OUP user menu

Do the eyes scan dream images during rapid eye movement sleep? Evidence from the rapid eye movement sleep behaviour disorder model

Laurène Leclair-Visonneau, Delphine Oudiette, Bertrand Gaymard, Smaranda Leu-Semenescu, Isabelle Arnulf
DOI: http://dx.doi.org/10.1093/brain/awq110 1737-1746 First published online: 16 May 2010

This article has a correction. Please see:


Rapid eye movements and complex visual dreams are salient features of human rapid eye movement sleep. However, it remains to be elucidated whether the eyes scan dream images, despite studies that have retrospectively compared the direction of rapid eye movements to the dream recall recorded after having awakened the sleeper. We used the model of rapid eye movement sleep behaviour disorder (when patients enact their dreams by persistence of muscle tone) to determine directly whether the eyes move in the same directions as the head and limbs. In 56 patients with rapid eye movement sleep behaviour disorder and 17 healthy matched controls, the eye movements were monitored by electrooculography in four (right, left, up and down) directions, calibrated with a target and synchronized with video and sleep monitoring. The rapid eye movement sleep behaviour disorder-associated behaviours occurred 2.1 times more frequently during rapid eye movement sleep with than without rapid eye movements, and more often during or after rapid eye movements than before. Rapid eye movement density, index and complexity were similar in patients with rapid eye movement sleep behaviour disorder and controls. When rapid eye movements accompanied goal-oriented motor behaviour during rapid eye movement sleep behaviour disorder (e.g. grabbing a fictive object, hand greetings, climbing a ladder), which happened in 19 sequences, 82% were directed towards the action of the patient (same plane and direction). When restricted to the determinant rapid eye movements, the concordance increased to 90%. Rapid eye movements were absent in 38–42% of behaviours. This directional coherence between limbs, head and eye movements during rapid eye movement sleep behaviour disorder suggests that, when present, rapid eye movements imitate the scanning of the dream scene. Since the rapid eye movements are similar in subjects with and without rapid eye movement sleep behaviour disorder, this concordance can be extended to normal rapid eye movement sleep.

  • rapid eye movement sleep
  • eye movements
  • rapid eye movement sleep behaviour disorder
  • dreaming


Rapid eye movement (REM) sleep has been characterized for over 50 years by bursts of eye movements, dream imagery and cortex activity that is more rapid and desynchronized than that of non-REM sleep (Aserinsky and Kleitman, 1953), as well as marked perfusion of the temporal and occipital brain areas (Maquet et al., 1996; Braun et al., 1998). These markers of CNS activation contrast with the generalized atonia of postural muscles (Jouvet and Michel, 1959). REMs are rapid saccades visible under the closed eyes of the sleeper, similar to saccades in the awake state when visual inputs are absent but imagined (Herman et al., 1983; Sprenger et al., 2010). The eyeballs rotate at a speed greater than 30° per second (Takahashi and Atsumi, 1997). REMs are present during 14–27% of REM sleep time, with 5–35 REMs per minute. REM activity peaks 5–10 min after the onset of a REM sleep period and then declines significantly 10 min later. The frequency of REMs increases from early to late REM sleep periods (Aserinsky, 1971).

Whether the directional properties of REM sleep eye movements are related to shifts of gaze related to dream imagery (the ‘scanning hypothesis’), or are random markers of brainstem activation, has fascinated sleep scientists since the discovery of REM sleep (Dement and Kleitman, 1957; Berger and Oswald, 1962; Jacobs et al., 1972). As evidence against the ‘scanning hypothesis’, subjects who have been blind since birth have REMs but no visual experience during their dreams (Gross et al., 1965). REMs persist in cats after lesion of the visual cortex (Jouvet, 1962) and in pontine preparations (Arnulf et al., 1998), although the number and complexity of these REMs is reduced. In these instances the REMs do not follow visual imagery. Foetuses and neonates display abundant REMs during agitated sleep (a precursor of REM sleep), despite their lack of vision. In monkeys, several binocular REMs are non-conjugated, meaning that there is no fixation point (Zhou and King, 1997). Several authors reject the scanning hypothesis and suggest that the REMs would rather be used for lubricating the ocular surface (Murube, 2008), warming the brain (Wehr, 1992), stimulating and stabilizing the circuits that have been insufficiently activated during waking (Herman and Roffwarg, 1983), affording intense endogenous stimulation to enhance the ontogenic development of the CNS (Roffwarg et al., 1966) or have no purpose, being random markers of brain activation. However, the possibility remains that there are central commands for eye movements related to dream imagery. In studies performed in normal humans, subjects are awakened during REM sleep (while their eye movements are monitored) and report their dreams. Scenes requiring a determining control of gaze are selected. This approach requires a relatively rapid arousal and rapid recollection of dream mentation, followed by the retrograde assessment of eye movement–dream imagery correspondence, i.e. the temporal alignment of connected events in the dream narrative and the eye movement recordings. A demonstrative example is a sleeper looking up and down during REM sleep, followed by his report that he dreamed of climbing up a series of ladders looking up and down as he climbed (Dement and Kleitman, 1957). Using this technique, some authors found a 70–80% correspondence (Dement and Wolpert, 1958; Roffwarg et al., 1962; Herman et al., 1984), while others have found only a 9–32% correlation (Moskowitz and Berger, 1969; Jacobs et al., 1972). Determining the correspondence between eye movements and dream imagery is challenging due to the use of varying and flawed methodologies, as well as amnesia and a lack of clarity in dream recall. Furthermore, in the awake state, the eyes and head work concertedly to produce gaze (Herman et al., 1984). Only with the summation of head and eye activity does an isomorphism between gaze and target become apparent. In normal REM sleep, atonia spares the extraocular muscles but not the neck muscle so that the head cannot move, rendering the parallel between observed eye movements and the subject’s description of gaze (in the dream) uncertain.

