Scaling of body temperature in mammals and birds
Summary
- 1
We examine variation associated with phylogeny in the scaling of body temperature in endotherms, using data from 596 species of mammal and 490 species of bird.
- 2
Among higher groups of mammals there is statistically significant scaling of body temperature with mass in Marsupialia (positive), Ferae and Ungulata (both negative). In mammalian orders where data are available for at least 10 species, scaling is negative in three orders (Carnivora, Erinaceomorpha and Artiodactyla), positive in one (Chiroptera) and not significant in seven others. There is no relationship apparent between the scaling of body temperature and the existence of gut fermentation. As expected, monotremes exhibit the lowest body temperatures, but within marsupials diprotodonts have a mean body temperature higher than several placental groups; the traditional ranking of body temperatures in the sequence monotremes – marsupials – placentals is thus misleading.
- 3
In birds, scaling relationships are significant only for Ciconiiformes (strongly negative) and Passeriformes (weakly positive).
- 4
When allowance is made for phylogenetic effects, there is no significant relationship between temperature in body mass in mammals overall, but an inverse and almost significant relationship in birds.
- 5
This study indicates a complex relationship between body mass, body temperature and metabolic rate in mammals and birds, mediated through ecology.
Introduction
Interest in the body temperature of mammals and birds extends back over two centuries (McNab 1970). The modern debate stems from a trio of papers by Scholander and colleagues (1950a,1950b, 1950c) and a short report by Rodbard (1950). In an analysis based on a very small data set, Rodbard (1950) argued that in both mammals and birds body temperature varied inversely with mass. These conclusions were challenged by Morrison & Ryser (1952), who identified errors in the body mass data by factors of between 3 and 30. McNab (1966), using a larger data set, was able to confirm Rodbard's result for birds, but not for mammals; he also demonstrated for the first time that body temperature varied among different groups of mammals. Since these studies, a considerable body of data has been compiled for body temperature in both mammals and birds, but no comprehensive reanalysis has been undertaken apart from a preliminary examination by White & Seymour (2003), who were concerned principally with metabolic rate rather than temperature.
The recent revival of interest in the scaling of metabolism as a fundamental factor in ecology (Brown et al. 2004) has prompted re-examination of many aspects of energetics. The central tenet of the metabolic theory of ecology is that the resting metabolism of any organisms is dictated primarily by two variables, namely body mass and body temperature (Gillooly et al. 2001). Whilst many studies have tested the assumptions and predictions of metabolic theory in terms of resting metabolism, there have been almost no recent studies of body temperature. An exception is that of White & Seymour (2003) who compiled an extensive data set for mammals and demonstrated a positive relationship between body mass and body temperature, precisely the opposite conclusion to Rodbard (1950); they also confirmed the long-held view that body temperature tended to be lower in marsupials than in placentals. None of the previous studies of body temperature in endotherms have allowed for potential phylogenetic differences in the scaling of body temperature within mammals or birds.
The body temperature of an organism is a balance between the rate at which heat is supplied and the rate at which it is lost (Fig. 1). For an endotherm the two sources of heat are metabolism and the environment. Losses occur through the skin, the rate of which is influenced strongly by external temperature and insulation, and also as a result of evaporation associated with respiration.
A conceptual model showing the factors influencing body temperature in endotherms. Solid arrows show heat flow, and line arrows indicate influence.
Because metabolic rate is a major source of the heat used to maintain body temperature in endotherms, and is influenced by ecology (Lovegrove 2000, 2003, 2004), there is likely to be a strong functional link between ecology and body temperature. However, there is also an important feedback in that the level of resting metabolism in endotherms is itself influenced by body temperature, just as it is in ectotherms (White, Phillips & Seymour 2006). This indicates that the overall relationship between ecology, metabolic rate and body temperature in mammals and birds is complex. The strong covariation between phylogeny and ecology in endotherms (McNab 1970, 2002) suggests that we should expect a strong phylogenetic signal in mammalian and avian body temperature. Whilst several recent studies have demonstrated an ecological or phylogenetic related variation in endotherm basal metabolic rate, there have been no recent detailed studies of body temperature.
Here, we present a phylogenetically based analysis of two large data sets of body temperature, for mammals and birds, which reveals important and hitherto unrecognized features. In undertaking this analysis, we focussed on two key questions:
- 1
Does mean body temperature, or the scaling of body temperature with body mass, vary phylogenetically within mammals and birds?
- 2
When allowance is made for any phylogenetic effects, does body temperature scale with body mass, and does this scaling differ between mammals and birds?
