ARTICLE IN PRESS

BRES-35418; No. of pages: 9; 4C:

BR AIN RE S EA RCH XX ( 2 0 06 ) XXX –X XX

a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m

w w w. e l s e v i e r. c o m / l o c a t e / b r a i n r e s

Research Report

Implicit oculomotor sequence learning in humans: Time course of offline processing Geneviève Albouy a , Perrine Ruby a , Christophe Phillips a , André Luxen a , Philippe Peigneux a,b , Pierre Maquet a,c,⁎ a

Cyclotron Research Centre (B30) University of Liège – Sart Tilman 4000 Liège Belgium Cognitive Science Department, University of Liège – Belgium c Department of Neurology, CHU Sart Tilman – Belgium b

A R T I C LE I N FO

AB S T R A C T

Article history:

Studies of manual and digital sequence learning indicate that motor memories continue to

Accepted 19 March 2006

be processed after training has ended, following a succession of identifiable steps. However, it is not known whether this offline memory processing constitutes a basic feature of motor learning and generalizes to the implicit learning of a sequence of eye movements. To assess

Keywords:

this hypothesis, we have created the serial oculomotor reaction time task (SORT).

Oculomotor sequence learning

Participants were trained to the SORT then tested after either 30 min, 5 h or 24 h. During

Offline memory processing

training, ocular reaction times decreased monotonically over practice of a repeated sequence, then increased when a different sequence was displayed, demonstrating oculomotor learning of the trained sequence. When tested 30 min after training, a significant gain in oculomotor performance was observed irrespective of the sequence learning. This gain was no longer present after 5 h. Remarkably, a gain in performance specific to the learned sequence emerged only 24 h after training. After testing, a generation task confirmed that most subjects learned implicitly the regularities of the sequence. Our results show that, as for manual or digital sequences, oculomotor sequences can be implicitly learned. The offline processing of oculomotor memories follows distinct stages in a way similar to those observed after manual or digital sequence learning. © 2006 Elsevier B.V. All rights reserved.

1.

Introduction

Motor skills primarily improve with practice. However, even after a single training session, motor performance is improved when tested at specific post-training intervals. This finding suggests that the motor memory is further processed offline, i.e. in the absence of any further practice. This process, globally referred to as consolidation, can lead to both enhanced performance and resistance to interference (Robertson et al., 2004a).

Present evidence indicates that memory consolidation for motor sequences involves several functional steps, which appear intrinsically dependent on several factors such as time, sleep, circadian rhythms or subject's awareness of the sequential material. The repetition of a finger sequence explicitly known in advance, as in the finger tapping task (FTT), speeds up monotonically during the initial training session (Fischer et al., 2002; Karni et al., 1998; Walker et al., 2002). Several hours after practice, performance returns to

⁎ Corresponding author. Fax: +32 43 66 29 46. E-mail address: [email protected] (P. Maquet). URL: http://www.ulg.ac.be/crc (P. Maquet). 0006-8993/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.brainres.2006.03.076

ARTICLE IN PRESS 2

BR AIN RE S EA RCH XX ( 2 0 06 ) XXX–X XX

immediate post-training levels if the participant remains awake but is enhanced if sleep intervenes (Fischer et al., 2002; Korman et al., 2003; Walker et al., 2002). A similar time course in performance is reported using a serial reaction time (SRT) task, if the sequence is explicitly known by the participants (Robertson et al., 2004b). In contrast, implicit sequence learning leads to a significant improvement after 12 h of wakefulness, with no further gain overnight (Robertson et al., 2004b). Finally, sequence learning was shown to be modulated by the circadian rhythm, through its interaction with sleep, especially REM sleep (Cajochen et al., 2004). All these experiments on motor sequence learning examined the acquisition of skilled movements of the fingers or of the hand. This situation leaves open the possibility that the reported time course of motor memory consolidation is specific to digital/manual tasks. Alternatively, sequence learning is a general human ability which might be acquired irrespective of the effector. This perspective is consistent with the hypothesis that sequence learning in serial reaction time tasks relies more on the spatial than perceptual or motoric aspects of the task (Willingham, 1999). Saccades are rapid eye movements that allow an individual to bring and maintain an object image onto the fovea. Saccades seldom occur in isolation. They rather take place in sequences. During visual exploration of the environment, the gaze trajectory depends on the visual scene. In humans, exploring a familiar face primarily focuses on internal facial features (eyes, nose, mouth), whereas the external facial features (hair, face shape) are mainly processed for unfamiliar faces (Bruce and Young, 1998). Sequences of saccades also support planning of hand movements by marking, in an anticipatory manner, landmarks on the object to grasp or on the target to reach (Johansson et al., 2001; Miyashita et al., 1996). Sequences of eye movements are also generated during mental imagery, when subjects mentally repeat a learned sensori-motor sequence encoded during observation (Brandt and Stark, 1997). Despite this evidence, some early studies reported that, in contrast to manual sequences, oculomotor sequences do not benefit from practice, suggesting that an oculomotor sequence could not be learned (Epelboim et al., 1995). Nevertheless, subsequent reports infirmed this observation. In the framework of neuroimaging studies, behavioural measures showed a decrease in ocular reaction time during oculomotor sequence learning (Kawashima et al., 1998) as well as shorter reaction times during the execution of overlearned oculomotor sequences rather than of new ones (Grosbras et al., 2001). However, these studies never investigated the effect of different delays between training and testing and only used explicitly learned, oculomotor sequences. Manual or digital serial reaction time tasks usually imply eye movements when stimuli are spatially distributed. When participants are asked to simply watch stimuli appearing in a repeating sequence, they showed a greater gain in explicit knowledge compared to motor performance (Howard et al., 1992). To the best of our knowledge, implicit oculomotor sequence learning has not yet been reported. In addition, none of the previous reports characterized the time course of performance during the first 24 post-training hours. At present, it is not yet known whether oculomotor memory is followed by offline processing as reported after manual sequence learning.

