Energetic cost of hovering Øight in a nectar-feeding bat measured with fast-response respirometry

ORIGINAL PAPER Y. Winter Energetic cost of hovering ¯ight in a nectar-feeding bat measured with fast-response respirometry Accepted: 28 April 1998 Abstract Hover-feeding glossophagine bats provide, in addition to the hummingbirds, a second vertebrate model for the analysis of hovering ¯ight based on metabolic measurement and aerodynamic theory. In this study, the power input of hovering Glossophaga soricina bats (11.9 g) was measured by standard respirometry and fast-response (<0.2 s) oxygen analysis. Bats needed 5±7 s after a rest-to-¯ight transition to return to a respiratory steady state. Therefore, only hovering events preceeded by a 7-s ¯ight interval were evaluated. _ VO2 during hovering ¯uctuated with a frequency of 3±5 Hz, which corresponded in frequency to the licking movement of the tongue. During hovering, bats often may have hypoventilated as indicated by reduced _ VO2and a respiratory exchange ratio (RER) well below the steadystate value of 1. Steady-state oxygen consumption (and derived power input) during hovering was estimated to be 27 (25±29) ml O 2 g )1 h )1 (158 W kg )1 or 1.88 W) in the 11.9-g bats as indicated by three independent ®ndings: (1) _ VO2was 26 ml O 2 g )1 h )1 after 6.5 s of hovering, (2) the mean RER during single hovering events was at its steady-state level of 1 only at oxygen uptake rates of 25±29 ml g )1 h )1 , and (3) when the oxygen potentially released from estimated oxygen stores was added to the measured oxygen uptake, the upper limit for oxygen consumption during hovering was found to be 29 ml O 2 g )1 h )1 . Hovering power input was about 1.2 times the value of minimum ¯ight power input (Winter and von Helversen 1998) and thus well below the 1.7±2.6 dierence in power output postulated by aerodynamic theory (Norberg et al. 1993). Mass speci®c power input was 40% less than in hummingbirds. Thus, within the possible modes of hovering ¯ight, Glossophaga bats seem to operate at the high-eciency end of the spectrum. Key words Hovering ¯ight power á Aerodynamics áFast-response respirometry áBat á Hummingbird Abbreviations RER respiratory exchange ratio (respiratory _ VCO2=_ VO2)áRQ respiratory quotient (metabolic _ VCO2=_ VO2)áSwing area, S hw and S aw areas of hand and arm wing, bwing span, l hw and l aw length of hand and arm wing Introduction The nectar-feeding Glossophaginae are a group of neotropical bats highly specialized for feeding from ¯owers by night (cf. Dobat 1985; von Helversen 1993). Like hummingbirds, ¯ower-feeding glossophagine bats usually hover while visiting their ¯owers, and they have evolved ¯ight kinematics specialized for ecient hovering in a way that appears to be unique among bats. That is, as the wing executes a backstroke its tip (the distal wing triangle) is supinated, so that the morphologically lower surface is turned upward. As a result, aerodynamically eective lift is generated during the backstroke as well as the stroke forward and down (von Helversen and von Helversen 1975; von Helversen 1986). It was the goal of this study to determine the metabolic power input during hovering ¯ight in Glossophaga. It is interesting to know this cost for both proximate and ultimate reasons. 1. Hovering ¯ight marks the lower endpoint of possible ¯ight speeds so that for an integrated understanding of ¯ight mechanics and energetics it will be necessary to know both the metabolic power input and aerodynamic power output during hovering ¯ight. 2. For the cost/bene®t analysis of a foraging bat, the contribution of the energy expended in hovering ¯ight is important. If hovering is energetically expensive, it may have acted as an important physiological constraint during the evolution of the behavioural algorithms for making foraging decisions. J Comp Physiol B (1998) 168: 434±444 ÓSpringer-Verlag 1998 Y. Winter Institut fu Èr Zoologie II, Universita Èt Erlangen, Staudtstrasse 5, D-91058 Erlangen, Germany e-mail: [email protected] Aerodynamic considerations suggest that the energetic cost of hovering ¯ight might be very high, well above the cost of level forward ¯ight (cf. Norberg 1990; Norberg et al. 1993). In previous studies with Glossophaga soricina, metabolic power input during forward ¯ight (Winter et al. 1993; Winter and von Helversen 1998) and aerodynamic power output during hovering and forward ¯ight (Norberg et al. 1993) have been determined. The analysis of hovering and forward ¯ight in Glossophaga based on kinematic and morphological data and aerodynamic theory predicts an aerodynamic power output during hovering that amounts to 170± 260% of the value during level forward ¯ight (Norberg et al. 1993). However, recent measurements of oxygen consumption by Glossophaga during hovering ¯ight at a feeder functioning as a respirometric mask did not con®rm such a large dierence (Winter et al. 1998). _ VO2during hovering was even lower than the oxygen demand during level forward ¯ight. The respiratory exchange ratio (RER) found in these measurements was 0.8, which is below the expected value of 1 (oxidization of nectar carbohydrates), and this discrepancy might indicate that the bats did not reach a steady state while hover-feeding. In this case the measured oxygen uptake would not re¯ect the actual energy consumption in the muscles so that de®nite conclusions about the power input cannot be drawn. These measurements were based on a conventional respirometric analysis system with a relatively slow response time, as compared to the duration of the events. Measurements