Journal of Neurophysiology

Error message

Notice: PHP Error: Undefined index: custom_texts in highwire_highwire_corrections_content_type_render() (line 33 of /opt/sites/jnl-jn/drupal-highwire/releases/20151124215058/modules/highwire/plugins/content_types/

Contribution of the Premotor Cortex to Consolidation of Motor Sequence Learning in Humans During Sleep

Michael A. Nitsche, Michaela Jakoubkova, Nivethida Thirugnanasambandam, Leonie Schmalfuss, Sandra Hullemann, Karel Sonka, Walter Paulus, Claudia Trenkwalder, Svenja Happe


Motor learning and memory consolidation require the contribution of different cortices. For motor sequence learning, the primary motor cortex is involved primarily in its acquisition. Premotor areas might be important for consolidation. In accordance, modulation of cortical excitability via transcranial DC stimulation (tDCS) during learning affects performance when applied to the primary motor cortex, but not premotor cortex. We aimed to explore whether premotor tDCS influences task performance during motor memory consolidation. The impact of excitability-enhancing, -diminishing, or placebo premotor tDCS during rapid eye movement (REM) sleep on recall in the serial reaction time task (SRTT) was explored in healthy humans. The motor task was learned in the evening. Recall was performed immediately after tDCS or the following morning. In two separate control experiments, excitability-enhancing premotor tDCS was performed 4 h after task learning during daytime or immediately before conduction of a simple reaction time task. Excitability-enhancing tDCS performed during REM sleep increased recall of the learned movement sequences, when tested immediately after stimulation. REM density was enhanced by excitability-increasing tDCS and reduced by inhibitory tDCS, but did not correlate with task performance. In the control experiments, tDCS did not improve performance. We conclude that the premotor cortex is involved in motor memory consolidation during REM sleep.


One of the most challenging endeavors of the last decades has been the identification of neurophysiological correlates of learning and memory formation and to develop rational treatment strategies on this foundation to improve the respective functions (Miniussi et al. 2008; Reis et al. 2008). Long-term potentiation (LTP)—i.e., the enduring strengthening of learning-related synaptic connections—has been identified as the likely physiological basis of learning and memory formation (Rioult-Pedotti et al. 2000; Stefan et al. 2006; Ziemann et al. 2004). Consequently, motor learning is enhanced by facilitatory stimulation of the primary motor cortex in healthy subjects (Galea and Celnik 2009; Hunter et al. 2009; Nitsche et al. 2003a; Reis et al. 2009). Similar effects were accomplished in chronic stroke patients with motor deficits (Fregni et al. 2005; Hummel et al. 2005).

For the consolidation of a motor memory, i.e., the stabilization and improvement of performance in later phases of the learning process, evidence for a beneficial effect of brain stimulation is scarce and has been discussed with some controversy. During motor memory stabilization, areas other than the primary motor cortex might be involved and the information about a specific motor activity stored in these secondary areas likely differs from that retained in the primary motor cortex. For the serial reaction time task (SRTT), in which subjects learn a finger movement sequence (Nissen and Bullemer 1987), the primary motor cortex seems to be involved in the early learning phase and encodes effector-specific information, whereas premotor and parietal cortices are important for memory consolidation and encode more abstract aspects of the task (Grafton et al. 1998; Honda et al. 1998; Maquet et al. 2000). When consolidation takes place seems to depend on task characteristics, such as explicit or implicit learning, declarative or procedural learning, and awareness of and attention to learning (for an overview see Diekelmann et al. 2010; Song 2009). For procedural motor learning, as explored in SRTT, a relevant contribution of rapid eye movement (REM) sleep has been proposed. In humans, REM sleep deprivation impairs memory (Empson and Clarke 1970; Empson et al. 1980; Karni et al. 1994; Smith 1993, 1995; Tilley and Empson 1978). Conversely, it was shown that procedural learning is followed by an increase of REM density (Smith et al. 2004) and recall performance correlates with the amount of REM sleep (Stickgold et al. 2000). Moreover, with regard to motor learning, cerebral areas activated during the learning phase of the SRTT are reactivated during REM sleep, particularly the premotor cortex (Maquet et al. 2000).

Here we aimed to explore the causal role of sleep-dependent activation of task-relevant cortical areas for motor memory consolidation by applying premotor brain stimulation during REM sleep in healthy humans who had learned an SRTT sequence before. Transcranial DC stimulation (tDCS) induces prolonged cortical excitability changes in humans: anodal stimulation increases and cathodal stimulation decreases it (Nitsche and Paulus 2000, 2001; Nitsche et al. 2003b). Blocking N-methyl-d-aspartate receptors (NMDARs) prevents the induction of aftereffects (Liebetanz et al. 2002; Nitsche et al. 2003c) and thus these share a certain similarity with LTP and long-term depression (LTD). In humans, anodal tDCS selectively of the primary motor cortex, but not of the premotor cortex, improved performance in the acquisition phase of the SRTT (Nitsche et al. 2003a). A similar result was obtained in a visuomotor tracking task (Antal et al. 2004).

