Sleep Loss Disrupts The Neural Signature Of Successful Learning Part 3

Data analyses

Behavior

Behavioral data were analyzed using R Studio (v.1.4.1717, R Studio Team 2021). Memory consolidation was indexed by the change in visuospatial memory accuracy between the immediate and delayed tests.

Delayed testing is a test method often used to test memory and learning effects. Its basic principle is that after a certain learning and memory task is completed, the test is delayed for some time to observe people's long-term memory and mastery of the learned knowledge.

Research shows that delayed testing can help us consolidate memories and improve learning. We usually learn and memorize over and over again when learning new knowledge, but delayed testing can help us memorize and recall more effectively, helping to deepen our understanding and memory of the knowledge we have learned. This deep memory is more likely to help us quickly recover and recall the information we need in the future.

Through delayed testing, we can discover blind spots in our learning, find out the content that is most difficult for us to remember, and study and re-master it in a targeted manner to more comprehensively master the required knowledge.

But we need to note that delay testing is not completely effective. It is mainly used to consolidate the knowledge we have already mastered, and will not help us learn concepts or knowledge points. Therefore, when learning new knowledge, we also need to take more notes and apply the knowledge we have learned to real life to deepen our understanding and memory of the knowledge.

In short, delayed testing is of great help to our learning and memory. It allows us to better consolidate knowledge memory and learn and master the required knowledge more effectively. Let us actively use delayed testing to improve our learning results and better master the required knowledge and skills during the learning process. It can be seen that we need to improve memory, and Cistanche deserticola can significantly improve memory because Cistanche deserticola is a traditional Chinese medicinal material that has many unique effects, one of which is to improve memory. The efficacy of minced meat comes from the various active ingredients it contains, including acid, polysaccharides, flavonoids, etc. These ingredients can promote brain health in a variety of ways.

Click know supplements to improve memory

For each participant and test, we computed an error score for each image by calculating the distance (cm) between the recalled location (image center) and the location where the image had appeared at passive viewing. We derived a retention index (RI) by subtracting the error score at the delayed test from the error score at the immediate test for each image and then averaging across images. 

A follow-up RI was calculated between the immediate and follow-up tests using the same method. To ease understanding (e.g. higher RI = better retention), we swapped the order of the RI subtraction to that which was preregistered. 

This change yields statistically identical results aside from the sign change. As preregistered, 1 participant was removed from analyses that included RISleepBenefit scores (see below) because their RI at the delayed test in the sleep deprivation condition was >3 SD from the mean.

Next-day learning was assessed by the learning index (LI), which equated to the percentage of correctly recognized images on the paired associates test. Between condition differences in RI and LI were analyzed using paired-sample t-tests with a significance threshold of P < 0.05. We report the "classical" Cohen's d as our effect size estimate because it is unaffected by experimental design and thus facilitates comparisons across different studies (R function: cohen, R package: LSR, Navarro 2015).

One of our primary aims was to investigate the relationship between sleep-associated consolidation and next-day learning and how SWA contributes to this relationship. To do this, we first quantified the benefit of sleep (vs. sleep deprivation) on the RI and LI. We subtracted (for each participant) the RI in the sleep deprivation condition from the RI in the sleep condition to derive a RISleepBenefit. 

Similarly, we subtracted (for each participant) the LI in the sleep deprivation condition from the LI in the sleep condition to obtain a LISleepBenefit. Positive scores on the RISleepBenefit and LISleepBenefit therefore indicate a sleep-associated improvement in performance. RISleepBenefit and SWA (see below) were entered as predictors of LISleepBenefit in a forced-entry multiple regression analysis. 

A Bayesian multiple regression analysis (R package: BayesFactor, Morey, and Rouder 2018) was used to test for evidence of the null (i.e. no relationship between sleep-associated consolidation [RISleepBenefit], SWA, and next-day learning [LISleepBenefit]). Exploratory correlations were computed using Pearson's R.

EEG (sleep)

Preprocessing

Sleep EEG data were partitioned into 30 s epochs and were scored in RemLogic 3.4 according to standardized criteria (Iber 2007). Epochs scored as sleep-stage N2 or SWS were exported to MATLAB 2019a using the FieldTrip toolbox (Oostenveld et al. 2011, v.10/04/18) for further analysis. Artifacts were identified and removed using FieldTrip's Databrowser (mean ± SD artifacts rejected, 3.5 ± 2.85), noisy channels were removed (4 channels across 4 participants), and 2 entire datasets were excluded due to excessive noise. The remaining data were band-pass filtered between 0.3 Hz and 30 Hz using Butterworth low-pass and high-pass filters.

