Metabolic adjustments to increasing foraging costs of starlings in a closed economy

Fluctuations in food supply are the rule in nature, and animals have evolved numerous adaptations to cope with such fluctuations. Knowledge of physiological or metabolic adjustments to variation in foraging success is essential for making predictions about behavioural choices such as when to leave a site in which food availability decreases, or what levels of energy reserves should be maintained in different environments. The response to diminishing foraging conditions has many dimensions: resting metabolic rate may be altered, mass reduced, and consequently, flight and thermoregulation costs may change (Tiebout,1991; Deerenberg et al.,1998; Bautista et al.,1998). Therefore daily energy expenditure (DEE), and thus requirements to remain in energy balance, are not simply proportional to the(foraging) activity.

An approach that has frequently been used to study the physiological consequences of food stress is complete or partial caloric restriction(Daan et al., 1989; Cherel et al., 1994), which by definition results in a decrease in DEE(Fig. 1A). However, when foraging success decreases in the real world, animals that are not sit-and-wait predators have to spend more time and energy foraging for the same amount of food. Everything else remaining equal, DEE is expected to be an accelerating function of foraging costs per reward, because the extra energy spent foraging must also be acquired, which again increases foraging time and energy expenditure, and so on (Fig. 1B). Surprisingly, contrary to this simple prediction,experimental tests found that DEE decreased with increasing foraging costs per reward (Deerenberg et al.,1998; Bautista et al.,1998), thereby superficially resembling the results of caloric restriction experiments. A possible explanation for this counter-intuitive result was suggested by Fotheringham(1998): in his experiments,starlings Sturnus vulgaris L. decreased food intake and body mass with decreasing foraging reward rate when the number of flights needed for a food pellet was fixed (e.g. 20 flights per reward). But when using variable reward rates (rewarding each flight with a reward with e.g. probability 1/20)they maintained food consumption and body mass (albeit that the range of foraging reward rates was small). Since earlier studies of the relationship between DEE and foraging costs per reward used fixed reward rates(Deerenberg et al., 1998; Bautista et al., 1998), this aspect of the experimental design may explain why they found a decrease in DEE with increasing foraging costs. Fotheringham speculated that cognitive processes, such as motivation or memory, caused the differential response to the variable and fixed reward rates. Whatever the mechanisms, since food reward rates experienced by free-living animals will typically be variable,applying variable foraging reward rates in laboratory studies may prove essential when extrapolating the results to natural conditions.

We studied captive starlings to test the prediction that birds increase DEE with increasing foraging costs. We further quantified the (energy) dimensions over which starlings adapt their physiology to the harshness of their environment (mass changes, time and energy budgets, pectoral muscle size). Following Fotheringham (1998)we rewarded foraging effort with a certain probability that the birds will be given a reward after a return flight, thereby creating a variable foraging reward rate. We set the foraging costs per reward at three different levels,referred to as a `rich', `intermediate' and `poor' foraging environment.

Starlings were individually housed in eight indoor flight cages(length×width×height: 5.45 m×0.68 m×0.80 m). Birds earned their food by flying between two perches, 5.00 m apart. Switches on the flight perches were connected to a computer, registering when a bird had landed. The reward (a food pellet) was released into a small tray in front of a feeding perch, which could be reached by hopping 20 cm from the nearby flight perch. The feeding perch was attached to a balance (P-1040 load cell,Tedea-Huntleigh, Cardiff, UK), and body mass data were automatically stored in a computer at 1 s intervals. Communication with the balances was established using a RS232 serial multiport (C218, Moxa Technologies, Taipei, Taiwan). Pellet dispensers were operated by a PLC (DL205, Koyo, Langerak, The Netherlands) and the activity data were stored in a PC.

The L:D cycle was 14:00 h:10:00 h and food could only be obtained during the light period. For practical reasons the light period was from 00:00 h to 14:00 h, and night started and ended with a 10 min period of dim light. Fresh drinking water was always present, and water for bathing was presented 1 day per week. As a source of complementary nutrients, two mealworms were given three times each week, except during the periods of energy intake measurements. Ambient temperatures were 16.3±0.1°C (mean ± s.e.m.) during the night and 17.1±0°C during the day.

