604

lo-J r cofLj - ck (Dr rem Dv~

Sampling Bias in Respirometry

Jack PHayes John R Speakman Paul A Racey

Department of Zoology University of Aberdeen Aberdeen AB9 2TN

United Kingdom

Accepted 11991

Abstract Comparisons of the metabolic rates ofspecies populations and treatment groups ofanimals are common Houieuer the data used in tbese comparisons may not be truly equivalent e report tbe effects of uarying (1) total time from which minimum metabolism is selected and (2) time over which metabolic rate is calshyculated (calculation interval) on estimates ofoxygen consumption for Microtus

agresus and Apodemus sylvaricus Oxygen consumption was measured at 10deg 20deg and 30deg G using open-circuit respirometry The lowest 15 min ofmetabolism were ]3 and 65 bigber for Microtus and Apodernus respectiuely uiben seshylected from a total monitoring period of30 min than toben selectedfrom a peshyriod of6 h Tbis demonstrates tbe importance cfstandardizing tbe duration of time from tobicb minimal estimates of metabolism are selected (t e tbe amount of time for which metabolism is measured) For Microtus minimum oxygen COIshy

sumption was 12 higher taben calculated ouer 60 mill tban u-ben calculated over 15 min We suggest that analyses relying on minimal (og basal) metabolic rates include calculation interval as a potential covariate

Introduction

Many factors (eg mass thermal environment feeding status reproducshytive state) affect the aerobic metabolism of animals (Peters 1983 Calder 1984 Schmidt-Nielsen 1984) Thermal environment and feeding status (fasted or not) are typically controlled when respirometry is used to meashysure the aerobic metabolic rates of individual animals (Bartholomew

bull Address correspondence to EG amp G Energy Measurements Inc Applied Ecology Department PO Box 127 Tupman California 93276

PhySIOlogical Zoology 65(3)604-619 1992 copy 1992 byThe University of Chicago All rights reserved 0031-935Xj92j6503-9141$0200

Respirometry Methods 605

1972 Calder 1984) Other variables such as amount of time between

handling animals and starting measurements and circadian and ultradian fluctuations (Aschoff and Pohl 1971 Heusner Roberts and Smith 1971

Prothero 1984) are less commonly controlled and may account for sigshynificant differences between measurements on different species (Kenagy and Vleck 1982) Despite the cautions that most researchers observe

there are two aspects of methodology in respirometry studies that are in need of careful analysis

It is not uncommon to read that metabolic determinations were made

until a low stable value was obtained The rationale behind this approach is to remove the metabolic cost of activity from the estimated value From a statistical perspective this censoring of data is a dubious practice What constitutes a low or stable value is subjective and potentially a source of substantial bias The longer an animals metabolism is monitored the

lower the animals metabolism for any set interval (eg lowest 5-min meshytabolism) is likely to be Clearly after determining the lowest 5-min metshyabolic rate for 1 h further measurement can only result in an equal or lower value Thus resting measurements obtained from varying the total length of monitoring of an animal are inappropriate The effect of varying total observation time will depend on many factors (eg species circadian and ultradian rhythms variation in the periods of total observation) However the effect of varying lengths of observation has yet to be established for any species

Another area of concern with respect to the comparability of data is the interval over which metabolic rate is calculated For example Dawson and Olson (1987) distinguished between summit metabolism (a term they asshycribed to maximal cold-induced O2 consumption over a period of more than 2 h) and peak metabolism (the maximal cold-induced O 2 consumption over a period of just a few minutes) Peterson Nagy and Diamond (1990) address the problem further when they review a number of studies in an attempt to define the highest sustainable metabolic rate Calculations of metabolism over different time scales are likely to yield different results (see eg Chappell 1984) although in some cases this may result from artifacts of measurement protocols rather than from real biological differshyences (see Hayes 1989)

In this article we address two issues (1) the statistical bias associated with censoring data or biased sampling and (2) the effect of varying the amount of time over which measurements are calculated We also examine the efficacy of one approach used to control for circadian andor ultradian variations in metabolism

606 J P Hayes J R Speakman and P A Racey

Material and Methods

Study Animals

Short-tailed field voles (Microtus agrestis Linnaeus) were obtained from the wild in Grampian Region United Kingdom (SrN) during May 1990 The voles were trapped in Longworth traps baited with carrot and supplied with hay The voles were obtained from three areas separated by at least 10 km On the day of capture voles were returned to the Department of Zoology where they were housed in plastic rodent cages Males and females were housed separately Typically five voles were kept per cage but in a few cages there were fewer than five voles Wood shavings for bedding and light straw for nesting material were provided Rodent chow (Special Diet Service CRM Diet) and water were provided ad lib Fresh cabbage was supplied at least once per week Voles were housed at room temperature (ca 22degC) on a 14LI0D photoperiod

Respirometry Protocol

Main Trials Oxygen consumption was measured with an open-Circuit resshypirometry system At each of three temperatures 10deg 20deg and 30degC three randomly selected males and three randomly selected females were meashysured We attempted to obtain measurements at OdegC as well but not all voles could sustain normothermia at Odegc Thus each of 18 voles was meashysured once During measurements the voles were held in a Perspex (Plexishyglas) chamber 9 ern X 8 ern X 7 em A perforated Perspex partition restricted the voles movement to a length of 65 em The vole was placed in the chamber over a wire-mesh grid to keep it separated from voided urine and feces

An upstream flowmeter (Alexander Wright 02S-L water displacement) was modified so that every time SOO mL had passed through the flowmeter a voltage signal was generated The time interval between signals was used to calculate the flow rate Air was dried immediately before and after the flowmeter All flow rates were corrected to STPD (OdegC 760 Torr) Mean flow ranged from 468 to 694 mLmin The metabolism chamber was placed inside a controlled-temperature cabinet (Gallenkamp) that regulated the internal ambient temperature (T plusmn01 degC) Inside the controlled-temperature cabishynet and upstream of the metabolism chamber there was a length of copper tubing about 6 mm in diameter and 1 m long that ensured incoming air was equilibrated with the T in the cabinet