Materials and methods
Data were compiled from a thorough search of bibliographic data bases, secondary references and the world-wide web. The initial compilation was undertaken in 1999; to this were added data from Lovegrove (2000), White & Seymour (2003), and a repeat search of bibliographic data bases in 2005.
As with all such compilations, data quality was highly variable. The principle difficulties encountered were poor or inappropriate temperature measurement and a lack of data on body mass. As with basal metabolic rate, body temperature should ideally be measured in resting, normothermic, post-absorbtive, inactive and conscious individuals. However, it can be difficult to achieve these conditions in mammals where the digestive tract supports significant fermentation, such as artiodactyls, macropods or lagomorphs, or in small active forms such as shrews (White & Seymour 2003).
Few measurements of body temperature in birds or mammals meet all of these criteria, and only the most reliable data were used for further analysis. Where a given species was represented by more than one study, the choice of data point was based upon data quality, highest number of observations and most recent date of measurement. The initial data set for mammals comprised 1237 observations, which was reduced to 596 (one data point per species). For birds the total data set of 896 observations was reduced to 490 species, with many species represented by separate data for males and females. Where body mass was not reported, representative mass data were taken from Silva & Downing (1995) for mammals, and Dunning (1993) or Sibley (2000) for birds.
For taxonomic analysis, species were assigned to families, orders and one higher taxon. An important aspect of this study was thus the selection of the taxonomic framework. In both mammals and birds recent molecular work has resulted in radical revision of long-established taxonomies. Whilst new species are still being discovered or defined, particularly in some groups of birds, taxonomy at the level of family and order is generally fairly stable and most changes have been concerned with establishing phylogenetically consistent groupings at higher taxonomic levels (e.g. Barker et al. 2004; Springer et al. 2005; Bininda-Emonds et al. 2007). Past experience suggests that not all systematic changes suggested by recent molecular work will prove robust to further study; for the analysis reported here we therefore chose to use widely accepted modern taxonomies, but not to update these in relation to the most recent molecular work where proposed changes have yet to achieve consensus.
In the case of birds, we chose the systematic arrangement of Monroe & Sibley (1993), with the higher taxon being parvclass. This classification is based on the original DNA study of Sibley & Ahlquist (1990), but modified significantly in the light of more recent molecular work. The early analyses of Sibley & Ahlquist (1990) suggested a number of radical changes to the traditional avian classification of Wetmore (1930, 1960) and Peters (1951). Not all of these changes have withstood the passage of time, and whilst contentious in some areas, the attempt by Monroe & Sibley (1993) to generate a taxonomic sequence that reflects phylogeny marks a significant advance on the traditional sequence, which Wetmore (1930) himself accepted as ‘necessarily arbitrary’. Acknowledging that avian systematics is in a state of flux, Cracraft, Barker & Cibois (2003) in their introduction to the third edition of the Howard & Moore checklist eschewed providing a classification above family level, specifically to avoid the need for ‘numerous arbitrary choices’. Because our analyses required a higher level taxonomy we have used the most recent attempt (Monroe & Sibley 1993), accepting that this is far from perfect but an important improvement on the traditional arrangement of Wetmore or Peters. Taxonomic queries resulting from data in the older literature were resolved by reference to Gruson (1976) or Dickinson (2003). For mammals, the classification used was based on Duff & Lawson (2004) with higher taxa being subclass (Prototheria), cohort (Marsupialia), magnorder (Xenarthra, Epitheria) or grandorder (Ferae, Lipotyphla, Archonta, Ungulata). Queries concerning older colloquial or scientific names were resolved by reference to Corbet & Hill (1986). Although a somewhat conservative choice, this taxonomy is used widely by mammalian ecologists. Recent molecular work has proposed a radical rearrangement of the mammalian orders within higher groups. The details vary between authors (e.g. in the placement of Eulipotyphla and the precise taxonomic level at which Glires and Archonta should be united), but there is an emerging consensus that Afrotheria, Laurasiatheria and Archonta plus Glires represent major lineages within the placental mammals. These differences change the precise values of summary statistics, but not the qualitative patterns or ecological conclusions. Where data have been summarized at higher taxonomic levels, we have therefore presented data according to both taxonomies.
Overall, we believe that our conclusions from statistical analyses of physiological variables at the level of family or order are thus likely to be reasonably robust for both birds and mammals, whereas the detail of those drawn from analyses at higher taxonomic levels may be affected by subsequent taxonomic revisions. A summary of the data is given in Supplementary Tables S1 and S3.
statistical analysis
Body mass data for both mammal and birds are highly skewed and were therefore log-transformed (natural logs) before analysis; a unit increase on the log scale corresponds to a 2·7-fold increase in mass.
