Participants

Young (age range 18–30 years old) healthy volunteers of all genders were recruited from KU Leuven and the surrounding communities. Inclusion criteria were: 1) right-handed, as assessed by the Edinburgh Handedness Inventory [17], 2) no prior extensive training with a musical instrument requiring dexterous finger movements (e.g., piano, guitar) or as a professional typist, 3) free of medical, neurological, psychological, or psychiatric conditions, including depression and anxiety as assessed by the Beck’s Depression and Anxiety Inventories [18, 19], 4) no indications of abnormal sleep, as assessed by the Pittsburgh Sleep Quality Index (PSQI [20]; 5), not considered extreme morning or evening types, as quantified with the Horne & Ostberg chronotype questionnaire [21], 6) free of psychoactive and sleep-influencing medications, 7) non-smokers, and 8) no trans-meridian trips or work night shifts in the three months prior to participation. One hundred and sixteen individuals met these inclusion criteria (across the two experiments) and thus initiated participation in an experimental session. All participants gave written informed consent before the start of the study and received compensation for their participation.

As the specific effects of sleep on the integration of novel motor information into memory has not been previously investigated, a literature-based a priori estimation of the hypothesized effect size is not possible. Nonetheless, given the identical task and an analogous experimental design, sample sizes were determined on the basis of the effect size measured in our previous work investigating the influence of compatible vs. incompatible ordinal structure on motor memory integration [7]. The primary effect of interest in this earlier work was the group main effect in the response time (RT) for novel transitions in the session 2 training phase, as assessed with a 2 (compatible vs. incompatible) x 20 (practice blocks) ANOVA. This group main effect was significant (F(1,36) = 5.29, p = 0.027, Ƞp2 = 0.128 / Cohen’s F = 0.383). Corresponding power analysis (Effect size f = 0.383, tails = 2; alpha = 0.05, Power = 0.80, correlation among repeated measures = 0.72) via the software G*Power [22] resulted in an estimated minimum of 21 subjects per group. In Experiment 1, we aimed to include 25 subjects per group to be slightly more conservative. To achieve this desired sample, 55 participants started the protocol. Five individuals were excluded from all data analyses due to not completing the experiment (n = 4) or failure to comply to experimental instructions (n = 1), resulting in a final sample size of N = 50 (34 females). Note that two subjects were excluded from analyses of Session 2 test runs due to a keyboard malfunction during one or more blocks of this run.

In Experiment 2, the estimated sample size was increased by 20% to account for higher between-subject variance as task completion was not completed in a controlled laboratory environment (as opposed to our previous work) but at the participants’ own home via the web-based data acquisition platform Psytoolkit [23, 24]. Of the 61 participants who started Experiment 2, 1 was excluded from data analysis for failure to comply to experimental instructions, resulting in the planned sample size of 60 participants. Following inspection of the data, a further 3 participants (1 belonging to the AM-PM, 2 to the PM-AM group) were excluded from analyses as they failed to appropriately perform the motor task (i.e., statistical outliers (> 3 SD from group mean) on sequence accuracy during Session 1), resulting in a final sample size for analyses of N = 57 (40 females).

General experimental procedures

In both experiments, participants completed two sessions of motor tasks (described below), separated by a delay of either approximately 4 (Experiment 1; Fig 1A) or 12 hours (Experiment 2; Fig 1B). During the first experimental session (Session 1), participants learned a specific motor sequence (Sequence 1). In the second experimental session (Session 2), integration of new movements into the motor sequence knowledge developed in Session 1 was assessed (see details below). Importantly, the effect of sleep on integration processes was assessed with the comparison between a sleep group (diurnal and nocturnal in Experiments 1 and 2, respectively) and a matched wake control group (see below for details). At the start of each session, the psychomotor vigilance task (PVT [25]) and Stanford Sleepiness Scale (SSS [26]) were administered to provide objective and subjective assessments of vigilance, respectively. Additionally, the beginning and end of Sessions 1 and 2, respectively, included runs of a pseudorandom Serial Reaction Time Task (SRTT) variant (see subsequent section for details), allowing the assessment of general motor execution independent of sequence learning per se.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Experimental design.

