Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
Contents lists available at SciVerse ScienceDirect
Comparative Biochemistry and Physiology, Part A
j ourna l homepage: www.e lsev ie r .com/ locate /cbpa
Estimatingfieldmetabolic rates for Australianmarsupials using phylogeny
Alexander Riek a,⁎, Jorn Bruggeman b
a Department of Animal Sciences, University of Göttingen, Albrecht-Thaer-Weg 3, 37075 Göttingen, Germanyb Department of Earth Sciences, University of Oxford, South Parks Road, Oxford OX1 3AN, United Kingdom
Article history:Received 9 November 2012Received in revised form 17 January 2013Accepted 18 January 2013Available online 1 February 2013
Keywords:Allometric scalingEnergy expenditureMarsupialsPrediction equationsPhylogenetic tree
Field metabolic rate (FMR) is a useful measure for the energy expenditure in free-ranging animals. Field met-abolic rates for species that have not been measured are usually predicted by allometric equations on thebasis of their body mass (BM). Phylogenetically informed methods improve estimates of both allometric re-lationships and species-specific FMR values by considering the evolutionary history of species. Further im-provement is possible by incorporating isolated measurements on BM and FMR, but most existing methodsforce the user to discard such incomplete data. In the present study the FMR of most Australian marsupialspecies was predicted for the first time using a phylogenetic method that was explicitly designed to handleincomplete data. This allows full use of the dataset containing 35 samples of FMR and 130 samples of BM.Cross-validation demonstrated that FMRs were estimated with high accuracy. The resulting prediction equa-tion was FMR (kJ day−1)=5.27 BM (g)0.69. Field metabolic rate and BM were highly phylogenetically corre-lated (r=0.96), i.e. FMR and BM co-evolved. Differences between species-specific and generic marsupialestimates of FMR revealed that herbivores have lower energy expenditure than carnivores. Specifically, her-bivorous macropods have on average lower relative FMR (kJ/d) (3.75±0.53 BM0.69; mean±SD) than carniv-orous dasyurids (7.64±0.84 BM0.69). Phylogenetically informed estimates for most extant Australianmarsupial species are now available.
Field metabolic rate (FMR) is an estimate of energy expenditure in afree-living animal. The FMR includes the total energy required for a cer-tain period of time to carry out all normal activities by an animal, such asthermoregulation, hunting, reproduction etc. The method to measureFMR, namely the doubly labelled water (DLW) method, was first pro-posed by Lifson et al. (1955). In brief, this method involves the enrich-ment of the body water of an animal with an oxygen and a hydrogenisotope. While the hydrogen isotope leaves the body as water only,the oxygen isotope is lost not only as water but also as respiratoryCO2. The difference between the elimination rates of hydrogen and ox-ygen isotopes can be used to estimate the CO2 production, which to-gether with the respiration quotient yields the O2 production and thusan estimate of the FMR (Nagy, 1987; Speakman, 1997; Butler et al.,2004). The DLW technique is currently themost reliable method for es-timating energy expenditure in free-ranging animals and has been usedto estimate the FMR in numerous wild mammalian species (for reviewsee Nagy, 1987; Nagy et al., 1999; Nagy, 2005; Capellini et al., 2010).
FMR is typically well described by an allometric function of BM. Thecorresponding allometric exponent has been, similarly to the exponent
of basal metabolic rate, the source of debate for many years (Nagy,1987; Koteja, 1991; Nagy, 1994; Nagy et al., 1999; Speakman, 2000;Nagy, 2005; Packard and Boardman, 2009; Capellini et al., 2010). Mostanalyses on the scaling of metabolic rates in mammals that propagatea universal exponent for all mammals failed to control for phylogenyexcept in a few recent studies (White et al., 2009; Capellini et al.,2010). Phylogeny plays an equally important role when reconstructingunknown values (e.g., FMR) of ancestral or present-day species. This isnot just because phylogenetically informed methods produce more ac-curate estimates of regression coefficients or allometric exponents:Garland and Ives (2000) demonstrated that a species' trait value canbe estimated with higher accuracy and reduced uncertainty if the posi-tion of its ancestor in the phylogeny is taken into account.