In this paper, we hypothesized that sequence of saccades can be implicitly learned by healthy human subjects. We also predict that the off-line processing of this motor learning, as assessed by later testing at specific post-training intervals, changes over time, as described for the manual sequence learning tasks. To specify these hypotheses, we examined the development of the specific sequential knowledge, and contrasted it with the mere improvement in basic oculomotor abilities. We developed an oculomotor sequence learning task based on the digital serial reaction time task described by Nissen and Bullemer (1987). In this new task, participants had to visually track a dot which, every 550 ms, adopts one of 4 possible positions on the screen (Fig. 1A). Unbeknownst to the participants, the successive dot positions followed a second-order 8-element sequence. In order to force the participants to fixate the dot at all time, there was a 20% probability at each move that the color of the dot would briefly turn from yellow to pale orange. Participants were instructed to notify this event by a key press. All subjects were trained to a total of 7 blocks and later tested during 5 further blocks (Fig. 1B). In each block the sequence is presented 15 times. In order to allow for a direct measure of sequence learning, another sequence, which did not share any common triplet of successive locations with the learned sequence, was presented 5 times during training (6th block of the training session, B6) and during testing (5 times in the 3rd and 15 times in the 5th blocks of the test session, B10 and B12). The exposure to the new sequence was limited in order to avoid major interferences with the learned sequence. Eye movements were recorded during both sessions, using electro-oculography (EOG). Oculomotor reaction times (oRTs) were measured on EOG recordings as the delay between the change in dot position and the first saccade initiated in the direction of the move (Figs. 1C–D). Different groups of subjects were tested at various intervals after training (30 min, 5 h or 24 h). A generation task, performed at the end of the test session, allowed us to discard any subject demonstrating explicit knowledge of the learned sequence.

2.

Results

Nine subjects were excluded from data analysis. Three subjects persistently stared at the center of the screen, probably catching the dot trajectory at the periphery of their visual field. Three other subjects failed to improve their oculomotor performance during the training session, as oRTs at block 5 did not differ from oRTs at block 1. Finally, the analysis of the generation task showed that 3 subjects (one in each group) acquired an explicit knowledge of the sequence during the SORT practice as they drew more than 3 triplets of successive locations common to the trained sequence. They respectively produced 6, 8 and 6 triplets common to the trained sequence. After exclusion of these subjects, the mean number of generated triplets common to the learned sequence over the 3 groups was 1.31 ± 0.71 (SD) triplets. Eventually, 17 subjects in the 30 min Group (mean age : 24 ± 2 years), 16 subjects in the 5 h Group (mean age : 23 ± 2.5 years), and 16 subjects in the 24 h Group

ARTICLE IN PRESS BR AIN RE S EA RCH XX ( 2 0 06 ) XXX –X XX

3

Fig. 1 – Description of the SORT, of the experimental design and of the data analysis. (A) Trajectory followed by the dot between 4 points. Arrows depict the trajectory of sequence 1 (S1). (B) Experimental design. Training and testing sessions consisted of 7 and 5 blocks, respectively. The unpracticed sequence was proposed to the participants on blocks 6, 10 and 12. (C) Raw EOG horizontal (first row) and vertical (second row) recordings during practice of the SORT. Vertical blue lines indicate dot moves and positions. (D) Automatic detection of oculomotor reaction times (oRTs) computed the delay between the dot move (vertical blue line : trigger) and the first saccade in the right direction (vertical red line : saccade initiation). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

(mean age : 22 ± 2.3 years) were included in the analysis. Mean age did not differ between groups (F(2,46) = 2.29, P = 0.11). The Pittsburg Sleep Quality Index questionnaire (Buysse et al., 1989) revealed that during the preceding month, mean sleep duration [30 min Group, 7 h 32 min ± 59 min; 5 h Group, 7 h 48 min ± 1 h 10 min; 24 h Group, 8 h 7 min ± 52 min; F(2,46) = 1.34, P = 0.26] and median subjective sleep quality [30 min Group, 3, interquartile interval (IQI) = 0.5; 5 h Group, 3, IQI = 0.5; 24 h Group, 3, IQI = 0.00; on a 4-step scale, from poor (1) to good (4); Mann– Whitney U tests] did not differ between groups over the month preceding the experiment (30 min versus 5 h, U = 109, P = 0.49; 30 min versus 24 h, U = 121, P = 0.80; 24 h versus 5 h, U = 114, P = 0.61). In the 24 h group, sleep duration and quality [from very poor (1) to very good (5)] between the two sessions were subjectively assessed using the St. Mary's Hospital sleep questionnaire (Ellis et al., 1981). During the night between training and test session, subjects slept on average 8 h 33 min ± 1 h 11 min and the median subjective sleep quality was 4.

2.1.

Basic properties of recorded eye movements

For each subject, we checked that the relationship between saccade duration and amplitude was linear, confirming that recorded saccades conform to the main sequence (Leigh and Zee, 1999). On average (over all

subjects, blocks and sessions), the mean angular velocity was 240 ± 20°/s.

