Oxygen limitation and thermal tolerance in two terrestrial arthropod species

Ectotherms challenged by heat or cold reach limits to physiological function that are set by processes operating at the cellular, organ or whole-animal levels, or by combinations thereof (e.g. Hochachka and Somero, 2002; Pörtner, 2002; Hoffmann et al., 2003; Pörtner et al., 2006; Chown and Terblanche, 2007). Perhaps the most compelling case for thermal limitation at the whole-organism level has emerged from experimental studies that use various physiological indicators to link failure of whole-organism function to limitations in oxygen supply of marine species – the hypothesis of oxygen and capacity limited thermal tolerance (OCLT) (Pörtner et al., 1999; Frederich and Pörtner, 2000; Pörtner, 2001; Pörtner, 2002). This hypothesis can be described as follows: when ectotherms cool their metabolism slows, leading to insufficient ATP production in ventilatory muscles, thus reducing oxygen supply to the tissues. For example, as temperature is lowered, spider crabs Maja squinado show declining ventilation rates and haemolymph P_O2 concentrations. In this progression, the lower pejus temperature (T_pI) at which function is first compromised and loss of aerobic scope starts is followed by a complete switch to anaerobic metabolism – the lower critical temperature (T_cI), and finally by molecular and cellular cold damage at the damaging temperature T_d (Pörtner et al., 1999; Frederich and Pörtner, 2000; Pörtner, 2001; Pörtner, 2002; Pörtner and Knust, 2007). The sequence is mirrored during warming: the rate of increase of the cardiac and ventilatory processes that fuel metabolism starts to diminish at the upper T_pII and disappears at the upper T_cII as metabolism switches from aerobic to anaerobic. Finally, the upper T_d is marked by heat damage to proteins, membranes and cells (Pörtner et al., 1999; Frederich and Pörtner, 2000; Pörtner, 2002; Mark et al., 2002).

Pörtner and colleagues have suggested that OCLT theory applies to both aquatic and terrestrial ectotherms, and have called for tests of its predictions in terrestrial ectotherms. At least within the arthropods, several studies have responded to this call, either implicitly or explicitly. Bridges and colleagues examined oxygen consumption and heart rates of two scorpion species, Opisthophthalmus flavescens and Parabuthus villosus, during heating (Bridges et al., 1997). Neither species appeared to become oxygen limited when exposed to heat stress. Woods and Hill established that hawk moth eggs, Manduca sexta, become oxygen limited at high but not low temperatures, but that this is the consequence of diffusion limitation, protein instability or some unknown mechanism (Woods and Hill, 2004). Both of these tests were, essentially, unintentional investigations of the OCLT.

In an explicit test of oxygen limitation of upper thermal limits using activity-based critical thermal maximum (CT_max), Klok and colleagues (Klok et al., 2004) found that the CT_max of the tenebrionid beetle Gonocephalum simplex was independent of oxygen limitation during heat stress, and that metabolic rate in this species only decreased at a very low oxygen concentration (2.5%). These authors also found that the isopod Armadillidium vulgare became oxygen limited during heat stress, showing inhibition of metabolic rate at 10% oxygen and a decline in CT_max at more severe hypoxia (2.5% oxygen). Activity-based CT_max and metabolic rate of Drosophila melanogaster both declined under ambient hypoxia of <10% (Lighton, 2007). Taken together, this evidence suggests that oxygen limitation of CT_max occurs in some but not all terrestrial arthropods. In contrast, few (if any) studies have experimentally investigated oxygen limitation of the critical thermal minimum (CT_min), although Sinclair and colleagues (Sinclair et al., 2004) suggested that the oxygen limitation hypothesis of Pörtner and coworkers (Pörtner et al., 1999) may be applicable to insects during cold stress (see also Chown and Terblanche, 2007).

These explicit tests of OCLT (Klok et al., 2004; Lighton, 2007) both used manipulation of ambient oxygen concentration, accompanied by measurement of short-term changes in whole-animal metabolic rates during dynamic or ramped temperature changes, i.e. modification of a technique pioneered by Lighton and Turner termed ‘thermolimit respirometry’ (Lighton and Turner, 2004). Metabolic rate is estimated from either oxygen consumption or carbon dioxide production. Critical temperatures are usually derived from inflection points on such gas traces, and are associated with cessation of activity of the experimental animal assessed either using automated activity detection or visually. These critical limits are in essence the CT_max or CT_min of the individuals concerned. That is, they represent knockdown temperatures from which individuals usually recover fully. They are not necessarily equivalent to lethal limits, particularly at low temperatures (for reviews, see Hoffmann et al., 2003; Chown and Nicolson, 2004).

