CO₂ Release Patterns in Drosophila Melanogaster: The Effect of Selection for Desiccation Resistance

Most terrestrial insects experience a ready availability of oxygen and limited supplies of water. They seek to limit water loss through a number of physiological specializations of the excretory system, the cuticle and the respiratory system (Edney, 1977; Hadley, 1994a). It has been suggested that the control of gas exchange is one of these specializations. Insects have long been known to ventilate the tracheae intermittently (e.g. Punt, 1950; Schneiderman and Williams, 1953), principally by the action of spiracular valves at the external openings of the tracheae. This intermittent release of respiratory gases has, at various times, been termed discontinuous respiration, closed-flutter-open cycles, closed-flutter-ventilation cycles and discontinuous ventilation. Lighton and Garrigan (1995) have recently adopted a more general descriptive term, ‘the discontinuous gas exchange cycle’ (DGC). We employ this term here, since it avoids any confusion between gas exchange and metabolic respiration. The definition of the DGC also makes no assumption about the diffusive or convective mechanism of gas exchange or about the state of the spiracles during each phase of the cycle.

In its classic form, as seen in Platysamia cecropia pupae (Schneiderman, 1960; Levy and Schneiderman, 1966a,b,c), the DGC consists of three phases which correspond to the condition of the spiracles: open, closed or fluttering. It has been hypothesized that the DGC conserves water, since water loss from the respiratory surfaces would occur only during periods in which the spiracular valves are open. The simplicity and clarity of the hypothesis that the DGC evolved as a means of water conservation has led to its historical acceptance, although direct experimental evidence is limited (for a discussion of the limitations of the current evidence, see Edney, 1977; Hadley, 1994b).

Recently, the role of the DGC in water conservation in adult insects has been called into question, notably by Hadley (Hadley and Quinlan, 1993; Hadley, 1994b), who found that cyclic CO₂ release in the lubber grasshopper Romalea guttata occurs only when they are motionless and that the DGC is disrupted when the insects are desiccated. It has also been shown that measurements of water loss during the open phase of the DGC in different insects indicate that the vast majority of water loss is cuticular, not respiratory (Machin et al. 1991; Lighton, 1992; Hadley and Quinlan, 1993). In contrast, Lighton et al. (1993) has found that ants with more watertight cuticles show a higher percentage of respiratory water loss during the open phase of the DGC. These results suggest that the DGC may play an important role in water balance in some, but not necessarily all, insects where the DGC occurs.

As gas exchange patterns have been studied in a greater variety of insects, it has become clear that the classic DGC of open, closed and flutter phases may be only one of many existing patterns. Differences in the relative length of the phases have been observed in ants (Lighton, 1988, 1990). In grasshoppers, there is gas leakage during the closed phase, no flutter phase and multiple bursts of CO₂ release during the open phase (Quinlan and Hadley, 1993; Hadley and Quinlan, 1993). Discrete ‘flutter bursts’ rather than flutter phases are seen in tenebrionid beetles (Lighton, 1991). A periodic DGC is absent in crickets (Quinlan and Hadley, 1982), in middle-aged grasshoppers (Hamilton, 1964), in active insects (e.g. Lighton, 1991) and at high temperatures (Hadley and Quinlan, 1993). Lighton and Berrigan (1995) even observed that the gas exchange patterns of different castes within an ant species can differ. They suggest that the DGC might be an adaptation, at least in these species, to habitats high in carbon dioxide. It would seem, therefore, that discontinuous gas exchange, which was initially thought to be a universal adaptation for water conservation in terrestrial insects, is actually quite variable in its expression within and between species. Its significance in the conservation of water has also been called into question.