Power spectral analysis

Due to significant noise in the occipital channels (as a result of electrodes detaching during the night in several participants), we only included frontal (F3 and F4), central (C3 and C4), and parietal (P3 and P4) channels in our spectral analysis of the sleep EEG data. Using functions from the FieldTrip toolbox, artifact-free N2, and SWS epochs were applied to a Fast Fourier Transformation with a 10.24-s Hanning window and 50% overlap. EEG power in the SWA (0.5–4 Hz) and fast spindle (12.1– 16 Hz) bands was determined by averaging across the corresponding frequency bins and channels.

EEG (learning)

Preprocessing

All 8 EEG channels (F3, F4, C3, C4, P3, P4, O1, and O2) were included in our analysis of learning. Data were highpass filtered (0.5 Hz), notch-filtered (49–51 Hz), and segmented into trials (−3 s to 4.5 s around stimulus onset). Trials for which participants did not respond were removed from the analysis (mean ± SD excluded trials, sleep: 0.1 ± 0.45, sleep deprivation: 5.11 ± 7.93, Priest et al. 2001). 

From scalp electrodes, eye blinks and cardiac components were identified and were removed using an independent components analysis, and noisy channels were interpolated via a weighted average of their nearest neighbors (14 channels across 6 participants and 2 conditions). Trials were visually inspected and data from 2 participants were removed due to excessive noise in multiple channels.

Time-frequency analyses

Time-frequency representations (TFRs) were calculated separately for lower (4–30 Hz) and higher frequencies (30– 60 Hz). Our preregistered upper bound was 120 Hz, but because our sampling rate was 200 Hz, the upper bound was above the Nyquist frequency and had to be lowered. For lower frequencies, data were convolved with a 5- 5-cycle Hanning taper in 0.5-Hz frequency steps and 5-ms time steps using an adaptive window length (i.e. where window length decreases with increasing frequency, e.g. 1.25 s at 4 Hz and 1 s at 5 Hz, to retain 5 cycles). 

For higher frequencies, data were convolved with tapers of the Slepian sequence (3 tapers), also in steps of 0.5 Hz and 5 ms with an adaptive window length. For this latter analysis, frequency smoothing was set to 0.4 of the frequency of interest (e.g. 20 Hz smoothing at 50 Hz). Artifact rejection was achieved via a data-driven approach applied separately to the analyses of lower and higher frequencies: power values that exceeded the 85th percentile across all time/frequency points and trials were removed from each participant's dataset.
TFRs were converted into percent power change relative to a −400 ms to −200 ms prestimulus baseline window. This window was chosen to mitigate baseline contamination by poststimulus activity while preserving proximity to stimulus onset (note that our poststimulus time window of interest started at 0.3 s, see below). 

Trials were divided into subsequently remembered and forgotten adjective-image pairings (based on the test phase 48 hours later). Because our 49–51 Hz notch filter overlapped with our gamma frequency range, we reran our higher frequency analysis (30–60 Hz) without a notch filter and the results in the gamma frequency range (40–60 Hz) were unchanged.

Statistics

TFR analyses were performed as dependent sample analyses and were corrected for multiple comparisons using FieldTrip's nonparametric cluster-based permutation method (1,000 randomizations). Clusters were defined by channel ∗ time while averaging across the frequency bands of interest (theta [4–8 Hz], alpha [8– 12 Hz], beta [12–20 Hz], and gamma [40–60 Hz], cluster threshold P < 0.05). 

Preregistered analyses in theta and gamma bands were 1-tailed, whereas exploratory analyses were 2-tailed. To reduce interference from early visual-evoked responses, the time window of interest was set from 0.3 s to 2 s (Friedman and Johnson 2000; Osipova et al. 2006). A factorial approach was used to assess the impacts of sleep deprivation (vs. sleep) on the neural correlates of encoding: We calculated the grand average TFR difference for subsequently remembered > forgotten adjective-image pairings within each condition (sleep and sleep deprivation) and then entered these contrasts into the cluster-based permutation analysis (Sleepremembered > forgotten > Sleep Deprivationremembered > forgotten). 

