Behavioural leads in evolution: evidence from the fossil record



There has been much discussion of the role of behaviour in evolution, especially its potential to lead morphological evolution by placing the organism in a novel selective environment. Many adaptations of living species can be imagined to have originated in this way, although documented examples are relatively few. A fruitful arena for testing hypotheses about behavioural innovation is the fossil record. Traditionally, the behaviour of fossil species has been deduced from their morphology, precluding the observation of a behavioural lead preceding morphological evolution. This circularity can be broken by examining behavioural proxies independent of the adaptive morphology itself. Examples applicable to fossil remains include dietary information (e.g. wear traces on teeth, stable isotopes) and trace fossils indicating locomotor mode (footprints). The signature of a behavioural lead would be an observed shift in behaviour from one horizon (or taxon) to another, followed later by a functionally-related morphological change. This pattern can be sought either in finely-stratified anagenetic sequences of fossils (stratophenetic approach) or among fossils with well-resolved species-level phylogenies (cladistic approach). An array of case studies from the literature is presented. These include feeding shifts in finely-resolved sequences of vertebrates ranging from freshwater fish to terrestrial ungulates, as well as locomotor changes crucial to major evolutionary transitions in the origin of tetrapods, birds, and humans. The latter examples highlight the role of behaviour in initiating exaptation (the requisitioning of structure for a new function). The case studies also illustrate the challenges of using fossil sequences to elucidate behavioural roles, including insufficient stratigraphic resolution and uncertainty over the adaptive function of observed traits. By the same token, they suggest criteria for choosing promising cases for research, as well as for formulating testable hypotheses about evolutionary modes. © 2013 The Linnean Society of London, Biological Journal of the Linnean Society, 2014, 112, 315–331.


There are various processes by which organisms adapt to their environment (Fig. 1). The core mechanism is adaptive evolution by natural selection, but adaptation can also be achieved by phenotypic plasticity, behavioural flexibility, or a shift of range to track a preferred habitat. There are significant feedbacks among these processes (Fig. 1). For example, range shifts are unlikely to leave the population in precisely the same environment and so natural selection may result. Phenotypic plasticity may be canalized into a fixed adaptation by genetic assimilation (Waddington, 1953; Lande, 2010) or, conversely, natural selection may favour adaptive flexibility in a highly variable environment, a process that has been termed variability selection (Potts, 1998). Finally, behavioural accommodation may allow the species to explore a new niche while the slower process of natural selection adapts the phenotype accordingly (Hardy, 1965). The latter process is the focus of the present review.

Figure 1.

Modes of adaptive change. A, range (distributional) changes, even when broadly ‘tracking’, place a population in a new selective regime. B, range reduction may result in extinction. C, ecophenotypic as well as genetic change may be adaptive. D, in highly variable environments, natural selection may favour adaptive phenotypic or behavioural flexibility (‘variability selection’: Potts, 1998). E, conversely, plasticity may be reduced by natural selection ‘genetically assimilating’ part of the reaction norm. F, behavioural plasticity may elicit phenotypic plasticity and vice versa. G, flexible behaviour may lead to natural selection on behavioural or other phenotypic traits: labelled the ‘Baldwin effect’ but see text for discussion. H, an organism's behaviour may modify its own environment (‘niche construction’: Odling-Smee et al. 2003) and hence its selective context.

Behaviour is crucial because it is the means by which adaptive form is applied to the environment. Although standard accounts of evolution explain the origin of adaptation by natural selection, they generally sidestep the question of the order in which a change in form, change in behaviour, and change in habitat come about. Darwin (1859: 183) himself stated: ‘It is difficult to tell, and immaterial for us, whether habits generally change first and structure afterwards; or whether slight modifications of structure lead to changed habits; both probably often change almost simultaneously’ (emphasis added; see also Fussey & Partridge-Hicks, 2006). He went on, however, to give examples where ‘habits have changed without a corresponding change of structure’, with the clear implication that, given time, the exploratory population would become physically adapted to its new niche. Subsequently, various authors have gone further and suggested that many key morphological features might never have evolved had the animal not first explored new habitat and/or developed new habits (Hardy, 1965; Maynard Smith, 1987). Would the ancestors of the polar bear have evolved webbed feet before they began swimming in pursuit of marine prey (Fig. 2)?

Figure 2.

Could the polar bear have evolved its webbed feet before it became a marine predator? A, coastal brown bear with flounder. B, swimming polar bear showing webbed feet. Photo credits: (A) Manuel Presti; (B) Wikimedia Commons.

Behavioural initiation of evolution includes (but is not restricted to) aspects of the ‘Baldwin Effect’. This concept (Baldwin, 1896; Simpson, 1953; Weber & Depew, 2003) emphasizes phenotypic accommodation during the lifetime of an organism as a precursor to longer-term, genetically-based adaptive change (i.e. evolution). The crucial initiating factor is generally seen as behavioural, and this is the focus of the present review, although many authors (including Baldwin himself) broadened the concept to encompass any acquired phenotypic response (Hall, 2003: 147, table 8.1; West-Eberhard, 2003: 151). A key distinguishing feature of the Baldwin effect is that the accommodation to the new situation is immediately ‘appropriate’, as opposed to genetic assimilation where an environmentally-induced trait may or may not be of adaptive value (Bateson, 2004). It therefore has more power, in a natural situation, to find the adaptive ‘needle in a haystack’ on which selection can act (Maynard Smith, 1987). Bateson (2012) has suggested the term ‘adaptability driver’ to replace the somewhat inappropriate and confused term ‘Baldwin effect’.

The behavioural response may be genetically determined, although it can also be the result of cognitive flexibility (Duckworth, 2009). Flexible behaviour, the ability cognitively to devise new solutions to problems, is mainly observed in mammals and birds, with some examples from other vertebrate and invertebrate groups (e.g. cephalopods). Selection can enhance the tendency to produce the behaviour, especially if it is persistent (often through learning) within the population. Additionally, and of particular interest here, further phenotypic adaptations to the new niche may evolve. The nature of the behavioural accommodation is likely to guide the direction of permanent adaptation (Hardy, 1965); for example, the dorsoventral flexion of the backbone in the locomotion of terrestrial mammals ensured the evolution of a horizontal tail-fluke (rather than vertical as in fishes, amphibians and aquatic reptiles), whenever they became secondarily aquatic, in seals, beavers, sea-cows, and cetaceans.

