HIPPOCAMPUS 24:1157–1168 (2014)
Hippocampal Slow EEG Frequencies During NREM Sleep are Involved in Spatial Memory Consolidation in Humans Fabio Moroni,1,2* Lino Nobili,3,4 Giuseppe Iaria,5 Ivana Sartori,3 Cristina Marzano,1 Daniela Tempesta,6 Paola Proserpio,3 Giorgio Lo Russo,3 Francesca Gozzo,3 Carlo Cipolli,2 Luigi De Gennaro,1 and Michele Ferrara6
ABSTRACT: The hypothesis that sleep is instrumental in the process of memory consolidation is currently largely accepted. Hippocampal formation is involved in the acquisition of declarative memories and particularly of spatial memories. Nevertheless, although largely investigated in rodents, the relations between spatial memory and hippocampal EEG activity have been scarcely studied in humans. Aimed to evaluate the effects of spatial learning on human hippocampal sleep EEG activity, we recorded hippocampal Stereo-EEG (SEEG) in a group of refractory epilepsy patients undergoing presurgical clinical evaluation, after a training on a spatial navigation task. We observed that hippocampal high-delta (2–4 Hz range) activity increases during the first NREM episode after learning compared to the baseline night. Moreover, the amount of hippocampal NREM high-delta power was correlated with task performance at retest. The effect involved only the hippocampal EEG frequencies inasmuch no differences were observed at the neocortical electrodes and in the traditional polysomnographic measures. The present findings support the crucial role of hippocampal slow EEG frequencies during sleep in the memory consolidation processes. More generally, together with previous results, they suggest that slow frequency rhythms are a fundamental characteristic of human hippocampal EEG during both sleep and wakefulness, and are related to the consolidation of different types of memories. C 2014 Wiley Periodicals, Inc. V
KEY WORDS: hippocampus; stereo-EEG; declarative memory; memory consolidation; local sleep
1
Department of Psychology, “Sapienza” University of Rome, Roma, Italy; 2 Laboratory of Psychology, Department of Specialized, Diagnostics and Experimental Medicine, University of Bologna, Bologna, Italy; 3 Centre of Epilepsy Surgery ‘‘C. Munari’’, Center of Sleep Medicine, Niguarda Hospital, Milan, Italy; 4 Institute of Bioimaging and Molecular Physiology, Section of Genoa, National Research Council, Genova, Italy; 5 Department of Psychology and Clinical Neurosciences, and Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada; 6 Department of Life, Health and Environmental Sciences, University of L’Aquila, L’Aquila, Italy Grant sponsor: Compagnia di San Paolo, Programma Neuroscienze 2008/09; Grant number: 3896 SD/sd; 2008.2130 (to M.F.); Grant sponsor: ESRS Sanofi-Aventis Research Grant 2008/10 (to F.M.), and Fondazione del Monte di Bologna e Ravenna 2011/13 (to C.C.). *Correspondence to: Fabio Moroni; Department of Psychology, “Sapienza” University of Rome, Via dei Marsi, 78, 00185 Rome, Italy. E-mail: [email protected] Accepted for publication 23 April 2014. DOI 10.1002/hipo.22299 Published online 2 May 2014 in Wiley Online Library (wileyonlinelibrary.com). C 2014 WILEY PERIODICALS, INC. V
INTRODUCTION The medial temporal lobe (MTL), and particularly the hippocampal formation, has been considered central for declarative memory in humans, playing a key role from the initial phase of memory formation to the final storage in widespread brain areas (Eichenbaum, 2004). Sleep, as emerged from a growing number of experimental studies, critically and actively supports the process of memory consolidation. In particular, new and weak memory traces are strengthened and transformed into a more stable and persistent form by sleep-dependent processes (Tononi and Cirelli, 2006; Diekelmann and Born, 2010). From these studies emerged that declarative memory and, in particular, spatial learning specifically benefits from NREM sleep (for a review, see Rasch and Born, 2013). To date, our knowledge about hippocampal neurophysiological activity related to memory formation mostly derives from animal studies on spatial learning, which is strictly dependent on the hippocampal formation (Squire et al., 2004). At the cellular level, it has been observed that neuronal ensembles activated during the waking behavior are then re-activated during posttraining slow-wave sleep (SWS) (e.g., Pavlides and Winson, 1989; Wilson and McNaughton, 1994; Skaggs and McNaughton, 1996) and REM sleep (Poe et al., 2000; Louie and Wilson, 2001). This "replay" of place cells activity during sleep has been interpreted as a process of memory trace strengthening and transferring towards other cerebral areas (Buzsaki, 1989, 1998; Born and Wilhelm, 2012). In humans, place cells active during virtual navigation and recall of navigation related memories have been recently identified (Miller et al., 2013). On the other hand, reactivation of place cells during sleep has not been replicated yet, although a neuroimaging study has reported a hippocampal reactivation during SWS, but not during REM sleep, following spatial learning that correlated with task performance improvement (Peigneux et al., 2004). From an electrophysiological point of view, rodent’s hippocampal EEG activity during movement and exploratory behavior, as well as during REM sleep, has shown to be characterized by the rhythmic slow activity (RSA or theta: 4–10 Hz). RSA is considered
1158
MORONI ET AL.
