Thermoregulation during mild exercise at different circadian times.

Abstract

Eight healthy subjects exercised at 90watts on a cycle ergometer on four occasions, at times close to the minimum, maximum rate of rise, maximum, and maximum rate of fall of their resting core temperature. The duration of exercise was determined by the time taken for the core (rectal) temperature to reach an equilibrium value. Forearm skin blood flow and temperature were measured regularly during the exercise, as were heart rate and ratings of perceived exertion. Sweat loss was calculated by weighing the subjects nude before and after the exercise. The rise of heart rate was not significantly different at the four times of exercise, though the rating of perceived exertion was greatest at 05:00 h. Resting core temperatures showed a significant circadian rhythm at rest (the timing of which confirmed that exercise was being performed at the required times), but the amplitude of this rhythm was decreased significantly by the exercise. The initial rate of rise of core temperature, and the total rise from the resting to the equilibrium value, were both inversely proportional to resting temperature. The time-course of the rise was accurately described by a negative-exponential model, but this model gave no evidence that the kinetics of the equilibration process depended upon the time of day. The thermoregulatory responses to the rise in core temperature--the amount of total sweat loss and rises in forearm skin blood flow and temperature--differed according to the time of exercise. In general, the responses were significantly greater at 17:00h compared with 05:00h, and at 23:00 h compared with 11:00 h. The results accord with predictions made on the basis of previous work by us in which core temperature rhythms have been separated into components due to the endogenous body clock and due to the direct effects of spontaneous activity. The results are discussed in terms of the ecological implications of the differing capabilities of humans to deal with heat loads produced by spontaneous activity or mild exercise at different phases of the circadian rhythm of resting core temperature.

Full text

Page 1

CHRONOBIOLOGY INTERNATIONAL
Vol. 21, No. 2, pp. 253–275, 2004
Thermoregulation During Mild Exercise at
Different Circadian Times
James Waterhouse,1,*Benjamin Edwards,1Paul Bedford,1Amber Hughes,1
Kirsty Robinson,1Alan Nevill,2Dietmar Weinert,3and Thomas Reilly1
1Research Institute for Sport and Exercise Sciences, Liverpool John Moores
University, Liverpool, UK
2School of Sport, Performing Arts and Leisure, University of Wolverhampton,
Walsall, UK
3Institute of Zoology, University of Halle, Halle, Germany
ABSTRACT
Eight healthy subjects exercised at 90watts on a cycle ergometer on four
occasions, at times close to the minimum, maximum rate of rise, maximum, and
maximum rate of fall of their resting core temperature. The duration of exercise
was determined by the time taken for the core (rectal) temperature to reach an
equilibrium value. Forearm skin blood flow and temperature were measured
regularly during the exercise, as were heart rate and ratings of perceived exertion.
Sweat loss was calculated by weighing the subjects nude before and after the
exercise. The rise of heart rate was not significantly different at the four times of
exercise, though the rating of perceived exertion was greatest at 05:00h. Resting
core temperatures showed a significant circadian rhythm at rest (the timing of
which confirmed that exercise was being performed at the required times), but the
amplitude of this rhythm was decreased significantly by the exercise. The initial
rate of rise of core temperature, and the total rise from the resting to the
*Correspondence: James Waterhouse, Research Institute for Sport and Exercise Sciences,
Liverpool John Moores University, Henry Cotton Campus, 15-21 Webster St., Liverpool
L32ET, UK; Fax: þ151-231-4353; E-mail: waterhouseathome@hotmail.com.
253
DOI: 10.1081/CBI-120037799
Copyright & 2004 by Marcel Dekker, Inc.
0742-0528 (Print); 1525-6073 (Online)
www.dekker.com