How close CT_max and CT_min are to T_c or T_p of the OCLT has not yet been fully clarified. However, because individuals can recover fully from knockdown without impairment to sensitive traits such as fertility (Hoffmann et al., 2003), it is clear that critical minima and maxima identified using movement are not equivalent to the T_d, as defined by Pörtner (Pörtner, 2001). Nonetheless, Klok and colleagues (Klok et al., 2004) equated the temperatures at which metabolism reaches minima (TMetMin) and maxima (TMetMax) with the start of respiratory breakdown – the lower (T_pI) and upper (T_pII) pejus temperatures of the OCLT. They also argued that the critical temperature of the OCLT (T_cII), at which metabolism switches from aerobic to anaerobic, corresponds to the pre-mortal valley evident on CT_max respirometry traces, which also coincides with a cessation of activity.

Given the small number of studies testing OCLT for terrestrial invertebrates, the significance of this proposed mechanism for understanding both short-term and evolutionary responses to temperature variation, and the importance of these responses in advancing the study of biogeography and climate variation (Lighton, 2007; Pörtner and Knust, 2007; Pörtner and Farrell, 2008), we examined OCLT in two model organisms by manipulating ambient oxygen concentrations in conjunction with thermolimit respirometry (Lighton and Turner, 2004; Klok et al., 2004; Lighton, 2007). We tested two predictions arising from the OCLT theory. First, if oxygen availability sets thermal limits, then hypoxia should cause contraction of the thermal range, manifested as increases in CT_min and TMetMin accompanied by declines in CT_max and TMetMax. This effect is likely to be more pronounced in species with separate circulatory and ventilatory systems (specifically the isopod, Porcellio scaber) than in those where gas exchange with the tissues is direct (specifically the insect, Tenebrio molitor) [see Klok et al. for rationale (Klok et al., 2004)]. Conversely, hyperoxia should alleviate oxygen limitation and cause expansion of the thermal range. Second, if thermal limits of these species are coupled by the common mechanisms proposed by the OCLT, then both critical temperatures should be affected by manipulation of ambient oxygen availability. If they are not, then oxygen availability is either not setting limits to thermal tolerance or is only one of several factors influencing such limits in terrestrial arthropods, setting them apart from the species for which this mechanism was originally proposed.

We emphasise that the study is not a two-species comparison: rather, we selected two arthropods as representatives of their respective widely different respiratory (oxygen delivery) systems: a tracheated insect, and a terrestrial crustacean that uses specialised appendages for diffusive gas exchange. We tested predictions arising from the oxygen limitation hypothesis separately upon each species.

Adult male Tenebrio molitor L. beetles, live mass (mean ± s.e.m.) 111.2±2.9 mg (CT_min experiments) and 109.5±2.9 mg (CT_max experiments), from a laboratory-bred colony kept for several generations and adult male Porcellio scaber Latreille isopods, live mass 65.1±4.4 mg (CT_min experiments) and 64.5±3.7 mg (CT_max experiments), collected in Stellenbosch, South Africa between May and August (late autumn and winter), were used. Isopods were identified using Hopkin (Hopkin, 1991). Males were used for both species following the finding that amphipod females (Gammarus pseudolimnaeus) were less tolerant of hypoxia than were males (Hoback and Barnhart, 1996). Individuals of both species were acclimated for at least 10 days in plastic containers in Labotec incubators (Labotec, Roodepoort, South Africa), with beetles at 25.0±1.0°C and isopods at 21.0±1.0°C, both with a 12 h L:12 h D cycle. Isopods were provided with soil, pieces of bark and leaves from their microhabitat. Because isopods are sensitive to desiccation (Cloudsley-Thompson, 1975), relative humidity inside containers was maintained at 94.7±8.3% using small cotton wool wads that were moistened twice daily. Relative humidity was verified with Hygrochron DS1923-F5 temperature/humidity iButtons with an accuracy of ±0.6% (Maxim/Dallas Semiconductor, Sunnyvale, CA, USA). Isopods were fed fresh vegetables, Pro-Nutro cereal (Bokomo Foods, Cape Town, South Africa) and decapitated mealworms; beetles were fed wheat bran, vegetables and fruit. Food was changed every second day. To eliminate the effects of specific dynamic action [reviewed for insects in Chown and Nicolson (Chown and Nicolson, 2004)], animals were fasted by removing them from the colony for 24 h before experimentation. Animals were weighed on a Mettler Toledo Analytical Ax504 balance accurate within 1 mg (Mettler Toledo Products, Greffensee, Switzerland) before every experiment described below.