One way to circumscribe these methodological problems (recall bias, retrograde assessment, neck–eye movement combination) in humans is to study subjects with REM sleep behaviour disorder (RBD). This recently described parasomnia is characterized by dream enactment during REM sleep (Schenck et al., 1986), when patients yell, kick, slap or catch invisible objects, with incomplete atonia. RBD affects subjects without any other disease (idiopathic RBD) or suffering from various neurological and neurodegenerative diseases, mainly synucleinopathies (Boeve et al., 2007). Using video and sleep monitoring (Oudiette et al., 2009), we observed several complex, non-violent dreams enacted during REM sleep, including an ex-smoker mimicking the quiet gesture of smoking a fictive cigarette during REM sleep (Supplementary Video 1). We suspect that as he turns his head rightward and downward to scratch his cigarette in an ashtray, eye and head movements are coordinated towards targeting the fictive cigarette in his hand. We therefore sought to determine how the eye and head movements matched with the suspected dreamt target in a series of patients with RBD.

Materials and methods


From January 2008 to September 2009, 63 patients were referred to the sleep disorders unit for suspected RBD. Patients underwent videopolysomnography. The diagnosis of RBD was confirmed by interview plus videopolysomnography in 56 patients, using international criteria for RBD (American Academy of Sleep Medicine, 2005). Subsets of the patient population had idiopathic RBD (n = 25), idiopathic Parkinson’s disease (n = 14), multiple systemic atrophy (n = 10), dementia with Lewy bodies (n = 5) and idiopathic narcolepsy (n = 2). Medical condition and treatments, in particular antidepressant or benzodiazepine drugs, were recorded (Supplementary Table 1). Seventeen age- and sex-matched healthy controls underwent the same procedure. The demographic and clinical characteristics of patients and controls are shown in Table 2.

Sleep monitoring

All patients and controls underwent videopolysomnography for one (n = 35/56 patients and 17 controls) or two (n = 21/56 patients) consecutive nights. The monitoring included Fp1-A2, C3-A2, C3-O1 electroencephalography (EOG) (see below), levator menti and tibialis anterior muscle surface electromyography, nasal pressure through nasal prongs, recording of tracheal sounds through a microphone placed at the surface of the trachea, thoracic and abdominal belts with gauge strains to assess respiratory efforts, electrocardiography, pulse oximetry, EEG-synchronized infrared video-monitoring and ambiance microphone recordings (Brainet®, Medatec Ltd, France). The patients were interviewed the next morning with regard to dream content. Sleep stages (in particular REM sleep without atonia), arousals, alpha rhythm on EEG, respiratory events and periodic leg movements were scored through visual inspection according to previously reported standard criteria and definitions (De Cock et al., 2007).

Rapid eye movement monitoring

Horizontal and vertical eye movements were recorded by EOG using four Ag–AgCl surface electrodes, two at the outer canthi of the eyes for horizontal movements and two placed periorbitally (at the superior and inferior parts of the middle of a single eyeball) for vertical movements, with an A2 (right mastoid) reference. The signal was acquired through an AC amplifier, filtered with low (0.16 Hz) and high (16 Hz) filters, sampled at a rate of 200 Hz and amplified at 200 µV/mm (Brainet®, Medatec Ltd, France). The saccades were calibrated during wakefulness and daylight, using the EOG channels monitoring the four eye movement directions. The subjects were asked to fixate on a target displaying crosses placed at the centre and at left, right, up and down locations. They had to look from the centre to each cross to obtain saccade amplitudes of 5° and 26°. During REM sleep, the eye movements were selected if they had a duration shorter than 500 ms and an amplitude greater than 3°. REMs density and REMs index were measured in the first 17 patients taking part prospectively in the study who were not being treated with antidepressants, and 17 controls. In addition, REMs density was measured in 14 patients taking antidepressants. To measure REMs density, REM sleep was divided in 3 s mini-epochs. These mini-epochs were classified as with and without REMs (Frauscher et al., 2009). The REMs density was the ratio between mini-epochs with REMs and the total number of mini-epochs during REM sleep. The REMs index corresponded to the number of REMs per minute of REM sleep. In addition, the REMs were classified according to their pattern as isolated (only one REM), clustered (two to four REMs separated by <1 s without REM) or burst (equal or more than five REMs separated by <1 s), as classified in cats (Jeannerod et al., 1965).