Page 2

ORDER REPRINTS
equilibrium value, were both inversely proportional to resting temperature. The
time-course of the rise was accurately described by a negative-exponential model,
but this model gave no evidence that the kinetics of the equilibration process
depended upon the time of day. The thermoregulatory responses to the rise in
core temperature—the amount of total sweat loss and rises in forearm skin blood
flow and temperature—differed according to the time of exercise. In general, the
responses were significantly greater at 17:00h compared with 05:00h, and at
23:00h compared with 11:00h. The results accord with predictions made on the
basis of previous work by us in which core temperature rhythms have been
separated into components due to the endogenous body clock and due to the
direct effects of spontaneous activity. The results are discussed in terms of
the ecological implications of the differing capabilities of humans to deal with
heat loads produced by spontaneous activity or mild exercise at different phases
of the circadian rhythm of resting core temperature.
Key Words:
Sweating.
Core temperature; Skin blood flow; Skin blood temperature;
INTRODUCTION
Physical activity imposes a metabolic heat load upon the body, and core
temperature rises above its resting value as a result. The rise is limited by recruitment
of thermoregulatory reflexes that promote sweating and cutaneous vasodilatation.
Once the core temperature has risen above the threshold for a thermoregulatory
reflex, there is a progressive increase in reflex activation that ultimately limits the rise
in core temperature.
There are circadian rhythms in the core temperatures at which sweating
and cutaneous vasodilatation are initiated—the sweating and vasodilatation
thresholds (Aoki et al., 1995, 1997, 2001; Stephenson et al., 1984; Tokura et al.,
1979; Waterhouse and Minors, 1995). Since these thresholds are in phase with the
circadian rhythm of core temperature, it is considered that the same circadian
oscillator (‘‘body clock’’) is responsible for these rhythms. It is probable that the
sympathetic nervous system is involved in some way (Aoki et al., 2001; Gordon,
1994; Reilly and Brooks, 1986; Stephenson et al., 1984). With sustained exercise, core
temperatures equilibrate at a raised value, probably indicative of a new set-point
(Davenne and Lagarde, 1995; Lind, 1963; Reilly and Brooks, 1986). These equi-
librium temperatures also show a circadian rhythm that is in phase with resting
temperature. However, the amplitude of this equilibrium rhythm might be lower,
particularly if the exercise is heavy (Reilly and Garrett, 1995), since there appears to
be a ‘‘ceiling’’ above which core temperature does not rise.
Cutaneous vasodilatation is also involved in the circadian rhythm of resting
core temperature. This rhythm is measured most accurately by undergoing a
constant routine protocol (Czeisler et al., 1985; Mills et al., 1978). In this protocol,
activity and sleep are prevented for a period of at least 24h, food intake is restricted
to identical snacks taken at equal intervals, and the light intensity, temperature, and
humidity of the environment are maintained constant. In these circumstances, a
circadian rhythm with an amplitude of about 0.5?C and peak at about 17:00h
254 Waterhouse et al.

Page 3

ORDER REPRINTS
(minimum at about 05:00h) remains (Krauchi and Wirz-Justice, 1994). The rhythm
is caused mainly by heat loss through the skin, though there is also a small
contribution from heat gain due to metabolism. There is a circadian rhythm of heat
loss from the distal limbs, rhythms of skin temperature and blood flow in these
regions showing peaks in the late evening—when core temperature is falling most
rapidly—and minima in the morning—when it is rising most rapidly (Aschoff and
Heise, 1972; Hildebrandt, 1974; Smolander et al., 1993). In other words, a resting
individual is in a heat-gain mode in the morning and in a heat-loss mode in the
evening. The rhythm in core temperature produced by these changes in vascular tone
is promoted also by behavioral rhythms; increases in activity in the morning and
decreases in the evening; and a preference for warmer ambient temperatures in the
morning and cooler ones in the evening (Gordon, 1994; Refinetti, 1998; Shoemaker
and Refinetti, 1996; Terai et al., 1985).
Since similar thermoregulatory mechanisms are involved in the response to a
heat load and in the production of the circadian rhythm of core temperature at rest,
it seems likely that interactions between these two processes might exist. There is
some evidence for such an interaction. Thermoregulatory responses to changes in
ambient temperature during sleep are less than those during the waking hours
(Haskell et al., 1981; Sagot et al., 1987); there are circadian rhythms in the sweating
and vasodilatation thresholds (see above); and circadian rhythms in the sensitivity of
thermoregulatory reflexes have been described (Refinetti and Menaker, 1992; Watts
and Refinetti, 1996). The sensitivity of a thermoregulatory reflex can be defined as
the relationship between its output (the increase in vasodilatation or sweating) and
its input (the rise of core temperature above the threshold of the reflex). Defined this
way, the sensitivities of the sweating and vasodilatation reflexes have been shown to
be less in the early morning than in the evening (Aoki et al., 1995, 2001, 2002; Torii et
al., 1995). Also, we previously showed that the response to the same amount of
moderate exercise in the morning differed from that in the late afternoon; in the
morning, the rise in rectal temperature was larger and the rise in skin blood flow
tended to be less (Aldemir et al., 2000).
Many of these studies have considered only two time-points. A fuller test of the
hypothesis that there is an interaction between the processes producing the circadian
rhythm of resting core temperature and those limiting the rise in core temperature
produced by exercise requires more time-points distributed throughout the 24h.
More specifically, it would be predicted (Prediction 1) that, during the morning rise
of resting core temperature, when the circadian system is in the heat-gain mode, the
mechanisms for losing the heat load due to the imposed exercise are less effective; by
contrast, during the evening, with resting core temperature in the heat-loss mode,
these mechanisms are predicted to be more effective. In addition, it would be
predicted (Prediction 2) that, during the middle of the active phase when core
temperature and activity are both high, thermoregulation—to limit the possibility of
hyperthermia—is more effective than during the middle of the resting phase, when
activity and core temperature are both low. To test these two predictions requires a
minimum of four time-points in order to be able to compare thermoregulatory
reflexes during the middle of the rising and falling phases (Prediction 1) and the peak
and trough (Prediction 2) of the circadian rhythm of resting core temperature.
Circadian Rhythms of Thermoregulation 255

