The Evolution of Activity Capacity

The animal kingdom consists of organisms of very diverse activity capacities, maximal performance levels of which an animal is capable. Some animals are completely sessile, some are slow as snails, some are fleet but tire quickly, and some possess seemingly endless stamina. Physiologists, behaviourists, ecologists and natural historians have been interested in quantifying these activity capacities, because animal athletic abilities are of intrinsic interest and have functional and ecological implications. Knowing activity capacities may, for instance, permit a biologist to identify those structures and functions that impose intrinsic limitations on performance, to define quantitatively the constraints within which any behaviour must occur (e.g. locomotor performance space, Bennett, 1989) or to predict the outcome of an encounter such as that between predator and prey. Most commonly examined are capacities for three different types of activity: burst speed, maximal exertion and endurance. Burst speed is the greatest velocity an animal attains over a short distance; maximal exertion is the work output during activity to rapid exhaustion, usually measured as distance travelled; and endurance is measured as the greatest speed an animal can sustain or the length of time an activity can be sustained. Different morphological, physiological and biochemical processes are thought to underlie and limit each of these different types of behaviour (see Bennett, 1989). Research has concentrated on defining these limits; only now have investigations of the intermittent use of these capacities begun (e.g. R. B. Weinstein and R. J. Full, in preparation). Capacities are generally measured as absolute, instead of size-relative, speeds or distances because interactions among organisms in natural conditions depend on absolute and not relative performance.

How did the great diversity of activity types come about in different animals and how is it maintained? There has recently been considerable interest in and speculation on the evolution of activity capacity (e.g. Bennett and Ruben, 1979; Arnold, 1983; Bakker, 1983; Huey and Bennett, 1986; Pough, 1989; Bennett and Huey, 1991), and it is currently one of the most fertile areas of interactive research between the disciplines of evolutionary biology and comparative physiology/ physiological ecology. It is a logical bridge between these fields because scenarios of natural selection in which activity capacity is crucial for survival can be relatively easily envisioned. These include, for example, predator–prey, reproductive or territorial interactions in which minor differences in speed or exertion may mean success or failure, perhaps with fatal consequences. Such direct and decisive selection on other physiological variables, such as rates of maintenance metabolism or sodium excretion, is less easy to envision. Interest in the evolution of activity capacity has ranged from investigations on its variability and importance in extant natural populations (microevolutionary studies) to its derivation in different major taxa, such as classes or phyla (macroevolutionary studies). As such, it touches on many fundamental issues in organismal biology and ecology. This essay examines some data and major issues that have been studied in regard to the evolution of activity capacities.

While scenarios of selection involving physical capacity are easy to imagine, are they in fact real? Do minor differences in activity ability translate into differential survivorship, reproduction and fitness? Or is adequacy enough? Are encounters in which performance capacity is important so infrequent or balanced by so many other factors that selection operates only against individuals with greatly impaired capacities, e.g. birds that cannot fly or fish that cannot swim? Do time and chance, as suggested in Ecclesiastes, outweigh athletic ability? Only empirical studies on natural populations can answer these questions, and thereby inform speculation in both physiological ecology and evolutionary biology on the natural significance of activity capacity.

Two conditions must obtain for a trait to be potentially susceptible to natural selection (Endler, 1986): the trait must be heritable, that is, have a genetic basis so that offspring resemble parents in regard to the trait, and it must be variable within the population. Laboratory and pedigree studies have demonstrated that activity capacities have both these attributes. Performance capacities such as burst speed and endurance have been found to possess significant levels of heritability [broadsense h² (see Falconer, 1989)=0.2–0.7] in mammals (Ryan, 1975; Langlois, 1980; Bouchard and Malina, 1983a,b; Tolley et al. 1983; Gaffney and Cunningham, 1988), reptiles (van Berkum and Tsuji, 1987; Garland, 1988; Tsuji et al. 1989; Jayne and Bennett, 1990a,b; Garland et al. 1990a) and insects (Caldwell and Hegmann, 1969; Curtsinger and Laurie-Ahlberg, 1981). Considerable intrapopulation variability in activity capacity also exists, even at birth in precocial organisms. For example, coefficients of variation are 16, 54 and 66%, respectively, for burst speed, maximal exertion and endurance in newborn snakes (Thamnophis sirtalis, Jayne and Bennett, 1990b). Even among adult individuals, variability remains similarly great (e.g. Garland, 1984). Individual differences in these capacities persist through time, even after 1 year under field conditions (Huey and Dunham, 1987; van Berkum et al. 1989; Huey et al. 1990; Jayne and Bennett, 1990a), and may possess significant physiological or morphological correlates (e.g. Garland, 1984; Miles, 1987; Gleeson and Harrison, 1988; Kolok, 1990). Thus, activity capacities are apparently governed by genetic factors, persist through time and are highly variable among individuals, factors necessary for the potential for selection on the characters.

