related species that could be traced through later ontogeny, but the small sample sizes, incomplete preservation, and lack of association of much of this material, as Lehman noted, makes it difficult to evaluate hypotheses about sexual differences, even if they are accepted. Ryan et al.'s (2001) study of a ceratopsian bone bed, where dimorphism could be presumed to emerge, turned up no significant patterns. A recent review of Ceratopsia (Dodson et al., 2004) did not accept sexual dimorphism as a general feature in this clade of dinosaurs. Soft-part features and behaviors that are not preserved in extinct taxa may well have contributed to sexual selection (e.g. Sampson, 1997). However, to invoke them for extinct groups of dinosaurs is outside the pale of homological and analogical comparison.

As for fossil birds, which are dinosaurs, we have almost no information about dimorphism; long tail feathers in the basal avialian Confuciusornis are suggestive (Chiappe et al., 1999), but this is not enough to establish evolutionary polarity. Because dimorphism (and not just inter-sexual difference) is generally low in other reptiles (Fig. 5), the EPB does not support sexual dimorphism in non-avian dinosaurs on the grounds of homological comparison.

Figure 5.

 We ask whether there is any support in the extant phylogenetic bracket of extinct dinosaurs for a broader distribution of dimorphism in ‘bizarre’ skeletal structures. This question would not be affirmatively answered by size dimorphism (as in crocodiles and some other reptiles), nor by soft part structures (as in birds) or behaviors, which cannot be assessed in fossils. The answer is ostensibly negative.

Comparative tests of the mate competition and species recognition hypotheses

Vrba (1984) used the example of degree of horn differentiation, which is usually greater in alcelaphine bovids (hartebeest, wildebeest, etc.) than in the related aepycerotines (impalas), to suggest an explanation for the greater species diversity through time of the former clade. Sampson (1999) suggested that sexual selection, not just natural selection, could be the motor of enhanced diversity in certain subclades over others. He proposed a Mate Recognition Hypothesis (MRH) by which selection for positive recognition of mates could lead to increased differentiation of populations and eventually greater rates of speciation in some lineages over others. This idea has a strong backing in recent research on the value of sexual selection in promoting differentiation of populations in a single species (e.g. Andersson, 1994; Price, 1998; Mendelson & Shaw, 2005). But like any other hypothesis that involves sexual selection, a degree of sexual dimorphism is required that is not found in dinosaurs.

We propose that species recognition is a simpler and more general explanation for the patterns seen in the distribution of bizarre structures in dinosaurs. Structures that promote species recognition allow individuals of a single species to recognize each other and distinguish conspecifics from members of other species. Advantages include banding together for protection from predators, parental care and the possible location of mates. As explained above (Display: Intraspecific), this is a broader and more hierarchical function than that proposed by the Mate Recognition Hypothesis, and it does not require sexual dimorphism. It can also involve many other kinds of cues than visual, let alone those related to bizarre structures. The fact that these various functions exist apart from simple mate recognition is witnessed by the appearance of bizarre structures, often in incipient form, in individuals not involved in mating at all.

If species recognition has been important in influencing macroevolutionary trends, it should have some empirical tests by which its effects can be differentiated from those of other hypotheses. We propose two.

First, the pattern of diversification of bizarre structures in clades should be relatively random: it should not show trends that could ostensibly be related to selection (merely size-related change would not qualify). An example, necessarily simplified, is presented in Fig. 6. In the diagram at left, the pattern of change documented through time shows clear directional trends. This kind of change is readily explained by selective forces, whether natural or sexual. The standard model is of variation in populations, followed by directional selection. This can represent improvement of a function (natural selection) or continued trends in mate preference (sexual selection; runaway sexual selection is an extreme condition). A gradation of forms is expected both within and among lineages: gradual improvement is expected in a single lineage, whereas adaptive divergence (for ecological or sexually selective reasons) should characterize differences among lineages.

Figure 6.

 Expected differences in macroevolutionary patterns of morphology between regimes governed by natural selection (or Fisherian sexual selection) and species recognition. Although these are extremes of a continuum of patterns, it would be expected that natural selection would cause more or less linear morphological trends through time as the function of a structure improved. Sexual selection might also direct the orthal elaboration of a difference into a morphological trend. But when species recognition is selected for, the object is to be recognizably different, not necessarily functionally better or more visually elaborate; so a range of morphology is possible, as long as the result is divergence. From Main et al. (2005).

