Estimating field metabolic rates of pinnipeds: doubly labelled water gets the seal of approval


*Correspondence author. E-mail:


  • 1Measures of the field metabolic rate of marine mammals are extremely difficult to make but they are essential for assessing the impacts of marine mammals on prey populations, and for assessing dive performance in relation to aerobic limits.
  • 2The doubly labelled water (DLW) method is an isotope-based technique for the estimation of the CO2 production, and hence energy expenditure, of free-living animals. Estimates of field metabolic rate (FMR) from DLW in pinnipeds to date are extremely high and at the upper range for most mammals. DLW has previously been validated in pinnipeds but logistical difficulties meant for these validations were less than ideal, and it has been hypothesised that DLW may overestimate FMR in these animals.
  • 3To test this hypothesis, we used DLW and simultaneous open-flow respirometry to determine the daily energy expenditures (DEE) of wild grey seals (Halichoerus grypus) held temporarily in a captive facility, during 4–5 days of simulated foraging at sea. Comparing DEE from DLW and respirometry, we found that DLW predicted DEE accurately (average difference between the two estimates was 0·7% SD = 17%n = 31), but as with validations of other species there was a large range of individual errors (from –39% to +44%).
  • 4The results dispel the doubts surrounding the use of DLW as a field method for estimating energy expenditure in grey seals, and by implication other pinnipeds, and simultaneously open a series of questions about their ability to maintain surprisingly high metabolic rates for protracted periods.
  • 5We make a number of recommendations for future studies of pinniped FMR using DLW. We suggest use of the Speakman two-pool calculation will be most appropriate. Studies should aim for enrichment levels as high as economically feasible but to at least 150 p.p.m. above background for the O2 isotope. Measurement periods should be extended between one and two half-lives (5–10 days for a typical foraging seal). We also encourage the calculation and presentation of estimates of precision in estimates of FMR.
List of abbreviations

Basal metabolic rate


Calculated aerobic dive limit


Daily energy expenditure


Doubly labelled water


Field metabolic rate


Respiratory quotient (ratio of carbon dioxide produced to oxygen consumed).


Rate of carbon dioxide production as calculated by DLW


Total body Energy


Total body water


Total body fat


Total body protein


Rate of carbon dioxide production as calculated by open flow respirometry


Rate of oxygen consumption as calculated by open flow respirometry


The marine ecosystem is increasingly perceived as being under threat from over-exploitation by fisheries activity. The top predators of the marine may be adversely affected by such activity, and/or may exacerbate the effects of fishing on prey species, and consequently come into direct competition with fishing activity. Our understanding of these interactions requires knowledge of the energy flux through marine mammal populations, as this sets their food requirements, which in turn determines their impact on fish stocks and potential competition with fisheries. Knowledge about energy expenditure of marine mammals while at sea is also important in the context of diving physiology and life history. For example, studies of the at-sea metabolic rates of some species of seals and sea lions have shown that they are operating close to their physiological limits (Costa & Gales 2003). Energetic or nutritional stress has been implicated in several marine mammal population declines; therefore, it is essential that reliable and accurate field metabolic rate estimation techniques are available.

There are four main approaches to estimate energy flux in free living animals: (i) a combination of quantification of costs of various activities using indirect calorimetry, and time and activity budget information from the field (e.g. Sparling & Fedak 2004); (ii) correlative techniques that link physiological measurements that can be telemetered from the wild, such as heart rate (HR), to laboratory observations of energy flux determined in the laboratory, typically also using indirect calorimetry (e.g. Bevan, Speakman & Butler 1995; Butler et al. 1992); (iii) measuring the changes in body mass and composition over time to provide an estimate of energy output; however, this is only possible when animals are fasting; and (iv) an isotope washout technique, called the doubly labelled water (DLW) method (Lifson & McClintock 1966; Nagy 1980, 1983; Speakman & Racey 1988; Speakman 1997, 1998), that involves dosing animals with isotopes followed by repeated captures to trace the isotope elimination, and provides an overall measure of energy expenditure between captures (Costa 1987; Reilly & Fedak 1991; Aquarone, Born & Speakman 2006). The DLW technique allows the estimation of CO2 production and hence energy expenditure from the differential elimination of isotopes of hydrogen and oxygen introduced into the body water. The O2 label is eliminated from the body by continuous flux through the body of both water and expired CO2, but the hydrogen label is only eliminated by the water flux. The difference between the two elimination rates is therefore correlated with CO2 production (Lifson, Gordon & McClintock 1955). Multiplying the difference in the gradients of the exponential declines in isotope enrichments over time by the size of the body water pool gives a quantitative estimate of CO2 production, but there are many complexities involved in correcting for differential distribution spaces of the labels and fractionation during elimination (Speakman 1997) providing a number of alternative calculation methods.