In keeping with the approach adopted by Klok and colleagues (Klok et al., 2004), CT_min and CT_max were first determined visually in normoxia by placing individuals in water-jacketed Perspex tubes connected to a water bath (Grant LTC 12-50, R150, Grant Instruments, Cambridge, UK) programmed to increase or decrease temperature at a rate of 0.25°C min^–1 after an initial equilibration period of 10 min (accepted as a standard measure, acknowledging that variation in rates may affect the determination of limits) (see Terblanche et al., 2007; Chown et al., 2009). CT_min experiments for beetles started at 20°C and for isopods at 15°C; CT_max experiments for both species started at 30°C. The knockdown temperature, at which animals failed to respond to mild physical stimulation, was taken as the endpoint. Animals were not removed from the tubes for inspection [a misinterpretation by Hazell et al. of previous work using this method (see Hazell et al., 2008)], but rather were stimulated with a fine plastic dowel.

For all other experiments, thermolimit respirometry was used (sensuLighton and Turner, 2004) to determine CT_min and CT_max in animals exposed to 2.5% O₂ (severe hypoxia), 10% O₂ (mild hypoxia), 21% O₂ (normoxia) and 40% O₂ (hyperoxia), with oxygen balanced by nitrogen in each case, and verified using a calibrated Ametek S3A-II Oxygen Analyzer (AEI Technologies, Pittsburgh, PA, USA) (see below for description of respirometry system). Isopod CT_min was not determined in 10% O₂, since no effect of oxygen concentration was observed between 21% and 2.5% O₂. Between 5 and 17 individuals were used for each oxygen concentration treatment.

A flow-through respirometry system (see Lighton and Turner, 2004; Klok et al., 2004) and activity detector (AD1 Activity Detector, Sable Systems International, Las Vegas, NV, USA) were used to record metabolic rates and motor activity, respectively. For each trial, a pressurised gas cylinder (AFROX Ltd, Cape Town, South Africa) fed one of the gas mixtures described above into scrubbers: first soda lime to remove CO₂, then silica gel and Drierite to remove water vapour. Air then passed through mass flow controllers (0–500 ml model, Side-trak, Sierra Instruments Inc., Monterey, CA, USA) which regulated flow rate at 100 ml min^–1, through the zero channel of an infra-red gas analyser (LiCor LI-7000 ‘IRGA’, Lincoln, NE, USA), over the animal in a 3.3 ml plastic cuvette, and finally back through another channel of the IRGA which recorded the difference in CO₂ concentration in air before and after it passed over the animal, every 1 s. The IRGA data were logged to a standard desktop computer for subsequent transformation and analyses. This differential reading permitted calculation of CO₂ emission per unit time, using LI-7000 software (LiCor). Baselines (between 24 s and 18 min) were recorded before and after an experiment using an identical setup without an animal.

Temperature was controlled by immersing animals in their cuvettes in a water bath (Grant R150, Grant Instruments Ltd) programmed to change temperature at 0.25°C min^–1 (Lighton and Turner, 2004). For beetle CT_min and CT_max, starting temperatures were 20 and 30°C respectively, and for isopods, 15 and 25°C, with a 10 min equilibration period. Temperatures in the cuvette were recorded every 1s with thermocouples (type T, 36SWG) connected to a PicoTech TC-08 temperature data logger using Pico Data Acquisition Software (Pico Technology Ltd, Cambridge, UK).

During the determination of isopod CT_max, air was humidified to prevent desiccation stress. The system was identical to that described above, with the following two exceptions. First, air leaving the gas analyser was directed to a 500 ml plastic jar two-thirds filled with distilled water plus 1–2 ml of a 1 mol l^–1 aqueous solution of sodium hydroxide to prevent CO₂ in the water from diffusing into the air stream. The plastic jar was placed in a water bath (Grant LTC 12-50, R150, Grant Instruments Ltd) set to start at 15°C (10°C cooler than the temperature of the animal) and ramp up at a rate of 0.35°C min^–1 (0.1°C faster than the ramping rate experienced by the animal). The humidified air then passed through a plastic coil of Bev-A-Line tubing (Thermoplastic Processes Inc., Georgetown, DE, USA) before passing over the animal in its cuvette, both inside a second water bath. The relative humidity inside the cuvette (77.8±0.9%) was verified with a HM34 relative humidity meter (accuracy ±2% RH; Vaisala Industrial Instruments, Vantaa, Finland). Second, air leaving the animal passed through a scrubber containing re-activated silica gel and re-activated Drierite [to improve response times, (White et al., 2006)] to remove water vapour before the air entered the LI-7000 IRGA, and to prevent condensation within the gas analyser cell.