Recent methods unify traditional allometric and phylogenetic anal-yses by partitioning observed variability in trait values among phyloge-netic and phenotypic components. This partition can bemade explicitly(Ives et al., 2007; Felsenstein, 2008a; Bruggeman et al., 2009) or implic-itly by introducing a factor (“phylogenetic signal”) that scales the influ-ence of the phylogeny (Pagel, 1999; Freckleton et al., 2002). Suchextensions have a notable side effect: where the classic allometric orphylogenetic relationship between two traits (e.g., FMR and BM) is dic-tated by joint observations on both traits, the extended analyses alsobenefit from information on the evolution of the individual traits. Con-cretely, every measured trait value is valuable, even if correspondingvalues for other traits of interest are unavailable. However, mostexisting phylogeneticmethods do not provide ameans of incorporating
599A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
such incomplete data. This is unfortunate, since incomplete datasets arethe rule rather than the exception: for instance, in the present study, 95out of 130 Australianmarsupial species lack a value for FMR, while theirBM is known. These 95 BM values would need to be discarded in mostanalyses, even though they carry potentially valuable information.
The recently developed PhyloPars web server (Bruggeman et al.,2009) estimates evolutionary parameters (e.g. phylogenetic regressioncoefficients or allometric exponents) as well as the trait values of allspecies in the phylogeny, while explicitly incorporating the informationfrom incomplete data. For the present study, we employ this method toobtain the best possible estimates for the FMR of Australianmarsupials,taking into account their placement in the phylogeny as well as theirknown BM. Our study is the first to predict unknown FMR valuesusing phylogenetic information. The resulting estimates of energy ex-penditure for all Australian marsupials will be of value especially inareas such as conservation biology, where it is at times impractical orimpossible to obtain the direct data. Thus these estimates are expectedto contribute to the conservation andmanagement of these animals, be-cause the knowledge of the total energy cost of free existence of a spe-cies allows predicting the resource availability for this species in itshabitat. Furthermore, it can help in establishing long-term conservationstrategies as shown for some arid-zone marsupials in Australia (Humeet al., 2004; Lapidge and Munn, 2012).
2. Methods
2.1. Database
Data on BM and FMR from 35 published studies were available forthe present analysis. Body mass data from an additional 95 Australianmarsupial species were taken from Strahan and van Dyck (2008).Thus the dataset included a total of 35 FMR and 130 BM values, leaving95 missing values for FMR. An effort was made to include all availablestudies that reported FMR in free-living, non-lactating and non-torpidadult Australian marsupials. Only marsupial species occurring inAustralia were chosen for the present study because for the remainingapprox. 150marsupial species, distributed over PapuaNewGuinea, sev-eral pacific islands, North- and South America, only four FMR valueswere available. The data on FMR and BMare available as supplementarymaterial (Appendix S1).
2.2. Phylogeny
The phylogeny in Newick format was derived from the recentlypublished mammalian supertree which includes 4510 species withupdated branch lengths derived from dated estimates of divergencetimes (Bininda-Emonds et al., 2007, 2008). The supertree for mam-mals was pruned to include only the species of concern for thisstudy, i.e. Australian marsupials (N=130). The tree is available assupporting information (Appendix S2).
2.3. Phylogenetic analysis
To estimate phylogeny-controlled values of the allometric exponentrelating FMR to BM, as well as the 95 unknown FMR, the known traitvalues of present-day marsupial species were combined with the phy-logeny in a “Brownian motion” model of evolution (Felsenstein, 1988).This model assumes trait evolution to be indistinguishable from a corre-lated randomwalk. The rate and direction of evolution are described bythe phylogenetic covariance matrix (Felsenstein, 2008a), which can beestimated from the combination of known trait values and phylogeny.The Brownian motion model can only account for linear relationshipsbetween traits. As a power law (allometric) relationship between BMand FMRwas expected, all trait values were log10 transformed to linear-ise this relationship. The phylogenetic regression coefficient thus comesto represent the allometric exponent relating FMR to BM.
Several methods exist to estimate evolutionary parameters from ob-served trait values and phylogeny. Such methods are generally basedupon Phylogenetically Independent Contrasts (PIC; PHYLIP; Felsenstein,2008b) or Phylogenetic Generalized Least Squares (PGLS; BayesTraits;Pagel et al., 2004).However,most existingmethods donot handle incom-plete datasets, as missing data preclude both the analytical estimation ofevolutionary parameters and the standard varieties of the ubiquitousExpectation-Maximization (EM) scheme (Dempster et al., 1977). Conse-quently, these methods cannot incorporate the information of the 95values of BM that lack an associated FMR. The recently developedPhyloPars web server (Bruggeman et al., 2009, http://www.ibi.vu.nl/programs/phylopars/) is an exception, as it is a PGLS-type methoddesigned specifically to handle missing data. In short, PhyloPars employsa restricted maximum likelihood (REML) approach to estimate theCholesky-transformed phylogenetic covariance matrix (Pinheiro andBates, 1996), using the iterative Broyden–Fletcher–Goldfarb–Shanno op-timization procedure. This method has several desirable properties: ituses all available data, produces a positive-definite, unbiased estimateof the phylogenetic covariance matrix, and converges rapidly. We usedthe PhyloPars web server to estimate the phylogenetic allometric expo-nent that links BM and FMR, as well as the FMR values of the 95present-day marsupials of which only BM is known.