2.2.

Oculomotor reaction times : training session

A repeated-measure analysis of variance (ANOVA) computed on median oRTs per block with block repetition (B1 to B5) as within-subject factor and test interval (30 min versus 5 h versus 24 h) as between-subject factors revealed a significant oRTs decrease over blocks, F(4,180) = 78.66, P < 0.0001 (Fig. 2). There was no significant group effect (P = 0.25) nor any significant block by group interaction (P = 0.38). In addition, a 2-way ANOVA revealed a main effect of sequence [learned (block B5) versus unlearned (block B6) sequence] on median oRTs, F(1,46) = 163.76, P < 0.0001, but failed to show a significant main effect of the test interval (30 min versus 5 h versus 24 h; P = 0.43) or of the interaction between test interval and sequence type (P = 0.15). Furthermore, it is to note that the subjects' performance on block B7 went back to the level of performance observed on block B5 as indicated by a 2-way ANOVA revealing no significant difference between performance on these two blocks (F (1,46) = 0.65, P = 0.42), no main effect of the test interval (P = 0.18) and no interaction between theses effects (P = 0.42). These data indicate that subjects from the 3 groups similarly improved and learned the regularities of the sequence of ocular movements during the training session.

ARTICLE IN PRESS 4

BR AIN RE S EA RCH XX ( 2 0 06 ) XXX–X XX

Fig. 2 – Oculomotor reaction times (ms) during training and testing sessions for the 3 groups. Black circles: 30-min delay. Red triangles: 5-h delay. Blue squares: 24-h delay. Error bars represent the standard error of the mean. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.3. Oculomotor reaction times : test versus training sessions Average median reaction times for blocks B5, B6, B8 and B10 appear in Table 1, and Figs. 2 and 3. A 3-way ANOVA computed using blocks B5, B6 (training session), B8 and B10 (test session) revealed a significant main effect of session (training versus test; F(1,46) = 12.1, P < 0.005) and of sequence type (learned [B5, B8] versus unlearned [B6, B10]; F(1,46) = 236.7, P < 0.0001), a significant session by group effect (F(2,46) = 4.8, P < 0.05), a significant session by sequence effect (F(1,46) = 6.5, P < 0.05) and a significant session by sequence by group interaction (F(2,46) = 3.6, P < 0.05). These data suggest a significant influence of the delay between training and test sessions on the improvement of oRTs between sessions and/or on the evolution of the differences in oRTs between the learned and the unpracticed sequence.

2.3.1.

Within-group effects

Using planned comparisons, we assessed the changes in oculomotor ability and the development of the (implicit) sequential knowledge, separately within each group. In subjects tested 30 min after training (Group 30 min), median oRTs significantly decreased from training to test

Table 1 – Average (SD) median reaction times (ms) for blocks B5, B6, B8 and B10

30 min Group 5 h Group 24 h Group

B5

B6

B8

B10

−46 ms (87) −29 (84) 2 (69)

124 (43) 127 (31) 118 (25)

−78 (96) −24 (78) −54 (70)

106 (57) 130 (24) 113 (37)

Negative values denote anticipation (see text).

Fig. 3 – Changes in oRTs (ms) between training and testing sessions (respectively left- and right-hand bar of each pair), for the learned (white bars) and unpracticed (gray bars) sequence, for the 3 delay groups. Error bars represent the standard error of the mean.

session for the learned sequence (32 ms; B5 versus B8; F = 6.5, P < 0.05). The oRTs also improved for the untrained sequence, although this gain did not reach significance (18 ms; B6 versus B10; F = 3.7, P = 0.059). The interaction between sequence type [trained versus untrained] and session [training versus test] was not significant (F = 0.87, P = 0.35). These results show that the gain in oculomotor performance during the 30-min posttraining interval similarly benefits to the trained and (although to a lesser extent) to the untrained sequence. In contrast, in subjects tested 5 h after the end of training (Group 5 h), no significant between-session difference in oRTs was observed, neither for the trained (5 ms increase in oRTs; F = 0.16, P = 0.68) nor for the untrained (3 ms increase in oRTs; F = 0.13, P = 0.7) sequence. Likewise, session by sequence interaction was not significant (F = 0.01, P = 0.91). These results indicate that the oculomotor performance measured 5 h after training is not improved as compared to levels reached after the end of the training session. Finally, in subjects tested 24 h after the end of training, oRTs were significantly decreased for the trained sequence (56 ms; B5 versus B8; F = 20.25, P < 0.0001) but not for the unpracticed sequence (5 ms; B6 versus B10, F = 0.29, P = 0.59). This difference was confirmed by a significant interaction between sequence type [trained versus untrained] and session [training versus test]; F = 12.7, P < 0.001. These results suggest that the 24-h interval was beneficial to oculomotor performance to the learned, but not to the unpracticed sequence.

ARTICLE IN PRESS BR AIN RE S EA RCH XX ( 2 0 06 ) XXX –X XX

2.3.2.