Anodal, cathodal, or sham tDCS were administered over the premotor cortex during REM sleep after motor sequence learning. Recall of previously learned sequential finger movements was tested either directly after stimulation or the next morning. Sleep architecture using visually scoring sleep stages (according to the criteria published by Rechtschaffen and Kales 1968) and REM density were monitored. To control for global effects of tDCS on reaction time, a control experiment with randomized sequential finger movements during tDCS was conducted (for course of the experiments see Fig. 1). To control for time-dependent but not sleep-dependent consolidation, we conducted a further control experiment, in which tDCS was applied 4 h after learning and thus approximately at a time after learning similar to that of the main experiments, but without intervening sleep. We hypothesized that anodal premotor tDCS during REM sleep would enhance SRTT recall, if this cortex is involved in REM sleep-associated consolidation of procedural memory, caused by an enhancement of task-related cortical activation. For the cathodal excitability-diminishing tDCS condition, we consequently hypothesized no effect or a deleterious effect on memory consolidation. If the effect was specific for motor learning consolidation, stimulation should not affect simple reaction time task performance and, if sleep is crucial for consolidation, stimulation during wakefulness should not improve performance relative to placebo stimulation.

Fig. 1.

Experimental course. The experimental courses for testing the modulation of procedural motor memory consolidation by transcranial direct current stimulation (tDCS) of the prefrontal cortex during rapid eye movement (REM) sleep (experiment 1) as well as the effect of tDCS of the same region on motor performance in a reaction time task, where no sequential learning processes are involved (experiment 2), and the design of experiment 3 exploring motor consolidation during wakefulness are shown. In experiment 1, subjects performed the serial reaction time task (SRTT), an implicit sequential finger movement learning task, in the early evening. Fifteen-minute anodal, cathodal, or placebo tDCS was performed in the second and, if needed, also in the third REM period during sleep. Group A (cathodal, anodal, placebo stimulation, applied randomized in a crossover design) was awakened immediately after the end of the last REM period and tDCS was administered for the performance of a recall test, in which reproduction of the learned sequences was demanded; the group was then allowed to sleep again. Group B (anodal and placebo stimulation, randomized crossover design) was not wakened and conducted the recall test the following morning at 7:05 a.m. In experiment 2, subjects performed a randomized sequential finger movement task, which was otherwise identical to the SRTT during the daytime. This group received anodal, cathodal, or placebo tDCS during performance of the task. In experiment 3, the participants learned the SRTT sequence during wakefulness. At 4 h after the learning procedure, anodal or sham tDCS was applied for 15 min. Immediately afterwards, as well as 8 and 24 h after sequence learning, a recall test was performed to explore consolidation.



In experiment 1 (anodal, cathodal, or placebo tDCS, SRTT), 20 healthy, right-handed, good sleepers, without subjective complaints about sleep disturbance in the past or during the last 4 wk [anodal tDCS group A (7 subjects, 5 females) mean age: 36.6 ± 10.6 (SD) years; anodal tDCS group B (7 subjects, 5 females) mean age: 33.7 ± 11.7 (SD) years; cathodal tDCS (6 subjects, all female) mean age: 36.7 ± 13.5 (SD) years] participated after giving written informed consent. Each subject participated in one of the real stimulations (anodal A, anodal B, or cathodal tDCS) and one placebo stimulation session. They underwent complete polysomnographic (PSG) recording within one adaptation night. During this night subjects slept in a separate room. Electroencephalographic (EEG,10–20 system), electrooculographic (EOG), and electromyographic (EMG) recordings of the submental and both anterior tibial muscles, electrocardiography, airflow, respiratory movements of thorax and abdomen, oxygen saturation, and body position were monitored to exclude sleep disorders. Only subjects with subjectively undisturbed sleep and a normal PSG recording according to age in the adaptation night, without increased periodic limb movements index >10/h (Zucconi et al. 2006), without alpha intrusions or with atypical sleep patterns, and without a clinically relevant sleep-related breathing disorder were defined as good sleepers and allowed to participate in the study.

In experiment 2 (random sequences), 12 healthy, right-handed subjects [26.3 ± 2.9 (SD) years, 6 female] were included after giving informed consent. We tested 12 subjects in this experiment because it has been shown in a foregoing study of our group that this number of subjects is sufficient to achieve significant results if differences exist between tDCS stimulation conditions (Nitsche et al. 2003a).

In experiment 3 (consolidation during wakefulness), 32 healthy, right-handed volunteers (24.75 ± 2.9 yr, 20 females) with subjectively undisturbed sleep participated. We included this relatively large group of subjects in this control experiment to be able to also detect relatively small effects, if present.

In all experiments, a repeated-measures design was performed for each stimulation condition (anodal/cathodal vs. placebo tDCS). Each subject received anodal or cathodal and placebo stimulation in different sessions separated by ≥1 wk to prevent carryover effects of task learning and stimulation. The order of the different stimulation conditions was randomized between subjects.

The study was approved by the local ethics committee and conformed to the Declaration of Helsinki; informed consent was obtained from the participants before the start of the experiments.