In the diagram at right in Fig. 6, however, there is no obvious trend in evolutionary change; the only objective of evolutionary change is to make a lineage different from other closely related lineages (e.g. Figs 3 and 4; Main et al., 2005: fig. 10). This pattern represents what would be more likely expected from the species recognition model. The direction and degree of difference are not important or predictable; not all possible dimensions of morphospace are expressed. Taxonomic diversity is not necessarily higher under either model.

Second, there could be evidence that at some point, several closely related species with divergent bizarre structures lived at the same time in environments that at least partly overlapped. In other words, several contemporaneous sympatric, parapatric, or partly allopatric species existed when these lineages were diverging. These differences might have been positively selected as a means to reinforce associations (including mating) with appropriate conspecifics. However, lineages may also continue to diverge in isolation from others simply because this kind of evolutionary change follows a natural flexibility of phenotype. So, white-crowned sparrows diverge at the local populational level at a very rapid rate, changing songs in ways instantly recognizable to human birdwatchers as well as to the birds themselves (Baptista, Bell & Trail, 1993; Bell, Trail & Baptista, 1998). These songs both reinforce populational identity and allow mate recognition. But the populations may not overlap geographically to any great extent. Drift may also play an important role, especially in small populations with some isolation (Mayr, 1963; Eldredge & Gould, 1972). Many evolutionary changes occur in lineages because certain organisms have the evolutionary ‘habit’ of changing regularly, not because they are adjusting to myriad continuous demands of natural or sexual selection. Female preferences can change quickly, and even ‘anticipate’ desirable variations that later appear in males (Futuyma, 2009).

In this way, we predict that the species recognition hypothesis can account for both the differentiation of related sympatric species and the anagenetic change in lineages that may indeed characterize much of dinosaurian evolution, including putative ontogenetic stages and sexual dimorphs (e.g. Evans, 2007).

Discussion: explaining bizarre structures in dinosaurs

Morphological diversification in the bizarre structures of dinosaurs does not seem to show clear patterns of directional evolution within clades. To date, no satisfactory adaptive explanation has been proposed and tested for the evolution of bizarre structures in any dinosaurian clade (not simply an individual species). The most recent phylogenetic analyses of these clades do not reveal trends in the morphology of these structures that indicate any directionality that can be attributed to adaptive improvement or sexual selection (Weishampel et al., 2004). We stress that this does not deny the importance of mechanical adaptation, sexual selection, or any other macroevolutionary process in dinosaurs; it simply concludes that to date there is no evidence that it has shaped any bizarre morphology in a clade. The fossil record (like the living record) provides only a sample of the diversity that has existed, and our phylogenetic reconstructions would be very different with a different or more complete sample.

The second test of the Species Recognition model supposes that several contemporaneous lineages in a clade with bizarre structures should overlap geographically to some degree during their divergence. The geographic and temporal distributions of some dinosaurian clades, including ceratopsians, pachycephalosaurs, lambeosaurines and ankylosaurs, suggest that these bizarre structures appeared in dinosaurian groups that lived at nearly the same time in nearly the same areas (Table 2). Until recently, a salient exception appeared to be the stegosaurs, whose low diversity was anomalous to this general model. However, new discoveries of stegosaurs have increased our knowledge of their diversity: Carpenter (2001) estimates that at least five stegosaur species are now known from the Morrison Formation of the western United States (although Galton & Upchurch (2004b) recognize only three). This would appear to simplify the problem, but there is an additional caveat: the species are not all contemporaneous (K. Carpenter, pers. comm., 2004), and there may be geographic differentiation as well within the Morrison.

Table 2.   Contemporaneous occurrences of several species of related dinosaurs with ‘bizarre’ structuresa
  • a

    All information from Chapter 24 of Weishampel et al. (2004). Note that even though several related taxa may be present in a formation, they have not necessarily been identified from the same localities or stratigraphic equivalences within formations. For implications, see discussion of Fig. 7 in the text.

  • b

    The dominance of stegosaur diversity has come to light mostly in the past few decades. However, in some formations, non-‘bizarre’ dinosaurs, such as theropods, sauropods and iguanodontids, are more diverse in the Middle and Late Jurassic.

  • c

    Ankylosaurs replace stegosaurs in the Early Cretaceous as the most diverse ‘bizarre’ dinosaur clade. However, the non-‘bizarre’ iguanodontids are still very diverse in places, and in China a great diversity of basal birds has been discovered in the past decade: were feathers ‘bizarre’ structures that functioned as devices of species recognition?

