G ENERAL D ISCUSSION

In the present study, we investigated whether one night of sleep might affect memory formation and consolidation for new individual faces. Our study suggests that, by contributing to better explicit face recognition, sleep might thus mediate important social and adaptive functions.

In our first experiment, we found improved explicit memory for faces that had been seen twice before a period of nocturnal sleep when compared to a similar period of daytime wakefulness. By contrast, implicit access to newly encountered faces was maintained independently of the presence or absence of a period of sleep. Experiment 2 showed that long-term face recognition tested after a period with intervening sleep was not affected by changes in the size of the faces between encoding and testing sessions, whereas a decrease in performance was observed after a period of wakefulness. This suggests that some plasticity mechanisms may occur at a high-level of visual processing, where faces are represented independently of their retinotopic coordinates.

The present demonstration of a modulation of face recognition after one night of sleep adds new support to previous behavioral evidence of sleep-dependent improvement in learning and memory consolidation (Karni et al., 1994; Stickgold, James et al., 2000; Gais et al., 2000;

Maquet, 2001; Peigneux et al., 2001; Gaab et al., 2004; Walker & Stickgold, 2004). While a previous study found increased implicit memory for faces in a task requiring face identification after REM sleep (Wagner et al., 2003), here we show for the first time that explicit recognition of individual faces can also benefit from a period of sleep, independently of sleep effects on implicit memory measures. Our results are consistent with a recent study on a word-list discrimination task showing enhanced explicit recognition for verbal material after sleep as compared with wakefulness, whereas implicit memory was not affected by sleep (Drosopoulos et al., 2005) and another one, using a serial reaction time task, which also shows specific explicit memory enhancement after sleep (Fischer et al., 2006). It should be noted here that performance in old/new recognition tasks may include some independent implicit component, through a feeling of familiarity coming to mind automatically in the presence of an old face, without any intentional recollection (Jacoby, 1991). However, the recognition task is a conscious memory test during which reference is made to the faces learned before. Thus, old/new responses in this task can be considered as an explicit memory measure (Schacter, 1987). Furthermore, experiments on amnesic patients showed normal

implicit memory but impaired recognition performance (Warrington & Weiskrantz, 1974;

Hamann & Squire, 1997). Most importantly, in our experiments we found no overnight increase for different measures of implicit memory, thus providing additional support for increased sensitivity of explicit rather than implicit memory formation to the beneficial effect of sleep (Fischer et al., 2006; Robertson et al., 2004).

In the present experiments, we found that repeated exposures to an individual face before one night of sleep significantly improved the subsequent recognition of this face. This observation is compatible with recent models of memory consolidation in which new memories are not consolidated at once, but instead need to be re-instantiated before they are consolidated (Stickgold & Walker, 2005; Walker et al., 2003). Another implication of our finding is that sleep effects might preferentially apply to certain memory traces, in particular those for stimuli that achieved a sufficient level of encoding. The importance of the number of pre-sleep stimulus repetitions was previously reported in rats by Smith and colleagues (Smith &

Lapp, 1986; Smith et al., 1980). They found that 100 trials on a two-way shuttle shock avoidance task given within one training session induced more subsequent REM sleep and improvement of performance as compared to the same total number of trials distributed over several days. Our experiments provide evidence for a similar mechanism in humans whereby the repetition of visual stimuli is required for sufficient memory traces to be formed before they can be further consolidated during human sleep.

Moreover, our results revealed that explicit memory was unaffected by changing the size of images, implying that overnight memory improvement relied on some abstract encoding rather than pure ‘retinotopic’ perceptual encoding of the new face stimuli. Therefore overnight improvement in face recognition might involve neuronal reorganization within high-level visual systems, where information coding is relatively invariant to the metric properties of the images, thus allowing successful recognition of faces even across large changes in visual appearance. By possibly promoting the abstraction process underlying the consolidation of new face representations (cf. Bruce & Burton, 2002), sleep could provide some optimal neurophysiological conditions for the integration of newly created representations into an organized network of conceptual knowledge (see Maquet, 2001).

Previous functional MRI studies showed that sleep could induce lasting changes in primary visual cortex activity after training on fine visual discrimination tasks (Schwartz et al., 2002;

Walker et al., 2005). Based on the present results, one could predict that sleep does not only affect the earliest cortical stages of visual processing, but might also induce task-related changes at higher-level stages of vision in the human brain, such as regions involved in face perception (e.g., Kanwisher et al., 1997; Haxby et al., 2001).

Below we consider a critical point concerning our assessment of sleep effects. Several procedures have been proposed to test for sleep-associated effects on cognitive performance.

In addition to the direct comparison of memory performance across sleep and wake delays as performed here, total or partial sleep deprivation protocols are also frequently used (Plihal &

Born, 1999; Plihal & Born, 1997; Wagner et al., 2003; Gais et al., 2000; Maquet, Schwartz et al., 2003; Orban et al., 2006, see also S. Fischer et al., 2002). Yet, all these experimental procedures present important limitations. Sleep-deprivation procedures typically disrupt normal nocturnal sleep, while altering many other factors associated with nocturnal sleep.