As well as behavioural accommodation to an existing habitat or resource, dispersal into a new habitat, selection of microhabitat or modification of habitat (all aspects of ‘niche construction’: Odling-Smee, Laland & Feldman, 2003; Odling-Smee et al., 2013) represent further aspects of behaviour that alter a species' ecology and may lead to evolutionary change (the ‘eco-evolutionary feedbacks’ of Post & Palkovacs, 2009). By providing ‘ecological inheritance’, niche construction provides a context of novel selection pressures that persists across generations, thereby enhancing the likelihood of adaptive evolution (Odling-Smee et al., 2013). Recent research has highlighted the importance of intraspecific behavioural (‘personality’) variation in determining the capacity of individuals to disperse, and their subsequent success in founding a population (Chapple, Simmonds & Wong, 2012; Wolf & Weissing, 2012), affecting the likelihood of subsequent adaptive change and speciation.

Therefore, following Hardy (1965), Bateson (2004) and others, a broad view of the roles of behaviour in evolution is taken in the present review. Figure 3 illustrates some alternative scenarios. In the routes highlighted by filled arrows, behaviour is crucial in moving the organism into a new environment, and/or in exploring new niche space with its existing morphology, in both cases leading to adaptive morphological change.

Figure 3.

Alternative pathways to achieving new form and function. The example is a hypothetical mammal adapting to a change from woodland browsing (round snout adaptive, pointed snout even more so) to grassland grazing (flat snout adaptive). At the circular junctions, any input can lead to any output. Pathways marked by solid arrows are the focus of the present study, and key potential contributions of behaviour are underlined. E, Environment; G, Genotype; M, Morphology; B, Behaviour; each can change from an initial state (1) to an altered state (2 or 3). Notes: ‘Niche construction’ may entail either moving to a new environment or altering one's environment. The ‘extinct’ pathways imply that, without immediate behavioural accommodation, natural selection or genetic accommodation may not be able to act fast enough to ensure survival. ‘Exaptation’ occurs when coping in a new environment because existing phenotype is being used in a new way (although natural selection can also exapt without behavioural help). ‘Red Queen’ evolution (Van Valen, 1973) implies evolutionary change in an unchanged environment and niche.

Most literature on behavioural leads in evolution has been theoretical, sometimes involving modelling (Hinton & Nowlan, 1987; Ancel, 1999; Zollman & Smead, 2010; Sznajder, Sabelis & Egas, 2012) but often essentially discursive (references in Turney, Bryson & Suzuki, 2008). Thus the plausibility of the process has been widely discussed, and many existing adaptations of organisms can be imagined to have arisen in this way. Demonstrated examples are, however, relatively few.

Losos, Schoener & Spiller (2004), studying brown anole lizards, Anolis sagrei, showed that, when introduced to islands with ground predators, the lizards adapted behaviourally by moving to higher perches out of reach of predation. This shift led, within 6 months, to significant selection for larger body size and hindlimb length on islands with predators, interpreted as adaptive to faster escape.

Cubo, Ventura & Casinos (2006) considered the water vole (Arvicola) complex in Europe, of which some populations are semi-aquatic and others are subterranean. Based on a phylogenetic study, they deduced that the semi-aquatic habit was primitive and, because only some subterranean populations show morphological adaptations to digging, they concluded that the unmodified diggers illustrate the first, behavioural stage in evolution of subterranean adaptation.

Sol, Stirling & Lefebvre (2005) found that species of passerine birds with larger relative brain size (taken to be a proxy for behavioural flexibility) generally have more subspecies than those with smaller brains, and this is not simply a result of greater dispersal ability. Assuming that subspecies richness implies local adaptation, the results suggest that behaviourally flexible lineages have undergone greater evolutionary diversification. At a higher taxonomic level, Nikolakakis, Sol & Lefebvre (2003) found that avian lineages with larger brains and a higher propensity for innovative behaviours tend to contain more species than less flexible lineages.

Badyaev's (2009) work on the North American house finch (Carpodacus mexicanus) demonstrates a remarkable cascade of adaptations in populations established in different environments, with the high dispersal ability in this group being considered a key initiating factor. Yet also for sympatric speciation, where dispersal is not a factor, likely examples among insects demonstrate the importance of behavioural factors such as assortative habitat or mate selection in initiating the process of divergence (Bolnick & Fitzpatrick, 2007).

Behaviour and Evolution in the Fossil Record

One of the key advantages of the fossil record is that it provides, in principle, the possibility of directly observing evolutionary processes in time series. Modes of evolution have been studied successfully in cases where the record is particularly good: generally, finely-stratified sequences with abundant remains of the target species in successive layers, allowing a statistical appraisal of change in morphological traits.

A major avenue of research has been in the adaptive function of such traits, and these, in turn, have been used to deduce behaviour of the species concerned, and sometimes the palaeoenvironments in which they lived. Morphological features of marine fossils, for example, often by analogy with living forms, differentiate a nektonic or benthic habitat. Among land vertebrates, limb proportions of extinct forms suggest their locomotory mode, whereas dental morphology provides clues to dietary category. Combining several species' adaptations, the spectrum of adaptive types across a fossil community has been successfully used to reconstruct its former environment. Considering locomotory adaptations, for example, a preponderance of arboreally-adapted mammals indicates forest, whereas species adapted to running on open, flat ground would imply a largely treeless plain (Kappelman et al., 1997). Turning to feeding adaptations, herbivorous mammals with high-crowned teeth, flat occlusal surfaces, a low-slung head, and a wide snout imply grassland, whereas browsing-adapted species have low-crowned, cuspate teeth, an elevated head, and narrow snout, and point to the presence of forest (Eronen et al., 2010).

Although such deductions of habitat and behaviour from the morphology of species are likely to be approximately correct for broad units of time and space, they must logically miss instances where, in an individual species, behavioural change has preceded morphological adaptation. Put another way, the Baldwin effect and related phenomena cannot be identified on the basis of behaviour deduced from morphology, because of the inherent circularity of the logic. To break this circularity, it is necessary to find indicators of behaviour that are independent of morphology per se. Changes in behaviour and morphology through time can then be separately traced, and evidence of behavioural lead sought, as signalled by a chronological lag between behavioural innovation and morphological response.