crucial in memory encoding, probably fulfilling the role of phase-synchronizer for neuronal ensembles belonging to different brain areas that can discharge together under the coordination of theta rhythm (Buzsaki, 2002; Battaglia et al., 2011). In humans, hippocampal electrophysiological activity has been mainly investigated by means of intracerebral recordings carried out in refractory epilepsy patients undergoing presurgical clinical evaluation. These studies have reported an increase of theta activity in the neocortex (Kahana et al., 1999; Caplan et al., 2001, 2003) and in the hippocampus (Ekstrom et al., 2005; Watrous et al., 2011) during spatial navigation tasks in virtual environments. Nevertheless, the existence of a prominent 2–4 Hz rhythm, below the theta range, in the spontaneous human hippocampal EEG activity has been repeatedly described during both resting wakefulness and sleep (Bodizs et al., 2001; Moroni et al., 2007, 2012; Clemens et al., 2009), and recently also during the execution of a spatial navigation task (Clemens et al., 2013). Hence, the prominent presence in human hippocampal recordings of slow (delta) EEG rhythms seems to be a relevant emerging finding. Indeed, in a recent study we described a peculiar bimodal distribution of hippocampal activity in the delta range, made up by a low frequency nonoscillatory activity (up to 2 Hz) and a faster oscillatory rhythm (2–4 Hz) (Moroni et al., 2012). Interestingly, we observed that this faster oscillatory activity at around 3 Hz is present, in the human spontaneous hippocampal SEEG signal, not only during wakefulness and REM sleep but even during NREM sleep. Previous studies have hypothesized that human hippocampal activity in the 2–4 Hz range could be the analogue of animal RSA, being related to behavioral activated states and REM sleep (Moroni et al., 2012; Clemens et al., 2013). Since we have observed that, at variance with animal literature, in humans this rhythm is largely present also during NREM sleep (Moroni et al., 2012), we asked which could be its peculiar functional role during NREM sleep. In particular, in this study we asked whether this hippocampal slow activity could be related to spatial memory consolidation processes during sleep. To this aim, we administered to a sample of epileptic patients undergoing presurgical examination a spatial memory task, namely the Cognitive Map Test (CMT) that strictly depends on hippocampal activity (Iaria et al., 2007; Ferrara et al., 2008). In particular, by recording intracranial EEG activity during both CMT execution and post-training night sleep, we evaluated the effects of task training on the hippocampal EEG activity during subsequent sleep, under the assumption of a specific involvement of the EEG frequencies below the theta range. The relations between sleep EEG frequencies and postsleep task performance were also evaluated.
MATERIALS AND METHODS Participants We recruited eight patients (6 M, 2 F; mean age: 26.75; age range: 18–38) with pharmacoresistant focal epilepsy. All Hippocampus
patients were candidates for surgical removal of the epileptic focus and underwent individual investigation with stereotactically implanted intracerebral multilead electrodes (SEEG) for the precise localization of the epileptogenic zone (Cossu et al., 2005). Patients were selected based on the presence of at least two electrode contacts localized within the hippocampus. We excluded from the study patients who did show features of hippocampal sclerosis. We also excluded patients with severe cognitive impairment; that is, scoring under the 95th percentile at the Raven’s colored progressive matrices (CPM) (Raven, 1996). Throughout the study, patients took their standard doses of anticonvulsant medications. The study protocol was approved by the Local Ethics Committee (Niguarda Hospital, Milan, Italy). Prior to SEEG electrode implantation, patients gave written informed consent for participation in this study and for publication of data. Two of the eight selected patients were discarded because considered outliers, since their task performance at postsleep retest did not fulfill the predetermined criterion of a decrease not larger than three standard deviations from that of a normative sample (see below). Therefore, all the subsequent analyses have been carried out on a sample of 6 patients (see Table 1).
Electrodes Placement and EEG/SEEG Recordings Stereo-EEG activity was recorded from platinum-iridium semiflexible multilead depth-electrodes, with a diameter of 0.8 mm, a contact length of 2 mm and an intercontact distance of 1.5 mm (Dixi Medical, Besancon France). The placement of electrode contacts was ascertained by postimplantation magnetic resonance imaging (MRI) scans. For the entire sample, we considered for analyses contacts localized within the anterior hippocampus of the right hemisphere (see Fig. 1). As the neocortical leads are concerned, the localization was more sparse and we selected contacts in the right hemisphere within the frontal lobe (middle frontal gyrus) in three patients, central operculum in two patients and temporal lobe (inferior temporal gyrus) in one patient (for details on locations see Table 1). Scalp EEG activity was recorded from two platinum needle electrodes placed during surgery at "10–20" positions Fz and Cz on the scalp. Electroocular (EOG) activity was registered at the outer canthi of both eyes and submental electromyographic (EMG) activity was acquired with electrodes attached to the chin. EEG and SEEG signals were recorded using a 192-channel recording system (Nihon-Kohden Neurofax-110) with a sampling rate of 1,000 Hz during resting wake and virtual navigation task, and using a 24 channels ambulatory system recording (XLTEK, TrexTM) with a sampling rate of 512 Hz during the baseline and experimental nights of sleep. Recording data were handled with a customized MatLab software (MatLab 7.0, The Matworks, Inc.). This software allowed us to modify montage settings and to apply digital filters to the signal. For all recorded channels, bipolar montages were calculated by subtracting the signals from adjacent contacts of the same depth-electrode to minimize common electrical noise and to maximize spatial resolution (Gaillard et al.,
M
M
M M
M
F
1
2
3 4
5
6
18
27
19 23
29
38
Lamotrigine 400 mg/day Clopazam 10 mg/day Carbamazepine 1200 mg/day Topiramate 400 mg/die Lamotrigine 300 mg/day Carbamazepine 700 mg/day Carbamazepine 800 mg/day Lacosamide 200 mg/day Topiramate 400 mg/day Lamotrigine 400 mg/day Oxcarbazepine 1500 mg/day Felbamate 1800 mg/day Clonazepam 4 mg/day
Medication (mg/day)
R
R
R R
RL
R
Hemisphere
C, T, I, P, O
C, T, I, O
F, T, I F, T,I, P
T
F, T
Sample lobe
a
R 5 right; L 5 left; C 5 central; F 5 frontal; O 5 occipital; P 5 parietal; T 5 temporal; I 5 insular. Position of the SEEG derivations submitted to EEG analysis, all from the right hemisphere. b Site of origin of the seizures.
Gender
Patient
Age (yrs)
Demographic, MRI Findings and Clinical Information for Each Patient
TABLE 1.
