LETTERS Boosting slow oscillations during sleep potentiates memory Lisa Marshall1, Halla Helgadottir1, �� Matthias Molle1 �� & Jan Born1 There is compelling evidence that sleep contributes to the long- term consolidation of new memories1. This function of sleep has been linked to slow (,1 Hz) potential oscillations, which pre- dominantly arise from the prefrontal neocortex and characterize slow wave sleep2���4. However, oscillations in brain potentials are commonly considered to be mere epiphenomena that reflect syn- chronized activity arising from neuronal networks, which links the membrane and synaptic processes of these neurons in time5. Whether brain potentials and their extracellular equivalent have any physiological meaning per se is unclear, but can easily be investigated by inducing the extracellular oscillating potential fields of interest6���8. Here we show that inducing slow oscillation- like potential fields by transcranial application of oscillating potentials (0.75 Hz) during early nocturnal non-rapid-eye-move- ment sleep, that is, a period of emerging slow wave sleep, enhances the retention of hippocampus-dependent declarative memories in healthy humans. The slowly oscillating potential stimulation induced an immediate increase in slow wave sleep, endogenous cortical slow oscillations and slow spindle activity in the frontal cortex. Brain stimulation with oscillations at 5 Hz���another fre- quency band that normally predominates during rapid-eye-move- ment sleep���decreased slow oscillations and left declarative memory unchanged. Our findings indicate that endogenous slow potential oscillations have a causal role in the sleep-associated consolidation of memory, and that this role is enhanced by field effects in cortical extracellular space. Slow oscillations reflect widespread ���up��� and ���down��� states of net- work activity. These oscillations are generated within the neocortex and are most prominent during slow wave sleep (SWS) the up and down states reflect, respectively, global neuronal depolarization with excitation, and neuronal hyperpolarization with neuronal silence2,3,9. Essentially owing to its synchronizing influence on neuronal activity within the neocortex and in dialogue with thalamic and hippocampal circuitry, the slow oscillation has been suspected to underlie the consolidation of memory during sleep3,10���13. The slow oscillation signal peaks at 0.7���0.8 Hz, although spectral components can extend into the slow delta band (1���4 Hz)9,14. We have examined the role of slow oscillations in memory consolidation by inducing them through transcranial application of oscillating potentials during early noc- turnal non-rapid-eye-movement (non-REM) sleep after a learning period. Hippocampus-dependent declarative memory was assessed by a paired-associate learning task, with memory retention measured by the difference in the number of words recalled when tested after sleep and the number recalled at learning before sleep. As expected from previous studies using the same procedure8,15, retrieval testing after sleep showed an increase in performance, compared to learning before sleep, in both the slow oscillation stimulation and the sham stimulation conditions (Fig. 1b). However, after slow oscillation stimulation this increase in memory (mean 4.77 words) was greater than that following sham stimulation (mean 2.08 words F1,12 5 7.96, P 5 0.01). In the stimulation condition, 36.50 6 1.24 words were recalled at learning before sleep, and this number increased to 41.27 6 1.21 at retrieval after sleep. In the sham condition perform- ance increased from 37.42 6 0.92 (learning) to 39.50 6 0.84 words after sleep. This improvement in retention following stimulation is striking considering that most subjects were medical students, who were highly trained in memorizing facts and already performed well in the sham condition. Note that our retention measure does not allow us to differentiate between a slowed decay and an actual gain in memory after stimulation16. To test whether slow oscillation stimulation specifically affected the formation of hippocampus-dependent declarative memory, we also tested subjects on a non-declarative, procedural finger-sequence tapping task17. Retrieval testing after sleep confirmed the character- istic overnight improvement in skill in both conditions (number of 1University of Lu ��beck, Department of Neuroendocrinology, Haus 23a, Ratzeburger Allee 160, 23538 Lu ��beck, Germany. 