These different approaches have been used to estimate metabolic rates and hence food intake in several pinniped species. Unfortunately, the estimates provided by the different methods differ greatly and are currently impossible to reconcile. In particular, there are discrepancies between metabolic rate measurements using the DLW technique and those obtained from time and energy budgets based on respirometry in captive conditions. At-sea field metabolic rates derived using DLW in pinnipeds are generally high, for example, harbour seals (Phoca vitulina) 6× basal metabolic rate (BMR) (Reilly & Fedak 1991) and Antarctic fur seals (Arctocephalus gazella) 4·6–7·4 times BMR (Costa, Croxall & Duck 1989; Arnould, Boyd & Speakman 1996a). These levels are exceptional among mammals more generally where the average field metabolic rate (FMR) is only 2–3× BMR (reviewed in Speakman 2000). In captive studies, such high rates have only been seen where seals have been pushed to their maximal rates of energy expenditure using swim flumes and additional drag from drag cups or weights. Such rates appear to be sustainable aerobically for only very short periods (Davis, Williams & Kooyman 1985; Fedak 1986; Elsner 1987; Williams, Kooyman & Croll 1991; Butler et al. 1992). In contrast, laboratory-based indirect calorimetry estimates of MR at realistic long-term exercise levels, combined with time-energy budgets from wild grey seals (Halichoerus grypus) generate estimates of at-sea FMR between 1·5 and 3× BMR (Sparling 2003).

In other cases where FMR has been estimated from methods other than DLW, for example, direct measurements of O2 consumption in freely diving Weddell seals (Leptonychotes weddellii) under ice (Castellini, Kooyman & Ponganis 1992) and using turnover of deuterium and changes in proximate body composition in free living harbour seals (Bowen, Oftedal & Boness 1992; Coltman et al. 1998), FMR estimates were around 2–3× BMR, that is, consistent with the estimates based on indirect calorimetry.

There are several possible explanations for this apparent mismatch. FMR estimates from wild seals may have only been taken from animals during periods of extremely high metabolic activity. Alternatively it may be difficult to replicate the conditions in the field to get correct estimates of energy demands by respirometry to multiply by field time budgets. Finally, there may be a discrepancy between the two main techniques used to estimate FMR in seals, and previous estimates of FMR for pinnipeds from DLW may have been overestimated.

Speakman (1993) first suggested that DLW might overestimate FMR of seals on the basis that there may be an unaccounted for additional flux of the O2 isotope in the ornithine–arginine cycle that becomes quantitatively significant in obligate carnivores. This view was reinforced after Boyd et al. (1995) used DLW concurrently with respirometry to measure the metabolism of California sea lions (Zalophus californianus) swimming in a flume. Moreover, the only DLW measurements available for large terrestrial carnivores (the African wild dog, Lycaon pictus: Gorman et al. 1998) have similar high levels of expenditure to the seals.

However, the overestimate by DLW compared to respirometry in the Boyd et al. (1995) study is compromised by the short duration of the measurement which meant the precision of the DLW method was poor. Although Costa (1987) validated the DLW method against food intake in a single non-swimming northern fur seal (Callorhinus ursinus) and found a reasonable correspondence this situation involved activities divorced from those representative of most FMR measurements on pinnipeds. In addition, validation of the method in domestic dogs (Speakman et al. 2001) suggested no gross overestimate by the DLW method relative to respirometry, indicating the flux of O2 in the ornithine–arginine cycle was not an issue.

This mismatch has direct management implications. What estimate of energy flux and hence food consumption should be used in fisheries and ecosystem models? As the DLW estimates are consistently about twice those derived from the alternative laboratory-based approach, it is clear that choice of method could lead to a factor of two differences in predicted food consumption. At present we do not know if this reflects real differences in metabolism between the laboratory and field, or if it is the consequence of some problem with the DLW method in its application to marine mammals.

Logistical problems involved have hitherto precluded a simultaneous cross validation of methods for marine mammal species in anything approaching realistic conditions. Here, for the first time we present the results from a large-scale validation of the DLW method for use in pinnipeds over realistic time-scales and foraging and diving schedules. In this study, we tested the hypothesis that the DLW method overestimates the metabolism of free-living seals.


