We are delighted by the constructive and thoughtful comments of Knell & Sampson (2010) on our original article (Padian & Horner, 2010). The reasons why so many kinds of dinosaurs evolved such bizarre or exaggerated features are not well understood, and different investigators come to the problem with different preconceptions and favored hypotheses, depending on their training. We all acknowledge that several factors may be at issue in given cases, as Darwin (1859) recognized in his original formulation of the problem. But we take issue with some fundamental assumptions that Knell and Sampson raise, which illustrate how academic fields often evolve.
Perhaps the central difference is that, in our view, mate recognition is not a category of sexual selection, but of species recognition (because an animal cannot consider mating with another unless it first recognizes that they are conspecific), and because mate recognition does not require sexual dimorphism in secondary characters; whereas, to Knell and Sampson, sexual selection does not require sexual dimorphism, and mate recognition is a more closely related concept to sexual selection.
In our view, Charles Darwin understood organismal biology better than anyone of his time, partly because he thought through problems so thoroughly. In devising his theory of natural selection, he realized that certain living animals bore some salient phenotypic characteristics, such as horns and antlers, that could not be readily explained through the agency of natural selection. He knew that these sorts of features (and their associated behaviors) would pose a threat to the acceptance of his theory of natural selection (because they would be seen as fatal exceptions), and he also understood that these features were not, in most cases, directly relevant to an individual's survival (i.e. ecologically adaptive). Rather, they helped an individual attract mates or repel rivals for mates. The opposite sex lacked these features (or did not use them in mating). To this phenomenon he gave the name sexual selection, and he explicitly defined it (1859, pp. 89–90 and nearly the same words in 1871, p. 243) by stating that these features were present in one sex and not the other, and that they either attracted mates or repelled rivals.
This was the genesis of the term sexual selection. Darwin needed this concept as a contrast to natural selection, in order to defuse a false argument of his detractors; he knew exactly what he was doing. Darwin cannot be ‘wrong’ about the definition of this concept, despite the protests or confusion of later authors, because he invented it, and his empirical basis for it is entirely valid; he was not ‘imprecise’ (paceCarranza, 2009). Myriad examples prove the presence of distinct, monosexual characters in species that are used to attract mates and repel rivals (Darwin, 1871; Andersson, 1994). Thus, the only possible definition of sexual selection requires sexual dimorphism (and not simply allometric sexual differences: Padian & Horner, 2010).
To say this does not deny that many factors are involved in the attraction of mates, the repulsion of rivals, success in mating and the differential production of offspring. But sexual selection is only a small part of this, and Darwin was not trying to explain all aspects of mate recognition, attraction, competition and reproductive success. The notion that sexual selection involves more or different aspects than those defined by Darwin is an historical error of misinterpretation in the scholarly literature that has sadly become entrenched. But one cannot change the definition of a term at will. This only creates confusion and fosters misinterpretation [consider the later misuses of Van Valen's (1973) original concept of the ‘Red Queen,’ which denoted the control of energy in an ecosystem by individual species]. We acknowledge that Darwin's term is widely misused in the recent literature, and unfortunately, this has brought confusion to an extremely interesting and productive field.
Futuyma (2009), whose textbooks have been the ne plus ultra of the field for many years, views sexual selection as a subtype of natural selection, and many biologists agree. There is an historical context for this misunderstanding. In the early 20th century, mathematical modelers of the Modern Synthesis needed to examine whether natural selection could be a viable force in populations (Mayr & Provine, 1980). The Darwin–Wallace hypothesis was that individuals that were more fit for their environments would wind up leaving their features to the next generation, in which they would be proportionally better represented. (That is a testable hypothesis that can be examined based on the presumed adaptive features of the individuals in question.) However, because these modelers could not quantify how well adapted a particular individual is, they modified the concept of ‘fitness’ by taking a shortcut. In their terms, the number of offspring would be an indirect measure of adaptive fitness. This proxy for Darwin's idea of ‘fitness,’ which became purely reproductive, had nothing to do with individual fitness to an environment (Waddington, 1967) and begged the question whether more offspring produced was a true measure of adaptive fitness, thus creating a tautology that creationists exploited for decades (Bethell, 1985; Johnson, 1991; Wells, 2000). (‘Who are the best adapted? Those who leave the most offspring. Why do they leave the most offspring? Because they are best adapted.’) But Darwin was not talking about how many offspring an individual leaves; he was talking about the potential to survive and eventually to leave offspring with one's adaptively superior traits. This link in the causal chain is important: without it, all discussion of selection merely centers on a competition to leave offspring, which ignores the core of Darwin's theory as he presented it in the Origin.
