The authors have declared that no competing interests exist.
Conceived and designed the experiments: JNC WED LMP NH PAL. Performed the experiments: JNC. Analyzed the data: JNC WED LMP NH PAL. Contributed reagents/materials/analysis tools: JNC WED PAL. Wrote the paper: JNC WED LMP NH PAL.
Sleep plays a role in memory consolidation. This is demonstrated by improved performance and neural plasticity underlying that improvement after sleep. Targeted memory reactivation (TMR) allows the manipulation of sleep-dependent consolidation through intentionally biasing the replay of specific memories in sleep, but the underlying neural basis of these altered memories remains unclear. We use functional magnetic resonance imaging (fMRI) to show a change in the neural representation of a motor memory after targeted reactivation in slow-wave sleep (SWS). Participants learned two serial reaction time task (SRTT) sequences associated with different auditory tones (high or low pitch). During subsequent SWS, one sequence was reactivated by replaying the associated tones. Participants were retested on both sequences the following day during fMRI. As predicted, they showed faster reaction times for the cued sequence after targeted memory reactivation. Furthermore, increased activity in bilateral caudate nucleus and hippocampus for the cued relative to uncued sequence was associated with time in SWS, while increased cerebellar and cortical motor activity was related to time in rapid eye movement (REM) sleep. Functional connectivity between the caudate nucleus and hippocampus was also increased after targeted memory reactivation. These findings suggest that the offline performance gains associated with memory reactivation are supported by altered functional activity in key cognitive and motor networks, and that this consolidation is differentially mediated by both REM sleep and SWS.
Slow-wave sleep and rapid eye movement sleep are associated with the reactivation and consolidation of a motor skill within distinct brain networks.
After a motor skill is learned, the memory undergoes "offline" processing so that improvement occurs even without further practice. Sleep has been shown to enhance this consolidation and, in the process, to reorganize the brain regions involved. However, it remains unclear how sleep does this, and whether different sleep stages have different contributions. One popular idea is that the memory trace is reactivated during slow-wave sleep—a period of sleep characterized by synchronized activity at a slow frequency and high amplitude, as recorded by electroencephalography (EEG)—which drives memory reorganization within the brain. To test this in humans, we took advantage of "targeted memory reactivation," where replay of specific memories is cued by presentation of a sound that was present during learning. After sleep, motor performance was faster for cued memories, suggesting that the trace was consolidated during sleep. Coupled with this, brain activation and connectivity in several motor-learning areas was enhanced for the cued memory. Furthermore, some changes in brain activity were associated with time spent in slow-wave sleep, while others were associated with time spent in rapid-eye movement sleep. These observations provide further insight into sleep's unique role in memory consolidation by showing that offline skill enhancement depends on the reactivation of specific memories, and the associated changes in neural activity may rely upon processing that unfolds across different stages of sleep.
Memory consolidation begins the moment new information is encoded and is a process where initially fragile memories are stabilised, strengthened, and reorganised in the brain [
The spontaneous reactivation of cerebral activity after learning is hypothesised to underscore such plasticity during sleep and the associated performance gains [
Overnight procedural memory consolidation is linked to enhanced functional activation within striatum, hippocampus, cerebellum, and motor cortical regions, as well as striato-hippocampal and medial prefrontal-hippocampal (mPFC-HPC) connectivity [
The role of rapid eye movement (REM) sleep in procedural memory reactivation and consolidation is debated [
We aimed to test this by cueing a modified version of the serial reaction time task (SRTT) [
(a) Learning (L) of the SRTT task consisted of interleaved blocks of the cued and uncued sequence, and also random blocks. (b) The cued sequence is replayed during periods of SWS in groups of 12 sequences (CUE) and equivalent periods of silence (NO-CUE). (c) Retest (R) of the SRTT takes place the following morning in the MRI scanner, followed shortly afterwards by the explicit memory test outside of the scanner.