Demonstrated direct proxies for behaviour in the fossil record include:

  • Dental microwear (microscopic scratches and pits on the occlusal surface of teeth) as evidence of diet.
  • Dental mesowear (the shapes into which tooth cusps wear through use) as evidence of diet.
  • Stable isotopes (especially 13C and 15N) preserved in hard tissues, as evidence of diet.
  • Preserved gut contents as evidence of diet.
  • Coprolites (fossil dung) as evidence of diet (provided they can be confidently linked to the culprit species).
  • Sub-annual properties (e.g. isotopic composition) of growth rings in shell, bone or tooth, as evidence of seasonal migration or dietary variation.
  • Abnormal individual growth rings as markers of stress, or of life-cycle events such as weaning.
  • Ontogenetic changes as a result of use (e.g. exaggerated muscle attachments or cortical thickening of limb bones)
  • Other marks on hard tissues as a result of activity in life (e.g. high incidence of breakage of mammalian carnivore teeth as a result of bone-cracking)
  • Marks on other individuals or species (e.g. tooth-marks on prey), provided they can be confidently linked to the species responsible.
  • Preserved trackways as evidence of locomotory mode (provided they can be confidently linked to the target species, and bearing in mind that their form is influenced not only by behaviour, but also by the morphology of the locomotory organ).
  • Aspects of ‘extended phenotype’, such as nests or hives.
  • Rarely, animals preserved ‘in the act’ (e.g. a female carrying a foetus, or mating insects preserved in amber).
  • Proxy evidence of habitat, from biotic and abiotic data in enclosing sediment (provided it can be confidently interpreted as the living, and not just depositional, environment of the target species).

Other examples of behavioural traces in the fossil record are described by Boucot (1990) and Boucot & Poinar (2012).

The standard expectation in palaeontology, of behavioural and morphological change being synchronous, is illustrated in Figure 4A. If behavioural change preceded the shift in morphology, however, the pattern shown in Figure 4B would be expected. The prediction is then an initial state with morphology M1 and behaviour B1; a subsequent stage where morphology is unchanged at M1 but behaviour has changed to B2 (similar to the +/– stage of Strömberg, 2006: fig. 3); and a final stage where morphology has changed, perhaps gradually over time, to M2. That an observed morphological change has occurred in response to a particular evidenced behavioural change is inevitably a matter of interpretation; for example, the development of flippers following the move from a terrestrial to an aquatic habitat can reasonably be interpreted as an adaptation to swimming. Issues of taxonomy also affect palaeontological interpretation: the studied fossil sequence might reflect anagenetic change in a single species or even population, or a sequence of species or populations of demonstrable close relationship but not necessarily a direct ancestor–descendent series. Whereas some of the examples presented below are stratophenetic (dependent on reliable relative ordering of strata, and with later populations assumed to be descended from earlier ones), other cases depend on a cladistic approach, with behavioural proxies treated as ‘characters’ mapped onto the cladogram to estimate their order of acquisition relative to morphological features.

Figure 4.

The predicted palaeontological pattern with (A) synchronous acquisition of new morphology and appropriate behaviour; (B) behavioural lead followed by acquisition of morphology. M1, M2, initial and new morphology; B1, B2, initial and new behaviour. Solid circle and line, morphology; dashed circle and line, behaviour. The intermediate phase M1, B2 shows new behaviour but retains initial morphology (equivalent to the cartoon with the smiley in Fig. 3).

Very few palaeontological studies have been undertaken with the explicit intention of testing for a behavioural role in morphological evolution: these include those of Miocene sticklebacks by Purnell et al. (2007), Miocene–Pleistocene ungulates by Strömberg (2006), and proboscideans by Lister (2013). However, a review of the literature reveals further published studies where appropriate behavioural and morphological data are available. In some of these cases the authors noted a chronological ‘offset’ between behavioural proxies and morphological change. A selection of these case studies is reviewed below. They range from microevolutionary examples with extensive material and detailed stratigraphic control, to major innovations in vertebrate evolution where evidence is more patchy but sufficient to frame hypotheses about the behavioural role and suggest avenues of future research.

Examples from the Fossil Record

Feeding ecology and body armour in Miocene sticklebacks

Bell, Travis & Blouw (2006) and Purnell et al. (2007) studied changes in body armour (bony spines and plates) in stickleback (Gasterosteus) fossils spanning 20 kyr of a varved (annually-stratified) sequence from the Miocene of Nevada (Fig. 5). The same studies also examined feeding ecology using dental microwear, grounded in studies of laboratory and wild populations of modern Gasterosteus that showed clear correlation of microwear score with benthic versus planktonic feeding habit. In the fossil samples, feeding habit, deduced from microwear, was significantly correlated to body armour, with more benthic-feeding samples showing greater armour development. At one point in the sequence, a substantial increase in body armour took place within 150 years. This was associated with a shift from planktonic to benthic ecology but, interestingly, the anatomical changes started 100 years after the habitat change was complete and themselves took a further 150 years to complete. Purnell et al. (2007: 1887) comment: ‘This evidence of an ecological shift preceding phenotypic change suggests that this part of the sequence may record rapid evolution driven by shifts in trophic ecology and adaptation to benthic niches’. Caution is expressed that population replacement cannot be ruled out, although the existence of intermediate phenotypes of all three armour characters scored, in the period of transition (Bell et al., 2006: fig. 2), argues in favour of an in situ evolutionary transition. The likely selective force for the development of body armour in a benthic habitat is predation. Moreover, studies of modern sticklebacks demonstrate that a transformation of this rapidity is plausible; a newly-colonizing lake population showed significant phenotypic change within only 20 years (Aguirre & Bell, 2012), while developmental research indicates that substantial increase or reduction in pelvic armour can be achieved through mutation of a single control gene (Chan et al., 2010).

Figure 5.

Dental microwear and body armour in sticklebacks. Scanning electron microscope image of (A) tooth of laboratory planktivorous fish (reproduced with permission; Purnell et al., 2007) showing lower feature density but greater length of individual features than (B) tooth of laboratory benthic fish; lines mark microwear features (image kindly provided by M. Purnell). C, heavily armoured Miocene individual with three dorsal spines (D1–D3), seven predorsal pterygiophores (R1–R7), and a fully expressed pelvis (P) (reproduced with the permission of the Palaeontological Society; Bell et al., 2006). Diagrams below the specimen indicate the range of pelvic phenotypes and their scores, from 1 (typical of planktonic populations) to 3 (typical of benthic populations).