A. Experiment 1 consisted of a Nap/No-Nap design. Participants completed Session 1 (S1) in the morning, followed by a 90-minute episode of either sleep or quiet wakefulness monitored with standard polysomnography. They then completed Session 2 (S2) in the afternoon. B. Experiment 2 consisted of an AM-PM / PM-AM design. Participants completed S1 either in the morning or evening, and S2 on the same evening or on the following morning, respectively. C. For both experiments, during S1 and S2 participants performed a motor task (SRTT) during which they learned separate motor sequences (S1: Sequence 1, corresponding to keys 4-7-3-8-6-2-5-1, in which 1 through 8 are the left pinky to the right pinky fingers, excluding the thumbs; S2: Sequence 2, corresponding to keys 4-7-2-8-6-3-5-1). Sequence 2 was highly similar to Sequence 1, as 75% of the elements were in the same ordinal position. The positions of keys 2 and 3 (represented in red) were switched in Sequence 2, resulting in 4 novel key transitions to be learned relative to Sequence 1.

https://doi.org/10.1371/journal.pone.0280591.g001

All participants followed a constant sleep/wake schedule (according to their own rhythm) for 4 days (3 nights) prior to the first experimental session. Compliance was assessed with self-reported sleep diaries, as well as wrist actigraphy recordings (ActiGraph, Pensacola, USA; Experiment 1 only). Sleep quality and quantity for the night preceding the experimental sessions were assessed with the St. Mary’s sleep questionnaire [27].

Participants were instructed to avoid alcohol, caffeine, daytime sleeping/napping, and heavy exercise during the 12h preceding each experimental session, and to avoid performing sequential finger movements during the time between the two experimental sessions.

Motor tasks

The motor task was identical to our earlier research [7]. Specifically, participants performed an explicit serial response time task (SRTT) coded and implemented in MATLAB ([28] MATLAB, 2020; Experiment 1) or Psytoolkit ([23, 24]; Experiment 2). The task consisted of an 8-choice response time task in which participants were instructed to react to cues shown on a screen. Eight squares, corresponding spatially to the 8 fingers used to perform the task (i.e., no thumbs) were presented on the screen. The color of the outline of the squares alternated between red and green, indicating periods of rest or practice, respectively. During practice, one of the squares was filled green, and participants were instructed to press the corresponding key with the corresponding finger as quickly and as accurately as possible. The cues either followed a pseudorandom sequence or a deterministic 8-element sequence. In the latter case, participants were explicitly informed that the order of key presses followed a sequential pattern, but were not given any additional information such as what the sequence was or how many keys it was composed of. For the pseudorandom condition, each of the eight keys/fingers appeared once in every run of 8 consecutive presses, ensuring that the distribution of keys/fingers was the same as the sequence condition. Each practice block contained 64 key presses (corresponding to eight repetitions of the eight-element sequence or eight consecutive random patterns of eight key presses) and was separated by 15-second rest periods.

During the first experimental session, participants completed 4 blocks of pseudorandom SRTT, designed to assess motor execution independent of a sequence to be learned. The pseudorandom SRTT was followed by the 24 blocks of sequential SRTT. The sequential runs for the first experimental session required participants to learn a specific key sequence (Sequence 1; 4-7-3-8-6-2-5-1, in which 1 through 8 are the left pinky to the right pinky fingers; see Fig 1C). During the second experimental session, participants performed a new motor sequence (Sequence 2; 24 blocks) in order to examine the integration of new movements into the pre-existing motor sequence knowledge developed in Session 1. As the goal of the present study is to test whether post-learning sleep, as compared with wakefulness, is necessary for integration processes to take place, Sequence 2 was designed as per our previous research in order to optimize integration [7]. As in our previous research, the ordinal structure of Sequence 2 (4-7-2-8-6-3-5-1) was highly compatible with the previously learned sequence, as 75% of the elements were in the same ordinal positions as Sequence 1 (i.e., only elements 2 and 3 were in new ordinal positions; see Fig 1C). With this manipulation, 4 of the transitions in Sequence 2 were already learned in the previous session (i.e., 4–7, 8–6, 5–1, 1–4), and thus performance on these transitions reflects memory retention across the two experimental sessions. The other 4 transitions in Sequence 2 were not previously learned and thus were novel (i.e., 7–2, 2–8, 6–3, 3–5). Performance on these transitions then reflects integration of novel information into the cognitive-motor schema (corresponding to the ordinal structure of Sequence 1) acquired in the previous session [7]. The second experimental session ended with 4 blocks of the pseudorandom SRTT. Performance on these 4 blocks was compared to that on the 4 blocks of pseudo-random SRTT completed prior to sequence learning in Session 1 to assess across-sessions changes in task performance solely due to motoric improvement, i.e., unrelated to schema acquisition and integration. It should be noted that in both experimental sessions, the 24 blocks of the sequential SRTT were divided into two runs: a “training” run that included 20 practice blocks and a “test” run of 4 blocks. The test run was administered approximately 1 minute after the training run ended and was designed to assess end-of-training performance following the further dissipation of mental and physical fatigue [29].