Firstly, we confirmed that sequence learning occurred prior to sleep by showing that RTs were significantly faster for sequence trials compared to random trials for both cued, t(21) = 9.2,
RT (ms) | Errors (%) | ||||
---|---|---|---|---|---|
Cued | Uncued | Cued | Uncued | ||
357.2 ± 51.0 | 351.3 ± 47.1 | 7.8 ± 4.7 | 5.8 ± 3.3 | ||
421.4 ± 31.6 | 422.4 ± 34.1 | 6.9 ± 4.0 | 7.6 ± 3.5 | ||
64.2 ± 32.7 | 71.1 ± 36.2 | 0.9 ± 6.2 | 1.8 ± 5.3 | ||
335.8 ± 39.9 | 342.7 ± 44.9 | 3.8 ± 2.5 | 4.7 ± 3.3 | ||
307.2 ± 51.7 | 304.6 ± 54.1 | 5.3 ± 2.8 | 4.7 ± 3.2 | ||
392.2 ± 30.2 | 400.0 ± 32.8 | 6.2 ± 3.6 | 6.7 ± 4.7 | ||
84.9 ± 40.7 | 95.4 ± 46.1 | 0.9 ± 3.1 | 2.0 ± 3.4 | ||
21.4 ± 24.0 | 8.6 ± 23.8 | 3.9 ± 4.0 | 1.6 ± 3.1 | ||
50.0 ± 25.6 | 46.8 ± 24.5 | 2.4 ± 3.0 | 1.1 ± 2.2 | ||
29.3 ± 22.1 | 22.5 ± 24.3 | 0.7 ± 5.1 | 0.9 ± 5.7 | ||
20.7 ± 32.8 | 24.3 ± 33.5 | 1.8 ± 6.1 | 0.2 ± 6.6 |
Early: mean performance during the first four blocks of each sequence at retest.
Late: mean performance during the last four blocks of each sequence at retest.
Improvement: the difference between presleep and postsleep performance.
Next, we explored early improvement in RTs due to TMR effects by subtracting initial blocks of sequence retest from the final blocks of sequence learning (presleep performance), providing a measure of “early sequence improvement.” Here, we found improvement for the cued sequence was significantly greater than the uncued, t(21) = 2.46,
Next, we compared cued and uncued sequence performance during the last blocks of the retest (late sequence improvement) and found that the advantage for the cued sequence was no longer present, t(21) = 0.58,
To summarise, we found a behavioural cueing effect at the beginning of retest, but cued and uncued sequence performance equalised across the duration of the retest block such that this was no longer present in late sequence blocks. In light of our previous findings [
Accuracy is typically high for motor sequence learning (MSL) tasks and does not change substantially across practice or offline [
There was a trend for more errors to be made for the cued sequence relative to the uncued sequence prior to sleep, t(20) = 1.86,
After sleep, there was a significantly greater error rate improvement for the cued relative to the uncued sequence at early blocks, t(20) = 2.46,
Considering all sleep electroencephalography (EEG) as 30 s epochs, 96% of CUE periods and 97% of NO-CUE periods were in SWS. All others were stage 2 and excluded from further EEG analyses. Duration of sleep stages and total sleep are displayed in
Duration (min ± Standard Deviation) | |
---|---|
32.7 ± 26.7 | |
217.9 ± 46 | |
102.1 ± 34.3 | |
84.2 ± 27.3 | |
440 ± 65 |
To examine the neural responses associated with a procedural memory that has undergone TMR during sleep, we contrasted activity during performance of the cued sequence with performance of the uncued sequence at retest (
(a) The basic comparison between cued and uncued showed reduced activity in left caudate (−20, 24, −10) for the cued sequence. (b) SWS was associated with enhanced activation in bilateral caudate (16, 8, 20 and −12, 20, 12), and bilateral hippocampi (26, −34, 2 and −22, −34, 6) for the cued sequence relative to the uncued. (b) REM sleep was associated with cueing related activity enhancement in left cerebellum (−32, −54, −44 and 20 −72, −26), left superior parietal cortex (−28, −56, 68 and 22, −54, 38), left sensorimotor cortex (SMC) (−40, −32, 68), left dorsolateral prefrontal cortex (dlPFC) (−30, 34, 28) and right premotor cortex (PMC) (42, −2, 32 and 42, −2, 58). These findings were whole brain corrected (