We have a parallel problem with the history of the term ‘sexual selection.’ Present-day experts acknowledge that its use is greatly confused (Clutton-Brock, 2007; Carranza, 2009). Arnold (1994) fostered some confusion by taking the ‘shortcut’ to reproductive success, defining the term as ‘selection that arises from differences in mating success (number of mates that bear or sire progeny over some standardized time interval)’ without incorporating Darwin's requirement of sexual dimorphism and the prior differential success in attracting mates and repelling rivals. Cornwallis & Uller's (2009) redefinition embodies decades of terminological deterioration in denoting the term as ‘any variation in direct fitness [the component of fitness gained by producing your own offspring] among different phenotypes caused by their ability to gain sexual partners, produce fertile eggs and generate offspring.’ For them, sexual selection is almost entirely about the number of offspring produced. And, for most biologists educated in the literature of population genetics, Darwinian fitness (the outcome of natural selection, for them) is purely a measure of how many offspring one leaves. No wonder so many biologists regard sexual selection as a subtype of natural selection. If both concepts reduce simply to leaving more offspring, why would one think otherwise?
But these revisionary definitions are misguided: there can be no concept of sexual selection without sexual dimorphism (and not just allometric size difference, as between male and females of many species). This does not mean that the hundreds of studies performed on mating factors are incorrect, misguided or invalid, just because they have misused Darwin's term. To the contrary, we are gifted with an incredible literature related to the interactions of the sexes; but only a small part of this pertains to what Darwin defined as sexual selection. Mate recognition, mate competition, mating success and reproductive output are fascinating topics on which many important papers have been published. But sexual selection is involved only when a feature or behavior that one sex has (or uses) that the other sex does not is used to repel rivals or attract mates, and so gain access to reproduction. That is where Darwin's (1859, 1871) concept begins and ends. We do not need to redefine or expand it (Clutton-Brock, 2007; Carranza, 2009); we need to return to its original meaning. To do so eliminates confusion related to consequent questions such as the function of bizarre structures in dinosaurs, to which we now turn in responding to some of Knell and Sampson's specific points.
1. With few exceptions, sample sizes for individual dinosaur species are too small to conduct statistical tests for the presence of sexual dimorphism. We agree, but as we showed, this has not stopped many paleontologists from arguing in its favor, without testing for other causes of variation, such as ontogeny or phyletic (anagenetic) change in evolution. On the other hand, we do have many well-represented dinosaur species, including those that bore bizarre structures, and these are tractable to statistical analysis.
2. Bovid males use their horns predominantly in competition for mates. Yes, and in these bovids sexual dimorphism is generally high; females usually lack horns, except in small species, in which both sexes typically have small horns. There are really no living animals related to or comparable with the Mesozoic dinosaurs that we discussed in these respects. Analogies to living animals and their patterns of behavior must therefore be tested stringently.
3. Species recognition has not been documented as a key factor in the evolution of exaggerated traits among any extant animals. We think it has generally not been examined, and the hypothesis certainly cannot be rejected. Very few living animals have exaggerated structures comparable with those of Mesozoic dinosaurs, and to our knowledge extensive studies of species recognition have not been carried out on those that do, although this would not be prohibitively difficult. The present is sometimes a key to the past, but it is not its universal arbiter.
4. We assume that traits under directional selection evolve slowly enough for directional change to be evident on phylogenies of extinct clades. We do not pretend to know how rapidly these changes occurred, or even what triggered them in these dinosaurs. There is increasing evidence of anagenetic change between species that previously were considered sister taxa (e.g. Evans, 2010; Scannella & Horner, 2010), and these changes may have taken thousands to tens of thousands of years or less, judging by biostratigraphic distributions. However, we are making a rather different point that we would not expect directional trends within clades in which species are simply evolving to be recognizably different from each other. This is a quite different process, and on a quite different scale, than for example the directional ‘runaway selection’ of sexual characters seen in living populations (e.g. Kirkpatrick, 1982; Andersson, 1994).
Knell & Sampson (2010) also cite the strong evidence that beetle horns are frequently used in mate competition, but are not known to function in species recognition. We agree, but these beetle horns are different in important respects from the structures we discussed in dinosaurs. First, they are often dimorphic, as Knell and Sampson noted. Second, large horns may deter predators on both males and females, whereas there is no evidence that the bizarre structures of dinosaurs deterred predators. We agree with Darwin (1859, p. 90): ‘Yet I would not wish to attribute all such sexual differences to this agency [sexual selection]: for we see peculiarities arising and becoming attached to the male sex in our domestic animals …, which we cannot believe to be either useful to the males in battle, or attractive to the females.’ In beetles, as Knell and Sampson describe, several morphological patterns and evolutionary processes are at work, and we do not wonder that their evolutionary trends are not simply directional.