Thermolimit respirometry data, extracted from LI-7000 software, were converted to μl CO₂ h^–1 and baselines were drift-corrected using ExpeData analytical software, version 1.0.16 (Sable Systems International). Activity and temperature data were then plotted with drift-corrected data.

The identification of critical thermal limits using visual observations of knockdown temperatures for individual animals is straightforward (e.g. Klok and Chown, 1997) (for a review, see Hoffmann et al., 2003). However, use of thermolimit respirometry combined with infrared activity detection to identify these same limits necessitates careful interpretation of combined time series data from gas analysers and activity detectors. Given the wealth of information on both metabolic state and gas exchange patterns provided by this last method, we used our data to address this challenge by comparing different methods of CT_min and CT_max determination [see discussions in Lighton and Turner (Lighton and Turner, 2004), Klok et al. (Klok et al., 2004) and Lighton (Lighton, 2007)].

We assessed four methods of estimating CT_min and CT_max. The first of these uses voluntary activity. We used the activity absolute difference sum (ADS) inflection estimates suggested and discussed comprehensively by Lighton and Turner (Lighton and Turner, 2004). Activity readings (in mV) comprise both positive and negative values, because the signal oscillates around zero. The ADS is the cumulative sum (over time) of the absolute differences between adjacent activity readings. The inflection point of this cumulative time series coincides with the cessation of voluntary motor control. To determine this inflection objectively, a period (3–8 min) around the inflection point of the activity ADS cumulative time series is selected, and an ordinary least squares linear regression fitted to the activity ADS data points, with time as the independent variable. The highest consecutive five residuals of this regression are identified, and the mean temperature corresponding to these five data points taken as the inflection point that represents the CT_min or CT_max (Lighton and Turner, 2004). These calculations were done using an Expedata macro (C. J. Klok and A. Kaiser, personal communication). CT_min or CT_max estimates made in this way are referred to throughout as aADS [see supplementary material Fig. S1, and figure 3 in Lighton and Turner (Lighton and Turner, 2004)].

For the second method, we used the same procedure on ADS residuals to identify inflection points that we refer to throughout as VI ( inflection) CTs (for explanation and example of calculation, see supplementary material Fig.S2). In this case, the selected period stretched from the pre-mortal valley to the start of the post-mortal peak (sensuLighton and Turner, 2004). Lighton and Turner (Lighton and Turner, 2004) found activity ADS-derived CT_max values (here, aADS CT_max) to be unbiased predictors of ADS-derived values (here, VI CT_max). The aADS and VI methods estimate inflection points using regressions fitted to time series data, for which consecutive data points are not independent of one another. Moreover, both activity and CO₂ emission do not necessarily increase linearly with time, as high activity may be interspersed with periods of inactivity. Given that non-independence and non-linearity of individual data points violate the assumptions of an ordinary least squares linear regression (Quinn and Keough, 2002), we compared these methods with two further methods for estimating critical limits.

Our third method, which we call the pre-mortal valley (PMV) method, used the pronounced burst of CO₂ release, the post-mortal peak or ‘ghost breath’ observed in ants (Lighton and Turner, 2004), tenebrionid beetles (Klok et al., 2004) and flies (Lighton, 2007). This burst signifies the loss of spiracular control. For Tenebrio molitor, we identified CT_min and CT_max and for isopods only CT_max for each individual as the temperature corresponding to the lowest point in the valley of CO₂ release just before the post-mortal peak. This method was not used on P. scaber CT_min data as we did not observe any brief spikes preceding the freezing point of these animals. This inflection point most closely corresponds to the T_c of the OCLT (Klok et al., 2004).

Fourth, because we observed a conspicuous and sudden CO₂ release in beetle CT_min traces which seemed to indicate loss of muscular control, we sought to identify the start of this burst by calculating the 95% confidence intervals (CI) around the mean for the 5 min period preceding it. We identified the first five points that fell outside the CI, and took the mean of the corresponding five temperatures as the CT_min. This method enabled us to identify a sudden deviation from a steady rate of CO₂ release. It could not be used for CT_max experiments, because both beetle and isopod increased continuously with increasing temperature, showing no plateau period for estimation of 95% CI.

Finally, within each trace we identified minimum and maximum metabolic rates (Min. and Max. ) by using ExpeData to find the lowest and highest five points in the CO₂ traces, respectively. The mean temperatures corresponding to these 5 s periods were identified as TMetMin and TMetMax, respectively, approximating the lower (T_pI) and upper (T_pII) pejus temperatures at which function worsens as metabolism becomes anaerobic and unsustainable (Frederich and Pörtner, 2000; Klok et al., 2004). We considered these pejus rather than critical temperatures because they occurred prior to the critical temperatures and therefore probably represent the onset of deleterious conditions.