Next, we investigated whether other factors were related to the cueing effect by running the same cued > uncued contrast again with five regressors added at the second level: duration (mins) of SWS (SWStime), duration of REM (REMtime), duration of stage 2 (Stage2time), the number of replayed sequences (replays) and the procedural cueing effect. Prior work has shown that the length of SWS predicts behavioural consolidation effects after both normal sleep [
Region | MNI x, y, z(mm) | No. of voxels | Peak T | Peak Z | Peak P(unc) |
---|---|---|---|---|---|
(Cued > Uncued) * SWS duration (mins) | |||||
Right caudate | 16, 8, 20 | 606 | 8.49 | 4.97 | <0.001 |
Left caudate | −12, 20, 12 | 624 | 6.22 | 4.24 | <0.001 |
Right heschl’s gyrus | 44, −16, 8 | 77 | 5.06 | 3.75 | <0.001 |
Right hippocampus | 26, −34, 2 | 139 | 4.3 | 3.38 | <0.001 |
Left hippocampus | −22, −34, 6 | 128 | 3.74 | 3.06 | 0.001 |
(Cued > Uncued) * REM sleep duration (mins) | |||||
Right superior parietal lobe | 22, −54, 38 | 677 | 5.98 | 4.15 | <0.001 |
Left dlPFC | −30, 34, 28 | 342 | 5.03 | 3.74 | <0.001 |
Left superior parietal lobe | −28, −56, 68 | 106 | 4.87 | 3.66 | <0.001 |
Left fusiform gyrus | −40, −72, −10 | 74 | 4.71 | 3.59 | <0.001 |
Right PMC | 42, −2, 32 | 123 | 4.69 | 3.58 | <0.001 |
Left cerebellum | −8, −70, −28 | 107 | 4.41 | 3.43 | <0.001 |
Left somatosensory cortex | −40, −32, 68 | 79 | 4.38 | 3.42 | <0.001 |
Left cerebellum | −32, −54, −44 | 73 | 4.15 | 3.3 | <0.001 |
Right cerebellum | 20, −72, −26 | 119 | 3.98 | 3.2 | 0.001 |
Right PMC | 42, −2, 58 | 80 | 3.88 | 3.15 | 0.001 |
Left supramarginal gyrus | −58, −44, 30 | 83 | 3.74 | 3.06 | 0.001 |
Right middle temporal gyrus | 54, −56, 10 | 112 | 3.63 | 3 | 0.001 |
Left superior temporal gyrus | −44, −38, 20 | 51 | 3.62 | 2.99 | 0.001 |
Right cerebellum | 14, −78, −50 | 66 | 3.4 | 2.86 | 0.002 |
Right supramarginal gyrus | 48, −38, 36 | 54 | 3.29 | 2.79 | 0.003 |
(Cued > Uncued) * stage 2 sleep duration (mins) | |||||
Right caudate | 16, 8, 22 | 223 | 6.1 | 4.19 | <0.001 |
Left cingulate gyrus | −14, −14, 34 | 83 | 4.01 | 3.22 | 0.001 |
(Cued > Uncued) * replayed sequences | |||||
Right primary motor cortex | 62, 2, 38 | 84 | 4.71 | 3.59 | <0.001 |
(Cued > Uncued) * procedural cueing effect (cued minus uncued RT) | |||||
Right orbitofrontal cortex | 16, 38, −8 | 51 | 4.73 | 3.6 | <0.001 |
Bilateral midbrain | 0, −18, −16 | 86 | 4.56 | 3.51 | <0.001 |
Right anterior cingulate | 20, 40, 14 | 133 | 4.23 | 3.34 | <0.001 |
Right middle frontal gyrus | 36, 10, −36 | 159 | 4.19 | 3.32 | <0.001 |
Left anterior cingulate | −2, 4, −6 | 108 | 4.07 | 3.25 | 0.001 |
The main effect of TMR across the whole brain, showing increased activity when considering SWS duration, REM sleep duration, stage 2 sleep duration, replays and the procedural cueing effect, voxel threshold of
Region | MNI x, y, z(mm) | No. of voxels | Peak T | Peak Z | Peak P(unc) |
---|---|---|---|---|---|
(Uncued > Cued) | |||||
Left caudate | −20, 24, −10 | 86 | 4.09 | 3.42 | <0.001 |
