5. INFLUENCE OF SLEEP ON FACE RECOGNITION AND ASSOCIATED ACTIVITY IN RIGHT
5.3.2. FMRI RESULTS
Having established that sleep after learning has an effect on subsequent explicit memory for faces and on their configural processing, we now go to the main focus of our study:
identifying sleep-related effects on brain activity, and more especially on the hippocampus and face-selective areas.
Memory for identity
First, we compared neural responses to previously learned faces (Whole faces) minus Novel ones (Table 2). This contrast revealed increased activity in the right and left superior frontal gyri, independently of the group. These regions are known to be recruited in high cognitive functions and particularly in working memory (du Boisgueheneuc et al., 2006) but also in the retrieval of voice identity (Relander & Rama, 2009). More importantly, this analysis revealed an increased activity in the right hippocampus (peak at 24x, -27y, -15z), for the night group and the 24 hours group compared to the day group (Figure 4). A similar pattern of activity was found very close from this peak in the right hippocampus (peak 24x, -24y, 15z), in the contrast Whole faces minus Distracters (Two halves and One half) and Novels. This result was consistent with a general involvement of the hippocampus in face memory (Haxby et al., 1996). The 24 hours group also showed increased activation in frontal areas, compared to both the day and the night group, for the Whole faces compared with the Novel faces. Frontal
regions have been implicated in face memory, and especially in the protection against false recognition (Rapcsak et al., 2001; Rapcsak, 2003).
Whole face Two halves One half Novels
Whole face > Novels, (Night&24h) > Day
T=3.41, p < .05 SVC
Whole face Two halves One half Novels
Whole face > Novels, (Night&24h) > Day
T=3.41, p < .05 SVC
Whole face Two halves One half Novels
Whole face > Novels, (Night&24h) > Day
T=3.41, p < .05 SVC
Figure 4: A. Increased response to the Whole face items in the right hippocampus for the night and 24 hours groups compared to the day group. Statistical maps are overlaid on average T1-structural scan and displayed at p < .005. B. Parameter estimates extracted from the hippocampal peak illustrate the selectivity of this difference for the Whole face condition. White bars: day group, black bars: night group; grey bars: 24 hours group. A 2-way ANOVA, with one between-group factor (Group: day, night, 24hours) and one within-group factor (Condition: Whole face, Two halves, One half, Novels) was performed on the parameter estimates. There was no main effect but a significant interaction (p <
.05). * p < .05, ** p < .01.
Table 2. Areas of significant activation during presentation of learned faces (Whole face items) relative to novel ones (p < 0.001, mimimal voxel size k = 5). Based on a priori hypotheses significances are corrected for small volumes of interest (SVCs) covering the right hippocampus.
Brain area Coordinates
x,y,z (mm) T value
p value
Whole face > Novels Day, Night and 24hours
R superior frontal gyrus 18, 6, 48 4.75 < .001 L superior frontal gyrus -18, -3, 48 3.28 < .001 Whole face < Novels Day, Night and 24hours
Nothing
Whole face > Novels Night > Day
L posterior hippocampus -39, -39, -6 4.16 < .001
R hippocampus 24, -27, -15 3.16 < .05 SVC
Whole face > Novels Night and 24hours > Day
R hippocampus 24, -27, -15 3.41 < .05 SVC
Whole face > Novels Night < Day
Nothing
Whole face > Novels Night and 24hours < Day
Nothing
Whole face > Novels Night > 24hours
Nothing
Whole face > Novels Night < 24hours
R middle frontal gyrus 33, 18, 30 4.19 < .0001 Whole face > Novels 24 hours > Day
L superior frontal gyrus -24, 3, 48 4.13 0.000 L posterior cingulate gyrus -3, -36, 45 3.69 0.000 Whole face > Novels 24hours < Day
Nothing
Memory for configuration
Then we tested our hypothesis on the sleep-dependent strengthening of configural information by contrasting the “Whole face vs. Two halves” conditions. Behavioral results showed that the day group relied more on features, making less distinction between faces with old features in the learned configuration (Whole face) and faces with old features in a new configuration (Two halves). Here are presented the results from the contrast Whole face minus Two halves, assessing the memory for faces’ identity, i.e. memory for faces containing both featural and configural information versus faces containing previously seen features but in another configuration (Table 3). Results showed a decreased activity in the left inferior occipital gyrus (Occipital Face area, OFA) for the Whole face compared to the Two halves items, independently of the group. (Figure 5). The parameters estimates of the indivual peak values for the OFA across the four conditions showed that there was in fact an increase of activitiy for the distracters in the day group. Thus, instead of a clear sensitivity to configuration, the OFA might process the parts and the day group might do more intense verification of these features.