To reflect the rationale of the cluster-based permutation test, we report effect sizes as Cohen's dz based on the average of the largest cluster (i.e. averaging across all channels and time points that contributed at any point to the largest cluster, Meyer et al. 2021).

Results

Sleep benefits memory consolidation

To assess overnight consolidation, we computed an RI from the immediate and delayed visuospatial memory tests (higher RI = better overnight retention, see Materials and methods). As expected, the RI was significantly higher after sleep than sleep deprivation (t(28) = 3.78, P < 0.001, d = 0.71, see Fig. 2a). To ensure that our findings were not driven by between-condition differences in fatigue at the delayed test, we also assessed memory retention between the immediate and follow-up test (which took place 48 h after the delayed test, thereby allowing for recovery sleep). 

As expected, the RI was still significantly higher in the sleep (vs. sleep deprivation) condition (t(28) = 2.18, P = 0.038, d = 0.44, see Fig. 2b), suggesting that sleep had facilitated overnight consolidation. There was no significant between-condition difference in visuospatial accuracy at the immediate test (mean ± standard error of the mean (SEM), sleep: 2.44 ± 0.10, sleep deprivation: 2.57 ± 0.10, t(28) = −0.98, P = 0.337, d = 0.19, BF01 = 3.28) and there was no difference in the benefit of sleep on retention (RISleepBenefit [i.e. sleep condition RI − sleep deprivation condition RI], see below) between participants who completed the sleep condition before or after the sleep deprivation condition (t(27) = 0.22, P = 0.828, d = 0.08, BF01 = 2.81).

Although response times on the PVT were slower in the morning after sleep deprivation (mean ± SEM, 399.00 ms ± 17.63) than sleep (289.15 ± 4.34, P < 0.001), there was no significant relationship between RISleepBenefit and PVTSleepBenefit (i.e. mean RT after sleep − mean RT after sleep deprivation, R2 = −0.15, P = 0.440, BF01 = 1.92). Similarly, as indicated by the SSS, participants felt less alert after sleep deprivation (mean ± SEM, 5.37 ± 0.15) than after sleep (2.27 ± 0.16). However, there was no significant correlation between RISleepBenefit and SSSSleepBenefit (i.e. mean rating after sleep − mean rating after sleep deprivation, R2 < −0.01, P = 0.991, BF01 = 2.46). An extended analysis of the PVT and SSS data is available in the Supplementary Material.
Sleep improves next-day learning

To assess encoding performance after sleep or sleep deprivation, we calculated an LI, which equated to the percentage of correctly recognized images on the paired associates test (this took place 48 h after encoding, following recovery sleep). As expected, encoding performance was significantly higher after sleep than sleep deprivation (t(29) = 12.19, P < 0.001, d = 2.17, see Fig. 2c), suggesting that sleep had benefited next-day learning. There was no significant difference in the benefit of sleep on new learning (LISleepBenefit; i.e. sleep condition LI − sleep deprivation condition LI, see below) between participants who completed the sleep condition before or after the sleep deprivation condition (t(28) = 0.37, P = 0.712, d = 0.14, BF01 = 2.75).

There was no significant relationship between LISleepBenefit and PVTSleepBenefit (R2 = −0.30, P = 0.113), although the evidence for the null remained inconclusive (BF01 = 0.86). Similarly, there was no significant correlation between LISleepBenefit and SSSSleepBenefit (R2 = −0.35, P = 0.056) with inconclusive evidence for the null (BF01 = 0.53).

No relationship between sleep-associated consolidation, slow wave activity, and next-day learning