Page 4

ORDER REPRINTS
There is indirect evidence to support these two predictions. This evidence comes
from attempts to purify measured core temperature data, that is, to separate them
into the component due to the direct effects of spontaneous activity and that due
to the body clock (Waterhouse et al., 1999, 2001; Weinert and Waterhouse, 1998;
Weinert et al., 2003). These methods require estimates of core temperature in the
absence of activity, and such estimates can be obtained by estimating the relationship
between temperature increase and activity. In one study (Waterhouse et al., 2001), a
group of ANCOVA models was used to estimate these relationships, with activity as
the covariant. In these models, the covariant was modelled using linear, quadratic,
and cubic terms. The linear model, though easiest to analyze, suffers from the
fact that its basic tenet—that core temperature continues to rise with increases in
activity—is only an approximation, the rise normally being halted by the process of
thermoregulation. The quadratic and cubic models are biologically superior to the
linear model in that they postulate a curvilinear relationship between activity and
temperature. In more recent studies upon gerbils (Weinert et al., 2003), the linear and
quadratic ANCOVA models were again used to describe the relationship between
activity and core temperature, but a third model, a negative-exponential model, was
added. This was because the quadratic model postulates that, after the turning point
has been exceeded, further rises in activity are predicted to cause a fall in core
temperature—and this is unacceptable biologically. The negative-exponential model
is the most acceptable biologically in that it postulates that the rise in core tem-
perature caused by increasing activity moves in a curvilinear fashion from a resting
temperature (the Lower Intercept Temperature, LIT) toward an equilibrium value
(the Upper Asymptote) as thermoregulatory reflexes are progressively recruited. We
found that the negative exponential model was also the most successful from a
statistical viewpoint, since the model accounted for the largest proportion of the
variance of a set of raw core temperature data.
Whichever purification method was used, however, there was the finding in
mice (Weinert and Waterhouse, 1998), humans (Waterhouse et al., 1999, 2001), and
gerbils (Weinert et al., 2003) that the coefficient(s) of the term(s) relating rise of
temperature to amount of activity indicated greater temperature increases for a given
amount of activity when the activity was performed around the trough rather than
around the peak of the resting circadian rhythm of core temperature. Also, the
coefficients indicated greater temperature increases for a given amount of activity
when it was performed during the rising (heat-gain mode) rather than the falling
(heat-loss mode) phase of the core temperature rhythm in the case of mice (though
this finding was not confirmed with humans or gerbils).
Since the more effective is thermoregulation, the less will be the rise in core
temperature caused by a given amount of activity, the purification methods have
provided evidence in favor of both of the above predictions with regard to
thermoregulatory efficiency. However, these analyses have dealt so far only with
spontaneous activity, and more direct measurements of thermoregulation, such as
sweating and increases in skin blood flow and skin temperature, have not been made.
The current report describes a series of experiments in which exercise has
been performed at four time-points during the 24h—close to the times of trough,
maximum rate of rise, peak, and maximum rate of fall of resting core temperature.
The two main aims of the study have been the following: 1) To investigate the
256Waterhouse et al.

