Participants were similar in habitual sleep characteristics based on PSQI reports, a subjective measure of sleep quality over the past 30 days (Sleep group mean PSQI score=4.8; Wake group mean PSQI score=5.4; F(1,80)= .97, p=.33). Subjectively estimated habitual sleep latency (F(1,80)= .16, p=.69) and habitual sleep duration (F(1,80)= .05, p=.82) also did not differ for PSQI reports across groups.
Full night PSG was measured for 25 of the participants in the Sleep group. Total PSQI, habitual sleep latency, and habitual sleep duration did not differ between Sleep groups with and without PSG (all p’s> .49). Subjective reports of total sleep time for the experimental night did not differ for those Sleep group participants with PSG (mean=403.5 min, SD= 68.1) and those without PSG (mean=362.7 min, SD=96.6) although this trended towards significance (F(1,52)= 3.12, p=.08). Note that, given this trend, behavioral measures were compared between Sleep group participants with and without PSG. Recognition accuracy and valence ratings did not differ between groups with the exception of arousal ratings which showed a greater decrease in the no PSG group, F(1,52)= 5.1, p=.03. Sleep latency was 16.8 min (SD= 13.4) for the Sleep group without PSG as reported in sleep diaries. This did not differ from subjective sleep latency for the Sleep group with PSG, 22.6 min (F(1,52)= 1.81, p= .19). Mean sleep efficiency in the sleep group participants with PSG was 92.84% (SD= 6.19).
Within the PSG group, a comparison of self-reported sleep time and sleep time as measured by PSG neared significance (t(24)= −1.67, p=.11) with participants’ subjective reports tending to underestimate total sleep time. Such misperceptions have been observed by others (e.g. Silva et al., 2007
; Carskadon et al., 1976
). PSG measured mean sleep efficiency was 82.84% (SD=6.19). Participants spent 23% (SD= 18.52) of total sleep time in NREM Stage 1 sleep, 35% (SD=15.12) in NREM Stage 2, 21.78% (SD=6.38) in slow wave sleep (SWS) and 23.67% (SD=9.28) in REM sleep.
presents hit rate (HR) and false alarms (FA) across groups and emotional valence categories. An ANOVA on HR revealed that the main effect of group (Sleep vs. Wake) was significant, F(1,80)=10.27, p=.002. That is, for both negative and neutral pictures, the Sleep group performed better than the Wake group (). The main effect of Valence was also significant with negative pictures remembered better than neutral pictures, F(1,80)=38.65, p<.001. The interaction between Valence and Group was not significant, F(1,80)=1.15, p=.29, suggesting that the benefit of sleep after encoding was not modulated by stimulus valence.
Memory accuracy across groups
Figure 2 Performance and sleep physiology measures. (A) Recognition accuracy in Session 2. (B) ΔValence was computed as the SAM Valence ratings for Session 2 minus SAM valence ratings for Session 1. (C) ΔArousal was computed as the SAM Arousal (more ...)
An ANOVA on FA revealed a significant main effect of Group, with Sleep group participants having fewer false alarms than the Wake group, F(1,79)=4.91, p=.03. Neither the main effect of Valence (F(1,79)= 1.25, p= .23) nor the Valence × Group interaction (F(1,79)= .69, p= .41) were significant.
Individual differences in habitual sleep quality did not confound group comparisons for recognition accuracy. By adding total PSQI scores to the mixed ANOVA model, the Group main effect remained significant F(1,80)=6.45, p=.013) with neither PSQI main effect nor PSQI × Group interactions. Given that gender was not balanced across groups, Gender was also added to the model. The Group main effect remained significant, F(1,80)=12.68, p<.001, but only a trend toward a Gender main effect (F(1,80)=2.96, p=.09, Males greater) emerged. Importantly, there were no significant Group × Gender or Valence × Group × Gender interactions (both p’s > .12).
It is possible that the analyses of Hit Rate reflect differences in response bias rather than memory. Based on the Signal Detection Theory (e.g. Macmillan & Creelman, 1991), memory discrimination is most accurately summarized by measures of sensitivity (the ability to discriminate hit rates and false alarms; d’) and response bias (tendency for conservative v. liberal responding; c). An ANOVA on d’ revealed a significant main effect of Group with better recognition memory discriminability in the Sleep group (F(1,79)= 14.63, p <.001), a significant main effect of Valence with better memory for negative pictures (F(1,79)= 20.87, p <.001) and no significant Group × Valence interaction (F(1,79)= .99, p =.32). Response bias, c, was similar between Sleep and Wake groups (F(1,79)= .05, p =.82). There was a main effect of Valence with more conservative response tendency for neutral pictures (F(1,79)= 25.32, p <.001) but no Group × Valence interaction (F(1,79)= .01, p =.93).