2.4. Phylogenetic signal
Phylogenetic generalized least squares estimates of regression coef-ficients (allometric exponents) and ancestral states offer considerableimprovement over those provided by traditional Ordinary Least Squares(OLS) methods, provided the traits under study exhibit a phylogeneticsignal. In other words, PGLS is useful only if more closely related speciesare more likely to have similar trait values. This can be tested by intro-ducing a parameter λ (0≤λ≤1) that scales internal branch lengths inthe phylogeny (Pagel, 1999; Freckleton et al., 2002). This parameter isusually interpreted somewhat vaguely as a measure of evolutionaryconstraint acting on the phenotypes. However, the role of λ is mathe-matically indistinguishable from a form of phenotypic variability inwhich the phenotypic covariancematrix is proportional to the phyloge-netic covariancematrix (cf. Felsenstein, 2008a). If all samples are furthertaken at the same distance (“time”) from the root of the tree, the pheno-typic variability is effectively equal for all observed species, as in othermethods that explicitly distinguish phenotypic variability (Felsenstein,2008b; Bruggeman et al., 2009). By introducing a single λ for both traits,its optimal value is dictated not only by joint observations on FMR andBM, but also on isolated measurements of either trait. This emphasizesthe need for a method that incorporates incomplete data. To determinethe phylogenetic signal, the PhyloPars procedure was modified to esti-mate λ. To verify whether λ was estimated correctly, the procedurewas repeated after eliminating the 95 incomplete rows from the datasetand switching the procedure from REML to (biased) maximum likeli-hood; the results of this modified PhyloPars analysis were found to beidentical to estimates generated by BayesTraits (Pagel et al., 2004).
We determined the strength of the phylogenetic signal (λ) in the twotraits (i.e. FMR and BM), first separately for the individual traits, then forboth in combination. To determine whether λ differed significantly from0 (no phylogenetic signal) and 1 (the signal expected from a randomwalk model), likelihood values were compared in a likelihood ratio test.The ratio between maximum likelihoods attained with free λ and fixedλ, multiplied by 2, produces a test statistic that is expected to followchi-square distribution with 1 degree of freedom (Pagel, 1997, 1999;Freckleton et al., 2002). The P value is then given by 1minus the cumula-tive density of the chi-square distributionwith 1 degree of freedom, at thevalue of the test statistic. The same method was used to determine theconfidence interval of the allometric exponent linking FMR and BM. Forthis purpose, the internal representation of the phylogenetic covariancematrix was redefined in terms of the regression coefficient of interest bymodification of the Cholesky transformation (Pinheiro and Bates, 1996);
Fig. 1. Relationship between fieldmetabolic rate (FMR) and bodymass (BM) inAustralianmarsupials derived by ordinary least squares regression (OLS, short-dashed line; FMR(kJ day−1)=10.32 BM (g)0.58, N=35, Pb0.001) and by phylogenetic generalized leastsquares (PGLS, solid line; FMR (kJ day−1)=5.27 BM(g)0.69, N=130, Pb0.001) regressionfor measured FMR values; the Y-intercept was calculated by forcing the regression linethrough the estimate for the root of the tree, following Garland et al. (1993) and Garlandand Ives (2000).
Table 2Phylogenetic generalized least squares models for the allometry between field metabolicrate (FMR) and body mass (BM) for Australian marsupials, with the maximum likelihood(ML) λ value for the relationship between FMR and BM and with the slope (b) and inter-cept. For the slope 95 % confidence intervals are listed between parentheses.
Species N MLλ
P forλ=1a
b (c.i.) log10Intercept
All marsupials 130 0.98 0.017 0.686 0.722
600 A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
specifically, by dividing the first off-diagonal element of the Cholesky de-composition by the first diagonal element.
2.5. Estimating missing FMR values
After estimating the phylogenetic covariances that characterizethe rate and direction of evolution, PhyloPars combines these withthe original dataset to infer missing trait values for ancestral andpresent-day species. These estimates can be written as a weightedsum of the known values of both traits, with weights determined bycorrelations between species (i.e., phylogenetic distance) and correla-tions between traits (i.e., allometric relationships). Thus, the estimatefor an unknown FMR value of a present-day marsupial incorporatesknowledge on the species' known BM as well as the FMR of its phylo-genetic neighbours.