Middle F gyrus x 5 23, y 5 58, z 5 8 Inferior T gyrus x 5 66, y 5 213, z 5 1
Middle F gyrus x 5 35, y 5 16, z 5 8 Middle F gyrus x 5 46, y 5 37, z 5 29 Central operculum x 5 53, y 5 19, z 5 1 Central operculum x 5 62, y 5 27, z 5 12
x 5 35, y 5 221, z 5 216
x 5 30, y 5 220, z 5 2213 x 5 28, y 5 217, z 5 217 x 5 30, y 5 233, z 5 22 x 5 25, y 5 221, z 5 216
Cortexa (Talairach coordinates)
x 5 29, y 5 214, z 5 218
Anterior hippocampusa (Talairach coordinates)
SEEG
R inferior P lobule
R inferior T gyrus
R anterior T gyrus R middle T gyrus
L temporo2mesial
R inferior F gyrus
Epileptogenic zoneb
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY 1159
Hippocampus
1160
MORONI ET AL.
FIGURE 1. Magnetic resonance imaging (MRI) scans and 3-D view (panel A) of intracranial electrodes implanted in the anterior hippocampus of a sample patient (panels B, C, and D coronal, sagittal, and axial views, respectively). Yellow circle indicates the location of the two electrode contacts considered for the analysis. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
2009). A 0.1 Hz high-pass filter was applied to bipolar EEG and SEEG signals, a 0.16–15 Hz band pass filter was applied to EOG signal and a 3 Hz high pass filter was applied to EMG signal.
Procedure SEEG recordings were carried out during three consecutive nights (adaptation, baseline, and postlearning), during resting wake, and during the CMT. Adaptation, baseline, and postlearning nights were recorded, respectively, during the second, third, and fourth night after electrodes implantation in four out of six patients. In the other two patients the order of baseline and postlearning nights was inverted being respectively the fourth and third night after electrode implantation. The partial counterbalancing of conditions was applied in order to control for a possible sequence effect on spontaneous EEG activity given by the passage of time and changes in clinical routine after electrodes implantation. The adaptation night was excluded from the analysis, as having been recorded with a Hippocampus
different recording system (Nihon-Kohden Neurofax-110) as compared to baseline and postlearning nights (XLTEK, TrexTM). The study began at about 8.30 p.m., at which time patients were connected to the polygraph and the recording started. At about 11 p.m. patients were requested to go to sleep. After about 7.5 h of sleep (at about 6.30 a.m.) patients were waked up. In the afternoon, following the baseline night (in four patients) and the adaptation night (in 2 patients), a resting wake with eyes opened was recorded (at 5 p.m.) and patients were then administered the CMT (learning 1 test phase, see below for details). The postlearning night, patients underwent the same procedure as followed during the baseline night. Finally, the morning following the postlearning night at 9.00 a.m., patients were retested on the CMT (retest phase).
Computerized 3-D Virtual Navigation Task The CMT has been designed to assess the individual’s ability to orient within a virtual environment. The test assesses two specific aspects of human topographical orientation; i.e., the
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY ability to form a mental representation of the environment, and the ability to use that mental representation for the purpose of orientation (Iaria et al., 2007; Ferrara et al., 2008). By presenting local properties (i.e., landmarks), the task requires that subjects, first, learn about the environment by forming a mental representation of it (i.e., a cognitive map) including the location of six landmarks (i.e., a bank, a church, a clinic, a police station, a post office, and a supermarket). Then, they are required to rely on that mental representation to travel between the different landmark locations available within the environment. Before the experimental sessions, patients underwent a practice session to make sure that they were comfortable navigating within the virtual environment. After the practice session, to ensure that the patients had proficient motor skills, they were asked to perform three control tasks. These tasks require the patient to complete a predetermined route within a virtual environment without stopping along the pathways. After this control session, patients were administered the experimental tasks, namely the learning and retrieval sessions. In the learning session, patients were required to form a mental representation of the virtual environment. In executing the task, patients were free to course through the environment using whatever path and strategy they choose in order to learn the layout of the environment, including the location and identity of the six environmental landmarks. The first assessment of the formation of the cognitive map did occur after seven minutes, based on a previous study showing that the cognitive map of this specific environment does not form earlier (Iaria et al., 2007). This assessment required the patients to report on a paper 2-D top-view outline of the city map the correct locations of the six landmarks they encountered in the environment. After this first assessment, the patients were re-assessed every two minutes. The learning phase ended when the patient correctly reported on the map the location of the six landmarks with an accuracy of 100%. After patients successfully performed the learning session, the patients were administered the retrieval task consisting of nine trials. The presence of six landmarks allowed for the assessment of 18 unique trials in which participants were asked to move from one location to another: nine of these trials were randomly selected to be used in the test phase of the study, and the remaining nine were used in the re-test phase. In each trial, the patients started facing a landmark and a sign reporting the landmark they need to reach, following the shortest pathway as quickly as possible. The test session was performed immediately after the learning session in the afternoon before the experimental night sleep (at about 6.30 p.m.), whereas the retest session was performed at least 2 hours after the following morning awakening (at about 9.00 a.m.). For each trial, we measured the time (seconds) that participants spent in reaching a target location in addition to the ideal time calculated to get from starting to target location. This additional time (i.e., time delay) has been shown, in fact, to be representative of spatial ability in the use of a cognitive map (Iaria et al., 2007).
1161
Behavioral Data Analysis The difference in time delay between trials in the test and re-test phases of the study was treated as dependent measure. Given the large inter-individual differences and intra-individual fluctuations in this kind of performance (Ferrara et al., 2008), we used the median time as a more stable index of central tendency. Moreover, to establish a criterion for the exclusion of possible outliers, we calculated the percentage of performance change (i.e. increment or decrement) between test and retest conditions in a sample of 100 subjects enrolled in our laboratory in two previous studies (Ferrara et al., 2008; Tempesta et al., 2012). This large sample showed a mean percentage of test-retest variation of 27.56% and a standard deviation of 29.41%. Since the test-retest retention interval in this sample was variable, a further analysis on a subsample of 20 subjects with the same retention interval of the present study confirmed the above results (mean percentage 5 27.53%; standard deviation 5 31.36%). Thus, patients who had a decrease larger than 3 standard deviations from that of the entire sample were excluded from statistical analyses.