Correspondence between rapid eye movements and behaviour during REM sleep behaviour disorder

We scrutinized, second by second, the video obtained during REM sleep in 35 consecutive patients suffering from moderate to severe RBD. When we observed a movement or a vocalization (regardless of type, amplitude and duration), we put a tag at the start and the end of the event to determine the total time spent in visible movements and vocalizations. Each behaviour was classified according to its complexity as previously described (Frauscher et al., 2007). The behaviours were considered as minor when they were myoclonic or too simple to be noticed by the bed partner (e.g. finger twitch). They were major when their amplitude was large or when more than one limb was moving (e.g. large arm movement or whole-body jerk). Complex behaviours were apparent ‘acting out’ of the dream, but not clear enough for researchers to determine the exact meaning of the behaviour (e.g. multiple, disordered gestures of the superior limbs). Scenic behaviours were apparent ‘acting out’ of the dream and easily understood by the observer (e.g. mimicking the gesture of smoking a cigarette). Each 3 s mini-epoch during REM sleep was scored as follows: (i) without REMs or behaviours; (ii) with REMs but without associated behaviours; (iii) without REMs but with behaviours; or (iv) with REMs and behaviours. For the last category, we determined whether the REMs associated with the observed behaviour were isolated, grouped or in bursts, and if they occurred before, during or after the observed behaviour.

Visual scanning of the dream scene during REM sleep behaviour disorder

From data collected for 56 patients (and 77 nights), we selected the scenic behaviours that seemed intentional and would require gaze control if they had been performed during wakefulness (e.g. to grab something on a fictive high shelf) from the RBD sequences. During these intentional behaviours, the patient’s attention, especially the head and limb movements, was directed towards the action. After carefully watching 5544 min of video obtained during REM sleep, we found only 19 apparently goal-oriented behaviours, in nine patients. We analysed these 19 RBD behaviours, image by image (24 images/s). Each behaviour was split into individual movements of the head and limbs, including movements of directed attention (grabbing a fictive object upward; one expects the eyes to move upwards before the head does) and head movements for adjustment and target refocusing (head and trunk bending to the left when the subject makes efforts to pull a rope on his right side; one expects the eyes to move rightwards when the head moves leftwards, to maintain the gaze on the target). In addition, to select only clear-cut, goal-oriented behaviours, we used three levels of security to measure the concordance between eye movements and behaviour. First, the observed eye movements were compared to the expected eye movements. The concordance between the eye movements and the direction of the head and limb was analysed in two spatial planes (vertical and horizontal, but not forward/backward) and in each direction of the plane (right/left and up/down) with reference to the subject (Table 1). The RBD-associated REMs were either concordant (same plane, same direction), in the opposite direction (same plane, but left instead of right), or in the opposite plane (vertical instead of horizontal). Second, to predict more finely the direction and amplitude of the eye movement, sequences normally associated with these behaviours during wakefulness and taking into account the adjustment of REMs when the head turns, a healthy awake control reproduced these behaviours while monitored, first with eyes open, then with eyes closed. The control performed several trials, among which we selected the best one, i.e. performed with eyes closed, and containing an artefact-free EOG. We compared the REMs observed in patients with the awake-predicted eye movements. ‘Unpredicted REMs’, whatever their direction, were not expected to contribute to waking imitation of the behaviour. Similarly, ‘missing REMs’ were present when awake controls mimicked the RBD and absent during the RBD behaviour in patients. Third, we restricted the analysis for ‘determinant’ REMs as ‘compulsory for realizing a key action in the dream’ (e.g. an upward REM before raising the arm to catch a high object). All predictions were performed prior to analysis to avoid any bias. Examples of wake-predicted REMs and RBD-associated REMs with similar behaviour are shown in Fig. 1.