Page 5

ORDER REPRINTS
suitability of the negative-exponential model for describing the time-course of
changes in core temperature during mild exercise rather than during spontaneous
activity; 2) To test directly the above predictions relating to the effectiveness and
sensitivity of thermoregulatory reflexes at different phases of the circadian rhythm of
resting core temperature—by comparing amounts of sweating and the relationships
between changes in skin blood flow or skin temperature of the forearm and changes
in core temperature.
METHODS
Subjects
Eight subjects (four males and four females) were recruited from the university
by word of mouth. The four females were studied at all stages of their menstrual
cycles. Some details of the subjects were as follows: age: females, 20–24 years; males,
20–22 years; height: females, 164–179cm; males, 170–182cm; body mass: females
56–92kg, males 65–94kg. All subjects lived a conventional lifestyle (sleeping at night
for 6–8h), were nonsmokers, participated regularly in exercise and training, and
were familiar with using the recumbent cycle ergometer. They abstained from
alcohol and heavy training for the 24h before the bouts of exercise.
The experimental procedures were fully explained to the subjects, any questions
were answered, and the equipment was shown to them prior to their giving signed
consent and taking part in the study. The study was approved by the university’s
Human Ethics Committee.
Protocol
Pilot studies had indicated that an exercise load of 90 watts would result in a rise
of core temperature that reached equilibrium in 30–45min and would not be found
by the subjects to be difficult to maintain. Each subject was studied at four times of
the day (05:00h, 11:00h, 17:00h, 23:00h). The study was carried out between
November 2002 and January 2003, when GMT was in operation. Exercise bouts
were separated by at least 72h, though the interval was generally about a week. The
order of testing times was randomly determined. When the test session was at
05:00h, the subjects slept in the laboratory from about 22:00h the previous evening
until 04:00h. This arrangement gave subjects adequate sleep and sufficient time to
wake fully before the bout of exercise.
Subjects wore the same kind of clothing to all experimental sessions, and the
laboratory temperature (20þ2?C) and humidity (60?5%) were controlled. Subjects
were asked to ensure that they were adequately hydrated before each test session.
Each test session was divided into two phases—an initial preparation and resting
phase, followed by the exercise phase. At the start of the experiment, subjects first
inserted a rectal probe 10cm beyond the external anal sphincter, towelled themselves
down, and were weighed nude to an accuracy of 0.1kg (Seca Ltd, Weighing and
Measuring Systems, England). They then sat on the cycle ergometer while the probes
Circadian Rhythms of Thermoregulation 257

Page 6

ORDER REPRINTS
(see next section) were attached. After being in the laboratory for at least 45min, the
resting (baseline) values of the measured variables were collected, this taking a
minimum of 15min.
During the exercise phase, the mild exercise was performed on a reclining cycle
ergometer (Kettler Ergometer Sxi, Germany). The duration of exercise was variable,
this being determined by the time taken for core temperature to reach an equilibrium
value. (Equilibrium was defined as having been reached when no change in rectal
temperature was measured over a period of 9min.) Immediately after the cessation
of exercise, the subject towelled himself/herself down and was weighed nude again.
Measurements
Core body temperature and forearm skin temperature were measured every
3min using a rectal probe and skin thermistor and a Squirrel data logger (Type 1002,
Grant Instruments, Cambridge). To keep the location of skin thermistor constant, its
placement was determined by reference to the base of the middle finger.
Skin blood flow was measured using Laser Doppler Flowmetry (Periflux System
5000, heater 4005, 408 standard probe, and 455 angled thermostatic probe; Perimed,
Ja ¨ rfalla, Sweden). The two lasers were attached the right forearm, approximately
1cm apart. The distance of the lasers to the base of the middle finger was measured
so that the location of the lasers could be replicated between experiments. The
arm and hand were rested on, and taped to, a table to reduce artiefacts data due to
movement. At the end of the exercise period, the Doppler heaters were turned up
to 42?C to produce maximal blood flow. All readings were calculated as a percentage
of this maximum. The record produced was a continuous one, but the mean value
every 3min during exercise was used in the analysis.
Heart rate was measured every 5min using a Polar heart rate monitor (Polar
Favour, Kempele, Finland) strapped to the chest wall directly above the heart.
The rating of perceived exertion, RPE, was measured every 5min during exercise
using the Borg scale (Borg, 1970).
The Negative-Exponential Model
The rise during exercise of core temperature from baseline to equilibrium was
investigated by the negative-exponential model. In this model, the temperature
during exercise, [Temperature]t, takes a curvilinear course from a Lower Intercept
Temperature (LIT, resting temperature) to an Upper Asymptote Temperature
(UAT, equilibrium temperature). The equation is:
½Temperature?t¼ ½UAT? ? f½UAT? ? ½LIT?g?Expð?K?tÞ
The constant, K, in the model is analogous to the time constant and is a measure of
shape of the curve.
Data sets from each of the four times of exercise and the eight subjects were
analyzed separately. Each data set consisted of the resting core temperature and
258Waterhouse et al.