To determine the accuracy of inferred trait values, PhyloPars per-forms leave-one-out cross validation: known trait values are one byone omitted from thedataset, and then re-estimated from the remainingvalues using the phylogeny. This produces a set of residuals – one foreach original known trait value – that characterize the quality of the in-ferred missing values; the mean of the absolute value of these residualsis used as measure of accuracy. To aid in judging this measure of accura-cy, cross-validation is repeated with two simple null models: a meanmodel in which the inferred value is equal to the mean of all remainingknown values for the trait of interest, and a nearest neighbour model inwhich the inferred value is equal to the phylogenetically nearest knownvalue.
3. Results
3.1. Phylogenetic signal and allometric model fit
Measured FMR and BM both exhibited individually strong phylo-genetic signals with maximum likelihood (ML) values of λ of 0.84and 0.96, respectively (Table 1). For BM alone and both traits in com-bination the phylogenetic signal was slightly but significantly different(Pb0.05) from λ=1, suggesting that a small fraction of the variabilityis not due to evolutionary constraints, but, e.g., to phenotypic variability.The fact that the phylogenetic signal λ differs significantly from 0 in allcases emphasizes the inappropriateness of OLS analysis and the needfor a phylogenetic correction of the trait data. The allometric relation-ship derived by the evolutionary model, using measured and phyloge-netically controlled estimates was FMR (kJ day−1)=5.27 BM (g)0.69
(Fig. 1, Table 2). The 95% confidence interval of the exponent was0.62–0.75, which notably excludes the exponent estimated with OLS(0.58, Fig. 1). This further confirms the need for phylogenetically in-formed analysis. We may note that the scale factor 5.27 applies to theevolutionary ancestor of all marsupials, and therefore is appropriatewhen estimating FMR for a general marsupial that lacks phylogeneticplacement.
3.2. Estimated FMR values
In our analysis we estimated FMR for 95 Australian marsupialspecies on the basis of their phylogeny, ranging in size from 4 g
Table 1Comparison between the strength of the phylogenetic signal (λ) in field metabolic rate(FMR), body mass (BM), and both traits in combination. Additionally, P values are givenfor two fixed values of λ, representing a non-phylogenetic model (λ=0; equivalent toOLS) and the original random walk model (λ=1).
Trait BM FMR FMR×BM
ML λ 0.96 0.84 0.98P for λ=0 b0.001 b0.001 b0.001P for λ=1 0.025 0.092 0.017
ML λ=maximum likelihood for λ.
(Planigale ingrami) to 44.5 kg (Macropus giganteus; see supporting in-formation, Appendix S1). When estimating missing trait values, theevolutionary model had the lowest cross-validation errors for bothtraits (BM and FMR), compared to the mean and nearest neighbourmodel, respectively (Table 3). FMR and BM were highly phylogeneti-cally correlated (r=0.96 and different from 0 with p=1.00000),which indicates FMR and BM tightly co-evolved. As a result, the pat-terns in the FMR estimates were dominated by the trends in BM(Fig. 2).
To investigate interspecific differences unrelated to BM, we com-pared the species-specific ratios of FMR to BM0.69. This quantity is equiv-alent to the allometric scale factor for FMR that is most appropriate forthat species and its descendants. High values of the scale factor indicaterelatively high energy expenditure of the species; low scale factors indi-cate low expenditure. The logarithm of the species-specific scale factoris a linear combination of log-transformed FMR and biomass. Since thejoint evolution of these variables is described by the Brownian motionmodel, any linear combination of the two, including the (log) allometric
a P for λ=0 was for all species b0.01.b Equal to the order Dasyuromorphia.c Includes the orders Peramelemorphia (N=8), Notoryctemorphia (N=2) and
Diprotodontia (N=71).
Table 3Cross-validation errors for body mass (BM) and field metabolic rate (FMR) generatedusing the PhyloPars program (Bruggeman et al., 2009; see text for details).