Respirometry Methods 607

The excurrent air from the metabolism chamber was connected to a large plastic syringe barrel A small-diameter tube inserted in the syringe barrel was used to subsample the excurrent air Carbon dioxide was removed with Carbosorb (BDH Poole UK active agent NaOH) and water was removed with silica gel The O2 concentration in the excurrent air was measured with an Applied Electrochemistry S3A dual-channel O2 analyzer A parallel line of reference air was sent to the other channel of the O2 analyzer to

minimize drift over the course of the measurements The difference between the O 2 content of the reference and animal air was continuously monitored with a BBC microcomputer and the mean recorded every 5 s

Each vole was weighed to plusmn001 g before being placed in the metabolism chamber The animals were not fasted before measurement but no food or water was available in the chamber Voles were placed in the metabolism chamber between 0707 and 1019 in the morning Data recording started after the animals had been in the chamber for about 1 h Measurements were continued for at least 6 h Voles were weighed again at this time

Auxiliary Measurements Additional measurements of O2 consumption were made in a larger metabolism chamber that allowed more scope for voluntary activity The setup was identical to that already described except that the metabolism chamber (8 cm high X 7 cm wide) was divided into three secshytions of length 25 ern 40 ern and 25 ern respectively and that flow rates ranged from 685 to 725 mLmin The vole was placed in the central large section and the smaller sections were filled with silica gel All measurements with this chamber were made at 20degC Six randomly selected female voles (excluding the voles previously used) were studied We also measured two Apodemus syluaticus (one male and one female) at 20degC Each Apodemus was measured in both the small and the large metabolism chamber following the same procedures as used for Microtus The female Apodemus was slightly sluggish after the measurement in the small metabolism chamber so her body temperature was measured It was slightly depressed (34degC)

Calculations

Calculation Interval We calculated steady-state O2 consumption using the appropriate equation from Hill (1972) For measurements made in the small metabolism chamber O2 consumption was calculated for the lowest 15 20 25 30 35 40 45 50 and 55 min of each hour and for the full hour itself This was repeated for each successive full hour for which data were obtained Thus 60 values (6 h X 10 intervals) were calculated for each individual

608 J P Hayes J R Speakman and PA Racey

For the large chamber which had a slower washout O 2 consumption was calculated for each full hour only

Sampling Bias For each animal the lowest 15-min interval of O 2 conshy

sumption occurring in the first 30 45 60 90 180 and 360 min of measureshyment was calculated

Subsampling of Runs If animals show pronounced ultradian or circadian rhythms (see Aschoff and Pohl1971 Heusner et al 1971 Kenagy and Vleck

1982) then it is important that measurements be made at the same phase of the cycle or that all phases of the cycle be sampled so that equivalent

data are obtained Devoting an entire day to measuring a single animal throughout a cycle will severely limit the number of animals that can be measured One approach to controlling for rhythmic variation in metabolism

but one that still uses analyzer time effectively is to set up an automated system and measure a number of animals simultaneously with each animal actually being monitored for a fraction of the total time (eg for a few

minutes every hour see Hayes Garland and Dohm 1992) The utility of this approach however depends on a number of factors including the

amplitude and period (length) of the rhythm the duration of the measureshyment and the timing of the measurement relative to the phase of the rhythm

To study the effect of subsampling we exhaustively subsarnpled the

complete metabolism recording for each animal as follows Oxygen conshysumption for the first 15 min of each of the 6 h was calculated Then the lowest of the six values was retained This procedure was repeated for all possible 15-min intervals of each hour For example O2 consumptions beshytween 003525 and 005025 013525 and 015025023525 and 025025

and 053525 and 055025 (hrnins time relative to the start of meashysurements) were calculated and the lowest of the six values determined

Next the lowest of the six values from between 003530 and 005030 01 3530 and 015030 023530 and 025030 and 053530 and 0550 30 was determined Finally the mean of all the lowest values was calculated

(hereafter referred to as the subsampling mean) The procedure we used to calculate the subsampling mean is biased because the first and last 15 min of each hour are sampled less intensively than the middle 30 min of each hour Despite this bias we think the method we used to calculate the subsampling mean is better than one based on randomly selecting IS-min

intervals within the hour because it more accurately reflects the way that measurements are likely to be collected if an automated subsampling system

is used

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 605

1972 Calder 1984) Other variables such as amount of time between

handling animals and starting measurements and circadian and ultradian fluctuations (Aschoff and Pohl 1971 Heusner Roberts and Smith 1971

Prothero 1984) are less commonly controlled and may account for sigshynificant differences between measurements on different species (Kenagy and Vleck 1982) Despite the cautions that most researchers observe

there are two aspects of methodology in respirometry studies that are in need of careful analysis

It is not uncommon to read that metabolic determinations were made

until a low stable value was obtained The rationale behind this approach is to remove the metabolic cost of activity from the estimated value From a statistical perspective this censoring of data is a dubious practice What constitutes a low or stable value is subjective and potentially a source of substantial bias The longer an animals metabolism is monitored the

lower the animals metabolism for any set interval (eg lowest 5-min meshytabolism) is likely to be Clearly after determining the lowest 5-min metshyabolic rate for 1 h further measurement can only result in an equal or lower value Thus resting measurements obtained from varying the total length of monitoring of an animal are inappropriate The effect of varying total observation time will depend on many factors (eg species circadian and ultradian rhythms variation in the periods of total observation) However the effect of varying lengths of observation has yet to be established for any species

Another area of concern with respect to the comparability of data is the interval over which metabolic rate is calculated For example Dawson and Olson (1987) distinguished between summit metabolism (a term they asshycribed to maximal cold-induced O2 consumption over a period of more than 2 h) and peak metabolism (the maximal cold-induced O 2 consumption over a period of just a few minutes) Peterson Nagy and Diamond (1990) address the problem further when they review a number of studies in an attempt to define the highest sustainable metabolic rate Calculations of metabolism over different time scales are likely to yield different results (see eg Chappell 1984) although in some cases this may result from artifacts of measurement protocols rather than from real biological differshyences (see Hayes 1989)

In this article we address two issues (1) the statistical bias associated with censoring data or biased sampling and (2) the effect of varying the amount of time over which measurements are calculated We also examine the efficacy of one approach used to control for circadian andor ultradian variations in metabolism