5. Living animals do not universally show the pattern we predicted, that species recognition traits would be expected to become exaggerated among close relatives living in sympatry or parapatry. Knell and Sampson claim that because this correlation is not universal in living animals, it ‘weaken[s] any inferences based upon the fossil record.’ This is an untenable application of actualism, because it posits that all biological possibilities must be realized in the present-day biota, and that a lack of universality in the present implies impossibility in the past. It seems preferable to propose and test criteria in specific cases, because the relationship between morphology and behavior is so complex.
Knell and Sampson propose that multiple contemporaneous, closely related species could also evolve under sexual selection, and we agree. But we predict differences between the consequences of sexual selection and those of species recognition. In a clade in which sexual selection is acting within several species, the focus is on selection on a range of phenotypes within that species, regardless of what other species are doing; whereas our hypothesis of evolution under species recognition predicts that species evolve so as to differentiate themselves from other species, not from members of their same species. We expect, as many studies of ‘runaway sexual selection’ have shown (Andersson, 1994), that morphological change in a species under this pressure will be relatively directional, whereas under species recognition, evolution merely has to produce differences from other species.
6. The fossil record of dinosaurs does not support the previous prediction either. In the several years since we began to develop the species recognition hypothesis and to try to devise some tests, new research has forced dinosaur specialists to rethink old paradigms. The exact duration of a dinosaur species (often <1 million years) says nothing by itself about how long it would take species within dinosaur subclades to make relatively minor but distinct changes to an already elaborate morphological structure (compare sister taxa in figs 3 and 4 in Padian & Horner, 2010). The species duration also seems to be irrelevant to the question of morphological stasis, inasmuch as the vast majority of paleobiological studies with adequate samples and controls shows a predominant pattern of stasis or random walk (Hunt, 2007). However, it increasingly seems that dinosaur species traditionally deemed distinct and overlapping in time in fact overlapped far less than previously thought, and in many cases did not represent cladogenetic splits but are rather anagenetic, chronological replacements of each other (Horner, Varricchio & Goodwin, 1992; Scannella & Fowler, 2009). In this way, 17 species of Triceratops were pared down to two morphologically distinct forms of a single anagenetic lineage that evolved through the Hell Creek Formation. This pattern is indeed not what we originally predicted for dinosaurs (although this does not negate our hypothesis as a general prediction). However, it may speak to regional biogeographic patterns in ways that we did not originally consider, if (for example) the latest Cretaceous Triceratops lineage in Montana diverged at some point from common ancestors with its sister lineage in the basins of Utah and New Mexico (Gates et al., 2010), if in fact they are distinct lineages in these regions.
7. The ‘costs’ of maintaining structures involved in species recognition should be less than for those involved in sexual selection. With all due respect, we think that there are too many complex and interrelated aspects of an animal's biology that cannot be accounted for by simple ‘cost’ models (e.g. Maynard Smith, 1982). What is the net ‘cost’ to an animal if it increases its survival and its reproductive representation in the next generation? Most major groups of dinosaurs, which dominated terrestrial environments for over 150 million years, evolved elaborate cranial and post-cranial structures that were arguably unnecessary to evolutionary success, inasmuch as most animals lack them. Clearly the ‘cost,’ for whatever cause, was less than the benefits. These ‘costs’ have to be assessed at a macroevolutionary level, not only at the level of whether a single individual incurs a greater ‘cost’ by deceiving or playing straight with other individuals.
Discussions of ‘cost’ in evolution had their genesis in ‘optimality theory’ that held that natural selection would be expected to optimize adaptation. But nearly all evolutionary biologists recognize that there are life history tradeoffs and developmental limitations involved in phenotypic plasticity, and that animals merely need to be ‘good enough’ to survive. Knell and Sampson discuss what ‘modeling studies have shown,’ but models cannot provide evidence: they are merely schemes for restating the consequences of first principles that are already accepted, or for framing data that have already shown a pattern. We do not understand how it is possible to know that elaborate structures in highly complex extinct animals ‘cost’ so much to their bearers that they could not be involved in species recognition as much as in any other evolutionary process.
Finally, Knell and Sampson suggest that the ‘cost’ of producing elaborate structures would be too high for species recognition, but worth the effort for sexual selection. We think that if these structures in dinosaurs were ‘expensive,’ it would be a waste for females to develop them as well; whereas, if they were important in recognizing other members of a species, then all the members would develop them.
