تاثیر خواب پس از یادگیری در مقابل بیداری بر روی حافظه شناخت برای چهره های با حالات مختلف

Abstract A beneficial effect of sleep after learning, compared to wakefulness, on memory formation has been shown in many studies using a variety of tasks. However, none of these studies has specifically addressed recognition memory for faces so far. The recognition of familiar faces, together with the extraction of emotional information from facial expression, is a fundamental cognitive skill in human everyday life, for which specific neural systems and mechanisms of processing have been developed. Here, we investigated the role of post-learning sleep for later recognition memory for neutral, happy, and angry faces. Twelve healthy subjects, after judging the emotional valence of the faces in the evening (learning phase), either slept normally in the subsequent night, with sleep recorded polysomnographically (sleep condition), or remained awake (wake condition) according to a cross-over design. Recognition testing took place in the second evening after learning, i.e. after a further night of regular sleep spent at home. Sleep after learning, compared to wakefulness, enhanced memory accuracy in recognition memory. This effect was independent of the emotional valence of facial expression. The response criterion at recognition testing did not differ between sleep and wake conditions. The amount of non rapid eye movement (NonREM) sleep during post-learning sleep correlated positively with memory accuracy at recognition testing, while time in REM sleep was associated with a speeded responding to the learned faces. Results suggest that face recognition, despite its dependence on specialized brain systems, nevertheless relies on the general neural mechanisms of sleep-associated memory consolidation.

. Introduction Evidence from animal and human studies supports the notion that sleep plays a crucial role in the consolidation of newly acquired memory traces (e.g., Buzsáki, 1998, Pennartz et al., 2002 and Stickgold, 2005). In humans, a beneficial effect of sleep on memory retention of previously learned material has been demonstrated for a broad variety of tasks and materials, where specific sleep stages are differentially implicated depending on the memory system addressed by the specific task (Gais and Born, 2004a, Maquet, 2001 and Walker and Stickgold, 2006). Regarding the fundamental distinction between hippocampus-dependent explicit (declarative) and hippocampus-independent implicit (nondeclarative) memory (Squire, 1992), the latter appears to benefit mainly from rapid eye movement (REM) sleep, the former from non-rapid eye movement (NonREM) sleep, in particular slow wave sleep (SWS) (Fowler et al., 1973, Karni et al., 1994, Plihal and Born, 1997, Plihal and Born, 1999, Tucker et al., 2006 and Wagner et al., 2003). The present study investigates the role of sleep in recognition memory for previously seen faces. Faces belong to the most important visual stimuli humans encounter in everyday situations. We usually identify the persons we know by recognizing their face, a process for which specific neuroanatomical systems have developed in the brain (Allison et al., 1994, Farah, 1996, Kanwisher, 2000 and Tsao et al., 2006). However, facial stimuli have seldom been used as learning material in sleep research (Clemens et al., 2005 and Wagner et al., 2003). Thus, although the ability to reliably recognize familiar faces is one of the most fundamental cognitive skills in human life, little is known so far about how memory for faces is influenced by sleep. In a previous study, we have investigated the effect of sleep on implicit face memory in a reaction time priming task and found evidence for a supporting effect specifically of sleep periods containing high amounts of REM sleep (Wagner et al., 2003). Here, we extend these findings by investigating the role of sleep in explicit memory for faces in a standard recognition memory procedure. Explicit recognition of familiar faces is a basic prerequisite for appropriate human social behavior. Importantly, recent findings in other memory tasks indicated that sleep can even be more relevant to explicit than implicit aspects of memory formation ( Fischer et al., 2006, Robertson et al., 2004 and Wagner et al., 2004). The only study on sleep-associated memory consolidation