SEEG/EEG Data Analysis Polygraphic recordings were scored by one of the authors (P.P.), who was blind to the experimental conditions. Resting wake with eyes opened and learning, test and retest sessions of the CMT were visually scored in 2 s epochs. Then, occurring artifacts, interictal spikes and pathological EEG signals were manually removed from SEEG traces. Sleep scoring was performed visually on scalp recordings (Fz–Cz, EOG, and EMG) following the standard criteria (Rechtschaffen and Kales, 1968) in 20-s epochs. We are unable to report sleep latency because, in the clinical setting, patient’s sleep was recorded with an ambulatory recording system in his/ her room, and time of "lights out" was not acquired. Thus the following polysomnographic (PSG) parameters were considered as dependent variables: (1) total sleep time (TST), defined as the sum of time spent in stages 1, 2, SWS, and REM; (2) percentage of each sleep stage (time spent in a sleep stage/TST); (3) wakefulness after sleep onset (WASO), expressed as the intrasleep time (minutes) spent awake. After sleep scoring, SEEG/EEG traces were cleared from artifacts, interictal spikes and pathological EEG signals through a manual rejection.
SEEG Power Spectra Analysis Power spectral analysis was conducted on anterior hippocampal derivations and on selected cortical derivations (see Table 1 for contact details). Spectral power for each derivation signal was computed using the Fast Fourier Transform (FFT—Welch method) applied to 2-s segments (Hamming window), with an overlapping period of 1 s, in the frequency range of 0.5–30.0 Hz. As the sleep recordings are concerned, resulting power spectral density was averaged in 20-s epochs. To enable comparison with the data of our previous study (Moroni et al., 2008), Hippocampus
1162
MORONI ET AL.
power spectral analysis was carried out on the first NREM period (stages 2, 3, and 4). The first NREM sleep period was chosen because of its sensitivity to postlearning plasticity processes (e.g., Huber et al., 2004) and of its proximity to the experimental manipulation. Due to a large inter-subject variability on the spike number and distribution and, consequently, on the incidence of rejected epochs across the first NREM period, the power density was divided into quintiles and then averaged. In this way the timecourse of each first sleep period was preserved, making it sure that equivalent sleep periods during the two nights contributed to the NREM power average.
Statistical Analysis Since SEEG power spectra show a large inter-subject variability, for each frequency bin EEG power density is expressed as a percentage of the total power corresponding to the entire frequency range (0.5–30 Hz). After percentage normalization, the resulting power density values were grouped in the following frequency bands: (a) low delta: 0.5–2.0 Hz; high delta: 2.1–4.0 Hz; Theta: 4.1–8.0 Hz: alpha: 8.1–12.0 Hz: beta1: 12.1–16.0 Hz: beta2: 16.1–30.0 Hz for waking recordings, and (b) low delta: 0.5–2.0 Hz; high delta: 2.1–4.0 Hz; theta: 4.1–8.0 Hz: alpha: 8.1–12.0 Hz: sigma: 12.1–16.0 Hz: beta: 16.1–30.0 Hz for sleep recordings. For each frequency band, waking SEEG power spectra were submitted to one-way repeated measure ANOVAs with Condition (resting wakefulness, learning session, test session, and retest session) as factor. Similarly, for each frequency band, sleep SEEG power spectra were submitted to a one-way repeated measure ANOVA with Night (baseline, postlearning) as factor, separately for hippocampal, and cortical derivations. Moreover, we performed one-way repeated measure ANOVAs to ascertain whether baseline and postlearning nights differed with respect to one or more PSG measures. Paired Student’s t test (two-tailed) was performed on behavioral data comparing task performance scores at presleep test and postsleep retest.
RESULTS Virtual Navigation Task Performance A first data inspection confirmed that the task is characterized by a large performance variability. In particular, two patients drastically worsened their performance at retest showing, respectively, a 153.3 and 172.5% increase of time delay to reach the target locations at retest. As their performance had a decrease larger than 3 standard deviations from the mean of the above reported normative sample, these two patients were excluded from all the successive analyses. No significant difference was observed between presleep test and postsleep retest performance in the sample of six patients (t 5 20.41; P 5 0.69, see Fig. 2 for patients performance scores). Hippocampus
FIGURE 2. Task performance during test and retest sessions for each of the six patients. Task performance was calculated as the difference between the time (s) spent to reach the target locations (landmarks) and an ideal minimum time for each of the nine test and retest trials. *Patients who were administered the task during the day after adaptation night, having the order of condition inverted.
Polysomnographic (PSG) Measures Repeated measure ANOVAs comparing the PSG variables in the baseline and postlearning nights did not show significant changes for any of the measures assessed (see Table 2), indicating that there were no substantial differences in the sleep macro-architecture of the two nights.
SEEG Power Spectra During the First NREM Sleep Period The power spectra analysis was carried out on equivalent NREM periods in the two nights (mean 6 SE: baseline: 3,583 s 6 335.5; postlearning: 4,290 s 6 837.2; t 5 20.73; P 5 0.50). The ANOVA indicated a significant difference between baseline and postlearning night for the high-delta (2.1–4.0 Hz) band (F1,5 5 14.21; P 5 0.013). In particular, a high-delta power increase was observed during the postlearning night compared to baseline (mean 6 SE: baseline: 13.51 6 1.13; postlearning: 14.3 6 1.15, see Fig. 3). Higher levels of highdelta power during the postlearning night as compared to baseline were observed in all of the six patients, comprising the two patients (#4 and 5) who had an inverted order of nights. Moreover, the high-delta power during postlearning night was strongly correlated with the performance difference between postsleep retest and presleep test (r 5 0.92; P 5 0.009). In
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY TABLE 2. Means and SEs of the PSG Variables of the Baseline and Postleaning Night Baseline
Postlearning
Variables
Mean
SE
Mean
SE
F(1,5)
P
Stage 1 (%) Stage 2 (%) SWS (%) NREM (%) REM (%) WASO (min) TST (min)
5.44 67.34 9.49 74 20.55 54.83 397.89
0.72 3.09 4.64 1.49 1.64 17.4 29.88
7.76 64.24 8.24 69.74 22.49 55.16 327.11
2.03 4.47 3.25 4.65 5.38 19.42 59.59
2.91 0.46 0.49 0.8 0.13 0.0002 1.15
0.15 0.53 0.53 0.41 0.73 0.99 0.33
1163
derivation (F1,5 5 2.60; P 5 0.17), indicating that the observed effects on percentage values are not a spurious effect of total power changes. Finally, to evaluate the robustness of our results, we performed the same ANOVAs with a Bonferroni correction. Since it is too conservative in the case of correlated outcome variables, the alpha level was adjusted by taking into account the mean intercorrelation (r 5 0.47) between the dependent variables (Sankoh et al., 1997; Perneger, 1998) and the P value after Bonferroni’s adjustment was set to 0.0134. The effect observed on hippocampal high-delta band is still significant (P 5 0.0130) after the Bonferroni’s adjustment.