23 :0 8 00 :0 8 01 :0 8 02 :0 8 03 :0 8 04 :0 8 05 :0 8 06 :0 8 W S1 S2 M T im e Declarative, non-declarative, control tests Stimulation ** 1 2 3 4 5 1 2 Sham Stimulation Stimulation Sham Speed 0 3 Recalled words 0 6 Learning Recall 4 3 2 1 REM W b c Declarative, non-declarative, control tests a Figure 1 | Slow oscillatory stimulation enhances declarative memory performance. a, Time-course of experiment. Indicated are time points of learning and recall of memory tasks, psychometric control tests, stimulation intervals, period of lights off (horizontal grey bar), and sleep represented by a hypnogram. W, wake 1���4, sleep stages 1���4. b, Performance on the declarative paired-associate memory task across the retention period of nocturnal sleep for stimulation and sham stimulation. Performance is expressed as difference between the number of correct words reported at recall testing and learning. The list contained 46 experimental word-pairs (**P , 0.01). c, Performance speed on the non-declarative procedural motor skill task across the retention interval expressed as the difference in the number of correctly tapped sequences per 30 s between recall testing and learning. Data are the means 6 s.e.m. Vol 444|30 November 2006|doi:10.1038/nature05278 610 Nature Publishing Group ��2006

correctly tapped sequences before sleep 17.74 6 1.30 (stimulation), 18.15 6 1.28 (sham), and at retrieval after sleep 19.77 6 1.52 and 20.69 6 1.46, respectively F1,12 5 67.70, P , 0.001). However, in contrast to declarative memory performance, the sleep-associated gain in performance (2.03 6 0.65 sequences) was not enhanced through slow oscillation stimulation (P . 0.6 Fig. 1c). Also, over- night changes in error rate did not differ between the two conditions (P . 0.25). Performance on two additional tasks, a declarative non- verbal paired-associate task and a procedural mirror-tracing task, likewise indicated that stimulation led to an improvement only for the declarative task (see Supplementary Information for a figure summarizing the main result). The ability of slow oscillation stimu- lation during early non-REM sleep to enhance retention of word- pairs and its failure to affect procedural skill are consistent with reports that hippocampus-dependent memories benefit mainly from early SWS, and procedural memories from REM sleep (which pre- vails during late sleep), although non-REM sleep might have com- plementary functions1,18,19. The efficacy of polarization over the prefrontal cortex in our study is in line with this region���s importance in the hippocampal���neocortical dialogue that is assumed to underlie the consolidation of hippocampus-dependent memories20. In a control experiment (n 5 8) using a protocol identical to that of the main experiment we shifted the timing of stimulation to the period shortly before awakening (05.45���06.15 h), that is closer to retrieval testing, which should increase any immediate non-specific effects of stimulation on cognitive function during retrieval21. However, under this condition retention of word-pairs remained unchanged and was similar to retention after sham stimulation (post-sleep retrieval with reference to learning: 3.21 6 1.43 versus 2.93 6 1.21 words, P . 0.8). These and further control tests of vigilance and general retrieval capabilities (see Supplementary Information) safely exclude any substantial non-specific contri- bution of slow oscillation stimulation to the improved declarative memory at retrieval testing. We examined sleep and electroencephalogram (EEG) activity more closely to understand the mechanisms that underlie the enhancement of memory performance. During the 5-min periods of acute stimulation, the induced potentials precluded sleep scoring (Fig. 2). However, the 1-min stimulation-free intervals yielded clear signals. During these intervals, more total time was spent in SWS after slow