that used facial stimuli to assess explicit memory so far was reported by Clemens and colleagues (2005), who primarily investigated verbal declarative memory (remembering names) but introduced the face memory task as a nonverbal control task. In this study, subjects learned face-name associations in the evening and were tested after a night of sleep in the next morning for verbal (free recall of the names) and nonverbal memory (face recognition task). Recognition memory for the faces was positively associated with total sleep time and the amount of NonREM sleep during the night (while overnight memory retention of the names correlated with the number of sleep spindles). Here, we compare directly the effects of sleep and wakefulness following learning on subsequent recognition of faces newly encountered at learning. To avoid confounds with circadian factors, learning and memory testing always took place at the same time of day in both conditions. A second aim of the study was to investigate the possible modulating role of emotional valence on sleep-associated face recognition. Apart from the identification of familiar persons, conveying emotional information via facial expression is the second major function of faces (see Calder & Young, 2005, for a recent discussion of both functions). To vary emotional valence of the learning material, we therefore included faces with neutral, positive (happy), and negative (angry) facial expression. Emotionally valenced material is typically better remembered than neutral material (a phenomenon called “emotional enhancement” in memory) and previous studies indicated that the effect of sleep on memory consolidation can differ depending on the emotional valence of the learning material, with memory consolidation for highly emotional material particularly benefiting from sleep periods containing high amounts of REM sleep (Cahill and McGaugh, 1998, Christianson, 1992, Hamann, 2001, Wagner et al., 2001 and Wagner et al., 2005). However, because these observations mainly refer to verbal free recall tasks, they do not necessarily hold for face recognition memory. Although emotional enhancement has also been reported for recognition memory tasks (Bradley et al., 1992 and Ochsner, 2000), several studies on recognition memory failed to confirm this and rather found evidence for an emotion-induced shift towards a more liberal response bias, i.e., a general tendency to classify emotional stimuli as familiar regardless of whether they were actually presented previously or not (Joyce and Kutas, 2005 and Windmann and Kutas, 2001). This has also been demonstrated specifically for face stimuli (Johansson, Mecklinger, & Treese, 2004). Apart from memory performance per se, this response bias may be also influenced by sleep. We therefore tested effects of post-learning sleep not only on memory accuracy, but also on response bias in recognition memory (see Snodgrass & Corwin, 1988).

Results 3.1. Face recognition Results for recognition memory measures (hit rate, false alarm rate, memory accuracy, response bias) are shown in Table 1, including pairwise t-tests between sleep and wake conditions. Sleep after learning, compared to wakefulness, generally enhanced memory accuracy (Pr), independent of facial expression (sleep 0.46 ± 0.05 vs. wake 0.36 ± 0.03, F (1,22) = 5.52, p = .038, for main effect of Sleep; p = .69 for Sleep × Valence interaction). Although both the hit rates and the false alarm rates contributed to this effect (as indicated by overall higher hit rates and lower false alarm rates after sleep than wakefulness), these measures per se were not significantly affected by sleep overall or in interaction with valence (p > .26, for all respective effects). Control analyses, which included the factor “Order” in the ANOVA, showed that the sleep effect on memory accuracy did not depend on whether the sleep condition was the first or second experimental night for a subject (p = .41), nor was there an overall order effect (p = .57). Table 1. Face recognition Sleep Wake Mean SEM Mean SEM p Hit rate Angry faces 0.63 0.07 0.59 0.06 .723 Neutral faces 0.69 0.06 0.61 0.05 .117 Happy faces 0.61 0.07 0.53 0.04 .298 All faces 0.64 0.06 0.58 0.03 .266 False alarm rate Angry faces 0.09 0.04 0.19 0.04 .082 Neutral faces 0.21 0.05 0.26 0.04 .365 Happy faces 0.24 0.06 0.22 0.04 .754 All faces 0.18 0.04 0.22 0.03 .325 Memory accuracy (Pr) Angry faces 0.53 0.08 0.40 0.07 .173 Neutral faces 0.48 0.06 0.35 0.04 .112 Happy faces 0.37 0.07 0.32 0.04 .491 All faces 0.46 0.05 0.36 0.03 .038 Response bias (Br) Angry faces 0.18 0.08 0.36 0.08 .204 Neutral faces 0.38 0.09 0.41 0.06 .686 Happy faces 0.39 0.07 0.30 0.06 .383 All faces 0.32 0.06 0.36 0.05 .962 Bold indicates that the value is the only significant pairwise comparison between sleep and wake conditions. Table options Regardless of sleep, both false alarm rates and memory accuracy (Pr), but not hit rates, tended to be influenced by valence (p = .08 and .10, respectively, for main effect of Valence on false alarm rates and Pr; p = .23 for hit rates), with lower false alarm rates for angry faces as compared to neutral and happy faces and, consequently, highest memory accuracy for angry faces (false alarm rates: angry 0.14 ± 0.03, neutral 0.23 ± 0.04, happy 0.23 ± 0.03; Pr: angry 0.47 ± 0.07, neutral 0.41 ± 0.04, happy 0.34 ± 0.04). The response bias (Br) was not overall affected by sleep or emotional valence (p = .64 and p = .19, for respective main effects), while both factors tended to interact in their influence on response bias (p = .09) due to a relatively enhanced conservative response bias (i.e., an inclination to answer “new”) specifically for angry faces in the sleep condition. Fig. 1b summarizes the results for recognition performance collapsed across the three valence categories. We additionally performed an analysis in which valence categories were formed on the basis of individual judgments of facial expression rather than by a priori classification. This analysis was performed for hit rates only, because individual judgments of facial expression were only obtained for old faces presented at learning. This analysis revealed the same pattern of results as the analysis of hits based on the a priori classification of valence categories. Although sleep enhanced hit rates numerically compared to wakefulness, this effect did not reach significance (sleep 0.66 ± 0.06 vs. wake 0.57 ± 0.03, p = .12, for main effect of sleep). Valence did not affect hit rates overall or in interaction with sleep (p > .30). 3.2. Reaction times Reaction time data are displayed in Table 2. Sleep did not exert a substantial effect on reaction times overall or in interaction with valence or the Old/New factor (p > .13, for all effects). Also, reaction times did not differ on the whole between old and new faces (p = .35 for main effect Old/New). A significant main effect of Valence (p = .04) indicated that across all conditions responses were generally faster for neutral as compared to happy and angry faces (neutral 857 ± 24 ms, happy 889 ± 20 ms, angry 883 ± 20 ms; p = .003, for neutral vs. happy; p = .07, for neutral vs. angry faces). Whereas the data in Table 2 show that this effect results mainly from the strong differences in this direction in the Old/Wake and the New/Sleep subconditions, Valence did not interact significantly with the two other factors (p = .69, for Valence × Old/New interaction; p = .19, for Valence × Sleep interaction; p = .14, for Valence × Old/New × Sleep interaction). Table 2. Reaction times (ms) Sleep Wake Mean SEM Mean SEM p Old Angry faces 848 40 949 31 .113 Neutral faces 876 31 846 38 .577 Happy faces 905 31 899 24 .878 All faces 876 25 898 25 .709 New Angry faces 851 27 883 35 .314 Neutral faces 822 36 884 36 .112 Happy faces 853 33 899 44 .365 All faces 842 29 889 36 .174 Difference old–new Angry faces −3 34 66 43 .225 Neutral faces 53 47 −38 33 .174 Happy faces 52 48 −1 48 .494 All faces 34 29 9 33 .584 Table options 3.3. Sleep Two participants, whose sleep could not be recorded completely due to technical problems, were excluded from sleep analysis. Polysomnographic recordings from the remaining subjects confirmed a normal distribution of sleep stages during consolidation sleep in the night after learning (stage 1 sleep, 3.2 ± 1.0%, stage 2 sleep, 59.3 ± 2.6%, SWS 18.7 ± 2.5%, REM sleep 17.4 ± 1.17%, wake time 0.3 ± 0.1%, total sleep time 441.8 ± 4.8 min, sleep onset 11.5 ± 2.4 min). Correlations between recognition performance and sleep parameters showed that in the sleep condition overall memory accuracy was strongly correlated with the amount of NonREM sleep during consolidation sleep (r = .79, p = .007; Fig. 2a). Hit rate or false alarm rate alone were not significantly associated with NonREM sleep (r = .36, p = .31, and r = −.42, p = .23, respectively). Separate analyses for the three valence categories of facial expression revealed that the association between memory accuracy and NonREM sleep was significant for angry and happy, but not neutral faces (angry: r = .79, p = .006, happy: r = .86, p = .001, neutral: r = .03, p = .93). The same pattern, although less pronounced, was found for the correlation between memory accuracy and total sleep time in the night after learning (all faces: r = .65, p = .042, happy faces: r = .66, p = .037, angry faces: r = .61, p = .063, neutral faces: r = .11, p = .76), but not for any of the single sleep stages within NonREM sleep (S1, S2, S3, S4, and SWS). REM sleep, in contrast to NonREM sleep, was negatively correlated with memory accuracy, although not significantly (r = −.53, p = .116) and tended to be associated with higher overall false alarm rates (r = .59, p = .072). Time spent in NonREM and REM sleep during post-learning consolidation sleep are ... Fig. 2. Time spent in NonREM and REM sleep during post-learning consolidation sleep are differentially correlated with explicit and implicit aspects of memory performance at recognition testing. NonREM sleep was associated with memory accuracy in recognition memory, i.e. explicit memory for the faces (a), while REM sleep was associated with a relative speeding of response time for old as compared to new faces, i.e. repetition priming (implicit memory) (b). Figure options Regarding reaction times, there was a strong negative correlation between REM sleep and reaction times for old faces (r = −.76, p = .012), but not for new faces (r = −.02, p = .95). Consequently, REM sleep was also associated with the difference between old and new faces in reaction times as an indicator of implicit memory (r = −.70, p = .024; Fig. 2b). This pattern was observed in all three valence categories, but did not reach significance in separate analyses for the three categories (angry: −0.54, p = .108, happy: −0.54, p = .108, neutral: −0.59, p = .071). Neither NonREM sleep overall nor any sub-stage of NonREM sleep correlated significantly with any reaction time measure. 3.4. Subjective ratings Ratings of subjective sleepiness, activation, motivation, boredom, concentration, and tension obtained at learning and recognition testing did not differ between sleep and wake conditions (sleep vs. wake means ± SEM at learning: sleepiness 3.08 ± 0.26 vs. 2.42 ± 0.31, activation 2.83 ± 0.35 vs. 3.25 ± 0.31, motivation 3.08 ± 0.38 vs. 3.42 ± 0.36, boredom 2.25 ± 0.33 vs. 2.00 ± 0.28, concentration 2.92 ± 0.38 vs. 3.42 ± 0.26, tension 2.25 ± 0.31 vs. 2.08 ± 0.29; at recognition testing: sleepiness 1.58 ± 0.26 vs. 1.92 ± 0.36, activation 3.83 ± 0.31 vs. 3.58 ± 0.23, motivation 3.67 ± 0.28 vs. 3.25 ± 0.18, boredom 2.17 ± 0.24 vs. 1.92 ± 0.23, concentration 3.42 ± 0.26 vs. 3.42 ± 0.19, tension 2.08 ± 0.34 vs. 2.00 ± 0.30; p > .19, for all main effects of Sleep and Sleep × Phase interactions). Independent of sleep vs. wake conditions, there was a general tendency to feel less sleepy and more activated at recognition testing than at learning (sleepiness, p < .01, activation, p < .05, for main effect of Phase), which probably reflects circadian influences as well as anticipatory effects (recognition testing, in contrast to learning, was not followed by an overnight stay at the laboratory). 3.5. Performance at learning To control for possible “baseline” differences in facial processing, we compared experimental conditions also with respect to task performance at learning. Identification of facial expression did not differ between sleep and wake conditions (means ± SEM for sleep vs. wake conditions: angry faces, 39.2 ± 2.6% vs. 37.5 ± 3.5%; neutral faces, 76.7 ± 3.3% vs. 78.3 ± 2.7%; happy faces 94.2 ± 2.3% vs. 94.2 ± 2.6%; p = .99, for main effect of Sleep; p = .92, for Sleep × Valence interaction). Independent of experimental sleep vs. wake conditions, identification of facial expression (as defined by the a priori valence categories) was distinctly better for neutral and happy faces as compared to angry faces and better for happy as compared to neutral faces (p < .001, for main effect of Valence and all pairwise comparisons).