Transcranial direct current stimulation

Direct current (1 mA) was induced via rubber electrodes covered with conductive electrode cream in experiment 1 and via saline-soaked sponge electrodes (surface of rubber and saline-soaked sponge electrodes, 35 cm2) in experiments 2 and 3. In sleep experiment 1, it was essential that the conductivity of the electrodes was stable for a relatively long time because tDCS took place some hours after the electrodes were fixed on the skin. Electrode cream guarantees such stable conductivity, whereas water-soaked electrodes would have run dry in the meantime and thus would have compromised stimulation under these circumstances. In the remaining experiments, tDCS was performed more or less immediately after positioning of the electrodes on the head and thus water-soaked electrodes, which are more comfortable for the subjects, were well suited. Anodal tDCS was delivered by a specially developed, battery-driven constant-current stimulator (Schneider Electronics, Gleichen, Germany). Constant-current flow was controlled by a voltmeter. In experiment 1, tDCS was delivered during the second REM period and, in some cases, with only a short second REM period also during the third REM period. We decided to apply tDCS during the second REM period because this is commonly more stable and lasts longer than the first REM period. Specifically, the duration of the first REM period has been reported to last between about 19 ± 5.5 (SD; Boukadoum and Ktonas 1988) and 16.8 ± 12.0 min (Gann et al. 1992), whereas REM period 2 lasts for 30.2 ± 11 min, REM period 3 for 25.3 ± 12 min, and REM period 4 for 43.5 ± 16.7 min in healthy young subjects (Boukadoum and Ktonas 1988). Total stimulation duration was 15 min. In experiment 2, stimulation was performed during the whole course of the experimental trial, which lasted for about 15 min. We performed tDCS during and not before task performance in experiment 2 to optimize the efficacy of tDCS because it was shown in a previous experiment that tDCS applied over the primary motor cortex during, but not before, task conduction enhances performance (Kuo et al. 2008; Nitsche et al. 2003a). In experiment 3, anodal or sham tDCS was performed 4 h after sequence learning during wakefulness, immediately before a short rehearsal of the task, for 15 min.

One electrode (to which the term anodal/cathodal stimulation refers) was placed contralaterally to the performing right hand, the other electrode ipsilaterally. A premotor-prefrontal electrode position was chosen by positioning the first electrode 3 cm anterior to C3 and the other contralaterally above the right orbit. This electrode position approximates the one chosen for transcranial magnetic stimulation (TMS) of the left dorsal premotor cortex in previous studies (Bestmann et al. 2005; Buhmann et al. 2004; Münchau et al. 2002; Rizzo et al. 2004) on the basis of neuroimaging studies (Fink et al. 1997; Schluter et al. 1998, 1999). Here, premotor TMS was administrated 2–3 cm anterior and in Bestmann et al. (2005) 1 cm medial to the primary motor cortex hot spot of small hand muscles. The position C3 (according to the international 10–20 EEG system) resembles the hand area of the primary motor cortex (Cui et al. 2000; Lemm et al. 2004; Serrien et al. 2003). Thus our electrode position (also taking into account the relatively large size of the tDCS electrodes) is assumed to cover not only the dorsal premotor cortex but also, to some degree, adjacent prefrontal areas. The efficacy of tDCS to modify cortical excitability and information processing is largely restricted to the area under the electrode, as shown by simulation, electrophysiological, and cognitive studies (Boros et al. 2008; Miranda et al. 2006; Nitsche et al. 2003a, 2007). For example, primary motor, but not premotor, tDCS (distance of electrode center ∼3 cm2) during learning of the SRTT affected performance. Thus a more widespread functional influence of tDCS on distant cortical areas is improbable. Placebo stimulation was performed by switching the stimulator on for 5 s and then slowly switching it off again. Hereby an initial itching under the electrode is perceived, but no lasting effects of the stimulation on cortical excitability are induced (Nitsche and Paulus 2000). A similar placebo stimulation protocol has been shown to be indistinguishable from verum stimulation (Gandiga et al. 2006).

Serial reaction time task (experiments 1 and 3)

The participants were seated in front of a computer screen at eye level behind a response pad with four buttons (numbered 1–4) and were instructed to push each button with a different finger of the right hand (index finger for button 1, middle finger for button 2, ring finger for button 3, and little finger for button 4). An asterisk appeared in one of four positions, horizontally spaced on a computer screen and permanently marked by dots. The subjects were instructed to press the key corresponding to the position of the asterisks as fast as possible. After a button was pushed, the go signal disappeared. The next go signal was displayed 500 ms later. The learning test consisted of 8 blocks of 120 trials. In blocks 1 and 6 the sequence of asterisks followed a pseudorandom order in that asterisks were presented equally frequently in each position and never in the same position in two subsequent trials. In blocks 2 to 5 as well as in blocks 7 and 8, the same 12-trial sequence of asterisk positions was repeated 10 times (e.g., abadbcdacbdc). The recall test consisted of only the first 3 blocks of the learning trial. Subjects were not told about the repeating sequence.

Random sequence control (experiment 2)

This experiment was identical to the learning trial of the SRTT, apart from the fact that not repetitive sequences, but rather a random order of stimuli, was presented in each block.

Experimental procedures

Experiment 1 (Fig. 1).

Subjects arrived in the sleep laboratory at 7:00 p.m. After reading the instructions, they were placed in front of a computer monitor and conducted the learning session of the SRTT. They were then allowed to relax by watching television, listening to the radio, or reading. At 10:30 p.m., the electrodes for polysomnographic recording (EEG montage T5–A2, C4–A1, T3–T5, T5–01, T4–T6, T6–02 according to the international 10–20 system; the EEG was adjusted to spare C3 and surrounding areas for the tDCS electrode placement), mental EMG electrodes, horizontal and vertical EOG electrodes, and tDCS electrodes were affixed and subjects went to bed at 11:00 p.m. Sleep parameters were continuously monitored during the night by a trained neurologist (MJ). Subjects received 15 min anodal, cathodal, tDCS, or placebo stimulation during the second REM period. If the second REM period lasted <15 min, the remaining stimulation duration was applied during the following REM period. One group (A, anodal/cathodal tDCS) of subjects was roused immediately after the REM period in which the last stimulation took place. By 5 min later, the subjects were seated in front of the computer and performed the SRTT recall test. Afterward, all subjects were allowed to go to sleep again. The second group (B, only anodal tDCS) was allowed to sleep until the next morning at 7:00 a.m. and performed the recall test of the SRTT at 7:05 a.m. Afterward the electrodes were removed and the subjects were allowed to go home. Each of both groups received anodal/cathodal or placebo stimulation in different sessions in randomized order. They received different, but equally difficult, SRTT learning sequences in the two nights; allocation of the respective sequences to the stimulation condition was randomized. Both sessions were separated by ≥1 wk to avoid carryover effects of tDCS. Aftereffects of 15 min tDCS lasted for 1–2 h (Nitsche and Paulus 2001; Nitsche et al. 2003b). Because the learning-related effects of SRTT performance might last much longer, different parallel versions of SRTT were applied in each session.