Next, we tested the hypothesis that overnight consolidation predicts next-day learning and that SWA contributes to this relationship. Because we aimed to target the relationship between sleep-associated memory processing and next-day learning, it was necessary to first quantify the positive impact of sleep (vs. sleep deprivation) on the RI and LI. We therefore subtracted both the RI and LI between the sleep and sleep deprivation conditions (on a participant-by-participant basis) such that positive scores on the resultant RISleepBenefit and LISleepBenefit metrics indicated a sleep-associated improvement in performance. SWA was defined as EEG power within the 0.5–4 Hz frequency band during sleep stages N2 and SWS (collapsed across all EEG channels). In a multiple regression model, we entered RISleepBenefit and SWA as predictors and LISleepBenefit as the outcome variable. Contrary to expectations, sleep-associated consolidation (RISleepBenefit) and SWA did not significantly account for next-day learning (LISleepBenefit, F(2, 24) = 1.51, R2 = 0.11, P = 0.242, see Fig. 3). No significant relationship was observed between RISleepBenefit and LISleepBenefit independently of SWA (B = 3.30, t(24) = 0.86, P = 0.399) or between SWA and LISleepBenefit independently of RISleepBenefit (B = −0.51, t(24) = −1.65, P = 0.111). RISleepBenefit did not significantly correlate with SWA (R2 = 0.21, P = 0.298). A follow-up Bayesian analysis revealed anecdotal evidence in support of the null (i.e. that sleep-associated consolidation and SWA did not account for next-day learning, BF01 = 2.04).

In a subsidiary analysis, we repeated this multiple regression but only entered data from the sleep condition into our model (i.e. the RISleepBenefit and LISleepBenefit were replaced with the RI and LI from the sleep condition alone). Our findings mirrored those of the foregoing analysis: sleep-associated consolidation (RI) and SWA did not significantly account for next-day learning (LI, F(2, 25) = 1.83, P = 0.181, R2 = 0.13, BF01 = 1.68). There was also no significant relationship between RI and LI independently of SWA (B = 4.46, t(25) = 1.67, P = 0.107) or between SWA and LI independently of RI (B = −0.25, t(25) = −1.16, P = 0.256), and no significant correlation was observed between RI and SWA (R2 = 0.28, P = 0.143).

We also explored whether RI in the sleep condition was correlated with other sleep parameters previously implicated in declarative memory consolidation: time (min) in SWS (Backhaus et al. 2006; Scullin 2013) and fast spindle power (12.1–16 Hz, Cox et al. 2012; Tamminen et al. 2010). However, no significant relationships emerged (all P > 0.368). Sleep data are available in Table 1.

Sleep deprivation disrupts beta desynchronization during successful learning

Finally, we tested the hypothesis that sleep deprivation disrupts theta and gamma synchronization at learning. However, no significant differences were observed in the theta (4–8 Hz) or gamma (40–60 Hz) bands when comparing TFRs between subsequently remembered and forgotten adjective-image pairings or between the sleep and sleep deprivation conditions, and there was no significant interaction between these contrasts (all P > 0.05).

In an exploratory analysis, we investigated the effect of sleep deprivation on beta desynchronization, an established marker of successful learning (Hanslmayr et al. 2009, 2011, 2012, 2014; Griffiths et al. 2016). Consistent with these previous findings, an overall reduction in beta power was observed during encoding of subsequently remembered (vs. forgotten) adjective-image pairings when combining the sleep and sleep deprivation conditions (corresponding to 2 clusters in the left hemisphere beginning at ∼1.5–1.7 s (P = 0.044, d = −0.66) and at ∼1.75–1.9 s (P = 0.038, d = −0.49) after stimulus onset (see Fig. 4a).

Interestingly, changes in beta power accompanying successful learning were significantly different in the sleep and sleep deprivation conditions (interaction, corresponding to a cluster in the left hemisphere at ∼0.5–0.7 s, P = 0.014, d = −0.33, see Fig. 4b and d). Whereas encoding of subsequently remembered (vs. forgotten) adjective-image pairings was associated with a downregulation of beta power after sleep (P = 0.005), an apparent upregulation of beta power emerged from the same contrast after sleep deprivation (P = 0.019, see Fig. 4c, although this latter post hoc test did not survive Bonferroni correction, alpha = 0.0125). Moreover, beta power was significantly reduced during the encoding of subsequently remembered pairings in the sleep (vs. sleep deprivation) condition (P = 0.001), but no such difference was observed during the encoding of subsequently forgotten pairings (P = 0.928).

To explore whether this significant interaction was driven by increased fatigue in the sleep deprivation (vs. sleep) condition, we correlated (for each participant) average beta power for the contrast Sleepremembered > forgotten > Sleep Deprivationremembered > forgotten (within the significant group-level cluster) with SSSSleepBenefit and PVTSleepBenefit. No significant relationships were observed (SSS: R2 = −0.20, P = 0.311, BF01 = 1.58, PVT: R2 = 0.18, P = 0.371, BF01 = 1.73), suggesting that the foregoing findings did not arise from between-condition differences in fatigue.

