• Positron emission tomography (PET) was used to identify the neuroanatomical correlates underlying ‘central command’ during imagination of exercise under hypnosis, in order to uncouple central command from peripheral feedback.
• Three cognitive conditions were used: condition I, imagination of freewheeling downhill on a bicycle (no change in heart rate, HR, or ventilation, ): condition II, imagination of exercise, cycling uphill (increased HR by 12 % and by 30 % of the actual exercise response): condition III, volitionally driven hyperventilation to match that achieved in condition II (no change in HR).
• Subtraction methodology created contrast A (II minus I) highlighting cerebral areas involved in the imagination of exercise and contrast B (III minus I) highlighting areas activated in the direct volitional control of breathing (n = 4 for both; 8 scans per subject). End-tidal P_CO2 (P_ET,CO2) was held constant throughout PET scanning.
• In contrast A, significant activations were seen in the right dorso-lateral prefrontal cortex, supplementary motor areas (SMA), the right premotor area (PMA), superolateral sensorimotor areas, thalamus, and bilaterally in the cerebellum. In contrast B, significant activations were present in the SMA and in lateral sensorimotor cortical areas. The SMA/PMA, dorso-lateral prefrontal cortex and the cerebellum are concerned with volitional/motor control, including that of the respiratory muscles.
• The neuroanatomical areas activated suggest that a significant component of the respiratory response to ‘exercise’, in the absence of both movement feedback and an increase in CO₂ production, can be generated by what appears to be a behavioural response.

Imagining movement (Decety et al. 1991) or attempting to contract muscles before and after regional anaesthesia (Gandevia & Hobbs, 1990; Jørgensen et al. 1993) or neuromuscular blockade (Secher, 1985; Gandevia et al. 1993) has been used to dissociate peripheral neural signals from central command. Attempted muscle contraction in these studies enhances the cardiovascular response. Nowak et al. (1999) reported that before axillary blockade, handgrip exercise activated the contralateral sensory motor area, supplementary motor area and ipsilateral cerebellum. This pattern of activation was not affected by regional anaesthesia even though the strength of contraction was reduced (Nowak et al. 1999), suggesting that the activated areas may be involved in generating command-related cardiovascular responses to exercise.

Mental imagery of treadmill exercise at rest (Decety et al. 1991) has been shown to increase heart rate (HR) and ventilation (); however, in more recent work (Wuyam et al. 1995) the response was only seen in athletes. Imagination of exercise has also been studied under hypnosis (Levin & Egolinsky, 1936; Nemtzova & Shatenstein, 1936; Agosti & Camerota, 1965; Daly & Overley, 1966; Morgan et al. 1976; Wang & Morgan, 1992) with all studies showing increases in and one (Arvidsson et al. 1970) showing increases in cardiac output. Hypnosis allows isolation from the environment and therefore permits a more directed imagination than is possible when awake (Morgan et al. 1973; Coe et al. 1980). In this state, we have used imagination of exercise at rest in experimentally naive volunteers as a cognitive tool to uncouple central command from peripheral feedback. We show in humans that a significant component of the cardiorespiratory response to the imagination of exercise can be generated in the absence of both movement feedback and any increase in carbon dioxide production. Using PET we have identified higher brain areas that may encode a cardiorespiratory motor programme.

A part of this study has been presented in abstract form (Thornton et al. 1997; Paterson et al. 2000).

All volunteers gave their informed consent whilst remaining experimentally naive regarding cardiorespiratory control. Experiments were carried out with local ethics committee approval and with permission from the Administration of Radioactive Substances Advisory Committee, London, to administer radioactivity to male subjects over 25 years of age. All experiments conformed to the Declaration of Helsinki.

Two studies were undertaken, all under hypnosis. Study 1: imagination of exercise whilst at rest. Study 2: neuroimaging during imagination of exercise whilst at rest.

Throughout this paper, ‘awake’ is used to describe the normal conscious state and ‘under hypnosis’ refers to the hypnotized state, which is clearly distinguishable from sleep.

In study 2, eight experimentally naive men who scored 6 or more out of 12 on the Stanford scale were familiarized with leg exercise cycle ergometry in the semi-reclined position in the awake state.

A transmission scan was performed using an external ¹³⁷Cs positron source to permit compensation for the field distortion occurring due to the subject's head. Scans occurred at 6 min intervals. Each consisted of a 30 s background scan (to allow correction for background activity), followed after a 15 s delay by a 20 s infusion of radiotracer. An activation scan of 90 s was then taken to coincide with the steady-state cardiorespiratory response. Suggestion of the required imagination was made just before the infusion of radiotracer commenced. The integrated counts for the activation scan, corrected for background activity, gave an estimate of regional cerebral blood flow. All subjects studied with PET had a structural MRI scan on a different date.