How then is the presence or absence of detectable natural selection on these characters assayed? The following procedure has been developed in conjunction with R. B. Huey (University of Washington) and S. J. Arnold (University of Chicago) (see Bennett and Huey, 1991, for a more comprehensive discussion). A group of gravid females is collected from a natural population and permitted to give birth or lay eggs, which later hatch in the laboratory. Activity capacities (i.e. burst speed, maximal exertion, endurance) of the neonates or hatchlings are measured directly, along with morphological and size characters of interest. The young are then released into their parental population. This population is subsequently resampled after long periods (e.g. 1 year) to determine which of the young survived. The survivors may be remeasured and released again, and other animals in the population may also be assayed. Very large sample sizes (several hundred animals) are required for these experiments. Selection is analyzed according to a variety of techniques (e.g. Lande and Arnold, 1983; Schluter, 1988; Jayne and Bennett, 1990b) that ask whether the survivors are a random subsample of the original group of animals released. If so, selection has not been detected; if they are not a random subsample, selection may be inferred to favour an extreme (directional selection) or the mean value (stabilizing selection) of the trait. As such factors as body size may induce false correlations and patterns within the data (see Bennett, 1987a), these are generally removed statistically before selection is analyzed.

An example of such a study is our work on a natural population of garter snakes (Thamnophis sirtalis) in northern California (Jayne and Bennett, 1990a,b). Reptiles are favoured model organisms for such studies because of high local population densities, low mobility and extreme precociality (Huey et al. 1983; Seigel et al. 1987). We measured burst speed, maximal exertion (as total distance crawled under pursuit) and endurance on 275 neonatal snakes born in our laboratory, as well as 382 field-caught snakes (including 86 neonates), and released all of them into their local population. The persistence of these individuals within the population was determined for periods as long as 3 years. When recaptured, an animal had its activity capacities measured again prior to release. As activity capacity in snakes depends greatly on body size (Heckrotte, 1967; Pough, 1978; Jayne and Bennett, 1990a), we analyzed size-corrected residuals. Activity capacities were significantly correlated with year-to-year survival in some age classes of this population (Table 1). During the first year of life, body size (length, not mass) was a significant correlate of survival (P=0.022), but none of the activity capacity traits were significantly involved. However, for yearling animals, burst speed [both size-corrected and absolute (P<0.01)] and maximal exertion were positively and significantly associated with survivorship, and endurance was nearly so. In animals 2 years of age and older, burst speed, both in absolute and in size-corrected terms, was also significantly correlated with survivorship (illustrated in Fig. 1). The intensity of selection on these activity capacities is similar to that previously determined for morphological characters under selection (Schluter, 1988). Since snakes in this population do not reproduce until they are 3–4 years old, some activity capacities are associated with differential survivorship prior to reproduction and may therefore have an impact on fitness.

Similar studies have been undertaken on natural populations of other groups of vertebrates, with mixed results. Unpublished work (quoted in Bennett and Huey, 1991) by R. B. Huey, T. Garland, J. S. Tsuji and F. H. van Berkum on fence lizards (Sceloporus occidentalis and by A. E. Dunham, R. B. Huey, K. L. Overall and R. A. Newman on canyon lizards (Sceloporus merriami) has found no evidence for selection on speed or endurance in either species. In contrast, Miles (1989) has reported a significant positive association between size-corrected burst speed and survivorship in the tree lizard (Urosaurus ornatus). In two different studies involving staged behavioural encounters, individuals with greater burst speed had greater success. Coho salmon (Oncorhynchus kisutch) with greater speed were more successful in avoiding predation by other fish (Taylor and McPhail, 1985) (these individuals were also larger and this result may be complicated by size effects). Faster fence lizards were found to be socially dominant in behavioural interactions with size-matched conspecifics (Garland et al. 1990b). Comparative studies on taxa from different natural thermal regimes show shifts in the thermal dependence of activity capacities consistent with the notion of adaptive evolution in response to selection on those capacities (endurance in populations of killifish, Fundulus heteroclitus, DiMichele and Powers, 1982; burst speed in species of scincid lizards, Huey and Bennett, 1987). Several studies, therefore, have found associations between activity capacities and survivorship, important behaviours or temperature adaptation. These results point to the significance of these capacities in evolution. However, the effects are clearly not universal, and selection for these factors may be only episodic, too difficult to detect or simply absent in some natural populations.