-1 -0.5 0 0.5 1
-36x –87y –12z, Left inferior occipital gyrus (OFA) Whole face Two halves One half Novels
Signal change
** *
-1 -0.5 0 0.5 1
-36x –87y –12z, Left inferior occipital gyrus (OFA) Whole face Two halves One half Novels
Signal change
** *
Figure 5: Decreased response to the Whole face items in the right OFA for the day group compared to distracters (Two halves and One half). White bars: day group, black bars: night group; grey bars: 24 hours group. A 2-way ANOVA, with one between-group factor (Group: day, night, 24hours) and one within-group factor (Condition: Whole face, Two halves, One half, Novels) was performed on the parameter estimates. There main effect of condition was at trend (p = . 06). There was no main effect of group (p = .42), nor interaction ( p = 0.20). Planned comparisons showed significant increases for the distracters in the day group compared with the Whole faces * p < .05, ** p < .01.
This contrast also revealed areas whose brain activity was modulated by sleep. Importantly, we found a significant increase of activity in the right fusiform gyrus for the Whole face items compared to the Two halves distracters. Activity in this region was significantly higher for subjects from the night group compared to subjects from the day group. We superimposed the image from the Whole face > Two halves brain activity to the Houses >
Houses&Landscapes&Scrambled result from the Face localizer task. We found that the fusiform region activated for the Whole face items lies within the Fusiform Face Area (Kanwisher et al., 1997) (Figure 6). This result suggests that sleep might increase face recognition performance by modulating the activity in the FFA, known to be critical for configuration processes (Rossion et al., 2000). However, the 24 hours group did not show a similar pattern. Other processes might allow good recognition performance for this group. An increased activity was observed for these subjects, in middle frontal and cingulate regions, for the Whole face compared with the Novel items. These regions have been thought to play a role in explicit retrieval (Schott et al., 2005; Buckner & Koutstaal, 1998).
Face localizer, mean R FFA
Whole face > Two halves,
& Night > Day Whole face > Two halves, Night > Day
Signal change
39x –45y –21z, Right Fusiform gyrus
One half Novels
Whole face > Two halves,
& Night > Day Whole face > Two halves, Night > Day
Signal change
39x –45y –21z, Right Fusiform gyrus
One half Novels
Figure 6: A. Increased response to the Whole face items in the right fusiform guys for the night group compared to the day group. B. Statistical maps are overlaid on average T1-structural scan and displayed at p < .005. Parameter estimates extracted from the right fusiform gyrus peak illustrate the selectivity of this difference for the Whole face condition. White bars: day group, black bars: night group; grey bars: 24 hours group. A 2-way ANOVA, with one between-group factor (Group: day, night, 24hours) and one within-group factor (Condition: Whole face, Two halves) was performed on the parameter estimates. There was no main effect but a significant interaction (p = .02). * p < .05, **
p < .01. C. Superimposition of the right fusiform gyrus activity from the Whole face > Two halves and the FFA activity from the Face localizer task. The localizer activity image is presented at the same threshold at the Whole face > Two halves contrast image’s one (p < .005).
Table 3. Areas of significant activation during presentation of learned faces (Whole face items) relative to Two halves distracters (p < 0.001, mimimal voxel size k = 5). Based on a priori hypotheses significances are corrected for small volumes of interest (SVCs) covering the right fusiform gyrus.
Brain area Coordinates
x,y,z (mm)
T value p value
Whole face > Two halves Day, Night and 24hours
L parahippocampal gyrus -33, -12, -33 4.57 < .001
L precentral gyrus -45, -12, 42 4.24 < .001
R precentral/frontal eye field 54, -9, 39 3.96 < .001 Whole face < Two halves Day, Night and 24hours
L fusiform gyrus -42, -69, -9 4.69 < .001
L inferior occipital gyrus (OFA) -36, -87, -12 4.28 < .001 Whole face > Two halves Night > Day
R inferior temporal gyrus 45, -63, -6 3.64 < .001
R fusiform gyrus 39, -45, -21 3.06 < .002, < .05 SVC
Whole face > Two halves Night and 24hours > Day
L insula -36, -3, 24 5.06 < .001
R medial frontal gyrus 6, -12, 51 3.94 < .001
L posterior cingulate -3, -36, 51 3.8 < .001
L posterior cingulate -3, -39, 21 3.78 < .001
Whole face > Two halves Night < Day
Nothing
Whole face > Two halves Night and 24hours < Day
Nothing
Whole face > Two halves Night > 24hours
Nothing
Whole face > Two halves Night < 24hours
Nothing
Whole face > Two halves 24 hours > Day
L middle frontal/precentral -36, -6, 30 5.4 0.000
R cingulate 6, -12, 51 3.85 0.000
Whole face > Two halves 24hours < Day
Correlation between hippocampal and fusiform activity
As the fusiform gyrus is a key structure for face recognition, and, as showed here, for configural processing, we tested whether activity in this brain region could be correlated to hippocampal activity, during recognition of learned faces compared with distracters and novel ones, and whether this correlation might be different for the sleeping and non-sleeping groups. Regarding the standard consolidation theory (Dudai, 2004, 2006), our hypothesis was that memory for faces was consolidated through hippocampal-fusiform connections, which may be stronger after a night of sleep, and weaker after a longer delay. We extracted values, for each participant, from the contrast Whole faces minus Distracters (Two halves and One half) and Novels, from the right fusiform gyrus (peak: 39x, -45y, -21z) and from the hippocampus (peak: 24x, -27y, -15z). Then, we performed correlation analysis between these two values, within each group of subject separately.