View this table:
Table 1

Definitions of the different REMs

All types of REMsREMs corresponding to directed attention and to target refocusing
Determinant REMsThese REMs correspond to directed attention and are considered as compulsory for the key action
Observed REMs (during RBD)
    Concordant directionThese REMs are in the same direction and plane as expected when looking at the direction of the behaviour
    Opposite directionThese REMs are in the opposite direction but in the same plane as expected (e.g. left versus right)
    Opposite planeThese REMs are in the plane opposite to expected (e.g. vertical versus horizontal)
    Unpredicted REMsThese eye movements are not observed when reproducing the goal-oriented behaviour when awake, but REMs are present during the goal-oriented behaviour in REM sleep.
    Missing REMsThese eye movements are observed when reproducing the goal-oriented behaviour when awake, but REMs are absent during the goal-oriented behaviour in REM sleep.
Figure 1

Comparison of eye movements between control awake eyes that are closed (upper two channels) and patients in REM sleep (lower two channels). In this example, the patient caught a fictive object on a shelf, and the control reproduces the same scene. EOG (in degrees) is recorded between the left external canthus (LOC) and the right external canthus (ROC) to monitor horizontal movements, and between the left supra-orbital (SOC) and left infra-orbital (IOC) positions to monitor vertical movements. The first vertical line corresponds to an upward head movement; the subjects raised their left arms after the second vertical line. The control subject looks upward four times before he raises his left arm and then looks upward once more after having raised his left arm and head. The patient looks upward twice before raising his left arm and looks upward once after having raised his head. In the patient, there is an unpredicted REM directed towards the left (on horizontal EOG).

Statistical analysis

As most measures were not normally distributed, demographics and sleep parameters of patients and controls were compared using a Mann–Whitney non-parametric two-paired test. The various REMs indexes were compared between patients and control groups using a Mann–Whitney non-parametric one-paired test. The correlation between REMs and limb movements was tested with the Spearman rank correlation. The coincidence of mini-epochs with REMs and the abnormal behaviour during RBD was examined by the Wilcoxon paired-sample test. The type of associated REMs and their time association with the abnormal behaviours, as well as the concordance between the wake-predicted REMs and RBD-associated REMs, were analysed by a Kruskal–Wallis test with Dunns post hoc comparison. The statistical software was GraphPad Prism 5 (GraphPad software, La Jolla, CA, USA). Results are reported as the mean (SD).


Sleep measures

The 17 controls and 56 patients with RBD were similar in age, sex ratio and body mass index (Table 2). The patients with RBD had a slower occipital alpha waking rhythm than controls, and 30.4% of them (and no controls) had a pathological alpha waking rhythm (<8 Hz). RBD patients exhibited shorter total sleep time, reduced sleep efficiency, longer REM sleep latency and higher N1 percentage than the controls. The REM sleep percentages and the various fragmentation indexes (arousal, respiratory event and periodic leg movements) were similar in both groups.

View this table:
Table 2

Demographics, clinical and sleep parameters in patients with RBD and controls

Number of patients5617
Age (years)64.7 (8.2)62.2 (7.1)0.11
Sex (% male)77820.64
Body mass index (kg/m2)26.1 (4.2)26.4 (3.7)0.78
Occipital alpha waking rhythm (Hz)8.7 (1.5)9.9 (1.3)0.006
Night-time sleep
    Total sleep time (min)392 (89)443 (66)0.023
    Sleep efficiency (%)76.5 (12.1)85.3 (5.9)0.009
Latency to (min)
    Sleep onset36 (29)26 (18)0.27
    REM sleep157 (119)82 (38)0.009
Sleep duration (% total sleep time)
    Stage N17.7 (6)4.6 (2.8)0.05
    Stage N249.5 (11.9)54.3 (7.9)0.10
    Stage N324.1 (9.1)21.5 (9.1)0.25
    REM sleep17.9 (6.7)19.5 (4.4)0.39
Sleep fragmentation (number/hour)
    Arousals21 (11)17 (9)0.12
    Periodic leg movements32 (38)16 (29)0.10
    Apnoea–hypopnoea11 (14)7 (6)0.76
  • Data are mean (SD). When a patient underwent two nights, only the first night was included.

REMs in REM sleep and coincident behaviour

REMs density, index and complexity were similar in 17 patients with RBD (and not on antidepressants) and in 17 controls (Table 2). REMs density did not differ between 17 patients who were not taking antidepressants (28.8 ± 12.6%) and 14 patients who were taking antidepressants (33.5 ± 14.4%; P = 0.16). REM density and index did not differ between patients with idiopathic RBD and patients with RBD associated with a synucleinopathy (idiopathic Parkinson’s disease, multiple systemic atrophy or dementia with Lewy bodies), as shown in Supplementary Table 2. There was a weak (R = 0.344) but significant (P = 0.043) correlation between REM density and the percentage of movements during REM sleep in the 35 patients with RBD (Fig. 2).

Figure 2

Correlation (R = 0.34, P = 0.04) between the percentage of REM sleep with movements (x-axis) and the density of REMs (y-axis). Each point represents a patient with RBD.