Experiment 2 (Fig. 1).

The random sequence control experiment was identical to the SRTT, with the exception that no repetitive sequences, but random stimuli, were presented in each block. In this experiment, subjects performed the tests during daytime. Anodal, cathodal, or placebo stimulation was administered during performance. The order of sessions was balanced between participants and single sessions were separated by ≥1 wk to avoid carryover effects. This control experiment was performed to rule out any effects of premotor stimulation on motor performance independent from sequence learning, i.e., task-routine–related changes of reaction time. Since former experiments (Kuo et al. 2008; Nitsche et al. 2003a) showed a maximum effect of tDCS on performance during stimulation, we decided to apply tDCS during instead of before task performance in this case. Since for this learning-independent task we had no reason to believe that the effect of tDCS on performance might be sleep dependent, the experiments were carried out during daytime between 9:00 a.m. and 6:00 p.m.

Experiment 3 (Fig. 1).

Subjects arrived in the laboratory before noon. After reading the instructions, they were placed in front of a computer monitor and conducted the learning session of the SRTT. They were then sent home and were allowed to perform recreational activities, but not to conduct massive motor or learning activities (e.g., playing an instrument) or to sleep. Four hours later, they arrived back in the laboratory and received 15 min anodal tDCS or placebo stimulation. Immediately after the end of stimulation, they performed the SRTT recall test. SRTT recall was repeated 8 and 24 h after learning. The order of the single sessions was counterbalanced between subjects and the sessions were separated by ≥1 wk to avoid carryover effects.

Data analysis

SRTT and random sequence.

In each trial, reaction time (RT) was measured from the appearance of the go signal until the first button was pushed. For each block of trials of a given experimental condition, mean RT was calculated for each subject separately, incorrect responses, and reaction times of <200 ms or >3,000 ms, or those that were >3SDs of the individual subject's mean response time were discarded. Mean reaction times were standardized to block 1 for each subject in each stimulation condition separately for the learning and recall conditions, to control for initial RT group differences. Furthermore, the SD of reaction times for each subject in every block was calculated as an index of variability of reaction times. An error rate (ER) was calculated to assess the number of incorrect responses for each block and each subject in each stimulation condition.

Statistical analyses were performed with separate ANOVAs (level of significance: 0.05) for each real tDCS (anodal/cathodal) versus sham stimulation condition (including the factors tDCS-condition and block) for RT, ER, and variability for absolute, and for RT additionally for standardized values. RT, ER, and variability value differences between the respective tDCS-conditions were compared by use of paired-sample, two-tailed, Student's t-test (level of significance: 0.05) within each block of the task when the ANOVAs revealed significant results. Additionally, t-tests were used to compare RT of block 1 with those of the remaining blocks and to compare RT of block 8 (learning session) with those of blocks 2 and 3 of the recall session in each condition. For experiment 1, some additional calculations were performed. Because in some subjects tDCS was delivered only in the second REM period and, in other subjects, in the second and third REM periods, we compared REM density and SRTT recall performance between both groups for group A (anodal tDCS) with t-tests for independent samples, to rule out an influence of these different modes of tDCS application on REM density and SRTT performance. To test a dependence of SRTT consolidation from anodal tDCS-induced REM density changes in group A, we performed a Pearson correlation between RT differences of the last block of the learning session and the second recall block (absolute and standardized values) and REM density of the third REM period. The critical P values were set to 0.05.

For the random sequence control experiment, RTs were standardized, as described earlier, and repeated-measures ANOVAs (factors block and stimulation condition) and post hoc t-tests were calculated for the absolute and standardized values. Critical P values were set to 0.05.

Sleep parameters.

Polysomnography was scored according to standardized criteria by Rechtschaffen and Kales (1968). Parameters of sleep macrostructure [sleep onset latency, REM sleep latency, time in bed, total sleep time, awake time (min, %), NREM 1 (min, %), NREM 2 (min, %), NREM 3 + 4 (min, %), REM (min, %)] were compared by use of paired-sample, two-tailed Student's t-test (level of significance: 0.05) for the tDCS and placebo-tDCS nights within groups A and B and between the groups.

Two blinded scorers counted the frequency of rapid eye movements for each 30-s period and individual means for each REM period were calculated (REM density according to Aserinsky 1969). Two-factorial, repeated-measures ANOVA was performed (level of significance: 0.05) for anodal tDCS versus placebo tDCS and cathodal tDCS versus placebo tDCS separately, since subjects were not identical in both groups. REM density served as a dependent variable, whereas REM period and stimulation condition served as independent variables. Post hoc paired-sample two-tailed t-tests (level of significance: 0.05) were calculated to reveal differences between stimulation conditions for each REM period and between REM periods within one stimulation condition. Furthermore, the REM densities of groups A and B for each REM period and placebo/real tDCS conditions were compared separately to control for group differences. To test the interrater reliability regarding the REM density analysis, we correlated the REM counts of both raters, using Spearman's rank-correlation test.