  • d

    This formation also contains, among non-‘bizarre’ taxa, three tyrannosaurids, at least two ornithomimosaurs, at least three oviraptorosaurs, and at least two dromaeosaurs.

  • e

    The diversity of non-‘bizarre’ taxa in these formations is proportionally comparable to those in the previous note. ?Denotes uncertain identification.

  • Ant, Antarctica; Brit, Britain; Fm., formation; Fr, France; Ger, Germany; NAm, North America.

Early Jurassic: Cryolophosaurus (Sinemurian-Pliensbachian, Ant), Dilophosaurus (Sinemurian-Pliensbachian, NAm, China)
Middle Jurassic:Stegosaurus, Lexovisaurus, ?Omosaurus (Bathonian-Callovian, Brit); Lexovisaurus (Callovian, Fr, Ger); Huayangosaurus, unidentified stegosaur (Bathonian-Callovian, China)
Late Jurassic: Five stegosaur species (Morrison Fm., Utah), four stegosaur species (Morrison Fm., Colorado), five stegosaur species (Morrison Fm., Wyoming) (all Kimmeridgian-Tithonian, NAm); stegosaurs, not diverse (mostly Dacentrurus armatus) (Oxfordian-Kimmeridgian, Brit; Kimmeridgian-Tithonian, Fr, Port); at least four stegosaur species (?Oxfordian, Upper Shaximiao Fm., China)b
Early Cretaceous: Six named ankylosaur species (Cambridge Greensand, Brit), two ankylosaur species (Wessex Fm., Brit) (late Aptian).
Many species of avialians (Jiufotang and Yixian Fms., China).c
Late Cretaceous: Five neoceratopsian (‘protoceratopsids’ and relatives) species (?mid-Campanian, Djadokhta Fm., Asia).
At least four ankylosaur species, at least 10 hadrosaur species, four pachycephalosaur species, at least 10 neoceratopsian species (Campanian, Dinosaur Park Fm., NAm);d equal or less diversity of these taxa in coeval and slightly younger formations such as the Bearpaw Shale and the Horseshoe Canyon Fms.;e at least five pachycephalosaurs, at least three neoceratopsians (Maastrichtian, Hell Creek and Lance Fms., Montana and neighboring states)

The lack of contemporaneity could have several explanations, including insufficient stratigraphic sampling to establish that more of these species lived at the same time than it now appears. However, another approach is phylogenetic. If these five species turned out to be morphotypes of a single anagenetic lineage, there would indeed be no evidence for contemporaneity. Would the hypothesis of species recognition thereby be weakened (Fig. 7, left)? In fact, such a result would weaken a hypothesis of anti-hybridization, but it would not weaken or test the hypothesis of positive assortative mating (Paterson, 1993). However, if phylogenetic analysis revealed that these species indeed represented different lineages, and their ‘ghost ranges’ indicated that they must have diverged from others at an earlier time, then at one time the test of contemporaneous species would have been passed (Fig. 7, right). It is not impossible that such a pattern could also indicate other processes than species recognition, such as sexual or social selection, but in concert with non-directional evolutionary change the indication would be rather more strongly in favor of species recognition. Phylogenetic analysis and further biostratigraphic sampling can test this hypothesis.

Figure 7.

 The ‘ghost’ of species recognition. Consider five species in a lineage, distributed sequentially through time; the real extent of temporal overlap among forms is unknown because the record is necessarily incomplete. In the figure at left, these five forms are part of a single anagenetic lineage. In this case there is no evidence that species recognition influenced the diversification of lineages (speciation), although species recognition, drift, or several other kinds of selective factors could account for the anagenetic change. In the figure at right, the same five forms belong to separate lineages with successively ancient divergence points. This pattern provides a basis for testing the hypothesis that species recognition, among other factors, may have been involved in the diversification of the clade. Although as in the figure at left, no more than one of the five species is found at the same time as another, phylogeny reveals their ghost lineages, and allows the inference that, for example, at time X at least four of the lineages may have been contemporaneous.