Page 7

ORDER REPRINTS
the 10 temperatures obtained during the first 30min of exercise. The method used
(SPSS, version 11) was iterative and required starting estimates of LIT, UAT, and K
to be inserted. (The values used were 37.0?C, 38.0?C and 0.005min?1, respectively,
and based upon inspection of the raw data and preliminary exploration of them.)
The output of the model, obtained when subsequent iterations made no significant
difference to the output from the model, was estimates of LIT, UAT, and K.
Other Statistical Methods
Two-way ANOVA with repeated measures has been used for analyzing many of
the variables for time-of-day and duration-of-exercise effects, both being within-
subject factors: Factor 1, Time of day (05:00h, 11:00h, 17:00h, and 23:00h); Factor
2, Duration of exercise (0min, 3min, 6min, etc.). On other occasions, one-way
ANOVA with repeated measures was used, the within-subject factor being time of
day. Corrections for violations of sphericity (Greenhouse-Geiser epsilon) were used,
the degrees of freedom being reported to the nearest whole number. The analyses
were by SPSS (version 11).
Analysis of covariance, ANCOVA, was used for assessing the relationship
between thermoregulation and core temperature at the four different times of
exercise. Resting and exercise forearm skin blood flow or skin temperature were
treated as the dependent variable and the rectal temperatures that were measured
at the same time were treated as the covariates. The covariates were entered as both
linear and quadratic terms, so allowing the effect of the increase in rectal
temperatures upon skin flow or temperature to be curvilinear. The output of the
model enabled the significance of the covariate to be assessed, as well as any differ-
ences between subjects (N¼8), time of day when the exercise was performed
(05:00h, 11:00h, 17:00h, and 23:00h), duration of exercise (0min, 3min, 6min,
30min), and interactions between these factors. The time-of-day effect was assessed
by comparing the linear coefficients at the four times of exercise. To achieve this, the
analysis was performed twice; once with the other three coefficients being compared
with the coefficient for the data at 05:00h as baseline, and once with the other three
coefficients being compared with the coefficient at 11:00h as baseline. The
ANCOVA models were run on a Minitab package (version 12).
Probability, P, values indicating or approaching significance are reported to
three decimal places. Nonsignificant values are reported correct to two decimal
places.
RESULTS
Heart Rate and Rating of Perceived Exertion
Mean heart rates at rest and during the first 30min of exercise are shown in
Fig. 1. Exercise caused a marked rise in heart rate (F1,9¼147.1, P<0.001), but there
was no significant difference (F2,14¼0.1, P¼0.94) in heart rate between the four
times of day, nor any significant interaction (F4,27¼1.4, P¼0.27). If either the
Circadian Rhythms of Thermoregulation 259

Page 8

ORDER REPRINTS
changes from resting values or the values during exercise only were considered,
then there was still no significant time-of-day effect (F2,14¼0.1, P¼0.87) nor
interaction (F4,26¼0.90, P¼0.47), but there was a significant duration-of-exercise
effect (F3,22¼13.9, P<0.001).
Ratings of perceived exertion (RPE) are shown in Fig. 2. The RPE scores
showed a maximum of about 12, which is between fairly light and somewhat hard on
the Borg scale. That is, the exercise was not regarded by the subjects as difficult to
maintain at any of the times of testing. Values for RPE during exercise showed
highly significant effects of time of day (F2,14¼7.93, P¼0.005) and duration of
exercise (F1,9¼9.9, P¼0.008), and a significant interaction (F4,27¼4.1, P¼0.011).
4
6
8
10
12
14
05 101520 2530
Duration of Exercise (min)
RPE (Borg scale)
05:00 h
11:00 h
17:00 h
23:00 h
Figure 2.
exercise.
Mean values of rating of perceived exertion at rest and during the first 30min of
60
70
80
90
100
110
120
0510 15 20 2530
Duration of Exercise (min)
Heart Rate (beats/min)
05:00 h11:00 h
17:00 h 23:00 h
Figure 1.Mean heart rates at rest and during the first 30min of exercise.
260Waterhouse et al.