Protocol 1: imagination of exercise, awake and under hypnosis Hypnotized subjects (n = 17) imagined themselves exercising for 2 min at the ‘heavy’ work rate with which they had been familiarized. It was suggested to the subjects that they were cycling up a steep hill with a grade equivalent to the heavy work rate. In order to assess the repeatability of the response, four of these subjects performed the protocol eight times at 6 min intervals. Imagination of exercise was repeated in eight of the subjects in the awake state. In five subjects bipolar surface EMGs were obtained during imagination of exercise under hypnosis from left and right quadriceps and hamstrings.

Protocol 2: imagination of freewheeling downhill under hypnosis In order to examine the cardiorespiratory effects of cognitive effort per se, hypnotized subjects (n = 8) were asked to imagine themselves on a bicycle freewheeling down a hill, whilst applying the brakes gently. This protocol was performed for 2 min.

Protocol 3: volitional hyperventilation under hypnosis Hypnotized subjects (n = 6) were told to ‘breathe faster’ for 2 min. Feedback on the breathing frequency required was provided by the hypnotist and was set at the level achieved during imagination of exercise.

In order to examine whether the magnitude of any hyperventilation was limited by hypocapnia, seven subjects during study 1 (protocol 3) had CO₂ added to the inspired air to maintain isocapnia.

Neuroimaging imagination of exercise at rest PET scanning was used to image neuro-anatomical correlates of ‘central command’. We employed cognitive subtraction methodology to create two contrasts (A and B) in two separate protocols with different subjects (n = 4 for both; 8 scans per subject). Three cognitive conditions were used: condition I, imagination of freewheeling downhill on a safe country road; condition II, imagination of exercise, cycling uphill on the same road; condition III, volitionally driven increase in f (under instruction of the hypnotist) to match that achieved in condition II whilst imagining a safe country road. The order of the conditions was randomized in a balanced design, with the restriction that condition III was always the last to be studied. In the first protocol, freewheeling was compared with cycling uphill to produce contrast A, highlighting cerebral areas involved in the imagination of exercise. In the second protocol, freewheeling was compared with a voluntary increase in f to produce contrast B, highlighting areas activated in the direct volitional control of f as opposed to indirect activation as a consequence of imagination of exercise. Any effects of both imagination and hypnosis were therefore constant throughout all conditions. All subjects in this study had CO₂ added to the inspired air to maintain isocapnia.

Breathing Analog signals were sampled at 200 Hz (Biopac MP100WS) and recorded on a Power Macintosh computer using Acqknowledge software for subsequent analysis. Breath-by-breath data from each subject were divided into 15 s epochs (study 1, protocols 1 and 3; study 2) or 30 s epochs (study 1, protocol 2) and averages were calculated for each epoch. The average value from 30 to 15 s prior to the onset of the instruction to imagine was taken as the control value; this was then subtracted from each epoch average. Changes were then averaged across the group and all data were expressed as means ±s.e.m. Comparison of imagined exercise with and without inspired CO₂ was evaluated by a repeated measures ANOVA and post hoc Tukey test.

Positron emission tomography Analysis of the PET data was performed using statistical parametric mapping (SPM96) (Frackowiak et al. 1997). Images were reconstructed as axial planes and realigned to remove the effects of head motion. They were spatially normalized to the Montreal Neurological Institute (MNI) template using the subject's own T1 MRI to guide this process and then smoothed by an isotropic 10 mm, full-width half-maximum, isotropic Gaussian kernel filter to account for individual variation in gyral anatomy and to improve the signal-to-noise ratio. The standard stereotaxic space was that of the Montreal Neurological Brain (Evans, 1993) but the coordinates presented in Table 2 have been corrected (A. Meyer-Lindenberg, unpublished data: http://www.mailbase.ac.uk/lists/spm/1998-06/0079.html) to those of Talairach space (Talairach & Tournoux, 1988) to facilitate comparison with previous studies.