Diet and hypsodonty in Tertiary to Quaternary ungulates

Many lineages of herbivorous mammals show modification of feeding adaptation in response to the spread of grasslands through the Tertiary and Quaternary, a trend in turn resulting from global cooling and aridification. The key dental adaptation, evolved in parallel in many taxa, was an increase in the crown height (hypsodonty) of the cheek teeth, generally interpreted as an adaptation to a more abrasive diet dominated by grasses, compared to the less abrasive browsing of broad-leaved plants by their ancestors. Although the broad pattern of adaptive change in several groups of mammals (horses, proboscideans, various artiodactyl groups) broadly correlates with environmental change, detailed studies of individual species and genera, comparing morphological change with dietary proxies such as stable isotopes and dental wear, are increasingly demonstrating a more complex picture.

African proboscideans

Lister (2013) collated data from various lineages of Proboscidea (elephants and their relatives) through the Miocene to Quaternary (approximately 20–0 Mya) of East Africa, using fossils from radiometrically-dated sites. Independent proxies were plotted for (1) vegetational change, based on δ13C in palaeosol carbonate; (2) diet, based on δ13C in tooth enamel; and (3) morphological adaptation, based on molar enamel ridge count and hypsodonty index. These data show that, with the beginning of the spread of C4 grasses at 10–8 Mya (Fig. 6A), various lineages of late gomphotheres, stegodonts, and early elephants switched to a diet containing a substantial proportion of grass, compared to the browsing habit of their ancestors (Fig. 6B). However, the major adaptive response, an increase in hypsodonty, is seen only in later representatives of the elephant lineages Elephas, Palaeoloxodon and Loxodonta, with only minor advances until 5 Mya, after which they show strong directional evolution for more than 3 Myr (Fig. 6C). This pattern is suggestive of an initial behavioural shift to grazing, leading eventually to morphological adaptation.

Figure 6.

Evolution of hypsodonty in East African Neogene to Quaternary proboscideans (reproduced with permission; Lister, 2013, where details of taxa can be found). A, palaeosol δ13C (proxy for vegetation; for sources and site symbols, see Lister, 2013). B, tooth enamel δ13C (proxy for diet). C, hypsodonty index in upper third molar. In (B) and (C), triangles, stegodonts and ‘gomphotheres’; circles, elephants; diamonds: tetralophodont/elephantid intermediate. Horizontal lines mark the interval 8–5 Mya when diet changed (B) but with little morphological response (C).

The ‘lag’ of several million years is remarkable, however, both in terms of long-term survival with apparently ‘suboptimal’ dentition and the length of time required for morphological adaptation to evolve. There are several possible explanations for the delay. First, the major increase in tooth crown height (by a factor of three in some lineages) required significant reorganization of cranial morphology and musculature (Maglio, 1973), which may have taken several million years to achieve. Second, there may have been other, more immediate, morphological adaptations that allowed the initial shift to grazing. The gomphotheres and early elephants that took up grazing after 10 Mya do have a slightly increased number of enamel ridges in their molars compared to their browsing predecessors, an adaptation to grass-eating that temporarily mitigated the pressure for hypsodonty increase (Lister, 2013). Third, it is possible that the hypsodonty increase was not an adaptation to grass-eating per se but, instead, to a later environmental change. The palaeosol data (Ségalen, Lee-Thorp & Cerling, 2007; Levin et al., 2011; Cerling et al., 2011; Fig. 6A) indicate that the spread of C4 grasses was progressive, commencing approximately 10 Mya and accelerating from 4 Mya, mirroring the hypsodonty trend in the proboscideans. Conceivably, the progressive opening and drying of the habitat led to an increase in grit and dust on plant food, abrading teeth and selecting for hypsodonty (Mendoza & Palmqvist, 2008; Damuth & Janis, 2011; Jardine et al., 2012; Lucas et al., 2013). In this case, the behavioural switch to grazing led to the development of hypsodonty more indirectly, by placing the species in a habitat that later imposed an additional selective force.

North American horses

Mihlbachler et al. (2011) examined hypsodonty increase in 31 taxa of North American equids through 60 Myr of the Tertiary and Quaternary, with dental mesowear as the independent dietary proxy (Fig. 7). A shift to a more grazing habit (probably of C3 grasses) commenced approximately 22 Mya in the ‘Anchitheriinae’ stem-group, coincident with the earliest documented spread of grass-dominated habitats (Strömberg, 2006), but with only minor changes in hypsodonty (Damuth & Janis, 2011). The major increase began approximately 18 Mya with the advent of the Equini and continued, in tandem with progressive increase in grazing, for a further 12 Myr (Fig. 7C). Mihlbachler et al. (2011) identify ‘Anchitheriine’ species (in the genera Kalobatippus and Parahippus, the latter believed to have given rise to the Equini) with low (browsing-adapted) molar crowns but mixed-feeding to grazing mesowear, suggesting that these ‘populations pioneering new habitats’ were ‘under the most intense selection for increased crown height’ (Mihlbachler et al., 2011: 1179, 1181). The authors comment: ‘These observations are consistent with a hypothesis of adaptation in which the selective regime precedes the morphological change’ (Mihlbachler et al., 2011: 1180). It may be significant that, among the anchitheres, Parahippus stands out as having a large number of individuals with particularly worn teeth (C. Janis, pers. comm.). Strömberg (2006) suggests that the lag of 4 Myr between grazing and hypsodonty increase in horses may indicate that the evolutionary rate was constrained by the complexity of the required changes in the whole adaptive complex, including cranial morphology and enamel microstructure, as well as digestive anatomy and physiology.

Figure 7.

Diet and hypsodonty in North American fossil equids. (A–C) molars showing (A) low crown with browsing mesowear; (B) intermediate crown with mixed-feeding mesowear; (C) hypsodont crown with grazing mesowear (Hulbert, 2004; reproduced with permission of B. MacFadden); (D) mesowear scores (data points, shaded envelope and white trendline) and mean molar hypsodonty indices (grey trendline) for North American equids arranged temporally (reproduced with permission; Mihlbachler et al., 2011). Note offset between white (diet) and grey (morphology) trends, especially the 16 Myr ‘lag’ between 34–18 Mya. Arrowed points are anchitheriine molars combining mixed-feeding mesowear with low crown height, and thus are hypothesized to have been under the most intense selection for hypsodonty increase.