None of the subjects had to be excluded from the experiments due to the exclusion criteria outlined earlier.

SRTT (consolidation during sleep, experiment 1)

After the end of the experiment, subjects were asked whether they were aware of a reoccurring sequence during SRTT performance. Over all conditions, seven subjects declared some awareness, but none of them was able to recall the sequence or parts of it, nor did the performance of this group differ from the overall results, as revealed by inspection of the data, similar to a foregoing experiment of our group (Nitsche et al. 2003a).

With regard to anodal tDCS (experiment 1a), for RT, the ANOVAs revealed a significant main effect of block for absolute and standardized values in both groups [recall after REM sleep (group A)/next morning (group B); Table 1)]. As shown in Fig. 2, this was due to a decrease of RT in the course of performance, with the exception of the random blocks 6 (learning) and 1 (recall condition), which did not contain the learned sequence. Conversely, the main effect of tDCS was not significant. The interaction between block and tDCS was significant only for group A in the standardized calculation. This is due to decreased RT in the recall test of the anodal tDCS relative to the placebo stimulation condition (Fig. 2) on the one hand and to slower RTs of the sequential recall blocks, compared with block 8 of the learning session, after placebo stimulation, on the other.

View this table:
Table 1.

Results for the two-factorial repeated-measurement ANOVAs conducted for the different stimulation conditions as applied to the SRTT (experiment 1), the randomized sequence experiment (experiment 2), and REM density

Fig. 2.

Anodal tDCS during REM sleep acutely improves recall in implicit motor learning. For the learned sequences, reaction times are shorter in the anodal stimulation recall condition compared with that in the placebo stimulation condition, if recall is performed immediately after REM sleep during the night. Reaction times of the last learning block are identical to the recall blocks for the anodal stimulation condition, but prolonged for the placebo stimulation, thus arguing for memory consolidation after anodal tDCS. This effect is significant if reaction times are standardized to the randomized sequence block 1 (B), but a trend is also visible for the absolute values (A). If recall is performed in the morning after the learning session, the beneficial effect of anodal tDCS disappears. Here performance is improved compared with the early learning phase in both stimulation conditions (C and D). Learning took place before sleep in both groups and recall immediately after REM phase during night or the next morning. Filled symbols indicate significant reaction time differences of anodal/placebo stimulation conditions relative to block 1, the asterisk marks significant differences between anodal and placebo stimulation conditions for a single block (2-tailed t-test, paired samples, P < 0.05). Error bars in this and the following figures represent SE.

With regard to the dependence of memory consolidation from the anodal tDCS-induced enhancement of REM density in group A, a trend for a correlation was noticed. The larger the REM density, the faster were the RTs of the second recall block, compared with the last block of the learning session (absolute values r = 0.620, P = 0.147; standardized values r = 0.682, P = 0.096). For group A (anodal tDCS), four subjects received tDCS only during REM period 2 and three subjects during REM periods 2 and 3 (average inter-REM period 84 min). Neither REM density nor SRTT recall performance differed between these groups, as revealed by the respective t-test (P > 0.05).

RT in the random block 1 (learning and recall) did not differ between or within the groups and stimulation conditions (absolute values calculation, P > 0.05), with the exception of a slower RT in the cathodal tDCS, compared with the respective placebo tDCS condition. Additionally, the RTs in the random block 6 (learning) did not differ significantly from the random block 1. For variability, the respective ANOVAs revealed no significant effects of block and tDCS condition or the interaction between these variables. For error rates, the main effect of block was significant in group B; all other variables did not differ between the respective conditions.

Results of the RT ANOVAs for the cathodal stimulation experiment (1b) show a significant main effect only for block. This was caused by a decrease of RT in the course of the experiment in both the cathodal stimulation and the placebo stimulation group, within the course of the experiment, as shown by the t-test, with the exception of the random blocks 6 (learning) and 1 (recall condition), which did not contain the learned sequence. RT in block 1 (learning) did not differ between the groups. In the recall condition RTs differed significantly between the random block 1 and the sequence blocks 2 and 3 only in the cathodal stimulation, but not in the placebo stimulation condition. Since RT of block 1 (recall) was slower after cathodal tDCS compared with placebo stimulation (Fig. 3), this difference might account for the significant reduction of RTs within the sequence blocks relative to performance in the random block 1. Consequently, RTs of sequence blocks 2 and 3 did not differ significantly between cathodal and placebo stimulation conditions. There was a significant difference between block 8 (learning session) and block 2 (recall session) in the placebo stimulation condition for standardized values, which, however, was not observable for block 3 of the recall session. For variability and error rates, the respective ANOVAs revealed no significant effects of block and tDCS condition or interaction between these variables. REM density and SRTT performance did not correlate significantly.

Fig. 3.

Effect of cathodal tDCS during REM sleep on recall in implicit motor learning. Reaction times are shorter in the cathodal stimulation sequence recall condition (recall blocks 2 and 3) compared with the random item condition (recall block 1), if recall is performed immediately after REM sleep during the night. However, the difference between the cathodal and placebo stimulation sessions did not reach significance. Since reaction time (RT) of the random stimuli block was significantly prolonged in the cathodal tDCS recall condition relative to the respective sham condition, the relative improvement of sequence recall compared with the random stimulus condition should be attributed to the random stimulus RT difference between the cathodal and sham stimulation groups. Compared with block 8 of the learning block, standardized reaction time of block 2 (recall block) in the placebo stimulation condition is significantly prolonged. Filled symbols indicate significant reaction time differences of cathodal/placebo stimulation conditions relative to block 1, the asterisk marks significant differences between cathodal and placebo stimulation conditions for a single block, and hash symbols mark differences between block 8 of the learning condition and blocks 2 and 3 of the recall condition (2-tailed t-test, paired samples, P < 0.05). A depicts absolute, B standardized RTs.