Various laboratories are presently conducting studies that attempt to clarify the role of the DGC by examining its taxonomic distribution among arthropods, or by examining differences in gas exchange patterns in insects from a variety of habitats. An alternative approach is the use of selection experiments. Such experiments can control precisely for phylogeny and can reveal changes in physiology in response to highly controlled selection criteria. We have been conducting such experiments using the fruit fly Drosophila melanogaster, which can be kept as large outbred populations in the laboratory and is amenable to laboratory selection for resistance to desiccation. Our experiment uses a series of 15 populations of D. melanogaster, five populations selected for survival under desiccation, five control populations and five ancestral populations. We describe the changes in CO₂ release patterns observed in these stocks of flies in response to selection for desiccation resistance.

We used 15 outbred populations of Drosophila melanogaster selected and maintained in the laboratory of M.R.R. (Rose, 1984; Rose et al. 1990; Graves et al. 1992). Five replicate stocks of flies were the ancestor populations. These stocks were formed in 1980 and maintained as large, outbred populations. They were selected for postponed senescence and are hence termed O (old) flies. They are currently on a 10-week generation cycle.

In 1988, each replicate O stock was used to start two additional populations, D (desiccated) flies, which were subsequently selected for desiccation resistance, and C (control) flies, maintained as controls for the D population. Thus, the O1 stock is the direct ancestor of the D1 and C1 stocks, and the O2 stock is the ancestor of both the D2 and C2 stocks, etc. (Fig. 1).

The D populations were raised in vials until 4 days post-eclosion, then selected for desiccation resistance by being placed as large populations in acrylic cages containing 150 g of Drierite desiccant wrapped in cheesecloth and containing no water or food (since our fly food contains water). When approximately 20 % of the flies remained alive (after approximately 72 h), they were given food and water and allowed to reproduce. The C and D populations differ only in access to water: while flies from the D populations were exposed to desiccating conditions, the C flies were maintained as controls with access to a water source but not to food. The C and D stocks were both on a generation time of approximately 3 weeks. At the time of these experiments, the C and D populations had undergone approximately 150 generations of selection.

All the stocks (O, C and D) were kept under 24 h light regime and maintained on banana–molasses food as both larvae and adults. The stocks were raised and selected at 25 °C.

For the present study, all 15 stocks were raised identically for two generations without overt selection pressure to eliminate parental and grandparental effects. During these two generations, the flies were treated in the following manner. Larvae eclosed after 10 days in the hatching vials. The adult flies were transferred into acrylic cages at day 14 and given food and yeast paste to stimulate egg-laying. Eggs were collected for the following generation and placed into 20 vials containing standard food with 60–80 eggs per vial. Matching replicates of stocks were reared on the same schedule: O1, C1 and D1 together, O2, C2 and D2 2 weeks later, etc.

Eggs for the experimental generation were collected from parental stocks five times, providing five experimental groups that would reach age 4 days post-eclosion every other day. Only female flies were used from the experimental generation. Females were used because only D females survive the desiccation selection treatment. The D females are inseminated by the males prior to selection.

For the separate hydration experiments, young adult females were taken from the O1, C1 and D1 stocks during routine maintenance. These flies were maintained in vials on banana–molasses food until placed in the respirometer.

Measurements were made on individual female flies in a stream of dry, CO₂-free air in a room maintained at 25±1 °C. Peaks of CO₂ release were measured by a Licor LI-6251 infrared CO₂ analyzer with Sable Systems data aquisition software. The noise level of this system is below 0.05 p.p.m. at our level of measurement. The respirometry chambers were 1 ml plastic syringes cut down to 0.5 ml, with a small amount of cotton placed at both ends to prevent the flies from leaving the chamber and to maximize turbulence in the air flow.

A series of valves controlled the air flow to an empty control chamber and six experimental chambers. For each experiment, two flies from replicate stocks of ancestor, control and desiccation-selected flies were placed individually in six chambers (e.g. O1, O1, C1, C1, D1, D1). Air flow from only one chamber was read at any one time. The air was passed through a Drierite/Ascarite/Drierite column to dry it and remove CO₂, and was then drawn through one chamber at a flow rate of 100 ml min⁻¹ using a vacuum. The release of CO₂ by the fly was averaged and recorded once per second. A pump was used to provide dry, CO₂-free air at a flow rate of 100 ml min⁻¹ through all chambers not being analyzed. The continuous flow of dry air over the insect created a strongly desiccating environment during the experiment.