No differences are observed between the two nights for each considered PSG variable.
SEEG Power Spectra During Resting Wakefulness and Spatial Navigation Memory Task
particular, the more was the power density in the high-delta band during postlearning night, the larger was the performance increase at retest (see Fig. 4). No other frequency band showed significant difference between conditions (see Table 3). With respect to the cortical leads, no significant difference was observed between baseline and postlearning night for all the frequency bands (see Table 3). Control analyses (ANOVAs) carried out on total EEG power (0.5–30 Hz) did not show any significant difference between nights for hippocampal (F1,5 5 0.20; P 5 0.67) and cortical
ANOVA carried out on SEEG power spectra density of the four waking sessions showed no significant effects for the hippocampal derivations. However, it is of note that high-delta power was tendentially higher in the learning phase and positively correlated (r 5 0.60, ns) with learning performance (i.e., the higher the delta power, the shorter the learning phase). ANOVA also showed a significant main effect of Condition in the low-delta band (0.5–2.0 Hz) for neocortical derivations (F3,15 5 3.53; P 5 0.04). Post hoc comparisons showed that low-delta power was significantly higher during the Resting Wakefulness condition as compared to the other conditions
FIGURE 3. Hippocampal relative SEEG power for low-delta (0.5–2.0 Hz), high-delta (2.1– 4.0 Hz), theta (4.1–8.0 Hz), alpha (8.1–12.0 Hz), sigma (12.1–16.0 Hz), and beta (16.1–30.0 Hz) frequency bands of the first NREM episode of baseline and postlearning nights. Highdelta activity during postlearning night is significantly higher than during baseline night (*P 5 0.013). Panels are scaled differently for each band for clarity purposes. Hippocampus
1164
MORONI ET AL.
DISCUSSION In the present study we aimed at evaluating for the first time the effects of a hippocampus-dependent spatial memory task on the ensuing hippocampal sleep SEEG activity. Our findings showed that hippocampal SEEG power in the highdelta range (2.1–4 Hz) increases during the first NREM period after task training compared to baseline; such a delta power increase is related with task performance improvement at retest. These findings support the hypothesis that, in humans, hippocampal EEG activity in the delta band is involved in spatial memory formation processes during sleep.
Hippocampal SEEG Power in the High-Delta Range (2.1–4 Hz) Increases During the First NREM Period After Training on a Navigation Task FIGURE 4. Scatterplot of the individual correlations between hippocampal stereo-EEG power in the high-delta during postlearning night and CMT retest-test performance difference (negative values indicate a performance improvement).
(see Table 4 for details), presumably due to the fact that patients were more relaxed and/or behaviorally less active. Also in this case, control analyses (ANOVAs) carried out on total EEG power (0.5–30 Hz) did not show any significant difference between the four waking conditions for hippocampal (F3,15 5 0.50; P 5 0.69) and cortical derivation (F3,15 5 0.71; P 5 0.56), indicating that the observed effects on percentage values are not a spurious effect of total power changes.
In a previous study, we reported the first direct evidence that slow hippocampal oscillations are involved in human memory consolidation. In that study, in a sample of epileptic patients, we showed that a short but intensive training on a sequential finger tapping task (SFTT) is followed by an increase in the amount of hippocampal SEEG power in the very low frequency range (0.5–1.0 Hz) during the first postlearning NREM period (Moroni et al., 2008). Here we found a similar postlearning power increase, but in the upper part of the delta frequency range (2.1–4 Hz). Such a dissociation can be interpreted in terms of the features of the memory tasks adopted in the two studies. Although the SFTT is a motor task, neuroimaging studies have demonstrated that the hippocampus is
TABLE 3. Repeated Measure ANOVAS Comparing Baseline and Postleaning Night for Hippocampal and Cortical Derivations Baseline
Postlearning
Mean
SE
Mean
SE
F(1,5)
P
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Sigma (12.1–16.0 Hz) Beta (16.1–30.0 Hz)
27.22 13.51 2.75 0.94 0.57 0.11
1.04 1.13 0.23 0.19 0.15 0.02
26.77 14.3 2.68 0.85 0.56 0.11
1.12 1.15 0.22 0.17 0.17 0.02
1.32 14.21 2.54 1.78 0.07 0.002
0.3 0.013 0.17 0.23 0.8 0.96
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Sigma (12.1–16.0 Hz) Beta (16.1–30.0 Hz)
39.62 5 1.2 0.82 0.43 0.06
1.03 0.42 0.06 0.24 0.16 0.02
39.3 5.24 1.29 0.79 0.43 0.06
0.69 0.21 0.11 0.19 0.09 0.02
Hippocampus
Cortex
Relative (percentage) power spectra for each frequency band were calculated on the first NREM period.