For both pseudo-random and sequential variants of the SRTT, the primary outcome variables were: 1) response time (RT), i.e. time between cue presentation and subject key press, and 2) accuracy. Individual trials were excluded from analyses when the measured response time was greater than 3 standard deviations above or below the participant’s mean response time for that block. In Experiment 1 and 2, an average of 1.68% (Nap: 1.70%; No-Nap: 1.66%) and 1.71% (AM-PM: 1.72%; PM-AM: 1.70%) of trials were excluded from analyses, respectively. The averaged response times for correct movement transitions (i.e., both response n and n-1 were correct) and the percentage of correct transitions were computed for each block of practice on the task. For both measures and consistent with our previous work [7], averaging was done across all transitions (sessions 1 and 2) as well as on two subsets (Session 2 only): a) novel transitions (i.e., those transitions not performed in Session 1; 7–2, 2–8, 6–3 and 3–5); and b) learned transitions (i.e., those transitions that were also performed in Session 1; 4–7, 8–6, 5–1 and 1–4).

In each experimental session and following performance of the SRTT, participants performed a generation task in which they were asked to self-generate (i.e., without receiving visual cues) the sequence learned during that session. The purpose of the generation task was to assess sequence knowledge. The task consisted of 24 key presses, ideally corresponding to three 8-element sequences. Percentage of correct transitions and percentage of correct ordinal positions were extracted as the dependent measures of interest.

Experiment 1

Approximately one week prior to the first SRTT session, subjects were invited for a habituation session during which they completed a nap monitored with polysomnography (PSG; acquisition details described below). This allowed participants to habituate to sleeping in our laboratory while wearing the PSG equipment.

Participants returned to the laboratory to complete two SRTT sessions in the same day, separated by a delay of approximately 4 hours (Fig 1A). During the delay between the two sessions, participants were afforded a 90-minute nap opportunity (Nap group) or watched a movie with minimal sound volume and no subtitles for the same duration (No-Nap group). These nap and quiet wakefulness episodes were both monitored with PSG. Thus, participants completed experimental Session 1 in the late morning (10am-1pm), the Nap or No-Nap conditions in the early afternoon (12–3:30pm) and Session 2 took place 30min after the Nap/No-Nap episode to ensure dissipation of sleep inertia (2pm-5:30pm).

Both habituation and experimental nap recordings were completed in a sound-dampening sleep laboratory, monitored with a digital sleep recorder (V-Amp, Brain Products, Gilching, Germany; bandwidth: DC to Nyquist frequency) and digitized at a sampling rate of 1000 Hz. Standard electroencephalographic (EEG) recordings were made from Fz, C3, Cz, C4, Pz, Oz, A1 and A2 according to the international 10–20 system. As the habituation session served to acclimate participants to sleep recordings and these data were not considered in formal analyses (other than sleep scoring), channels Fz, Pz and Oz were omitted to accelerate the procedure. A2 was used as the recording reference. A1 served as a supplemental individual EEG channel that could be used as a back-up reference if necessary. An electrode placed on the middle of the forehead served as the recording ground. Bipolar vertical and horizontal eye movements (electrooculogram: EOG) were recorded from electrodes placed above and below the right eye and on the outer canthus of both eyes, respectively. Bipolar submental electromyogram (EMG) recordings were made from the chin. Electrical noise was filtered using a 50 Hz notch.