We measured FMR of temporarily captive wild grey seals within a purpose-built simulated foraging setup, which also functioned as a respirometry chamber. All seals were caught in the wild, from local haul-out sites and taken by boat to the captive facility of the Sea Mammal Research Unit in St Andrews. Seals were released back into the wild after a maximum period of 1 year. While at SMRU, the animals were housed in outdoor seawater pools at ambient temperature and fed a diet of herring (Clupea harengus), mackerel (Scomber scombus) and sprat (Sprattus sprattus) supplemented with vitamins (Aquavits, International Zoo Vet Group, Keighley, UK).

We simulated 5-day long foraging trips, whereby the seal was ‘at sea’ for 4–5 days, alternating periods of foraging (diving between surface and underwater feeding device to feed) with periods of rest. During this time seals were kept in a large pool (40 × 6 × 2·5 m) with the surface covered by panels of aluminium mesh preventing the seal from surfacing anywhere other than a clear acrylic breathing chamber situated in one corner of the pool. This allowed for continual measurement of gas exchange. Each animal underwent three or four of these simulated foraging trips, each measurement consisting of a range of workloads and food intakes. Workload was manipulated by altering the schedules of feeding dives during simulated foraging. Because of the difficulty of serial capture and blood sampling (also likely to be a constraint for studies on wild seals) we used the two-sample approach where only a background, initial (equilibration) and final isotope determination is made to track the turnover of both isotopes over the experimental period.

At the beginning of each week a seal was taken into an enclosure and weighed on a platform scale (Avery, ± 0·1 kg). It was then sedated using an intravenous injection (dose of 0·005 mL kg−1) of zoletil (Virbac, France). A blood sample (10 mL) was taken from the extradural vein for determination of isotope background levels (Speakman & Racey 1987: method D), before intravenous injection of a weighted dose of isotopically labelled water containing 2H and 18O. The syringe used to inject the DLW was weighed empty, and then containing the DLW. On injection, the syringe was flushed fully four times to ensure that all DLW had been injected into the seal. The seal was then left in the enclosure with no access to food or water for a 3-h equilibration period, after which a further 10-mL blood sample was taken. Immediately after this second blood sample, the seal went into the main pool and the aluminium mesh panels were closed down. The panels remained closed until the end of the trial, 4–5 days later. During this entire time the only place that the seal could surface was the breathing box. At the end of the 5-day period the seal was taken out of the main pool set up into an enclosure where they were weighed and on a subset of individuals, a final isotope dilution body composition was carried out. Blood tubes were centrifuged immediately after collection at 1000 g for 15 min and then plasma (50 µL) samples were drawn-off and the flame was sealed into Vitrex precalibrated pipettes (Camlab, UK) for analysis.

simulation of foraging

The simulated foraging set up is described in detail in Sparling et al. (2007). Briefly, seals were trained to swim from the breathing box (the surface) to an automatic feeding device (prey patch), situated 80 m away from the breathing box. The feeder is an aluminium box that houses a conveyor belt driven by an electric motor. The experimenter placed fish into slots on the belt that were then presented to the seal at an underwater window. A video camera was mounted above this underwater window recorded the seals’ presence at the feeder. Fish was presented on the feeder over a series of dives where prey encounter rate (PER) remained constant within a given dive, but varied between dives.


O2 consumption and CO2 production during the entire 5-day trial was measured using an open-flow respirometry system connected to the breathing box. The breathing chamber had an inlet, which opened to the outside, and an outlet, which was connected by 1·5–in. diameter flexible hosing to a pump, situated inside the laboratory (c. 6 m away). Another section of this flexible hose, 1·5-m long, was attached to the inlet, acting as ‘dead space’ so that none of the seals’ expirations were lost through the inlet. Ambient air was drawn through the box at a rate depending on the animals’ requirements (200–400 L/min), sufficient to make the change in O2 concentration during breathing around 1% and to prevent a build-up of CO2 levels. Flow was maintained and monitored using Sable Systems Flow Kit 500H (Sable Systems International, Las Vegas, NV). A 500-mL min−1 subsample was pumped at positive pressure through a drying column, and then to a CO2 analyser (Sable Systems CA10a, accuracy 1%, resolution 0·001%, zero drift) and an O2 analyser (Sable Systems FC-10a, accuracy 0·1%, resolution 0·0001%). Outputs of the gas analysers (and the flow rates from the FlowKit) were connected to the serial ports of a desktop PC. The PC stored fractional O2 and CO2 concentration, pressure, temperature and flow rate with a time stamp once per second. Analyser drift was minimal, but if it occurred it was corrected for during data analysis. The system had a lag of c. 30–40 s from when the seals began breathing (or bleeding nitrogen gas into the chamber) until the first deflection on the O2 analyser, and a 95% response time of c. 1·5 min. CO2 analyser was calibrated daily using a 5% CO2 in nitrogen mix (BOC gases). O2 analyser was calibrated daily using ambient air and oxygen free nitrogen gas (Fedak, Rome & Seeherman 1981). Rates of O2 consumption inline image and CO2 production inline image were calculated using the following equations:


where V is the flow rate at STPD, inline image and inline image and CO2 fractions of the air before it passed by the seal and inline image and inline image were the fractions after it had passed the seal (Arch et al. 2006).