In 35 consecutive patients with RBD, the total duration of abnormal behaviours during REM sleep was 6.1 ± 5.8 min per night, corresponding to 9.2 ± 7.3% of REM sleep time devoted to movement. Sixty-six percent of these behaviours were minor, 18.8% were major, 13.5% were complex and 1.8% were scenic. In these patients, the mini-epochs without REMs were 2.2 times more frequent than mini-epochs with REMs (Table 3). There were as many body movements with REMs as without REMs during REM sleep. The relative occurrence of body movements was 2.1 times greater in REM sleep with REMs compared to REM sleep without REMs. This was the case for all categories (minor, major, complex or scenic) of behaviours (Table 4). When behaviours were associated with REMs, the eye movements were isolated (48.1 ± 13.8%), clustered (37.3 ± 10.3%) or bursting (14.6 ± 12.6%, Dunns post hoc P < 0.05, F = 62.44). As for the temporal link between REMs and behaviours, 43.5 ± 8.2% of REMs were concomitant with behaviours. We observed that 35.2 ± 8% of REMs preceded the behaviours, and 21.4 ± 5.7% of REMs followed the behaviours (Dunns post hoc, P < 0.05, F = 69.92). The coincidence between all types of abnormal behaviour and REMs was also found to be significant in patients with idiopathic RBD and in patients with synucleinopathy-associated RBD (Supplementary Table 3).

View this table:
Table 3

REM density and index in patients with RBD and controls

REM density (%)28.8 (12.6)26.8 (7.5)0.46
REM index (number/min)
    Total10.7 (7)9.7 (3.7)0.46
    Isolated2.5 (0.9)2.5 (0.6)0.47
    Clustered4.2 (1.9)3.8 (1.3)0.41
    In bursts4 (5.2)3.4 (2.4)0.34
REMs complexity ratio (%)
    Clustered + burst/total28.3 (16.7)32.2 (11.9)0.17
    Burst/total71.8 (9.8)72 (8.7)0.41
  • Data are mean (SD).

Eye and limb movement concordance during RBD

Among 56 patients with RBD and 77 nights, there were only 19 apparently goal-oriented behaviours in nine patients. Patient 1 was attacked by lions and glanced at his aggressor while escaping. Patient 2 (i) ordered an employee on his right-hand side; (ii) pulled a fictive object down; (iii) waved to someone called ‘Annie’ on his right side; and (iv) pulled an imaginary object up then put it on his right side. Patient 3 (i) called someone on his right side; (ii) strangled an attacker just below him while shouting ‘I will break your head against this wall’; and (iii) defied someone on his left. Patient 4 (i) was attacked on his left then on his right side; and (ii) started a fight on his left side. Patient 5 (i) grabbed an object below him; and (ii) grabbed something on his far left side. Patient 6 talked to someone on his right side. Patient 7 smiled and kissed someone standing on her right side. Patient 8 kicked and defied an aggressor on his left side. Patient 9 (i) grabbed an object upward then climbed a ladder; (ii) grabbed something below him; (iii) pushed someone on his left-hand side; and (iv) grabbed an object on his left side. The mean duration of the goal-oriented behaviours was 14.2 ± 7.1 s. After the healthy controls mimicked the 19 goal-oriented behaviours, we observed 163 eye movements during wakefulness (named ‘predicted REMs’), corresponding to 8.6 REMs per motor sequence. When analysing the REMs of patients during RBD, we found 162 REMs during the goal-oriented behaviours. REMs were observed in 58.3% of the cases for which we had predicted eye movements, and were absent in 41.7% of such cases (missing REMs). Nonetheless, when there were eye movements, their plane (horizontal versus vertical) and direction (up versus down, left versus right) were concordant with the direction of the RBD-associated motor behaviour in 82% of cases (Table 5 ). An example of REMs and synchronous goal-oriented behaviour is shown in Fig. 3 and Supplementary Video 2. Concordant REMs were more frequent than REMs in an opposite direction and REMs in an opposite plane, and as frequent as unpredicted REMs. As for determinant REMs, they were predicted in 16/19 sequences and recorded during sleep behaviour in 10/19 sequences (62% of sequences). When present, they were concordant with the direction of the motor behaviour in 90% of cases. Concordant, determinant REMs occurred more often than REMs in an opposite direction, REMs in an opposite plane, and were as frequent as unpredicted REMs.