Identifying the actual agents of selection on activity capacity through focal animal observations has been suggested as an alternative approach to microevolutionary studies (Pough, 1989). This, however, to my mind is a secondary question, to be approached only after the likelihood of selection has been established with the type of experiment outlined above. If selection effects are undetectable in those observations, it is unlikely to be profitable to search for agents. Identification of such agents is highly problematic, and a series of studies on different species of anuran amphibians has failed to produce correlations between activity performance and characters thought to be related to fitness (Wells and Taigen, 1984; Sullivan and Walsberg, 1985; Walton, 1988a). Studies on small numbers of individuals may, however, fail to detect selection even if it is occurring, because of low statistical power and the stochastic nature of the selection event, or may miss the event if it occurs at a different age from that studied. Ideal studies of this type would require continuous observation of very large numbers of animals in the field throughout their entire lifetime, or at least until they are postreproductive. Such studies are generally impossible.

Considerable interest has been directed towards the evolution of activity capacity and its metabolic and physiological support among and within different animal taxa. As is not uncommon in evolutionary studies, such macroevolutionary considerations do not flow seamlessly from microevolutionary observations and speculation, so these areas of thought regarding activity capacity have largely developed independently of each other. Macroevolutionary studies have been directed primarily towards the analysis of qualitative differences in metabolic pathways among different animal phyla and quantitative differences in metabolic capacities within phyla. These differences determine rates of ATP generation and hence ability to support the skeletal muscle contraction on which activity depends.

Most animals use aerobically based metabolism to fuel both maintenance and how to moderate levels of activity. As activity level (e.g. speed) increases, oxygen consumption rises to meet increased demand for ATP production. The quantitative pattern of this increase depends primarily on locomotor mode (e.g. walking–running, swimming) (Schmidt-Nielsen, 1972; Tucker, 1975) and not on taxon. In running animals, for instance, the patterns of speed dependence of oxygen consumption and size-adjusted net cost of transport are very similar in insects (Full et al. 1990), crustaceans (Herreid, 1981), reptiles (Gleeson, 1979; Walton et al. 1990), mammals and birds (Taylor et al. 1982). Maximal oxygen consumption sets an upper limit on this kind of sustainable behaviour. Endurance and maximal oxygen consumption are well correlated both within and among taxa according to differences in body size, athletic ability and body temperature (Brett, 1972; Taylor, 1982; Bennett and John-Alder, 1984; Garland, 1984; Taylor et al. 1987; Full et al. 1988). The evolution of endurance capacity is therefore intimately linked with, and depends on, evolutionary changes in maximal capacities to consume oxygen.

Activity requiring power output in excess of that supported by oxygen consumption is fuelled by anaerobic metabolism. Part of this may be provided by endogenous stores of ATP or muscle phosphagens (creatine or arginine phosphate), but these are present in small quantities in comparison to metabolic demand. Additional ATP generation is provided by catabolic anaerobic pathways. A considerable diversity of such pathways is known in the animal kingdom (Hochachka, 1980; Hochachka and Somero, 1984; Gäde and Grieshaber, 1986), with a correspondingly diverse array of end products, energetic efficiencies and rates of energy production. In a highly interesting article, Gnaiger (1983) has argued on thermodynamic grounds (challenged by Watt, 1986) that rate of power output and efficiency of energy transformation cannot be optimized simultaneously within any one of these pathways. Hence, some anaerobic pathways, those terminating in lactic acid, alanine or octopine, emphasize high rates of ATP formation and relatively low ATP yield per unit of substrate catabolized. Others, those terminating in succinate or propionate, have a greater energetic efficiency of substrate use but a low capacity for power output. The former class of pathways is therefore used in fuelling short-term bursts of activity (when ATP demand exceeds aerobic ATP supply) and can be useful in enhancing activity capacity. The latter pathways are used for enduring environmental hypoxia and permit long-term survival in anoxia but do not contribute to activity enhancement. Lactic acid production is used as the principal anaerobic pathway during activity in vertebrates (Bennett, 1978), urochordates and echinoderms (Ruben and Parrish, 1991), crustaceans (McMahon, 1981), arachnids (Prestwich. 1988). leeches (Zebe et al. 1981) and some insects (Harrison et al. 1989; activity metabolism in flying insects is not anaerobically supplemented). Octopine formation occurs during intense activity in molluscs, both cephalopods (Grieshaber and Gäde, 1976) and bivalves (Gäde, 1981), and sipunculids (Grieshaber and Zebe, 1978). Accumulation of alanopine during activity has been demonstrated in some annelids siegmund et al. 1985). The evolutionary antiquity of the formation of these compounds during activity is still unknown, both because of uncertainties in the phylogeny of the invertebrate groups and because of lack of information on end product accumulation in many of them. (The literature on anaerobic product formation in invertebrates is reviewed in Gade and Grieshaber, 1986.) The evolution of the capacity for bouts of intense activity therefore generally depends on the development of capacities for rapid formation and tolerance of high concentrations of lactic acid or octopine. Production of the former is associated with rapid exhaustion, the exact physiological causation of which is still in dispute.