Table 2 Brain activations under hypnosis during imagination of exercise and voluntary hyperventilation

A small volume correction factor was applied only to areas activated where a prior hypothesis existed concerning areas known to be involved in the control of breathing (Worsley et al. 1996). These regions of interest were hand-edited on a copy of the SPM canonical T1 template, using Analyze AVW software (Robb et al. 1989). Each region of interest image was entered into SPM96 and the corrected P value for peak voxels within these regions was calculated (M. Brett, unpublished data: http://www.mrc-cbu.cam.ac.uk/Imagining/vol-corr.html). SPM96 corrects for multiple comparisons using a Euler characteristic. This characteristic is dependent on the number of resolution elements (resels) with a given image, which itself depends on the number of voxels within the image and the 3-D smoothness of that image. A small volume correction reduces the number of resels in the search volume to that of the brain region of interest in question; thus the threshold for accepting a given Z score as significant (P < 0.05) is reduced (Worsley et al. 1996).

The peak work rate attained during the incremental exercise test was 172 ± 8 W (mean ±s.e.m.), corresponding to a peak _O2 of 2332 ± 187 ml min⁻¹. The heavy work rate assigned to each subject was of the order of 103 ± 5 W (_O2 of 1356 ± 83 ml min⁻¹).

Table 1 Cardiorespiratory data in subjects, awake and hypnotized, in both actual and imagined heavy exercise

Imagination of exercise under hypnosis (protocol 1) In the absence of hypnosis there was no effect on cardiorespiratory parameters of imagining heavy exercise (n = 8, Table 1B). In 17 resting hypnotized subjects, imagination of the previously performed heavy exercise task resulted in hyperventilation and hypocapnia (Table 1C, Fig. 1 and Fig. 2) and an increase in HR (Table 1C). The dynamics of the and HR responses over the 2 min period of imagination were well fitted by monophasic exponentials (group data; : t_{50 %} = 16 s, goodness of fit of regression coefficient r² = 0.98; HR: t_{50 %} = 5 s, r² = 0.96). The magnitude of the increase in ventilation was ca 30 % of that seen during actual performance of the exercise task under hypnosis (Table 1A and C) but occurred via an increase in f with no change in V_T. The magnitude of the ventilatory response to imagination of exercise was not affected by the addition of CO₂ to the inspired air to maintain eucapnia (n = 7, repeated measures ANOVA, P = 0.345, Fig. 3). The magnitude of the HR response to the imagination of exercise was 12 % of that seen during actual exercise (Table 1A and C).

Figure 1 Imagination of exercise under hypnosis at rest

Figure 2 Imagination of heavy exercise under hypnosis

Figure 3 Imagination of exercise under hypnosis at rest whilst maintaining isocapnia

In the four subjects who repeated the protocol eight times, the magnitude of the ventilatory responses was similar between repeats. In this subgroup, the coefficient of variation of the steady-state ventilatory response was 9 %. There was no obvious movement in the legs during imagination of exercise, and in five subjects the lack of rhythmic motor activity was confirmed by the absence of any significant increase in quadriceps or hamstring EMG activity (Fig. 4).

Figure 4 EMG activity during voluntary contraction and imagination of exercise

When asked about the experience, subjects gave vivid accounts (‘…I was really into it, I felt like I had to give the pedals a real shove to get them to go round…‘) and reported sensations that included tiredness in the legs (‘I knew that my legs weren't moving, but they felt really tired and I was battling against myself to keep going.‘) and sweating (‘…my legs started tingling as if I was about to break out into a sweat, it was like I was blushing…‘). Seven subjects mentioned that they noticed their breathing had increased.

Imagination of freewheeling downhill under hypnosis (protocol 2) Imagination of freewheeling downhill under hypnosis did not result in any changes in cardiorespiratory variables (n = 8, Table 1D).

Volitional hyperventilation under hypnosis (protocol 3) The instruction to breathe faster resulted in an immediate and sustained increase in ventilation with a square-wave response. The magnitude of volitional hyperventilation was not affected by maintaining isocapnia (n = 6, Table 1E).

Figure 5 Activations during imagination of exercise and voluntary hyperventilation with associated respiratory responses

Table 2 shows the coordinates for the most activated voxels within relevant brain areas that are also significant (P < 0.05). For contrast A (imagined exercise), these areas are shown in Fig. 6 as through projections in three anatomical views; some of these areas are further illustrated (Fig. 5) on a rendered representative brain (Evans, 1993). The activations in the right dorso-lateral prefrontal cortex and supplementary motor areas in the midline remained significant after correction for multiple comparisons (Pm). The activations in the right premotor area, both superolateral sensorimotor cortices, midline thalamus and the cerebellum bilaterally were significant following small volume correction (Psmvc), but not following multiple comparsions. A further analysis correlating the intensity of a given voxel with for all data in contrast A showed that the right premotor area, right dorso-lateral prefrontal cortex and left postero-lateral cerebellum remain significant (Z > 4.8) after correction for multiple comparisons. In the absence of a prior hypothesis, the right parietal association area narrowly missed being significantly activated after correction for multiple comparsions (Pm = 0.07; x = 62, y =−40, z = 19, Z = 4.6). Of interest, there was a non-significant activation in the left insula cortex (Z = 3.4 Psmvc = 0.11). For contrast B (voluntary hyperventilation), Table 2 and Fig. 5 also show significant activations in the supplementary motor area and the lateral sensorimotor areas. There was no activation of the insula cortex.