View this table:
Table 5

Concordance of the direction of REMs with apparently goal-oriented behaviours in RBD

Observed REMs
Concordant directionOpposite directionOpposite planeUnpredictedPF
All types of REMs
    Total number788967n.a.n.a.
    Percentage of all observed REMs82%8.4%9.5%n.a.n.a.
    Number/motor sequence4.1 (4)0.4 (0.6)0.5 (0.9)3.5 (4)<0.00125.6
    REMs/total/patient52.3 (31.2)7 (13.3)5.9 (13.3)34.8 (28.8)<0.00130.6
    Percentage of observed REMs/patient79.4 (28.3)10.2 (17.3)10.4 (25)<0.00121.7
Determinant REMs
    Total number18203n.a.n.a.
    Percentage of all observed REMs90%10%0%n.a.n.a.
    Number/motor sequence1.1 (1.4)0.1 (0.3)00.2 (0.4)<0.00117.5
    REMs/total78.5 (32.1)12 (31.6)09.5 (17.1)<0.00122.5
    REMs/total/patient88 (31.6)12 (31.6)0<0.00116
  • P for comparison between the directions (concordant versus opposite plane and direction), using Kruskal–Wallis test, with Dunns post hoc. Data are mean (SD).

  • n.a.: not applicable.

View this table:
Table 4

Coincidence between REMs (phasic REM sleep) and motor behaviours during REM sleep in 35 patients with RBD behaviour

Mini-epochs with REMsMini-epochs without REMsP
Mini epochs of 3 s (n)444 (280)984 (539)<0.001
With motor behaviour (n)96 (106)96 (80)0.52
With motor behaviour (%)23.2 (17)11.3 (9.6)<0.001
Motor behaviour (n, total)61 (64)31 (29)<0.001
    Minor38 (41)23 (21)<0.001
    Major13 (13)5 (7)<0.001
    Complex10 (13)3 (5)<0.001
    Scenic1 (3)0.3 (1)<0.001
  • REMs = rapid eye movements during REM sleep.

Figure 3

Examples of REM (lower panel) and synchronous RBD-behaviour (upper panel) when the patient dreamed that he climbed on a ladder. The low part of the figure represents a 20 s epoch of REM sleep with RBD. The two EOG channels monitor in degree the horizontal movements between LOC and ROC and the vertical movements between SOC and IOC. C3-A2 is the EEG channel, EMG channels monitor the muscle activity of the right (EMG4) and left (EMG5) extensor carpi. Each vertical line on the figure corresponds to the frame of video shown on the higher part of the figure (for full video, see Supplementary Video 2). (1) The patient is resting, a leftward and upward gaze begins; (2) the head is directed towards the top and left part of the space, and the left arm is raised, while REMs have just ended; (3) the head is still directed towards the top and left, the left arm is up, and the REMs begin to move rightwards and downwards; (4) the head is more directed to the bottom and the right, and the REMs are over; (5) the patient raised his left leg so as to climb on a ladder, while the REMs start to move upwards; (6) the head is more directed upwards (the patient trying to climb higher).


Among REMs that accompanied goal-oriented motor behaviour during RBD, 82% were directed towards the action of the patient. When restricted to determinant REMs, the level of concordance increased to 90%. These results indicate that the limbs, hands, head and eyes move together in a coherent, directional manner during RBD. The final gaze is placed in the direction and plane of the fictive target. The main addition of this study to preceding experiments on the scanning hypothesis (based on dream-recall and a posteriori correlation) is the presence of an external, dreamer-independent, ‘on line’ check of the dream scenario. This is key to avoid the numerous biases associated with the a posteriori recall method (total or partial amnesia, censure, reconstruction of a scenario, clarity of recall, simplicity of the narrative, intervention of an external investigator). We chose to restrict our study mostly to grabbing behaviours or salutations, i.e. simple, purposeful, goal-oriented behaviours in which fine coordination between the gaze and hand movements is obvious. The behaviours (including various types of movements, from mild jerk of an arm to scenic, nearly subtitled, complex behaviours) occurred twice as frequently during REM sleep with REMs compared to sleep without REMs, again supporting an association between the phasic generators of REMs and body movements. There was also a weak but significant correlation between the density of eye movement and the percentage of body movements that occurred during REM sleep. This association of body and eye movements in 35 patients with RBD is concordant with two previous studies in 8 and 12 patients but differs with respect to the type of behaviour associated with REMs. In eight patients, REMs were synchronous with major and complex but not minor behaviours (Frauscher et al., 2009). In 12 patients, REMs were also related to primitive, purposeful and semi-purposeful movements (Manni et al., 2009). In line with previous studies, the more complex the behaviour, the more frequently it was associated with REMs. Scenic, complex and major behaviours were 2.6–4.7 times more frequent during REM sleep with REMs compared with REM sleep without REMs. In contrast, minor behaviours were 1.6 times more frequent in REM sleep with REMs compared to REM sleep without REMs. In line with these findings, healthy volunteers report more active dreams when awakened during bursts of REMs (Berger and Oswald, 1962).