Food pellets (Trouvit Europe Eel, Trouw, France) consisted of proteins(44%), fat (30%) and fibre/ash (20.0%), complemented with vitamins and minerals (manufacturer's specifications). Average fresh mass of one pellet was 0.020 g, with water content 4%. Energy content was 24.7 kJ g^-1 dry mass; one pellet therefore contained 474 J.

The experiment was performed with eight males caught in the wild and housed in an outdoor aviary until the experiment. All birds had prior experience with the system. At the start of the experiment all birds experienced a foraging reward rate of 2.0 return flights per pellet (f/p; the `rich' environment). After 1 week the foraging reward rate of four birds was gradually decreased to 6.3 f/p over a period of 3 weeks (the `poor' environment). The other four birds remained in the rich environment. One bird kept losing mass when on a rate of 5.6 f/p, and this bird was therefore kept on a foraging reward rate of 5.0 f/p. The birds stayed on these schedules for a further 2 weeks, and during the last days of this period energy expenditure measurements were taken on all eight birds. Next, the rich and poor feeding conditions were gradually changed to the opposite state. In this transition period all birds remained on a foraging reward rate of 4.0 f/p (`intermediate' environment) for 1 week to measure their energy budget. Subsequently, foraging reward rates of the birds formerly in the rich condition was steadily decreased over a period of 4 weeks, until one bird was on a rate of 6.3, one at 5.6 and one at 5.0 f/p. The fourth bird died halfway through the experiment from an unidentified disease and was left out from all analyses. The four birds that were initially in the poor environment were gradually brought to a rich, 2.0 f/p regime.

For basal metabolic rate (BMR) measurements the birds were taken from their cages at the end of the light period and kept for the night in an open air flow system for measuring rates of O₂ consumption and CO₂ production. Starlings are post-absorptive after 60-75 min(Karasov, 1990; Levey and Karasov, 1994). During a measurement, a bird was sitting on a perch inside a dark 24 l Plexiglas box at a temperature of 26.5°C, which is within the thermoneutral zone of starlings (Kendeigh et al., 1977; Biebach, 1979, 1984). The air flow rate was controlled by mass-flow controllers (5850S, Brooks, Rijswijk, The Netherlands), that were calibrated with a bubble flow meter(Levy, 1964), set to deliver 40.0 l h^-1. In- and out-flowing air was dried by passing through a molecular sieve (3Å, Merck, Darmstadt, Germany). Gas analysis was done using a paramagnetic O₂-analyser (Servomex Xentra 4100,Crowborough, UK) and CO₂-analyser (Servomex 1440). The system was calibrated before each measurement session using two 3-digit precision gas mixtures of 20.0% O₂/0% CO₂ and 21.0% O₂/1.0%CO₂ in N₂. Measurements were recorded at 9 min intervals. The rate of oxygen consumption was calculated from these measurements and converted to the energy equivalent, while correcting for the respiratory quotient, according to Brody(1945). BMR was taken to be the minimum value of a 30 min running mean. Body mass was measured before and after respirometry. Because the temperatures in the flight cages (on average 16.8°C) were within the thermoneutral zone(Kendeigh et al., 1977),respirometer measurements could be applied for night-time energy expenditure(E_night, kJ 10 h^-1) estimates in the flight cages without temperature corrections.

To estimate the metabolic rate of the starlings during the day when not flying, a separate series of trials was undertaken in which a different group of starlings was kept in smaller cages measuring 40 cm×80 cm×40 cm(height×width×depth), precluding flight activities. Eighteen birds had ad libitum food, and an additional six birds were restricted to 70% of the ad libitum food intake per day, a reduction similar to the 73% of food eaten by starlings in a poor environment compared to a rich environment. These individual trials lasted approx. 7 days, at the end of which their metabolic rate was monitored for 24 h by respirometry, while maintaining the light:dark schedule.