Hippocampus
0.63 1.15 1.09 0.1 0.0009 0.22
0.46 0.33 0.34 0.75 0.98 0.66
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY
1165
TABLE 4. Means and SE of the Hippocampal and Cortical SEEG Power Spectra for Each Band During the Four Waking Conditions (Resting Wakefulness (RW), Learning (Lr), Test (Ts), and Retest (Rt) RW
Lr
Ts
Rts
Mean
SE
Mean
SE
Mean
SE
Mean
SE
F(3,15)
P
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Beta1 (12.1–16.0 Hz) Beta2 (16.1–30.0 Hz)
26.47 10.94 3.16 1.36 0.8 0.27
1.11 0.65 0.22 0.28 0.19 0.04
26.32 12.16 2.88 1.17 0.78 0.26
1.39 0.92 0.18 0.23 0.21 0.04
25.89 11.79 2.92 1.21 0.96 0.3
1.21 1.03 0.16 0.19 0.25 0.04
28.27 10.35 2.56 1.16 0.96 0.28
1.83 0.56 0.28 0.22 0.27 0.04
1.27 1.95 2.58 1.27 2.33 1.42
0.32 0.16 0.09 0.32 0.11 0.27
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Beta1 (12.1–16.0 Hz) Beta2 (16.1–30.0 Hz)
18.52 8.36 5.4 1.63 1.38 0.9
3.26 1.17 1.19 0.28 0.26 0.33
15.37 8.37 6.14 1.88 1.46 1.04
3.36 0.97 1.85 0.4 0.29 0.45
14.76 7.26 5.8 2.04 1.56 1.31
3.04 0.85 1.57 0.42 0.28 0.48
14.64 7.49 6.45 1.98 1.36 1.18
2.5 0.79 138 0.27 0.22 0.34
3.53 0.89 1.07 0.97 0.33 1.46
0.04 0.46 0.39 0.43 0.8 0.26
Hippocampus
Cortex
Repeated measure ANOVAs results are also reported. Comparisons were conducted on relative (percentage) power spectra for each frequency band.
involved also in the consolidation of procedural memories, and of motor sequence-based skills in particular (Walker at al., 2005; Albouy et al., 2013). In the present study we used a truly hippocampusdependent spatial memory task, that has been shown to rely on both proper functioning and structural integrity of the hippocampus (Iaria et al., 2007, 2008), showing that a training on this task modifies hippocampal activity during the ensuing sleep period. Moreover, patients who showed more high-delta power during postlearning NREM sleep had better performance at task retest than patient who showed less high-delta power. This effect was localized at the anterior hippocampus and no differences were observed at neocortical regions. Moreover, postlearning and control nights did not differ on PSG measures, suggesting that the hippocampal high-delta power increase was strictly local and not dependent on macrostructural sleep aspects. We could hypothesize that the different delta ranges involved in sleep-related memory consolidation in our previous study (0.5–1.0 Hz) and in the present one (2.1–4.0 Hz), could be ascribed to the involvement of different and specific consolidation mechanisms. It is plausible that synaptic downscaling, typically conveyed by low-delta activity and in particular by the
Hippocampal Slow EEG Frequencies During NREM Sleep are Involved in Spatial Memory Consolidation in Humans Fabio Moroni,1,2* Lino Nobili,3,4 Giuseppe Iaria,5 Ivana Sartori,3 Cristina Marzano,1 Daniela Tempesta,6 Paola Proserpio,3 Giorgio Lo Russo,3 Francesca Gozzo,3 Carlo Cipolli,2 Luigi De Gennaro,1 and Michele Ferrara6
ABSTRACT: The hypothesis that sleep is instrumental in the process of memory consolidation is currently largely accepted. Hippocampal formation is involved in the acquisition of declarative memories and particularly of spatial memories. Nevertheless, although largely investigated in rodents, the relations between spatial memory and hippocampal EEG activity have been scarcely studied in humans. Aimed to evaluate the effects of spatial learning on human hippocampal sleep EEG activity, we recorded hippocampal Stereo-EEG (SEEG) in a group of refractory epilepsy patients undergoing presurgical clinical evaluation, after a training on a spatial navigation task. We observed that hippocampal high-delta (2–4 Hz range) activity increases during the first NREM episode after learning compared to the baseline night. Moreover, the amount of hippocampal NREM high-delta power was correlated with task performance at retest. The effect involved only the hippocampal EEG frequencies inasmuch no differences were observed at the neocortical electrodes and in the traditional polysomnographic measures. The present findings support the crucial role of hippocampal slow EEG frequencies during sleep in the memory consolidation processes. More generally, together with previous results, they suggest that slow frequency rhythms are a fundamental characteristic of human hippocampal EEG during both sleep and wakefulness, and are related to the consolidation of different types of memories. C 2014 Wiley Periodicals, Inc. V
KEY WORDS: hippocampus; stereo-EEG; declarative memory; memory consolidation; local sleep
1
Department of Psychology, “Sapienza” University of Rome, Roma, Italy; 2 Laboratory of Psychology, Department of Specialized, Diagnostics and Experimental Medicine, University of Bologna, Bologna, Italy; 3 Centre of Epilepsy Surgery ‘‘C. Munari’’, Center of Sleep Medicine, Niguarda Hospital, Milan, Italy; 4 Institute of Bioimaging and Molecular Physiology, Section of Genoa, National Research Council, Genova, Italy; 5 Department of Psychology and Clinical Neurosciences, and Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada; 6 Department of Life, Health and Environmental Sciences, University of L’Aquila, L’Aquila, Italy Grant sponsor: Compagnia di San Paolo, Programma Neuroscienze 2008/09; Grant number: 3896 SD/sd; 2008.2130 (to M.F.); Grant sponsor: ESRS Sanofi-Aventis Research Grant 2008/10 (to F.M.), and Fondazione del Monte di Bologna e Ravenna 2011/13 (to C.C.). *Correspondence to: Fabio Moroni; Department of Psychology, “Sapienza” University of Rome, Via dei Marsi, 78, 00185 Rome, Italy. E-mail: [email protected] Accepted for publication 23 April 2014. DOI 10.1002/hipo.22299 Published online 2 May 2014 in Wiley Online Library (wileyonlinelibrary.com). C 2014 WILEY PERIODICALS, INC. V
INTRODUCTION The medial temporal lobe (MTL), and particularly the hippocampal formation, has been considered central for declarative memory in humans, playing a key role from the initial phase of memory formation to the final storage in widespread brain areas (Eichenbaum, 2004). Sleep, as emerged from a growing number of experimental studies, critically and actively supports the process of memory consolidation. In particular, new and weak memory traces are strengthened and transformed into a more stable and persistent form by sleep-dependent processes (Tononi and Cirelli, 2006; Diekelmann and Born, 2010). From these studies emerged that declarative memory and, in particular, spatial learning specifically benefits from NREM sleep (for a review, see Rasch and Born, 2013). To date, our knowledge about hippocampal neurophysiological activity related to memory formation mostly derives from animal studies on spatial learning, which is strictly dependent on the hippocampal formation (Squire et al., 2004). At the cellular level, it has been observed that neuronal ensembles activated during the waking behavior are then re-activated during posttraining slow-wave sleep (SWS) (e.g., Pavlides and Winson, 1989; Wilson and McNaughton, 1994; Skaggs and McNaughton, 1996) and REM sleep (Poe et al., 2000; Louie and Wilson, 2001). This "replay" of place cells activity during sleep has been interpreted as a process of memory trace strengthening and transferring towards other cerebral areas (Buzsaki, 1989, 1998; Born and Wilhelm, 2012). In humans, place cells active during virtual navigation and recall of navigation related memories have been recently identified (Miller et al., 2013). On the other hand, reactivation of place cells during sleep has not been replicated yet, although a neuroimaging study has reported a hippocampal reactivation during SWS, but not during REM sleep, following spatial learning that correlated with task performance improvement (Peigneux et al., 2004). From an electrophysiological point of view, rodent’s hippocampal EEG activity during movement and exploratory behavior, as well as during REM sleep, has shown to be characterized by the rhythmic slow activity (RSA or theta: 4–10 Hz). RSA is considered
1158
MORONI ET AL.