The evolution of activity capacities and their metabolic bases in the vertebrates has engendered considerable interest and speculation. The scenarios proposed will form the basis for the ensuing discussion of the macroevolutionary development of activity capacity within a major taxon. This concentration of interest is not simply a reflection of vertebrate self-absorption. It is due primarily to an excellently documented palaeontological history, a great diversity of types and mechanisms of locomotion exhibited by members of the group, and a very large behavioural and physiological data base.

It is believed that the original vertebrates, the ostracoderms, were active, predatory carnivores (Jollie, 1973; Northcutt and Gans, 1983). They probably possessed metabolic patterns, and hence activity capacities, not very different from those of most modern fish (Ruben and Bennett, 1980; Ruben and Parrish, 1991). These would have included aerobically supported maintenance metabolism and a modest capacity for aerobically sustained swimming. Bouts of intense activity would have been fuelled by anaerobic glycolysis, resulting in lactic acid formation. These assertions come from phylogenetically based studies on the comparative activity physiology of contemporary animals. Lactic acid formation has been shown to occur during activity in teleosts (see references in Bennett, 1978), elasmobranchs (Wells and Davie, 1985), agnathans and a cephalochordate (Ruben and Bennett, 1980), and a free-swimming urochordate and an echinoderm (Ruben and Parrish, 1991). Given this pattern of trait distribution in outgroup taxa and extant vertebrates (Fig. 2), the most parsimonious assumption is its presence in the original ancestral vertebrates (Ruben and Bennett, 1980; Ruben and Parrish, 1991). Likewise, maintenance levels of oxygen consumption, adjusted for body size and temperature, are similar in extant ectothermic vertebrates and multicellular invertebrates (reviewed by Hemmingsen, 1960) and were, therefore, probably similar in vertebrate ancestors. Maximal rates of oxygen consumption for most of the taxa relevant to such an analysis (i.e. those in Fig. 2) are unknown and their study would be a valuable contribution in this regard. At this point, the assumption of moderate aerobic scopes for activity by ostracoderms does not seem unwarranted. In its qualitative aspects, therefore, the vertebrate metabolic plan appears to be an inheritance from deuterostome ancestors and not a novel invention of the group.

Aerobic scopes of modern fish are relatively low in comparison with those of endothermic vertebrates or flying insects (Beamish, 1978; Bennett, 1978; Taylor et al. 1981; Bartholomew, 1981). However, fish have a considerable range of sustainable speeds supported by these aerobic capacities and, consequently, relatively great endurance. This endurance capacity is due to the low cost of swimming as a locomotor mode (Fig. 3): net costs of transport (mlO₂ g⁻¹ mass km⁻¹) of swimmers are only about one-tenth those of walkers and runners (Schmidt-Nielsen, 1972; Tucker, 1975). Hence, even limited aerobic capacity can support a wide range of speeds, in comparison with a similar condition in terrestrial animals (Fig. 3). Anaerobic metabolism is reserved for very intense bursts of activity, avoiding temporal delays in the augmentation of oxygen consumption and permitting attainment of speeds beyond those supportable aerobically. Burst speeds in fish exceed sustainable speeds by a factor of only two or three (Table 2; see references in Beamish, 1978). In fish, therefore, the pattern of metabolic support coupled with locomotor cost is conducive to aerobically supported activity over a wide range of speeds and behaviours.