However, 78.3% of REMs were not associated with body movements in the patients with RBD. The dreamers may be passively ‘watching’ moving pictures in this part of their dreams without taking part in the action. It will be necessary to collect the accompanying dreams to support this hypothesis, which has not yet been systematically tested. Alternatively, the system that blocks motor atonia may be partially restored at these periods. If this is true, we should collect data on active dreams, with fictive movements of the dreamer, but no movement in reality. This hypothesis (some active dreams are transiently blocked by the REM sleep atonia) is supported by the observation that only 9% of REM sleep time was associated with visible behaviours in these patients with RBD, suggesting that we only accessed a small (but interesting) part of the ‘iceberg’ of dream content (assuming, as it was partially demonstrated by Dement and Wolpert in 1958, that dreaming take place throughout REM sleep). Conversely, there were body movements without REMs. Control of vision may not be necessary in this part of the inner scenario. Eventually, the dreamer may be watching something at a distance or just fixedly staring at some object, as Dement and Kleitman (1957) showed by awakening subjects after periods of ocular quiescence during REM sleep.

Do these results mean that the dreamer actually watches the dream images in RBD? This cannot be known for certain. The eyes are closed, and EOG is an indirect measure of one eye’s position, such that we do not know if the movements of the left and right eyes converge to a fixation point, or whether certain final REMs adjust the image within the fovea, as would occur when a patient who is awake watches a scene. Alternatively, one common system may simultaneously activate dream images as well as eye and body movements in a coherent fashion. This scenario would support the results from several experiments, including the presence of REMs in the absence of any kind of vision (in neonates, congenitally-blind humans, cats without visual cortex, pontine cats), as well as the temporal association between ponto-geniculo-occipital (PGO) spikes and REMs in cats. In animals, PGO spikes are generated in the pons (caudoventral pontine tegmentum) simultaneously with REMs but not before (Vanni-Mercier and Debilly, 1998). In addition, some PGO spikes are not associated with REMs, while REMs are always associated with PGO spikes. PGO spikes have been proposed to generate dream images and other hallucinatory aspects of dreams because they project to visual, parietal and temporal sensory brain areas (Jouvet, 1967). During RBD in cats with brainstem lesions, the motor sequences begin with a unilateral PGO spike coupled to a homolateral single REM, as if the cat were watching an invisible target as he moves his head and eye laterally (Sastre and Jouvet, 1979). Later, REMs and PGO spikes occur in bursts as the cat develops more complex behaviours. Furthermore, isolated high-amplitude REMs are related to orienting behaviour in cats, while REMs occur in bursts during generalized movements (jumping, attacks or phasic unorganized movements) and are absent during continuous slow body movements such as licking (Soh et al., 1992). Recently, event-related functional MRI analysis time-locked to the occurrence of REMs revealed that the pontine tegmentum, ventroposterior thalamus, primary visual cortex, putamen and limbic areas were activated in association with REMs, providing evidence for the presence of PGO spikes in humans (Miyaushi et al., 2009).

One surprising phenomenon was the presence of ‘missing REMs’. In 41.7% of goal-oriented behaviours, REMs were absent when they were expected. Furthermore, when patients displayed major or complex behaviours, the associated REMs were mostly isolated—not grouped or bursting as one would expect. These results suggest that the saccades during REM sleep differ from the saccades during wakefulness. In humans, eye saccades have a similar velocity during REM sleep and wakefulness in the dark, especially when imagining a visual scene (Herman et al., 1983; Sprenger et al., 2010). Yet some REM characteristics (round-shouldered appearance, looped trajectories, slow, oblique and torsional patterns) render these eye movements unlike any of those occurring during wakefulness (Fuchs and Ron, 1968; Jacobs et al., 1971). We suspect that the REMs observed during RBD are rough movements. We previously found that the limb movements during RBD in patients with Parkinson’s disease are different from their limb movements during the waking state (De Cock et al., 2007). These movements do not exhibit characteristics of parkinsonism (no tremor, bradykinesia or dystonia) and are on the contrary fast but jerky and rough, suggesting they are unfiltered by the basal ganglia. Motor-related brain activity before REMs during REM sleep is thought to differ from the mechanism that controls eye movement during wakefulness (Abe et al., 2008). This phenomenon may be responsible for the missing REMs. The saccades are controlled during waking by frontal association areas (frontal eye field, supplementary eye field and dorsolateral prefrontal cortex), the precuneus prefrontal cortex, the supplementary motor area, posterior cingulate cortex and the posterior part of the superior parietal cortex (Johnston and Everling, 2008). Notably, the dorsolateral prefrontal cortex, which is hyperperfused during waking saccades in the dark, is hypoperfused during REMs (Peigneux et al., 2001). These findings support the relative hypo-activity of associative prefrontal areas during REM sleep (Maquet et al., 1996). In a magneto-encephalography study, the frontal eye fields and midpontine nuclei were activated during both waking and REM sleep saccades, but with differences in timing, amplitude, time scale and sequence of activation. These results suggest that, during REMs, the midpontine nuclei drive eye movement and the frontal eye fields receive feedback from this activation (Ioannides et al., 2004). During REMs, compared to closed eyes during waking, the activity level is higher in the supplementary motor area and lower in the inferior parietal and precuneus cortex. These series of experiments suggest that some saccades would be missing during hand–eye coordinated movements during REM sleep because some brain areas that control saccades would be deactivated by the general REM sleep process. As an example, the vestibulo-ocular reflex is abolished during normal REM sleep in cats (Flandrin et al., 1979). Since PGO spikes are not always associated with eye movements, one may imagine that the limb motor sequence is not always associated with eye movements.

