A total of 110 chemical immobilization procedures using Telazol (1:1 mixture of tiletamine and zolazepam) were done on adult female Weddell seals as part of a study on maternal energy expenditure and lactation energetics. One hundred and six of these were by IV injection and 4 were by IM injection. Some individuals were immobilized more than once during the course of their lactation period (5–6 weeks), but no individuals were immobilized more than three times. Females were caught on the sea ice at Erebus Bay, Antarctica (77° 51' S, 166° 45' E) during the breeding season (October to December) of 2002 and 2003. Individuals were identified by flipper tags attached in previous years as part of a long-term tagging study , and ages ranged from 6 to 22 years old.
Females were easily approached on the ice and pups were relocated several metres away to avoid potential injury. Subsequently, a canvas bag was placed over the female's head , after which the majority of individuals remained in a prone position without struggle. The few that were slightly agitated would commence a 'rolling' behaviour and could not be restrained effectively on the ice. However, this behaviour typically ceased within 2–3 minutes. Females were then injected with Telazol intravenously via the extra-dural vein in the lumbar region  using a 5 ml syringe and 15 cm (6") 18G spinal needle, or intramuscularly in the rear flank with a 10 ml syringe and 9 cm (3.5") 18G needle. We attempted to give dosages of 0.5 mg/kg  and 0.75 mg/kg  IV and IM, respectively. Dosages at the first capture were calculated using an estimate of female body weight based on researcher's previous experience working with phocids. For additional captures, dosages were calculated by estimating mass loss rates through lactation. Drug induction and recovery times were recorded and the respiratory rate and volume of air moving (as estimated by listening to breathing sounds) were monitored throughout procedures. Induction time (seconds) was defined as the time from injection until the animal did not respond to a tap on the nose . The recovery time (minutes) was defined as the time from immobilization until the seal responded to a tap on the nose by moving and raising its head and maintaining its head in a raised position for ~10 seconds . This was repeated several times to ensure complete recovery. An endotracheal tube, oxygen, doxapram hydrochloride (Dopram®, Wyeth, Baulkham Hills, Australia) and flumazenil (Anexate®, Roche, Castle Hill, Australia) were available in the event of respiratory arrest.
After immobilization, females were weighed to the nearest 1 kg using electronic scales and standard body length and six girth measurements [G1-G6, 6] were recorded. The precise dosages of tiletamine:zolazepam were calculated for each female based on measured weights. Body composition (i.e., proportion of lipid and lean tissues) was measured using an isotopically labelled water technique. A 10 ml blood sample was collected to measure background isotope levels followed by the IV injection of a pre-weighed dose (to the nearest 0.1 mg) of 222 MBq tritiated water (HTO) into the extradural vein. The syringe was flushed with blood twice to ensure complete isotope delivery. A second blood sample (10 ml) was taken approximately 150 minutes after initial injection for the calculation of dilution space and body composition. Houser & Costa  found that HTO equilibration occurs within 90 minutes of an intravenous injection of northern elephant seal (Mirounga angustirostris) pups. Equilibration occurs in southern elephant seal (M. leonina) pups within 120 minutes of administration (IV; K.E. Wheatley, unpublished data). Therefore, we considered 150 minutes to be sufficient time before collecting a second blood sample. All samples were stored at -20°C until analysis.
Plasma samples were analysed for HTO activity using liquid scintillation spectrometry. Plasma samples (100 μl) were distilled in triplicate using the method of Ortiz et al. . For each vial of water recovered, 4 ml of EcoLite scintillate (ICN, Costa Mesa, USA) was added and HTO activity was counted for 15 minutes using a Beckman LS6500 scintillation counter. Correction for quenching was made by automatic external standardization. Calculations of body composition were done as described by Reilly & Fedak .
Data analysis and calculations
We did not obtain body composition data for 37 captures, but for 11 of these animals we obtained composition data for captures before and after the capture in question. Body composition for this intermediate capture was estimated by interpolation, assuming the change in composition was linearly proportional to a change in mass.
A set of generalized linear models (GLM) and penalized quasi-likelihood [PQL, 32] generalized linear mixed-effects models (GLMM) were constructed to examine the relationships between recovery and induction time and the various state variables. GLMs extend the standard regression model by (1) distributing the response y about its expected value μ according to a distribution F (e.g., normal, gamma, binomial, etc.), and (2) entering the predictors x1, x2,., x
into the model through the linear predictor η, which is related to the expected response μ by a monotonic link function η
) . GLMMs are linear models that include both fixed and random effects, where random effects are those associated with individual experimental units drawn at random from a population [e.g., individuals as in this study, 34]. GLMMs offer the advantage of partitioning variances due to the effects under investigation (fixed) and those that do not contribute to the hypotheses being tested (random).
Model comparison used Kullback-Leibler information to assign relative strength of evidence (Akaike's Information Criterion corrected for small samples [AICc, 26, 35]) to each model in the set . To compare a more complex model a to a simpler model b, we employed the information-theoretic evidence ratio (ER = AIC
weight of model a ÷ AIC
weight of model b) to quantify the relative support of a versus b, and used the per cent deviance explained (%DE) to determine structural goodness-of-fit of model a (test for model adequacy). Higher ER values indicate higher likelihoods of the tested model relative to model b (e.g., the null model).
The weights of evidence (w+
) for each predictor were calculated by summing the model AIC
) over all models in which each term appeared. However, the w+
values are relative, not absolute because they will be > 0 even if the predictor has no contextual explanatory importance . Therefore, a baseline for comparing relative w+
across predictors is required to ascertain which predictors are relevant. We randomized the data for each predictor separately within the dataset, re-calculated w+
, and repeated this procedure 100 times for each predictor. The median of this new randomized w+
distribution for each predictor was taken as the baseline (null) value (w+0). For each term the relative weight of evidence (Δw+) was obtained by subtracting w+0 from w+
. Predictors with Δw+ of zero or less have essentially no explanatory power .
To account for repeated captures (measurements), a series of GLMMs were constructed to examine relationships between induction and weighted recovery times and the age, total body lipid, stage of lactation and total number of captures. Examination of the residuals for the GLMMs determined that the gamma error distribution family and an identity link function were the most appropriate for each analysis. All statistical analyses were done using the R Package [Ver. 2.0.1, 36]. Values are presented as mean ± one standard error (SE) unless otherwise stated.