consisted only of single values for total oxygen uptake during single hovering events and did not allow the monitoring of instantaneous _ VO2. To circumvent this problem, I used in the present study a fast-response oxygen analysis technique (response time <0.2 s) to be able to determine changes in _ VO2during the 5±7 s hovering events. A hovering duration of 5±7 s, while still short if compared with the minute-long hovering ¯ights of hummingbirds or hawkmoths, should be sucient for a Glossophaga to come into the respiratory steady state. This is suggested by the results of an indirect estimate of the duration of the unsteady phase which was obtained here. It is critical to take into account such unsteady eects if respirometric data are to be used for inferences about the power input during hovering. Materials and methods This study was based on four individuals (2 females, 2 males) of Glossophaga soricina antillarum (Rehn 1902) (Phyllostomidae: Glossophaginae) which had been born and raised in a breeding colony of about 60 bats maintained in a tropical greenhouse at Erlangen University (the breeding stock originated from Jamaica). The bats ¯y about freely and feed by hovering in the greenhouses, so that they are well trained for ¯ight. The mean body masses and wing measures of the individuals used for the respirometric measurements are given in Table 1. Flight cage and nectar feeder The experimental procedures and equipment have been described previously (Winter and von Helversen 1998; Winter et al. 1998) and only a brief account is given here. Measurements were conducted in a climate-controlled room (22 °C1°, 57% rel. humidity, 0.975 kPa air pressure) with the photoperiod set to 12 h:12 h L:D. The body mass of the experimental animal was monitored during a measurement series without handling the animal, as a bat always returned for resting to a roost that was suspended from an electronic balance. The arti®cial nectar feeder was used to function as a respirometric mask. The ¯ow of nectar (a solution of glucose, fructose and sucrose) into this feeder was achieved by a computeroperated pump and was thus under the control of the experimenter. Reducing the rate of nectar ¯ow led to a prolongation of the hoverfeeding visits by the bats. As a modi®cation of the respirometric feeder mask described earlier (see drawing in Winter et al. 1998), a computer-controlled lid was added directly above the feeder opening which folded down automatically after a bat had departed from the feeder and stayed closed for 40 s preventing access to the nectar. This temporary food restriction caused the bats to feed especially intensively after the ¯ap had opened, increasing the duration of their visits to as much as 7 s (as compared with a maximum of 4.4 s with the previous arrangement). Respirometry Rates of O 2 and CO 2 exchange of a bat hovering at the ¯owthrough feeder mask were determined with a standard respirometric analysis system as described in Winter et al. (1998). The system was modi®ed, however, for the fast-response _ VO2measurements; for this, the oxygen sensor cell (Ametek [now AEI Technologies] N-37M with 0.25-ml internal sensor volume) was placed upright in the ¯ight cage and connected by a 3-cm tube to the respirometric mask. To minimize washout delay, no ®lter or desiccant was placed between sensor and mask. Components behind the gas outlet of the sensor were: desiccant, ®lter, mass ¯ow controller (Bronkhorst) and membrane pump (Wisa). The ¯ow rate through the sensor was constant at 800 ml min )1 standard temperature and pressure, dry (STPD). As shown before (Winter et al. 1998) this ¯ow rate was more than twice the minimum rate for capturing all exhaled gases from Glossophaga. The sensor was calibrated under these operating conditions with injections of N 2 (N 2 -dilution technique, Fedak et al. 1981). At this high ¯ow rate, the baseline of the sensor showed ¯uctuating drifts. A potential error resulting from such drifting was avoided, however, by determining the baseline before and after each of the 5to 10-s measurement intervals. Data were collected at a rate of 50 Hz. Total volumes of O 2 exchanged during hovering were calculated by integrating the signal envelope of oxygen depletion, and then subsequently corrected for CO 2 production using equation 3b from Withers (1977), with respiratory quotient (RQ) values taken from Table 1 Morphological data for 4 individuals of Glossophaga soricina antillarum (Rehn 1902). Mass was mean body mass during the hovering measurements, (Swing area, S hw and S aw areas of hand and arm wing, bwing span, l hw and l aw length of hand and arm wing). De®nition of morphological quantities as in Norberg and Rayner 1987. S includes areas of body and tail membrane (ca. 10%). Wings were measured (using Sigma Scan software) from scanned photographs of the bats held on their backs with outstretched wings Sex Mass (g) S (m 2 ) S hw (m 2 ) S aw (m 2 ) b (m) l hw (m) l aw (m) F 1 11.2 0.0111 0.00235 0.00265 0.271 0.072 0.044 F 2 12.7 0.0114 0.00248 0.00267 0.276 0.073 0.044 M 1 11.4 0.0100 0.00217 0.00232 0.265 0.067 0.044 M 2 12.3 0.0114 0.00243 0.00265 0.276 0.073 0.043 435 mean values determined during this study with the standard respirometric setup (Fig. 3C). As ¯ow rates were determined with a mass ¯ow controller behind the sensor, an additional correction for STP was not necessary. The air passing through the sensor was not dried until after the sensor, so that the drop in O 2 -concentration measured during a hovering event was partially caused by dilution with exhaled water