It has long been recognized that much ecological data contains a strong evolutionary signal, and that failure to allow for this phylogenetic non-independence can lead to an increase in type I error, and erroneous results (Felsenstein 1985). For many ecologists, the approach of choice to circumvent this problem is the use of phylogenetically independent contrasts (PIC, Felsenstein 1985) or phylogenetic generalized least-squares regression (PGLS, Grafen 1989). PIC and PGLS do, however, come with their own set of assumptions and potential difficulties (Muñoz-Garcia & Williams 2005). In particular, if the evolution of physiological traits varies significantly across lineages, and differs from the stochastic (Brownian) assumption, then analysis using PIC or PGLS can provide a biased estimate of parameters that is non-independent across clades (Price 1997).
At present, we lack a suitable species-level phylogeny for the 1086 taxa in this study. Even the most recent mammal supertree, which is 98% complete is less than 50% resolved at the species level (Bininda-Emonds et al. 2007). The use of PIC or PGLS is thus not an option for this study, and we have therefore used a multi-level regression model, which combines regression (for the trend) with a nested structure (for the random variation). This approach is similar to, but not identical with, the nested anova approach advocated by Harvey & Pagel (1991) for analyses where the phylogeny is unknown. Instead of using nested anova to determine the choice of taxonomic level, our analysis works with responses at the species level rather than the means at some higher level, but allows for dependencies amongst the observations (e.g. responses on species in the same genera are more similar than responses on species in different genera, etc). The nested structure for the inter-correlations amongst the responses is entirely empirical, and has the advantage of being relatively simple, and practically feasible in the absence of the complete phylogeny required by the more sophisticated methods such as PIC or PGLS.
We used a model incorporating random variation at different phylogenetic levels (between higher groups, between orders within groups, between families within orders, between genera within families and between species within genera), and a linear relationship between temperature and log body mass. If yijklm denotes the temperature for the ith group, jth order, kth family, lth genera and mth species in the nested hierarchy and xijklm denotes the corresponding value for loge body mass, the basic model is.
where a is the intercept and b is the slope in the linear relationship. The terms H, O, F, G and S are random effects for higher group, order, family, genera and species, which are assumed to vary independently and follow a normal distribution with zero mean and variances VH, VO, VF, VG and VS, respectively. In some models the slope b was allowed to vary to test for differences in slopes among groups and orders. For birds, the model was augmented to allow for observations on males and females of the same species by adding an extra random component (between individuals within species), and effects for differences between sexes. Models were fitted using the method of residual maximum likelihood (REML) (Patterson & Thompson 1971), implemented using the statistical package Genstat (Genstat 2005). In general, P-values are presented to indicate the strength of evidence for an effect, but for brevity ‘significance’ means statistical significance (5% level).
Results
Analysis of the complete data sets for birds and mammals but ignoring phylogenetic effects (ordinary least-squares regression of body temperature on log body mass) suggests positive scaling in mammals and an inverse relationship for birds, identical to the results of McNab (1966). However, once phylogeny is allowed for, there is no statistically significant relationship between temperature and body mass in mammals (P = 0·40), whereas in birds an inverse relationship is apparent that is close to statistical significance (P = 0·052; Table 1).
| Taxonomic grouping | Variance (% total) | |
|---|---|---|
| Mammals | Birds | |
| Higher group | 2·50 (40) | 0·100 (7) |
| Order (within higher group) | 1·48 (23) | 0·479 (32) |
| Family (within order) | 0·702 (11) | 0·628 (42) |
| Genus (within family) | 0·388 (6) | 0·065 (4) |
| Species (within genus) | 1·26 (20) | 0·032 (2) |
| Sex (within species) | nd | 0·183 (12) |
| Total variance | 6·34 | 1·49 |
| Slope of trend with loge body mass (SE) | 0·036 (0·043) | –0·093 (0·048) |
| P | 0·40 | 0·052 |
For mammals overall, the estimated slope (the effect size) corresponds to an increase in body temperature of 0·6 K (SE: 0·7) over the range of log body masses from 2 to 18 (7 g to 62 500 kg). For birds overall the effect size corresponds to a decrease in body temperature of 0·9 K (SE: 0·5) over a logarithmic body mass range of 2–12 (7 g to 160 kg). The power for detecting an effect size of 3 SE is about 85%. For mammals, this would mean a slope of +0·13 K per 2·7-fold increase in body mass, or +2·1 K over the full range of body mass in this study. For birds, the corresponding figures would be –0·14 K for a 2·7-fold increase in body mass, and –1·4 K over the full range of body masses.