An overall reduction in beta power was also observed for the sleep (vs. sleep deprivation) condition when combining subsequently remembered and forgotten adjective-image pairings corresponding to 2 clusters in the left (∼1–1.5 s, P = 0.014, d = −0.63) and right hemisphere (∼1.3–1.7 s, P = 0.038, d = −0.47). Given the previously reported links between alpha (8– 12 Hz) desynchronization and successful learning (Griffiths et al. 2016; Weisz et al. 2020), we also explored activity in this frequency band (same contrasts as above), but no significant effects were observed (all P > 0.05).

Actigraphy

Hours slept during the 48-h interval between the delayed and follow-up tests (as estimated via wristwatch actigraphy) were applied to a 2 (Condition: Sleep/Sleep Deprivation) ∗ 2 (Night: 1/2) repeated-measures ANOVA (R function: anova_test, R package: rstatix). Note that 2 participants were not included in this analysis due to technical problems with the actigraphy device. There was a main effect of Night (F(1, 27) = 62.47, P < 0.001, ηp 2 = 0.70), indicating that all participants slept for longer on night 1 than on night 2. A significant Condition ∗ Night interaction (F(1, 27) = 14.21, P < 0.001, ηp 2 = 0.35) also emerged, with Bonferroni-corrected post hoc tests indicating that sleep duration was longer in the sleep deprivation (vs. sleep) condition on night 1 (mean ± SEM hours sleep, sleep deprivation: 9.03 ± 0.45, sleep: 7.52 ± 0.24, P = 0.006) but shorter on night 2 (sleep deprivation: 5.33 ± 0.23, sleep: 6.00 ± 0.25, P = 0.036). There was no main effect of Condition (F(1, 27) = 3.07, P = 0.091, ηp 2 = 0.10).

It is possible that the longer duration of sleep on the first night after learning in the sleep deprivation (vs. sleep) condition augmented the consolidation of newly learned adjective-image pairings, potentially mitigating the initial impact of sleep loss on encoding. To test this possibility, we correlated the between-condition difference in sleep duration on the first night after learning (sleep deprivation condition − sleep condition) with the LISleepBenefit. However, no significant relationship emerged (R2 = −0.06, P = 0.756, BF01 = 2.33).

Discussion

Sleep provides a benefit over wake for retaining memories and also for learning new ones (Gais et al. 2006; Yoo et al. 2007; Talamini et al. 2008; Payne et al. 2012; Alberca-Reina et al. 2014; Kaida et al. 2015; Durrant et al. 2016; Tempesta et al. 2016; Cairney, Lindsay, et al. 2018; Cousins et al. 2018; Gaskell et al. 2018; Ashton et al. 2020; Ashton and Cairney 2021). Some suggest that these benefits can be explained by an active role of SWS and associated neural oscillations in shifting the memory retrieval network from the hippocampus to the neocortex and thus restoring the hippocampal capacity for new learning (Walker 2009; Born and Wilhelm 2012; Rasch and Born 2013; Klinzing et al. 2019). In the current study, we tested the hypothesis that the extent to which individuals consolidate new memories during sleep predicts their ability to encode new information the following day and that SWA contributes to this relationship. Although we observed a benefit of sleep (relative to sleep deprivation) on our measures of overnight consolidation and next-day learning, we found no evidence of a relationship between the 2 measures or with SWA.

Given the importance of sleep for new learning, we further sought to understand how sleep deprivation affects the neural correlates of successful encoding. Interestingly, whereas learning of subsequently remembered (vs. forgotten) associations was associated with a downregulation of 12–20 Hz beta power after sleep (as reported in previous work, Hanslmayr et al. 2009, 2011, 2012, 2014; Griffiths et al. 2019), no significant difference in beta power emerged after sleep deprivation. These findings suggest that an absence of sleep disrupts the neural operations underpinning memory encoding, leading to suboptimal performance.

Sleep benefits overnight consolidation and next-day learning

Previous work has shown that sleep supports memory consolidation (Gais et al. 2006; Talamini et al. 2008; Payne et al. 2012; Durrant et al. 2016; Cairney, Lindsay, et al. 2018; Gaskell et al. 2018; Ashton et al. 2020; Ashton and Cairney 2021) and subsequent learning (McDermott et al. 2003; Yoo et al. 2007; Kaida et al. 2015; Tempesta et al. 2016; Cousins et al. 2018). In keeping with these studies, we found that memory retention and next-day learning benefited from overnight sleep relative to sleep deprivation.

