Still Missing Some Significant Ingredients

IN THIS ISSUE OF SLEEP GENZEL ET AL.¹ PRESENT SURPRISING DATA ASKING US TO RECONSIDER WHAT WE THINK ABOUT SLEEP-DEPENDENT MEMORY consolidation. Classically, according to the dual-process hypothesis² it is believed that REM sleep facilitates procedural skill memory, whereas slow wave sleep (SWS) facilitates declarative memories.

However, surprisingly only rarely we see direct relationships of SWS or REM amounts after learning with overnight changes in declarative or procedural memory performance. In the current study by Genzel and colleagues subjects were deprived once each from SWS and REM sleep, and were once allowed to spend an undisturbed night. Memory performance was tested at 21:00 as well as 60 hours later, to bypass acute fatigue effects. Although deprivation of SWS and REM sleep (SWSD, REMD) led to significant reduction of the respective sleep stage (SWSD: 26.5 vs. 105.2 min, REMD: 12.1 vs. 81.6 min), procedural (finger tapping) as well as declarative memory consolidation (word pair association) remained unaffected. Yet stage 2 (S2) sleep spindles were found to be correlated with declarative memory performance after the undisturbed night. Unfortunately, the authors did not include a non-learning baseline night which makes it difficult to clearly dissociate trait-like (spindles are known to be generally elevated in good learners³) from immediate learning effects. Generally this is a caveat of many studies of this kind. That is, authors either do not relate their findings to overnight gains but simply to post-learning performances, or they relate absolute measures from the learning night (such as sleep spindles) to memory performance. However, to see if the change in sleep after learning is related to the change in memory performance overnight, it is important that these change scores be correlated.

The experimental paradigm most often used to compare effects of REM with those of SWS was designed by Ekstrand and colleagues.⁴ This paradigm takes advantage of the fact that SWS predominates during early sleep and REM during late sleep. Memory retention is then compared across sleep periods of equal length but with either REM or SWS predominating. Plihal and Born² extended that paradigm and used word-pair lists and mirror-tracing skills to assess whether the declarative and procedural memory systems can be related to SWS and REM, respectively. Although much data derived from that paradigm support the dual-process hypothesis, it is evident that plenty of studies to not fit this simple idea (Table 1).

Table 1.

Empirical Evidence Indicating that the Dual-Process Model of Sleep-Dependent Memory Consolidation is Incomplete

Authors name - study details	Verbal Declarative Learning (e.g., Word Pair Association) →SWS-dependent consolidation?	Procedural Motor Refinement (e.g., PR, FT task) →REM-dependent consolidation?	Complex Procedural Motor Learning (e.g., mirror tracing task) →REM-dependent consolidation?
Fogel et al. (2007)5	Theta and Sigma in REM↑	PR: S2 sleep ↑, SpD ↑, Spindle Duration ↑, 12–14 Hz activity ↑	REM density ↑
Fogel – Smith (2006)6	-	PR and others: S2 sleep ↑, SpD ↑; SpD Change x Memory Change	-
Smith – MacNeill (1994)7	-	PR: REMD no effect on memory performance (yet, late 4 hr sleep deprivation has)	-
Moroni et al. (2008)8	-	FT: SWS ↑, hippocampal 0.5–1 Hz ↑; 0.5–1 Hz x Memory Change	-
Rasch et al. (2008)9 –REMD by selective serotonin and norepinephrine reuptake inhibitors*	REMD no (negative) effect	FT: REMD had positive effect on the performance change overnight	REMD no (negative) effect
Hornung et al. (2007)10 –Old aged subjects (60–82 yrs)	REMD and S2 awakenings no effect	-	REMD no effect
Tamaki et al. (2008)11	-	-	Fast spindle duration and amplitude (13–16 Hz) ↑; absolute SpD and SpA x Memory Change
Morin et al. (2008)12	-	FT: Spindle number and duration ↑, >13 Hz and 18–20 Hz ↑	-
Nishida – Walker (2007)13	-	FT: S2 sleep x Memory Change; (fast) SpA (at [learning – non-learning hemisphere]) x Memory Change	-
Walker et al. (2002)14	-	FT: late S2 sleep x Memory Change	-
Schmidt et al. (2006)15 – Subjects spent 4 hour naps	11.25–13.75Hz ↑ after difficult encoding condition only; SpD change x Memory Change	-	-
Schabus et al. (2004)16	SpA change x Memory Change	-	-
Gais et al. (2002)17	SpD ↑; absolute SpD x memory performance	-	-

As easy procedural motor learning tasks we here consider the often used pursuit rotor (PR) and finger tapping (FT) task. As complex procedural motor tasks we consider the classically used mirror tracing task. For declarative learning the word-pair association task is considered. Note that with the term change we refer to changes in sleep (mechanisms) from non-learning nights to learning nights.