The gas analysers (LI-7000) that we used have a maximum drift accuracy of 0.1 p.p.m. °C^–1 (LI-7000 CO₂/H₂O Analyser Instruction manual, LiCor). at or below these drift values might potentially be indistinguishable from equipment noise (see also Lighton, 2007). During cooling, CO₂ release remained below the drift threshold value and was thus indistinguishable from equipment noise in the case of eight isopod individuals (one in the 40% O₂ group, three in the 21% O₂ group and four in the 2.5% O₂ group). These individuals were excluded from analyses comparing minimum (Min. ) but included in analyses comparing CT_min ascertained by aADS. For all beetles, CO₂ release during cooling was well above the noise level of the gas analyser for all individuals; the same was true for CO₂ release of both species during CT_max experiments.

Within each oxygen concentration, Shapiro–Wilks tests revealed that most thermal limits identified using motor and respiratory activity were not normally distributed. Therefore, values obtained with the different methods (aADS, VI, PMV and CI methods) were compared using Wilcoxon paired tests, with P-values adjusted to correct for multiple pairwise comparisons (Hochberg and Benjamini, 1990). On individual traces, we ascertained whether TMetMin and TMetMax always occurred in the same position relative to the PMV critical temperatures (equivalent to the T_csensu the OCLT) using Wilcoxon paired tests. Mann–Whitney U-tests (likewise corrected for multiple comparisons) were used to determine whether thermolimit critical temperatures derived from the different methods differed from knockdown temperatures. Pairwise tests with this powerful correction were deemed more appropriate than a Kruskal–Wallis ANOVA because critical temperatures determined using activity or various inflection points on traces are all derived from the same data, and hence are not independent of each other. This violates an assumption of ANOVA.

Between oxygen treatments, we compared each of the measured traits (CT_min, CT_max, Min. and Max. , and TMetMin and TMetMax) using generalised linear model (GLZ) ANCOVAs with body mass as the covariate, an assumed normal error distribution, and an identity link function. Separate- or homogenous-slopes models were chosen as appropriate, and significant pairwise differences between oxygen concentrations identified by non-overlapping confidence intervals of weighted marginal means. If body mass did not prove a significant covariate, we used Kruskal–Wallis ANOVAs followed by multiple comparisons of mean ranks. All statistical analyses were carried out in Statistica 8.0 (Statsoft Ltd, Tulsa, OK, USA). Unless otherwise stated, values reported are means ± 1 s.e.m.

We assumed that the first 30 min of each CO₂ trace constituted an equilibration period for gases moving through the Drierite column; therefore, we removed this period from all traces. Beetle CT_min traces showed a spike in CO₂ release (Fig. 1), observed in 10 out of 12 (40% O₂), 14 out of 14 (21% O₂), 15 out of 15 (10% O₂) and 12 out of 13 (2.5% O₂) traces. This spike corresponded closely with the visually determined knockdown temperature of this species and was assumed to indicate CT_min. It was absent from isopod traces (Fig. 2).

During CT_max experiments, beetle increased with temperature, followed by a smooth CO₂ decline, then increased again to form a sharp second peak (Fig. 3). Isopod increased until a peak was reached, whereafter CO₂ declined, and this was followed by a second peak (Fig. 4).

To compare thermolimit critical temperatures with those obtained during the knockdown method, only thermolimit data collected in normoxia were used. For beetles, thermolimit respirometry CT_min values obtained with the aADS and PMV methods were significantly higher than knockdown temperatures, as were isopod CT_min values estimated with aADS (Table 1). The VI, PMV and CI methods could not be applied to isopod CT_min data because these animals lack the post-mortal peak in CO₂ release. Reduced spiracular activity in species with spiracles also means that the VI method may prove less reliable than others because of the lack of any point on the trace that can be associated clearly with the change in the ADS (supplementary material Fig. S2).

For both species, CT_max values obtained with the thermolimit methods were lower than knockdown values, except in the case of isopod PMV CT_max values (Table 1).

We compared the three different thermolimit methods with each other at each of the oxygen concentrations (Tables 1 and 2) and found the following: beetle CT_min values obtained by aADS were significantly higher than those obtained with the VI and PMV methods at all oxygen concentrations, and were significantly higher than the CI method estimates for all but 2.5% oxygen. Beetle CT_min estimates using PMV and CI did not differ, so applying either to CT_min data is probably appropriate. For both species, CT_max estimates obtained with aADS and VI did not differ from each other in 10%, 21% and 40% O₂, but were lower than those obtained with the PMV method. At 2.5% O₂, cessation in activity (aADS) preceded the decline in metabolic rate, and CT_max values of all three methods differed from each other.