DEE was estimated from food consumption, faeces production and mass change. Food consumption was measured by weighing the food in the pellet dispensers at 48 h intervals. All faeces were collected from the cages, and from the plastic sheets covering the floors for this purpose. Faeces that remained on the sheets after initial cleaning were removed with moist towels of known dry mass. Faeces and towels were weighed after drying for 3 days at 70°C. Energy content of dried food samples and of individual faeces samples were measured using a bomb calorimeter (C5000, IKA, Heitersheim, Germany). DEE was calculated from the metabolisable energy intake (MEI) and body mass changes,according to the equation DEE=I-E-P, where I is the gross energy intake, E the energy excreted and P the energy cost of tissue accumulation or energy catabolised from stored tissue, all in kJ day^-1. P was estimated to equal mass change × -18.0 kJ g^-1 by accurately measuring energy budgets of captive starlings that showed mass changes (S. Verhulst,unpublished data; for method, see Masman,1986). The assimilation efficiency was calculated as gross energy intake minus energy content of the faeces divided by gross energy intake.

Body mass was measured automatically to the nearest 0.1 g when birds were on their feeding perch, and with an ordinary balance whenever a bird was handled. In this study we used the average mass during the last hour of the active period. Relative pectoral muscle thickness was measured using a `muscle meter' developed at Max Planck Research Centre for Ornithology (Seewiesen,Germany), which measured, to the nearest 0.1 mm, the distance from the breast surface to a virtual plane perpendicular to the sternum crest, 3.0 mm sideways of the sternum. Three measurements were taken at the location where the sternum protruded furthest from the centre of the body, and the average value was used.

Flight speed was measured early in the morning from video recordings taken from outside the cages through a one-way screen. The time from take-off to landing was measured with a stopwatch for a sample of flights.

All mean values are given ± standard error of the mean(s.e.m.). We analysed data with Generalised Linear Models (GLM) and Generalised Linear Mixed Models (GLMM) using SPSS (v. 12.0, SPSS Inc.), except where otherwise stated. We controlled for individual differences by including individual as fixed (GLM) or random (GLMM) effect in the models.

The number of flights and flight time per day increased fourfold with decreasing food availability (see Table 1 for all summary data). Flight speed in the poor environment was 4% lower than in the rich environment, which contributed to the effect of foraging environment on daily flight time. Food intake increased significantly with decreasing food availability, while assimilation efficiency was independent of foraging environment. Body mass decreased slightly during the measurement periods (-0.22±0.095 g day^-1), but the rate of mass loss within measurement periods was independent of food availability(F_2,12=0.06, P=0.95; using individual slopes within measurement period as dependent variable). Thus birds used on average 3.95±1.72 kJ day^-1 from their body stores. Combining data on food consumption, assimilation efficiency and mass change, we found that DEE increased with decreasing food availability; in the poor environment birds spent 43% more energy compared with the rich environment(Table 1). Time of the treatment, i.e. whether it was the first, second or third measurement series in the experiment, had no effect on DEE(Fig. 4; GLMM: F_1,12=0.01, P=0.9), nor did the order of treatments (i.e. poor or rich first; F_1,5=2.33, P=0.19).

BMR decreased by 20% from the rich to the poor environment(Table 1). Within individual birds, BMR was positively related to mass (F_1,13=35.3, P<0.001) and the slope [logBMR on log(body mass)] was 1.02±0.17 W g^-1. Such a steep slope is in agreement with other within-individual studies (see overview in Battley et al., 2001). Metabolic rate decreased over the course of the night and in intermediate and rich conditions reached minimum values after 6-8 h(Fig. 2). In the poor environment, minimum values were reached earlier, after ±4 h. We tested this by comparing the slopes of metabolic rates (MRs) plotted against time for each measurement session; these slopes decreased with decreasing food availability (Fig. 2; F_2,12=17.6, P<0.001). While BMR decreased by 20% from the rich to the poor environment, E_nightdecreased by 27%. BMR was independent of treatment order (GLMM: F_1,5=0.86, P=0.40) or time of measurement (GLMM: F_1,12=4.19, P=0.063), although the latter effect approached significance.

Total nocturnal mass-specific MR decreased with decreasing food availability in the same way as whole-body MR(Fig. 2). In contrast,mass-specific BMR (BMR_ms) was independent of food availability(Table 1).