8. Species recognition signals should vary less within a species than those adapted for sexual selection. The argument for this statement is that high levels of variation would increase the probability of error. We think the converse, that advantages in mating opportunities in natural populations are based predominantly on variation: namely, the males with the showiest antlers, the gaudiest plumage or the most pleasing song are likely to succeed. In order to be successful, males need to match this practical maximum as closely as possible. This would appear to select for decrease in variation. On the other hand, under species recognition, members of a species merely need to be more similar to each other than they are to members of other species, to avoid confusion.
Knell & Sampson (2010) also claim that strong positive allometry in these exaggerated structures are evidence for mate competition and against species recognition. But the evidence is often to the contrary, and sometimes in dinosaurs with exaggerated structures ‘positive allometry’ is not so simple or does not apply at all. A very small Triceratops with a skull 30 cm long (Goodwin et al., 2006) (adult skulls reach 3 m) imitates elders of his species in aspects of horn and frill ornamentation, yet he is years away from reproducing. Mid-sized Triceratops have horn and frill configurations that are still different from full-grown forms (Scannella & Horner, 2010). And the related pachycephalosaurs went through some staggering ontogenetic changes in skull form well before sexual maturity (Horner & Goodwin, 2009). These features and changes are in our view better explained within the context of species recognition, because they were irrelevant to mating and would have been of no use when interacting with other species (apart from mutual differentiation). In contrast, we propose that these ontogenetic morphs are examples of status recognition within these species, because they show the social status of individuals at various ontogenetic stages. Lambeosaurine hadrosaurs, on the other hand, display species-specific cranial crests that do not appear until the animals are about two-thirds grown in linear size (about half in mass), approximately at the time they would have been maturing sexually (Evans, Reisz & Dupuis, 2007; Lee & Werning, 2008). As in cassowaries, which also develop their cranial crests in both species at the same approximate point in growth, there is no sexual dimorphism in these features. They ostensibly show sexual maturity, and so they are also advertisements of status recognition, as the mature morphs of ceratopsians and pachycephalosaurs must have been. We regard these signals of mating receptivity as tools for mate recognition, a subset of species recognition.
Darwin (1859, 1871) admitted freely that the features of some animals could have had several functions, and in some cases the line between natural selection and sexual selection was difficult to draw. As we noted in our paper, and as Knell and Sampson agree, we see no reason not to be pluralistic about possible hypotheses. Our original paper had several aims. First, we showed that ‘functional’ arguments for bizarre structures generally fail, and no case has it been established that a hypothesized adaptive function has been improved within a dinosaurian lineage, as natural selection theory would require. Second, we argued that phylogenetic analysis of groups is essential to testing the hypothesis of adaptive trends (Knell and Sampson agree on the value of both of these aims). Third, we showed that hypotheses of sexual selection in dinosaurs are without evidence, because sexual dimorphism (and not simply possible sexual difference in minor features) has never been demonstrated. (Knell and Sampson disagree with our insistence that Darwin's definition be respected, but they do not dispute our conclusion; moreover, they differ with us in thinking that mate recognition is related to sexual selection, whereas we see it as related to species recognition.) Fourth, we showed that every prediction of the mate recognition hypothesis that is not untestable (Sampson, 1999) also applies to species recognition; in our view, mate recognition is most likely simply one function of species recognition (along with protection, care of young and so on). (Knell and Sampson demur, although we do not see any testable evidence for the mate recognition hypothesis in dinosaurs.) Finally, we proposed that species recognition is a simpler and better supported hypothesis to explain these bizarre structures in dinosaurs. We freely admit that our two tests are not perfect, because other evolutionary factors could always be involved. But, ceteris paribus, we hypothesize that natural and sexual selection should be expected to produce trends that are more linear than those from species recognition, because the only aim of the latter is to be different. We acknowledge that behavior could be involved in ways that we cannot perceive: for example, the accessory hornlets and marginal ornamentations of ceratopsians could be present in both sexes and only used by one, which would satisfy Darwin's definition.
But the bottom line is that dinosaurs were not exactly like any living animals. They altered their bizarre structures so much during ontogeny that what we can now recognize as developmental stages of a single species have traditionally been recognized as separate species and even genera. For example, the pachycephalosaurs Dracorex, Stygimoloch and Pachycephalosaurus are now known to be ontogenetic stages of the same species, even though their cranial ornamentations are grossly different (Horner & Goodwin, 2009). As noted above, the 17 named species of Triceratops now appear to be reducible to one species with two anagenetic morphs that succeed each other through time; in addition, the genus Torosaurus now appears to be the adult form of Triceratops (Scannella & Horner, 2010). No living vertebrates do anything like this, and it testifies to the complex social structure of these dinosaurs. If we try to explain their biology using untested or untestable analogies to living forms, or to accept a proposed function of a structure simply on the basis of what it ‘looks like’ it might do, we should expect to overlook important insights into some of the most marvelous animals ever to have walked the Earth.