Between-group effects

Using planned comparisons, we compared the changes in oculomotor performance and the development of (implicit) sequential knowledge across experimental groups. For the trained sequence, improvement in median oRTs from training to test sessions (B5 versus B8) was significantly higher after a 30-min than after a 5-h interval (37 ms; F = 4.27, P < 0.05), and significantly higher after a 24-h than after the 5h interval (61 ms; F = 12.01, P < 0.005), but not significantly different after the 24-h than after the 30-min interval (F = 2.11, P = 0.15). For the untrained sequence, median oRTs improvement from training to test sessions (B6 versus B10) was not significantly different after a 30-min than after a 5-h interval (21 ms; F = 2.58, P = 0.11), after a 24-h than after a 5-h interval (8 ms; F = 0.41, P = 0.5), or after a 24-h than after a 30-min interval (F = 0.91, P = 0.34). Finally, we found that difference in oRTs for the trained versus the untrained sequence (thought to reflect sequence knowledge) was enhanced from the training to the test session after the 24-h interval, more than after the 5-h delay (53 ms; F = 6.73, P < 0.05). A trend was detected for the session by sequence interaction showing a larger enhancement in sequence knowledge after the 24-h than after the 30-min interval (37 ms; F = 3.64, P = 0.06). No similar interaction effect was found between the 30-min and the 5-h interval (F = 0.52, P = 0.47).

2.4.

Sequence generation

At debriefing after the end of test session, 26 subjects out of 49 (53%) answered “yes” to the last question “Did you noticed that the dot was following always the same trajectory during the SORT task?” [8/16 (50%) in the 24 h Group, 11/16 (68%) in the 5 h Group, 7/17 (41%) in the 30 min Group]. However, none of them could explicitly recall the trajectory as, on average over upon the 3 groups, they drew only 1.31 (SD 0.71) triplets common to the learned sequence. A one-way ANOVA failed to disclose a significant Group [30 min versus 5 h versus 24 h] effect on the percentage of generated triplets common to the trained sequence (F(2,44) = 1.4, P = 0.25).

2.5.

5

in detail and are only reported for the sake of completeness (Fig. 4). The average percent detection of the color changes rose from 2.29 (SD 4.22)% during the first block to 29.34 (SD 20.86)% during the 11th block. The data show that the detection task was hard enough to challenge subjects' perceptual abilities, despite clear evidence for the development of an improvement in oculomotor abilities and the emergence of an implicit sequential knowledge. For the training session, a repeatedmeasure analysis of variance (ANOVA) computed on the percent correct detection of the change in dot color with block repetition (B1 to B5) as within-subject factor and test interval (30 min versus 5 h versus 24 h) as between-subject factors revealed a significant increase in detection over blocks, F(4,46) = 23.87, P < 0.00001. There was no significant group effect (P = 0.97) nor any significant block by group interaction (P = 0.86). A 2-way ANOVA revealed a main effect of sequence [learned (block B5) versus unlearned (block B6) sequence], (F(1,46) = 16.46, P = 0.0002), the performance decreasing when the unpracticed sequence is proposed to the subjects. There was neither a significant main effect of the test interval (30 min versus 5 h versus 24 h; P = 0.58) nor any interaction between test interval and sequence type (P = 0.61). A 3-way ANOVA computed on blocks B5, B6 (training session), B8 and B10 (test session) assessed the changes in detection from training to testing session. It revealed a significant main effect of session (training versus test; F(1,46) = 13.23, P < 0.0001) and of sequence type (learned [B5, B8] versus unlearned [B6, B10]; F(1,46) = 27.44, P < 0.0001), and a significant session by sequence by group interaction (F(2,46) = 4.00, P = 0.025). There was no significant session by group effect (P = 0.10), nor any significant session by sequence effect (P = 0.94). Using planned comparisons, we compared the changes in detection across

Detection of the changes in dot color

In the present study, the dependent parameter consisted in oRTs. The detection of the changes in dot color was required only to ensure that the participants stared at the target at all times. This was formally assessed on EOG recordings where the timing, amplitude and direction of eye movements were carefully checked (see Experimental procedures). In addition, the detection of color changes does not specifically reflect sequence learning. Although the detection of the changes in dot color is made easier by anticipation (and shortening in reaction times), it also depends on other abilities, such as adaptation of saccade amplitude or shape and color perception, which are also likely to evolve over time due to motor and perceptual learning. For these reasons, the measures concerning the detection of the changes in dot color will not be discussed

Fig. 4 – Percent detection of the changes in dot color (%) during training and testing sessions for the 3 groups. Black circles: 30-min delay. Red triangles: 5-h delay. Blue squares: 24-h delay. Error bars represent the standard error of the mean. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

ARTICLE IN PRESS 6

BR AIN RE S EA RCH XX ( 2 0 06 ) XXX–X XX

sessions and experimental groups. For the trained sequence, improvement in detection from training to test sessions (B5 versus B8) was significantly larger only after a 30-min as compared to a 5-h interval, F(1,46) = 5.26, P = 0.026. For the untrained sequence, percent detection from training to test sessions (B6 versus B10) was significantly larger after a 24h than after a 30-min (F(1,46) = 6.39, P = 0.015) or a 5h interval (F(1,4) = 4.6, P = 0.037). Finally, we found that differences in detection for the trained versus the untrained sequence blocks were larger from the training to the test session after the 30-min interval than after the 24-h delay (F (1,46) = 7.91, P < 0.007).

3.

Discussion

The aim of the present study was to design a new task to explore oculomotor sequence learning in humans. Our results show that human participants are able to implicitly learn a sequence of saccades. All subjects were submitted to the same amount of training and were tested only once, after variable delays. Furthermore, the difference in performance between the learned and the unpracticed sequence provide evidence of the specific memory-based changes. Consequently, any change in performance between the training and test sessions, occurring without any further practice, can be thought to be related to the specific offline memory processing. During the 24 h following the end of a single training session, performance followed a distinct time course. As compared to the performance achieved at the end of training, performance increases after 30 min but falls back to late training levels 5 h later. The day after, performance is again significantly enhanced. These results show that, as for manual sequence learning, several phases can be identified in the offline processing of oculomotor memories.