However, as with the African proboscideans discussed above, it is possible that grass-eating did not by itself impose selection pressure, and began in a partially-closed habitat with relatively little airborne dust to select for hypsodonty, or by grazing tall grass with relatively little adhering soil. As extensive areas of grassland opened in North America after 22 Mya (Strömberg, 2006), a behavioural shift into more open habitat, and/or a switch to shorter grasses, would have elicited the selective regime. However, any interpretation is limited by a lack of direct proxies for the dust or grit levels experienced by feeding ungulates. Strömberg (2006) cites studies of dental microwear and the sedimentary context of fossil horses suggesting little evidence of significant grit ingestion during critical phases of horse evolution and therefore considers adaptive lag a more plausible explanation.

Other examples

Studies of several other groups of fossil mammals also indicate an apparent discrepancy between feeding and morphology (Mihlbachler & Solounias, 2006; Strömberg, 2011; Jardine et al., 2012), even if there are not always sufficient data to examine chronological trends. Three examples are given below.

In a study of the extinct North American ruminant group Dromomerycidae, Semprebon, Janis & Solounias (2004) found that, in general, adaptive morphology correlated well with diet indicated by microwear and mesowear, but in some cases did not. For example, species of the Late Miocene Cranioceras show tooth-wear indicative of mixed-feeding (i.e. incorporating grass as well as browse) but retain a skull morphology similar to ancestral browsing species. The authors suggest that ‘some dromomerycid taxa might have been eating food materials that they were not optimally adapted to handle efficiently’, and that ‘skull morphological changes may lag behind actual dietary practices’ (Semprebon et al., 2004: 438, 440). In the contemporaneous and closely-related Pediomeryx, a similar mixed-feeding profile is associated with derived cranial features more adapted to grazing, suggesting a second stage in the process. This study is important in that it considers dietary adaptations other than molar hypsodonty (cf. Jardine et al., 2012). Although the evolution of hypsodonty might depend on selection as a result of dietary grit and hence not coincide with a shift to grass-eating, cranial adaptations (for example the broader muzzle in grazers than browsers) relate directly to the cropping and chewing of plants of different types and heights (Mendoza, Janis & Palmqvist, 2002). They therefore would have been expected, if their modification were coincident with behavioural change, to have changed in step with the proxy evidence of dietary shift.

Studying oreodonts (an extinct group of North American artiodactyls of uncertain affinity), Mihlbachler & Solounias (2006) found wide variation in diet (evidenced by mesowear analysis) through the Cenozoic, both within and between species, which did not always correlate to hypsodonty. These include low-crowned species with periodically significant grass intake that may have been behaviourally ‘pushing the boundary’, together with derived, relatively hypsodont species assumed to be capable of eating a variety of plant types and therefore under relaxed section, illustrating the two stages (if not in ancestor–descendent relationship) of the process of adaptation.

MacFadden & Shockey (1997) examined a range of herbivorous mammals in the Pleistocene of Bolivia, determining feeding habit (C4 grazing versus C3 browsing) from δ13C in tooth enamel. Most species showed a correlation between C3 intake and hypsodonty index but three brachydont species showed a strong grazing signature. These were the gomphothere proboscidean Cuvieronius (analogous to the grazing gomphotheres of the African Late Miocene; see above) and two camelids: a species of Llama, and a species comparable to Vicugna. Damuth & Janis (2011: 751) comment: ‘Were such animals … caught at a moment of time when they were struggling to maintain themselves in the face of an inappropriate diet?’. This would imply an interval of strong selection rarely caught in the fossil record (Mihlbachler & Solounias, 2006), even if, in this example, there is no opportunity to look for an evolutionary response in morphology, since the assemblage is geologically very recent. The alternative explanation, a lack of selective pressure because of low dust levels in a semi-closed environment, cannot be invoked in this instance because palaeobotanical evidence indicates a very largely open habitat (97% grass pollen) with only scattered trees and shrubs (MacFadden & Shockey, 1997).

The origin of tetrapod locomotion

The origin of terrestrial locomotion in land-living vertebrates entailed many morphological innovations in the transition from fins to limbs, including the development of feet with digits, pelvic apparatus connected to the vertebral column by a sacrum, limbs able to move in a ‘walking’ fashion, and associated musculature. Current evidence suggests that the first limbed vertebrates were primarily aquatic and that digitated limbs evolved before the ability to walk on land (Clack, 2009).

King et al. (2011) studied locomotion in the living Protopterus, a member of the lungfish, the extant sister-group to the tetrapods. This fish ‘walks’ underwater using its elongated bony fins to propel itself against the substrate. The predominant use of the pelvic (hind) fins, the alternating as well as bounding gaits, and the body held aloft above the substrate, all recall tetrapod locomotion and may provide one model for its origin. Although lungfish are not the direct ancestors of tetrapods, Protopterus is of interest in the present context in that it shows walking behaviour preceding ‘any obvious morphological specialisation for walking’ (King et al., 2011: 21149). It is possible, therefore, that behaviour of this kind prefigured the evolution of quadrupedal walking, with or without digits, and subsequently the origin of digitated feet. The authors point out that the other extant sarcopterygian (lobe-finned fish), the coelacanth Latimeria, also uses an alternating ‘gait’ in its fin movements (albeit not against the substrate), so this behaviour may be primitive for the group that included the ancestors of tetrapods.

In the fossil record, some hints of modification of forelimb elements toward a support function are seen in the most derived tetrapodomorph fish (Panderichthys and Tiktaalik), although the pattern of origin of the digitated limb remains essentially unknown (J. Clack, pers. comm.). Recent biomechanical analysis of the skeleton of the early tetrapod Ichthyostega (approximately 380 Mya) suggests propulsion by the front limbs pushing side-by-side against the substrate, with the hind limbs incapable of walking and probably used as paddles (Pierce, Clack & Hutchinson, 2012). This mode of locomotion differs from that observed in the living sarcopterygians (see above). It may represent an enhancement of aquatic adaptation in a lineage not directly ancestral to tetrapods. If primitive for tetrapods, however, it predicts a complex and currently undocumented behavioural transition to terrestrial walking.

Remarkable, early tetrapod tracks are preserved in the Late Devonian (approximately 395 Mya) of Poland (Fig. 8; Niedźwiedzki et al., 2010). These include isolated impressions of digitated feet that could have been made by a creature similar to Ichthyostega or Acanthostega. There are also series of smaller prints arranged in trackways, implying a pattern of walking involving all four limbs in alternating strides with the body clear of the substrate, which are features of terrestrial tetrapods. The trackway impressions are simple, without impressions of digits, and it is uncertain whether their maker had a fully-formed foot or not. It is therefore not yet possible to confirm whether quadrupedal walking behaviour arose after the acquisition of a digitated foot, which is implied if the Ichthyostega model of locomotion is primitive, or before it, as suggested by the Protopterus study.