Although this was a preregistered investigation of sleep's role in learning and memory and was motivated by prior work on the same topic (Gais et al. 2006; Yoo et al. 2007), it is important to consider the extent to which our findings can disentangle the memory effects of sleep from the disruptive influences of sleep deprivation. Extended periods of wakefulness give rise to various cognitive impairments (Krause et al. 2017), meaning that poorer performance in the sleep deprivation (vs. sleep) condition could reflect the indirect consequences of sleep loss rather than a direct absence of sleep (indeed, participants in the current study were slower to respond on the PVT and reported being less alert on the SSS after sleep deprivation than sleep). Focusing first on our assessment of overnight consolidation, generalized cognitive impairments arising from sleep deprivation could have impaired retrieval performance, creating the impression of a sleep-associated improvement in retention. 

While this is a reasonable concern given the sleep–memory effects observed at our delayed test (which followed immediately after the overnight interval), it does not explain why the retention advantage in the sleep condition was still present 48 h later (once sleep-deprived individuals had had ample opportunity for recovery sleep). Moreover, we observed no significant relationship between the benefits of sleep (vs. sleep deprivation) on memory retention and between conditions differences in SSS scores or PVT response times, suggesting that our findings were not driven by the general cognitive impairments that accompany sleep deprivation. It is therefore reasonable to conclude that our data reflect a positive impact of sleep on memory consolidation. To what extent this memory benefit of sleep can be explained by an absence of wakeful interference (such as that experienced in the sleep deprivation condition) or an active sleep-dependent consolidation mechanism, however, cannot be inferred from our data.

Turning to our analysis of next-day learning, although the assessment phase also took place 48 hours after encoding, the initial learning phase occurred immediately after sleep or sleep deprivation. We therefore cannot rule out the possibility that the apparent improvement in encoding performance after sleep was influenced by generalized cognitive impairments following sleep deprivation. Importantly, however, we think that our EEG data provide reasonable evidence that an absence of sleep does in itself disrupt new learning. Specifically, if our effects were driven by nonspecific cognitive deficits following sleep deprivation, one would expect to have observed only generalized differences in EEG activity between the sleep and sleep deprivation conditions (i.e. only a main effect of condition across all encoding trials). By contrast, a significant interaction showed that beta desynchronization was amplified in the sleep (vs. sleep deprivation) condition, specifically on trials for which adjective-image pairings were subsequently remembered. This impact of sleep on beta desynchronization during successful learning was not predicted by between-condition differences in SSS scores or PVT response times, and no between-condition difference in beta power emerged for pairings that were subsequently forgotten (see Fig. 4b and c). Because beta desynchronization is an established neural marker of semantic processing during successful learning (Hanslmayr et al. 2011, 2014; Griffiths et al. 2016), these findings may suggest that the neural mechanisms of encoding are indeed disrupted by an absence of sleep. Further support for this view is available below, where we outline how the brain may engage in compensatory learning strategies when semantic processing pathways are compromised by sleep deprivation (see "Sleep loss disrupts effective learning").

Because our RI was based on tests for the same items at the immediate, delayed, and follow-up sessions, it is possible that our data were influenced by retrieval practice effects (i.e. memories that undergo retrieval practice are typically better remembered than those that do not, Roediger and Karpicke 2006; Carpenter et al. 2008). That is, the retention advantage observed after sleep (vs. sleep deprivation) at the delayed test might have been maintained at the follow-up test as a result of retrieval practice. However, given that memories strengthened through retrieval gain little benefit from sleep-associated consolidation (Bäuml et al. 2014; Antony et al. 2017; Antony and Paller 2018), then, under a retrieval practice hypothesis, the immediate test should have nullified any later impact of sleep on retention. While it may still be argued that a between-condition difference in retention at the delayed test was driven by nonspecific impairments following sleep deprivation, this would not explain why the memory advantage in the sleep condition was still present 48 hours later (once recovery sleep had taken place). We therefore think that retrieval practice effects cannot provide a reasonable explanation of our findings.