Valence and Arousal
Session 1 subjective reports of valence and arousal allow us to probe baseline differences between groups, at least with respect to emotional reactivity. Importantly, there were no baseline differences. Sleep and Wake groups rated neutral (F(1,79)=.20, p=.65) and negative (F(1,79)=.26, p=.61) items similarly for Valence. Likewise, Arousal ratings were similar for Sleep and Wake groups (neutral: F(1,79)=1.65, p=.28; negative: F(1,79)= 3.53, p=.074). Although we note the trend towards significance of the latter comparison, by focusing on delta (ΔArousal), the change in ratings over the break, this possible difference is accounted for. Notably, this analysis also suggests that the subjective ratings did not differ by time of day, discounting circadian influences on this task.
The ANOVA for ΔValence revealed a significant Group × Valence interaction, F(1,80)= 3.81, p= .054. ΔValence for negative pictures was higher in the Wake group (mean =.65, SEM=.09) as compared to the Sleep group (mean=.39, SEM = .07) indicating that the initial negative ratings were attenuated more (i.e., more neutral) after a period of wake relative to sleep (). For neutral pictures, ΔValence was similar across Sleep (mean= −.14, SEM= .04) and Wake (mean= −.13, SEM= .06) groups.
Likewise, arousal ratings for negative pictures were more attenuated over the 12-hr break for the Wake group (mean ΔArousal=−1.17; SEM= .27) as compared to the Sleep group (mean =−.60; SEM = .18). ΔArousal for neutral pictures was reduced more over wake (mean=−.22; SEM=.21) than sleep (mean=−.15; SEM=.12; ). However, the main effect of group was not significant (F(1,80)= 1.51, p=.22) nor was the Group × Valence interaction (F(1,80)= 1.29, p=.25).
Given that initial emotionality for targets and foils was similar (based on a preliminary study), we compared session 2 ratings for targets (‘old’ stimuli) and foils (‘new’ stimuli), a measure of sleep-dependent emotional processing used previously (Wagner et al., 2002
). Specifically, the mean rating of foils was subtracted from the mean rating of targets in each valence category. The target-foil difference for mean valence did not differ for Sleep and Wake groups (F(1,80)= .22, p
= .64) nor was there a Group × Valence interaction (F(1,80)= .16, p
= .69). Similarly the target-foil difference in mean arousal ratings did not differ for the Sleep and Wake groups (main effect of Group, F(1,80)= 2.05, p
= .16; Group × Valence interaction, F(1,80)= .34, p
= .56). This may suggest some level of generalization of the emotional response for targets to foils. However, caution must be taken in interpreting this response. While normative data were used to balance targets and foils a priori, by basing our analyses on subjective categorization, targets and foils may differ in subjective emotionality for the experimental groups and targets and foils were not balanced for number of items. Moreover, old items are being rated for the second time at test, and may innately differ from new items that are rated for the first time at test.
Relationship between Memory and Valence
Of particular interest in the present study was whether changes in memory are associated with changes in emotional reactivity over sleep. ΔValence did not significantly correlate with hit rate for negative pictures (Wake: r=.25, p=.19; Sleep: r=−.096, p=.49). Likewise, for neutral pictures, the correlations between ΔValence and hit rate were not significant for either the Sleep (r=.07, p=.62) or Wake (r=.09, p=.65) groups. ΔArousal was also not a significant predictor of hit rate for negative (Sleep: r=−.17, p= .22; Wake: r=−.19, p=.32) or neutral pictures (Sleep: r=−.15, p=.27; Wake: r=.06, p=.78). Likewise, when d’ is used as a measure of memory, all correlations are non-significant.
Associations between Behavior and Sleep Physiology
Late-night REM has been hypothesized to be critical to emotional memory consolidation (Wagner et al., 2001
). Thus, we focused our comparisons on behavioral measures of negative pictures and measures of REM sleep (). We found no significant relationship between REM sleep time and hit rate for negative pictures (r= −.09, p=.67), ΔValence (r= −.15, p=.48), nor ΔArousal (r= .06, p=.78).