Polysomnographic data from both the nap and no-nap intervals were scored using an automated algorithm [30]. The algorithm uses two EEG channels (C3 and C4, referenced to the right and left mastoids, respectively), which are then averaged, and two EOG channels (left and right EOG, referenced to the opposite side mastoids). The resulting three-channel PSG data is then band-pass filtered between 0.3–45 Hz (EEG channel) or between 0.3–12 Hz (EOG channels), downsampled to 100Hz, and separated into 30s epochs. For each epoch, a fast Fourier transform is applied using a Hamming window of length 128, resulting in a spectrogram representing the time-frequency composition of the raw data. Each epoch is classified as belonging to a certain sleep stage according to a two-stage classification method: in the first stage, a deep convolutional neural network assigns sleep stage probabilities to the epoch based on its spectrogram; in the second stage, a multilayer perceptron generates revised sleep stage probabilities for the epoch by weighing the most probable sleep stage of the 5 preceding and succeeding epochs. The resulting most probable sleep stage is then selected as final score for that epoch. The automated sleep scores were then visually checked epoch by epoch by an experimenter. Where a mismatch arose between the experimenter’s and the algorithm’s score, the experimenter’s score, determined according to standard criteria [31], was retained for the epoch.

For the nap sessions, in addition to sleep scoring, we extracted sleep features known to be involved in motor memory processes. Specifically, sleep spindles and slow waves were automatically detected in NREM stages 2 and 3 sleep epochs using the YASA python package [32]. The spindle detection algorithm, based on previous work by Lacourse and colleagues [33], uses three detection methods: first, a Short-Time Fourier Transform performed on consecutive 2 sec epochs, with 200 ms overlap, to estimate the relative power in the sigma frequency band over the broadband EEG signal. Second, a sliding window (300 ms, step size 100 ms) is used to compute the Pearson correlation coefficient between the broadband EEG data and the spindle frequency band (12–15 Hz) signal. Third, the same sliding window is used to compute the root mean squared (RMS) within the defined spindle frequency band. A spindle is detected when two of the three following conditions are met: the spindle band–broadband correlation exceeds 0.65; the relative spindle band power equals or exceeds 0.2; the moving RMS exceeds the mean RMS + 1.5 standard deviations. Events longer than 2 sec or shorter than 0.5 sec are discarded. Events closer than 500 ms are merged into single spindles. The slow wave detection algorithm, based on previous work by Massimini et al. [34] and Carrier et al. [35], uses a Finite Impulse Response filter with transition band of 0.2 Hz to apply a band-pass filter between 0.3 and 1.5 Hz to the raw EEG data. Events are defined as slow waves if they meet the following criteria: duration of the negative peak between 0.3 and 1.5 sec, duration of the positive peak between 0.1 and 1 sec, amplitude of the negative peak between 40 and 300 μV, amplitude of the positive peak between 10 and 200 μV. After detection of negative and positive peaks, the algorithm finds the closest positive peak per each negative peak and computes peak-to-peak amplitude (PTP). Only PTPs between 75 and 500 μV are retained for analysis. Density and amplitude of spindles and slow waves were computed separately per each midline channel Fz, Cz, and Pz and then averaged across channels (but see S5 Table for channel-level data). Spindle and slow wave density were computed as the total number of spindle or slow wave events detected divided by the total time spent in sleep stages N2 and N3, per participant.

Experiment 2

The two experimental sessions were separated by a delay of approximately 12 hours (Fig 1B). Participants either performed the first session in the morning and the second in the evening of the same day (AM-PM group) or performed the first session in the evening and the second session in the morning of the following day (PM-AM group). All morning sessions were completed between 8.00 and 10.00 AM, and evening sessions between 8.00–10.00 PM.

During the delay between the two experimental sessions, participants followed their habitual daily schedule, which contained either a full night of sleep (PM-AM group) or 12h of continuous wake (AM-PM group). Participants of the PM-AM group were instructed to continue to follow the regular sleep schedule (i.e., bed/wake time in a 2h window around their habitual schedule and minimum of 7h of sleep), while participants of the AM-PM group were instructed not to nap between the two sessions. Compliance to these instructions was assessed in the two groups with the St Mary Hospital questionnaire.

Experimental sessions were completed at the participants’ own home, on their own computer, via a web-based platform [23, 24] that has been used in previous motor sequence learning research [36]. Prior to each experimental session, participants were instructed to remove all distractions from their environment such as computer or phone notifications, and to close the door to their room. To ensure participants’ compliance to experimental instructions, subjects were visually monitored through a videocall service. Participants were required to keep both their sound and video active throughout the entire session.