Paired Wilcoxon tests corrected for multiple comparisons (Hochberg and Benjamini, 1990) showed that within individual respirometry traces, TMetMin and TMetMax consistently occurred at less extreme temperatures than did the PMV CTs (Table 1 footnote). The exception to this pattern was isopod TMetMax in 2.5% O₂, which occurred at a higher temperature than did PMV CT_Max. Within each oxygen treatment, the range of metabolic function (TMetMax – TMetMin) did not change systematically with oxygen concentration: for beetles in 40%, 21%, 10% and 2.5% O₂ it was 36.3, 38.7, 37.2 and 35.1°C; and for isopods in 40%, 21% and 2.5% O₂ it was 43.7, 42.2 and 42.5°C. Formal statistical comparisons were not possible because these values are calculated by the subtraction of means from each treatment.

Neither species appeared to be oxygen limited during cooling. CT_min was unaffected by oxygen manipulation in beetles when determined with the aADS (Wald χ²=1.5, P=0.7), VI (χ²=4.5, P=0.2), PMV (χ²=0.6, P=0.9) and CI methods (χ²=4.0, P=0.3). The same was true for isopod CT_min determined with aADS (Wald χ²=1.8, P=0.4) (see also Table 1). Minimum metabolic rate (Min. ) of both species was unaffected by oxygen manipulation (Wald χ²=4.0 and 5.3, P=0.3 and 0.07 for beetles and isopods, respectively), and the same was true for TMetMin (Wald χ²=3.2 and 2.4, P=0.36 and 0.31 for beetles and isopods, respectively).

In contrast, CT_max was reduced by extreme hypoxia in both species. For beetles, CT_max values in 2.5% O₂ were significantly lower by 6.9°C (aADS), 3.9°C (VI) and 3.6°C (PMV) than in 21% O₂ (Wald χ²=204.6, 38.5 and 15.4, respectively, P<0.000001 for aADS and VI, P<0.004 for the PMV method). This difference was more pronounced in isopods, with CT_max in 2.5% O₂ lower than normoxic values by 10.5°C (aADS), 6.5°C (VI) and 5.8°C (PMV) (Wald χ²=129.1, 81.9 and 105.2, respectively, P<0.00001 in all cases). Moreover, for isopods, aADS and VI CT_max in 2.5% O₂ were also significantly depressed relative to corresponding values in 10% O₂. For beetles, maximum metabolic rate (Max. ) was significantly depressed at 2.5% O₂ relative to that at all other oxygen levels (Wald χ²=47.1, P<0.00001). For isopods, Max. at 2.5% O₂ was only significantly depressed relative to that at 21% O₂ (Wald χ²=12.6, P<0.05). TMetMax for beetles was significantly lower in 2.5% O₂ than in normoxia (Wald χ²=18.7, P<0.0005), but was unaffected by oxygen concentration in isopods (Wald χ²=7.1, P>0.07). See Table 1 for all actual values of the above variables.

When animals enter chill-coma they lose motor activity because of slowing of the movement of potassium and calcium across cell membranes through voltage-gated channels (Mellanby, 1939; Esch, 1988; Goller and Esch, 1990; Hosler et al., 2000; Kovac et al., 2007). The slowed ion movement causes muscle action potentials progressively to decrease in amplitude and increase in duration until they are abolished (Esch, 1988; Goller and Esch, 1990; Hosler et al., 2000). Cooling an animal to its chill-coma temperature causes a sudden burst in muscle action potential (Hosler et al., 2000), as observed in fruit flies, honey bees, bumblebees, wasps, beetles, flies, butterflies and moths (Esch, 1988; Goller and Esch, 1990; Hosler et al., 2000). During our cooling experiments, most beetle metabolic rate traces showed a prominent spike in followed by a sharp decline. This spike preceded the freezing point of these animals, previously found to be –16.8±2.5°C (M.M.S., S.J., S.A.B., J.S.T. and S.L.C., unpublished data), and corresponded closely with the knockdown CT_min of these animals. We therefore suggest that spiracles steadily lose their ability to close, allowing an efflux of CO₂ (see also Kovac et al., 2007). Then, as the animal enters chill-coma, spiracular control is lost, as indicated by a burst in muscle action potential. Relaxation of spiracular muscles typically (but not always) causes spiracles to open, rather than close (Chapman, 1998). If this is so, then the burst in muscle action potential should either slightly precede or coincide with the spike in CO₂.