The transition of vertebrates from an aquatic to a terrestrial habitat entailed major changes in many structural and functional systems (Randall et al. 1981; Carroll, 1988). Oxygen in the aerial environment is more easily accessible because of its increased diffusivity and concentration (Dejours, 1981). But in spite of this increased accessibility, aerobic capacities were not expanded: resting and maximal rates of oxygen consumption in reptiles and amphibians are very similar to those of fish of equal size and body temperature (Hemmingsen, 1960; Brett, 1972; Bennett, 1978, 1982; Withers and Hillman, 1979). Paradoxically, in spite of increased environmental oxygen availability, endurance capacities, measured as sustainable speeds, decreased in comparison with values measured in fish. Sustainable walking speeds of salamanders are less than 3 m min⁻¹ (Full, 1986; Feder, 1986; Else and Bennett, 1987; Full et al. 1988), less than one-tenth those of fish swimming (Table 2). Sustainable speeds of reptiles are greater (Table 2), but still far below those of fish (Fig. 3). This decrement in endurance is the result of greatly increased locomotor costs in the terrestrial environment. Net costs of transport are nearly 10 times greater in terrestrial walkers and runners than in swimmers of equal body size (Schmidt-Nielsen, 1972; Tucker, 1975). The biomechanics underlying this difference are still not fully understood. It may relate to the costs of sequential acceleration and deceleration of limbs and vertical displacement of centre of mass (Alexander, 1977), but the equality of locomotor costs of snakes and limbed reptiles (Walton et al. 1990) suggests that such explanations are less than complete, as snake locomotion has neither of these features. Factors involving the rate of cycling of skeletal muscle contraction may be of greater importance (Full et al. 1990; Kram and Taylor, 1990). In any event, greatly increased locomotor costs coupled with a lack of increment in aerobic capacity meant that limits of endurance were reached at much lower speeds and levels of exertion than in their aquatic ancestors (Fig. 3; Bennett, 1985).

How was capacity for burst speed affected by the transition to land? Data on burst speed capacities of amphibians are scant: tiger salamanders (Ambystoma tigrinum) attain peak running speeds of only about 10m min⁻¹ (Else and Bennett, 1987; Shaffer et al. 1991), slow in comparison to fish (Table 2). Burst locomotor performance of anuran amphibians is generally measured as jump distance (e.g. Zug, 1978), which is not directly comparable to other measurements reported here. Lizards are capable of running at 10–30 times sustainable speeds for short bursts (Table 2; Bennett, 1982; Garland, 1982, personal communication), speeds that are substantially greater than those of fish. These bursts of activity are fuelled by ATP or creatine phosphate catabolism and anaerobic glycolysis and lactic acid formation, the capacity for which is very great in lizards (Bennett and Licht, 1972) and some amphibians (Bennett and Licht, 1974). In comparison to fish, a much greater portion of the total behavioural repertoire of these amphibians and reptiles is thus accessed only by anaerobic metabolism (Table 2). Activity capacities of these terrestrial tetrapods are consequently short on endurance, but their maximal exertion and burst speed can be very great. Their behaviour is typically limited to rather slow movements, punctuated with periods of very intense bouts of activity of short duration.

This pattern of activity capacities of the terrestrial ectothermic vertebrates was fundamentally altered during the evolution of the mammals and birds. Endurance increased substantially in these lineages. Maximal sustainable speeds of mammals are 6-7 times greater than those of lizards of equal size and body temperature (Table 2, Fig. 3; Bennett, 1982; T. Garland, personal communication). Sustained jumping flight characteristic of birds is beyond the aerobic capabilities of modern reptiles (Bennett and Ruben, 1979). Burst speed, however, has apparently has been greatly modified: maximal speeds of lizards and mammals of equal size are not significantly different (Table 2; Garland, 1982, personal communication).

What is responsible for these differences in activity capacity? Most approaches to this question have restricted themselves to comparisons of mammals and reptiles, usually lizards. Such comparisons are made because of their similarity of body form and relative ease of measurement (in comparison to birds) of functional capacities during activity. These approaches implicitly assume that metabolic and performance traits that are widely shared among extant reptilian taxa are representative of the ancestral condition from which mammals evolved. Even though endothermy was evolved completely independently in the two lines, mammalian and avian systems appear highly convergent in many of their metabolic aspects (compare allometric equations for many such variables in Calder, 1984). Mammals may, therefore, potentially serve as adequate models in such comparisons for both endothermic taxa. In mammals and birds, both basal and maximal levels of oxygen consumption and aerobic scopes are approximately 6–10 times greater than those of extant reptiles of the same body mass and temperature (Bartholomew and Tucker, 1963; Brett, 1972; Bennett, 1978, 1982; Taylor et al. 1981). Net costs of walking and running are equal in mammals and reptiles (Bakker, 1972; Gleeson, 1979; Bennett and John-Alder, 1984), so the greater aerobic scope of the former affords a much greater range of sustainable speeds that can be maintained without supplementary anaerobic metabolism (Table 2, Fig. 3). Capacities for anaerobic metabolism, both anaerobic scope and anaerobic capacity (Bennett and Licht, 1972), are rather similar in mammals and reptiles (Bennett, 1978, 1982; Ruben and Battalia, 1979), as are activities of glycolytic enzymes (Bennett, 1972a) and myofibrillar ATPase (Gleeson et al. 1980b) in saurian and mammalian muscle. Hence, the lack of divergence in burst speed and maximal exertion capacities of the two groups is not particularly surprising.