Sleep macrostructure

With the exception of REM density, no sleep parameter differed significantly between placebo and tDCS nights within groups A and B or between the groups (P > 0.05). In experiment 1a (anodal tDCS), the ANOVA revealed significant main effects of tDCS condition and REM period as well as a significant interaction between both variables for REM density. As shown in Fig. 4, this is caused by a significant increase in REM density in the anodal stimulation condition, which is much more prominent in REM periods 2 and 3, compared with the placebo stimulation. In REM period 1 (prestimulation), REM density was identical for both stimulation conditions. In experiment 2b (cathodal tDCS), the ANOVA resulted in a significant main effect for REM period. The t-test revealed a significant lower REM density solely during the second REM period under cathodal tDCS compared with the placebo stimulation condition.

Fig. 4.

tDCS modifies REM density during and after stimulation. Whereas REM densities are identical in REM period 1 and thus before tDCS for both stimulation conditions, REM density is increased in REM periods 2 and 3 in the anodal tDCS, but decreased during REM period 2 in the cathodal tDCS condition relative to the placebo stimulation condition. Stimulation was performed in REM period 2 and, in some subjects, also in REM period 3. Filled symbols indicate significant REM density differences of anodal/cathodal/placebo stimulation conditions relative to REM period 1; the asterisks mark significant differences between anodal/cathodal and placebo stimulation conditions for a given REM period (2-tailed t-test, paired samples, P < 0.05).

The interrater correlation was ≥0.937 in all analyses of sleep parameters.

Random sequence (experiment 2)

Here the ANOVAs revealed a significant main effect of block in the standardized RT calculation. All other main effects and interactions were not significant (see Table 1); t-tests revealed only slight differences for the cathodal tDCS in two later occurring blocks in relation to block 1; anodal, cathodal, and placebo RT did not differ within single blocks. As depicted in Fig. 5, the significant block effect is most probably due to slight variations between the blocks. However, these were similar for all stimulation groups and did not follow a linear trend. For variability, the respective ANOVAs revealed no significant effects of block and tDCS condition or the interaction between these variables. For error rates, only the main effect of block was significant.

Fig. 5.

tDCS does not reduce reaction time independent from sequence learning. Stimulation of the premotor/prefrontal cortex did not affect reaction time in a random sequence task compared with the placebo tDCS condition [absolute values A, standardized values B (2-tailed t-test, paired samples, P < 0.05)], with the exception of small RT increases in late blocks of cathodal stimulation relative to block 1. Thus a learning-independent effect of tDCS on motor reaction time cannot sufficiently explain its effect on sequence recall.

SRTT (consolidation during wakefulness, experiment 3)

For RT, the ANOVAs conducted for absolute and standardized values revealed significant main effects for the factor block, but not for tDCS nor for the interaction between both factors (Table 1). The main effect for the factor block is caused by reduced RTs, compared with the random block 1, in the later sequences for both learning and recall (Fig. 6), with the exception of the random block 6, which did not contain the learned sequence. In relation to the sequence block 8 of the learning procedure, the absolute RTs of the sequence recall blocks were shorter, although this effect was no longer significant for the standardized RTs. RTs did not differ between anodal or sham tDCS in any block. For error rates and variability, the ANOVAs resulted in significant effects only for block, which was caused by lower error rates and variability in later blocks.

Fig. 6.

There was no effect of anodal tDCS on SRTT performance during daytime. Depicted are the absolute (A) and standardized (B) reaction times of SRTT performance, when anodal or sham tDCS was performed 4 h after sequence learning without intervening sleep. Block 1 in the learning and recall conditions and block 6 in the learning condition are random; the remaining blocks are sequence blocks. Recall 1 was performed 4 h, recall 2 was performed 8 h, and recall 3 was performed 24 h after learning. Anodal tDCS did not enhance performance relative to sham stimulation. Filled symbols depict RT differences vs. block 1 (2-tailed t-test, paired samples, P < 0.05); error bars are SE values.

The subjects reported no side effects of tDCS nor were they able to tell whether they received placebo or real tDCS when asked.


As shown by the results of this study, tDCS of the premotor cortex during REM sleep is able to improve recall in a sequential finger movement learning task, if tested immediately after stimulation. Reaction time was significantly shortened for the learned sequences during recall if preceded by excitability-enhancing anodal tDCS, compared with placebo stimulation. Specifically, RTs after anodal tDCS were identical to the last block of the learning session, whereas they were prolonged after placebo stimulation, if subjects were awakened immediately after stimulation. This is in accordance with former functional imaging studies, implying an important role of the premotor cortex in motor memory consolidation (Grafton et al. 1995, 1998; Hazeltine et al. 1997; Honda et al. 1998). Since stimulation was administered effectively during REM sleep and there was a trend for a positive correlation between memory consolidation and anodal tDCS-generated REM density increase, the results additionally favor a relevance of sleep, especially the REM periods, for these processes.