Figure 6 Areas of brain activated during imagination of exercise

Steady-state cardiorespiratory responses within the scanner during contrast A were as follows. Condition I caused no significant change in or HR (Fig. 5i). Condition II increased from 10.8 ± 0.5 to 19.3 ± 2 l min⁻¹, and HR from 64 ± 3 to 71 ± 5 beats min⁻¹; the dynamics of the increase in and HR followed an exponential (: t_{50 %} = 16 s; HR: t_{50 %} = 5 s). For contrast B, condition I also caused no significant change in HR or (Fig. 5iii). Condition III caused no change in HR (from 59 ± 5 to 60 ± 5 beats min⁻¹), but increased from 8.8 ± 0.2 to 13.6 ± 1.5 l min⁻¹ and f from 14 ± 2.2 to 26.2 ± 3.6 breaths min⁻¹; V_T decreased from 0.7 ± 0.1 to 0.6 ± 0.1 l and P_ET,CO2 decreased from 42.9 ± 1.4 to 41.8 ± 1.1 Torr.

The imaging results of the present study have identified the neuroanatomical correlates likely to be responsible for a ‘central command’ mechanism underlying ‘exercise’ hyperpnoea without movement feedback.

The supplementary motor area (SMA) is activated both during single leg actual exercise (Fink et al. 1995) and in the present study of imagined exercise. It is well established that the SMA/PMA (both planning and programming areas; Roland et al. 1980; Eccles, 1982), cerebellum (Decety et al. 1988) and the dorso-lateral prefrontal cortex (Fuster, 1997), are concerned with volitional/motor control, including that of the respiratory muscles (Fink et al. 1995). Mental imagery of novel precise movements activates all these areas, except for the motor cortex (Ingvar & Philipson, 1977). Moreover imagination of exercise in athletes demonstrates that (entirely in f) increases in proportion to the degree of imagined effort (Decety et al. 1993).

We interpret the activation of the dorso-lateral prefrontal cortex (seen in contrast A but not in contrast B) to reflect the task dictated by the experimental paradigm of imagining exercise without its execution, but at the same time evoking a cardiorespiratory response that mimics that seen in real exercise. The dorso-lateral prefrontal cortex is crucial in the self-generation of action or in actions generated on the basis of working memory (Frith et al. 1991). The output projections of the dorso-lateral prefrontal cortex are to the SMA and PMA but not to the motor cortex itself. If the movement is complex and novel, there is increased involvement of the prefrontal areas (Passingham, 1993), although it has not been established whether these areas are activated during task-familiar actual leg exercise (Fink et al. 1995). However, Pedersen et al. (1998) have shown activation of the left prefrontal cortex (middle frontal gyrus) during uncued movements of the right index finger, suggesting that Brodmann area 9 is directly related to motor command. The dorso-lateral prefrontal cortex receives information from the parietal association areas known to process information regarding perception of oneself in relation to the environment. This area narrowly missed significance in the present study. It is worth noting that the thalamus (significantly activated) has a projection to the parietal association areas.

In contrast B the SMA was significantly activated, but there were no activations in the superolateral sensorimotor cortices, which is a surprising finding given previous reports that have shown these areas to be activated during awake volitional inspiration (e.g. Gandevia & Rothwell, 1987; Colebatch et al. 1991). However, there was a strong activation in the lateral sensorimotor cortex that has been previously reported (but never explained) in volitional driven breathing, particular with the use of a face mask (Fink et al. 1996; Murphy et al. 1997). Activation of the lateral sensorimotor cortices presumably reflects sensorimotor changes in the face and/or upper airway (Penfield & Boldrey, 1937). The focus on changing breathing frequency as required in condition III of contrast B may have caused increased self-awareness of such anatomical structures.

We did not observe any significant activation in the sub-thalamic area (Fig. 7A; Eldridge et al. 1981). We also did not visualize any significant activation that we were confident of in the pontomedullary area. This latter observation could reflect limits of the scanner field of view down to the pontomedullary margin (Fig. 7B) together with ‘edge’ effects from activations of the medial cerebellum (Fig. 6, sagittal).