There are several limitations in this work. First, goal-oriented behaviours, where visual control is imperative, were rare in our series. We observed only 19 goal-oriented behaviours, including 163 REMs, in 9/56 patients, 77 nights and 5544 min of REM sleep. In addition, only 9% of REM sleep included abnormal behaviours in patients with RBD, including two-thirds of minor movements. Scenic RBD is however rare in the context of the sleep laboratory (accounting here for only 0.16% of total behaviours). The chance of observing scenic RBD in patients with severe RBD may increase with increased numbers of monitored nights. We doubled the number of observation nights in 21 patients with severe RBD. Implementing the series with patients suffering from idiopathic RBD, who have more frequent and severe behaviours (Iranzo et al., 2005), could help in this regard. Nonetheless, patients with RBD represented half of the sample population studied here. Notably, the patients in our series were representative of other patients with RBD: predominantly male, mean age of 64 years, with a slowed occipital alpha rhythm (Schenck and Mahowald, 2002; Gagnon et al., 2004). As a second limitation, the exact dream mentation during REMs was not known for most patients, except when they were spontaneously awakened by their behaviours (e.g. the patient chased by lions, who escaped by leaping out of the bed and woke up as he placed his hand on the floor). The cross-correlation between REMs, observed behaviour and dream mentation would require researchers to wake the patients up just after the behaviour, which is difficult for various reasons. The behaviours are unpredictable, sudden and brief (the investigator should decide very quickly when to enter the room and wake the patient). REM sleep without atonia is difficult to score as such ‘on-line’. Violent movements affect EOG monitoring, which led us to reject some motor sequences from the analysis. Waking up patients would prevent further REM sleep. Finally, dream recall has many biases, including patients who do not recall any dreams even after having displayed scenic motor sequences (Scaglione et al., 2005). The major strength of the RBD model is that, in cases with scenic behaviours, the dream report from the dreamer is not necessary, as the behaviour is so demonstrative (especially when speeches accompany the behaviour) that any observer having the same cultural background will be able to determine its nature: smoking, kissing, gesturing ‘thumbs up’, clapping, inspecting the army, giving a lesson (Oudiette et al., 2009). However, predicting the exact REM direction, based on the observed behaviour during RBD, is sometimes difficult. We used three different levels of confidence for these predictions. First, we restricted this analysis to simple, goal-oriented behaviours. Second, we made a sub-analysis restricted to determinant REMs, with the hypothesis that there was little risk, for example, that a human makes a hand salutation gesture or a kiss without looking in the direction he sends his gesture. We surely missed the part of the inner scenario that was not translated into a given behaviour (hence possibly explaining the supplementary REMs), but not the behaviour-associated determinant REMs. Third, an awake control imitated the behaviour, and his REM direction was compared to those of the dreamer. This imitation is of course partial, as nobody awake can have in mind a scenario identical to that of the dreamer. Therefore, REMs should be less (but not more) numerous during awake asleep than behaviours. These three levels of analysis resulted in the same conclusion, i.e. that REMs, when present, were in the direction appropriate with the observed behaviour.

Can the results obtained in patients with RBD be extended to the REMs of normal subjects? Here the index and density of REMs were similar in the patients with RBD and in the healthy controls, confirming two previous studies in smaller series (Lapierre and Montplaisir, 1992; Dauvilliers et al., 2007). Even the complexity of REMs (not studied previously) was unchanged. Such similarities suggest that the mechanism causing RBD, which unblocks some movements and behaviours during REM sleep, does not unblock certain additional REMs. Therefore, it is also probable that REMs in normal subjects are directed towards the targeted dream images.

In conclusion, our results support a tough link between the dream action (when goal-oriented) during RBD and the direction of the gaze, which can be extended by analogy to normal dreamers in REM sleep. If REMs are always temporally synchronized with mental imagery during REM sleep, they could be a reliable marker of gaze direction during dreams.


Federation pour la Recherche sur le Cerveau (AO-2007, laboratory grant to I.A.); Fondation pour la Recherche Medicale (AO-2008, student grant to L.L.V).

Supplementary material

Supplementary material is available at Brain online.


  • *These authors contributed equally to this work.

  • Abbreviations:
    rapid eye movement
    REM sleep behaviour disorder


View Abstract