isotope sample analysis

Samples of plasma in capillaries were vacuum distilled into glass Pasteur pipettes (Nagy 1983) and the water obtained used for isotope-ratio mass spectrometric analysis of 2H and 18O. The 2H analysis was performed on hydrogen gas, produced by on-line chromium reduction of water (Morrison et al. 2001; Speakman & Krol 2005).

For analysis of 18O enrichment in blood samples, water distilled from blood was equilibrated with CO2 gas using the small sample equilibration technique (Speakman et al. 1990). For analysis of 18O : 16O ratios, equilibrated water samples were admitted to an ISOCHROM µ GAS system (Micromass, UK), which uses a gas chromatograph column to separate nitrogen and CO2 in a stream of helium gas before analysis by IRMS.

For estimation of the injectate enrichment, the original injectate was diluted with tap water (five different solutions, ± 0·0001 g), in proportions similar to those expected in the seals. Mass spectrometric analysis of 2H and 18O was performed on five subsamples of each solution and five subsamples of tap water. The enrichment of the injectate was calculated for the five different solutions (Prentice 1990; Speakman 1997), and then averaged. We used isotopically characterized gases of H2 and CO2 (CP grade gases, BOC Ltd) in the reference channels of the IRMSs. Reference gases were characterized every 3 months relative to SMOW and SLAP (Craig 1961) supplied by the IAEA. Each batch of samples was run adjacent to triplicates of three laboratory standards to correct for day-to-day differences in mass spectrometer performance and inlet cross-contamination (Meijer, Neubert & Visser 2000). All isotope enrichments were measured in δ per mille relative to the working standards and converted to p.p.m., using the established ratios for these reference materials. The measures of isotope enrichment in blood samples were based on analysis of five subsamples (2H) or two subsamples (18O); all subsequent calculations were performed on the mean values.

calculating energy expenditure and body composition with dlw

CO2 production was calculated for each trial using several different published models each making different assumptions about fractionation of isotopes, evaporative water loss and different combinations of body water pool estimates. Each model is described briefly below but for a full discussion of the different calculations and their assumptions see Speakman (1997). The Lifson & McClintock (1966) model (eqn 7·3 in Speakman 1997) uses only the O2 dilution space to estimate of the body water pool size. Fractionation factors were derived at 25 °C and evaporation is assumed to be 50% of water loss. Nagy (1980; eqn 17·13) is a simpler calculation ignoring the fractionation corrections. Speakman (1997) derived a new equation (eqn 7·17), which assumes only 25% of water loss is fractionated and fractionation factors reflecting a mix of kinetic and equilibrium processes at 37 °C.

These models are collectively called one-pool models and make the assumption that the oxygen and hydrogen dilution spaces are the same. However, hydrogen partakes in other exchange reactions in the body and thus spreads to a slightly larger pool than oxygen (Sheng & Huggins 1971; Culebras & Moore 1977). Coward, Prentice & Murgatroyd (1985; eqn 7·34) suggested that the calculation should involve the elimination rate (ko and kd) and for each isotope multiplied by their respective dilution spaces (No and Nd). Models such as this, utilizing both dilution spaces are called two-pool models. Schoeller et al. (1986) devised a two-pool model with revised fractionated factors (eqn 7·40 in Speakman 1997). Speakman (1987, 1993) showed that in theory two-pool models will be more appropriate in larger animals (above 10 kg). Speakman (1997; eqn 7·43) derived a two-pool model equivalent in all other respects to the one-pool model represented by eqn 7·17.

Initial isotope dilution spaces (mol) were calculated using the plateau method (Halliday & Miller 1977) and were converted to millilitres assuming a molecular weight of body water of 18·020. This technique has been previously validated against chemical analysis in seals (Arnould, Boyd & Speakman 1996b). Final dilution spaces were measured in 16 of the 32 trials and were estimated for the remaining 16, from final body mass, assuming the same percentage of body mass as measured for the initial dilution spaces (average difference between body water as percentage of body mass between initial and final across the 16 trials where both were measured was < 1%). CO2 production rates from both respirometry and DLW were converted into energy expenditure using energy equivalents calculated from the measured respiratory quotient (RQ) using the Weir equation (1949).

calculation of uncertainty

We used a Monte Carlo simulation approach to incorporate uncertainty in all input parameters to calculate confidence limits for our body composition and daily energy expenditure (DEE) calculations (after Speakman 1995).