Figure 7 Transverse and sagittal activation during imagination of exercise

We observed a small increase in HR during imagination of exercise although there was only a non-significant activation in the left insula cortex. During volitionally driven breathing there was no change of HR or any activation in the left insula. Others (Williamson et al. 1997), however, have reported left insula activation during mild leg exercise with HR changes significantly greater than in the present study. There is evidence that this area projects to the ventrolateral medulla, lateral hypothalamus and nucleus tractus solitarius (Yasui et al. 1991). Electrical stimulation of the left insula cortex in the rat increases HR (Saper, 1982), suggesting that this area may be involved in the withdrawal of vagal tone to the sino-atrial node thereby allowing the rapid increase of HR at the onset of exercise. The lack of activation in our study may reflect the small change in HR and the limitation of small volume correction in a small area of brain.

The increases in seen during imagination of exercise under hypnosis in this study were via increases in f with no steady-state alterations in V_T. This is consistent with previous findings obtained in awake subjects imagining exercise (Decety et al. 1991; Wuyam et al. 1995) and in hypnotized subjects (Agosti et al. 1965), although another previous study of imagined exercise under hypnosis reported hyperventilation with increases in both f and V_T (Arvidsson et al. 1970). We have no good explanation for the fact that V_T did not change during imagination of exercise in the present study, but we speculate that drives to breathe may act independently on f and V_T.

The use of hypnosis as a tool to modify the perception of exercise load was first reported by Nemtzova & Shatenstein (1936) and has been used by others (Morgan et al. 1973, 1976). Thornton et al. (1997) reported that subjects performing constant moderate exercise hyperventilated and become hypocapnic if told that the work rate had increased when in fact it had not changed. In this study the hyperventilation occurred via increases in f with no steady-state changes in V_T. The magnitude of the ventilatory response was similar to that expected from an actual increase in work rate. Krogh & Lindhard (1913) were the first to describe the effect of altered perception on ventilatory responses to exercise. ‘In one experiment the subject J.L. was told to work at a rapid rate and with a heavy load (3 kg) but at the signal to begin, the current was not turned on to the magnets of the ergometer. The resulting respiration during the first few seconds was practically the same as if there had been a heavy load, though the actual work performed was insignificant…‘

A problem related to the use of hypnosis is that there is no satisfactory objective measure of the depth of the hypnotic state. Given the behaviour of the subjects (immobility, automatic responding) and the vividness of the subjects’ descriptions of hypnosis (partial amnesia, altered perception of the passage of time) our medical hypnotists were confident that the subjects were in a deep hypnotic state. There was also no evidence that the depth of hypnosis affected the magnitude of the results as the subjects exhibiting the largest cardiopulmonary responses were not necessarily those who were classed as being in the deepest hypnotic state.

A major limitation of PET studies is that the amount of radioactivity that can be given to any subject is clearly restricted. It was not possible therefore to have the same subject complete all three protocols in study 2 (PET). This necessitated two independent groups and prevented us from confidently studying a three-way interaction among conditions. As a consequence the validity of a statistical comparison between the results of contrasts A and B assumes that the two groups of subjects activate and deactivate the same areas and to the same degree when imagining the control state (freewheeling downhill). We cannot establish this point from our data. Nevertheless we have done a comparison of the results from contrasts A and B using a mixture of conjunction analysis and masking. Specifically, we have loaded up data from both contrasts into one design matrix and looked across the two protocols for commonalities or lack of commonalities in activation and deactivation areas. We found that many of the areas we were interested in behaved differently between the two protocols. However, it would be misleading to give the impression from these data that the behaviour in these areas was quantitatively different. The activations seen in contrast A (imagination of leg exercise, no movement feedback) are similar to those areas previously reported by Fink et al. (1995) during actual leg exercise. Moreover, for the same respiratory drive seen in contrast B (volitional hyperventilation), the dorso-lateral prefrontal cortex, PMA, superolateral sensorimotor cortex and cerebellum (activated in contrast A) were not activated in contrast B, highlighting a clear difference between the two cognitive conditions that resulted in the same behavioural end-point.

J.M.T. held an MRC DPhil studentship. We are grateful to the McDonnell-Pew Centre for Cognitive Neuroscience, Oxford, for supporting the neuroimaging study. A.G. was supported by a grant from The David and Frederick Barclay Foundation. We also thank Dr Richard Wise and Dr Paul Grasby from the MRC Cyclotron Unit, Hammersmith Hospital, London, UK, for advice on the analysis of the PET data.