^*It has to be mentioned that the pharmacological intervention might have activated various immediate early genes leading to the observed paradox memory gains. SpD, spindle density (number of sleep spindles per minute of S2 sleep); SpA, spindle activity (measures of spindle duration and amplitude); REMD, REM deprivation. x, direct regression of measures; ↑, measure going up after learning.

Surprisingly, and as in the study by Genzel and colleagues, REMD does not always block procedural memory consolidation as would be expected.^7,9,10 The lack of effect might be attributable to the residual amounts of the deprived sleep stage. It is therefore possible that the 12 min of REM in the REMD condition of the present study were sufficient for REM sleep to exert its positive effect on memory. In this context it is especially worth noting that even ultra short episodes of sleep (6 min) have been reported to exert beneficial effects on memory.¹⁸ Furthermore, in some sleep deprivation studies subjects are allowed to immediately return to sleep after being awakened questioning whether subjects were truly awake at all. Genzel and colleagues adequately controlled for that effect by asking subjects to perform simple arithmetic calculations for the duration of 2 minutes before going back to sleep. In general, the present study appears well conducted and results are not explainable by simple methodological flaws.

Interestingly, S2 sleep filling almost 50% of the total sleep time at night has found little attention in models of sleep-dependent memory consolidation. Yet the distinct waxing and waning 12–15 Hz oscillatory patterns, termed sleep spindles have long been postulated to provide a physiological brain state supporting synaptic plasticity.¹⁹ Many of the effects seen when depriving subjects of early SWS might really originate from depriving the brain of early sleep spindling rather than SWS per se. Evidence accumulates indicating that sleep spindles serve declarative memory consolidation as well as procedural motor skills (Table 1).

Of course there is also positive evidence for the dual-process model, as illustrated by the study from Rasch and colleagues²⁰ demonstrating that odor cuing during SWS can improve the retention of hippocampus-dependent declarative memories but not of hippocampus-independent procedural memories. Likewise the study from Marshall and colleagues²¹ beautifully demonstrated that boosting slow oscillations by transcrainal direct current stimulation during SWS can improve declarative but not procedural memories overnight.

But when going beyond the tasks mentioned in Table 1, there are also many more studies contradicting the dual-process model. In a study by Gais and colleagues²² using the visual discrimination task, discrimination skills for example improved over early SWS, and not late REM sleep as expected for a procedural skill. Stickgold and colleagues²³ found that SWS in the first quarter as well as REM sleep in the last quarter of the night served the memory consolidation of this procedural task, pointing to a multi-step process where SWS and REM fulfill different functions in turns.

Altogether the picture emerges that some ingredients are still missing in order to arrive at a satisfactory model for sleep-dependent memory consolidation.

In 2004 Smith and colleagues²⁴ proposed a model where the complexity of a motor task defines whether S2 or REM sleep will be needed for memory consolidation. In that model tasks which simply require motor skill refinement such as the pursuit rotor or finger tapping task can be subserved by S2 sleep mechanisms. REM on the other hand will be needed if complex motor tasks requiring the learning of new rules—such as the mirror tracing task—are to be learned. Indeed, this model appears to account for some of the discrepancies seen in the literature. According to Smith and colleagues it is therefore not surprising that Genzel's easy finger tapping task did not respond to REMD. However, further refinements of the dual-process model are needed. Most of all we need to turn away from the macroscale simply studying sleep architecture changes and focus more on fine-grained analyses of specific sleep mechanisms such as the distinguishable slow (<13 Hz) and fast (>13 Hz) sleep spindles,²⁵ PGO waves, REM densities, or actual slow wave activity (SWA); and also respect their often local experience-dependent nature. Furthermore, it has been shown that the regular occurrence of NREM-REM cycles, might be more crucial for retention of verbal material across sleep than a specific sleep stage per se.²⁶

We should not forget to also pay more attention to the exact nature of the studied tasks, the individual traits of our subjects (age, sex, a priori abilities), or the exact behavioral tests performed (free or cued recall, delayed tests of performance, susceptibility to interference).

In the end the story appears simply much more complex than can be explained by a simple model. Therefore, the scientific search must continue to unravel the many facets of sleep-dependent memory consolidation and come up with a more comprehensive theory integrating up to now divergent results.

DISCLOSURE STATEMENT

Dr. Schabus has indicated no financial conflicts of interest.