DLW estimates

To compute one estimate of DEEdlw, we'd generally use the mean isotope values from four replicate analyses. These replicate analyses gave us an indication of the analytical variability. For each isotope datum, four values were drawn at random from a distribution described by the mean of the replicate analyses for that sample and the standard deviation either (i) calculated as the average standard deviation across isotope samples from all 3 years of the study or (ii) the average standard deviation from the year the samples were from (see Table 1). The mean of the four resulting values were then used to calculate the rate of carbon dioxide production (rCO2). Repeating this process 1000 times gives a distribution of DEE estimates from which we can calculate confidence limits (Speakman 1995).

Table 1.  Variation in replicate hydrogen (H) and oxygen (O) isotope analyses in the 3 years of the study
 Mean number of replicate analysesMean standard deviation of replicatesMean standard error

Respirometry estimates

We used the manufacturer's accuracy levels to describe uncertainty in each parameter in the equations for calculating the rate of O2 consumption (VO2) and the rate of CO2 production (VCO2). For a particular experiment, each iteration of the calculation drew each parameter from a distribution described by the mean value of that parameter from that experiment and a standard deviation which was calculated using the manufactures quoted accuracy estimate as a co-efficient of variation (these were 1% of full scale measurement for measurement of FCO2 and flow rate, and 0·1% for measurement of FO2).

For comparison, we also carried out empirical tests using gas dilutions where we bled a mixture of 5% CO2 in nitrogen gas into the breathing chamber at a measured rate. This gas affected the downstream gas fractions in a similar manner to the respiratory gas exchange of an animal – i.e. elevated CO2 fraction and depleted O2 fraction. Expected values of VO2 and VCO2 were compared to values derived from the downstream respirometer flowmeter and gas analysis after the calibration gas stream was diluted and mixed into the respirometer airstream at the empty animal chamber.

Body composition changes

We used the same approach as for DLW to calculate error in dilution spaces from variation in isotope analysis. From these we also calculated error in our body composition measures. Because TBF and TBP are estimated using TBW as a percentage of body mass, to estimate error in fat and protein we had to incorporate a value for uncertainty in our measurements of body mass. Uncertainty in body mass is likely to be a result of differences in wetness of animal, amount of food in digestive tract as well as variability in the scales. We carried out repeat weighing of animals at different stages of fur wetness and at different times of day and estimated a CV of 1·5%. So for the Monte Carlo simulations, the standard deviation was set at 1·5% of measured mass.


We carried out 32 comparisons of DLW to respirometry over 3 years, using nine female grey seals (three adults and six juveniles < 1 year). The measurement period ranged from 96·5 to 120 h. During the comparisons, seals swam 70 ± 1·8 km. For full details of all trials see Supplementary Table S1.

isotope dilution

Dilution space ratios were between 0·975 and 1·078, with an overall mean of 1·022 (see Supplementary Table S2). TBW of the seals at the start of each measurement ranged from 46% to 65% of total body mass (mean 53% ± 4). We estimated precision of absolute TBW to be 0·3%. Expressed as a percentage of total body mass, however, and therefore including variability due to imprecision in our estimates of body mass, precision increases to 1·63%. As fat and protein content are estimated using TBW%, precision of these values were 5%–7%. In addition, because changes in body composition during each experiment rely on two measurements of TBW% the error in Δfat are very large (Table 2).

Table 2.  Estimates of precision in various stages of estimation of body composition changes
 Mean valueMean CV%
Nd (mls)32 104  0·27
TBW (mls)30 982  0·27
TBW (% body mass)    53  1·63
TBF (kg)    16·6  6·7
TBF (% body mass)    27·4  5·2
ΔFat (kg)     1·45268

daily energy expenditure (DEE)

We recorded a range of DEEresp estimates ranging from 8·74 to 26·18 MJ day−1 (Fig. 1). These values, along with corresponding DEEdlw estimates and the ratio between the two estimates are detailed in Supplementary Table S3. Adult values averaged 20·10 ± 4·2 MJ day−1 and juveniles averaged at 11·23 ± 1·7 MJ day−1. These values correspond to 1·9 and 2·6 times predicted basal metabolic rates (Kleiber 1975), and 0·9 and 1·2 times predicted field metabolic rates for mammals of a similar mass (Nagy 2005). DEEdlw estimates covered a similar range from 7·29 to 27·89 MJ day−1.