mammals
Analysis of the components of variation across taxonomic levels indicated that in mammals most of the variance was at the higher taxonomic levels (higher group and order; Table 1) indicating that these are the most appropriate levels to examine for scaling effects. Breakdown of body temperature according to higher group shows clearly that monotremes (Prototheria) exhibit the lowest body temperatures amongst mammals, but that marsupials as a group have higher body temperatures than some eutherians: the median body temperature for Marsupialia was higher than for either Xenarthra or Lipotyphla (Fig. 2a). Using the alternative higher group classification, the median body temperature for marsupials exceeds that for Xenarthra, but not any other group (Fig. 2b). Although the division of mammals into three fundamental lineages of monotremes, marsupials and placentals has a long history, and is supported by recent molecular work, both marsupials and placentals are highly variable groups in terms of ecology, basal metabolism and body temperature, so grouping at higher taxonomic level masks considerable variation which only becomes evident with analysis at lower taxonomic levels such as order.
Analysis of body temperature in mammalian higher groups. Box plots of temperature data for each higher group (subclass, cohort, magnorder or grandorder), where the group contains data for four or more species. (a) Analysis by higher order classification of Duff & Lawson (2004). (b) Analysis by higher order classification of Bininda-Emonds et al. (2007).
Applying a statistical model where the slope of the body temperature/body mass scaling relationship is allowed to vary across higher groups reveals strong evidence for differences among slopes (Wald test, W = 30·83, P < 0·001). Scaling of body temperature with log body mass is significant and positive in marsupials, whereas in Ferae and Ungulata it is significant and negative (Table 2). These scaling relationships are also significant in models that ignore phylogenetic affects. In two higher groups (Epitheria and Lipotyphla) the scaling relationships are significant only when phylogeny is ignored (Fig. 3), and in Archonta there are severe statistical problems of convergence because of the inadequate replication at lower taxonomic levels. In general, allowing for phylogeny does not affect the estimate of slope (an exception in mammals being for Lipotyphla; Fig. 3a), but increases the SE as would be expected by allowing for the positive correlation among the observations within lower taxonomic levels (phylogenetic pseudoreplication). The scaling relationships for Marsupialia, Ferae and Ungulata are shown in Fig. 4. In the Marsupialia, the scaling relationships for the different component orders would appear to broadly covary, suggesting that it is valid to fit an overall scaling relationship for marsupials as a group. In Ferae, however, the cimolestids (pangolins) are clearly different from the carnivores, and the relationship in the Ungulata is driven primarily by artiodactyls.
| Higher group | Order | n | Slope | SE | P |
|---|---|---|---|---|---|
| Mammals | |||||
| Marsupialia | Dasyuromorpha | 23 | 0·10 | 0·18 | 0·062 |
| Diprontodontia | 34 | 0·11 | 0·092 | 0·24 | |
| Didelphimorphia | 12 | 0·21 | 1·14 | 0·28 | |
| Epitheria | Lagomorpha | 11 | –0·16 | 0·16 | 0·34 |
| Rodentia | 251 | 0·044 | 0·071 | 0·54 | |
| Ferae | Carnivora | 58 | –0·17 | 0·073 | 0·026 |
| Lipotyphla | Erinaceomorpha | 10 | –0·81 | 0·24 | 0·01 |
| Soricomorpha | 28 | –0·16 | 0·24 | 0·5 | |
| Archonta | Chiroptera | 74 | 0·56 | 0·18 | 0·002 |
| Primates | 17 | 0·29 | 1·26 | 0·23 | |
| Ungulata | Artiodactyla | 20 | –0·25 | 0·089 | 0·011 |
| Birds | |||||
| Galloanserae | Anseriformes | 10 | 0·11 | 0·26 | 0·67 |
| Picae | Piciformes | 14 | 0·03 | 0·18 | 0·87 |
| Passerae | Ciconiiformes | 41 | –0·47 | 0·086 | < 0·001 |
| Passeriformes | 175 | 0·11 | 0·054 | 0·046 | |
Correlation of the slope of the relationship between body temperature and body mass (natural log transformed) when phylogenetic effects are ignored (OLS, ordinary least-squares) or allowed for (phylogenetic: hierarchical statistical model). Black symbols are taxonomic groups with statistically significant slopes in both models, grey symbols represent slopes that are significant under OLS but not phylogenetic analyses, and the open circles are significant in neither model. The lines show the 1 : 1 relationship. (a) Mammal higher groups; the outlier is Lipotyphla. (b) Mammalian orders; the outliers are Primates (above the line) and Soricomorpha (below the line). (c) Bird higher groups. Statistical analyses are summarized in Supplementary Tables S2 (birds) and S3 (mammals).