When foraging in the poor environment, birds had 15.6±3.1 g lower body mass as compared to the rich environment, and mass in the intermediate environment was intermediate (Table 1). The range in body mass changes was large, from 3.8 to 26.1 g,and one cause of this variation was the time spent in the cages(Fig. 3): 6.8±1.6 g was lost from the first to the last treatment. Note that initial body mass in the groups starting with different treatments was very similar(Fig. 3).

In the poor environment the pectoral muscle was 1.1±0.2 mm thinner than in the rich environment (Table 1), but between the intermediate and poor environment there was no difference (paired t-test, t₆=0.62, P=0.56).

Flight costs of starlings were previously estimated, in the same flight cages, at 20.5±0.93 W (Hambly et al., 2004) using labelled bicarbonate(Speakman and Thomson, 1997; Hambly et al., 2002). If flight costs were independent of food availability, the energy allocated to flight would have quadrupled (E_flight in Table 2A), due to the fourfold increase in flight time. However, flight costs were measured in the rich environment only, and daily flight costs estimated using instantaneous flights costs of 20.5 W did not fit the energy budget, in particular in the poor environment (given the observed time spent flying and a flight cost of 20.5 W the energy expenditure when not flying would have to be substantially lower than BMR, which is impossible; see below for calculations). We therefore made independent estimates of flight costs for all three experimental treatments using the DEE, activity and respirometer measurements. We estimated how much energy was spent on activities other than flight, and calculated flight costs under the assumption that the difference between the non-fly budget and the total energy expenditure was spent on flight.

Nocturnal energy expenditure was measured directly(E_night, Table 1). For non-flying birds during day-time we multiplied MR_night with a scaling factor determined independently using 24 h respirometer measurements on a different group of starlings. These birds stayed at 20.8±0.2°C, were either on ad libitum food or food-rationed, had drinking water available and experienced a L:D cycle of 12 h:12 h. We used respirometer boxes that were small enough to constrain activity, and most of the time was spent sitting (P. Wiersma, unpublished observations). From these measurements we calculated MR_day and MR_night. MR_day was correlated with mass(r=0.44, N=24, P=0.032), but more importantly,MR_day and MR_night were strongly correlated(Fig. 4). We calculated the ratio of MR_day to MR_night (1.62±0.03), and applied this figure to our current MR_night data to predict MR_day. Our estimate can be compared with Aschoff and Pohl's allometric relationship for passerines(Aschoff and Pohl, 1970), on the basis that a ratio of 1.42 for RMR_α/RMR_ρis predicted. The small difference with our value may be explained by the fact that Aschoff and Pohl used birds permanently in the dark while our birds had light during the day, which is likely to result in higher energy expenditure. Using the estimated MR_day in the budget resulted in estimated flight costs of 17.5±0.9 W. Estimates for the rich and intermediate environments did not differ from Hambly et al.'s measured value (one sample t₁₄<1.64, P>0.12; Hambly et al., 2004), but the estimate from the poor environment was significantly lower(t₁₄=2.72, P=0.017). We therefore estimated flight costs in the rich and intermediate environments at 20.5 W, and consider 17.5 W our best estimate for the poor environment.

DEE increased from 154 to 220 kJ day^-1 from the rich to the poor environment (Fig. 5), and the latter is within the range of brood provisioning starlings, i.e. 200-331 kJ day^-1 (Westerterp et al.,1982; Ricklefs and Williams,1984). To our knowledge, this is the first experiment where sustained DEE was elevated by manipulating foraging conditions, providing a potentially valuable new tool to study the physiology of hard work. Note,however, that although DEE increased, we did not observe an accelerated increase in DEE with increasing foraging costs (DEE increased 38.5 kJ from rich to intermediate environment, and 27.9 kJ from intermediate to poor environment). Thus our results only partly match the pattern that we originally predicted (Fig. 1B). This is not surprising, however, since in formulating this prediction we assumed that everything besides foraging costs remained constant, while the starlings demonstrated various energy saving adjustments when food availability decreased.