606 J P Hayes J R Speakman and P A Racey

Material and Methods

Study Animals

Short-tailed field voles (Microtus agrestis Linnaeus) were obtained from the wild in Grampian Region United Kingdom (SrN) during May 1990 The voles were trapped in Longworth traps baited with carrot and supplied with hay The voles were obtained from three areas separated by at least 10 km On the day of capture voles were returned to the Department of Zoology where they were housed in plastic rodent cages Males and females were housed separately Typically five voles were kept per cage but in a few cages there were fewer than five voles Wood shavings for bedding and light straw for nesting material were provided Rodent chow (Special Diet Service CRM Diet) and water were provided ad lib Fresh cabbage was supplied at least once per week Voles were housed at room temperature (ca 22degC) on a 14LI0D photoperiod

Respirometry Protocol

Main Trials Oxygen consumption was measured with an open-Circuit resshypirometry system At each of three temperatures 10deg 20deg and 30degC three randomly selected males and three randomly selected females were meashysured We attempted to obtain measurements at OdegC as well but not all voles could sustain normothermia at Odegc Thus each of 18 voles was meashysured once During measurements the voles were held in a Perspex (Plexishyglas) chamber 9 ern X 8 ern X 7 em A perforated Perspex partition restricted the voles movement to a length of 65 em The vole was placed in the chamber over a wire-mesh grid to keep it separated from voided urine and feces

An upstream flowmeter (Alexander Wright 02S-L water displacement) was modified so that every time SOO mL had passed through the flowmeter a voltage signal was generated The time interval between signals was used to calculate the flow rate Air was dried immediately before and after the flowmeter All flow rates were corrected to STPD (OdegC 760 Torr) Mean flow ranged from 468 to 694 mLmin The metabolism chamber was placed inside a controlled-temperature cabinet (Gallenkamp) that regulated the internal ambient temperature (T plusmn01 degC) Inside the controlled-temperature cabishynet and upstream of the metabolism chamber there was a length of copper tubing about 6 mm in diameter and 1 m long that ensured incoming air was equilibrated with the T in the cabinet

Respirometry Methods 607

The excurrent air from the metabolism chamber was connected to a large plastic syringe barrel A small-diameter tube inserted in the syringe barrel was used to subsample the excurrent air Carbon dioxide was removed with Carbosorb (BDH Poole UK active agent NaOH) and water was removed with silica gel The O2 concentration in the excurrent air was measured with an Applied Electrochemistry S3A dual-channel O2 analyzer A parallel line of reference air was sent to the other channel of the O2 analyzer to

minimize drift over the course of the measurements The difference between the O 2 content of the reference and animal air was continuously monitored with a BBC microcomputer and the mean recorded every 5 s

Each vole was weighed to plusmn001 g before being placed in the metabolism chamber The animals were not fasted before measurement but no food or water was available in the chamber Voles were placed in the metabolism chamber between 0707 and 1019 in the morning Data recording started after the animals had been in the chamber for about 1 h Measurements were continued for at least 6 h Voles were weighed again at this time

Auxiliary Measurements Additional measurements of O2 consumption were made in a larger metabolism chamber that allowed more scope for voluntary activity The setup was identical to that already described except that the metabolism chamber (8 cm high X 7 cm wide) was divided into three secshytions of length 25 ern 40 ern and 25 ern respectively and that flow rates ranged from 685 to 725 mLmin The vole was placed in the central large section and the smaller sections were filled with silica gel All measurements with this chamber were made at 20degC Six randomly selected female voles (excluding the voles previously used) were studied We also measured two Apodemus syluaticus (one male and one female) at 20degC Each Apodemus was measured in both the small and the large metabolism chamber following the same procedures as used for Microtus The female Apodemus was slightly sluggish after the measurement in the small metabolism chamber so her body temperature was measured It was slightly depressed (34degC)

Calculations

Calculation Interval We calculated steady-state O2 consumption using the appropriate equation from Hill (1972) For measurements made in the small metabolism chamber O2 consumption was calculated for the lowest 15 20 25 30 35 40 45 50 and 55 min of each hour and for the full hour itself This was repeated for each successive full hour for which data were obtained Thus 60 values (6 h X 10 intervals) were calculated for each individual

608 J P Hayes J R Speakman and PA Racey

For the large chamber which had a slower washout O 2 consumption was calculated for each full hour only

Sampling Bias For each animal the lowest 15-min interval of O 2 conshy

sumption occurring in the first 30 45 60 90 180 and 360 min of measureshyment was calculated

Subsampling of Runs If animals show pronounced ultradian or circadian rhythms (see Aschoff and Pohl1971 Heusner et al 1971 Kenagy and Vleck

1982) then it is important that measurements be made at the same phase of the cycle or that all phases of the cycle be sampled so that equivalent

data are obtained Devoting an entire day to measuring a single animal throughout a cycle will severely limit the number of animals that can be measured One approach to controlling for rhythmic variation in metabolism

but one that still uses analyzer time effectively is to set up an automated system and measure a number of animals simultaneously with each animal actually being monitored for a fraction of the total time (eg for a few

minutes every hour see Hayes Garland and Dohm 1992) The utility of this approach however depends on a number of factors including the

amplitude and period (length) of the rhythm the duration of the measureshyment and the timing of the measurement relative to the phase of the rhythm

To study the effect of subsampling we exhaustively subsarnpled the

complete metabolism recording for each animal as follows Oxygen conshysumption for the first 15 min of each of the 6 h was calculated Then the lowest of the six values was retained This procedure was repeated for all possible 15-min intervals of each hour For example O2 consumptions beshytween 003525 and 005025 013525 and 015025023525 and 025025

and 053525 and 055025 (hrnins time relative to the start of meashysurements) were calculated and the lowest of the six values determined

Next the lowest of the six values from between 003530 and 005030 01 3530 and 015030 023530 and 025030 and 053530 and 0550 30 was determined Finally the mean of all the lowest values was calculated

(hereafter referred to as the subsampling mean) The procedure we used to calculate the subsampling mean is biased because the first and last 15 min of each hour are sampled less intensively than the middle 30 min of each hour Despite this bias we think the method we used to calculate the subsampling mean is better than one based on randomly selecting IS-min

intervals within the hour because it more accurately reflects the way that measurements are likely to be collected if an automated subsampling system

is used

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

606 J P Hayes J R Speakman and P A Racey

Material and Methods

Study Animals

Short-tailed field voles (Microtus agrestis Linnaeus) were obtained from the wild in Grampian Region United Kingdom (SrN) during May 1990 The voles were trapped in Longworth traps baited with carrot and supplied with hay The voles were obtained from three areas separated by at least 10 km On the day of capture voles were returned to the Department of Zoology where they were housed in plastic rodent cages Males and females were housed separately Typically five voles were kept per cage but in a few cages there were fewer than five voles Wood shavings for bedding and light straw for nesting material were provided Rodent chow (Special Diet Service CRM Diet) and water were provided ad lib Fresh cabbage was supplied at least once per week Voles were housed at room temperature (ca 22degC) on a 14LI0D photoperiod