Page 9

ORDER REPRINTS
The interaction arises because RPE rose progressively throughout exercise at some
times but not others. If the times 05:00h and 17:00h were compared (one-way
ANOVA), then the effects of time of day (F1,7¼26.1, P¼0.001), duration of exercise
(F2,11¼8.18, P¼0.008), and the interaction (F2,16¼5.4, P¼0.014) remained, the
values at 17:00h continued to rise throughout exercise while the values at 05:00h
tended to remain constant. When the times 11:00h and 23:00h were compared, then
the time-of-day effect was not significant (F1,7¼2.1, P¼0.19), but there were still
significant effects of duration of exercise (F1,10¼8.9, P¼0.009) and of an interaction
(F3,4¼4.1, P¼0.029). In this case, the RPE values tended to reach a plateau at
11:00h but continued to rise at 23:00h.
Rectal Temperature
Rectal temperatures at rest and during the first 30min of exercise are shown in
Fig. 3. Inspection of this figure indicates that the core temperature at rest varied with
the time of day, values at 05:00h being the lowest. Group cosinor analysis indicated
that there was a significant (P¼0.022) circadian rhythm of resting core (rectal)
temperature: mesor?SD¼37.13?0.21?C; amplitude (95% confidence interval,
95% CI)¼0.35?C (0.09–0.61?C); acrophase (95% CI)¼17.8h (15.8–20.7h).
Exercise produced a clear rise in rectal temperature, but the rate of rise depended
upon the time of day, as can be seen more clearly in Fig. 4, where changes in rectal
temperature from resting values are shown. Statistically, the results when changes
from baseline values were considered were time-of-day effect, F2,13¼5.0, P¼0.026;
duration-of-exercise effect, F1,10¼111.9, P<0.001; interaction effect, F4,27¼5.5,
P¼0.003. The time-of-day and interaction effects confirm that rectal temperatures
rose from baseline values more quickly at some times of the day than others. When
changes in rectal temperature with exercise at 05:00h vs. 17:00h were compared,
the effects of time-of-day, duration-of-exercise, and the interaction effects retained
36.40
36.60
36.80
37.00
37.20
37.40
37.60
37.80
036912151821242730
Duration of Exercise (min)
Core Temperature (°C)
05:00 h11:00 h17:00 h23:00 h
Figure 3.Mean rectal temperatures at rest and during the first 30min of exercise.
Circadian Rhythms of Thermoregulation261

Page 10

ORDER REPRINTS
significance (F1,7¼13.2, P¼0.008; F2,10¼92.4, P<0.001; F3,18¼11.0, P<0.001,
respectively). By contrast, a comparison between 11:00h and 23:00h indicated a
significant effect of Duration of exercise (F2,12¼74.9, P<0.001), but no significant
effect of time of day (F1,7¼1.0, P¼0.34) or interaction between the main effects
(F2,12¼1.7, P¼0.23).
The time taken for the rectal temperatures to equilibrate varied from 30–63min,
and so the vast majority of equilibrium temperatures is not shown in Fig. 3. There
was no significant time-of-day effect (one-way ANOVA) when the times taken to
reach equilibrium were considered (F2,12¼0.8, P¼0.46). The equilibrium tempera-
tures themselves are shown in Fig. 5, from which it can be seen that the lowest
temperature remained at 05:00h. A significant (P¼0.032) circadian rhythm
remained, the parameters for which were mesor¼37.63?0.27?C; amplitude¼
0.25?C (0.03–0.47?C); acrophase¼18.0h (13.3–21.0h). Comparing the parameters
of the single cosinor analyses at rest and at equilibrium (Wilcoxon tests) indicated
a significant rise of mesor (P¼0.012) and a tendency for the amplitude to fall
(P¼0.089), but no significant change to the acrophase (P¼0.16). The fall in ampli-
tude of the equilibrium temperatures resulted from the fact that the rise in rectal
temperatures above resting values was greatest at 05:00h (Fig. 6). One-way ANOVA
confirmed this conclusion (F2,16¼11.1, P¼0.001).
The Negative-Exponential Model
Estimates of LIT and UAT from the negative-exponential model were compared
with the measured values for resting and equilibrium temperatures, respectively. One
-0.10
0.00
0.10
0.20
0.30
0.40
0.50
0.60
03691215 1821242730
Duration of Exercise (min)
Rise in Core Temperature (°C)
05:00 h
11:00 h
17:00 h
23:00 h
Figure 4.
exercise.
Mean changes from baseline of rectal temperatures during the first 30min of
262Waterhouse et al.