What structural and functional modifications underlie the evolution of greater aerobic capacity in birds and mammals? Virtually all aspects of oxygen-transporting systems have been increased, quantitatively and sometimes qualitatively. Pulmonary surface area and diffusing capacity have been greatly increased in the evolution of the alveolar and parabronchial lungs (Perry, 1983), as has the ventilatory apparatus itself, to permit high levels of pulmonary ventilation during activity (Ruben et al. 1987; Carrier, 1987). Oxygen-carrying capacity and oxygenaffinity of the blood have been greatly increased (Bennett, 1973; Pough, 1979). Complete structural separation of the cardiac ventricles has occurred and maximal cardiac output has increased greatly (Gleeson et al. 1980a). Activities of mitochondrial enzymes have increased substantially (Bennett, 1972a; Else and Hulbert, 1981). Else and Hulbert (1981) undertook a quantitative examination of the differences in aerobic capacities of the ‘reptilian and mammalian machines’. They concluded that the increments in aerobic capacity in the latter were due to a number of different factors, including relatively larger internal organs and later mitochondrial volume and surface area. Taylor and Weibel (1981) have argued that all portions of the oxygen transporting and utilizing apparatus should have similar maximal flux capacities, that is that there should be no particular limiting element in the chain and that no part should possess excess transport capacity. If such parity were maintained, the augmentation of all the elements of oxygen transport would have had to have evolved in concert and simultaneously. Given their diverse nature and probable diverse genetic control, this synchronization seems unlikely. Withers and Hillman (1988) have inferred a cardiovascular limitation on maximal oxygen consumption in anuran amphibians; a similar restriction in this or another element of the system may possibly have occurred in primitive reptiles. Probably the evolution of increased aerobic capacities in birds and mammals involved the sequential improvement in whatever aspect of the system formed its most limiting element at any particular time.

Where in the mammalian and avian lineages did high metabolic rates and endothermy appear? Are they inventions within the classes Mammalia and Aves or did they develop in groups ancestral to these? Since most characters relating to oxygen consumption and thermoregulation do not fossilize, there has been much heat but little light generated from very indirect evidence about the metabolic condition of mammal-like reptiles and dinosaurs. Bennett and Ruben (1986) proposed the following criteria for evaluating a character presumably associated with endothermy in extinct organisms: it must not be present in any extant ectotherm and it must have a clear and direct association with high metabolic rate. Further, the presence of such characters in sister taxa suggests ancestral inheritance. Reviewing the evidence, Bennett and Ruben (1986) concluded that the therapsid ancestors of mammals probably possessed high aerobic capacities and may have been well on the path to endothermy prior to the evolution of mammals per se. This conclusion was based on two lines of evidence. First, cynodont therapsids possessed elaborate turbinal bones in the nasal cavity, indicative of the conditioning of extensive volumes of ventilated air. These are not present in any ectothermic vertebrate and are directly associated with capacities for high levels of oxygen uptake. Second, all physiological and morphological features related to high levels of oxygen consumption and endothermy are shared by the mammalian sister taxa Prototheria (monotremes) and Theria (marsupials and placentals). These groups diverged fairly early in mammalian evolution and remain distinct in many features of their reproduction, endocrinology and neurology. The possession of so many diverse systems related to high levels of oxygen consumption and endothermy (e.g. diaphragm, four-chambered heart, hair) is due either to unprecedented levels of convergence or ancestral inheritance.

The presence or absence of endothermy in dinosaurs has been a topic of considerable discussion and sometimes acrimonious debate over the past two decades (see references in Thomas and Olson, 1980; Czerkas, 1986; Bakker, 1986). Both proponents and opponents have elaborate arguments and counterarguments too long and involved to be repeated here. I point out, however, that the of the characters proposed in support of dinosaur endothermy meet the aforementioned criteria of absence in all extant ectotherms and a clear and direct association with high metabolic rates. Further, as the birds are a monophyletic group, no extant endothermic sister taxon exists from which ancestral inheritance of endothermy can be inferred. I therefore personally consider the contention of dinosaur endothermy unproved.