3.1.

Implicitness of oculomotor sequence learning

About half of the participants (53%) reported having noticed that ‘the trajectory of the dot followed a given trajectory’. However, we further used a generation task, conducted manually, to test their explicit sequential knowledge. We used a stringent criterion (no more than 3 triplets common to the trained sequence) to identify the subjects showing any evidence for an explicit awareness of the sequence. Only 3 subjects reached the criterion. These 3 subjects were discarded from the analysis. These results suggest that while most subjects detected some regularities in the dot trajectory, very few acquired an explicit knowledge of the underlying deterministic 8-element trained sequence. Even if subjects were informed before the generation task that they have finish the test performing the new sequence, it remains possible that some interferences from the last block could have masked some explicit sequence knowledge. In addition, low rate of awareness of the learned sequence might be due to the “dual task” condition which the subjects were involved in. Dual tasks reduce the propensity for subjects to become aware of the sequence regularity (Keele et al., 2003). This would explain that only 3 participants out of 58

became aware of the 8-element sequence, a significantly smaller rate than the proportion reported by others although they used a longer (12 or 10-element) sequence (Robertson et al., 2004b; Honda et al., 1998).

3.2. Online and offline processing of oculomotor sequence memories An important result of the present study is that the processing of oculomotor memory follows identifiable phases similar to those reported for manual motor sequence learning. During training, ocular reaction times steadily decreased across blocks. The initiation of saccades soon preceded the dot moves, suggesting their (implicit) anticipation by the subjects. The trained and untrained sequences have the same abstract structure although they have a different surface structure. Consequently, the score provides a measure of the performance which is specifically attributable to the knowledge of the underlying sequential structure of the training sequence. However, when a new sequence was presented, oRTs return to naive levels, thereby demonstrating the effects of the exposure to the learned sequence. These results replicate the time course of motor learning classically reported in manual serial reaction time tasks (Nissen and Bullemer, 1987). Likewise, as for manual sequence learning tasks (Walker, 2005), off-line oculomotor learning seems to follow several identifiable steps after training has ended. The different offline gain in performance described cannot be attributed to the additional block of practice of the learned sequence on block B7, as indicated by the lack of gain in performance between the two last training blocks of learned sequence. Oculomotor performance was not different from late training levels, after a delay of 5 h. In contrast, it was significantly enhanced 24 h later. Nevertheless, in manual sequence learning, absence of improvement over the day was observed only when the sequence was explicitly known by the participants (Robertson et al., 2004b; Walker et al., 2002), whereas implicit sequence learning can be followed by a gain in performance after as little as 4 h after training (Press et al., 2005). The absence of improvement 5 h after training might be due either to the change of effector, or to methodological differences between the tasks such as the interstimulus interval (Destrebecqz et al., 2005) or the length (Pascual-Leone et al., 1993) or structure (Stadler and Neely, 1997) of the deterministic sequence. This absence of improvement is unlikely to be confounded by a circadian effect. Performance to the Psychomotor Vigilance Task (PVT) improves over the day in parallel with the increase in core body temperature (Cajochen et al., 1999). In our case, training occurred in the morning and testing in the afternoon. A circadian modulation would have facilitated the performance during the testing session, whereas no enhancement was observed. Furthermore, there is not any significant circadian effect on performance of the untrained sequence. This study was not designed to assess the influence of sleep on oculomotor sequence learning. Although our results are consistent with an effect of sleep on oculomotor

ARTICLE IN PRESS BR AIN RE S EA RCH XX ( 2 0 06 ) XXX –X XX

memory, they are equally compatible with a simple time effect, as described for implicit sequence learning using a manual serial reaction time task in which performance improved after 12 h of wakefulness (Robertson et al., 2004b), before sleep intervened. This specific issue is left open for future experiments.

3.3.

Early boost in motor sequence learning

Another finding in this study was to show an early boost in performance 30 min after the end of training. Using the finger tapping task, we recently showed that performance temporarily improved when tested 5 to 30 min after training (C. Hotermans, P. Peigneux, A. Maertens de Noordhout, G. Moonen, P. Maquet, submitted). A similar increase in performance level is observed in oculomotor sequence learning 30 min after training. As for the manual task, this gain in performance is no longer observed 5 h later. The similarity in time course between manual and oculomotor tasks speaks for the generality of the boost in motor skill learning. The boost might be attributed to the dissipation of fatigue which would accumulate with practice and impair performance. Alternatively, the boost phase might reflect an early structuring of the memory trace. In the present study, we are not in a position to favor any of these possibilities. Nevertheless, the latter hypothesis appeared most likely since Classen et al. (1998) showed that rapid plastic changes occurred 0 to 20 min after practice of repetitive simple thumb movements. Furthermore, for the manual task, we were able to show that the early boost in performance was predictive of performance levels eventually achieved 48 h later in the absence of any further practice.

3.4.