Page 11

ORDER REPRINTS
of the analyses (out of the total of 32: 8 subjects?4 times of day) involved many
more iterations than the others and gave results for UAT and K (but not for LIT)
that were very different from the others. It has been excluded from further treatment.
The mean differences?SD ([model prediction] minus [measured value]) for LIT
were ?0.01?0.04?C, and for UAT were þ0.02?0.04?C. These results are also
shown as Bland-Altman plots in Fig. 7. They indicate that there was no statistically
significant systematic error nor change in error with mean temperature.
Group cosinor analysis of the LITs gave a significant (P<0.005) fit with the
following parameters: mesor¼37.10?0.22?C; amplitude 0.40?C (0.17–0.63?C);
37.20
37.40
37.60
37.80
38.00
05:00 h11:00 h17:00 h 23:00 h
Time of Exercise
Equilibrium Temperature (°C)
Figure 5.Mean equilibrium temperatures (þSE) at the end of exercise.
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
05:00 h11:00 h 17:00 h23:00 h
Time of Exercise
Rise of Core Temperature (°C)
Figure 6.Mean changes from baseline (þSE) of rectal temperatures at equilibrium.
Circadian Rhythms of Thermoregulation263

Page 12

ORDER REPRINTS
acrophase 17.7h (15.3–18.9h). These values were close to those found by using the
measured resting temperatures (see above). Comparing the parameters of the single
cosine curves fitted to the LITs with those fitted to the measured data ([LIT—
measured], Student’s paired t-tests) gave paired differences that were statistically
nonsignificant (mesor: ?0.04?0.04?C, P¼0.30; amplitude: 0.03?0.04?C, P¼0.53;
acrophase: ?0.4?0.4h, P¼0.35).
Group cosinor analysis of the UATs gave a fit only verging on statistical
significance (P¼0.067). The curve had the following parameters: mesor¼37.68?C;
amplitude¼0.31?C; acrophase¼17.9h. Comparing the parameters of the single
cosine curves fitted to the UATs with those fitted to the measured data ([UAT—
measured equilibrium value], Student’s paired t-tests) gave paired differences that
were statistically nonsignificant (mesor: 0.05?0.03?C, P¼0.13; amplitude: 0.06?
0.06?C, P¼0.36; acrophase: 0.3?0.2h, P¼0.28).
That is, the results from the model accorded with those from experimental
observation. This applied also when the rises in core temperature from resting to
equilibrium values were considered. The mean difference between [UAT-LIT] and
[measured equilibrium value—resting value] was þ0.02?0.06?C, and the results are
Lower Intercept Temperature
-0.20
-0.10
0.00
0.10
0.20
36.0036.5037.0037.5038.00
Mean temperature (°C)
Model-Measured (°C)
Upper Asymptote Temperature
-0.20
-0.10
0.00
0.10
0.20
36.50 37.0037.5038.0038.50
Mean temperature (°C)
Model-Measured (°C)
Figure 7.
model with direct observations of temperatures. Above: lower intercept (baseline)
temperatures; Below: upper asymptote (equilibrium) temperatures. For more details, see text.
Bland-Altman plots. Comparison of estimates from the negative-exponential
264Waterhouse et al.

Page 13

ORDER REPRINTS
shown as a Bland-Altman plot in Fig. 8. They indicate that there was no statistically
significant systematic error nor change in error with mean temperature.
Forearm Skin Blood Flow and Temperature
Changes from resting values in forearm skin blood flow during exercise are
shown in Fig. 9; mean values rose to about 30% of maximal blood flow. There was
a significant effect of duration of exercise (F2,10¼28.2, P<0.001), but effects due
to time of day (F2,12¼0.5, P¼0.57) and the interaction between the main effects
-10.00
0.00
10.00
20.00
30.00
036912151821242730
Duration of Exercise (min)
Increase in Blood Flow (% Max)
05:00 h11:00 h17:00 h23:00 h
Figure 9.
exercise.
Mean increases above baseline in forearm blood flow during the first 30min of
Rise in Temperature
-0.20
-0.10
0.00
0.10
0.20
0.00 0.50
Mean Temperature (°C)
1.001.50
Model-Measured (°C)
Figure 8.
equilibrium from estimates made by the negative-exponential model and from direct
observations of the temperatures. For more details, see text.
Bland-Altman plot. Comparison of the rises in temperature from baseline to
Circadian Rhythms of Thermoregulation 265