Respirometry Protocol

Main Trials Oxygen consumption was measured with an open-Circuit resshypirometry system At each of three temperatures 10deg 20deg and 30degC three randomly selected males and three randomly selected females were meashysured We attempted to obtain measurements at OdegC as well but not all voles could sustain normothermia at Odegc Thus each of 18 voles was meashysured once During measurements the voles were held in a Perspex (Plexishyglas) chamber 9 ern X 8 ern X 7 em A perforated Perspex partition restricted the voles movement to a length of 65 em The vole was placed in the chamber over a wire-mesh grid to keep it separated from voided urine and feces

An upstream flowmeter (Alexander Wright 02S-L water displacement) was modified so that every time SOO mL had passed through the flowmeter a voltage signal was generated The time interval between signals was used to calculate the flow rate Air was dried immediately before and after the flowmeter All flow rates were corrected to STPD (OdegC 760 Torr) Mean flow ranged from 468 to 694 mLmin The metabolism chamber was placed inside a controlled-temperature cabinet (Gallenkamp) that regulated the internal ambient temperature (T plusmn01 degC) Inside the controlled-temperature cabishynet and upstream of the metabolism chamber there was a length of copper tubing about 6 mm in diameter and 1 m long that ensured incoming air was equilibrated with the T in the cabinet

Respirometry Methods 607

The excurrent air from the metabolism chamber was connected to a large plastic syringe barrel A small-diameter tube inserted in the syringe barrel was used to subsample the excurrent air Carbon dioxide was removed with Carbosorb (BDH Poole UK active agent NaOH) and water was removed with silica gel The O2 concentration in the excurrent air was measured with an Applied Electrochemistry S3A dual-channel O2 analyzer A parallel line of reference air was sent to the other channel of the O2 analyzer to

minimize drift over the course of the measurements The difference between the O 2 content of the reference and animal air was continuously monitored with a BBC microcomputer and the mean recorded every 5 s

Each vole was weighed to plusmn001 g before being placed in the metabolism chamber The animals were not fasted before measurement but no food or water was available in the chamber Voles were placed in the metabolism chamber between 0707 and 1019 in the morning Data recording started after the animals had been in the chamber for about 1 h Measurements were continued for at least 6 h Voles were weighed again at this time

Auxiliary Measurements Additional measurements of O2 consumption were made in a larger metabolism chamber that allowed more scope for voluntary activity The setup was identical to that already described except that the metabolism chamber (8 cm high X 7 cm wide) was divided into three secshytions of length 25 ern 40 ern and 25 ern respectively and that flow rates ranged from 685 to 725 mLmin The vole was placed in the central large section and the smaller sections were filled with silica gel All measurements with this chamber were made at 20degC Six randomly selected female voles (excluding the voles previously used) were studied We also measured two Apodemus syluaticus (one male and one female) at 20degC Each Apodemus was measured in both the small and the large metabolism chamber following the same procedures as used for Microtus The female Apodemus was slightly sluggish after the measurement in the small metabolism chamber so her body temperature was measured It was slightly depressed (34degC)

Calculations

Calculation Interval We calculated steady-state O2 consumption using the appropriate equation from Hill (1972) For measurements made in the small metabolism chamber O2 consumption was calculated for the lowest 15 20 25 30 35 40 45 50 and 55 min of each hour and for the full hour itself This was repeated for each successive full hour for which data were obtained Thus 60 values (6 h X 10 intervals) were calculated for each individual

608 J P Hayes J R Speakman and PA Racey

For the large chamber which had a slower washout O 2 consumption was calculated for each full hour only

Sampling Bias For each animal the lowest 15-min interval of O 2 conshy

sumption occurring in the first 30 45 60 90 180 and 360 min of measureshyment was calculated

Subsampling of Runs If animals show pronounced ultradian or circadian rhythms (see Aschoff and Pohl1971 Heusner et al 1971 Kenagy and Vleck

1982) then it is important that measurements be made at the same phase of the cycle or that all phases of the cycle be sampled so that equivalent

data are obtained Devoting an entire day to measuring a single animal throughout a cycle will severely limit the number of animals that can be measured One approach to controlling for rhythmic variation in metabolism

but one that still uses analyzer time effectively is to set up an automated system and measure a number of animals simultaneously with each animal actually being monitored for a fraction of the total time (eg for a few

minutes every hour see Hayes Garland and Dohm 1992) The utility of this approach however depends on a number of factors including the

amplitude and period (length) of the rhythm the duration of the measureshyment and the timing of the measurement relative to the phase of the rhythm

To study the effect of subsampling we exhaustively subsarnpled the

complete metabolism recording for each animal as follows Oxygen conshysumption for the first 15 min of each of the 6 h was calculated Then the lowest of the six values was retained This procedure was repeated for all possible 15-min intervals of each hour For example O2 consumptions beshytween 003525 and 005025 013525 and 015025023525 and 025025

and 053525 and 055025 (hrnins time relative to the start of meashysurements) were calculated and the lowest of the six values determined

Next the lowest of the six values from between 003530 and 005030 01 3530 and 015030 023530 and 025030 and 053530 and 0550 30 was determined Finally the mean of all the lowest values was calculated

(hereafter referred to as the subsampling mean) The procedure we used to calculate the subsampling mean is biased because the first and last 15 min of each hour are sampled less intensively than the middle 30 min of each hour Despite this bias we think the method we used to calculate the subsampling mean is better than one based on randomly selecting IS-min

intervals within the hour because it more accurately reflects the way that measurements are likely to be collected if an automated subsampling system

is used

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 607

The excurrent air from the metabolism chamber was connected to a large plastic syringe barrel A small-diameter tube inserted in the syringe barrel was used to subsample the excurrent air Carbon dioxide was removed with Carbosorb (BDH Poole UK active agent NaOH) and water was removed with silica gel The O2 concentration in the excurrent air was measured with an Applied Electrochemistry S3A dual-channel O2 analyzer A parallel line of reference air was sent to the other channel of the O2 analyzer to

minimize drift over the course of the measurements The difference between the O 2 content of the reference and animal air was continuously monitored with a BBC microcomputer and the mean recorded every 5 s