At the start of each respirometry run, one female fly at age 4 days post-eclosion was anesthetized using CO₂ gas, placed in an experimental chamber and allowed 20 min to recover, after which the respirometry recording was activated. This process was repeated with subsequent flies placed in the respirometer at timed intervals so that each fly was subjected to identical treatment prior to recording. Each cycle of data acquisition recorded from the empty control chamber for two separate 6 min periods and from each experimental chamber containing a fly once (for 18 min). This cycle was repeated every 2 h until all flies were dead. Since the flies from the D populations had significantly longer survival times, the recording procedure was changed after the death of the O and C flies to measure values for the two D flies more frequently until their death. Repeating the recordings in this manner allowed us to examine the CO₂ release patterns of the flies as they spent increasingly more time in the desiccating environment of the respirometry chamber.

Each set of replicates (e.g. O1, O1, C1, C1, D1, D1) was run five times over 5 days. The order of the flies in the eight chambers was systematically varied to eliminate any chamber effects. By the end of the experiment, we had collected data for 8–10 flies from each of the 15 populations.

To test whether the ventilatory patterns of our Drosophila melanogaster populations are affected by hydration state, an additional experiment was carried out. Adult flies were gathered from O1, C1 and D1 populations as representative samples of our 15 populations. These flies were divided into two treatments: hydrated or dehydrated (N=15 for each treatment per population). The release of CO₂ from the flies was then measured. Individual females were placed in modified respirometry chambers made from 3 ml syringes. The flies either had access to a wet filter-paper wick (hydrated treatment) or were separated from the wick by a sponge barrier and dry air flow (dehydrated treatment). This protocol served to equalize the humidity of the excurrent air samples and matched any effects of the water-filled wick acting as a CO₂ sink or source. At the start of each run, six flies from one population were anesthetized and placed individually in six chambers: three hydrated and three dehydrated. CO₂ release was measured as described above, with repeats continuing until the death of the dehydrated flies.

Measurements of survival time in the respirometer, metabolic rate and gas exchange pattern were obtained for 8–10 flies from each of the 15 stocks. We recorded CO₂ release for each fly for 18 min once every 2 h from the time they were placed in the respirometer until death. A precise determination of death is difficult in small insects, but we found that the disappearance of peaks on the CO₂ recording corresponded with the loss of visible movement in the fly and was quickly followed by a decrease in CO₂ release to 0 p.p.m. We felt confident that the change in CO₂ release from large peaks to uniform release could be used to indicate death. Zero CO₂ levels in the control chamber were used as a baseline zero recording for each repeat. Although each recording interval lasted 18 min for each fly, we used only the last 16 min in our data analysis to reduce any effect caused by the switch from flushing to sampling air flow.

Behavior often has a strong effect on insect gas exchange (e.g. Kestler, 1991; Hadley, 1994a). In experiments with other insects, data on DGCs are often collected when the animal is motionless, thus removing the effects of increased CO₂ production due to muscular ventilation of the tracheae and air sacs. To a large extent, however, Drosophila melanogaster are never motionless. Flies from the C and O populations often walked and groomed continuously (but did not fly) until they were severely desiccated. In spite of this limitation, our observations established that large, noticeable peaks of CO₂ release in active flies were not correlated with isolated behavioral events such as struggling to exit the chamber, righting or bouts of flight. Non-quantitative observations indicated that flies from the D populations were much more quiescent. Using the D flies, we were able to establish that peaks of CO₂ release were not due to any visible activity. Large, occasionally rhythmic releases of CO₂ were observed in flies that appeared entirely quiescent.