Hereby, the effects were stimulation polarity dependent: only anodal tDCS, and thus an excitability enhancement, significantly improved performance. Anodal stimulation clearly enhanced sequence recall not only relative to random stimuli, but also compared with the sham stimulation group. At first sight, cathodal tDCS also seems to have improved sequence recall compared with the random stimulus condition. However, inspection of the data shows that this is probably mainly due to a significantly prolonged RT in the random stimulus condition after cathodal tDCS, compared with placebo tDCS. Consequently, cathodal stimulation did not significantly improve recall of the learned sequence relative to sham stimulation. The reason for prolonged RT solely in the random stimulus block after cathodal tDCS remains unclear. Since a similar but minor effect was seen in experiment 2 for the experiment exploring the effects of premotor tDCS only on random blocks, it might be speculated that excitability diminution of the premotor cortex by cathodal tDCS impairs movement preparation, which takes place in the premotor cortex, by reducing task-related activation. However, since this effect had a relatively large variability, small size, and was not present in the learned sequences, it cannot be excluded that it was caused by chance.

Comparing SRTT performance in block 8 of the learning session and during recall for the anodal stimulation condition shows another interesting fact with regard to memory consolidation. Whereas performance did not differ between the respective blocks after anodal tDCS, when subjects were awakened directly after real tDCS, in the sham stimulation condition recall was worse. However, if subjects were allowed to sleep until the morning, performance was identical for both conditions after real or sham tDCS and did not differ from learning block 8. Since this disturbing effect of awakening during REM sleep was specific for the learned sequences, it cannot be attributed to an unspecific effect of the awakening procedure. If consolidation, however, is taking place in REM sleep and anodal tDCS during REM sleep improves consolidation, this result makes sense: here anodal stimulation should already have enhanced consolidation and thus made it more resistant to interference compared with sham stimulation. In the condition of a nondisturbed sleep (recall test in the morning), REM-dependent consolidation should have been completed during the night and thus no major further improvement via tDCS should be expected. This specific pattern of results—i.e., increasing the velocity of improved performance, but not the terminal extent of improvement—has been demonstrated in a former SRTT experiment of our group for anodal tDCS of the primary motor cortex in healthy subjects (Nitsche et al. 2003a).

The beneficial effect of anodal tDCS on learning processes was previously demonstrated in other studies (Antal et al. 2004; Kincses et al. 2004; Marshall et al. 2004; Nitsche et al. 2003a) and is most probably caused by the ability of an externally induced excitability enhancement to bring learning-related synaptic connections nearer to their modification threshold. More specifically, because the premotor cortex participates in motor learning consolidation and motor learning is suggested as being NMDAR-dependent and accompanied by cortical excitability enhancements (Rioult-Pedotti et al. 2000), the excitability enhancement induced by anodal tDCS should have enhanced memory consolidation by these mechanisms. A likely candidate mechanism for this effect is the membrane-depolarizing effect of anodal tDCS (Nitsche et al. 2003c), which would increase the probability of strengthening of learning-related NMDARs (Frégnac et al. 1990).

An unspecific effect of anodal tDCS on reaction time independent of sequence learning can be ruled out, since subjects showed no relevant reduction of RT with regard to the random sequences in the recall condition (block 1). Moreover, in the control experiment 2, where only random sequences were presented, tDCS did not show any effect in the direction of a performance improvement. In all experiments involving SRTT, the participants showed similar learning curves, with reduced RT for the blocks containing the learned sequence (blocks 2–5, 7, and 8), but not the random stimuli blocks (1 and 6 during learning) during the course of the experimental session, thus ruling out that prior learning differences caused the results. The results can also not be attributed to differences of error rate or variability because these were grossly identical for all conditions. Since, with the exception of REM density, no sleep parameters differed with regard to the respective stimulation conditions, an unspecific effect of tDCS on sleep should not have caused the effects either. A confounding effect of sleep inertia to the results is improbable because this should have affected all groups tested to the same degree. Importantly, the RTs of the first, random, recall block are identical to the random blocks (1 and 6) during learning. In the case of sleep inertia, these should have been increased.

In experiment 3, tDCS was performed immediately before SRTT sequence recall, but without intervening sleep. Here anodal tDCS did not have a consolidation-improving effect, compared with that of sham stimulation. Similar to the group awakened immediately after REM-phase anodal tDCS and to the group awakened the morning after sham or real tDCS, SRTT performance was similar to that of the last block of the learning sequence. This hints to the possibility that in the daytime stimulation group some kind of consolidation has already taken place during wakefulness, on the one hand, but that sleep disruption might impair memory consolidation, on the other. Since anodal tDCS of the premotor cortex prevented this disruption of consolidation, it might be speculated that consolidation again becomes fragile during REM sleep, which is supported by studies showing that REM deprivation impairs memory consolidation (Empson and Clarke 1970; Empson et al. 1980; Karni et al. 1994; Smith 1993, 1995; Tilley and Empson 1978). Moreover, it might be that during sleep memory contents are consolidated other than those during daytime. Indeed it has been shown that for motor sequence learning, sequence information is consolidated during daytime, whereas goal-directed information is consolidated during sleep (Cohen et al. 2005). The latter might be localized in the premotor cortex.

In principle, at least for the group that received tDCS during the second and third REM periods, it could be argued that the aftereffects of tDCS on non-REM sleep have caused the results. However, this does not seem probable for the following reasons. First, we have shown in a recently conducted experiment that the impact of anodal tDCS on SRTT performance does occur if tDCS is performed only during, but not before, the actual learning process (Kuo et al. 2008). Second, SRTT performance did not differ between the REM-2 only and REM-2/REM-3 groups (anodal tDCS, group A), which would have been expected if we had relevantly influenced information processing during non-REM sleep. Finally, the trend for a correlation between SRTT performance and REM density in this group offers direct evidence for a relationship between REM sleep and motor memory consolidation. However, since we did not perform tDCS during non-REM sleep, it cannot be definitely ruled out that stimulation during these sleep periods might also have improved SRTT performance or REM density. This should be explored in future experiments.