Thus, reduced performance after sleep-deprivation may reflect unspecific effects of stress, sleepiness, and reduced attention on cognitive performance rather than impaired memory (Belenky et al., 2003; Drummond et al., 2000). Conversely, comparisons across normal periods of sleep or wakefulness constitute more ‘ecological’ measures that can provide important information about memory processes during naturally occurring periods of sleep or of wakefulness. This approach may also be useful to plan training schedules for rehabilitation or educative purposes. Nevertheless, comparing normal sleep and wake periods implies circadian variations that can also affect cognitive performance. It is however unlikely that circadian cycles influence might account for the present pattern of results. Indeed, in Experiment 1, recognition performance levels for items from test phase 3 as well as RTs from the three test phrases did not differ between day and night groups, thus suggesting similar subjects’ behavior in our tasks during evenings and mornings. Moreover, circadian effects could lead to low performance for faces seen twice in the morning for the night group and a high level for those seen twice at night for the day group, which was not the case. Therefore, while complex circadian-sleep interactions as well as individual differences in circadian arousal (May et al., 2005) might have ad some non-specific noise in the data, statistically significant results reported in the present study plausibly reflect genuine effects of sleep on memory performance assessed in normal physiological conditions.

Finally, there is an ongoing debate concerning whether priming for newly learned unfamiliar faces can occur. Here, we found neither MEE in a pleasantness task nor RT priming effects in a gender discrimination task for studied unfamiliar faces(unlike Goshen-Gottstein & Ganel, 2000). However, we found faster RTs in the old/new recognition task for faces that had been seen twice compared to those that had been seen only once. These latter results are in line with previous findings by Ellis et al. (1990) and Wagner et al. (2003) who also found repetition priming only in a task requiring face recognition. Identification implicates relatively high levels of face processing, whereas gender discrimination might be performed at early levels of face processing (e.g., structural encoding). Pleasantness ratings may have been performed relying on superficial esthetic features than on deeper encoding processes. Our findings therefore suggest that implicit memory traces for the same face stimuli can be expressed in task requiring a deep processing of individual characteristics of the faces (pleasantness judgment or identification), but not (or to a lesser extend) in tasks based on shallow processing such as gender discrimination tasks.

What are the possible neurophysiological mechanisms underlying sleep-associated improvement in face recognition? Evidence ranging from cell recordings in rodents (Louie &

Wilson, 2001; Qin et al., 1997; Wilson & McNaughton, 1994) to PET and EEG studies in

humans (Maquet et al., 2000; Peigneux et al., 2004; Huber et al., 2004) demonstrated experience-dependent reactivation of brain structures previously activated during task performance at wake. Moreover, fMRI studies using sleep-dependent learning tasks found persistent changes in brain regions that were selectively implicated in the processing of information critical to the task (Maquet, Schwartz et al., 2003; Schwartz et al., 2002). Finally, Mograss and colleagues used ERPs to investigate sleep-dependent face learning (Mograss et al., 2006; Mograss et al., 2008). They found an increased old/new effect after nocturnal sleep and associated modifications of early and late frontal and posterior components. They interpreted these changes as possible manifestations of reduced interference inhibition, increased contextual processing, and facilitation of episodic memory.

Whether similar sleep-related neurophysiological mechanisms of local replay and plasticity would also apply to complex visual learning remains to be studied. While there is no direct evidence showing that overnight improvement of explicit face recognition might involve neuronal replay within the face networks during sleep, introspective reports reveal that face representations are frequently generated during sleep, constituting a recurrent visual category in our dream imagery (Kahn et al., 2002; Kahn et al., 2000; Schwartz & Maquet, 2002).

Overnight improvement in face recognition as found in the present study might rely on a stabilization and refinement of facial traits processed in the FRU, implicating neural changes in particular face-responsive brain regions such as the fusiform cortex. Sleep might thus promote the integration of new information within dedicated perceptive or cognitive modules, mediated by local reorganization within highly specialized brain regions. Hence, sleep may provide a favorable condition for the modular architecture of brain functions to be expressed and consolidated. Our study also suggests that by contributing to better explicit face recognition, sleep might thus mediate important social and adaptive functions.

How permanent such sleep-related effects on explicit memory are, where in the brain does neural plasticity underlying improved face recognition occur, and to what extend explicit memory for other visual categories might also benefit from a period of sleep remain important questions for future research.

Acknowledgements

The authors thank E. Sforza and P. Maquet for many insightful comments. This work was supported by the Swiss National Science Foundation (grants #3100A0-102133, #310000-114008 to S. S. and NCCR Affective Sciences #51NF40-104897).

5. Influence of sleep on face recognition and associated activity in right