The peaks of CO₂ release recorded from D. melanogaster were not periodic or of regular height. While fast-Fourier transformations and autocorrelation analyses are useful for assessing periodicity, we were more interested in quantifying peak height and frequency in our non-periodic data. We therefore used the standard error of a regression through the data to provide an objective measure of peak height and frequency. The recordings of CO₂ release were fitted using a linear regression, and we calculated the standard error of that regression (SER) as a measure of the ventilatory pattern. Repeated measurements from the empty control chamber gave an average SER of 0.01326 μl CO₂ h⁻¹. This value was subtracted from all other measurements of SER to remove the signal noise of the respirometry equipment.

The statistical ‘units’ in the comparisons of O, C and D selection regimes are the 15 populations. We compared the SER values of these populations in two ways: by comparing the largest peak size and frequency that the populations are capable of generating (maximum SER), and by comparing the average SER present at different times in the desiccating environment of the respirometry chamber (average SER). Maximum SER provides an estimate of the size of the largest peaks of CO₂ a fly can produce, while average SER is useful for describing changes in the peak height with time or desiccation. We measured the maximum SER for a population by averaging the maximum SERs of the 8–10 flies used from that population. To find the maximum SER for a single fly, we calculated the SERs for the full recording interval for each of the 6–50 recording intervals obtained for each fly (the number of intervals was dependent on the survival time). The highest SER from all the intervals for that fly was averaged with the maximum SERs from the other flies in the same population to obtain the population maximum SER. The five population SERs from each selection regime (O1–5, C1–5, D1–5) were also averaged to produce treatment maximum SERs for the O, C and D populations. The average SER values were gathered at each time interval by taking the SER values from all live flies within a population, and using the mean SER to represent that population at that time.

Comparisons of survival time and maximum SER between selected stocks were carried out using a single-factor analysis of variance (ANOVA) on 15 population means (d.f.=2, N=15). Correlations between survival time and SER were calculated using a linear regression (d.f.=1, N=15). The same correlations within individual populations were also checked with linear regressions (d.f.=1, N=8–10). Comparisons of metabolic rate at all the times between hours 2 and 10 were calculated using a 3(stock) × 5(time) repeated-measures ANOVA on all 15 population means (d.f. stock=2, d.f. time=4, d.f. interaction=8; N=75). The SER comparisons were analyzed similarly at all the times between hours 0 and 10. After finding a significant effect of selected stock for both SER and metabolic rate, single-factor ANOVAs were then calculated on the grouped data to allow for post-hoc comparisons. The comparison between hydrated and dehydrated SER on three combined populations was analyzed using single-factor ANOVAs (d.f.=1, N=45). All post-hoc comparisons were analyzed using the Student–Newman– Keuls test at the 0.05 significance level.

The selection regime for each experimental treatment (O, C, D) has been repeated in each of five populations. We defined the maximum SER value as the average of the maximum responses observed in all populations from a single selection treatment (see Materials and methods). These treatment means ± S.E.M. give the differences in SER between the O populations (0.48±0.02 μl CO₂ h⁻¹), the C populations (0.70±0.03 μl CO₂ h⁻¹) and the D populations (1.38± 0.14 μl CO₂ h⁻¹).

Fig. 2 gives examples of the CO₂ release patterns observed in the O, C and D flies. We illustrate patterns that provide a calculated SER near the maximum SER. In flies from the O populations (Fig. 2A) and the C populations (Fig. 2B), these patterns have irregular peaks of moderate height. The peaks from the flies of the D populations (Fig. 2C,D) are larger. Note that in the D populations, large values of SER are possible either from irregular peaks with heights 1–3.5 times those of the lowest CO₂ levels (Fig. 2C) or from highly regular peaks with heights 1–1.5 times the lowest CO₂ level (Fig. 2D). In our recordings from 50 D flies from five populations, the irregular peaks were the most common pattern observed. Regular, periodic peaks such as those shown in Fig. 2D were never seen in the C or O populations.