What occasioned the evolution of increased aerobic capacity in the mammalian and avian lineages? Given the difficulty in determining selective factors on morphology, physiology and performance in extant populations (see microevolutionary section above), it is probably foolhardy to attempt it for unknown organisms now extinct that lived in poorly understood environments. Two factors, however, have been suggested as major selective advantages. The first is metabolic thermoregulation. Body temperatures of birds and mammals are both higher and more constant than those of most ectothermic vertebrates. High body temperatures are suggested to have been favoured to maximize catalytic capacity of thermally dependent rate processes (Heinrich, 1977; Hochachka and Somero, 1984; Bennett, 1987b). Constancy of temperature presumably also favours stability of those processes and homeostasis in general, although these purported advantages are more nebulous and the logic behind them often circular. The evolution of endothermic homeothermy in these lineages required the elevation of maintenance rates of aerobic metabolism. Placing an insulatory layer on a vertebrate ectotherm does not confer significant homeothermy, as Cowles (1958) demonstrated directly by outfitting lizards with mink coats. Increased insulation, above that afforded by the reptilian integument, is necessary for homeothermy, at least in small endotherms, but is in itself not sufficient to produce it. Increments in maintenance metabolic rate occurred independently in both avian and mammalian lineages, converging on very similar levels in the two groups, when differences in body temperature are taken into account (equations in Peters, 1983; Calder, 1984). It is important to note that this metabolic increment is fundamentally different from that in other ‘endothermic’ animals, which elevate heat production by skeletal muscle contraction (e.g. insects, Kammer, 1981; tunas, Stevens and Neill, 1978; brooding pythons, Hutchison et al. 1966) or modified skeletal muscle tissue (e.g. billfish heater muscle, Block, 1987). The high rates of maintenance metabolism of birds and mammals are not myogenically based; they are generated primarily by visceral organs and the central nervous system. In humans, nearly 70% of resting metabolic rate is accounted for by the metabolism of the heart, kidneys, brain, liver and intestines, even though they constitute only 7% of body mass (Aschoff et al. 1971). Skeletal muscle tissue can produce high levels of heat in the cold or during activity, but is largely uninvolved in the production of the high basal metabolic rate of birds and mammals.

A difficulty in explaining the evolution of increased aerobic capacity on thermoregulatory grounds alone is its energetic cost. Maintenance metabolic rates must be increased very substantially above reptilian levels before significant homeothermy and high body temperatures can be achieved. Basal metabolic rates of birds and mammals exceed those of extant reptiles of equal size and body temperature by a factor of 6–10, and this differential is substantially greater at lower ambient temperatures (Dawson and Bartholomew, 1956; Bennett and Dawson, 1976). Minor increments (e.g. a doubling) in maintenance metabolic rates would not have been effective in establishing homeothermy or raising body temperature, but would have added substantially to energetic demands of fieldactive animals with little apparent benefit. The problem for the thermoregulatory explanation of the evolution of high metabolic rate lies in justifying intermediate stages in selective terms, presumptive adaptive valleys between adaptive peaks. Even if the final goal of endothermic homeothermy is worthwhile in terms of fitness, the condition cannot evolve unless all intermediate stages also confer sequentially greater fitness benefits. In a character as complex as endothermic temperature regulation, involving so many physiological systems, it is highly doubtful that it could have developed in its entirety without many intermediate stages. The problem becomes even more complex if ancestral animals had the capacity for behavioural thermoregulation of high body temperatures, as do many extant reptiles. In this case the benefits of high and relatively constant temperature would already have been available without increased metabolic cost. One argument has been advanced (Crompton et al. 1978; Taylor, 1980) that endothermy evolved in the mammalian line in response to nocturnality, when behavioural thermoregulation at high temperature is precluded. This argument loses force, however, if endothermy was already present in the reptilian ancestors of these nocturnal mammals (see above, Bennett and Ruben, 1986).