Figure 8.

Early tetrapod locomotion. A, trackway from the Devonian of Poland, showing manus and pes prints in diagonal stride pattern; presumed direction of travel from bottom to top. B, generic Devonian tetrapod based on Ichthyostega and Acanthostega, fitted to the trackway. Reproduced with permission (Niedźwiedzki et al., 2010).

The origin of avian flight

Extant birds possess a highly-modified complex of skeletal structures used in flight, as well as associated soft-tissue features of musculature, respiratory system, nervous system, and so on. It is now widely accepted that birds evolved from bipedal theropod dinosaurs (Chiappe, 2007), and recent years have seen the discovery of a remarkable array of feathered dinosaurs and early birds, especially from China. These finds illustrate the assembly of the avian body plan through a series of increasingly bird-like transitional anatomical stages. At the same time, many different models for the origin of flight have been proposed, from arboreal gliding to terrestrial jumping or running, including catching prey (Heers & Dial, 2012: table 1). Strikingly, almost all of these models are centred on behaviour, proposing that feathers, proto-wings, and associated structures evolved initially for functions other than flight but, through their ability to provide lift, they became modified in stages for increasingly sophisticated flight.

Direct evidence of locomotory behaviour is limited in known fossil theropods and birds. However, juveniles of living birds show transitional skeletal and feather morphologies that are remarkably similar to those of extinct theropods, and their locomotor capabilities provide convincing evidence with respect to form–function relationships in the extinct forms, and the behavioural stages leading to the development of true flight. In this way, the well-documented theropod record can be interpreted to illustrate the stepwise acquisition of a character complex by successive, alternating innovations in behaviour and morphology.

Heers & Dial (2012) show how, in living birds such as the chukar (Alectoris chukar), juveniles lack many of the flight adaptations of adults, with unfused thoracic vertebrae and sacrum, small pelvis, very small keel (sternum extension that attaches flight muscles), and feathers that are symmetric (vanes the same width on either side of shaft). In all these features, they are similar to theropods, and as young birds mature these features develop in partial ‘recapitulation’ of the evolutionary sequence in the fossil record. Of particular interest are the ways young birds utilize these morphologies in locomotion. In chukar, 7–8-day-old chicks engage in wing-assisted ‘flap-running’ up inclines, and controlled flapping descent (e.g. from a perch), their underdeveloped wings and feathers providing limited but useful aerodynamic force. Older, 18–20-day-old individuals, with more developed skeleton and feathers, are additionally capable of brief episodes of flight. Given the similarity of their anatomy to theropods with proto-wings and symmetric feathers, and support from biomechanical reconstructions (Hutchinson & Allen, 2009), these behaviours appear likely for bird ancestors and would have provided the selective context in which true flapping flight could evolve.

The evolution of bipedality in hominins

The origin of habitual bipedality in humans, from a largely arboreal, quadrupedal ancestor, involved a suite of anatomical modifications to the skeleton. In the hind limb, the foot transformed from a grasping structure to a weight-bearing platform with shorter toes, a large non-opposable hallux, a large heel, and an arch. The hip and knee joints enlarged and the vertebral column was placed closer to the hip joint. Leg length increased and the femur became slightly angled medially to form the ‘bicondylar angle’ bringing the knees under the body during walking.

One of the earliest known hominin fossils, Orrorin, already shows at approximately 6 Mya clear bipedal adaptations in its hip joint, specifically a spherical and anteriorly-rotated head and elongated neck of the femur. Other parts of the skeleton, including curved phalanges, indicate a retention of tree-climbing adaptation (Richmond & Jungers, 2008). The foot of Ardipithecus, more than 4 Mya old, similarly has several features suggestive of bipedalism but retains ape-like features such as a very divergent big toe. It was probably both a tree-climber and an occasional upright walker. Various species of Australopithecus, in the range 4.0–1.5 Mya, show further modifications of the foot toward bipedal adaptation, including a more aligned hallux and the presence of a longitudinal arch, but, overall, their skeletons suggest a facultative ability to both walk and climb trees.

Direct behavioural evidence comes in the form of preserved footprints, both the shape of the individual print and the pattern of walking indicated by trackways. The earliest are from the Laetolil beds, Tanzania, approximately 3.7 Mya and generally attributed to Australopithecus afarensis (Fig. 9). Recent experimental and simulation work (Raichlen et al., 2010; Crompton et al., 2012) has demonstrated that the Laetoli hominins walked erect with weight transfer similar to the economical extended-limb bipedalism of modern humans. This bipedal functionality was implemented largely by soft-tissue innovation but with an internal bony configuration differing from that of modern humans; for example, in the less expanded hallux (big toe) (Crompton et al., 2012).

Figure 9.

The origins of human bipedality. A, footprint trail cf. Australopithecus afarensis from Laetoli, Tanzania. B, false-colour depth maps of (left) Laetoli mean footprint and (right) modern human. Redder shades within the ‘footprints’ indicate greater mean depth. Reproduced with permission (Crompton et al., 2012). C, wild chimpanzee walking bipedally while carrying three papayas (one in each hand and one in mouth). Reproduced with permission (Carvalho et al., 2012).

A further line of evidence on behaviour is provided by epigenetically sensitive traits that are modified by an individual's activity pattern (Ward, 2002). In Orrorin, for example, cortical bone is thicker on the lower side of the femoral neck than on the upper side, in contrast to the situation in great apes but similar to that in humans. The difference is considered to be a result of bone remodelling in response to the configuration and usage of the limb abductor muscles in life, and is interpreted as providing ‘direct evidence for frequent bipedal posture and locomotion’ in Orrorin (Galik et al., 2004: 1453). The bicondylar angle of the femur is also an epigenetically labile trait: in a study of normal and non-ambulatory children, the existence of the angle was found to be the result of a habitual bipedal gait; it does not form in individuals who engage only in intermittent bipedality (Ward, 2002). Its presence in Australopithecus is therefore taken to demonstrate that it was a habitual biped.

Overall, hominins from Orrorin to Homo habilis show a varying mosaic of ape-like and human-like locomotory morphology (Harcourt-Smith & Aiello, 2004; Richmond & Jungers, 2008), an essentially human bony foot first appearing in Homo erectus. A footprint trail from Ileret, Kenya, at approximately 1.5 Mya, and assigned to Homo ergaster/erectus, is of modern form, with more longitudinally-aligned hallux and a narrower instep than the Laetoli prints (Bennett et al., 2009), and a strong ball and hallux impression indicating the antero-medial weight transfer that is the hallmark of modern human walking.