As discussed in Materials and methods, Drosophila melanogaster, especially in the five O and five C populations, were continuously active in the respirometry chamber. It is very likely that the noncyclic nature of CO₂ release in these two treatments is at least partially due to their continual walking and grooming. We made visual observations of the flies in the respirometer at the same time as recording CO₂ release and found that large peaks of CO₂ release, such as the spikes above 10 μl CO₂ h⁻¹ in Fig. 2C, were not correlated with isolated bouts of activity.

Fivefold replication of the populations allows us to test for statistically significant differences between treatments. On the basis of survival time in the stream of dry air in the respirometer, the desiccation-selected populations can resist desiccation better than the control or ancestral flies (P<0.01). If survival time is plotted against maximum SER for the 15 populations, there is a significant positive correlation (r=0.847, P<0.01). Post-hoc tests indicate that the survival time and maximum SER of the D populations are significantly different from those of the O and C populations (P<0.05), but the O and C treatments are not significantly different from each other with respect to either characteristic (Fig. 3).

While the correlation between maximum SER and survival time is highly significant at the treatment level, it is not seen within single populations (Fig. 4). Correlations between maximum SER and survival time for values from individuals from populations C5, D5 and O5, for example, have r values of 0.33 for C5, 0.23 for D5 and 0.03 for O5. None of the 15 stocks (N=8–10 individuals for each) showed a significant within-population correlation between survival time in the respirometer and maximum SER.

If discontinuous gas exchange is a physiological response to desiccation, one might expect the CO₂ release pattern to change as the insect desiccates. Fig. 5 illustrates the SER of the CO₂ release versus time in the respirometer. We analyzed the mean SER of the desiccation-resistant, control and ancestor populations over the first 10 h, the period during which all 15 populations were alive. Mean SER is significantly different between selection treatments (P=0.04) and with time (P<0.01). There is also a strong interaction between stock and time (P<0.01), indicating that there is a difference in how the treatment groups respond to desiccation over time.

All flies show similar SER values at hour 2. The C and D stocks show identical and increasing mean SER values until hour 8. At that time, these populations diverge in their response. The C flies show a decrease in SER for several hours prior to death. The D flies show higher and somewhat more variable SER values for many hours until levels decline shortly before death. The O flies showed no increase in SER above initial levels, but a similar decrease near death. Post-hoc tests indicate that SER values in the O flies are significantly different from those in the C and D flies during hours 0–10.

The above results demonstrate that the flies exhibit non-uniform CO₂ release in the respirometer under our experimental conditions. In addition, flies which have undergone selection for desiccation resistance show an increase in maximum SER compared with flies that have not undergone such selection. One possible explanation for these results is that the CO₂ release pattern observed in the respirometer is a specific response to the desiccating conditions. If so, access to water during respirometry should cause a reduction in SER. We therefore examined the CO₂ release pattern of flies in the respirometer under conditions in which they were or were not allowed access to free water in the form of a wet wick. We investigated whether flies reduce levels of discontinuous gas exchange in the presence of water, a hypothesis that makes no assumptions regarding the selection treatments of the flies. We used flies from the O1, C1 and D1 populations as a representative sample of our 15 populations. Flies provided with a water source maintained a lower SER (0.44±0.05 μl CO₂ h⁻¹, mean ± S.E.M.) than flies in desiccating conditions (0.78±0.08 μl CO₂ h⁻¹, P<0.01). Sample CO₂ release patterns of a hydrated and dehydrated fly are shown in Fig. 6.

We wished to determine whether the larger peaks in CO₂ production in the D populations were associated with an increased overall rate of CO₂ release. We found that the mean rate of CO₂ release differed significantly with selection treatment (P=0.019) (Fig. 7) but post-hoc tests indicate that it is the C populations rather than the D populations that have a slight but significantly higher average CO₂ release (4.7 μl CO₂ h⁻¹ for the C flies compared with 3.9 μl CO₂ h⁻¹ for the O flies and 4.1 μl CO₂ h⁻¹ for the D flies, P<0.05). Correction for dry mass removed any significant difference in the rate of CO₂ release between treatments during hours 2–10. Because the dry mass of the flies changed during desiccation in the respirometer, data are presented in Fig. 7 as μl CO₂ h⁻¹ per fly, not per milligram dry mass. There was no significant change in the rate of CO₂ release with time (P=0.070) during hours 2–10 when all populations were alive. The interaction term was significant, indicating that the treatment groups respond differently as time in the respirometer increases (P<0.01, see Fig. 7).