vapour. In the absence of detailed information on respiratory parameters in Glossophaga, it was not possible to estimate the quantity of this exhaled water vapour with sucient precision. Therefore, I performed a `biological calibration' of the fast-response oxygen analysis system. For this I compared the mean signal integrals from the fast-response system with an equivalent dataset (hovering events of equal duration) from the standard respirometric set-up which was calibrated conventionally. This gave a factor by which the signals from the fast-response system had to be multiplied in order to arrive at the same mean value of oxygen uptake for both data sets. To optimize the ¯ow of gas, the respirometric mask was altered slightly from our earlier version depicted in Winter et al. (1998). The gas outlet at the mask was moved to the far end of the tube to achieve a straight ¯ow of the gas from the opening of the feeder tube to the gas exit. To prevent the bat from blocking the gas outlet of the mask with its tongue during feeding, the upper half of the feeder tube was protected with wire mesh placed inside the tube. The fast-response measurements were performed with only one animal, as this measurement setup without ®lter and desiccant places the oxygen sensor at a high risk of damage (droplets of saliva or nectar entering the 750 °C ceramic sensor cell may seriously damage it). Lag of _ VO2response after rest-to-work transition When a resting bat takes o to ¯y, its respiratory gas exchange does not immediately re¯ect steady state rates. The duration of the transient phase of gas exchange was determined indirectly by analyzing the relationship between the oxygen uptake during a hovering event and the duration of the total ¯ight interval preceding this hovering event. The duration of each ¯ight interval was measured with an accuracy of 1 s by comparing the time when the balance connected to the roost signalled take-o with the time of interruption of the light beam at the respirometric mask. For the analysis, hovering events were grouped into 1-s classes of `¯ight duration before hovering' and the mean oxygen uptake during hovering was determined for each of these classes. Data plots revealed an asymptotic curve, from which the minimum ¯ight duration (where _ VO2was the same as after longer ¯ights) was determined (Fig. 1). Controls Swallowing and breathing The swallowing of nectar might inhibit breathing and therefore oxygen uptake during hovering. To test for such a possible handicap caused by swallowing nectar _ VO2was measured during hovering visits at the feeder under two dierent conditions: (1) the bat received a regular nectar reward (with swallowing) or (2) the bat did not receive a nectar reward (without swallowing). For these measurements, the computer-controlled nectar pumps were activated such that the bat received a nectar reward only during every second visit at the feeder. Thus hovering events with swallowing (nectar reward) alternated with hovering events without swallowing (no nectar reward). Supporting body mass on the feeder mask To test for the possibility that a bat may support some of its weight on the feeder mask during hovering, I measured the vertical force exerted by the bat on the feeder. The feeder mask for this measurement was ®xed on top of a programmable electronic Mettler PM-100 balance, 30 cm above the balance plate. To prevent the air accelerated downward by the hovering bat from exerting a force on the balance plate, the plate was shielded from this air current by a piece of cardboard (with a small hole for the feeder holder). Measurements were performed with three individuals of Glossophaga soricina with a mean body mass during the measurements of 11.4 g. For ®nal data analysis, the force measurements during the last 250 ms of a hovering event were not included because during the bat's departure the balance became unstable. Results Lag of _ VO2response after rest-to-work transition The time interval during which a bat had been in ¯ight prior to hovering at the feeder was variable. The minimum ¯ight interval before a hovering event was 2 s, which was the time needed to ¯y from the resting place directly to the feeder. On most occasions, however, a bat had been in ¯ight for longer than 2 s before a given hovering event. Plotting the `duration of ¯ight interval before hovering' against the mean _ VO2during hovering (Fig. 1) revealed that oxygen uptake during hovering was reduced when the hovering event had been preceded by only a short ¯ight interval, less than 5±7 s in duration. Therefore, the data analysed in this study were restricted to hovering events that were preceded by a ¯ight interval of at least 7 s to be sure that measurements of _ VO2during hovering were not in¯uenced by non-steady-state respiration following the rest-to-¯ight transition. Mean gas exchange during single hovering events Measurements of gas exchange (O 2 and CO 2 ) with the standard respirometric setup were performed for the Fig. 1 Mean rates of oxygen consumption (STPD) during single hovering events of Glossophaga bats as a function of the duration of the ¯ight interval prior to hovering (values are means  1 SD from 4 individuals). The ¯ight interval given here is the time between the departure from the balance-roost and the arrival at the photocellequipped respirometry feeder mask. The dotted vertical line at seven seconds indicates the beginning of the plateau phase, in which _ Vo2 during hovering was independent of the pre-hovering ¯ight interval. The accuracy of timing was 1 s 436 four individuals