Model Cross-validation error
log10 BM log10 FMR
(g) (kJ day−1)
Evolutionary model 0.206 0.094Mean model 1.020 0.422Nearest neighbour model 0.291 0.404
601A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
scale factor, must evolve according to the Brownian motion model aswell. Moreover, the expected value and covariance of the linearcombination is completely specified by the expectation and covarianceof log FMR and log biomass. Therefore, we can calculate the expectedscale factor directly from previous results. The results showed thatspecies in the order Dasyuromorphia had relatively higher FMRs thanthe species in the orders Peramelemorphia or Diprotodontia, respective-ly (Fig. 3). To determine whether the evolution of all marsupials is
Fig. 2. Fieldmetabolic rates (●=measured,○=estimated using PhyloPars) across the phylogeet al., 2008, see text for details). The best estimate for the FMR of a species not included in the
nonetheless well-described by a model with a single allometric expo-nent, we separated the dataset into carnivorous marsupials (order:Dasyuromorphia, N=49) and non-carnivorous marsupials (orders:Peramelemorphia, N=8; Notoryctemorphia, N=2; Diprotodontia,N=71). The resulting allometric exponents of these individual sub-groups did not differ significantly from the value of 0.69 estimated forall marsupials combined (Table 2). Within these orders herbivorousmacropods had with 3.75±0.53 kJ day−1 g−0.69 (mean±SD) on aver-age the lowest relative FMR (range: 2.31–5.44 kJ day−1 g−0.69) andcarnivorous dasyurids with 7.64±0.84 kJ day−1 g−0.69 the highest(range: 5.51–9.90 kJ d−1 g−0.69). Since there was a clear difference inrelative FMRwithin the order Diprotodontia betweenMacropodiformes(i.e. kangaroos, wallabies etc.) and the remaining species in this order(i.e. Vombatiformes and Phalangeriforms, Fig. 3) we further dividedthe dataset into macropods (N=35) and the remaining species in thisorder (N=36). The resulting PGLS regression equations differed inslope and intercept withmacropods having a substantially higher expo-nent (0.86) than the value estimated for all marsupials combined (0.69;Fig. 4, Table 2). However, it should be noted that the confidence interval
ny of Australianmarsupials (constructed from themammalian Supertree, Bininda-Emondstree is equal to the estimate for its nearest evolutionary ancestor.
Fig. 3. Species-specific scale factor for field metabolic rates (●=measured, ○=estimated using PhyloPars) as allometric function of biomass, across the phylogeny of Australianmarsupials (constructed from the mammalian Supertree, Bininda-Emonds et al., 2008, see text for details). The expected FMR of a species and its descendants is obtained by takingthe product of the scale factor and the species’ body mass raised to the power of 0.69. This approach may be used to estimate the FMR of species omitted in the figure from its knownbody mass and ancestry.
602 A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
for the macropod exponent (0.694–1.019) only just excludes the bestestimate for all marsupials combined. Evidence for a change in allome-tric exponent during marsupial evolution is therefore weak at best.
4. Discussion
Our study is the first estimating FMR values for most Australianmarsupials using a phylogenetically informed method. The results in-dicate that FMR exhibits a clear phylogenetic signal, underlining theneed for phylogenetic correction when estimating allometric expo-nents and unknown FMR values.
4.1. Phylogenetic signal
Our results show that the PGLSmodel, which takes the phylogeneticsignal in the data into account, fits the data significantly better than theOLS model and thus should be preferred when analysing FMR. The
strength of the phylogenetic signal for the relationship between FMRand BM, estimated by the ML λ was also significantly different from 1,indicating that a small fraction of the signal is due to phenotypic vari-ability (e.g. intraspecific variation). Therefore, assuming a value for λsuch as standard PIC/PGLS (λ=1) or OLS (λ=0), can over- or underes-timate the influence of shared evolution and will likely lead to errone-ous results and conclusions (Capellini et al., 2010).
4.2. The relevance of incomplete observations
Problems often encountered in phylogenetic analysis are either thelack of molecular data on a certain species and the resulting difficultyto place this species into the phylogeny, or the lack of traitmeasurementsfor a species with known phylogenetic position. A species with a lack ofeither of the aforementioned data is most commonly excluded from theanalysis. In our case we had a fully resolved phylogeny for Australianmarsupials derived from a recently published mammalian supertree
Fig. 4. Relationship between field metabolic rate (FMR) and body mass in Australianmarsupials derived by phylogenetic generalized least squares (PGLS) for carnivorousmarsupials (red line, dots), non-carnivorous marsupials (blue line, all open symbols),Diprotodontia excluding macropods (black line, triangles) and macropods (greenline, squares). For PGLS equations see Table 3; the Y-intercept was calculated by forc-ing the regression line through the estimate for the root of the tree, following Garlandet al. (1993) and Garland and Ives (2000).