Each vole was weighed to plusmn001 g before being placed in the metabolism chamber The animals were not fasted before measurement but no food or water was available in the chamber Voles were placed in the metabolism chamber between 0707 and 1019 in the morning Data recording started after the animals had been in the chamber for about 1 h Measurements were continued for at least 6 h Voles were weighed again at this time

Auxiliary Measurements Additional measurements of O2 consumption were made in a larger metabolism chamber that allowed more scope for voluntary activity The setup was identical to that already described except that the metabolism chamber (8 cm high X 7 cm wide) was divided into three secshytions of length 25 ern 40 ern and 25 ern respectively and that flow rates ranged from 685 to 725 mLmin The vole was placed in the central large section and the smaller sections were filled with silica gel All measurements with this chamber were made at 20degC Six randomly selected female voles (excluding the voles previously used) were studied We also measured two Apodemus syluaticus (one male and one female) at 20degC Each Apodemus was measured in both the small and the large metabolism chamber following the same procedures as used for Microtus The female Apodemus was slightly sluggish after the measurement in the small metabolism chamber so her body temperature was measured It was slightly depressed (34degC)

Calculations

Calculation Interval We calculated steady-state O2 consumption using the appropriate equation from Hill (1972) For measurements made in the small metabolism chamber O2 consumption was calculated for the lowest 15 20 25 30 35 40 45 50 and 55 min of each hour and for the full hour itself This was repeated for each successive full hour for which data were obtained Thus 60 values (6 h X 10 intervals) were calculated for each individual

608 J P Hayes J R Speakman and PA Racey

For the large chamber which had a slower washout O 2 consumption was calculated for each full hour only

Sampling Bias For each animal the lowest 15-min interval of O 2 conshy

sumption occurring in the first 30 45 60 90 180 and 360 min of measureshyment was calculated

Subsampling of Runs If animals show pronounced ultradian or circadian rhythms (see Aschoff and Pohl1971 Heusner et al 1971 Kenagy and Vleck

1982) then it is important that measurements be made at the same phase of the cycle or that all phases of the cycle be sampled so that equivalent

data are obtained Devoting an entire day to measuring a single animal throughout a cycle will severely limit the number of animals that can be measured One approach to controlling for rhythmic variation in metabolism

but one that still uses analyzer time effectively is to set up an automated system and measure a number of animals simultaneously with each animal actually being monitored for a fraction of the total time (eg for a few

minutes every hour see Hayes Garland and Dohm 1992) The utility of this approach however depends on a number of factors including the

amplitude and period (length) of the rhythm the duration of the measureshyment and the timing of the measurement relative to the phase of the rhythm

To study the effect of subsampling we exhaustively subsarnpled the

complete metabolism recording for each animal as follows Oxygen conshysumption for the first 15 min of each of the 6 h was calculated Then the lowest of the six values was retained This procedure was repeated for all possible 15-min intervals of each hour For example O2 consumptions beshytween 003525 and 005025 013525 and 015025023525 and 025025

and 053525 and 055025 (hrnins time relative to the start of meashysurements) were calculated and the lowest of the six values determined

Next the lowest of the six values from between 003530 and 005030 01 3530 and 015030 023530 and 025030 and 053530 and 0550 30 was determined Finally the mean of all the lowest values was calculated

(hereafter referred to as the subsampling mean) The procedure we used to calculate the subsampling mean is biased because the first and last 15 min of each hour are sampled less intensively than the middle 30 min of each hour Despite this bias we think the method we used to calculate the subsampling mean is better than one based on randomly selecting IS-min

intervals within the hour because it more accurately reflects the way that measurements are likely to be collected if an automated subsampling system

is used

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

608 J P Hayes J R Speakman and PA Racey

For the large chamber which had a slower washout O 2 consumption was calculated for each full hour only

Sampling Bias For each animal the lowest 15-min interval of O 2 conshy

sumption occurring in the first 30 45 60 90 180 and 360 min of measureshyment was calculated

Subsampling of Runs If animals show pronounced ultradian or circadian rhythms (see Aschoff and Pohl1971 Heusner et al 1971 Kenagy and Vleck

1982) then it is important that measurements be made at the same phase of the cycle or that all phases of the cycle be sampled so that equivalent

data are obtained Devoting an entire day to measuring a single animal throughout a cycle will severely limit the number of animals that can be measured One approach to controlling for rhythmic variation in metabolism

but one that still uses analyzer time effectively is to set up an automated system and measure a number of animals simultaneously with each animal actually being monitored for a fraction of the total time (eg for a few

minutes every hour see Hayes Garland and Dohm 1992) The utility of this approach however depends on a number of factors including the

amplitude and period (length) of the rhythm the duration of the measureshyment and the timing of the measurement relative to the phase of the rhythm

To study the effect of subsampling we exhaustively subsarnpled the

complete metabolism recording for each animal as follows Oxygen conshysumption for the first 15 min of each of the 6 h was calculated Then the lowest of the six values was retained This procedure was repeated for all possible 15-min intervals of each hour For example O2 consumptions beshytween 003525 and 005025 013525 and 015025023525 and 025025

and 053525 and 055025 (hrnins time relative to the start of meashysurements) were calculated and the lowest of the six values determined

Next the lowest of the six values from between 003530 and 005030 01 3530 and 015030 023530 and 025030 and 053530 and 0550 30 was determined Finally the mean of all the lowest values was calculated

