Results

The systematic literature search returned 2749 articles. We determined 16 relevant for our synthesis based on the inclusion and exclusion criteria. A further five articles were added based on citation searching and our knowledge of the field (figure 1).

There are no studies reporting effects of exercise as a zeitgeber (or indeed otherwise) on the human SCN. This is expected given the difficulty in accessing the human SCN. Furthermore, there are no reports of chronotype changes following exercise intervention. Evidence for a phase response curve to exercise in humans has been reviewed by Edwards et al 26 and our systematic search revealed no new studies that have attempted to demonstrate a phase response curve to exercise in humans. Edwards et al note that these studies represent supporting evidence that exercise might be a human zeitgeber but fall short of conclusively claiming that an exercise-zeitgeber has been demonstrated for humans.26 Thus, evidence for phase response curves will only be considered in brief.

There is abundant evidence of phase-shifting effects of exercise on commonly measured proxies of circadian rhythm (table 2). Exercise, depending on time of day or internal time of stimulus application, has been shown to phase-shift the melatonin, thyroid stimulating hormone (TSH) and BT rhythms.19 27–32 However, whether this effect is the result of masking or whether it persists after stimulus removal is unclear. From one study, no significant effect from the first exercise stimulus was observed, rather periodic bouts were required to manifest melatonin phase changes.19 This would suggest a zeitgeber effect rather than a masking effect.19 Buxton et al observed counter-intuitive phase-shifting responses to exercise at certain times with morning exercise eliciting phase delays and evening exercise eliciting phase advances.27 The authors plotted a tentative phase response curve based on their data.27 That the responsiveness to exercise followed a different timing pattern than would be expected for light may suggest the phase-shifting was the result of exercise per se and not light exposure.26 Van Reeth et al also plotted a tentative phase response curve to exercise.30 In a shortened forced sleep–wake schedule study, exercise did not aid adaptation to the new schedule, rather the melatonin ascending limb was phase-delayed compared with non-exercise by day 8.31 Furthermore, the phase points of the rhythm shifted in a temporally different manner with the descending phase not statistically different between groups after the initial 8 days but different after 14 days.31 In other words, some phase-shift differences took longer to manifest than others. If the effects observed are immediate effects rather than entraining effects, the time-period for manifestation of phase-shifts suggests an adaptation to the immediate effects. We would find this more difficult to reconcile than entrainment. In a longitudinal case study, high intensity exercise delayed the expression of clock genes in human hair cells compared with a following period of no exercise.33 Unfortunately, the first non-exercise circadian measurements were taken ~3 weeks after the exercise period ended; thus, a drifting of the phase was not tested.33 Added to this, phase differences were observed between the initial exercise-period circadian measurements and the second exercise-period circadian measurements.33 Hence, it is difficult to conclude that exercise was the zeitgeber. For instance, seasonal effects on phase of entrainment might have been detected. A repeat of this study, with more immediate observations after removal of the exercise stimulus could provide interpretable evidence of an exercise-induced circadian stimulus.

Several studies demonstrate that habitual external timing of exercise can (in a time-dependent manner) differentially affect the diurnal variation in performance (table 3).34–39 These studies are longitudinal involving repeated measures testing with several groups exercising at different times per day. However, we do not know how the different exercise schedules may have affected the timing of other zeitgeber exposure. For instance, morning exercise groups, as opposed to evening exercise groups, may be waking up earlier and being exposed to light earlier in the day. Over several weeks, they will entrain to this earlier light exposure, and their performance rhythm will also be expected to advance, which could explain the observations. On the other hand, some studies noted no differences in morning or evening BT between morning-trained and evening-trained groups which suggests either light exposures are not different between groups or are not sufficiently different to differentially affect the BT rhythm.34 36 A case study of exercise earlier in the day, measuring BT at several time points, identified changes in BT amplitude post-intervention but not in the observed phase timing of peaks or troughs.38 Another study demonstrates changes in diurnal variation of performance but not in the diurnal rhythm of testosterone or cortisol.39 This leads to the question: “why might the performance rhythm phase-shift separately from other well-described circadian rhythms such as BT, cortisol or testosterone?” A ‘masking effect’ of exercise on exercise performance could offer an explanation. However, this non-circadian effect needs to have lasted from 48 hours up to 2 weeks after the last exercise session as per the timing of the post-intervention testing session in these studies.34 36 37 What ‘washout’ period would be required to observe circadian effects from masking effects remains open. One study indicated no effect of habitual time of exercise on diurnal variation in performance.40 Different to the other studies reported here,34–39 the authors did not observe a diurnal variation in performance to begin with.40 They suggest that the tropical environmental conditions in Guadeloupe may have played a role.40 Remarkably, even an acute bout of submaximal cycling performed the day before a time-trial improved performance when it occurred at the same time-of-day as the time-trial.41 The authors suggest chronotype preference, fatigue (glycogen and lactate levels), circadian phase-shift and time familiarisation as the potential explanations, but that the latter is the most likely.41 However, other studies have observed phase-shifting of performance independent of BT34 36 ; thus, a circadian phase-shift of performance rhythm independent of BT may be possible.

The following studies demonstrate the potential for exercise to interact with the human circadian system to improve health (table 4). Chen et al, after detecting associations between poor rest–activity rhythms and poor sleep in patients with lung cancer,42 found that 6 months of prescribed individual-specific exercise improved objective sleep in patients with lung cancer compared with a “usual care” control group.43 The effect was strongest in individuals with poorer rest–activity rhythms.43 That the exercise targets were specific to each patient based on activity scores at the beginning of the trial could, at least in part, control for general fitness improvements affecting sleep. Thus, that this effect was observed more strongly in patients with poorer rest–activity rhythms indicates a potential circadian component to the exercise effect. Although time of day of exercise was not noted, 3 months of exercise improved intradaily rest–activity rhythmicity measured 1 month postintervention in an elderly population.44 The participants had discontinued the exercise regimen in this post 1 month period.44 Moreover, the effects were not attributed to changes in light exposure as no participants increased outdoor activities.44 There were no fitness and pre-exercise rest–activity rhythm associations, thus, improved fitness is likely to play only a minor role, if any, in the observed improved rest–activity rhythms. Exercise, depending on time of day, could differentially augment nocturnal blood pressure “dips”, time spent in different sleep-stages and sleep-onset latencies in the 24 hours following the stimulus.45 Consecutive measurements or measurements following a longer “washout” period were not taken. Thus, potential zeitgeber effects cannot be distinguished from masking effects. One month of evening exercise in Milan, prior to travel to New York, proved better than 1 month of morning exercise in terms of improving sleep and the circadian rhythm of rest–activity post flight.46 The authors report that the time-of-day of sport contributed to the circadian rhythm and sleep response to transmeridian travel.46 The authors state that bedtimes and waking times were set for each subject in this study, but they were not reported in the manuscript.46 Thus, both circadian phase and daily light exposure could have significantly contributed to faster adaptation to sleeping at the prescribed bedtime in New York in the evening-trained group.