During heating, CO₂ release forms two prominent peaks (see Lighton and Turner, 2004; Klok et al., 2004; Lighton, 2007). The first appears to be a consequence of an increase in the generation of CO₂ as metabolic rate rises with temperature. After this peak, insects are said to lose control over spiracular activity, and spiracles open, releasing a burst of CO₂ (Klok et al., 2004). In non-tracheated isopods, this second post-mortal CO₂ peak (Lighton and Turner, 2004) may result from the release of buffered CO₂ from body fluids, or enhanced mitochondrial activity (Wigglesworth, 1972; Lighton and Turner, 2004; Lighton, 2007). The second peak was sudden and steep in beetles and gradual for isopods, which exhibited smooth, wide peaks. The differences probably result from slower movement of this gas through isopod haemolymph than through the air-filled tracheae of the beetle.

Alternatively, the fact that CO₂ from isopods passed over a desiccant before entering the gas analyser may also have affected CO₂ traces for these animals. When damp, silica gel and Drierite have an affinity for CO₂ (Withers, 2001; White et al., 2006). The extent of this affinity can be investigated by estimating the washout of CO₂ from both the cuvette and the scrubber column (see Withers, 2001; White et al., 2006). If the washout from the cuvette is faster than that from the scrubber column, the risk of CO₂ build-up inside the latter is possible, thus diluting the animal's CO₂ release pattern and resulting in smoother traces (Gray and Bradley, 2006). The washout times from both the cuvette and scrubber column were calculated, and found to be approximately 9 and 124s, respectively. It is thus possible that the CO₂ signal from isopods was smoothed, removing transient CO₂ peaks.

For both species, CT_min and CT_max determined with thermolimit methods (activity- and -based ADS residuals, PMV and CI methods) were higher and lower, respectively, than the corresponding knockdown counterparts. The differences between knockdown and thermolimit-determined critical temperatures can be attributed to the fact that the animals were encouraged to move during the former method. The temperature at which the animals no longer responded to touch was recorded as the knockdown temperature. Conversely, CT_min and CT_max derived from thermolimit respirometry data represent voluntary activity, in the absence of stimulation. Although it is known that starting temperature may affect critical thermal limits (Terblanche et al., 2007), this effect was not important in the present study, as knockdown temperatures differed from thermolimit-determined critical temperatures irrespective of whether starting temperatures differed between methods or not.

Beetle CT_min values determined with aADS were higher than those determined with -based thermolimit methods (VI, PMV and CI methods). This is unsurprising, since voluntary movement often ceases before the onset of chill-coma (Mellanby, 1939). By contrast, cessation of voluntary activity during heating only preceded events in severe hypoxia (2.5% O₂). Animals generally displayed low activity during exposure to 2.5% O₂, probably because activity increases metabolic rate, and hence oxygen demand. Depression of activity was also observed in D. melanogaster exposed to oxygen concentrations below 5% during heating (Lighton, 2007).

Here, we have shown that within one species, critical thermal limits estimated with different methods may differ by as much as ∼5 and 6°C for CT_min and CT_max, respectively. Because the method adopted may influence the outcome of measures of thermal tolerance in animals (Terblanche et al., 2007; Chown et al., 2009), careful consideration needs to be given to the measure to be used, which should always be carefully reported. For analysis of respirometry data, we favour the use of either the PMV or CI methods rather than activity-based ADS residuals, which reflect cessation of voluntary movement while stimulated movement may still be possible. Moreover, unpublished work shows that voluntary activity in tsetse flies (Glossina spp.) does not change distinctly as temperature increases, and no reliable CT_max can be estimated from aADS (C. J. Klok, J.S.T. and S.L.C., unpublished data). By contrast, the PMV and CI methods probably reflect loss of spiracular control, and also show closer correspondence with visually determined knockdown temperatures than does the aADS method.

Isopods, which lack spiracles, produced CO₂ traces without sharp bursts. It was therefore difficult to use -based ADS residuals to determine CT_min for these animals. This highlights the need for a method of CT_min estimation for animals exchanging gases by diffusion through specialised regions of their cuticle, such as pleopod exopodites, rather than through tracheae. In such cases, verification of endpoints could be obtained using visual recordings (filming) of animal behaviour to observe onset of spasms or loss of the righting response.

Thermolimit CT_max has been defined as loss of spiracular control at the end of the plateau phase preceding a plummet in CO₂ release (Lighton and Turner, 2004; Klok et al., 2004). However, if loss of spiracular control coincides with a final opening of spiracles as the closer muscles of spiracles fail (Chapman, 1998), this endpoint should rather occur at the base of the post-mortal peak. It is therefore suggested that the PMV method of determining CT_max be favoured for animals that show twin peaks in CO₂ release during heating.