Page 14

ORDER REPRINTS
(F3,18¼0.9, P¼0.45) were nonsignificant. However, this approach does not take
into account the fact that there were significant differences in the changes in rectal
temperature at different times of day. This is illustrated in Fig. 10, which relates the
changes in skin blood flow to those in rectal temperature. As this figure shows, even
though the changes in blood flow might not have been significantly different, they
took place against different amounts of thermal load, that is, different rises in rectal
temperature. What is required is an indication if the relationship between changes of
skin blood flow and changes in rectal temperature shows a time-of-day effect. The
ANCOVA enabled this analysis to be performed, investigating if there are effects of
time of day upon skin blood flow that were independent of the covariant, changes in
rectal temperature.
Initial exploration indicated that the quadratic term was not significant;
therefore the covariant was investigated as a linear term only. It was found to be
highly significant (F4,236¼8.9, P<0.001). The ANCOVA was performed twice, first
with the 05:00h data as the baseline value and, second, with the 11:00h data as the
baseline value. With the 05:00h data acting as a baseline value, the coefficient
relating Skin Blood Flow to Rectal Temperature had a value of 21.23 (P¼0.001);
the value for the coefficient at 11:00h was significantly greater (by þ14.27,
P¼0.008); that at 17:00h was less, but not significantly so (by ?6.78, P¼0.26); and
that at 23:00h was significantly greater (by þ21.31, P¼0.001). When the data
at 11:00h acted as a baseline, then the coefficient was 35.50 (P<0.001); the coeffi-
cients at 05:00h, 17:00h, and 23:00h were, respectively, significantly less (by ?14.27,
P¼0.008), significantly less (by ?21.05, P¼0.001), and nonsignificantly greater
(by þ7.03, P¼0.27). In summary, these results indicate that there were significant
differences in the relationships between changes in skin blood flow and changes
in rectal temperature when measured at different times of the day. However, the
0
5
10
15
20
25
30
0 0.1 0.20.30.40.5 0.6
Rise in Rectal Temperature (°C)
Rise in Blood Flow (% Max. Flow)
05:00 h11:00 h17:00 h23:00 h
Figure 10.
above baseline during the first 30min of exercise.
A comparison of mean increases in forearm blood flow and rectal temperatures
266Waterhouse et al.

Page 15

ORDER REPRINTS
differences were not statistically significant in the cases of Predictions 1 and 2, that is,
comparing 11:00h vs. 23:00h and 05:00h vs. 17:00h, respectively.
Figure 11 shows the changes from resting values of skin temperature during
exercise at different times of day. At all times, there was an initial decrease, but later
changes consisted of rises in skin temperature that were greater at some times of the
day than others. However, two-way ANOVA did not show any significant effects of
time of day (F2,16¼1.6, P¼0.23), duration of exercise (F1,8¼1.1, P¼0.34), or the
interaction between these two factors (F4,25¼1.7, P¼0.19), due to the large amount
of interindividual variation associated with this measurement. Again, the important
issue is whether the relationship between changes in skin temperature and changes
in rectal temperature depended upon the time of day. Figure 12 plots the changes in
skin temperature compared with changes in rectal temperature. It draws attention to
the fact that the rises of rectal temperature were accompanied initially by falls of skin
temperature but that the recovery of skin temperature and its subsequent rise above
baseline values were far more rapid at 23:00h than at other times of day.
Initial explorations with the ANCOVA model indicated that the covariates
exerted a highly significant effect (F5,235¼31.1, P<0.001), and that they could be
described by statistically significant negative linear and positive quadratic terms.
With the 05:00h data acting as a baseline, the linear coefficient relating skin
temperature to rectal temperature had a value of ?1.32 (P¼0.022), and the
quadratic coefficient was þ4.68 (P<0.001). The same quadratic coefficient was used
for the data at the other three times, and the changed value of the linear coefficient
was investigated. The value for the linear coefficient at 11:00h was less negative,
but not significantly so (by þ0.33, P¼0.26); that at 17:00h was significantly less
negative (by þ0.97, P¼0.002), and that at 23:00h was significantly greater and
positive (by þ3.58, P<0.001). When the data at 11:00h acted as the baseline, then
the linear coefficient was ?1.00 (P¼0.07). The linear coefficients at 05:00h, 17:00h,
and 23:00h were, respectively, more negative, but not significantly so (by ?0.33,
-0.40
0.00
0.40
0.80
0369 1215 182124 27 30
Duration of Exercise (min)
Change in Skin Temperature (°C)
05:00 h11:00 h17:00 h23:00 h
Figure 11.
exercise.
Changes from baseline in forearm skin temperatures during the first 30min of
Circadian Rhythms of Thermoregulation 267