Left occipital lobe | −26, −102, 4 | 111 | 4.07 | 3.41 | <0.001 |
(Uncued > Cued) * SWS duration (mins) | |||||
Left somatosensory cortex | −62, −16, 40 | 80 | 4.62 | 3.54 | <0.001 |
Right somatosensory cortex | 62, −10, 42 | 89 | 4.14 | 3.29 | 0.001 |
Right middle frontal gyrus | 42, 28, 50 | 55 | 4.02 | 3.22 | 0.001 |
(Uncued > Cued) * REM sleep duration (mins) | |||||
Right caudate | 22, 24, 20 | 248 | 5.42 | 3.92 | <0.001 |
(Uncued > Cued) * stage 2 sleep duration (mins) | |||||
Right subgenual cingulate | 8, 8, −14 | 96 | 4.75 | 3.61 | <0.001 |
Left primary motor cortex | −38, −32, 70 | 51 | 4.1 | 3.27 | 0.001 |
(Uncued > Cued) * replayed sequences | |||||
Right caudate | 14, 8, 18 | 69 | 5.34 | 3.88 | <0.001 |
Right middle temporal gyrus | 50, −34, −8 | 253 | 4.47 | 3.46 | <0.001 |
Right precuneus | 20, −40, 0 | 84 | 3.54 | 2.94 | 0.002 |
The main effect of TMR across the whole brain, showing decreased activity that was associated with SWS duration, REM sleep duration, stage 2 sleep duration, replays, and the procedural cueing effect, voxel threshold of
When time spent in stage 2 was examined [(cued > uncued) * Stage2time] the pattern of activity was similar to that observed with the SWS regressor, with significantly increased activity for the cued sequence in the same right caudate cluster, and decreased activity in left SMC.
REM has also been linked to motor sequence memory reactivation and consolidation [
The number of sequences that were played as cues during sleep could potentially have influenced changes in functional activity. When examining the number of replays as the second-level regressor [(cued > uncued) * replays], we found increased activity in the right SMC, indicating that more replays were associated with a larger response for the cued sequence in these areas. Conversely, activity in the right caudate decreased for the cued sequence; therefore, the replays regressor showed the reverse pattern to the SWS regressor.
Next, the amount of overnight improvement in performance has previously been linked to neurophysiological changes [
After identifying localised activation differences associated with TMR during SWS, we sought to examine the functional connectivity of task-related regions that showed sensitivity to TMR. This was achieved with four psychophysiological interaction (PPI) analyses seeded in right and left hippocampus and right and left caudate nucleus, all based on peak coordinates identified in relation to the SWStime regressor [(cued > uncued) * SWStime]. Each analysis explored how connectivity from the seed region to the whole brain differed between cued and uncued sequences (
Crucially, both hippocampal seeds showed enhanced connectivity with key motor regions during cued relative to uncued sequence performance. Left hippocampus (−22, −34, 6) showed greater connectivity to right PMC, right putamen, left putamen, and thalamus, and a cluster spanning bilateral thalamus and midbrain (
A PPI analysis revealed enhanced connectivity for the cued sequence between left hippocampus (−22, −34, 6) and right putamen (36, −2, 4) and PMC (58, 4, 22). Contrasts displayed as sagittal and coronal projections superimposed on a standard MNI brain. Colour bar indicates t-values. Anatomical labelling based on peak z-score location.