The likelihood that the ancestors of H. erectus, whichever among the array of known species they were, showed facultative terrestrial and arboreal behaviour, implies that positive selection was stronger on the terrestrial mode, leading to anatomical modification to optimize for terrestrial locomotion at the expense of tree-climbing. Alternatively, it has been suggested that arboreal behaviour cannot be confidently deduced from the retention of primitive anatomical traits and that A. afarensis and related species were already obligate bipeds (Harcourt-Smith & Aiello, 2004). This scenario only strengthens the cardinal importance of their bipedal behaviour in providing the selective context for ‘modernizing’ their anatomy.

Recent studies of orangutans (Pongo pygmaeus) and chimpanzees (Pan troglodytes) graphically illustrate alternative possible first steps towards bipedality in an essentially arboreal human ancestor. The two examples occur in very different contexts but both rely crucially on behavioural innovation, in species lacking the morphological adaptations to bipedality discussed above. Modern chimpanzees adopt temporary bipedality at ground level to carry objects by hand (Fig. 9C; Carvalho et al., 2012). Orangutan, by contrast, walk bipedally along branches, while the hands are used for feeding, balance or weight transfer (Thorpe, Holder & Crompton, 2007). Evidence that one of these modes was at the origin of hominin bipedality would be a convincing case of behavioural flexibility leading morphological evolution. Such evidence might come in the form of early (> 6 Mya) bipedal trackways made by an essentially unmodified foot, especially if linked to a species considered to be a potential ancestral hominin.


Fossil sampling

The case studies discussed above vary in the degree to which they meet the ideal requirements for demonstrating behavioural leads in evolution. However, all fulfil the key requirement of providing at least some behavioural proxy data independent of adaptive morphology. In one case, the theropod-bird transition, the model is somewhat stretched because the behavioural evidence for proto-flight comes not from the fossils themselves but from the behaviour of morphologically-similar modern analogues.

The second requirement is for an adequate fossil record, either in the form of a finely-stratified sequence of ancestor–descendent populations (stratophenetic approach) or a well-resolved cladogram of closely-related taxa (cladistic approach). The varved Miocene sequence of Nevada (Purnell et al., 2007) is a near-ideal system, allowing a lag of only 100 years to be observed between behavioural and morphological change in the stickleback fish. In many other areas of the fossil record, it may be impossible to determine whether an observed ‘simultaneous’ appearance of new morphology and matching behaviour reflects genuinely simultaneous acquisition, or a behavioural lead too short to be observed with the given stratigraphic resolution. We must also acknowledge that some evolutionary responses, especially where suitable genetic variation and developmental pathways are already available, may be essentially synchronous with the behavioural driver (Post & Palkovacs, 2009), so no ‘lag’ would be detectable even in the most finely-stratified fossil sequences.

The theropod-bird and horse studies are examples of successful analysis of behaviour and morphology based on cladistically-ordered sequences of species. Other, potentially interesting examples were excluded from discussion because of incomplete knowledge of relationships. For example, Miocene beavers Stenocastor show evidence for swimming behaviour in their claws but had not evolved a flattened tail (Hugueney & Escuilllé, 1996). However, Stenocastor is not considered close to the ancestry of modern Castor, and current evidence is insufficient to establish whether swimming adaptations evolved once or more than once in the group.

Functional interpretation

Some of the cited examples illustrate evolution in a character with direct functional relationship to the observed behaviour (e.g. the evolution of hypsodonty in mammals switching to more abrasive food). In other examples, morphology has changed in response to other aspects of the new niche (e.g. the development of body armour in sticklebacks switching to feeding in the benthos).

Whether an observed morphological change has occurred in response to a particular behavioural change is, however, a matter of interpretation. As noted by Strömberg (2006), temporal coincidence of behavioural and morphological change is evidence for their adaptive link, but the longer the ‘lag’, the harder it is to apply this criterion. The other main line of evidence is functional interpretation of the observed morphology, although this may not be obvious or can be erroneous. For example, the question of whether hypsodonty in ungulates is adaptive for grass-eating per se, or to life in an open environment, is disputed, as discussed above. Similarly, the cranio-dental morphology of certain australopithecine hominin species was assumed to be adapted to cracking hard foods, but this is not substantiated by microwear and isotopic studies (Grine et al., 2012). Either the morphology in question evolved for a different function, or else it was not, at the time of sampling, being utilized for the function for which it was originally selected. There is also, however, the issue of correct interpretation of the behavioural proxy. Lucas et al. (2013) provide evidence that scratches in hominin dental microwear are primarily the result of abrasion by mineral particles, and may not accurately reflect food type (e.g. grass-eating) as previously assumed.

Finally, we must recognize that an animal's potential repertoire of behaviour is not decoupled from its anatomy: physical structure constrains which behaviours are possible. For example, biomechanical study of the early tetrapod Ichthyostega (Pierce et al., 2012) suggests restricted shoulder and hip mobility, so the animal lacked the necessary rotary motions to push the body off the ground and move the limbs in an alternating sequence. The paddle-like hind limb, moreover, was quite strongly adapted to aquatic locomotion. Quadrupedal walking may not have evolved from an Ichthyostega-like morphology but, if it did, initiation of walking would have had to wait for permissive morphological change. Only once appropriate structures are available, are they capable of being behaviourally co-opted to new function.

Evolutionary pattern and process

The ideal sequence illustrating a behavioural lead is the three-stage pattern illustrated in Figure 4, including the crucial transitional phase of unaltered morphology but modified behaviour (M1, B2). The initial behavioural shift might be seen in all sampled individuals of a species, or only in some individuals or populations. The latter would be interesting in suggesting behavioural variation, with the more exploratory individuals testing the new niche space. Such individual variation is seen in Figure 6 (African proboscideans) and Figure 7 (North American horses).

It is impossible, however, given the incompleteness of fossil remains, to be certain that no morphological innovations were providing adaptation during the transitional interval of apparently purely behavioural accommodation. This would include soft-tissue or physiological changes not preserved in the fossil record, although we can at least search thoroughly for any change in preserved hard parts. A degree of anatomical modification during the phase of behavioural innovation does not preclude the essential role of the behaviour in precipitating further morphological change. An example is seen in the case of African proboscideans, where minor dental modifications accompany the switch to grazing but are later followed by much more profound ones (Lister, 2013). The new behaviour and niche, albeit with the aid of already slightly modified anatomy, have elicited further anatomical specialization.