Our results can be compared with two earlier experiments with starlings foraging in a closed economy. Bautista et al.(1998) used fixed reward rates and found that DEE decreased from 144 kJ to 107 kJ when the environment changed from rich to poor. Fotheringham(1998) used variable reward rates (as did we) and found that DEE was approximately 169 kJ day^-1independent of reward rate [estimated by us from gross food intake of birds with variable reward rates: approx. 19 g day^-1, and energy content and assimilation efficiency from Bautista et al.(1998), who used the same food]. This is comparable to the DEE of our birds in the rich environment(Table 1). The poor environment in our experiment was much harsher than the poorest environment in Fotheringham's experiments: birds in his experiments flew ±4 km in the poorest environment, which is only 13% of the flight distance in the poor environment in our experiment.

In the poor environment BMR was 19.6% lower than in the rich environment(Table 1), resulting in an energy saving of 14.6 kJ day^-1, in qualitative agreement with comparable studies (Deerenberg et al.,1998; Bautista et al.,1998), and the observed effect of exercise (independent of foraging) in zebra finches (Nudds and Bryant, 2001). The energy savings on BMR were probably for a large part due to mass changes because BMR_ms did not differ between foraging environments. This contrasts with the results of lower BMR_ms in the poor environment in related studies(Deerenberg et al., 1998; Bautista et al., 1998). That BMR_ms was constant is remarkable considering the strong effect of food availability on body mass (-20% in the poor environment). Either body composition did not undergo major changes (which seems unlikely given that the birds became lighter, not smaller), or energy was saved in other ways, e.g. by lowering body temperature (T_b), in which case the constant BMR_ms may be coincidence. Hypothermic responses to cope with energy shortage are common in birds (McKechnie and Lovegrove, 2002), which suggests that this may be a plausible explanation. Savings during the entire night were larger than expected on the basis of BMR alone because metabolic rate decreased more quickly during the night when foraging in the poor environment(Fig. 2). This could also be due to an accelerated decrease of T_b in the course of the night.

Although our finding that BMR was lower in the poor environment is consistent with comparable studies, it contrasts with the notion that BMR is adjusted to work load (Gelineo,1964; Arieli et al.,1979; Daan et al.,1989; Speakman and McQueenie,1996; Williams and Tieleman,2000; Speakman and Selman,2003). According to this notion we would have predicted BMR to be higher in the poor environment, because of the increase in daytime energy expenditure this induced. Apparently, birds make different physiological adjustments under different ecological circumstances, but it is not evident what triggers these different responses. One could argue that metabolic rate was somehow constrained in the poor environment, forcing the birds to reduce their BMR. However, this is not consistent with the observation that a decrease in BMR was already present in the intermediate environment, while DEE increased further when birds were foraging in the poor environment. Possible constraints on DEE will be discussed below. The experimental protocol followed by our starlings can be looked upon as a training scheme for endurance training, and we see similarities with studies on human exercise physiology. Westerterp (2001) points out that `novice' trainees for the half-marathon lose body mass and concomitantly lower night-time metabolism. Only professional athletes achieve an increase of BMR at the same mass during training (hinting that a suite of changes are involved). It seems our starlings `acclimate' to a training programme in much the same way as `average' human beings do.

We previously estimated starling flight costs in our system at 20.5 W(Hambly et al., 2004; rich environment only). On the basis of this estimate, our birds would have spent 39.7 and 167.4 kJ day^-1 on flying in the rich and the poor environment respectively (E_flight in Table 2A). However, these estimates are too high to fit the energy budget in the poor environment. Given flight costs of 20.5 W, and the nocturnal energy expenditure as measured, the remaining energy in the poor environment for diurnal metabolic rate when not flying would be 0.87×BMR, which is clearly impossible (see Table 2A for calculations). Given that total and nocturnal energy expenditure can be measured with reasonable precision, this indicates that flight costs in the poor environment must have been lower than 20.5 W. This would in itself not be surprising,given the low mass in the poor environment, and the effect of mass on flight costs (Pennycuick, 1975; Rayner, 1979; but see Kvist et al., 2001). Note,however, that we could not confirm this directly in our experiment, in the sense that we found no treatment effect on our indirect estimates of flight costs (Table 2A). However, on the basis of an (interspecific) allometric equation of flight costs(Nudds and Bryant, 2000), one would predict that flight costs should decline by 17% from the rich to the poor environment. This is very close to our budget results (-15%, from 20.5 W to 17.5 W), although we acknowledge that direct measurements of flight costs in different foraging environments will have to be carried out to ascertain lower flight costs in the poor environment.