Region | MNI x, y, z(mm) | No. of voxels | Peak T | Peak Z | Peak P(unc) |
---|---|---|---|---|---|
Cued > Uncued | |||||
−12, 20, 12 | |||||
Bilateral thalamus and midbrain | 6, −22, −4 | 1,108 | 6.76 | 4.77 | <0.001 |
Right temporal pole (middle) | 34, 6, −38 | 124 | 5.41 | 4.16 | <0.001 |
Left fusiform gyrus | −46, −54, −20 | 75 | 4.15 | 3.46 | <0.001 |
Left lingual gyrus | −32, −92, −14 | 57 | 4 | 3.36 | <0.001 |
16, 8, 20 | |||||
Left superior parietal cortex | −14, −76, 60 | 127 | 5.13 | 4.01 | <0.001 |
−22, −34, 6 | |||||
Left middle temporal gyrus | −46, −32, −6 | 168 | 4.53 | 3.68 | <0.001 |
Right putamen and insula | 36, −2, 4 | 102 | 3.82 | 3.25 | 0.001 |
Left putamen and thalamus | −26, −24, 0 | 55 | 3.76 | 3.21 | 0.001 |
Right PMC | 58, 4, 22 | 62 | 3.72 | 3.19 | 0.001 |
Right middle temporal gyrus | 56, −36, −10 | 71 | 3.52 | 3.05 | 0.001 |
Bilateral midbrain and thalamus | 2, −16, −2 | 78 | 3.52 | 3.05 | 0.001 |
26, −34, 2 | |||||
Right fusiform gyrus | 34, −56, −16 | 85 | 4.48 | 3.66 | <0.001 |
Left caudate | −4, 26, 2 | 53 | 4.18 | 3.47 | <0.001 |
Uncued > Cued | |||||
−12, 20, 12 | |||||
Left cerebellum | −24, −76, −36 | 60 | 4.96 | 3.93 | <0.001 |
PPI analysis of connectivity between four seed regions and the rest of the brain, voxel threshold of
We show that targeted reactivation of a procedural memory alters functional activity and connectivity of motor memory networks in the human brain. The enhanced response speed for the cued sequence occurred alongside increased caudate and hippocampal responses associated with time spent in SWS, and increased hippocampal-caudate connectivity. Furthermore, REM sleep was linked with cueing-related activity changes in additional motor regions including SMC, PMC, and cerebellum. Together, these results support distinct contributions of different sleep stages to consolidation, with SWS facilitating consolidation in key subcortical regions (striatum and hippocampus) that are known to support sequence learning [
The finding that REM is associated with changes in functional activation, even though TMR occurred during preceding periods of SWS, is of particular significance because it suggests a link between NREM memory reactivation and the processing that occurs in subsequent REM sleep. REM’s role in the reactivation and consolidation of procedural memories is controversial [
Importantly, we found cueing-related increases in the responses of bilateral caudate nuclei and hippocampi that were associated with time in SWS. Long-term systems consolidation for motor skills is characterised by a gradual shift of the representation from caudal to ventral striatum, alongside decreasing recruitment of corticocerebellar circuits [
Interactions between the striatum and hippocampus have been suggested to underscore procedural learning and consolidation, perhaps mediated by connections with mPFC [
Our observation that TMR-dependent changes in caudate and hippocampal activity were related to SWS adds to our previous finding that SWS is linked to the impact of TMR on behavioural measures [
While the SWS-associated functional increases for the cued memory occurred in subcortical regions known for their roles in many types of learning, REM-associated increases related to cortical and cerebellar regions that are for the most part motor learning specific. These include increased activity in bilateral cerebellum, right ventral PMC and left SMC. SMC is critical for processing complex finger movements [
Interestingly, REM and SWS were linked to opposing patterns of activity in SMC and striatum: cueing-related SMC responses decreased in association with SWS and increased in association with REM, while right caudate activity decreased in association with REM and increased in association with SWS. Furthermore, an additional motor cortical region involved in directing eye movements and visual attention, the FEF [
Our connectivity analysis also revealed that for the cued sequence, the left hippocampus was more connected to right PMC, and bilateral thalamus and midbrain. Thalamus and striatum form a segregated loop with cortical motor regions in order to process motor information [
Notably, although longer SWS duration was associated with enhanced response in bilateral caudate nuclei after TMR, comparison of cued and uncued sequences without any additional covariates showed decreased activity in left caudate. An overnight decrease in striatal response is in line with some prior MSL studies that contrast the brain activity associated with sleep and wake retention intervals [
The stage 2 regressor identified a small number of regions that were also associated with SWS, with increased activity for the cued sequence in the right caudate and decreased activity in left SMC. Stage 2 duration has previously been correlated with procedural consolidation [
The number of sequences played to participants was positively correlated with SWS duration (r = 0.49,
Neuronal firing sequences in rodents [
There are some limitations to the current work. Our behavioral cueing effect was apparent during early blocks at retest but not later blocks where our sequence-specific measure was calculated. As a result, we cannot be certain whether the cueing effect at early sequence blocks was due to unspecific improvements in sensorimotor mapping or sequence learning. Thus, it is possible that TMR integrated the tones more tightly with the cued sequence, allowing faster responses. Prior work in which only sequence-specific skill has been shown to benefit from sleep-dependent consolidation [
It should also be noted that the two sequences share some basic features (e.g., triplets such as 2-4-3), meaning it is possible that TMR benefitted these features in the uncued sequence. The fact that we find significant performance differences in early blocks and in our prior work [
To conclude, we show that TMR of a procedural memory alters functional activity and connectivity changes in striatum and hippocampus, and this altered function may explain the behavioural effects associated with TMR. We provide tantalising hints that REM sleep, as well as SWS, after TMR is important for neural changes that support enhanced postsleep performance of a procedural skill. These findings further our understanding of sleep’s unique role in memory consolidation by showing that offline skill enhancement depends on the reactivation of specific memory traces, and the associated changes in neural activity rely upon processing that may unfold across several subsequent stages of sleep.