The parallel improvement of recall and REM density accomplished by anodal stimulation argues for a common neuronal basis of both mechanisms. This hypothesis is underscored by a trend for a positive correlation between REM density and memory consolidation in the anodal tDCS condition of group A. One candidate mechanism could be pontine wave activity, which has been proposed as being associated with REM sleep and improving memory consolidation in animals (Datta et al. 2003, 2004). Possibly, premotor/prefrontal anodal tDCS could positively modulate cortical effects of this activity by its excitability-enhancing effects and thus enhance not only memory consolidation but also REM density. However, in our experiments relatively large electrodes were used, which may cover both the premotor cortex and the frontal eye field (FEF). Thus alternatively by enhancing excitability of the FEF, which participates in the generation of REM-associated eye movements (Hong et al. 1995), tDCS could have improved synaptic transmission in this area and hereby increased REM density. In this case, increased REM density would be caused by increased motor activity, but would not necessarily reflect changes of sleep microstructure. The dissociation of REM density and recall in the cathodal stimulation condition (i.e., no effect on recall and reduced REM density after cathodal tDCS) is more in accordance with the latter explanation. Future experiments, accomplishing more selective stimulation of the premotor cortex, are needed to clarify this issue to a greater extent.

In a former study by our group, anodal tDCS was applied on different cortices during the learning phase of the same task (Nitsche et al. 2003a). Here, only stimulation of the primary motor cortex was effective, whereas premotor stimulation did not modulate RT in the SRTT. This is in accordance with the hypothesis that the primary motor cortex is involved primarily in the acquisition phase of motor learning, whereas premotor areas participate prominently in the consolidation of those memory traces.

Premotor tDCS was solely effective when recall was tested immediately after stimulation during the night, but not in the group tested the morning afterward. It should be stressed that the results of the current experiment support a role of the premotor cortex in procedural motor memory consolidation, but that this does not exclude the involvement of other areas, such as supplementary motor area and parietal cortices, which are also involved in motor learning and represent a memory-relevant network (Praeg et al. 2006). Moreover, the results of the present study do not allow one to make assumptions about the relative contribution of these different areas to consolidation. Especially for possible future clinical application, e.g., in motor rehabilitation, it will be important to quantify the effects of tDCS over different cortical areas during learning and consolidation on performance to design-optimized stimulation protocols.

Taken together, the present results provide further evidence for a prominent involvement of the premotor cortex in the consolidation of motor memory during REM sleep. Moreover, since tDCS did not disturb sleep and had relatively specific functional effects on cognitive and sleep parameters, it may evolve as an interesting tool for sleep research in the future.


This work was supported by European Union–Marie Curie Fellowship Training Stipend HPMT-CT-2001-0043 to M. Jakoubkova. Part of the project was supported by Federal Ministry of Education and Research of the Federal Republic of Germany Project Grant 13-6287 3.


No conflicts of interest, financial or otherwise, are declared by the author(s).


  1. Antal et al., 2004.
  2. Aserinsky, 1969.
  3. Bestmann et al., 2005.
  4. Boros et al., 2008.
  5. Boukadoum and Ktonas, 1988.
  6. Buhmann et al., 2004.
  7. Cohen et al., 2005.
  8. Cui et al., 2000.
  9. Datta et al., 2003.
  10. Datta et al., 2004.
  11. Diekelmann and Born, 2010.
  12. Empson and Clarke, 1970.
  13. Empson et al., 1980.
  14. Fink et al., 1997.
  15. Frégnac et al., 1990.
  16. Fregni et al., 2005.
  17. Galea and Celnik, 2009.
  18. Gandiga et al., 2006.
  19. Gann et al., 1992.
  20. Grafton et al., 1998.
  21. Hazeltine et al., 1997.
  22. Honda et al., 1998.
  23. Hong et al., 1995.
  24. Hummel et al., 2005.
  25. Hunter et al., 2009.
  26. Karni et al., 1994.
  27. Kincses et al., 2004.
  28. Kuo et al., 2008.
  29. Lemm et al., 2004.
  30. Liebetanz et al., 2002.
  31. Maquet et al., 2000.
  32. Marshall et al., 2004.
  33. Miniussi et al., 2008.
  34. Miranda et al., 2006.
  35. Münchau et al., 2002.
  36. Nissen and Bullemer, 1987.
  37. Nitsche et al., 2007.
  38. Nitsche et al., 2003c.
  39. Nitsche et al., 2003b.
  40. Nitsche and Paulus, 2000.
  41. Nitsche and Paulus, 2001.
  42. Nitsche et al., 2003a.
  43. Praeg et al., 2006.
  44. Rechtschaffen and Kales, 1968.
  45. Reis et al., 2008.
  46. Reis et al., 2008.
  47. Rioult-Pedotti et al., 2000.
  48. Rizzo et al., 2004.
  49. Schluter et al., 1999.
  50. Schluter et al., 1998.
  51. Serrien et al., 2003.
  52. Smith, 1993.
  53. Smith, 1995.
  54. Smith et al., 2004.
  55. Song, 2009.
  56. Stefan et al., 2006.
  57. Stickgold et al., 2000.
  58. Tilley and Empson, 1978.
  59. Ziemann et al., 2004.
  60. Zucconi et al., 2006.
View Abstract