Statistical analyses

All analyses were performed using IBM SPSS Statistics for Windows, version 26 [37]. In case of violation of the sphericity assumption, we applied Greenhouse-Geisser corrections for ε≤0.75, Huynh-Feldt corrections for ε>0.75 [38].

For each experiment, group differences in vigilance [25, 26] prior to the experimental sessions were assessed with separate session (2 levels) x group (2 levels) ANOVAs for each experiment. Sleep quantity and quality [27] for the night prior to the first motor task session were compared between groups with an independent-samples t-test for each experiment. Performance measures from the pseudo-random variant of the SRTT, reflecting general motor execution, were analyzed with separate group (2) x block (4) ANOVAS for each session. Performance on the sequential SRTT runs from the first session was assessed with group x block ANOVAs separately for each experiment, run (training (20 blocks) / test (4 blocks)) and dependent measure (RT / % correct). Last, performance on the generation task from Session 1 was analyzed with separate independent samples t-tests for the dependent measures of percentage of correct transitions generated and percentage of correct ordinal positions generated.

The primary, confirmatory data analyses aimed to test whether post-learning sleep resulted in enhanced memory integration compared with wakefulness (i.e., Nap vs. No Nap for Experiment 1 and PM-AM vs. AM-PM for Experiment 2). Separate two-way repeated-measures ANOVAs with the between-subject factor group and the within-subject factor block were conducted on the Session 2 sequential training and test runs to assess the effect of sleep on task performance. The analysis was performed on all movement transitions, as well as on two transitions subsets: a) novel transitions (investigating integration processes); and b) learned transitions (investigating memory retention processes). Similarly, two-way, repeated-measures ANOVAs with the between-subject factor group and the within-subject factor transition type (novel/learned) were conducted on response time and % correct transitions from the Session 2 post-training test run. This analysis aimed at comparing performance related to memory integration and retention between the two groups after extensive practice on the task (i.e., after the Session 2 training run; see [7]).

Additionally, for Experiment 1 only, we examined potential relationships between behavior and sleep features extracted from the nap period. Pearson correlations were performed between the extracted NREM sleep features (i.e., density and amplitude of spindles and slow waves) and a performance index (PI) on novel and learned transitions in Session 2. The PI is an aggregate speed-accuracy measure used extensively in previous work [39–41] and is computed with the following formula: (1)

Where x = blocks of trials; SeqDur = sequence duration (i.e., reflecting movement speed); errors = proportion of incorrect transitions.

PI was averaged across the 20 training blocks of Session 2 and normalized by the individual’s average PI on the 4 blocks of pseudorandom SRTT completed during Session 2 (i.e., reflecting motor execution on Session 2 independent of a sequence to be learned). As outlined in our pre-registration, this speed-accuracy measure was used solely for conducting correlations with relevant sleep features to reduce the number of correlations conducted (i.e., as opposed to correlating with speed and accuracy measures separately). For completeness, however, exploratory correlational analyses with separate speed and accuracy measures can be found in the supplement.

Experiment 1: Effect of diurnal sleep

Participant characteristics, vigilance and general motor performance.

Participant characteristics, sleep quality, quantity and duration for the night(s) preceding the experimental session, as well as subjective and objective measures of vigilance are provided in Table 1. Results from the corresponding statistical analyses can be found in the S1 Table. Performance on the pseudorandom SRTT, used to assess general motor execution prior to and following the sequential SRTT during sessions 1 and 2, respectively, is depicted in S1 Fig and results of the corresponding analysis are summarized S2 Table. In brief, and as expected, participants in the two experimental groups did not differ with regards to any of the reported measures (e.g., general characteristics, sleep prior to learning, vigilance and general motor execution).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Participant demographics and sleep and vigilance scores for each experimental group.

https://doi.org/10.1371/journal.pone.0280591.t001

Sleep during Nap / wake interval.