Twenty-five (16 males) healthy participants aged 18–35 y (mean age = 23.8 y, SD ± 4.2) volunteered. Three were excluded because of ceiling performance at learning, falling asleep during the fMRI scanning session, and disrupted SWS as a result of cueing. Data from the remaining twenty-two (14 male) participants aged 18–35 y (mean age = 23.5 y, SD ± 4.3) were analysed. Prestudy questionnaires determined that participants had no history of psychiatric diseases, neurological, sleep, or motor disorders and kept a normal sleeping pattern in the week prior to the experiment. Participants were free of any form of medication, except for females using the contraceptive pill. They were asked to abstain from caffeine and alcohol 24 h prior to testing and between test sessions and to avoid napping on the experimental day. All participants were right-handed, confirmed by a score of 80% or more on the Edinburgh Handedness scale [
Participants arrived at 7–8pm for the first session and were fitted for polysomnography (PSG). They then performed an adapted SRTT [
Each sequence was accompanied by pure tones, four high-pitched tones were used for one sequence (fifth octave; A/B/C#/D), and four lower pitched tones were used for the other (fourth octave; C/D/E/F). Participants performed 20 blocks of each sequence, followed by four random blocks containing no repeating sequence, (“R” displayed centrally), containing high- (two blocks) and low-pitched tones (two blocks)
Trials contained an auditory tone and visual cue in one of four spatial locations, corresponding to a four-button box used with all fingers of the left hand (
Participants were invited to sleep overnight in the Neuroscience and Psychology of Sleep (NaPS) Laboratory at the University of Manchester, where they were monitored with PSG. Lights out was at 11pm. Brown noise was presented throughout the night. During periods of SWS, one sequence’s tones were replayed just loud enough to be audible above the brown noise (at approximately 48 dB), in the same order as learning and at a speed similar to mean presleep performance, in blocks of 2 min replay (CUE), followed by 2 min silence (NO-CUE). Replay was only instigated after 3 min of what the experimenter considered to be stable SWS, in line with AASM criteria. Sequence A and B were counterbalanced across cued and uncued conditions, and tones (high/low pitch) were counterbalanced across sequences. Cues were stopped upon signs in the EEG of arousal or leaving SWS.
Participants were woken up at 7–8am. The retest session took place 11am–12pm during fMRI, consisting of 24 sequence blocks (12 cued and 12 uncued), followed by 24 random blocks (12 blocks containing cued tones and 12 containing uncued tones). “REST” was displayed centrally during 15 s breaks between blocks. Order of learning (i.e., whether participants began a session with sequence A or B), replay, and retest was counterbalanced across participants. Lastly, free recall was measured outside the scanner with participants marking sequence order on paper. The Stanford Sleepiness Scale assessed alertness prior to learning and retest sessions [
All experimental scripts were executed using MATLAB 6.5 (The MathWorks Inc., Natick, MA, 2000) and Cogent 2000 (Functional Imaging Laboratory, Institute for Cognitive Neuroscience, University College, London). Sounds were presented via a pair of Sony noise cancelling headphones during the learning session, via PC speakers during sleep replay, and via an MR compatible headphone system (MR Confon) during retest (fMRI). A serial four-button box attached to a Domino multicontroller from Micromint recorded participant responses, with a time resolution of approximately 1 ms.