during 1651 hovering events with a duration of up to 6.9 s. From these standard measurements only mean values of _ VO2and _ VCO2were obtained for a single hovering event. Of these events, 970 were preceded by a ¯ight interval of at least 7 s duration and were included in the following analysis. The total volume of oxygen taken up during a single hovering event increased linearly with the duration of that event (Fig. 2). It is interesting to note that the variability of oxygen uptake changed with the duration of the hovering events. Among the hovering events of less than 4 s duration, oxygen uptake varied by about a factor of two around the mean (regression line in Fig. 2) while hovering events that lasted for longer than 4 s were always accompanied by a high rate of oxygen uptake. The mean rate of gas exchange (ml g )1 h )1 ) for whole hovering events also varied with the duration of the event (Fig. 3): it was high during long and short hovers but low during hovers between 1.5 s and 4 s duration. As can be seen by comparing Figs. 2 and 3, the lower mean rate of gas exchange during hovering events of medium duration, was mainly caused by the occurrence of a larger number of such events with a low total oxygen uptake. Parallel to this, the respiratory exchange ratio (RER, as opposed to the cellular RQ) was around 1 during short hovering visits but dropped to a value around 0.8 during hovering events of medium duration (Fig. 3C). Only hovering events lasting longer than 5 s were again accompanied by a higher RER, which approached 0.9 during 6-s hovers. Ventilation intensity and RER A temporary deviation of the RER from the steady-state value of 1 may be caused by hypoventilation (see Discussion). If this is the case, a low RER should be associated with a low rate of ventilation (hypoventilation) and the RER should increase at higher rates of ventilation. Such a relationship was found when the mean rate of gas exchange calculated as _ VO2_ VCO2=2 was plotted against the RER. At RER 1, mean gas exchange was around 27 (25±29) ml g )1 h )1 , but dropped signi®cantly at RER 0.8. It appears justi®ed to use the mean rate of gas exchange as an index for ventilation intensity, as the two rates were signi®cantly correlated with each other (Fig. 4, inset). This is expected when higher rates of gas exchange are caused by a higher ventilation intensity. Fast-response _ VO2 The deviation of the RER from the steady state value of 1 (Fig. 3C) and the change in _ VO2during longer (as compared to shorter) hovering events indicated that standard respirometry, which only gave a single average over a complete hovering event, was insucient for obtaining an estimate of the steady state oxygen demand. This is because the instantaneous rate of oxygen uptake probably changed during hovering. For this Fig. 2 Oxygen uptake (STPD) of Glossophaga bats (n4) during hover-feeding at a respirometry mask. The regression line (V o2 [llg )1 ] 5.55 t [s]) was drawn to show that variability of oxygen uptake was not constant over the range of hovering event durations. Instead, during hovering events of longer than 4 s duration, oxygen uptake was less variable and almost always above average. The data shown are from hovering events that were preceded by a ¯ight interval of at least7sduration(seeFig.1) Fig. 3A±C Rates of gas exchange (STPD, means  1 SD) of four Glossophaga bats (see Table 1) during single hover-feeding events at a respirometry mask. The data in Aare the individual mean values and data in Bcombine the means from A. The respiratory exchange ratio (RER) in Cwas calculated from the data in B. It represents the mean RER from total hovering periods and not the instantaneous ratio during a hovering event. Data are from hovering events that were preceded by a ¯ight interval of at least 7 s duration (see Fig. 1) 437 reason I employed a fast-response oxygen analysis system (response time <0.2 s) in order to monitor directly changes in _ VO2during and towards the end of longer hovering events. For this purpose, only hovering events longer than 5 s were analysed (n15). The measurements revealed large ¯uctuations in _ VO2during hovering (Fig. 5A), probably caused by a modulation of breathing intensity. During the ®rst second of a hovering ¯ight these oscillations were of small amplitude, at a frequency of about 5 Hz. Thereafter, the amplitude increased and the frequency decreased to about 3 Hz, which most likely corresponded to the licking frequency of the tongue. Observations of the hovering bat with an infrared video system during the measurements showed that licking frequency was higher during the ®rst second, before the bat had started to imbibe the nectar solution, and dropped to a lower rate after the bat had contacted the nectar. The response time of the O 2 -analysis system was still too slow to resolve the breathing frequency, which would be expected to be synchronized with the wingbeat cycle at about 15 Hz. The mean rate of _ VO2as determined from the 15 measurements obtained, increased during a hovering event, reaching a value of 26 ml O 2 g )1 h )1 after 7 s of hovering (Fig. 5B). Controls Swallowing and breathing To test for the possibility that the swallowing of nectar interferes with breathing, _ VO2was determined during a measurement series in which the bat received a nectar portion during only every second feeder visit (Fig. 6). During this experiment, a bat hovered for a longer duration when it did receive a nectar reward at the feeder than when it did not. As observed before (Fig. 3B), mean _ VO2was higher when a hovering event