Scaling of body temperature with body mass for three mammalian higher groups where the relationship is significant in a phylogenetically based analysis. The different symbols represent individual orders within the higher groups. (a) Marsupiala. (b) Ferae, with cimolestids (pangolins) shown as open circles and carnivores as black symbols. (c) Ungulata, with artiodactyls shown as black symbols and all other orders as open symbols.
Body temperature and body mass data are available for 26 mammalian orders, but of these only 11 contain data for 10 or more species (see Supplementary material). Even in these there are problems for estimation of variation between families (within orders) and between genera (within families) because of small sample sizes. Thus for the marsupial orders Dasyuromorpha and Didelphimorphia there are data for only one family, and within Lagomorpha (rabbits and hares), Erinaceomorpha (hedgehogs and allies) and Soricomorpha (shrews and allies) there is minimal replication with data for only two families. In the Artiodactyla, there are data for only one species per genus, and so variation between genera and species is totally confounded. Also in Erinaceomorpha, Primates and Didelphimorphia the number of species exceeds that of genera by only one to three, so there is little scope for separating the two components of variance.
Despite these difficulties, there is strong evidence for differences in the slopes of the scaling relationship across the different orders (Wald test: W = 60·41, P < 0·001; Table 2). Allowing for phylogenetic effects, there is a significant positive relationship between body temperature and log body mass in Chiroptera (bats), and significant negative scaling in Erinaceomorpha, Carnivora and Artiodactyla (Fig. 5). The scaling relationships for Diprotodontia (positive), Rodentia and Soricomorpha (both negative), are non-significant once phylogenetic effects were allowed for, and there is no significant relationship for Dasyuromorpha, Didelphimorphia, Lagomorpha or Primates either ignoring or allowing for phylogeny. As with the analysis of the higher groups, in most cases allowing for phylogenetic effects does not change the slope of the scaling relationship, but does increase the SE and hence changes the significance level of the observed relationship (Fig. 3).
Scaling of body temperature with body mass for three mammalian orders where the relationship is significant in a phylogenetically based analysis. (a) Erinaceomorpha (negative scaling). (b) Artiodactyla (negative scaling). (c) Chiroptera (positive scaling).
birds
In birds, data are available for five parvclasses, although Coraciae contains data for only two species. As reported by McNab (1966), the lowest body temperatures in birds are found in ratites (Fig. 6) and penguins. In our data set comparably low temperatures are found in a single species of kingfisher and loon, and two species of Australasian frogmouth. The low temperature recorded for ratites appear, however, to be simply part of a general scaling relationship evident for all non-passerines (sensu stricto: all birds excluding parvclass Passerae; Fig. 7). This relationship is strongly negative and highly significant; it remains significant after removal of the outlier for the brown kiwi, Apteryx australis. Extending the analysis for ‘non-passerines’ to include groups within parvclass Passerae not traditionally regarded as passerines (i.e. all birds except those in the order Passeriformes) produces a much reduced slope, but this broader relationship includes hummingbirds (order Trochiliformes) where the scaling relationship is positive, though not statistically significant. We must conclude that non-passeriformes sensu lato is a heterogeneous assemblage in terms of thermal relationships, and hence any attempt to derive an overall scaling relationship for temperature and body mass for this group is probably invalid.
Analysis of body temperature in bird higher groups. Box plots of temperature data for each parvclass containing data for four or more species.
Scaling of body temperature with body mass for three bird orders where the relationship is significant in a phylogenetically based analysis, confining analyses to taxa where sex was determined. (a) Non-passeriforms (sensu stricto: all species excluding parvclass Passerae), with ratites shown as open symbols. The outlier is the brown kiwi, Apteryx australis. (b) Ciconiiformes, with data for suborders Charadrii (open symbols) and Ciconii (black symbols) plotted separately. (c) Passeriformes, with data for Tyranni (black symbols) and Passeri (open symbols) plotted separately.