Functional MRI data were acquired using an eight-channel head coil with a Siemens 3T Allegra MR scanner. The BOLD signal was recorded with T2*-weighted fMRI images obtained via a gradient echo-planar imaging (EPI) sequence. We acquired 50 oblique transaxial slices at 25-degree tilt, in an ascending sequence, voxel size 3 x 3 x 2.8 mm, matrix size of 64 x 64, flip angle of 80 degrees, repetition time (TR) of 2,960 ms, and echo time (TE) of 30 ms. A structural T1-weighted image was also acquired, using a 3D IR/GR sequence with a matrix size of 224 x 256 x 176, cubic voxels with isotropic resolution of 1 mm3, TR of 2,040 ms, TE of 5.57 ms, inversion time of 1,100 ms, and flip angle of eight degrees.
Three measures of behavioural performance improvement were calculated by comparing presleep to postsleep performance in terms of both accuracy and RT: (1) Early sequence improvement: mean performance on the first four blocks of SEQ_C and SEQ_U at retest subtracted from mean performance on the last four blocks of SEQ_C and SEQ_U at learning. This measure identifies whether any cueing effects are present immediately upon retest. (2) Late sequence improvement: mean performance on the last four blocks of SEQ_C and SEQ_U at retest subtracted from mean performance on the last four blocks of SEQ_C and SEQ_U at learning. This measure identifies whether any cueing effects are present toward the end of the retest session. (3) Sequence-specific improvement: this utilised a well-established method to separate sequence skill from learning of sensorimotor mapping between response keys and visual stimuli, by subtracting sequence from random performance [
We also calculated how strong the cueing effect was for each participant, in terms of their RT performance and their explicit sequence knowledge. The “procedural cueing effect” was obtained by subtracting early sequence improvement (defined above) for the cued from the uncued sequence. The “explicit cueing effect” was calculated by subtracting explicit sequence knowledge for the cued from the uncued sequence (note that this was only measured after consolidation) [
Mixed ANOVA and paired sample
Electrodes were affixed at standard locations, F3, F4, C3, C4, C5, C6, CP3, CP4, CP5, CP6, P7, P8, O1, and O2, referenced to the combined mean of left and right mastoid, according to the 10–20 system. Also attached were left and right electro-oculagram, left and upper electromyogram and forehead ground electrode. Impedance below 5Ω was verified, and the digital sampling rate was 200 Hz. Data were scored according to standard criteria [
Functional imaging data were analysed using Statistical Parametric Mapping 8 software (SPM8; Wellcome Department of Cognitive Neurology, London, UK). The first two volumes of each functional EPI run were removed to allow for T1 equilibration. Two participants were excluded from analysis for excessive movement >3.5 mm. Functional images were realigned to correct for motion artifacts using iterative rigid body realignment, minimizing the residual sum of squares between all scans and the first scan. Functional images were then spatially normalised to the Montreal Neurological Institute brain (MNI space), resampled to voxel size 2 x 2 x 2 mm. Lastly, a spherical Gaussian smoothing kernel (full-width half maximum = 8 mm) was applied to each participant’s normalised data.
Statistical analysis of MRI data at the single subject level used the general linear model (GLM) [
Contrast parameter images were generated for each participant with balanced linear t-contrasts, including one-sample
All analyses were whole brain corrected via a Monte Carlo simulation [
We examined the functional connectivity between regions using PPI. Four separate PPI’s were conducted. Each spherical seed region (radius 6 mm) was based on peak coordinates of the group response to the cued > uncued contrast with SWStime as a second-level covariate. Coordinates were chosen from the results when SWStime was the regressor of interest because this identified regions that were hypothesised a priori to show changes in functional connectivity after TMR, based on previous work [
(XLSX)
SRTT performance at training, testing of explicit sequence knowledge, and correlations between behaviour and EEG features (fast spindles and slow oscillations).
(DOCX)
We would like to thank Martyn McFarquhar for helpful comments regarding data analysis.
dorsolateral prefrontal cortex
electroencephalography
echo-planar imaging
frontal eye fields
fusiform face area
functional magnetic resonance imaging
general linear model
hemodynamic response function
Montreal Neurological Institute
medial prefrontal cortex
medial prefrontal-hippocampal
motor sequence learning
Neuroscience and psychology of sleep
nonrapid eye movement
premotor cortex
psychophysiological interaction
polysomnography
rapid eye movement
reaction time
standard error of the mean
sensorimotor cortex
serial reaction time task
slow-wave sleep
echo time
targeted memory reactivation
repetition time