Since the discovery and characterization of the DGC (see Introduction), a number of experimental approaches have been used to examine the role of discontinuous gas exchange in water conservation. These include studies examining changes in spiracular opening rates under humid or dry conditions (Miller, 1964; Krafsur, 1971), water loss measurements with the spiracles forced open using CO₂ (Mellanby, 1934; Bursell, 1957; Miller, 1964), measurement of water loss during different phases of the DGC (Kestler, 1985; Lighton, 1988; Machin et al. 1991; Hadley and Quinlan, 1993) and comparisons of the DGC in mesic versus xeric species (Lighton, 1992; Quinlan and Hadley, 1993). We have used an additional experimental approach, that of laboratory natural selection, as a tool for exploring the association of ventilatory pattern with water conservation. The advantages of the use of laboratory natural selection are as follows: (1) the number of groups or replicate populations that can be compared is subject to the control of the experimenter, (2) the number of subjects from each group can be maximized since populations are maintained in the laboratory, (3) the phylogenetic relationships between all groups are known exactly, and (4) the selection pressures can be defined and replicated and are thus much better understood than historical conditions for wild-caught species.

Our selection study began with five replicate populations of the ancestral O treatment. By deriving one desiccation-selected (D) and one control (C) population from each ancestral (O) population, we produced a balanced phylogeny; each D population is more closely related to its control and ancestor populations than to other D populations. This maximizes our capacity to distinguish the effects of selection from the effects of shared phylogeny. All five D populations show a three-to fourfold change in desiccation resistance, expressed as hours of survival in dry air, compared with both the ancestral and control populations. Since this differentiated desiccation resistance is expressed following exposure to identical rearing conditions for two generations (see Materials and methods), it is clearly a result of genetic differentiation and not environmental acclimation.

We used these stocks to search for correlations between desiccation resistance and the pattern of CO₂ release. If control of CO₂ release has no effect on resistance to desiccation, we would expect no change in CO₂ release pattern in the D treatment. We found, however, that the D populations do show an increase in the size and/or frequency of CO₂ peaks as measured by the standard error of the regression through the CO₂ release recordings (SER). This increase occurred independently five times under the same selection regime, and similar increases did not appear in the five control or the five ancestral stocks. These results indicate that CO₂ release pattern can vary with selection. The fact that the pattern varies in different populations under identical conditions indicates a genetic control of the DGC.

This link between increased SER and increased desiccation resistance could occur in two ways. Selection for resistance to desiccation might act directly on the control of gas exchange because of its value in reducing water loss. Alternatively, CO₂ release might only be linked to the trait that selection operated upon, either through pleiotropy or through a physiological effect. For example, selection for desiccation resistance may increase quiescence under stress, which allows greater periodicity in CO₂ release. Our finding that the SER increased in the D populations establishes that selection for desiccation resistance affects gas exchange. Whether the DGC is the target of selection could be elucidated using additional experiments, such as by relaxing selection in the D populations or by selecting for both high activity and desiccation resistance in other populations of Drosophila melanogaster.

Because of our interest in the effect of desiccation on patterns of CO₂ release, and the above observation that desiccation-resistant flies show enhanced SER values, we chose an additional method of testing the relationship between CO₂ release peak size and hydration state by examining the pattern of CO₂ release of hydrated flies in the respirometer. When insects were placed in the respirometer with a water source, the SER was significantly reduced compared with that of flies from the same populations that did not have access to water.