lasted longer. However, there was no indication that nectar intake interfered negatively with oxygen uptake. If the swallowing of nectar had had a negative eect on breathing, one would expect lower values of _ VO2during visits when nectar was consumed. Such an eect was not observed (ANCOVA, P> 0.7, f0.101, n114, hovering duration as covariate). Supporting body mass on the feeder mask during hovering A bat supported only a small fraction of its body mass by leaning on the feeder mask (as measured with the feeder on an electronic balance). At the beginning of a hovering event a bat pushed its head into the feeder Fig. 4 Change of RER with the mean rate of gas exchange, which is used here as an index of lung ventilation intensity. _ Vco2and _ Vo2 during single hovering events, were positively correlated to each other (inset regression line). Thus higher rates of gas exchange were most likely caused by a higher ventilation intensity, so that the mean rates of O 2 and CO 2 exchange ([Vco 2 +Vo 2 ]/2)canbeusedasanindexof ventilation intensity. The horizontal dashed line at RER 1 indicates the steady-state value (RER metabolic RQ) expected during nectar-sugar catabolism. Data as in Fig. 3 Fig. 5A±B Rates of oxygen uptake of a Glossophaga bat while hovering at a respirometry mask for at least 5 s. Measurements were made with a fast-response oxygen analysis system (response time <0.2 s). (A) The ®ne line without symbols is an original trace obtained with 50-Hz sampling. The circles are means of the original data calculated over 500-ms intervals (2 Hz). (B) Oxygen uptake during 15 separate hovers (®ne lines, 2 Hz as in A)andtheoverall mean (bold line, n15). The lines are dotted from the time when the bat had left the feeder; the overall mean (bold line) was obtained while the bat was still at the feeder. Note in (A) that the oscillation of the original trace changed during the hovering event, most probably in synchrony with the licking frequency of the tongue. Before the ®rst nectar reward had been given, licking frequency was around 5 Hz (with a low variability in O 2 -uptake) and dropped to around 3 Hz after the bat started imbibing the nectar (with high variability in O 2 - uptake). The bat's tongue could not block the gas outlet hole of the mask because a mesh restricted the tongue's access to the bottom half of the feeder tube. Data are from male 1 (see Table 1 and Fig. 3A) 438 mask from below and thereby produced an upward oriented force (lift) against the mask. This lift force ranged between )5mN and )16 mN (Fig. 7). After about 1 s the initial lift force turned into a downward oriented force (leaning onto the mask) with a magnitude of up to 5.5 mN (equivalent to a body mass support of 0.56 g). Thus, a bat supported less than 5% of its body mass by leaning on the feeder mask. Discussion Lag of _ VO2response after a rest-to-work transition The goal of this study was to elucidate the metabolic power input required for hovering by a nectar-feeding glossophagine bat. One of the diculties in studying the hovering energetics of Glossophaga is that the individual hovering ¯ights are very brief; they almost always last less than 1 s when the bat visits ¯owers in the ®eld, and even in the laboratory their duration is well below 10 s. Deducing power input by respirometry can be dicult during short-term activities, as the gas exchange transients between the lung and the atmosphere give a distorted re¯ection of events taking place in the tissues because of intervening circulatory delays and the buffering eects of gas stores. It was crucial for this study to have at least a rough estimate of the duration of the transient phase after a rest-to-work transition, and to be able to perform the measurements of oxygen uptake during the steady-state phase of hovering. An indirect estimate of the duration of this transient phase in Glossophaga is provided by the relationship found here between the mean _ VO2during hovering ¯ight and the duration of the preceding interval of forward ¯ight (Fig. 1): after making the transition from resting to ¯ying a Glossophaga took about 5±7 s to reach its new respiratory steady-state. Before this period of adjustment had elapsed, the oxygen uptake measured during hovering was distinctly lower than that found after longer ¯ights. For this reason, only the measurements preceded by a ¯ight interval of at least 7-s duration were evaluated in the present study. In our previous analysis of oxygen uptake by Glossophaga during hovering ¯ight (Winter et al. 1998), this aspect was not taken into account, and hence the gas consumption rates reported in the former study were generally lower. _ VO2and level forward ¯ight One way to test whether the method chosen here gave realistic results is to compare the turnover rates measured during brief hovering ¯ights (<1 s) with the previously determined cost of forward ¯ight in Glossophaga (Winter et al. 1993; Winter and von Helversen 1998). After the switch from one work level (forward ¯ight) to another (hovering ¯ight), the respiratory rate should initially still re¯ect the rate of the previous work level. The previous estimate of forward ¯ight cost was based on an indirect method in which both daily energy expenditure and the daily duration of ¯ight activity were quanti®ed. Flight cost was then estimated by relating 24h ¯ight activity with daily energy expenditure using multiple regression analysis, taking into account metabolic expenditures for resting. The result of these measurements for Glossophaga is bracketed by estimates of ¯ight cost in small vespertilionid