A detailed characterization of sleep characteristics and features during the 90-minute nap opportunity is provided in Table 2. Briefly, participants in the nap group slept on average 58.0 ± 19.5 minutes and presented a sleep latency of 15.5 ± 14.0 minutes. All participants reached stage Non-REM 2 (N2) and maintained it for an average of 33.0 ± 14.5 minutes. Participants in the No-Nap group remained awake during the 90-minute interval, with an averaged sleep duration of 0.0 ± 0.5 minutes (maximum = 1.0 min).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Average sleep characteristics for participants in the Nap group (Experiment 1).

https://doi.org/10.1371/journal.pone.0280591.t002

Motor task performance.

Performance on all transitions of the sequential SRTT is depicted in Fig 2 and the output of the corresponding statistical analyses is presented in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Performance on all transitions in Experiment 1.

Mean response time (in seconds) and % correct transitions per block of sequence task (20 blocks of training, 4 blocks of test administered directly after training) are shown separately for the Nap group (blue, N = 25) and No-Nap group (orange, N = 25). The experimental manipulation (nap/no-nap) took place between Session 1 and Session 2. Shaded areas represent the standard error of the mean (SEM).

https://doi.org/10.1371/journal.pone.0280591.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Sequential SRTT performance in Experiment 1.

https://doi.org/10.1371/journal.pone.0280591.t003

In Session 1 of the sequential SRTT (i.e., prior to the experimental manipulation), participants successfully learned the motor sequence, as evidenced by the main effect of block for the dependent measure response time (Fig 2, Session 1; Table 3A). These improvements in speed did not come at the expense of decreased accuracy, as the percentage of correct transitions remained stable across practice blocks. As expected, there were no differences between the Nap and No-Nap groups during initial learning, as evidenced by the lack of significant group effects as well as group x block interactions for both speed and accuracy measures. A nearly identical pattern of results was observed in the short test administered immediately after the Session 1 training run. Namely, RT decreased across blocks whereas accuracy remained high and stable. Most importantly, there were no differences between groups for any of the performance measures. These results collectively demonstrate that motor learning did not statistically differ between groups prior to the experimental manipulation in Session 1.

To determine whether post-learning sleep facilitated performance on the new sequence in Session 2, we assessed group differences in behavioral performance on the post-sleep/wake SRTT sequence task (Fig 2, Session 2; Table 3A). Analogous to Session 1, a significant main effect of block on response time was observed, indicating successful learning of the second motor sequence, while accuracy remained high and stable. The two experimental groups did not differ on either RT or accuracy in the Session 2 training run, as evidenced by the lack of group and group by block effects. Similarly, there were no group differences in the short post-training test run for both dependent measures.

The sequence learned in Session 2 contained movement transitions that had been previously learned in Session 1 (learned transitions) as well as transitions that were new (novel transitions). We repeated our analyses on these two subsets of movement transitions separately to investigate memory retention and integration processes, respectively (Fig 3; Table 3B, 3C). Results were nearly identical to those obtained on the global analyses of all transitions detailed above. Significant block effects were observed for RT on both learned and novel transitions. Note there was a significant (p = 0.048) effect of block on accuracy for novel transitions only, possibly indicating a decrease in performance difficulty with practice on the novel transitions during learning of Sequence 2. However, there were no significant differences between the two experimental groups in the training or test runs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Performance on novel and learned transitions in Session 2 of Experiment 1.

Mean response time (in seconds) and % correct transitions per block (20 blocks of training, 4 blocks of test administered directly after training) are shown separately for the Nap group (blue, N = 25) and No-Nap group (orange, N = 25). Shaded areas represent the SEM.

https://doi.org/10.1371/journal.pone.0280591.g003

Last, repeated-measures ANOVAs with the between-subject factor group and the within-subject factor transition type (novel vs. learned) conducted on the Session 2 test revealed no group differences in performance on the two transition types after extensive task practice (response time–main effect transition type: F(1,48) = 0.14, p = 0.71, η² = 0.003; main effect group: F(1,48) = 1.65, p = 0.21, η² = 0.033; interaction: F(1,48) = 0.13, p = 0.72, η² = 0.003; accuracy–main effect transition type: F(1,48) = 2.58, p = 0.12, η² = 0.051; main effect group: F(1,48) = 0.45, p = 0.51, η² = 0.009; interaction: F(1,48) = 0.04, p = 0.85, η² = 0.001).