Correlations between REM sleep measures and memory, ΔValence and ΔArousal for negative pictures
Given that Wagner et al. (2001)
measured REM in a 3 hr sleep opportunity late in the night, with participants achieving 30 mins of REM sleep, we focused further on REM sleep in the third quarter of the night (REM3%) with the justification that this may be most similar to the 30 mins obtained by Wagner’s participants. Supporting this, our participants averaged 30.78 mins of REM in the third quarter. We also separately considered REM4% as REM is thought to be richer as the night progresses (Mallick et al., 2011
). REM3% correlated significantly with ΔValence; notably, contrary to predictions (e.g., Walker, 2009
), this relationship was negative (r= −.41, p=.04). That is, those participants with more REM3% had less attenuation of negative reactivity (). ΔArousal did not correlate significantly with any of the REM measures (all p
’s>.28). Likewise, none of the correlations between REM4% and behavioral measures were significant (). Given that REM latency has also been shown to predict emotional memory processing (Nishida et al., 2009
), we examined whether this measure (time to enter REM), predicted behavioral measures. We found no significant correlations between REM latency and hit rate (r =.17; p
=.43), ΔValence (r =.22; p
=.28) or ΔArousal (r =−.29; p
=.16). Finally, a multiple regression analysis showed that neither ΔValence, ΔArousal, nor REM (REM3%, REM4%, REM latency) were significant predictors of hit rate for negative pictures (p’s
>.15 for all coefficients; adjusted R2
= .06, n.s.).
Given the role of SWS on memory consolidation, we investigated the relationship of SWS and behavioral measures, separately for negative and neutral pictures. Neither total SWS time nor SWS1% reliably predicted hit rate, ΔValence, or ΔArousal (all p’s> .21, see ). Likewise, SWS2% did not significantly correlate with behavioral measures with the exception of hit rate for neutral pictures where this correlation was, notably, negative (r= −.44, p=.02).
Finally, we probed the relationship between NREM Stage 2 and behavioral measures. NREM Stage 2 contains sleep spindles that are thought to underlie neural plasticity (Andrade et al., 2011
). NREM Stage 2 in the last two quarters of the night did not correlate with any behavioral measures nor were significant correlations observed between behavior and total NREM Stage 2 (all p
’s > .46). All correlations are in the same direction and non-significant when d’ is used.
The Morning and Evening groups were included to examine whether differences observed between the Sleep and Wake groups reflect circadian influences on cognition. A comparison of valence and arousal ratings in the Encoding Phase, fails to support a circadian account: The Morning and Evening groups rated Valence (negative: F(1,22)=.19, p=.66; neutral: F(1,22)=.06, p=.81) and Arousal (negative: F(1,22)= 1.58, p=.22; neutral: F(1,22)=.14, p=.71) similarly.
A mixed ANOVA for hit rate revealed that the main effect of Group (Morning vs. Evening) was not significant, F(1,22)=2.69, p=.12. The Valence × Group interaction was also not significant, F(1,22)=.22, p=.64. However, there was a main effect of Valence, F(1,22)= 4.07, p=.04. Negative pictures were remembered better than neutral pictures as seen in the 12h groups as well.
Similarly, ANOVA for recognition memory discriminability (d’) revealed no significant main effect of group (F(1,22)= 2.73, p=.12) and no Group × Valence interaction (F(1,22)= .24, p=.63). However, there was also no main effect of Valence in d’ measures, F(1,22)= .89, p=.36. ANOVA for response bias (c) revealed no significant main effect of Group (F(1,22)= 1.31, p=.27), no main effect of Valence (F(1,22)= 1.71, p=.21) and no significant Group × Valence interaction (F(1,22)= 3.42, p=.08).
A mixed ANOVA for ΔValence revealed no significant effect of Group (Morning vs. Evening), F(1,22)= .30, p=.60. ΔValence for negative pictures was significantly different from zero in both groups (Evening: t(11)= 4.82, p=.001; Morning: t(11)= 6.44, p<.001). Thus, even a 45-min interval was enough to reduce the negativity of the perceived valence (i.e., items are rated as less negative). Interestingly, ΔValence over 45 mins awake (Morning and Evening combined) was similar to the ΔValence observed following 12 hrs awake (F(1,50)=.64, p=.42) and was not significantly different from ΔValence for the 12h Sleep group (F(1,76)= 1.98, p= .16).
ΔArousal ratings also did not differ for the Morning and Evening groups (F(1,20)= .41, p=.53). For both Morning and Evening groups ΔArousal for negative pictures was not different from zero (Evening: t(11)= −.56, p=.59; Morning: t(11)= .73, p=.48). Although ΔArousal did not change within 45 mins, following 12 hrs of wake, ΔArousal was greatly reduced (45 min groups v. Wake group; F(1,50)=10.9, p=.002). ΔArousal following 45 min wake was not significantly different from ΔArousal following 12h sleep (F(1,76)= 3.25, p= .08). Together, these results suggest a differential effect of time on valence and arousal.