Sequence-specific knowledge develops with time

Another result of this study was to demonstrate that a specific improvement in reaction time for the learned sequence appears only 24 h after the end of training, an interval which includes a sleep period. During the first hour, performance changes to the same extent for both the learned and the new sequences. Indeed, no significant sequence by session interaction was observed 30 min and 5 h after training. These findings speak for an increase in basic oculomotor ability or the development of a familiarity to the essential features of the task, rather than the acquisition of a genuine sequential knowledge. In contrast, 24 h after training, performance improved significantly more for the learned than the new sequence. In addition, this differential improvement was more pronounced after a 24h delay than for earlier test sessions (30 min and 5 h). These results suggest that sequential knowledge, implicitly acquired during training, is further processed offline during the post-training period. This processing requires several hours before it can be effectively applied and manifest itself behaviorally. Again, this qualitative change in motor skill could arise progressively with time. Alternatively, it might be promoted by sleep periods. At this stage, we do not rule out any of these alternatives although we favor the latter hypothesis. Indeed, previous PET studies suggested that the sequential informa-

7

tion acquired after training to a serial reaction time task was processed during post-training REM sleep (Maquet et al., 2000; Peigneux et al., 2003), and demonstrated significant learningrelated changes in functional integration between cortical regions and basal ganglia (Peigneux et al., 2003). Our results indicate that motor sequence learning represents a basic human ability that can take place through ocular as well as finger or manual movements. Likewise, in manual sequence learning, we have found that ocular motor memories continue to be processed after training has ended, which eventually lead to an improvement of performance at later testing. The time course of this offline processing follows similar phases for ocular and finger movements, with an initial increase of performance occurring early in time, about 30 min after the end of learning, then decreasing back to training level in the following hours. We also found that a specific knowledge of the trained sequence emerged only 24 h later.

4.

Experimental procedures

4.1.

Subjects

Fifty-eight subjects gave their written informed consent to take part in this experiment approved by the Ethics Committee of the Faculty of Medicine of the University of Liège. They were paid for their participation. All were righthanded healthy subjects with normal uncorrected vision. None of them reported any neurological, psychiatric, traumatic or ophthalmic history. After the training session to the serial oculomotor reaction time task (SORT; see below), subjects were tested at various time intervals, either after 30 min, 5 h, or 24 h. In order to avoid any circadian confound, subjects in each group were systematically trained and tested at various circadian phases. In the 24 h Group, training and test sessions for a given subject occurred at the same time of the day, scheduled between 9:30 am and 7:00 pm. In the 5 h Group, subjects were trained between 9:30 am and 2:30 pm, then tested 5 h after between 2:30 and 7:30 pm. They were asked not to sleep between the training and the test session. Subjects in the 30 min Group were trained and tested between 9.30 am and 7.30 pm.

4.2.

Serial oculomotor reaction time (SORT) task

The SORT task was programmed and run using COGENT 2000 (http://www.vislab.ucl.ac.uk/Cogent/) implemented in MATLAB (Mathworks Inc., Sherbom, MA) on a PC system (1.4 GHz) with a 17-inch screen (refresh rate 60 Hz). A head chin rest positioned at 50 cm of the screen restricted head movements to ensure that the participant's straight gaze constantly point towards the center of the screen (on horizontal and vertical axes). In the SORT task, a yellow dot (diameter : 0.68°) is displayed at one of 4 possible screen locations (1–4; visual angle : 21° × 23°; see Fig. 1A). The dot stays at any given location for a constant duration of 550 ms, then abruptly moves to another location. Unbeknownst to subjects, the trajectory followed by

ARTICLE IN PRESS 8

BR AIN RE S EA RCH XX ( 2 0 06 ) XXX–X XX

the dot, i.e. the sequence of its positions, follows one of the following two second-order 8-element sequences: 2 4 3 1 4 2 1 3 (S1) or 3 1 2 4 1 3 4 2 (S2). The 2 sequences are identical in terms of dot locations and transitions frequency but differ by the sub-sequences of three elements they contain. To ensure that the subject fixates the dot at all time, there is a 20% chance at each position that the color of the dot turns to an isoluminant orange color for 34 ms. They are instructed to press a key as soon they perceive the color change. Subjects are told that the optimal strategy to detect the transient color change is to keep their gaze on the dot at all time.

4.3.

Experimental procedure

The experimental design is outlined in Fig. 1B. During the training session, subjects performed 7 consecutive blocks (B1 to B7). During blocks B1 to B5 and block B7, a given sequence, thereafter referred to as the trained sequence, was repeated 15 times per block (duration 66 s). To assess the extent to which subjects learned the trained sequence, block 6 consisted in 5 repetitions of the other sequence, thereafter referred to as the untrained sequence. Assuming that oRTs improvement reflects oculomotor response preparation and anticipation of the next stimulus, reaction times should increase on block 6 only if participants acquired specific knowledge about the sequential regularities characteristic of the trained sequence presented over the other blocks. Inter-blocks interval was 60 s. During the test session, subjects performed 5 further blocks (B8 to B12). The trained sequence was repeated 15 times during blocks B8, B9 and B11. The untrained sequence was presented 5 times during block B10 and 15 times during block B12. At the end of the test session, subjects were debriefed using a standardized procedure to assess their level of explicit knowledge of the sequence gained during SORT practice. First, subjects were asked whether they had noticed anything special about the task. Second, they were asked whether they had noticed some regularities or repetitions in the positions of the dot. Third, they were asked if they had noticed that the dot was always following a given trajectory. After debriefing, it was made clear to the subjects that the dot moved according to a predetermined sequence. They were then asked to generate the trained sequence by drawing it with a pencil on a sheet of paper (generation task). If subjects did not explicitly recollect the sequence trajectory, they were encouraged to produce their best guess. The number of generated triplets (i.e., chunks of three successive locations in the trajectory) belonging to the learned sequence was computed. Subjects were considered as having an explicit knowledge of the learned sequence if they generated more than 3 triplets common with the learned sequence. The threshold was set as the mean number of triplets common to the learned sequence in 1000 randomly generated sequences (mean = 1.08, SD = 1.04) plus 2 standard deviations. This conservative threshold was used in order to discard any subject showing any explicit knowledge of the trained sequence.