It is worth highlighting that the results presented above (see Table 3) did indicate trends for significant group differences in response time during the Session 2 training runs for all, learned and novel transitions (all p = 0.07). Specifically, and consistent with our hypotheses, RT was faster for the Nap group as compared to No-Nap. A visual inspection of the data, however, suggests that at least a portion of this marginally significant difference can be explained by non-significant group differences during Session 1 (i.e., prior to the Nap / No-Nap experimental manipulation). To examine this possibility, we conducted exploratory, follow-up analyses on RT for the Session 2 training run but normalized to RT during the Session 1 test run. The normalization was computed for each individual by dividing the response time of each block of Session 2 by the response time averaged across the 4 blocks of Session 1 test. This allowed to neutralize any effect on Session 2 data of spurious group differences in RT present prior to the experimental manipulation. These analyses revealed that after normalization, the trends for significant group differences in RT during Session 2 training were no longer present (see S2 Fig).

These results collectively demonstrate that a post-learning nap did not significantly influence schema-mediated motor memory consolidation processes.

Experiment 2: Effect of nocturnal sleep

Motor task performance.

Performance on all transitions of the sequential SRTT is depicted in Fig 4 and the output of the corresponding statistical analyses is presented in Table 4. Response time during performance of the Session 1 sequential SRTT (Fig 4, Session 1; Table 4A) significantly improved across blocks, demonstrating successful learning of the motor sequence, but did not differ between the two experimental groups. An analogous ANOVA conducted on the Session 1 post-training test detected no significant effect of block nor group differences in response time, indicating that both groups reached a stable and comparable performance post-training. Accuracy remained high and stable during the training run, but there was a significant effect of block on accuracy in Session 1 test (all transitions), caused by decreased accuracy on the last test block (average 89%, compared with 91% in the first three blocks). However, there were no differences in accuracy across groups. These results also collectively indicate that there were no time-of-day effects on performance in Session 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Performance on all transitions in Experiment 2.

Mean response time (in seconds) and % correct transitions per block of sequence task are shown separately for the PM-AM (blue, N = 28) and AM-PM (orange, N = 29) groups. Shaded areas represent the SEM.

https://doi.org/10.1371/journal.pone.0280591.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Sequential SRTT performance in Experiment 2.

https://doi.org/10.1371/journal.pone.0280591.t004

Using analogous analyses to those performed on Experiment 1, we compared post-sleep/wake performance to determine whether nocturnal sleep facilitated learning of the new motor sequence. During the Session 2 training run, we observed no significant group differences in either response time or accuracy, nor were there significant block × group interactions (see Fig 4, Session 2). Response time significantly improved across blocks, an effect that was not observed for accuracy. Similar analyses repeated on the test run revealed no effect of block, group, or their interaction for both response time and accuracy. We repeated analyses on the two subsets of learned and novel movement transitions and again observed no group differences (Fig 5, see Table 4B, 4C for detailed statistics). Collectively, these results suggest that the two experimental groups successfully learned the second motor sequence, with no differences in performance between groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Performance on novel and learned transitions in Experiment 2, Session 2.

Mean response time (in seconds) and % correct transitions per block of sequence task are shown separately for the PM-AM (blue, N = 28) and AM-PM (orange, N = 29) groups. Shaded areas represent the SEM.

https://doi.org/10.1371/journal.pone.0280591.g005

The group x transition type (novel vs. learned) ANOVA conducted on the Session 2 test revealed no group differences, but a trend for a significant effect of transition type on response time (main effect transition type: F(1,55) = 3.19, p = 0.08, η² = 0.055; main effect group: F(1,55) = 0.01, p = 0.931, η²<0.001; interaction: F(1,55) = 0.61, p = 0.44, η² = 0.011) and a highly significant effect of transition type on accuracy (main effect transition type: F(1,55) = 26.24, p<0.001, η² = 0.323; main effect group: F(1,55) = 1.64, p = 0.21, η² = 0.029; interaction: F(1,55) = 0.25, p = 0.62, η² = 0.005). These effects were driven by reduced response time and increased accuracy for learned transitions (RT–learned: 385 ± 90ms, novel: 396 ± 99ms; accuracy–learned: 94 ± 2%, novel: 92 ± 4%).

Consistent with Experiment 1, the results of the second experiment did not reveal any beneficial effect of sleep on the integration of novel motor information into a pre-existing schema.