At the next lower taxonomic level, data are available for 16 bird orders, but only seven contain data for 10 or more species (see Supplementary material). In contrast to the mammalian data, many studies of body temperature in birds record the sex of the individual measured. Many groups of bird exhibit marked sexual dimorphism in size or colouration, and this is frequently accompanied by a difference in body temperature. Analysis of the distribution of variance across taxonomic levels indicated that whilst as expected much of the variance is at the higher taxonomic levels, there is also marked variance at the level of individuals within species (Table 1), reflecting strong sexual differences in body temperature. Many of the families and orders examined in this study have data where the sex is recorded, though there are significant systematic differences in body temperature between sexes only for Galliformes (game-birds), Passeriformes (song-birds) and Gruiformes (cranes, though here there are only two species in the data set). We therefore chose to confine our analyses to those higher groups and orders containing more than 10 species with temperature, body mass and sex all recorded. Only four orders satisfy these criteria: Anseriformes, Piciformes, Ciconiiformes and Passeriformes.
Analysis of the trends in body temperature with body mass for these four orders, allowing for phylogenetic effects, indicates significant scaling in only two orders (Table 2). In Ciconiiformes, body temperature decreases with body mass, whereas in Passeriformes the trend is positive (Fig. 7). These two orders are very large and so were also analysed following separation into suborders. More recent molecular evidence has challenged the validity of the large order Charadriiformes as defined by Monroe & Sibley (1993); this order was therefore split into Charadrii (sandgrouse, waders, plovers, gulls, terms, skimmers, skuas and auks) and Ciconii (birds of prey, grebes, tropicbirds, cormorants, herons, storks, frigatebirds, loons and tubenoses). Scaling of body temperature is significantly negative in both suborders, though after allowing for phylogenetic effects the slope is steeper in Ciconii than in Charadrii. Similarly, Passeriformes were split into Tyranni (comprising essentially the New World suboscines and New Zealand wrens) and Passeri (everything else). Here the phylogenetic analyses reveal different scaling with Tyranni showing no significant slope, but Passeri exhibiting strong and significant positive scaling.
Discussion
The analyses reported here reveal hitherto unrecognized aspects of the pattern of body temperature in mammals and birds that are correlated with phylogeny. Interpretation of these patterns, however, must allow for potential sources of error or bias in the results.
Data quality is an important factor, especially as many measurements in the literature fall outside the criteria established for meaningful measures of metabolic rate or body temperature. For this study, analyses were confined to the highest quality data, comprising 51% of all the mammal records and 43% of bird records culled from the literature. Errors may arise where body mass was not recorded at the time of temperature measurement, and a representative body mass has to be taken from the literature (Savage et al. 2004; Farrell-Gray & Gotelli 2005). This error is impossible to quantify, but is unlikely to have introduced a bias (i.e. for body mass to have been systematically under or overestimated). Sensitivity analysis indicates that random errors in body mass would need to be substantial for the estimated scaling to be in error by 5% (see Appendix S1, and Tables S1–S4 in the Supplementary material).
The most important source of measurement error is probably the circadian rhythm in body temperature. Body temperature in birds and mammals has long been known to exhibit a circadian rhythm (Aschoff 1982; Mortola & Lanthier 2004), and in mammals this has been shown to be closely synchronized with daily rhythms of locomotor activity (Refinetti 1999). In birds, facultative hypothermia is widespread phylogenetically, and a circadian rhythm is particularly marked in small species that depend on insects or nectar for food (McKechnie & Lovegrove 2002). The magnitude of the circadian rhythm in body temperature is typically inversely related to body size (Aschoff 1982; Mortola & Lanthier 2004) and this in turn makes the potential error in the estimate of maximum body temperature also size-dependent. For metabolic rate the effect is small above a body mass of about 50 g. Errors from circadian rhythms will thus manifest themselves as a triangular shape to the scatter plot, where the variance in body temperature is inversely related to body mass. This source of error is not the only cause of such triangular plots, indeed this shape is strongly indicative of a range of interacting factors at work, but it is potentially important in relation to both body temperature and metabolic rate data.
Analysis of the residuals suggested that errors from circadian rhythms may be present in just three groups, namely passerine birds, rodents and bats. For larger groups such as marsupials, carnivores and non-passerine birds there is no indication that errors from circadian rhythms are significant. Neither is there any indication for insectivores (Lipotyphla), despite this group containing some taxa with small body size. If errors from circadian rhythms are dominating the pattern, then an apparent positive scaling relationships will be biased towards a stronger relationship than exists in nature and the real relationship might best be estimated from the upper bound of the scatterplot. The latter was estimated using the technique of Blackburn, Lawton & Perry (1992). For all three groups, bats, rodents and passerine birds, the upper bound scaling relationship is not significant (in all cases P > 0·05), whereas in bats and passerines the overall plot indicates a small positive scaling of body temperature with mass. An alternative approach is to use a weighted regression. In Chiroptera, applying a hierarchical statistical model with variance increasing exponentially with decreasing body mass reduces the slope slightly to 0·50 (SE 0·15). Overall these analyses indicate that data for groups containing taxa of larger body size are entirely robust to any errors from circadian rhythms, and that the effects are likely to be small in other groups.