bats using a dierent Fig. 6 Mean oxygen uptake during single hovering events of a Glossophaga bat at a respirometric feeder mask. Nectar food was available only during every other hovering visit to the feeder (®lled circles). Imbibing nectar strongly aected the duration of a hovering visit but had no apparent eect on the rate of oxygen uptake during hovering. There was no dierence in mean oxygen uptake between hovering events with and without a food reward (ANCOVA, P>0.7,f0.101, n114, hovering duration as covariate) Fig. 7 The downward oriented vertical force exerted on the feeder mask by three hovering Glossophaga soricina bats. Negative values were caused by an upward oriented force (lift) on the mask. A lift force was produced by a bat during the initial insertion of its head into the mask. Data are means  1 SD from three individuals hovering for durations between 1.5 and 2 s (open circle: n31, mass 11.3 g; ®lled circle: n20, mass 10.5 g; triangle: n36, mass 12.5 g). These data include the values from two individuals published previously (Winter et al. 1998). The vertical force applied by a bat during hovering onto the feeder mask was measured (sampled at a rate of 7 Hz) with the feeder mask (without respirometry tube or photocell electronic wiring) mounted 30 cm above an electronic balance (Mettler PM-100). The opening of the feeder mask was oriented downwards at a 45°angle. The balance was shielded from the air accelerated downwards by the hovering bat. As the balance became unsteady during the bat's takeo, data from the last 0.25 s of a hovering event were not included 439 method (doubly labelled water method, Speakman and Racey 1991) and ®ts well with the data for ®ve other species of glossophagine bats ranging in body mass from 7 g to 28 g (Winter and von Helversen 1998). As expected, _ VO2during short hovering events (<1 s, 22.2  1.7 ml O 2 g )1 h )1 , Fig. 3B) was not signi®cantly dierent from our previous estimate of _ VO2during forward ¯ight at medium speed (23.8  1.8 ml O 2 g )1 h )1 for 11.7 g bats, Winter et al. 1993; Winter and von Helversen 1998). In addition, the RER also corresponded to the expected steady-state value of 1 (Fig. 3C, catabolism of sugar carbohydrates). The measurements of _ VO2obtained during this study during short hovering visits were thus in general agreement with the expected value for the cost of forward ¯ight. _ VO2during hovering Two phenomena make it dicult to measure the steadystate oxygen consumption of Glossophaga in hovering ¯ight: (1) the brief duration of hovering and (2) the presumed tendency to hypoventilate while hovering (see below). Three independent ®ndings of this study, however, indicate that the steady-state oxygen consumption of hovering 11.9 g Glossophaga was between 25 and 29 ml O 2 g )1 h )1 (or 298±345 ml O 2 h )1 ): (1) _ VO2measured with the fast-response O 2 analyser was 26 ml O 2 g )1 h )1 after 6.5 s of hovering (Fig. 5), (2) it was only when the O 2 uptake was high, 25±29 ml O 2 g )1 h )1 , that the RER was at its steady-state level of 1 (Fig. 4), and (3) if an estimate of the oxygen potentially released from body oxygen stores was added to the measured oxygen uptake, the upper limit for oxygen consumption during hovering was found to be 29 ml O 2 g )1 h )1 (Fig. 8). These three points will be discussed in detail below. Fast-response measurements of O 2 -uptake The rate of oxygen uptake was 26 ml g )1 h )1 (Fig. 5) after 6.5 s of hovering ¯ight, a duration that should be suciently long for the respiratory steady state to be reached. A resting bat which initiated ¯ight needed only 5±7 s to come into respiratory steady state (Fig. 1) although metabolic power increased 3±10-fold (depending on the thermoregulatory eort during resting). In contrast, ¯ight power for hovering ¯ight is less than 1.5 times as great as that for forward level ¯ight. RER and hypoventilation Nectar-feeding Glossophaga consume nectar-sugar; hence their metabolic RQ during feeding activity should be 1, as is the case in other nectar-feeders such as hummingbirds (Suarez et al. 1990). During this study, however, the RER was often well below 1 (Fig. 3C) and this may have been caused by hypoventilation. Hypoventilation leads to a change in pulmonary partial pressures of O 2 (decrease) and CO 2 (increase). In turn, these changes of partial pressure in¯uence the rates of alveolar gas exchange. This eect, however, is dierent for CO 2 than for O 2 because the CO 2 -binding curve of the blood is much steeper than the corresponding O 2 - binding curve. Therefore, changes in pulmonary partial pressures will cause a transient suppression of CO 2 exchange, whereas the initial eect on the alveolar oxygen exchange is only small (e.g. Widdicombe and Davies 1991; Scheid 1996). During the short-duration phenomena investigated here, reduced ventilation could thus lead to a reduction in measured rates of gas exchange and a concomitant drop in RER (Fig. 3). This explanation is supported by the observation that higher rates of mean gas exchange (determined here as VO2VCO2=2were positively correlated with a higher RER (Fig. 4). The steady-state value of 1 (RER  metabolic RQ) was attained only with gas-exchange rates of 25±29 ml O 2 g )1 h )1 . This range brackets the Fig. 8 Estimate of maximum power input during hovering ¯ight in Glossophaga. Filled circles and corresponding regression line aare the mean values of measured oxygen uptake during hovering ¯ights of diering durations (data taken from Fig. 3B; regression aVo 2 [ul g )1 ]5.80 t [s], calculated from the plotted means). Small dots