Only a few previous studies have compared the patterns of gas exchange in hydrated and dehydrated insects. Hamilton (1964) found that locusts respired discontinuously only when very young (and water loss was high) and when starved (no water source). He concluded that the DGC was initiated to save water. Loveridge (1968) showed that dehydrated locusts ventilated the abdomen more slowly, and argued that this served to save water. These early studies showed a link between hydration and patterns of gas exchange – apparently so convincingly that little similar work followed for many years. Recent studies have provided more ambiguous results. Machin et al. (1991) found that dehydrated cockroaches ventilate more regularly and more often. They attributed this to a decreased capacity for CO₂ storage due to declining blood volume. Hadley and Quinlan (1993) found that desiccated grasshoppers have a less organized DGC than that of hydrated individuals, which led Hadley (1994b) to suggest that the DGC is not important for water conservation in this species.

Our work on Drosophila melanogaster adds information about a small dipteran to this body of data. The large surface area to volume ratio of Drosophila melanogaster makes them particularly vulnerable to water loss, so we might expect ventilatory control to be important in this species. Our findings support this expectation, since Drosophila melanogaster show a CO₂ release pattern that changes with desiccation.

The opening and closing of spiracles during the DGC has historically been demonstrated by measuring CO₂ release and O₂ uptake. In a classical DGC, levels of CO₂ release fall to zero during the closed phase and rise in single bursts during the periodic open phases. Drosophila melanogaster show mostly non-periodic release of CO₂ (see Fig. 2) without any reduction to zero release between peaks (see Fig. 2D). Does this mean that the gas exchange pattern observed is not a DGC? D. melanogaster are too small to measure O₂ uptake in single individuals, so we cannot obtain temporally correlated data on CO₂ release and O₂ uptake with the technology available.

Our evidence indicates that the CO₂ release pattern of fruit flies is responsive to the environment and to selection, and is apparently adaptive. As yet, we have little information regarding the mechanism of gas exchange control in these insects. Drosophila melanogaster have thoracic and abdominal spiracles, and large thoracic and cranial air sacs. The CO₂ peaks observed in the present study may be due to spiracular control or to forced ventilation through partially open spiracles. CO₂ release measured during the ‘closed’ phase of the periodic pattern seen in the D flies may represent leakage through partially open spiracles or leakage through the cuticle while the spiracles are closed.

Our limited knowledge of the respiratory physiology of this insect limits our understanding of what aspect of its physiology has responded to selection for resistance to desiccation. Casual observation of the D flies indicates they are less active than the C and O flies. It may be that this selection causes a decrease in activity and that quiescent flies show increased spiracular control. In addition, the D flies may have developed morphological differences such as changes in the spiracular muscles, or physiological differences such as increased capacity for buffering CO₂, increased tolerance for CO₂, or neural pathways that better control periodic release.

Even if the gas exchange patterns seen in Drosophila melanogaster are not regarded as discontinuous gas exchange because they are too far removed from the classical patterns, they are clearly capable of change under the influence of selection and they respond to hydration levels in the flies. Our future work will seek to elucidate the source of improvement of gas exchange pattern in the D populations.

In conclusion, we have established that ventilatory control increases after many generations of selection under desiccating conditions. We have also found changes in respiration within a single generation that show a strong link between the SER and hydration state. It is clear, however, that insects use many other methods to conserve water, including habitat choice, behavior, cuticular lipid type, absorption of water from excretory products and adjustment of hemolymph volume. The large variability in gas exchange patterns reported across different insect species is logical given this range of water conservation options. Only insects with potentially high rates of respiratory water loss would be expected to show a strong DGC. Our current data support the traditional hypothesis that the DGC has evolved as a mechanism to conserve water. We should be able to provide additional insights from future studies of the mechanism of CO₂ release in these insects and a determination of the relationship between water loss and the SER.

The authors would like to thank A. F. Bennett, A. Gibbs, R. K. Josephson and two anonymous referees for their comments on early manuscripts. We also thank A. Gibbs for providing additional equipment for these experiments. This research was carried out with support from NSF grant IBN 9507435 and NIH grant AG 09970.