Figure 1.

Correlation between DEE (daily energy expenditure in MJ d−1) as measured by DLW (Speakman 1997 two-pool model) and by respirometry. The dashed line is the line of unity.

The original method of calculation of rCO2 using DLW advocated by Lifson & McClintock (1966) resulted in a mean overestimate of 0·5% (SD = 17·4). Individual estimates using this method varied from an underestimate of 38% in experiment G3 to an overestimate of 44% in experiment B2. When using the two-pool model by Coward et al (1985; using individual ratios for dilution space) DEE is underestimated on average by 7% (SD = 17·8), individual errors ranged from –42% to +30%. Schoeller et al. (1986; two-pool model using fixed estimate of dilution space ratio of 1·03) generally underestimated DEE. Errors varied between –30% and +37%, with a mean error of –5% (SD = 17·5). The alternative approach of using the observed mean dilution space of 1·02 (Speakman 1997; two-pool model) produced errors of between –39% and +44% with a mean error of 0·5% (SD = 17·5) (Fig. 1). See Supplementary Table S3 for details of all individual model calculations. No significant differences were found between DEEdlw and DEEresp when using the Speakman (1997) two-pool (paired sample t = 1·09, df = 31, P = 0·284), Lifson and McLintock (t = 0·81, P = 0·424) or Speakman (1997) one-pool models (t = –1·55, P = 0·131). However, significant differences were indicated between DEEdlw and DEEresp when using the Nagy (1980) (t = –2·50, P = 0·018) or Schoeller (1986) equations (t = 2·19, P = 0·036). The random scatter of points on a Bland–Altman plot (Fig. 2) demonstrates good agreement between the two methods (respirometry and DLW using the Speakman 1997 two-pool model) (Bland & Altman 1987).

Figure 2.

Bland–Altman plot of differences between two methods.

For all calculations of precision in DLW, we used the results from the Speakman (1997) two-pool model. We calculated an overall precision of our DLW estimates at 7%. Supplementary Table S4 shows the confidence intervals calculated for both estimates of DEE. Confidence intervals overlapped in 75% of all trials. When the mean standard deviation across all isotope analyses (across the three different years of analysis) was used in the simulations, there was a negative relationship between the CV of the DLW estimate and both the initial enrichment above background and the difference between the initial and final enrichments (Fig. 3).

Figure 3.

Variation in precision of calculated DEEdlw with variation in (a) level of initial isotope enrichment in p.p.m. above background levels and (b) the extent of depletion of the initial isotope dose (initial minus final enrichment in p.p.m.). See text for description of precision estimation.

Incorporating the manufacturers estimates of measurement accuracy in respirometry calculations gave an overall CV of 2·5% for our estimates of DEEresp. This compared to a CV of 2·1% calculated by carrying out empirical tests comparing expected and measured volumes of calibration gases flowed through the system. In calculating confidence intervals for our respirometry estimates of DEE, we used the wider estimate of uncertainty.

food intake and mass balance

Mean body mass change over the period of the DLW measurements was +1·7 kg (Table 3), ranging from a decrease in 3·2 kg to an increase in 8·6 kg. Food intake varied greatly between individual trials – between 4·3 and 32 kg during the trials with an average of 13·5 kg. There was a significant correlation between food intake and mass change (R2 = 39·2%, P < 0·0001). We converted food intake into energy intake using the established energy content of the diet depending on fish species consumed (7·6 kJ/g for herring, 7·5 kJ/g for mackerel, 7·9 kJ/g for sprat). For those trials where we performed body composition determinations at the start and end, we calculated the change in body fat and protein, and converted this into changes in energy content (using the values of 39·6 kJ/g for fat and 16·8 kJ/g for protein) and added this to the energy expended to give an estimate of actual energy intake. For all the trials, we also estimated energy intake using four different assumptions about the composition (and thus energy content) of mass change: (i) changes in mass were due to changes only in water content and the energy content of the animal did not change; (ii) 100% of the mass change was fat; (iii) 100% of the mass change was lean tissue; and (iv) the mass lost or gained was identical in composition to the starting body composition.