show the distribution of the corresponding raw data (from Fig. 2). It was assumed that bats built up an oxygen de®cit during hovering and that the observed failure to sustain hovering for longer than 6.5 s was caused by exhaustion of internal oxygen stores and/or anaerobic capacities. Power input was thus fuelled by both respiratory oxygen uptake and the depletion of oxygen stores and/or anaerobic capacities. Measured oxygen uptake for a 6.5 s hovering event was 37.7 llg )1 (as given by regression a). The capacity for oxygen storage and/or anaerobic metabolism was assumed to equal the value for a hummingbird, estimated at 15 llO 2 per g body mass (see text). This value of 15 llg )1 was added (line b) to the measured oxygen uptake of 37.7 llg )1 during a 6.5 s hovering event. The resulting value of 52.7 llg )1 is the estimated maximum steady state oxygen demand for 6.5 s of hovering. This data point was connected to the origin by line c. The slope of line ccorresponds to a _ Vo2during hovering of 29.2 ml g )1 h )1 .Linedgives the minimum respiratory oxygen uptake necessary for sustaining hovering ¯ight. This was calculated by substracting estimated oxygen storage capacity (15 llg )1 )fromthe estimated steady state demand given by line c. Note that virtually all measurements of oxygen uptake (small dots) were in fact above this proposed lower limit of respiratory oxygen demand (line d) 440 directly measured _ VO2of 26 ml O 2 g )1 h )1 after 6.5 s of hovering. Why Glossophaga might have evolved a general tendency to reduce breathing during hovering is unclear. The imbibing of nectar per se did not suppress oxygen uptake. A comparison of hovering events with and without nectar availability did not indicate a dierence in _ VO2(Fig. 6). One possibility is that hypoventilation during the feeding from ¯owers might prevent the inhalation of pollen while the head is within a ¯ower's corolla. An additional reason may stem from the fact that the bell-shaped ¯owers, typical of some glossophagine-pollinated plants, are especially small, often no larger than a `head-mask' for the bat (Vogel 1968, 1969; von Helversen 1993, 1996) and thus may be virtually sealed by the head during hovering (Fig. 9). In addition, with a tidal volume for each breath of about 2 ml of air s )1 divided by breathing frequency (3 to 15 Hz. Fig. 5) the convective exchange into the ¯ower corolla may be low (tidal volume estimated with oxygen extraction eciency equal to 0.2 as in 100-g Phyllostomus during ¯ight, Thomas et al. 1984). With the ¯ow of fresh air into the ¯ower corolla impeded, the limited gas exchange possible might not be worth the ventilatory eort (although this eort may be low due to the mechanical linkage of ventilation to wingbeat frequency, cf. Thomas 1987). Capacity for oxygen storage and/or anaerobiosis On average, the measured mean _ VO2during a hovering event did not exceed 22 ml g )1 h )1 (Fig. 3B) which was below the estimated steady state demand of 25±29 ml O 2 g )1 h )1 . Therefore, it is most likely that the ATP for hovering metabolism partially came from oxygen stores and/or anaerobic processes. This remained unaccounted for by the respirometric measurements. The amount of oxygen storage and/or anaerobic capacity is unknown for Glossophaga, but data from hummingbirds may allow an approximation of this value. Data obtained by Chai and Dudley (1996) during their investigations of the limits of hovering ¯ight performance may be used for this purpose. Normally, hummingbirds are capable of sustaining hovering ¯ight for extended periods of up to several minutes. When Archilochus hummingbirds, however, are forced to hover in a hypoxic and hypodense air/helium mixture with 13% O 2 , they fail to hover after 6 s (Chai and Dudley 1996). In such hypodense air, the ¯ight muscle power output for hovering is about 20% higher than in normal air; the oxygen uptake during hovering remains at the normoxic, steady state rate of 45 ml g )1 h )1 . Thus, the birds build up a 20% oxygen de®cit per second. When they terminate hovering ¯ight after 6 s they have built up a de®cit corresponding to the oxygen demand of 1.2 s of hovering ¯ight or 15 ll per g of body mass. This is consistent with the ®nding that when the same hummingbirds are forced to breathe pure helium during hovering, the birds perform a ``ballistic descent'' after 1.5 s of hovering (Chai and Dudley 1996). Here, they presumably over-exploit their oxygen stores and/or anaerobic capacity. Using the same capacity for oxygen storage (and/or anaerobic capacity) of 15 llO 2 g )1 for a Glossophaga bat, one can estimate an upper limit of power input during hovering. Such an estimate is presented in Fig. 8. During hovering ¯ights of 6.5 s duration (the maximum duration observed during this study) a Glossophaga had an estimated maximum of 52.7 llO 2 g )1 available for metabolism which corresponds to an expenditure of 29 ml O 2 g )1 h )1 . These 52.7 llg )1 result when the measured uptake of 37.7 llO 2 g )1 during a 6.5 s hovering ¯ight is added to the estimated oxygen store depletion (and/or anaerobic capacity) of 15 llO 2 g )1 . There is no reason to assume the evolution of an Fig. 9 Hover-feeding Glossophaga soricina at a Paliavana prasinata (Gesneriaceae) ¯ower. Note that the bell-shaped corolla ®ts like a head-mask on the bat's face. (photograph taken by O. v. Helversen in the laboratory at a plant grown from seeds collected by M. Sazima in Vitoria, Espirito Santo, Brazil) 441 especially high tolerance to anaerobic metabolism in Glossophaga as compared to hummingbirds, in view of the fact