We found a significant positive correlation between hippocampal and fusiform activity in the night group (R² = .2971, p = .05; Figure 6). There was no correlation at all in the day group (day group: R² = .0020, p = .88). The 24 hours group showed a tendency towards a correlation (R² = .1488) but this was not significant (p = .27).
These results suggest a period of 12 hours with intervening sleep might strengthen hippocampal-fusiform interactions underlying the recognition of newly learned face identities, and that these interactions tend to weaken across longer periods of time.
R2= 0.002
Figure 6: Correlation analysis between fusiform and hippocampal activity, for the day group, night group and 24 hours group separately. Each dot represents a single subject.
5.4. Discussion
The present fMRI study had two main aims: (1) to assess whether sleep could improve explicit memory for faces; (2) to test whether this potential beneficial effect of sleep might rely on a strengthening of the configural information in faces; (3) to investigate whether sleep modulates activity in the hippocampus and in face-selective brain areas. In summary, results showed higher recognition accuracy and configural processing for subjects who slept between learning and testing sessions. This effect was associated with a sleep-related increase in right hippocampal activity selectively for Whole faces (same features, same configuration).
Activity in the right fusiform gyrus showed increased activity for Whole faces compared to the One half face distracters (same features, different configuration) in the sleeping groups, consistent with an involvement of this brain region in the processing of configural information in faces. Finally, activity changes in hippocampus and fusiform correlated selectively during the processing of old faces.
These findings show a beneficial effect of sleep on explicit memory, as also found in a word memory task (Drosopoulos et al., 2005), and in a face recognition task (Wagner et al., 2007).
The effect we found could not be related to a change in recognition criteria. They could not be related either to circadian effects, as both the night group and the 24hours group, tested in the morning for the night group, and in the evening for the 24 hours group, both showed increased recognition performance. Finally, behavioral performance of the 24 hours groups allows to consider that the increase performance in the night group was not linked, or not only linked, to a lack of sensory interference during the period of sleep.
In this study, sleep does not only increase the recognition performance for the previously learned faces, but also enhances the accuracy of the recognition. Contrary results could have been found, given the increased familiarity found by Jacoby et al. (1989). They observed that, after 24h with intervening sleep, names encountered once had a higher probability to be judged “famous” as newly encountered names, in a famous/non-famous discrimination task.
A pilot study was realized using the Remember-Know paradigm (Yonelinas, 1999), in order to disentangle familiarity/recollection contributions to memory. Unfortunately, we could not obtain reliable data, the number of items falling in the different response categories being too heterogenous to sustain reliable fMRI statistical analysis. However, our data using a simple old-new recognition test were consistent with a very recent study showing sleep-related decrease of false recognition for word memory (Fenn et al., 2009). These authors argue that sleep may consolidate item-specific details, thus facilitating accurate recollection.
In our study, the item-specific memory consolidation seems to rely on a strengthening of the
effect. The effect was related in the night group to an increased activity in the right fusiform gyrus, and more exactly in the so-called FFA. This confirmed the role of this region in configural processing, as previously mentioned by (Rossion et al., 2000). The OFA, the other face-selective area, showed increased activity for the distracters in the day group. Thus, the day group seems to rely more on features whereas the 2 other groups might rely on configural, more abstract memory traces. Linking our results to the Bartlett’s model of face recognition (Bartlett et al., 2009), we could assume that sleep favors the configural processing component, lying in the FFA, whereas a comparable period of wakefulness trigger the part-based processing component, lying in the OFA.
Finally, this study shows that the effect of sleep on face memory is associated with increased correlation in brain activity between the hippocampus and the FFA. This was true for the night group but not any more for the 24 hours group, even if the correlation slope was much sharper in the 24 hours group than in the day group. Regarding the standard consolidation theory (Dudai, 2004, 2006), our hypothesis was that memory for faces was consolidated through hippocampal-fusiform connections, which may be stronger after a night of sleep, and weaker after a longer delay. These results suggest that a period of 12 hours with intervening sleep might consolidate face memory through strengthened hippocampal-fusiform interactions in a first place, and through a progressive disengagement of the connections in the longer term. Frontal regions, playing a crucial role in the prevention of false memory for faces (Rapcsak et al., 2001) might be involved in these longer-term memory improvement, as suggested by the increased frontal activity for the 24 hours group compared to both the day group and the night group.
To conclude, sleep might be an optimal state for consolidation process to occur. Importantly, these findings show that sleep make one’s brain more resistant to misleading information.
Acknowledgements
We thank P. Maquet for helpful comments and F. Andersson, K. Ndiaye, and A. Achaibou for technical assistance during the scanning sessions. This work has been supported by the Swiss National Science Foundation (grant 310000-114008 to S.S.) and the Schmidheiny and Boninchi Foundations (grants to S.D.).
GENERAL DISCUSSION