Table 3.  Mass and body composition changes over each 5-day trial as predicted by isotope dilution
TrialΔMass (kg)ΔFat (kg)ΔWater (kg)ΔProtein (kg)ΔTBE (MJ)
A1 1·2 4·07–1·98–0·891451·7
A2–3·0 4·32–5·16–2·021345·9
A4–3·2 0·83–2·92–1·07 134·2
B1–0·2 1·89–1·48–0·61 636·3
B2 1·6 2·36–0·52–0·29 882·9
D1 1·0 0·94 0·05–0·03 367·2
D2–0·6 0·01–0·46–0·16 –24·7
D3 3·0 1·72 0·96 0·26 726·3
D4 0·0 0·73–0·52–0·22 251·7
E1 3·4 1·99 1·06 0·28 839·5
E2–0·4 0·64–0·74–0·29 201·0
E3 3·8–2·13 4·15 1·56–563·5
E4–1·2 0·43–1·17–0·44  92·6
G1 8·6–2·45 7·82 2·87–453·0
G2 7·6 3·12 3·31 1·031421·4
G3 3·2 3·42–0·04–0·171323·0

Assuming that the mass change consisted of lean mass or water consistently underestimated food intake and assuming fat overestimated suggesting that the body composition of the mass change was a mixture. Assuming that the mass change was of the same proportion as in starting body composition gave the best predictions of energy intake (Fig. 4).

Figure 4.

Predictions of energy intake given different assumptions of the composition of mass change between the start and end of experiment. The different assumptions were: (a) water; (b) lean tissue; (c) fat; (d) same composition as initial; and (e) as estimated by initial and final isotope dilution. Energy Intake (MJ) = Energy Expended (MJ as estimated by DLW) + ΔBody Energy (MJ).


The good correspondence between DEEdlw and DEEresp in this study lends confidence to the utility of the DLW method for estimating the FMR of seals. Most validation studies of the DLW method report good agreement between the mean values averaged over a group of individuals. Taking the 0·5% mean difference over all individual validations, our results are well within the range of previous validation studies in terrestrial mammals, which have an average discrepancy of 2·23% (Speakman 1997).

Precision increased with increasing initial isotope enrichment above background and also increased with an increase in the difference between initial and final enrichments. The latter effect depends on the relative contribution of water turnover and metabolism to washout of the O2 isotope. However, the initial enrichment is within the control of the researcher. These data suggest that while injecting more isotope is more expensive, the payback for this expense is an estimate with greater precision. Figure 3 provides an empirical picture of the likely nature of this trade-off in a free-living seal, which may be useful in the design of future DLW studies of pinnipeds to aid researchers in choosing the necessary injection levels of isotopes to achieve given levels of precision.

In the only other validation of the DLW technique against respirometry in marine mammals Boyd et al. (1995) reported that DLW overestimated DEE of California sea lions (Z. californianus) on average by between 36% and 46% depending on the calculation method employed. However, the duration of this validation was not ideal, depletion of oxygen and hydrogen isotopes was only 14% and 9%, respectively (compared to 46 ± 13% and 38 ± 12% in the present study), and precision in the resulting estimate, therefore, was very low (overall CV of 35% compared to the 7% calculated in this study). This comparison highlights the importance of carrying out DLW trials over long enough periods. Nagy (1983) recommended at least one and preferably two half-lives, and we concur with this recommendation. In seals this would mean a period of between about 5 and 10 days. Our data extend the only other validation of the method over this duration in seals made by Costa (1987) who compared the DLW method to material balance in a single individual on land, and found reasonable correspondence between the techniques. Although the cost of O2 isotopes can limit the amount used for dosing animals, particularly for such large animals as seals, our data support the suggestion from Speakman (1997) that the lowest dose used in large mammals should be at least 150 p.p.m. excess for the O2 isotope. Although it may appear tempting to reduce the isotope dose to enable more animals to be measured, the resultant quality of the data makes this a false economy.

The difference in dilution spaces between the two isotopes, with hydrogen space exceeding O2 space by an average of 2%, is consistent with other studies of mammals (Speakman 1997) and is slightly lower than that found in other studies of marine mammals and humans (Speakman, Nair & Goran 1993).

Although some studies of DLW present estimates of precision in individual metabolic rate determinations (e.g. Corp, Gorman & Speakman 1999), the same cannot be said for most published uses of isotope dilution estimation of body composition. Our estimates of 1·6% and 5·2% CV in estimates of TBW and TBF as percentage of body mass, respectively, are larger than some of the reported changes over the trials highlighting the fact that the precision of the method is not sufficient to detect small changes over short time-scales. Precision in estimates of changes in body composition could be improved by decreasing error in body mass estimation. However, we found that variation in the amount of food seals had in their digestive system had a large effect on mass. Such factors are likely to be difficult to control for in the field situation.