Given that recovery sleep after sleep deprivation is characterized by a homeostatic increase in SWS (Borbély 1982; Borbély et al. 2016), one might have expected the overnight consolidation of newly learned adjective-image pairings to be amplified in the sleep deprivation (vs. sleep) condition, potentially tempering the initial impact of sleep loss on encoding. Although we did not record sleep EEG during the time that participants were away from the laboratory (and thus have no insight into homeostatic increases in SWS after sleep deprivation), we did monitor sleep behavior with wristwatch actigraphy. Participants slept for longer during the first night after learning in the sleep deprivation (vs. sleep) condition, but this between-condition difference in sleep duration was not significantly correlated with the magnitude of sleep's benefit for learning. This suggests that longer recovery sleep in the sleep deprivation condition did not meaningfully influence the impact of sleep loss on new learning.

It is worth noting, though, that an enhanced consolidation of newly learned adjective-image pairings in the sleep deprivation (vs. sleep) condition (due to longer or deeper recovery sleep) could have obscured a relationship between sleep-associated memory retention and next-day learning in our multiple regression analysis. However, the same null effects were observed when our analysis was restricted to data from the sleep condition alone rather than the subtractions between the sleep and sleep deprivation conditions (as was done in our primary analysis). Hence, no relationship between overnight consolidation and next-day learning was observed when the influence of sleep deprivation (and the putative enhancement of sleep-associated consolidation during recovery sleep) was removed from our data.

No link between sleep-associated consolidation and next-day learning

If memory consolidation during SWS supports a shift in the memory retrieval network from the hippocampus to the neocortex, then sleep-associated consolidation of hippocampus-dependent memories should predict next-day learning of new, hippocampally-mediated associations, and SWA should facilitate this relationship. However, we observed no such effects in our data, suggesting that new learning in the hippocampus may not be contingent on hippocampal memory processing during the preceding night of sleep.

An alternative interpretation of these null effects is that our experimental paradigm could not provide an adequate test of our hypothesis. Although we reasoned that the use of 2 conceptually different hippocampus-dependent tasks would prevent our findings from being influenced by retroactive or proactive interference, qualitative differences between these tasks might have negated our ability to detect a relationship between sleep-associated memory consolidation and next-day learning. This is nevertheless a speculative suggestion that can be addressed in future research (e.g. by using a paired-associates task to assess both overnight memory retention and subsequent encoding).
Although our study was motivated by the assumptions of the Active Systems framework (Walker 2009; Born and Wilhelm 2012; Rasch and Born 2013; Klinzing et al. 2019), it is important to also consider our findings in the context of homeostatic synaptic downscaling, which is regarded as another fundamental mechanism through which sleep supports learning and memory (Tononi and Cirelli 2014, 2016). From this perspective, sleep is the price the brain pays for waking plasticity to avoid an accumulation of synaptic upscaling. Because synaptic renormalization should mainly occur during sleep (when neural circuits can undergo a broad and systematic synaptic downscaling), a night of sleep deprivation would prevent the restoration of cellular homeostasis and impair next-day learning. Several theoretical accounts of sleep-associated memory processing have made progress in reconciling the key tenets of the Active Systems and Synaptic Homeostasis frameworks, suggesting that these processes work in concert to support global plasticity and local downscaling, respectively, and in doing so, prepare the hippocampus for future encoding (Lewis and Durrant 2011; Genzel et al. 2014; Klinzing et al. 2019). Interestingly, whereas global memory replay and consolidation have been linked to slow (<1 Hz) oscillations, downscaling and forgetting are associated with delta waves (1–4 Hz) in local networks (Genzel et al. 2014; Kim et al. 2019). How interactions between global slow oscillations and local delta waves regulate overnight memory processing is therefore pertinent to further understanding the relationship between sleep-associated consolidation and next-day learning.