It is evident that the interplay of behavioural and morphological change is an ongoing, sequential process, especially in the building of complex character assemblages such as that associated with flight in birds: each anatomical modification is likely to be associated with a shift in behavioural repertoire, creating the context for the next innovation, and so on.

Behaviour leading morphology is, however, not the only possible pattern of change. Many advantageous mutations do not require behaviour at all to be positively selected; for example, the first mutant theropod with proto-feathers could have been selectively favoured thanks to enhanced thermoregulation. Only later might behaviour have been involved in exapting the feathers for flight. In another model, mutation affecting morphology might lead the animal to adopt novel behaviours to survive, which, if successful, could result in selective spread of the new phenotype: the inverse to the ‘behavioural lead’ model. A celebrated example is the goat born without front legs that learnt to walk bipedally (West-Eberhard, 2003: 51). Dover's (2000) concept of ‘adoptation’ postulated genetic processes leading to anatomical change, organisms then seeking out an appropriate habitat or niche. Finally, theory has suggested that behaviour may in some circumstances inhibit morphological adaptation (Duckworth, 2009). Provided independent behavioural proxies and an adequate fossil record are available, it is in principle possible to test between these various modes, or at least between their predictions regarding the temporal relationship of behavioural and morphological change.

A further potentially important factor is the role of phenotypically plastic morphology, if an adaptive region of the reaction norm becomes fixed by selective narrowing of the developmental spectrum (Fig. 1: genetic assimilation). Phenotypic plasticity (Lande, 2010; Piersma & van Gils, 2010) complicates the tracing of the behaviour–morphology interaction in the fossil record, in that observed morphological change might be part of a pre-existing reaction norm rather than adaptively selected during the time-period under study. It is not always possible to determine from the fossil record whether an observed morphological change is genetically-determined (i.e. an evolutionary change) or an expression of phenotypic plasticity (ecophenotypic change). The latter is often limited to relatively simple changes in body size or form, and is more readily reversible (Lister & Green, in press); even so, more complex, long-term phenotypic changes could be initiated, at least, ecophenotypically (Pfennig et al., 2010). However, this is a general issue in the interpretation of adaptive morphology in the fossil record, and is not limited to the study of behavioural leads. An ecophenotypically altered morphology could itself also interact with behavioural innovation in leading to fixed phenotypic change.

External factors influencing the process (the ‘change of environment’; Fig. 1) include not only physical factors such as climate and topology, but also other animal or plant species in a competitive, synergistic, predatory or prey relationship to the species under consideration. Competition, for example, is a likely driver of the original behavioural shift in many cases, and may subsequently select for character displacement (Pfennig & Pfennig, 2012). The external environment can also be altered by behaviour, not only at the level of the individual organism (niche construction) but, also, in larger-scale ‘ecosystem engineering’, such as burrowing and bioturbation, affecting many species as well as those responsible for the initial behaviour (Erwin, 2008).

A key conclusion of this review is recognition of the potential importance of behaviour in the process of exaptation (cf. Strömberg, 2006). Behaviour can forge a new use for a structure provided it has some functionality in that role (i.e. it is ‘preadapted’) and can become exapted by further modification. Thus, the proto-wings of theropods, if originally evolved for functions other than flight, could have provided lift when flapped in an appropriate fashion. Similarly, the tetrapod hind-foot, if it initially evolved for swimming (as in Ichthyostega), may later have been exapted for terrestrial locomotion. Behavioural exaptation to a new role can be predicted almost always to lead to evolutionary structural refinement.

The behaviour of an organism is often more proximal to selection pressure than the structure itself, constituting the activities that allow the individual to survive and reproduce. Selection on behaviour will ‘drag’ with it the morphology, or any other aspect of phenotype, that is utilized in, or facilitates, the behaviour. For example, it was suggested above that preferential selection for terrestrial behaviour in hominins, from an ancestor with a dual, facultative terrestrial and arboreal habit, led to terrestrialization of the locomotory apparatus.

We must be careful not to regard a lineage as ‘heading’ to a supposed optimal condition seen today or in its terminal members. If the behaviour and morphology of a fossil species were in the process of adapting, it was toward evolutionary equilibrium with its environment at the time. Nonetheless, the adaptive process, especially the evolution of complex structure, must take time, implying that species are not always optimally adapted to a rapidly changing environment. Nor is it teleological to regard the modern human foot as more optimally adapted to bipedal walking (e.g. in mechanical or energetic terms) than that of our ancestors who first took up bipedality.

Forward view

This brief review illustrates the potential of the fossil record, with its chronological perspective over different time-scales, and available behavioural proxies as well as hard-tissue morphology, to illustrate the role of behaviour in phenotypic evolution.

Together with controlled present-day experiments such as those of Losos et al. (2004), palaeontology can lead us, beyond mere plausibility arguments, to concrete examples. Accordingly, we need to treat trace fossils as evidence of the behavioural factor in evolution rather than merely as proxies for morphology. This will allow us to progress beyond the common palaeontological assumption of a lock-step between morphology and behaviour and, instead, explore the chronological relationship between them and test hypotheses of process. The length of time that morphology takes to evolve, following a behavioural innovation, is one important question, both in terms of survival capacity during the ‘lag’ period, and the rate of construction of new adaptations. Another is the likely life-history consequences of a period of ‘sub-optimal’ adaptation; for example, switching to abrasive food with low-crowned teeth is likely to reduce life-span and select for rapid reproduction and, perhaps, reduced body size.

The behavioural model can also lead to specific, testable predictions in particular cases. For example, occasional bipedality in chimpanzees, as a model for the first stages in human upright walking, would predict bipedal trackways made by a hominin ancestor with still-unmodified ape-like postcranial morphology.

As the above discussion makes clear, only in some cases are fossil data sufficient to allow the behavioural role in evolution to be teased from its morphological consequences. By the same token, however, we can move proactively to hypothesis-testing by seeking and selecting case studies where data fulfil the necessary requirements.


I am most grateful to Jenny Clack, Paul Barrett, Isabelle de Groote and Robin Crompton who kindly read parts of the manuscript and provided invaluable suggestions. I am also very grateful to Christine Janis, John Odling-Smee and Mark Purnell for their insightful comments that significantly improved the final paper. Finally, I thank Richard Hulbert and Bruce MacFadden for help with acquiring published images.