A second explanation suggested for the evolution of increased metabolic rates in the mammalian and avian lineages is the increased capacity for aerobically supported activity (Regal, 1978; Bennett and Ruben, 1979). Increased levels of maximal oxygen consumption directly expand the range of levels of sustainable exertion and the behaviours that depend on them (e.g. Fig. 3). Animals with greater aerobic limits have greater endurance in a variety of different activities that may influence fitness, such as prey pursuit, predator avoidance, territory maintenance and courtship. It is relatively easy to visualize a direct benefit in selective terms to such increased endurance capacity, particularly in animals, such as reptiles, that have limited aerobic capabilities and must otherwise depend heavily on anaerobic metabolism, with its ensuing exhaustion, to support vigorous activity. Ancestral animals in both the mammalian and avian lineages seem to have been progressively more capable of vigorous activity, as shown by general skeletal rearrangements, such as a change in limb suspension and reduction of skeletal mass (Romer, 1966; Olson, 1971; Kemp, 1982; Carroll, 1988). The expansion of aerobic capacities to support endurance-related behaviours is, thus, plausible in its progressive accrual of benefits during its evolution (i.e. no apparent adaptive valleys) and is also in accord with general palaeobiological patterns of the groups in question.

The difficulty for the ‘aerobic capacity’ (Taigen, 1983) theory of the evolution of endothermy is explaining the concomitant rise in maintenance metabolic rate with that of maximal oxygen consumption (Bennett and Ruben, 1979). If increased aerobic capacity during activity were the only consideration, it seems that a more satisfactory solution would have been to expand maximal aerobic rates without increasing maintenance rates. Thereby, the animals could benefit from both increased endurance and low maintenance costs, the latter being characteristic of ectothermy (Bennett and Dawson, 1976; Pough, 1983). A general linkage between maximal and maintenance aerobic metabolic rates was proposed to explain increments in the latter as the former is increased (Bennett and Ruben, 1979). In all groups of vertebrates, even within individual poikilotherms at different body temperatures, maximal oxygen consumption exceeds maintenance by an average factor of approximately 10. The physiological basis of this linkage is unclear, especially as maximal rates are largely myogenically based and maintenance rates are not. This factorial difference is not fixed or absolute, and comparisons among disparate species of similar body size can show differences in maximal rates without similar differences in maintenance rates (e.g. Bennett, 1972b, 1978; Ruben, 1976; Taylor et al. 1987). In phylogenetically controlled comparisons, which are inherently more reliable (Felsenstein, 1985; Huey, 1987), significant positive correlations between these factors occur among groups of anuran amphibians (Taigen, 1983; Walton, 1988b). Such correlations have not been found within species (Pough and Andrews, 1984; Garland and Else, 1987; Garland et al. 1987). Thus, this association remains a generality, without a firm mechanistic explanation or intraspecific correlational basis. Two factors, however, bear further investigation. First, all previous studies refer to phenotypic correlations among metabolic rates. Genetic correlations may have very different patterns and link traits evolutionarily without obvious phenotypic associations (Arnold, 1987). Examinations of the genetic relationships between maximal and maintenance levels of oxygen consumption would prove helpful in understanding their association. Second, the relationship between intraspecific and interspecific correlation and its relationship to the analysis of evolutionary pattern is just beginning to be analyzed (Emerson and Arnold, 1989). Further theoretical and experimental studies on these general associations would also be valuable.

In summary, both explanations for the evolution of high metabolic rates associated with endothermy have positive features and difficulties. Perhaps a combination of thermoregulatory and activity capacity factors were involved simultaneously or sequentially. The early stages of its evolution may have been dominated by selection for increased endurance, and its later stages by selection for homeothermy. Unifactorial explanations for the evolution of any complex character are almost certainly incomplete.

Understanding the evolution of such complex characters as activity capacities, in both their mechanistic bases and adaptive significance, is a fundamental goal do organismal biology. In the studies cited above, an excellent beginning has been made, both conceptually and experimentally, in addressing the issues. In my opinion, we would benefit from further studies on the operation of selection on these characters in natural populations of diverse taxa, followed by studies attempting to quantify locomotor behaviour under field conditions and to identify selective agents. The bridge between microevolutionary studies on extant populations and diversification of capacities among higher taxa is a difficult one to build, but one well worth the attempt. Comparative studies on different populations within a species or among congeneric species with diverse activity capacities may be useful in uncovering the physiological and morphological bases, and perhaps the ecological determinants, of activity capacity. Experimental manipulations, where feasible, of such factors as food or mate availability or predation may be particularly useful in examining their influence as agents of selection. Choice of appropriate experimental groups will obviously be crucial to the success of these studies. Analysis of ancient historical pattern, for instance the selective causation and mechanism of the evolution of endothermy, will probably always remain speculative. However, the use of phylogenetically based analyses and the future availability of new analytical techniques may help to resolve some of the major questions in this area.

I thank R. J. Full, R. B. Huey and J. A. Ruben for critically reading and commenting on the manuscript and S. M. Reilly for help preparing the figures. Supported by NSF Grants DCB88-20128 and BSR90-18054.