that hovering durations of Glossophaga within the natural environment are short (normally below 1 s, personal observation, Tschapka 1993). Thus it seems unlikely that hovering power input in Glossophaga exceeded 29 ml O 2 g )1 h )1 . Power input during hovering The combined evidence for the amount of power input during hovering ¯ight in Glossophaga can be summarized as follows. Our initial measurements revealed a _ VO2of 16.7 ml O 2 g )1 h )1 during 4 s hovering ¯ights (Winter et al. 1998). Since then we learned that a Glossophaga, after departing from its roost, needs to have been in ¯ight for about 5±7 s before achieving a respiratory steady state (Fig. 1). When respiratory measurements during hovering were restricted to hovering events preceeded by a ¯ight interval of at least 7-s duration, the mean O 2 measured during single hovering ¯ights (4±7 s) was 21 ml g )1 h )1 (Fig. 3B). However, even here, the deviation of the RER from the steady state value of 1 (Fig. 3C) and the change in _ VO2during longer (as compared to shorter) hovering events indicated that standard respirometry, which only gave a single average over a complete hovering event, was insucient for obtaining an estimate of the steady state oxygen demand. This is because the instantaneous rate of oxygen uptake probably changed during hovering. For this reason I employed a fast-response oxygen analysis system (response time <0.2 s) in order to directly monitor changes in _ VO2during and towards the end of longer hovering events. These measurements revealed that _ VO2increased from an initial value below 20 to a value of 26 ml g )1 h )1 after 6±7 s of hovering (Fig. 5), and this time interval should be sucient for a hovering Glossophaga bat to come into respiratory steady-state. A dierent line of evidence arrived at the same estimate. Anaylsis of the data obtained by standard respirometry (one value of mean _ VO2for a single hovering event) showed that only at higher rates of gas exchange (25± 29 ml g )1 h )1 ) was the RER during a hovering event at its steady state rate of 1 (Fig. 4). A maximum value for _ VO2 of 29 ml g )1 h )1 was also suggested when an estimate of oxygen storage capacity was added to the mean 22 ml O 2 g )1 h )1 consumed during 6.5 s hovering events (Fig. 8). Steady state oxygen demand of hovering Glossophaga soricina was thus found here to be 27 (25±29) ml O 2 g )1 h )1 which corresponds to a power input of 158 W kg )1 or 1.88 W for the 11.9-g bats. Additional in¯uences on hovering The hovering during the measurements may have been in¯uenced by some additional factors which are unlikely to signi®cantly aect the conclusions drawn from this study. For completeness they are listed below, but as most have been discussed in detail previously (Norberg et al. 1993; Winter et al. 1998), this discussion is not reiterated at full length. 1. Bats approached the feeder following a pendulumswing-like ¯ight path with the dead centre at the level of the feeder, thus converting kinetic into potential energy during the ®nal approach path. The energy saved for hovering from this eect, however, is unlikely to last for more than a few tenths of a second. 2. Due to the ground eect, the downwardly induced air ¯ow may recirculate within a con®ned space (Rayner and Thomas 1991). As the bottom of the ¯ight cage here was made of netting that let air through and the solid ¯oor was approximately six wing spans from the hovering bat, this potential eect was likely to be very small (Norberg et al. 1993). In addition, Bartholomew and Lighton (1986), who compared measurements of _ VO2in hummingbirds within enclosures (with a potential ground effect) to those in free-ranging birds, found no dierence. 3. During hovering, bats partially supported their body mass by leaning on the respirometry mask, despite its being oriented downwards at a 45°angle (Fig. 7). Measurements of the vertical force exerted on the feeder mask (with an electronic balance) showed, however, that bats supported less than 5% of their body mass, so that the associated energy savings were equally small. 4. Bats have exceptionally large skin areas and may lose carbon dioxide through their well vascularised, thin ¯ight membranes (Herreid II et al. 1968). During hovering ¯ight, however, cutaneous gas exchange would be expected to amount to less than 1% of the total carbon dioxide output (Winter et al. 1998) and thus cannot explain the observed deviation in RER. 5. While it is a common experience for adult humans that we cannot simultaneously breathe and swallow, this is rather the exception than the rule within mammals (Tillmann and Wustrow 1982). The measurements performed here demonstrated that the swallowing of nectar did not inhibit oxygen uptake as _ VO2was the same during hovering visits with and without swallowing of nectar (Fig. 6). Power input and power output This study complements two earlier studies carried out at our laboratory with Glossophaga soricina,on(1) power input during level forward ¯ight (Winter et al. 1993; Winter and von Helversen 1998) and (2) aerodynamic power output during hovering and level forward ¯ight (Norberg et al. 1993). We can thus now compare the ¯ight costs of two modes of ¯ight (forward and hovering ¯ight), as well as two dierent ways of estimating the cost of ¯ight: aerodynamic theory and metabolic measurements. The cost of level forward ¯ight at medium speed in Glossophaga has been estimated at 1.63  0.12 W (Winter et al. 1993; Winter and von Helversen 1998) for individuals with nearly the same average mass (11.7 g) as used in the present study 442