4.4.

EOG data recording and analysis

Vertical and horizontal electro-oculograms were recorded on a bipolar montage (Mowrer et al., 1936; Fig. 1C), using a Synamp

acquisition system (Neuroscan, NeuroSoft, Sterling, Virginia). Sampling rate was 1000 Hz. Bandwidth was set between 0.05 and 200 Hz. The ground electrode was positioned on the forehead. Oculomotor reaction times (oRTs) were defined as the delay between the change in dot position and the first saccade initiated in the direction of the move (Fig. 1D). Note that as learning of the sequential regularities progresses, subjects are more likely to anticipate the position of the next dot, i.e. to perform the ocular saccade before the dot actually start moving, in which case a negative oRT is recorded. For each trial, saccade initiation is identified in a time window starting 400 ms before and ending 300 ms after the change in dot position. Windows of successive trials overlap by 150 ms, allowing for the detection of extremely early or late oRTs. Oculomotor reaction times identification used a 2-step procedure for each individual session. At each step, the correlation between a template and the horizontal EOG recordings was computed at each point in time, using a sliding window. In a first step, a saccade recorded in a typical subject (amplitude: 8°; duration: 35 ms) was used as a template to detect representative saccades on individual horizontal EOG recordings. All saccades identified during this first step were then averaged to produce a subject-tailored template for each saccade of a given sequence. In a second step, for each saccade of the sequence, the same procedure as in the first step was applied, using the average template. When a horizontal saccade component was found, its corresponding vertical component was searched for on the vertical electro-oculogram (VEOG). In the case of a discrepancy between the onsets of horizontal and vertical components, the oRT was computed as the latest of the two components. If no correct saccade was found in the analysis time window, the oRT was arbitrarily set to 300 ms, as it is unlikely that any useful saccade would be generated afterwards. Results of the automated computation processes were visually controlled separately for each trial by one of the authors (PR). Any misidentified saccade was discarded from the analysis (on the average, this occurred in 1.4% of saccades during training and 1.9% of saccades during testing). As the distribution of oRTs was likely to be skewed towards progressively shorter (or negative) values as learning progresses, the median oRT for each block was taken as the dependent variable in subsequent statistical analysis. Statistical analyses were conducted using STATISTICA 6.1 software (StatSoft, France). To investigate practice-related improvement of performance during the training session, a repeated-measure analysis of variance (ANOVA) was computed on mean reaction times for blocks B1 to B5 of the learned sequence. Next, to test for the acquisition of sequence knowledge, a 2-way ANOVA compared oRTs between the trained (B5) and the untrained (B6) sequence. We additionally verified that the performance on B7 (trained sequence) was comparable to the level of performance observed on B5. In order to test for possible gain in performance between sessions, we performed an ANOVA with oRTs in blocks B5, B6, B8 and B10 as repeated-measure. The difference in oRTs between blocks B5 and B8 reflected the gain in performance on

ARTICLE IN PRESS BR AIN RE S EA RCH XX ( 2 0 06 ) XXX –X XX

the trained sequence whereas the change in performance related to the unpracticed sequence was assessed by the difference in oRTS between blocks B6 and B10. These 2 gains, as well as their differences, i.e., the sequence by session interaction, were subsequently used to explore betweengroup differences in sequence knowledge acquisition.

Acknowledgments This study was supported by the Belgian Fonds National de la Recherche Scientifique (FNRS), the Fondation Médicale Reine Elisabeth, the Research Fund of ULg, and PAI/IAP Interuniversity Pole of Attraction P5/04. PM is supported by FNRS, GA by a PhD grant from the French ‘Ministère de la Recherche’, and PR by a ‘Marie Curie Individual fellowship’ grant from the EU Commission. REFERENCES