603A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
(Bininda-Emonds et al., 2007, 2008). However from the 130 marsupialspecies included in the phylogeny, FMR values for 95 species were miss-ing. Conventional phylogenetic analyses exclude such incomplete rows,particularly at the stage that estimates the parameters of the evolution-ary process (i.e., phylogenetic signal, covariances, and regression coeffi-cients). In fact, reintroduction of the incomplete rows for the purposeof reconstructing missing values after phylogenetic parameters havebeen estimatedwould in principle be feasible in any conventional meth-od. However, if we repeat our analysis with incomplete rows removed,we find that this directly affects the estimates for the phylogeneticparameters: the phylogenetic signal drops from0.98 to 0.88 and the allo-metric exponent linking FMR to BM decreases from 0.69 (0.62–0.75) to0.66 (0.59–0.74). In other words, exclusion of incomplete rows canlead to reduced confidence and biases in estimates of the phylogenic sig-nal and regression coefficients. There is a clear need for phylogeneticmethods that can handle missing data.
4.3. Quantifying the value of additional measurements
In addition to inferring FMR values of unmeasured species, the phy-logenetic model can guide future experiments: it can identify whichnew measurements would most reduce uncertainty in inferred values.If we define a set of desired trait values (e.g., FMR for all marsupials, orfor a specific clade) and a set of measured trait values (e.g., the presentdataset plus a potential new FMR measurement), the model prescribesthat the variance of the desired trait values, conditional on the observa-tions, is a function of covariances only (e.g., Johnson and Kotz, 1972).These covariances are specified by the phylogeny and themodel param-eters (phylogenetic covariances and phylogenetic signal); they do notdepend on the measured values themselves (Freckleton et al., 2002).We can thus enumerate all species with missing FMR, and for each spe-cies quantify to what extent measurement of its FMR would reduce es-timate uncertainty beyond that obtained with the present dataset. Inthis manner we may identify new FMR measurements that wouldmost improve the estimates of interest. As the result of such a procedureis completely dependent on chosen species of interest, we cannotpresent general conclusions here. For instance, if we choose to optimize
the FMR estimate for the common ancestor of all marsupials, overall un-certainty is most reduced by measuring the FMR of Notoryctes sp.,followed by one of the Phalangeroidea. If we alternatively choose to op-timize the FMR estimate for the clade that includes Sminthopsis,Ningaui and Planigale, uncertainty is most reduced by measuring theFMR of Planigale sp., followed by Sminthopsis laniger (see also Figs. 2and 3).
4.4. Field metabolic rate in marsupials
As already known for other clades, FMR inmarsupials is highly dom-inated by trends in BM, i.e. BM and FMR were highly phylogeneticallycorrelated (r=0.96, different from 0 with P=1.00000). As a result,the most intriguing patterns are not found in absolute FMR=kBM(g)0.69 (Fig. 2), which mirrors BM, but in the species-specific allometricscale factor k=FMRBM−0.69. This scale factor can be thought of as a rel-ativemeasure of FMR, corrected for the influence of BMand leaving onlyinfluences of other physiological and ecological traits. Our results show,that non-carnivorous marsupials such as macropods have substantiallylower relative FMR compared to carnivorous marsupials, such asdasyurids (Fig. 3). In an extreme example, the FMR of the carnivorousTasmanian devil (Sarcophilus laniarius, BM: ~9 kg) is with 4.1 MJ/dnearly twice as high as the FMR for the similar sized herbivorous rockwallaby (Petrogale xanthopus, FMR: 2.2 MJ/d, BM: ~8.9 kg). Comparingour two PGLS regression lines for carnivorous and non-carnivorousmar-supials, it seems that on average, carnivorous marsupials have at thelower endof theBMrange (i.e. 5–100 g) a 27–44%higher energy expen-diture and at the upper end of the BM range (i.e. 100–9000 g) a 14–27%higher energy expenditure compared to non-carnivorous marsupials.Similar results were found for placental carnivores and herbivores(Carbone et al., 2007; Riek et al., 2007) albeit including fewer numbersof species. This difference can generally be explained by the fact that car-nivores have to expend a substantial amount of energy for hunting andcapturing their prey (Carbone et al., 2007). This emphasizes, that FMR inmarsupials is not only influenced by BM but also by diet. Thus, our re-sults would suggest for the first time that this difference in energy ex-penditure between carnivorous and non-carnivorous animals, so faronly described for placentals, is also present in marsupials.