Finally, we return to the test of the Mate Recognition Hypothesis that Sampson (1999) proposed. We found that in every criterion, mostly related to higher rates of speciation and habitat shifts, the concept of ‘species recognition’ could be substituted for the terms related to sexual selection without any apparent difference in results. The exception was his fourth criterion (speciation will often be correlated with vicariance events rather than the formation of peripheral isolates), which we suggest is untestable in the fossil record, and in any case would not discriminate between sexual selection and species recognition as a cause. In summary, the criteria for MRH and SRH are very similar in outline; but in any given case the tests of these characters, and perhaps the characters themselves, would be quite different operationally for the same animals. For example, the horns of ceratopsians might satisfy all four (five) criteria listed above for both MRH and SRH, but would not pass the test of high sexual dimorphism required for sexual selection; on the other hand, they appear to pass the two tests of the species recognition hypothesis (non-directional variation of bizarre structures and several sympatric species). Moreover, without a clear demonstration of sexual dimorphism, the MRH reduces to the social selection hypothesis (Hieronymus et al., 2009).

Conclusions

Our purpose is not to insist that species recognition has been the only cause of the evolution of bizarre structures in dinosaurs, nor that adaptation, social selection and sexual selection have been unimportant in dinosaurian evolution. We merely ask in each case: how would we test this? We conclude that the hypotheses of mechanical function and sexual display that have predominated for decades as general explanations of the evolution of these structures in dinosaurian clades are unfounded. When we test the hypothesis that presumed functions of these structures have evolved in their clades, we find no evidence; hence the notions that these structures are ‘adaptations’ fail the criteria proposed by evolutionary biologists (Greene, 1986; Williams, 1992; Rose & Lauder, 1996; Padian, 2001). Furthermore, sexual dimorphism has not been strongly established for any bizarre structures in dinosaurian lineages, even though mild dimorphism has been statistically demonstrated in at least one lineage and may be plausible in others. If criteria of sexual behavior other than those based on sexual selection (which requires sexual dimorphism: Darwin (1871) are to be proposed, they should be justified on grounds that are more stringent than weak analogies to very different living organisms.

We stress that no evolutionary hypothesis can be regarded as a ‘default’ explanation (i.e. if a certain class of explanation fails, then another one is automatically strengthened or must be accepted by default). Hypotheses must be independently tested, or they are not scientific. In many or most cases, definitive tests will not be possible. We have proposed two tests of a Species Recognition hypothesis, and there may be others. In our view, most dinosaurian bizarre structures pass these tests, but they do not pass the tests of adaptation or of sexual display. The importance of social selection (Hieronymus et al., 2009) remains to be tested in dinosaurs beyond individual species. This does not mean that these structures were not adaptive or used in attracting mates; we simply have no evidence on these points at present.

Our hypothesis is that the Species Recognition Hypothesis is simpler and more general in explaining the evolution of bizarre structures in dinosaurs than those of mechanical function, social selection, or sexual selection/mate recognition. Rigorous tests of these complementary evolutionary hypotheses should be applied to other lineages besides dinosaurs. Bizarre structures are common in many fishes, as well as other reptiles. In birds, sexual dimorphism, display and selection are well-established phenomena that have clearly had a very strong role in shaping avian evolution. The expression of bizarre structures in mammals, notably ungulates, is entailed in a constellation of ecological characteristics that greatly complicate their explanation (Jarman, 1974; Perez-Barberia, Gordon & Pagel, 2002).

Finally, we emphasize that a given structure may have several purposes, and that even in living animals it is often difficult to determine the uses of particular structures, their evolutionary histories, and even how the animals are communicating. In this respect the hypotheses of paleobiologists are largely interpreting the shadows on the wall of Plato's cave. We persist in efforts to explain these structures because they were of obvious use to their bearers, and this is in principle discoverable.

Acknowledgments

We thank S. Bar-David, J. Brashares, V. de Buffrenil, K. Carpenter, P. Cross, P. Dodson, J.O. Farlow, E. Hebets, T. Hieronymus, R. Irmis, C. Janis, E. Lacey, B. Lundrigan, S. Patek, A. de Ricqlès, M.J. Ryan, S.M. Sampson, K.M. Scott, A.B. Shabel, L.M. Witmer and many other colleagues and reviewers for constructive comments and suggestions, without implying their agreement with all our points. UCB undergraduates Jasmeet K. Dhaliwal and Sylvia Moses provided research support. R. Irmis and A. Lee provided technical support. This work was supported by the University of California Museum of Paleontology and the Committee on Research of the University of California, Berkeley. This is UCMP Contribution No. 2012.

Ancillary