(hereafter referred to as the subsampling mean) The procedure we used to calculate the subsampling mean is biased because the first and last 15 min of each hour are sampled less intensively than the middle 30 min of each hour Despite this bias we think the method we used to calculate the subsampling mean is better than one based on randomly selecting IS-min

intervals within the hour because it more accurately reflects the way that measurements are likely to be collected if an automated subsampling system

is used

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 609

Data Analysis

Statistical analyses were performed with Minitab (Ryan Joiner and Ryan 1985) and SAS (SAS Institute 1985) Repeated-measures ANOVA was used to test for effects of calculation interval (I 5-60 min) temperature (10deg20deg and 30degC) and hour (I-6) As appropriate both univariate and multivariate analyses are reported Since repeated-measures of body mass could not be made throughout the course of each respirometry run without disturbing the animals (which would have permitted a repeated-measures ANCOVA to remove body mass effects) the repeated-measures analyses were pershyformed on residuals from ANCOVAs on body mass by temperature Actually since these ANCOVAsremoved all variation due to temperature the adjusted means for temperature were added back to the residual before performing the repeated-measures ANOVA Note that adjusted means and not raw means had to be added to the residuals because raw means contain the effects of differences in body mass between temperatures In practice this distinction was of little consequence because the mean mass of animals at the different temperatures was very similar Repeated-measures ANOVA with a profile transformation was used to test for significant differences between adjacent levels for the lowest 15 min of O2 consumption from the first 30 45 60 90 180 and 360 min (SAS Institute 1985) This ANOVA was done on the raw data and not those adjusted for differences in mass among temperatures because the slopes of O2 consumption on mass were significantly different among temperatures for the 180middot and 360-min measurements In all analyses the repeated measures of metabolism were highly intercorrelated so the nominal significance levels of the univariate analyses must be interpreted with caution Thus for within-subject (vole) effects (ie calculation interval and hour of run) we report the Greenhouse-Geisser and Huynh-Feldt adshyjusted probability levels but these adjusted probabilities may still be too liberal in rejecting the null hypothesis (Freund Littell and Spector 1986) All significance levels are reported for tests based on SAS Type III (ie partial) sums of squares

Results

Mean body mass was 264 253 and 266 g respectively at 10deg 20deg and 30degC Oxygen consumption increased with temperature and calculation inshyterval (fig 1) The least-squares means (ie adjusted to the grand mean body mass of 261 g) of the lowest 15 min for each hour were 122 192 and 257 mLmin at 30deg20deg and 10degC respectively Temperature signifishy

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

610 J PHayes J R Speakman and P A Racey

-c E--g 30

bull 10degC -c m 20degCo bull 30degCa 20

E en s oo 10

s Q) C) gtshy~ 00

15 20 25 30 35 40 45 50 55 60

Calculation Interval (min) Fig 1 The mean and standard error jor lowest O2 consumption ojMicroshytus agrestis at 10deg 20deg and 30deg C for each calculation internal (n = 6

uoles for each bar)

cantly affected 0 1 consumption (Fl l = 438 P lt 00001) but sex did not (Fll4 = 330 P = 00906) Oxygen consumption varied significantly with calculation interval according to both the multivariate (Wilkss A= 00529 F96 = 119 P = 00034) and the univariate (F9 126 = 806 both GreenhouseshyGeisser and Hunyh-Feldt adjusted probabilities P lt 00001) tests Both the multivariate (Wilkss A = 0719 FS1O = 0781 P = 05858) and univariate (Fs7o = 066 Greenhouse-Geisser and Hunyh-Fe ldt adjusted probabilities 05834 and 06495 respectively) analyses indicated that hour of the run did not affect O2 consumption (fig 2)

Lowest 0 1 consumption was correlated with the log of the calculation interval (fig 3) The slopes of the relationship between lowest 0 1 conshysumption and calculation interval differed significantly among individual voles A pooled analysis (ie a separate slopes separate intercepts regression mode I) indicated that calculation interval explained a significant (P lt 005) amount of the total variation in lowest O 2 consumption

Increasing the time from which the lowest I 5-min interval of metabolism was selected resulted in lower estimates of 0 1 consumption (fig 4) A profile ANOYA (contrasts between adjacent levels) indicated that significant breaks occurred between 60 and 90 min and between 180 and 360 min Across all three temperatures mean O2 consumption for the lowest 15 min in the first 30 min of a run was 192 as opposed to 170 mLmin for the lowest 15 min

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Fig 2 Variation in O2 consumption ofMicrotus agrestis over the 6 b of

respirometry runs at 10deg20deg and JOdege Each bar is the least-squares

mean of O2 consumption for the full hour (n = 6 voles a teach

temperature)

out of all 360 min of the run Thus measuring the voles for only 30 min would have resulted in an estimate of O 2 consumption 13 higher than the estimate after measuring them for 6 h

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

612 J P Hayes J R Speakman and PA Racey

- 30

--I E

E 25 c o QE 20 ~ en c o o 15

c (I) C)

gt- 4---r--r------------r-------------r----gtlt 10 o 10 20 30 40 50 60

Calculation Interval (min) Fig 3 Lowest O2 consumption aMicrotus agrestis plotted as a function

ofcalculation interval The curves fitted to the plotted points are the least-squares models 002 consumption vs log (calculation interval) at

each temperature

The subsampling mean O 2 consumption was 183 mLmin The estimate

obtained using the subsampling approach was not appreciably lower than

that obtained from monitoring voles for the first 90 min of a run (182 mLI

min fig 4) Thus for these data there would have been little advantage to

subsampling instead of measuring individuals for 90 min each

The lack of benefit from subsampling for Microtus and the relatively small

differences associated with total time voles were monitored reflects the abshy

sence of significant hourly variation in metabolism and the fact that there

was no elevation in metabolism early in the runs One potential explanation

for this is that the voles did not have sufficient space to exhibit significant

voluntary movements or activity However hourly O2 consumption at 20degC

did not differ between voles in the small chamber and those in the larger

chamber allowing more space for voluntary activity (ANCOVA with mass as

a covariate FJ bull9 = 115 P = 0312)

In contrast to Microtus agrestis O 2consumption of Apodemus syluaticus

was higher during the first 1 or 2 h than the last 4 or 5 h of measurement

(fig 5) In the small metabolism chamber the lowest 15 min of metabolism

in the first 30 min averaged 148 mL 02min and the lowest 15 min out of

the complete 6 h averaged 0898 mL 02min (fig 6) The subsampling

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

bullbull

Respirometry Methods 613

C 20-E-I ~E 19 bull c o c E j CtJ C o o 17

18 j bull

bull c CD C) gtshygtC 16 +-------------r---r-o

o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 4 The lowest 15 min of O2 consumption plotted against tbe total time

from tubicb tbe lowest 15-min interval was selected for Microtus agrestis Eacb point is tbe mean for 18 individuals (6 eacb at 10deg 20deg and