Brandt, S., Stark, L., 1997. Spontaneous eye movements during visual imagery reflect the content of the visual scene. J. Cogn. Neurosci. 9, 27–38. Bruce, V., Young, A., 1998. In the Eye of the Beholder. Oxford Univ. Press, New York. Buysse, D.J., Reynolds, C.F.r., Monk, T.H., Berman, S.R., Kupfer, D. J., 1989. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 28, 193–213. Cajochen, C., Khalsa, S.B., Wyatt, J.K., Czeisler, C.A., Dijk, D.J., 1999. EEG and ocular correlates of circadian melatonin phase and human performance decrements during sleep loss. Am. J. Physiol. 277, R640–R649. Cajochen, C., Knoblauch, V., Wirz-Justice, A., Krauchi, K., Graw, P., Wallach, D., 2004. Circadian modulation of sequence learning under high and low sleep pressure conditions. Behav. Brain Res. 151, 167–176. Classen, J., Liepert, J., Wise, S.P., Hallett, M., Cohen, L.G., 1998. Rapid plasticity of human cortical movement representation induced by practice. J. Neurophysiol. 79, 1117–1123. Destrebecqz, A., Peigneux, P., Laureys, S., Degueldre, C., Del Fiore, G., Aerts, J., Luxen, A., Van Der Linden, M., Cleeremans, A., Maquet, P., 2005. The neural correlates of implicit and explicit sequence learning: interacting networks revealed by the process dissociation procedure. Learn. Mem. 12, 480–490. Ellis, B.W., Johns, M.W., Lancaster, R., Raptopoulos, P., Angelopoulos, N., Priest, R.G., 1981. The St. Mary's Hospital sleep questionnaire: a study of reliability. Sleep 4, 93–97. Epelboim, J.L., Steinman, R.M., Kowler, E., Edwards, M., Pizlo, Z., Erkelens, C.J., Collewijn, H., 1995. The function of visual search and memory in sequential looking tasks. Vision Res. 35, 3401–3422. Fischer, S., Hallschmid, M., Elsner, A.L., Born, J., 2002. Sleep forms memory for finger skills. Proc. Natl. Acad. Sci. U. S. A. 99, 11987–11991. Grosbras, M.H., Leonards, U., Lobel, E., Poline, J.B., LeBihan, D., Berthoz, A., 2001. Human cortical networks for new and familiar sequences of saccades. Cereb. Cortex 11, 936–945. Honda, M., Deiber, M.P., Ibanez, V., Pascual-Leone, A., Zhuang, P., Hallett, M., 1998. Dynamic cortical involvement in implicit and explicit motor sequence learning. A PET study. Brain 121 (Pt 11), 2159–2173.

9

Howard, J.H., Mutter, S.A., Howard, D.V., 1992. Serial pattern learning by event observation. J. Exper. Psychol., Learn., Mem., Cogn. 18, 1029–1039. Johansson, R.S., Westling, G., Backstrom, A., Flanagan, J.R., 2001. Eye–hand coordination in object manipulation. J. Neurosci. 21, 6917–6932. Karni, A., Meyer, G., Rey-Hipolito, C., Jezzard, P., Adams, M.M., Turner, R., Ungerleider, L.G., 1998. The acquisition of skilled motor performance: fast and slow experience-driven changes in primary motor cortex. Proc. Natl. Acad. Sci. U. S. A. 95, 861–868. Kawashima, R., Tanji, J., Okada, K., Sugiura, M., Sato, K., Kinomura, S., Inoue, K., Ogawa, A., Fukuda, H., 1998. Oculomotor sequence learning: a positron emission tomography study. Exp. Brain Res. 122, 1–8. Keele, S.W., Ivry, R., Mayr, U., Hazeltine, E., Heuer, H., 2003. The cognitive and neural architecture of sequence representation. Psychol. Rev. 110, 316–339. Korman, M., Raz, N., Flash, T., Karni, A., 2003. Multiple shifts in the representation of a motor sequence during the acquisition of skilled performance. Proc. Natl. Acad. Sci. U. S. A. 100, 12492–12497. Leigh, R.J., Zee, D.S., 1999. The Neurology of Eye Movements. Oxford Univ. Press, New York. Maquet, P., Laureys, S., Peigneux, P., Fuchs, S., Petiau, C., Phillips, C., Aerts, J., Del Fiore, G., Degueldre, C., Meulemans, T., et al., 2000. Experience-dependent changes in cerebral activation during human REM sleep. Nat. Neurosci. 3, 831–836. Miyashita, K., Rand, M.K., Miyachi, S., Hikosaka, O., 1996. Anticipatory saccades in sequential procedural learning in monkeys. J. Neurophysiol. 76, 1361–1366. Mowrer, O., Ruch, T., Miller, N., 1936. The corneo-retinal potential difference as the basis of the galvanometric method of recording eye movements. J. Physiol. 114, 423–428. Nissen, M.J., Bullemer, P., 1987. Attentional requirements of learning: evidence from performance measures. Cogn. Psychol. 19, 1–32. Pascual-Leone, A., Grafman, J., Clark, K., Stewart, M., Massaquoi, S., Lou, J.S., Hallett, M., 1993. Procedural learning in Parkinson's disease and cerebellar degeneration. Ann. Neurol. 34, 594–602. Peigneux, P., Laureys, S., Fuchs, S., Destrebecqz, A., Collette, F., Delbeuck, X., Phillips, C., Aerts, J., Del Fiore, G., Degueldre, C., et al., 2003. Learned material content and acquisition level modulate cerebral reactivation during posttraining rapid-eye-movements sleep. NeuroImage 20, 125–134. Press, D.Z., Casement, M.D., Pascual-Leone, A., Robertson, E.M., 2005. The time course of off-line motor sequence learning. Brain Res. Cogn. Brain Res. 25, 375–378. Robertson, E.M., Pascual-Leone, A., Miall, R.C., 2004a. Current concepts in procedural consolidation. Nat. Rev., Neurosci. 5, 576–582. Robertson, E.M., Pascual-Leone, A., Press, D.Z., 2004b. Awareness modifies the skill-learning benefits of sleep. Curr. Biol. 14, 208–212. Stadler, M.A., Neely, C.B., 1997. Effects of sequence length and structure on implicit serial learning. Psychol. Res. 60, 14–23. Walker, M.P., 2005. A refined model of sleep and the time course of memory formation. Behav. Brain Sci. 28, 51–104. Walker, M.P., Brakefield, T., Morgan, A., Hobson, J.A., Stickgold, R., 2002. Practice with sleep makes perfect: sleep-dependent motor skill learning. Neuron 35, 205–211. Willingham, D.B., 1999. Implicit motor sequence learning is not purely perceptual. Mem. Cogn. 27, 561–572.