Sleep loss disrupts effective learning

Successful learning is associated with left-lateralized beta desynchronization ∼0.5–1.5 s after stimulus onset (Hanslmayr et al. 2009, 2011, 2012, 2014; Griffiths et al. 2016, 2021). Consistent with these prior studies, we observed a decrease in beta power ∼0.3–2 s after stimulus onset during the encoding of subsequently remembered (vs. forgotten) associations, and this was most pronounced in the left hemisphere. Beta desynchronization is thought to reflect semantic processing during successful memory formation (Hanslmayr et al. 2011; Fellner et al. 2013); as beta power decreases, the depth of semantic processing increases (Hanslmayr et al. 2009). More broadly, neocortical alpha/beta oscillations have been linked to the processing of incoming information during episodic encoding (Griffiths et al. 2021). For our learning task, participants were instructed to form vivid mental images or stories that linked the adjective and image of each pairing. The observed downregulation of beta power during successful learning might thus reflect an engagement of information processing operations, possibly involving semantic representations, allowing these novel associations to be bound together into 1 coherent episode and committed to memory.
Importantly, however, the change in beta power that accompanied successful learning differed according to whether participants had slept or remained awake across the overnight interval. Whereas encoding of subsequently remembered (vs. forgotten) adjective-image pairings was associated with beta desynchronization after sleep, no significant difference in beta power emerged from the same contrast after sleep deprivation. Hence, a protracted lack of sleep appeared to disrupt semantic processing operations when participants were successfully forming new memories. This interpretation is in line with previous behavioral findings where sleep-deprived individuals have had difficulty in encoding semantically incongruent stimulus pairs (Alberca-Reina et al. 2014). The sleep-deprived brain might thus rely on alternative processing routes when committing new information to memory. Indeed, prior studies have shown that sleep deprivation leads to compensatory neural responses during learning (Chee and Choo 2004; Drummond et al. 2004) and recognition (Sterpenich et al. 2007).

What might be the nature of this alternative route to learning after sleep deprivation? It is interesting to note that we observed an upregulation of beta activity during successful (vs. unsuccessful) learning in the sleep deprivation condition (although this difference did not survive a Bonferroni correction for multiple comparisons). Increases in beta power have been linked to working memory and active rehearsal (Tallon-Baudry et al. 2001; Hwang et al. 2005; Onton et al. 2005; Deiber et al. 2007), suggesting that sleep-deprived individuals may engage in more surface-based rehearsal strategies due to semantic processing pathways being compromised by an absence of sleep.

It is important to note that the foregoing findings on beta desynchronization arose from an exploratory analysis that was not preregistered and should therefore be treated with caution until such time that they are replicated in confirmatory research.

Conclusion

We investigated whether memory consolidation in sleep predicts next-day learning and whether SWA contributes to this relationship. Furthermore, we investigated how the neural correlates of successful learning are affected by sleep deprivation. Although sleep improved both memory retention and next-day learning, we found no evidence of a relationship between these measures or SWA. Whereas beta desynchronization-an established marker of semantic processing during successful learning-was present during the encoding of subsequently remembered (vs. forgotten) associations after sleep, no such difference in beta power was observed after sleep deprivation. An extended lack of sleep might therefore disrupt our ability to draw upon semantic knowledge when encoding novel associations, necessitating the use of more surface-based and ultimately suboptimal routes to learning.

Acknowledgments

We are grateful to Marcus O. Harrington for help with data collection and Jennifer E. Ashton for help with the experimental materials. We also thank members of the Sleep, Language, and Memory Group at the University of York for fruitful discussions of the data.

Supplementary material

References

Alberca-Reina E, Cantero JL, Atienza M. Semantic congruence reverses effects of sleep restriction on associative encoding. Neurobiol Learn Mem. 2014:110:27–34.

Antonenko D, Diekelmann S, Olsen C, Born J, Mölle M. Napping to renew learning capacity: enhanced encoding after stimulation of sleep slow oscillations. Eur J Neurosci. 2013:37(7): 1142–1151.

Antony JW, Paller KA. Retrieval and sleep both counteract the forgetting of spatial information. Learn Mem. 2018:25(6):258–263.

Antony JW, Ferreira CS, Norman KA, Wimber M. Retrieval as a fast route to memory consolidation. Trends Cogn Sci. 2017:21(8): 573–576.

Ashton JE, Cairney SA. Future-relevant memories are not selectively strengthened during sleep. PLoS One. 2021:16(11):e0258110.

Ashton JE, Harrington MO, Guttesen AÁV, Smith AK, Cairney SA. Sleep preserves physiological arousal in emotional memory. Sci Rep. 2019:9(1):5966.

Ashton JE, Harrington MO, Langthorne D, Ngo H-VV, Cairney SA. Sleep deprivation induces fragmented memory loss. Learn Mem. 2020:27(4):130–135.

For more information:1950477648nn@gmail.com