30deg C) Tbe arrows indicate significant breaks between adjacent points

as determined by a repeated-measures ANOVA uritb a profile transformashy

tion The means across all temperatures were plottedfor ease of visualshy

ization but tbe actual analysis was performed on tbe individual data points (see text)

mean O2 consumption was 101 mL 02min Hourly O2 consumption was higher in the large metabolism chamber (fig 5)

Discussion

Sampling Bias

Monitoring metabolism for varied lengths of time introduces systematic bias into estimates of minimum metabolic rates (fig 4) For Microtus samshy

pling for 6 h instead of 30 min resulted in a mean difference of 13 for the lowest 15 min of metabolism Even though the voles we studied were all recently captured from the wild they were quite docile and almost all of them would sit in sorneones hand without apparent agitation The docility of the voles may explain why there was so little variation in metabolism over the 6 h of each run If voles showed a fright response or any marked

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

614 J P Hayes J R Speakman and P A Racey

-cshyE-

30 ADOdemus J E- 25 bull Large chamber

c 0= 20

m Small chamber

c E 15 ~ rn c 0 10 0 c CD 05 IC) gtshygtlt0 00

1 23456

Fig 5 Mean O2 consumption arid standard error bar for each of the 6 b

of the respirometry runs at 20deg C for two Apodemus sylvaticus

response to handling we would have expected the initial hour of the run

to have been higher than the later hours (fig 2) The absence of significant

differences throughout the 6 h of the runs is somewhat surprising given that

ultradian and circadian rhythms have been documented in Microtus (Erkishy

naro 1969 Lehmann 1976 Daan and Slopsema 1978) Perhaps this is because

ultradian rhythms in metabolism are less pronounced during the inactive

phase which is when we measured metabolism than during the active

phase (Gerkerna and Daan 1985 Kleinknecht Erkert and Nelson 1985)

Moreover an hour-by-hour analysis may not be affected by ultradian rhythms

that would be detected by more sensitive techniques such as periodogram

autocorrelation or spectral analysis

Problems with sampling bias may be greater with more active species

For Apodemus sampling for only 30 min resulted in estimates of O 2 conshy

sumption 65 higher than sampling for 6 h (fig 6) Apodemus syluaticus

is much more active than Microtus agrestis In contrast to Microtus Aposhy

demus had elevated metabolism during the first hour or two (fig 5) The

greatest effect of variation in the amount of time from which metabolic rates

are sampled will occur when some measurements are obtained from animals

that are exhibiting an elevated metabolism in response to handling and

others are not For example if some Apodemus were measured for 1 hand

Hour of Run

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 615

18 AQQdemus5 E

J 16 E-c 14 o a

12E J CD C 10o o c 08CD 0) gtshygtlt 06 +--------------r----r--shyo o 60 120 180 240 300 360

Time from which lowest 15 min were selected (min)

Fig 6 The lowest 15 min of O2 consumption plotted against the total time from which the lowest 15-min intervals were selected for two Apodemus

sylvaticus Compare with jig 4 for Microtus agrestis

others for 2 or 3 h the estimates of their metabolic rates could be very different (fig 6)

Differences in the period of time from which the metabolic rates of animals

are sampled may produce biologically significant differences in estimates of metabolic rate Effects of 13-65 are as great or greater than would be

expected from not accounting for circadian rhythms (Kenagy and Vleck 1982 Stupfel et al 1987) In ecological studies such as McNabs (1986) analysis of variation in basal metabolic rate with food habits differences of

13-65 in the estimates of metabolic rate could markedly affect the inshyferences drawn Thus it is clearly important to standardize procedures for evaluating the metabolic rate of individuals

Subsampling

One way to control for effects of an initial elevation of metabolism in reshysponse to handling and for ultradian or circadian fluctuations is to make extended measurements of individuals It is desirable however to be able

to measure more than one animal per analyzer per day This can be done by monitoring several animals in parallel with data being collected for each animal for part of each hour (or whatever time period desired see Hayes

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

616 J P Hayes J R Speakman and PA Racey

et al 1992) Using this subsampling approach with Microtus resulted in estimates of metabolism very similar to those obtained by monitoring each animal for the same total length of time at the start of a respirometry run Thus for these voles the additional complexity and cost of an automated switching system to subsample each animal does not appear to be warranted However consecutively measuring animals for 90 min after placing them in a metabolism chamber may have resulted in values different from those we found for the initial 90 min of our 6-h runs If several individuals were measured consecutively in a day then different parts of the circadian and possibly ultradian cycle would be sampled

While the data for Microtus suggest that there would have been no benefit to subsampling instead of consecutively measuring individuals the lack of benefit is due to the very limited temporal variation exhibited during the course of measurements (fig 2) For Apodemus subsampling 15 min per hour for the 6 h of each run (X = 101 mL 02min) would have produced results substantially different (1896) from those obtained when taking the lowest 15 min out of the first 90 min of a run (X = 120 mL Odmin) In general the same total time devoted to subsampling may not always be better than consecutive measurements on individuals throughout the day but it is unlikely to be worse

Calculation Interual

The comparability of measurements calculated over different time periods is a subject that has received very little attention Dawson and Olson (1987) distinguished between what they called peak metabolism (measurements of 5-10 min) and summit metabolism (measurementsgt 2 h) They reported that peak metabolism of Blarina breuicauda was 6-15 higher than sumshymit metabolism depending on the acclimation conditions Some of the difshyference they reported between peak and summit metabolism however may be due to using a cold stress that exceeded the thermogenic capacity of their animals resulting in hypothermia and an inability to maintain heat production Hayes (1989) found that Peromyscus maniculatus can maintain maximal metabolic rates for 2 h at levels very similar to their Lrnin maximum Chappell (1984) reported maximal O2 consumption during exercise was 71 higher when calculated for the highest 1 min versus the highest 5 min of a 6-min run For maximal O2 consumption during cold exposure he reported a difference of 32 between estimates calculated for 2 min and 8 min