Strikingly, and irrespective of the estimation method employed, all CT_min estimates and TMetMin in both T. molitor and P. scaber were unaffected by changes in ambient oxygen concentration of more than an order of magnitude, from 40% to 2.5%. Moreover, thermal ranges did not contract in hypoxia. These findings suggest that over short time scales and in resting animals, oxygen delivery to the tissues is unlikely to set thermal limits fully in these species. Thus the hypothesis of OCLT (Pörtner et al., 1999; Frederich and Pörtner, 2000; Pörtner, 2001; Pörtner, 2002) is not fully supported by the current data.

Low temperature slows both aerobic metabolism (oxygen demand) and ventilation (oxygen supply) (Pörtner, 2002; Storey and Storey, 2004). Upon cooling, metabolic requirements in both study species were probably minimal, so the animals did not become effectively oxygen limited. Minimal metabolism can be accomplished and maintained during cooling by either abandoning or minimising metabolic control (Makarieva et al., 2006). Freeze-avoiding species, such as T. molitor and P. scaber (E. Marais and S.L.C., unpublished data), that show a gradual decline in metabolic rate as temperature decreases and maintain metabolic rate above baseline values, are in keeping with the minimum metabolic control strategy. Animals thus show minimal metabolic activity during cooling that is independent of oxygen availability, a finding that does not support the OCLT hypothesis.

Both study species became oxygen limited during heating in extreme (2.5% O₂) hypoxia, illustrated as a decline in CT_max, T_MetMax (for beetles only), and Max. . The magnitudes of the declines in various CT_max estimates in isopods were approximately twice the corresponding values for beetles, as found previously (Klok et al., 2004). Beetle thermal tolerance thus appears to be less sensitive to hypoxia than is the case for isopod thermal tolerance, as predicted to some extent by the OCLT (Pörtner, 2001; Pörtner, 2002). These arthropod groups have structurally and functionally different gas exchange systems (Klok et al., 2004). Beetles take up air through spiracles; the air then fills the trachea and is transported in gaseous form to the tissues where oxygen is unloaded (Loudon, 1989). Isopods, on the other hand, absorb oxygen across the integument of the pleopod exopodites (Cloudsley-Thompson, 1975). Oxygen entering the haemolymph is bound to haemocyanin, with a small fraction dissolved in the haemolymph, and then transported to the tissues (Cloudsley-Thompson, 1975). Klok and colleagues (Klok et al., 2004) suggested that the additional binding step of oxygen to haemocyanin may delay oxygen delivery to the tissues in isopods and in other marine invertebrates with similar gas exchange systems. Indeed, gas exchange systems of the isopods Ligia occidentalis and Alloniscus perconvexus limited these animals in normoxia, since mild hyperoxia (25% O₂) caused increased oxygen uptake (Wright and Ting, 2006). Additionally, oxygen is known to diffuse faster through air than through water (Prosser, 1973), and thus probably faster to the tissues of beetles than to those of isopods. Beetles would therefore have more efficient gas exchange systems than isopods. Differences in gas exchange system may then be one of the reasons why the beetles in this study better tolerated stressful temperatures and hypoxia than did the isopods.

For both a representative of arthropods with single-step tracheal gas exchange systems and a representative terrestrial crustacean with a two-stage system involving oxygen absorption by cuticular diffusion followed by transport in the haemolymph, upper thermal limits are influenced by ambient oxygen availability under extreme conditions, but lower thermal limits are not. This outcome is in keeping with studies suggesting that thermal maxima and minima in terrestrial invertebrates are decoupled (Chown, 2001; Hoffmann et al., 2005). It also suggests that OCLT (Pörtner, 2002) is perhaps not as significant a mechanism for limiting thermal tolerance in terrestrial species as it appears to be for their marine counterparts. However, the precise nature of the relationship between CT_min, CT_max, TMetMin and TMetMax on the one hand, and the pejus and critical temperatures discussed by Pörtner (Pörtner, 2001; Pörtner, 2002) on the other, requires further scrutiny before this outcome can be more widely generalised. Doing so, and extending investigations to other terrestrial ectotherms by using both short- and long-term experiments to measure several performance and biochemical traits, would seem to be the most promising avenues for further exploring the relevance of Pörtner's (Pörtner, 2001; Pörtner, 2002) important idea to terrestrial organisms.

We thank Elrike Marais and Jacques Deere for their assistance, and C. J. Klok and A. Kaiser for providing us with their thermolimit macro. Erika Nortje, Henry Davids and Charlene Janion provided valuable assistance in the laboratory. The referees provided incisive and helpful reviews for which we are grateful.