phylogenetic patterns in body temperature
Interest in the body temperature of mammals and birds has a long history (McNab 1970). A number of analyses have suggested that any overall relationship might be confounded by mixing a variety of different relationships for different taxa (McNab 1970; White & Seymour 2003), and this is confirmed clearly by the analyses presented here. In an evolutionary context it makes no sense to construct a scaling relationship for mammals or birds overall, because different groups of mammals and birds have evolved quite different body temperature scaling relationships (Table 2).
Analyses that allow for phylogenetic structure reveal a number of important evolutionary features in the mammalian data. It has long been recognized that monotremes (Prototheria) have the lowest body temperatures amongst mammals, and the four monotremes in this data set were amongst the very lowest body temperature recorded. However comparably low temperatures were recorded for the silky anteater Cyclopes didactylus, the marsupial mole Notoryctes caurinus and nine species of tenrec. Thus although monotremes as a group do indeed exhibit the lowest temperatures amongst the mammals, the traditional ranking of increasing body temperatures in the sequence monotromes – marsupials – placentals is misleading. Two groups of placentals (armadillos, anteaters and sloths: Xenarthra, and insectivores: Lipotyphla) have lower median body temperatures than do marsupials overall (though in Lipotyphla only slightly so, and this disappears when Eulipotyphla are incorporated into Laurasiatheria). Furthermore, marsupials do not form a coherent group themselves in terms of body temperature, with median values increasing in the sequence didelphids – dasyurids – diprotodonts; whilst the data suggest an overall relationship for marsupials may be valid, the statistical analyses could not confirm positive scaling for any individual order within the Maruspialia (though the small number of species for which we have data, and the narrow range of body masses constrain our ability to detect any scaling trend that might exist). The highest individual mammalian body temperatures are for species of ungulate and carnivore, and it may be because these two groups dominated early work on body temperature that the traditional ranking developed. The highest mean temperatures were recorded for mustelids (14 species), bovids (9 species) and lagomorphs (11 species), and placentals had the highest values of body temperature and body mass in the overall data set.
White & Seymour (2003) reported a positive scaling between temperature and body mass in marsupials, and this is confirmed in the data set analysed here (Fig. 4a). The strong scaling for all marsupials is dominated by diprotodonts, although both dasyurids and didelphids also exhibit positive (but non-significant) scaling. A positive scaling relationship is also exhibited by bats (Fig. 3c), as previously reported by McNab (1969). Rodents show a positive scaling but here the relationship is not quite significant (Table 2). A strongly negative scaling of temperature with body mass is seen in artiodactyls (Fig. 5b); this relationship remains statistically significant even after removing the data for Hippopotamus. Insectivores (sensu lato: Lipotyphla) also show a strongly negative scaling, which is similar in the two orders dominating the data set for this group, Soricomorpha (shrews and allies) and Erinaceomorpha (hedgehogs and allies). Although the scaling for these two orders is significant (Table 2), for Lipotyphla overall there is no significant scaling (P > 0·05). The most complex relationship is that exhibited by the grandorder Ferae. Here the order Carnivora exhibits a significant negative scaling of temperature with body mass, and the marine taxa (one otariid and three phocid seals) are not inconsistent with the trend established for the non-marine carnivores. The cimolestids (pangolins), however, are clearly different from the overall trend within Ferae (Fig. 4b), and this group is increasingly regarded as distinct from the Carnivora within the Laurasiatheria (Bininda-Emonds et al. 2007).
thermal ecology and body temperature
For both birds and eutherian mammals, in those higher groups or orders containing species of large body size, scaling of body temperature is negative: in non-passerine birds, artiodactyls and carnivores, big species have lower body temperatures than small species. Marsupials are thus strikingly different in having some species of large body size, but a strongly positive scaling of body temperature. Simplistic surface area to volume considerations might suggest that larger species would be warmer (an idea that underpins the concept of inertial homeothermy), although this ignores the important role of respiratory heat and water loss in the thermal balance of vertebrates. The body temperature of endotherms is carefully regulated, and this negative scaling of body temperature indicates that the set-point for body temperature is under selection.