oscillation (170.77 6 17.78 s) than after sham stimulation (124.62 6 19.17 s, F1,12 5 6.03, P , 0.05 Table 1). The times spent in the different sleep stages during the 60 minutes after stimulation and for the whole night were comparable between stimulation and sham conditions (Supplementary Table 1). The increase in SWS is a plausible explanation for the ability of slow oscillatory stimulation to improve memory, particularly as this increase was presumably also present during the periods of acute stimulation. However, visual scoring of SWS relies strongly on an undifferentiated estimation of the presence of slow wave rhythms. We suspected that the transient increase in SWS during the 1-min intervals between periods of stimu- lation reflected a temporally limited enhancement of EEG slow oscil- lations, and that this was the specific process that, during this sleep stage, mediated the memory improvement. Indeed, spectral analysis of EEG activity during the five 1-min intervals between stimulation periods confirmed that stimulation acutely facilitated endogenous slow oscillations. Stimulation dis- tinctly enhanced EEG power within the slow oscillation band (0.5��� 1.0 Hz, F1,12 5 11.67, P , 0.01 at the frontocentral recording site, Fz Fig. 3a). A slight increase in power in adjacent low (1���1.5 Hz) delta frequencies and in the 1���4-Hz delta band failed to reach significance (P . 0.06, for all comparisons) indicating that the effect of stimu- lation was focused on the slow oscillation band. Interestingly, slow oscillation stimulation simultaneously enhanced EEG power within the slow spindle frequency range (8���12 Hz, peaking at ,10.5 Hz, F1,12 5 13.12, P , 0.01 at Fz Fig. 3a) as well as spindle counts (see Supplementary Information). The effects were most pronounced for the first three inter-stimulation intervals (Fig. 3b), with a prefrontal maximum, although they spread to the other recording sites (F1,12 . 7.43, P , 0.02, for overall effects of stimulation). The con- junct increase in slow oscillation and frontal spindle activity agrees well with the notion that neocortical slow oscillations drive the thalamic generation of spindles3,9,14 and emphasizes that stimulation induces a physiologically coherent pattern of activity in this system. 1 s 1 s 1 s 5 ��V 25 ��V 25 ��V a b Figure 2 | Synchronization of slow oscillatory EEG activity. a, EEG recordings during the last seconds of a 5-min stimulation period (shaded areas) and first few seconds of a stimulation-free interval of two individuals at prefrontal sites (Fz). b, Corresponding mean 6s.e.m. across all subjects and stimulation periods over the parietal cortex (where the EEG is least contaminated by the ceasing stimulation artefact). Positivity upward. Note entrainment of the slow oscillatory EEG activity to the slow oscillatory rhythmic stimulation. Hatched bar indicates time interval of stimulation- induced phase changes in the 0.78���0.98-Hz and 1.37���1.56-Hz bins of the EEG signal. Table 1 | Sleep parameters during transcranial stimulation and sham conditions Slow oscillation stimulation (n 5 13) Stimulation (mean 6 SEM) Sham (mean 6 SEM) Theta stimulation (n 5 5) Stimulation (mean 6 SEM) Sham (mean 6 SEM) Awake 0.8 6 0.8 1.5 6 1.5 Awake 12.0 6 12.0 0.0 6 0.0 S1 17.7 6 10.3 30.8 6 12.1 S1 18.2 6 18.2 24.1 6 24.1 S2 110.8 6 14.2 143.1 6 20.4 S2 181.6 6 38.0 144.1 6 30.6 S3 113.1 6 18.3 85.4 6 16.2 S3 59.3 6 34.2 108.1 6 24.4* S4 57.7 6 17.1 39.2 6 18.7 S4 29.8 6 13.4 23.9 6 17.5 SWS 170.8 6 17.8 124.6 6 19.2* SWS 89.1 6 42.4 132.0 6 35.0 Time (s) spent in different sleep stages during the 1-min stimulation-free intervals between the periods of stimulation for the main experiment (top) testing slow oscillation stimulation (0.75 Hz) and in a supplementary experiment (bottom) testing theta stimulation (5 Hz). Sleep scoring based on 10-s intervals. Awake and S1 sometimes occurred in just one case. Asterisk indicates a significant (P , 0.05) increase in SWS with slow oscillation stimulation and a significant (P , 0.05, one-tailed) decrease in S3 sleep with theta stimulation. S1���S4, sleep stages 1���4. NATURE|Vol 444|30 November 2006 LETTERS 611 Nature Publishing Group ��2006