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 617

The time period over which O2 consumption was calculated (calculation interval) had a significant effect on estimated O 2 consumption (fig 3) Oxshyygen consumption of Microtus agrestis was 12 higher when calculated over 60 min than when calculated over 15 min This is large enough to significantly influence the inferences that might be drawn from comparative studies of variation in metabolic rates Minimum O2 consumption increased with the logarithm of calculation interval (fig 3) One reason for the increase of O 2 consumption with calculation interval is that shorter calculation inshytervals allow the selection of minimum values that exclude periods when voles were active In some respirometry studies activity is monitored along with metabolism so that minimum values can be selected from within the nonactive period only Another reason for the effect of calculation interval is that metabolism even during resting is not a constant value but a distrishybution of values that may be affected by slight changes in body temperature hormone levels and a host of other underlying physiological processes Shorter calculation intervals result in selecting minimum estimates that are progressively lower (ie close to the lower tail of the distribution) Longer calculation intervals will result in damping extreme values by averaging them with values that are less extreme It is important to recognize that estimates from different calculation intervals will vary and this should be accounted for when comparing data for which calculation intervals differ

Summary

This article describes the effects of sampling on estimates of metabolic rates Circadian cycles will influence the effects of sampling bias (eg varying the total time from which lowest values are selected might have had quite a different effect if measurements were started 1 h before the voles active phase began) Thus our analyses reflect when in the circadian cycle meashysurements were made If measurements had been initiated at some other part in the circadian cycle the results may have been different Future reshysearch is needed to determine how sampling bias may vary in different parts of the circadian cycle

In summary we make the following recommendations The period of time that animals are measured within any respirometry study should be standardized Moreover as has been previously suggested (Heusner 1965 Kenagy and Vleck 1982) animals should be monitored long enough to

control for responses to handling and for circadian rhythms Subsampling (monitoring several animals in parallel) is an efficient approach to this probshy

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

618 J P Hayes J R Speakman a1d P A Racey

lem particularly when it is necessary to measure large numbers of animals Differences in calculation interval are also large enough to have an effect on comparative studies Biologists cannot always work with data collected in precisely the same way but calculation interval should be included as a covariate in comparative analyses of respirometry data

Acknowledgments

This research was supported by Natural Environment Research Council grant GR36945 to PAR and]RS We thank G Hays and P Webb for commenting on the manuscript

literature Cited

AsCHOFF] and H POHL 1971 Rhythmic variations in energy metabolism Proc Fed Am Soc Exp BioL 29 1541- 1552

BARTHOLOMEW G A 1972 Energy metabolism Pages 44-72 in M S GORDON ed Animal physiology principles and adaptations Macmillan New York CALDER W A 1984 Size function and life history Harvard University Press Camshy

bridge Mass 431 pp CHAPPELL M A 1984 Maximum oxygen consumption during exercise and cold exshy

posure in deer mice Peromyscus maniculatus Respir Physiol 55367 - 377 DAAN 5 and S SLOPSEMA 1978 Short- term rhythms in foraging behaviour of the

common vole Microtus arualis] Camp Physiol 127B215--227 DAWSON T] and] M OLSON 1987 The summit metabolism of the short-tailed

shrew Blarina breuicauda a high summit is further elevated by cold acclimation Physiol Zool 60631- 639

ERKINARO E 1969 Der Phasenwechsel der lokomotorischen Aktivitat bei Microtus agrestis (L) M aroalis (Pall) und M oeconomus (PalL) Aquila (Ser Zool) 8 1-29

FREUND R J R C LITTELL and P C SPECTOR 1986 SAS system for linear models SAS Institute Cary NC 210 pp

GERKEMA M P and S DAAN 1985 Ultradian rhythms in behavior the case of the common vole (Microtus arualis) Pages 11-31 in H SCHULZ and P LAVIE eds Ultradian rhythms in physiology and behavior Springer Berlin

HAYES ] P 1989 Field and maximal metabolic rates of deer mice tPeromyscus maniculatus) at low and high altitudes Physiol ZooI 63732 - 744

HAYES] P T GARLANDJR and M R DOHI-I 1992 Individual variation in metabolism and reproduction of Mus are energetics and life history linked Funct Ecol 6 (in press)

HEUSNER A 1965 Sources of error in the study of diurnal rhythm in energy metabshyolism Pages 3 -12 in] ASCHOFF ed Circadian clocks North-Holland Amsterdam

Respirometry Methods 619

HEUSNER A A] C ROBERTS and R E SMITH 1971 Time as a factor in metabolic studies of Peromyscus Acta Physiol Acad Sci Hung Tomus 40 1-11

HILL R W 1972 Determination of oxygen consumption by use of the paramagnetic oxygen analyzer] Appl Physiol 33261-263

KENAGY G] and D VLECK 1982 Daily temporal organization of metabolism in small mammals adaptation and diversity Pages 322 - 338 in] AsCHOff S DAAN and G A GROOS eds Vertebrate circadian systems Springer Berlin

KLEINKIfECHT S H G ERKERT and] E NELSON 1985 Circadian and ultradian rhythms of activity and Oj-consumprlon in three nocturnal Marsupialian species Petaurus breuiceps Phalangeridae Dasyuroides byrnei Dasyuridae Monodelpbis domestica Didelphidae Z Saugetierkunde 50321- 329

LEHMANN U 1976 Short-term and circadian rhythms in the behavior of the vole Microtus agrestis (L) Oecologia 23185-199

McNAB B K 1986 The influence of food habits on the energetics of eutherian

mammals Ecol Monogr 561-19 PETERS R H 1983 The ecological implications of body size Cambridge University

Press Cambridge 329 pp PETERSON C c K A NAGY and] DIAMOND 1990 Sustained metabolic scope Proc

Nat Acad Sci USA 872324-2328 PROTHERO J 1984 Scaling of standard energy metabolism in mammals 1 Neglect

of circadian rhythms J Theor BioI 1061-8 RYAN B F B L JOINER and T A RYAN JR 1985 Minitab handbook PWSmiddotKent

Boston 385 pp SAS INSTITUTE 1985 SAS users guide statistics SAS Institute Cary NC 956 pp SCHMIDT NIELSEN K 1984 Scaling why is animal size so important Harvard University

Press Cambridge Mass 241 pp STUPfEL M V GOlJRLET L COURT] MESTRIES A PERRAMON and P MERAT 1987

Periodic analysis of ultradian (40 min lt 1lt 24 h) respiratory variations in laboratory vertebrates of various circadian activities Chronobiologla 14365-375