crucial in memory encoding, probably fulfilling the role of phase-synchronizer for neuronal ensembles belonging to different brain areas that can discharge together under the coordination of theta rhythm (Buzsaki, 2002; Battaglia et al., 2011). In humans, hippocampal electrophysiological activity has been mainly investigated by means of intracerebral recordings carried out in refractory epilepsy patients undergoing presurgical clinical evaluation. These studies have reported an increase of theta activity in the neocortex (Kahana et al., 1999; Caplan et al., 2001, 2003) and in the hippocampus (Ekstrom et al., 2005; Watrous et al., 2011) during spatial navigation tasks in virtual environments. Nevertheless, the existence of a prominent 2–4 Hz rhythm, below the theta range, in the spontaneous human hippocampal EEG activity has been repeatedly described during both resting wakefulness and sleep (Bodizs et al., 2001; Moroni et al., 2007, 2012; Clemens et al., 2009), and recently also during the execution of a spatial navigation task (Clemens et al., 2013). Hence, the prominent presence in human hippocampal recordings of slow (delta) EEG rhythms seems to be a relevant emerging finding. Indeed, in a recent study we described a peculiar bimodal distribution of hippocampal activity in the delta range, made up by a low frequency nonoscillatory activity (up to 2 Hz) and a faster oscillatory rhythm (2–4 Hz) (Moroni et al., 2012). Interestingly, we observed that this faster oscillatory activity at around 3 Hz is present, in the human spontaneous hippocampal SEEG signal, not only during wakefulness and REM sleep but even during NREM sleep. Previous studies have hypothesized that human hippocampal activity in the 2–4 Hz range could be the analogue of animal RSA, being related to behavioral activated states and REM sleep (Moroni et al., 2012; Clemens et al., 2013). Since we have observed that, at variance with animal literature, in humans this rhythm is largely present also during NREM sleep (Moroni et al., 2012), we asked which could be its peculiar functional role during NREM sleep. In particular, in this study we asked whether this hippocampal slow activity could be related to spatial memory consolidation processes during sleep. To this aim, we administered to a sample of epileptic patients undergoing presurgical examination a spatial memory task, namely the Cognitive Map Test (CMT) that strictly depends on hippocampal activity (Iaria et al., 2007; Ferrara et al., 2008). In particular, by recording intracranial EEG activity during both CMT execution and post-training night sleep, we evaluated the effects of task training on the hippocampal EEG activity during subsequent sleep, under the assumption of a specific involvement of the EEG frequencies below the theta range. The relations between sleep EEG frequencies and postsleep task performance were also evaluated.
MATERIALS AND METHODS Participants We recruited eight patients (6 M, 2 F; mean age: 26.75; age range: 18–38) with pharmacoresistant focal epilepsy. All Hippocampus
patients were candidates for surgical removal of the epileptic focus and underwent individual investigation with stereotactically implanted intracerebral multilead electrodes (SEEG) for the precise localization of the epileptogenic zone (Cossu et al., 2005). Patients were selected based on the presence of at least two electrode contacts localized within the hippocampus. We excluded from the study patients who did show features of hippocampal sclerosis. We also excluded patients with severe cognitive impairment; that is, scoring under the 95th percentile at the Raven’s colored progressive matrices (CPM) (Raven, 1996). Throughout the study, patients took their standard doses of anticonvulsant medications. The study protocol was approved by the Local Ethics Committee (Niguarda Hospital, Milan, Italy). Prior to SEEG electrode implantation, patients gave written informed consent for participation in this study and for publication of data. Two of the eight selected patients were discarded because considered outliers, since their task performance at postsleep retest did not fulfill the predetermined criterion of a decrease not larger than three standard deviations from that of a normative sample (see below). Therefore, all the subsequent analyses have been carried out on a sample of 6 patients (see Table 1).