Furthermore, by calculating separate PGLS regression lines formacropods and the rest of the species in the order Diprotodontiashowed that most of the macropods expend less energy than anyother marsupial of the same size. While it has been already reportedthat some macropods have exceptionally low energy expenditures(Nagy et al., 1999) we show that this holds true for most of the speciesin their sub-order. Reasons for these low energy expenditures aremost likely due to the evolution of their distinct digestive capabilities,their manner of reproduction and their mode of locomotion. Dawsonand Taylor (1973)were the first to show, that although energy expendi-ture in kangaroos increases rapidly with increasing speed until up toabout 10 km/h comparable to most other land mammals, the amountof energy expended remains the same once they start to hop withspeeds of up to 35 km/h. A similar sized animal running at the samespeed would consume nearly twice as much energy (Tyndale-Biscoe,2005). Furthermore the evolution of a foregut fermentation systemallowedmacropods to substantially reduce energy expenditure throughthe symbiotic relationship with bacteria, enabling them to make effec-tive use of the energy content of plants by breaking down the celluloseof plant cell walls and thus releasing their content. Additionally, the cel-lulose is fermented by the bacteria to short-chain fatty acids, an energysource for the animal that would otherwise be unavailable without theenzymes from the bacteria to hydrolyse the cellulose (Van Soest, 1994;Hume, 1999; Tyndale-Biscoe, 2005). Therefore, it could be argued thatalthough foregut fermentation is not 100% efficient (6-10% are lost viamethane; Van Soest, 1994) a large extent of the digestion process isdone by the microbes which potentially saves energy for the animal
604 A. Riek, J. Bruggeman / Comparative Biochemistry and Physiology, Part A 164 (2013) 598–604
and thus reduces overall energy costs for digestion, translating to someextent into lower FMRs.
Comparing our findings onmarsupials with published results fromplacentals (White et al., 2009; Capellini et al., 2010) shows that afterphylogenetic correction marsupial FMR seems to scale differentlywith BM compared to placentals (0.69 vs. 0.74). Importantly howev-er, in our study, unlike in previous analysis (Nagy et al., 1999; Whiteet al., 2009; Capellini et al., 2010), we included most of the Australianmarsupials, including 95 species for which only BM was known. De-spite lacking information on FMR – the trait of interest – inclusionof these incomplete rows has distinct benefits. Foremost, these con-tribute by fine-tuning the estimate for the joint phylogenetic signaland the allometric slope. Subsequently, the added BM values enablemore reliable reconstruction of FMR values for Australian marsupialspecies that have not been measured yet (n=95) or are impossibleto measure (e.g., FMR of evolutionary ancestors). Thus, the predictionof energy expenditure under free-ranging conditions, as measured byFMR, could help contributing to the conservation and management ofAustralian marsupials. Specifically, predictions could be used to esti-mate the impact of different population sizes on resources (i.e. energyavailability) in certain habitats. Furthermore, knowing the energy ex-penditure of endangered or locally extinct marsupials could help inestablishing captive populations for future release (Lapidge andMunn, 2012).
Acknowledgements
The work of A.R. was supported by a research grant from theAlexander von Humboldt Foundation, Germany. The work of J.B.was supported by the Netherlands Organisation for Scientific Re-search (NWO) through the Rubicon program, grant 825.09.018, andby a Junior Research Fellowship at St. John's College, Oxford.
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.cbpa.2013.01.007.
References
Bininda-Emonds, O.R.P., Cardillo, M., Jones, K.E., MacPhee, R.D.E., Beck, R.M.D., Grenyer,R., Price, S.A., et al., 2007. The delayed rise of present-day mammals. Nature 446,507–512.
Bininda-Emonds, O.R.P., Cardillo, M., Jones, K.E., MacPhee, R.D.E., Beck, R.M.D., Grenyer,R., Price, S.A., et al., 2008. The delayed rise of present-day mammals. Nature 456,274.
Bruggeman, J., Heringa, J., Brandt, B.W., 2009. PhyloPars: estimation of missing param-eter values using phylogeny. Nucleic Acids Res. 37, W179–W184.
Butler, P.J., Green, J.A., Boyd, I.L., Speakman, J.R., 2004. Measuring metabolic rate in thefield: the pros and cons of the doubly labelled water and heart rate methods. Funct.Ecol. 18, 168–183.
Capellini, I., Venditti, C., Barton, R.A., 2010. Phylogeny and metabolic scaling in mam-mals. Ecology 91, 2783–2793.
Carbone, C., Teacher, A., Rowcliffe, J.M., 2007. The costs of carnivory. PLoS Biol. 5, e22.
Dawson, T.J., Taylor, C.R., 1973. Energetic cost of locomotion in kangaroos. Nature 246,313–314.