In the course of the experiment the birds lost 6.8±1.6 g body mass(Fig. 3). At the same time flight speed increased slightly by 0.12±0.04 m s^-1(Table 1), which may be due to the lowering of body mass. Treatment had no effect on assimilation efficiency,but there was a small increase over time: 0.037±0.012 from the first to the third measurement (Table 1). Treatment order had no effect on any variable except mass-specific BMR, with birds that started in the rich environment having a 0.99±0.36 mW higher BMR_ms. Since the effect was small and statistically weak (P=0.04), suggesting it may be a spurious result,we will not further discuss it.

Compared with free-living birds, mass was exceptionally low when birds foraged in the poor environment (Cramp and Perrins, 1994), suggesting an energy shortage despite the substantial increase in foraging effort and DEE. Energy shortage is also suggested by the accelerated decline in metabolic rate in the course of the night. Given that food availability was in principle unlimited, one could argue that the birds were starving in the midst of plenty. We now examine some hypotheses that could explain why birds did not increase their foraging effort in the poor environment to the level required to maintain the same mass as in the rich environment. To this end we first summarise our findings(Fig. 6) to illustrate the effects of the metabolic adjustments to foraging conditions on the total energy budget. We extended this overview with estimates of the required energy and time budget for the hypothetical case of birds in the poor environment that maintain mass and activity specific metabolic rates as in the rich environment. By lowering mass in the poor environment, and hence maintenance metabolism and probably flight costs, the birds achieved a flight time reduction of 34%, and a 37% reduction in DEE compared with the expected DEE in the absence of such responses (these two percentages differ slightly due to the effect of foraging environment on flight speed; Table 1). Reduced flight costs and maintenance metabolism resulted in approx. 15% reduction in DEE, but the flight time reduction brings about the greatest energy savings, namely approx. 24%. The 27% energy saved during the night gives rise to a further 5% lower DEE.

We now discuss hypotheses that could explain why birds in the poor environment did not maintain body mass as in the rich environment.

The length of the working day (14 h) could be insufficient to collect the food required to maintain mass. This seems unlikely however, because in the poor environment birds still spent only 16.2% of the light period on flying. This increases to 42% of daytime spent on foraging when food handling time is taken into account. Using our energy expenditure measurements(Table 1), we calculated that the flight time should have increased to 25% to maintain mass in the poor environment. Including turning time on the perch and food handling time this would add up to 65% (Fig. 6),which would still leave 35% `free time'. Since there were no other notable time-consuming activities, we conclude that time was not a limiting factor. Energy expenditure can be constrained by extrinsic or intrinsic constraints(Tinbergen and Verhulst,2000), but available foraging time seems the only extrinsic constraint present in our system. Since this does not seem an adequate explanation we now turn to intrinsic constraints.

It has been suggested that sustained energy expenditure is constrained to approximately 4×BMR (Drent and Daan,1980), and birds spent 3.7×BMR in the poor environment,close to this limit. Furthermore, the required DEE to maintain mass in the poor environment (353 kJ; Fig. 6),combined with the BMR in the rich environment, would result in 4.7×BMR. Although this is high, it is still within the range of values observed (Daan et al., 1990). More importantly, DEE in the poor environment was at the lower end of the range of values reported for free-living starlings feeding nestlings(Fig. 5), suggesting that some increase in DEE was certainly possible. We did not notice symptoms indicating exhaustion, such as difficulties in flying or lethargic behaviour, but cannot rule out that such effects would have occurred when birds had increased their foraging effort to the level required to maintain mass. In conclusion, there was at least some scope for an increase in DEE from an energetic perspective,suggesting that metabolic constraints do not explain why mass was not maintained in the poor environment.