McNab (1970) developed a formal biophysical argument whereby the body temperature of a mammal or bird is a balance between metabolic rate (which governs the rate at which heat is produced) and conductance, which is heavily influenced by insulation and governs the rate of which heat is lost. Seebacher (2003) has also developed a sophisticated model for ectotherms which demonstrates clearly the importance of insulation in regulating body temperature. Body temperature is thus influenced by selection acting on aspects of both anatomy and physiology, and the thermal characteristics of a particular group such as monotremes, ratites, marsupials, bats or passerines are neither primitive nor advanced, but an adaptation to the habitat and ecology of the group in question (as has been shown clearly for metabolic rate by Lovegrove 2000). There is no evidence from this preliminary study that variations in the body temperature of birds or mammals represent an evolutionary sequence, or the gradual acquisition of improved endothermic physiology (as was originally proposed: Sutherland 1899; Wetmore 1921). Rather, we are faced with a series of more-or-less independent evolutionary adjustments to the costs and benefits of maintaining a particular body temperature, influenced by ecology and constrained by phylogeny.
Our analysis has demonstrated that the heterogeneity observed in the scaling of body temperature in both mammals and birds is correlated with phylogeny. There is, of course, a very strong covariation between ecology and phylogeny, and McNab (1970) has argued cogently that assigning all of the causality to phylogeny (or ecology) alone is giving artificial primacy to one driver over another, when in reality they are tightly coupled. Thus, although our analysis is concerned with variability associated with phylogeny, the immediate driver is ecology.
An intriguing problem posed by the results of our analysis is why some groups exhibit positive scaling and others negative scaling in body temperature. Some groups of mammals undergo significant fermentation in the gut, and it is possible that this process may provide a significant source of body heat as well as nutrient. The scaling relationships for body temperature in the major groups of gut fermenters are, however, quite different: in artiodactyls it is strongly negative, in marsupials (where the data are dominated by diprotodonts) it is strongly positive, and lagomorphs show no significant relationship (Table 2). There is thus no consistent pattern across different phylogenetic groups in the extent to which heat from gut fermentation influences body temperature.
As has long been known, birds as a whole have higher body temperatures than mammals, and ratites have the lowest body temperatures of birds, close to the warmest mammals. Picae (woodpeckers, barbets and allies) have the warmest median body temperatures for birds, whilst the highest individual temperatures are in passerines.
Concluding remarks
Whilst the inherent inadequacies of the data constrain any conclusions we reach, a number of features do appear to be robust. Birds clearly tend to have a higher body temperature than mammals, and both ratites and monotremes have the lowest body temperatures of their class. The traditional view of mean body temperature increasing among mammals in the sequence monotremes – marsupials – placentals, and in birds from non-passerines to passerines, masks much underlying heterogeneity; these simplistic rankings are thus misleading and should be abandoned.
An important conclusion from the analyses reported here is the absence of evidence for an overall scaling relationship for either mammals or birds (though note the earlier discussion concerning the power of the analyses to detect such scaling). Rather there are individual scaling relationships for individual lineages within mammals and birds.
Although the development of the metabolic theory of ecology (Brown et al. 2004) was a stimulus to our analysis of body temperatures, our results do not constitute a strong test of either the assumptions underpinning the theory, or of its predictions. They do, however, indicate that resting metabolic rates predicted from the theory will exhibit significant heterogeneity, driven by the varying scaling patterns of body temperature across different groups of mammals and birds. A recent analysis demonstrated variation in the scaling exponent of resting metabolic rate across mammalian orders (Duncan, Forsyth & Hone 2007). If body temperature is a major determinant of resting metabolic rate (a central tenet of the metabolic theory of ecology) then we would predict that the scaling of resting metabolic rate would be lower in those mammalian orders with negative scaling of body temperature than in those with positive scaling of body temperature. A comparison of the scaling exponents for metabolic rate and body temperature reveals no significant relationship at the level of order (Pearson r = 0·514, P > 0·1). This indicates a complex relationship between body mass, body temperature and resting metabolic rate (Fig. 1) and leaves open the intriguing question of whether evolution has adjusted resting metabolic rate through changes to body temperature, or whether body temperature is simply a consequence of the resting metabolic rate that has evolved for a particular environment and ecology.
Acknowledgements
The initial stimulus for this work was the serendipitous discovery of Rodbard's work in an old ecology textbook, whilst browsing in a second-hand book-store. The initial compilation of data was undertaken with outstanding care and attention to detail by Pascale Perry. We are extremely grateful to Barry Lovegrove, Craig White and Roger Seymour for making their own compilations of data available, to Adrian Friday for discussion and helpful suggestions as the analysis progressed and to Brian McNab and two other referees for detailed and constructive criticism which improved the paper significantly.