Most previous field applications of the DLW method to pinnipeds have used the Nagy (1983) equation. This equation is a one-pool model equation which makes no correction for fractionation effects. In the current validation, this equation resulted in an overestimate of the simultaneous respirometry by about 13%. Using a two-pool approach, and taking into account fractionation effects both lead to reductions in the DLW estimate. The extent of the reduction compared with the Nagy equation depends on the details of the assumptions made about the relative pool sizes and, with respect to fractionation, the proportion of water lost in fractionated form and the detailed fractionation factors. The Lifson and McClintock equation is also a one-pool equation like the Nagy equation, but it attempts to correct for fractionation. When Lifson and McClintock made their fractionation correction they used the only available fractionation factors at the time, which were for effects at 25 °C. They also assumed that 50% of water loss is fractionated. In the current study using this equation reduced the estimated DEE by DLW by about 15% giving a very good match to the respirometry data. However, one-pool models are probably inappropriate for large animals like seals (Speakman 1987, 1993). The Coward & Prentice (1985) equation combines the Lifson and McClintock assumptions about fractionation with a two-pool approach. This resulted in a decreased estimate of metabolism by DLW of about 20% below that derived by the Nagy equation and 7% lower than the simultaneous respirometry. Schoeller et al. (1986) pointed out that the fractionation assumptions made by Lifson & McClintock (1966) probably result in an over-correction for fractionation effects. The Schoeller et al. (1986) equation makes more realistic assumptions for fractionation, but it combines these with a fixed assumption for the pool size ratio (1·036), which for these seals is too large. Consequently, because these effects cancel out, the Schoeller equation also underestimated the DEE compared to simultaneous respirometry. The correct fractionation effects make the estimate larger, but the excessive pool size ratio reduces the estimate again. The result is an underestimate of the respirometry again by 5%. The Speakman two-pool model uses the observed pools size ratio and therefore overcomes the problem of making a fixed assumption, as in the Schoeller equation, and it also makes more realistic estimates of the fractionation effects. This equation also gives an answer that is almost identical to the simultaneous respirometry.

Although on the face of it there appears little to choose between the Lifson and McClintock, and Speakman two-pool equations, we would argue that the latter equation is a better choice. This is because the Lifson and McClintock equation reaches the correct answer only because its faults in this situation cancel each other. That is, it ignores the two-pool effect, but overestimates fractionation effects. This fortuitous cancelling of errors may not occur in all circumstances. Accordingly using the Speakman two-pool model will be more robust to variations in the conditions under which measurements are made. As expected the Speakman one-pool model gives an estimate that is too high because it makes realistic fractionation assumptions but does not account for the two-pool effect.

Interestingly, the difference between the Nagy and Speakman two-pool equations of 12% in this study is very close to the difference of 14% found in studies of Australian sea lions (Costa & Gales 2003) and New Zealand sea lions (Costa & Gales 2000).

Our results do not support the hypothesis that DLW significantly overestimates the metabolic rates of pinnipeds (Speakman 1993; Boyd et al. 1995). This suggests that field metabolic rates of pinnipeds have been measured during periods of very high energy expenditure (e.g. Reilly & Fedak 1991; Costa & Gales 2000, 2003; Aquarone et al. 2006). These high rates of energy expenditure have implications for the impact of pinniped populations on their prey and open a series of questions about their ability to maintain surprisingly high metabolic rates for protracted periods.

In conclusion, our validation of the DLW method in seals was carried out under conditions and over durations comparable to those encountered during field applications of the method. The use of either the Speakman (1997) two-pool model or the Lifson & McClintock (1966) one-pool model yielded the best estimates of DEE. We suggest use of the Speakman two-pool calculation will be most appropriate for future studies. We recommend that future studies of large mammal FMR using DLW aim for enrichment levels as high as economically feasible but to at least 150 p.p.m. above background for the O2 isotope (as suggested by Speakman 1997). Measurement periods should be extended to between one and two half-lives as recommended by Nagy (1983) (5–10 days for a typical foraging seal). This has the added advantage that animals are more likely to be in energy balance over the measurement period if the measure is protracted (Speakman et al. 1994). We encourage the calculation of precision using the variability in input parameters and endorse the recommendation of Speakman (1997) that raw data from such studies should be presented (in appendices or electronic supplements) to allow easy comparison with future studies. We also conclude that, where the body composition changes are likely to be small relative to the uncertainty in isotope determination of TBW, TBF and TBP, there is nothing to gain from performing isotope dilution both at the start and the end of the DLW trial and equally satisfactory results can be gained from assuming that the composition of any mass change is the same as the starting body composition.


S. Moss provided invaluable assistance with animal handling; A. Zollinger and S. Brando helped with simulated foraging trials; P. Thomson and P. Redman carried out the isotope analysis. The Natural Environment Research Council UK provided funding for this work (NER/D/S/2003/00650).