Digestive bottlenecks can play an important role in foraging decisions(Kersten and Visser, 1996). Starlings have food retention times of 53-59 min, independent of diet(Levey and Karasov, 1994), and we therefore assumed that hourly intake rates over the light period are equal to, or below, maximum sustainable rates. The average food consumption for the different birds was low at 0.56-0.80 g h^-1, well below the maximum hourly consumption rates (range 0.98-1.4 g h^-1). We therefore conclude that DEE was not constrained by a digestive bottleneck.

When flight costs increase with mass, at some mass level the net energy gain of foraging will become negative, and this could have prevented birds from maintaining high mass in the poor environment. We therefore calculated foraging efficiency and net intake rate in the poor environment for birds with their actual mass and the mass maintained in the rich environment. The effect of body mass (via flight and resting metabolism costs) on these foraging currencies was rather small (Table 3). Both currencies became much less attractive when foraging reward rate decreased, but lowering of flight costs through mass loss did not result in substantial improvement. Certainly, no currency would drop to negative values. There is therefore no indication that effects on foraging currencies constrained mass changes.

Fotheringham (1998) showed that under variable rewards the birds maintained mass and food consumption,but when reward rates were fixed the birds' mass and food consumption decreased when foraging costs increased. We therefore used variable reward rates in our experiments, but cannot exclude the possibility that further modifications in the reward schedule would result in even higher work rates in the poor environment.

Given the absence of conspicuous constraints that prevent birds from maintaining a higher mass and energy expenditure in the poor environment, the response to deteriorating foraging conditions can be considered as an optimality problem, i.e. in terms of fitness consequences of different options. Although we cannot estimate the fitness benefits of maintaining high mass, it seems reasonable to assume that if benefits of high mass vary they are likely to be higher in the poor environment, because the probability that food availability drops below a critical level due to stochastic variation is higher when foraging conditions are poor. This suggests that failure to maintain mass in the poor environment is related to fitness costs associated with the high foraging effort this requires(Fig. 6). DEE in the poor environment was slightly lower than in brood-rearing starlings(Fig. 5), but brood-rearing birds accrue fitness benefits from a high DEE (i.e. offspring production),which are absent in our experiments. Absence of such benefits may be part of the explanation why birds were not motivated to further increase their DEE. There is growing evidence that there are trade-offs between work rate and different aspects of somatic maintenance and repair such as immune function(e.g. Sheldon and Verhulst,1996; Verhulst et al.,2005), and protection against oxidative damage(Wiersma et al., 2004; Alonso-Alvarez et al., 2004). It is therefore plausible that such effects would entail a cost of increasing work rate, but whether such effects really occur in our study system remains to be demonstrated. Nevertheless, given that we consider all other hypotheses less likely, and that trade-offs between work rate and somatic maintenance have been demonstrated in other systems, we consider the costs of increasing work rate the most likely explanation for our finding that birds do not maintain mass in the poor environment.

In conclusion, when trying to understand the effect of food availability on animal behaviour it is important to be aware of the flexibility of the energy budget. For example, energy saving that may be achieved through physiological adjustments may have significant consequences for individual-based modelling exercises, which explore the relationship between food supply, individual behaviour and population dynamics. Surprisingly, to our best knowledge field data on the relationship between foraging costs per reward and DEE or BMR still have to be collected.

We are grateful to Wim Beukema, Roy Voll, Ger Veltman, Jan Koenes, Wim Venema and Roelie van Zonneveld for their help with development and construction of the flight cages. Sjoerd Veenstra, Roelie Veenstra-Wiegman and Tosca Boeré looked after the birds, Leon Steyvers wrote software for data processing and Gerard Overkamp assisted with the respirometer and bomb calorimeter measurements. Ulf Bauchinger and Herbert Biebach kindly provided us with a flight muscle meter. We thank Rudi Drent, Joost Tinbergen, Ineke van der Veen, Ulf Bauchinger and Yvonne Verkuil for comments on the manuscript. Construction of the flight cages was supported by NWO-SLW grant 803-30.165 to S.V., who was further supported by the Technology Foundation STW, applied science division of NWO and the technology programme of the Ministry of Economic Affairs, and an NWO Vici-grant.