Dempster, A.P., Laird, N.M., Rubin, D.B., 1977. Maximum likelihood from incompletedata via the EM algorithm. J. R. Stat. Soc. B 39, 1–38.
Felsenstein, J., 1988. Phylogenies and quantitative characters. Ann. Rev. Ecol. Syst. 19,445–471.
Felsenstein, J., 2008a. Comparative methods with sampling error and within-speciesvariation: Contrasts revisited and revised. Am. Nat. 171, 713–725.
Felsenstein, J., 2008b. PHYLIP (Phylogeny Inference Package) version 3.68. Distributedby the author. Department of Genome Sciences, University of Washington, Seattle.
Freckleton, R.P., Harvey, P.H., Pagel, M., 2002. Phylogenetic analysis and comparativedata: a test and review of evidence. Am. Nat. 160, 712–726.
Garland, T., Ives, A.R., 2000. Using the past to predict the present: confidence intervalsfor regression equations in phylogenetic comparative methods. Am. Nat. 155,346–364.
Garland, T., Dickerman, A.W., Janis, C.M., Jones, J.A., 1993. Phylogenetic analysis of co-variance by computer-simulation. Syst. Biol. 42, 265–292.
Hume, I.D., 1999. Marsupial nutrition. Cambridge University Press, Cambridge, UK.Hume, I.D., Gibson, L.A., Lapidge, S.J., 2004. Conservation of Australian arid-zone mar-
supials. In: Gordon, M.S., Bartol, S.M. (Eds.), Experimental approaches to conserva-tion biology. University of California Press, Berkely, pp. 122–131.
Ives, A.R., Midford, P.E., Garland, T., 2007. Within-species variation and measurementerror in phylogenetic comparative methods. Syst. Biol. 56, 252–270.
Johnson, N.L., Kotz, S., 1972. Distributions in statistics: continuous multivariate distri-butions. Wiley, New York.
Koteja, P., 1991. On the relation between basal and field metabolic rates in birds andmammals. Funct. Ecol. 5, 56–64.
Lapidge, S.J., Munn, A.J., 2012. Seasonal field metabolic rate and water influx of captive-bred reintroduced yellow-footed rock-wallabies (Petrogale xanthopus celeris). Aust.J. Zool. 59, 400–406.
Lifson, N., Gordon, G.B., McClintock, R., 1955. Measurement of total carbon dioxide pro-duction by means of D2
18O. J. Appl. Physiol. 7, 704–710.Nagy, K.A., 1987. Field metabolic rate and food requirement scaling in mammals and
birds. Ecol. Monogr. 57, 111–128.Nagy, K.A., 1994. Field bioenergetics of mammals - what determines field metabolic
rates. Aust. J. Zool. 42, 43–53.Nagy, K.A., 2005. Field metabolic rate and body size. J. Exp. Biol. 208, 1621–1625.Nagy, K.A., Girard, I.A., Brown, T.K., 1999. Energetics of free-ranging mammals, reptiles,
and birds. Annu. Rev. Nutr. 19, 247–277.Packard, G.C., Boardman, T.J., 2009. A comparison of methods for fitting allometric
equations to field metabolic rates of animals. J. Comp. Physiol. B 179, 175–182.Pagel, M., 1997. Inferring evolutionary processes from phylogenies. Zool. Scr. 26,
331–348.Pagel, M., 1999. Inferring the historical patterns of biological evolution. Nature 401,
877–884.Pagel, M., Meade, A., Barker, D., 2004. Bayesian estimation of ancestral character states
on phylogenies. Syst. Biol. 53, 948–957.Pinheiro, J.C., Bates, D.M., 1996. Unconstrained parametrizations for variance-covariance
matrices. Stat. Comp. 6, 289–296.Riek, A., van der Sluijs, L., Gerken, M., 2007. Measuring the energy expenditure and
water flux in free-ranging alpacas (Lama pacos) in the Peruvian Andes using thedoubly labelled water technique. J. Exp. Zool. A 307, 667–675.
Speakman, J.R., 1997. Doubly labelled water: Theory and practice. Chapman and Hall,London.
Speakman, J.R., 2000. The cost of living: field metabolic rates of small mammals. Adv.Ecol. Res. 30, 177–297.
Strahan, R., van Dyck, S., 2008. The Mammals of Australia. New Holland Publ., Sydney,Australia.
Tyndale-Biscoe, H., 2005. Life of marsupials. CSIRO Publishing, Collingwood, Australia.Van Soest, P.J., 1994. Nutritional Ecology of the Ruminant. Cornell University Press, Cornell,