Electrodes Placement and EEG/SEEG Recordings Stereo-EEG activity was recorded from platinum-iridium semiflexible multilead depth-electrodes, with a diameter of 0.8 mm, a contact length of 2 mm and an intercontact distance of 1.5 mm (Dixi Medical, Besancon France). The placement of electrode contacts was ascertained by postimplantation magnetic resonance imaging (MRI) scans. For the entire sample, we considered for analyses contacts localized within the anterior hippocampus of the right hemisphere (see Fig. 1). As the neocortical leads are concerned, the localization was more sparse and we selected contacts in the right hemisphere within the frontal lobe (middle frontal gyrus) in three patients, central operculum in two patients and temporal lobe (inferior temporal gyrus) in one patient (for details on locations see Table 1). Scalp EEG activity was recorded from two platinum needle electrodes placed during surgery at "10–20" positions Fz and Cz on the scalp. Electroocular (EOG) activity was registered at the outer canthi of both eyes and submental electromyographic (EMG) activity was acquired with electrodes attached to the chin. EEG and SEEG signals were recorded using a 192-channel recording system (Nihon-Kohden Neurofax-110) with a sampling rate of 1,000 Hz during resting wake and virtual navigation task, and using a 24 channels ambulatory system recording (XLTEK, TrexTM) with a sampling rate of 512 Hz during the baseline and experimental nights of sleep. Recording data were handled with a customized MatLab software (MatLab 7.0, The Matworks, Inc.). This software allowed us to modify montage settings and to apply digital filters to the signal. For all recorded channels, bipolar montages were calculated by subtracting the signals from adjacent contacts of the same depth-electrode to minimize common electrical noise and to maximize spatial resolution (Gaillard et al.,
M
M
M M
M
F
1
2
3 4
5
6
18
27
19 23
29
38
Lamotrigine 400 mg/day Clopazam 10 mg/day Carbamazepine 1200 mg/day Topiramate 400 mg/die Lamotrigine 300 mg/day Carbamazepine 700 mg/day Carbamazepine 800 mg/day Lacosamide 200 mg/day Topiramate 400 mg/day Lamotrigine 400 mg/day Oxcarbazepine 1500 mg/day Felbamate 1800 mg/day Clonazepam 4 mg/day
Medication (mg/day)
R
R
R R
RL
R
Hemisphere
C, T, I, P, O
C, T, I, O
F, T, I F, T,I, P
T
F, T
Sample lobe
a
R 5 right; L 5 left; C 5 central; F 5 frontal; O 5 occipital; P 5 parietal; T 5 temporal; I 5 insular. Position of the SEEG derivations submitted to EEG analysis, all from the right hemisphere. b Site of origin of the seizures.
Gender
Patient
Age (yrs)
Demographic, MRI Findings and Clinical Information for Each Patient
TABLE 1.
Middle F gyrus x 5 23, y 5 58, z 5 8 Inferior T gyrus x 5 66, y 5 213, z 5 1
Middle F gyrus x 5 35, y 5 16, z 5 8 Middle F gyrus x 5 46, y 5 37, z 5 29 Central operculum x 5 53, y 5 19, z 5 1 Central operculum x 5 62, y 5 27, z 5 12
x 5 35, y 5 221, z 5 216
x 5 30, y 5 220, z 5 2213 x 5 28, y 5 217, z 5 217 x 5 30, y 5 233, z 5 22 x 5 25, y 5 221, z 5 216
Cortexa (Talairach coordinates)
x 5 29, y 5 214, z 5 218
Anterior hippocampusa (Talairach coordinates)
SEEG
R inferior P lobule
R inferior T gyrus
R anterior T gyrus R middle T gyrus
L temporo2mesial
R inferior F gyrus
Epileptogenic zoneb
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY 1159
Hippocampus
1160
MORONI ET AL.
FIGURE 1. Magnetic resonance imaging (MRI) scans and 3-D view (panel A) of intracranial electrodes implanted in the anterior hippocampus of a sample patient (panels B, C, and D coronal, sagittal, and axial views, respectively). Yellow circle indicates the location of the two electrode contacts considered for the analysis. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
2009). A 0.1 Hz high-pass filter was applied to bipolar EEG and SEEG signals, a 0.16–15 Hz band pass filter was applied to EOG signal and a 3 Hz high pass filter was applied to EMG signal.
Procedure SEEG recordings were carried out during three consecutive nights (adaptation, baseline, and postlearning), during resting wake, and during the CMT. Adaptation, baseline, and postlearning nights were recorded, respectively, during the second, third, and fourth night after electrodes implantation in four out of six patients. In the other two patients the order of baseline and postlearning nights was inverted being respectively the fourth and third night after electrode implantation. The partial counterbalancing of conditions was applied in order to control for a possible sequence effect on spontaneous EEG activity given by the passage of time and changes in clinical routine after electrodes implantation. The adaptation night was excluded from the analysis, as having been recorded with a Hippocampus
different recording system (Nihon-Kohden Neurofax-110) as compared to baseline and postlearning nights (XLTEK, TrexTM). The study began at about 8.30 p.m., at which time patients were connected to the polygraph and the recording started. At about 11 p.m. patients were requested to go to sleep. After about 7.5 h of sleep (at about 6.30 a.m.) patients were waked up. In the afternoon, following the baseline night (in four patients) and the adaptation night (in 2 patients), a resting wake with eyes opened was recorded (at 5 p.m.) and patients were then administered the CMT (learning 1 test phase, see below for details). The postlearning night, patients underwent the same procedure as followed during the baseline night. Finally, the morning following the postlearning night at 9.00 a.m., patients were retested on the CMT (retest phase).
Computerized 3-D Virtual Navigation Task The CMT has been designed to assess the individual’s ability to orient within a virtual environment. The test assesses two specific aspects of human topographical orientation; i.e., the
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY ability to form a mental representation of the environment, and the ability to use that mental representation for the purpose of orientation (Iaria et al., 2007; Ferrara et al., 2008). By presenting local properties (i.e., landmarks), the task requires that subjects, first, learn about the environment by forming a mental representation of it (i.e., a cognitive map) including the location of six landmarks (i.e., a bank, a church, a clinic, a police station, a post office, and a supermarket). Then, they are required to rely on that mental representation to travel between the different landmark locations available within the environment. Before the experimental sessions, patients underwent a practice session to make sure that they were comfortable navigating within the virtual environment. After the practice session, to ensure that the patients had proficient motor skills, they were asked to perform three control tasks. These tasks require the patient to complete a predetermined route within a virtual environment without stopping along the pathways. After this control session, patients were administered the experimental tasks, namely the learning and retrieval sessions. In the learning session, patients were required to form a mental representation of the virtual environment. In executing the task, patients were free to course through the environment using whatever path and strategy they choose in order to learn the layout of the environment, including the location and identity of the six environmental landmarks. The first assessment of the formation of the cognitive map did occur after seven minutes, based on a previous study showing that the cognitive map of this specific environment does not form earlier (Iaria et al., 2007). This assessment required the patients to report on a paper 2-D top-view outline of the city map the correct locations of the six landmarks they encountered in the environment. After this first assessment, the patients were re-assessed every two minutes. The learning phase ended when the patient correctly reported on the map the location of the six landmarks with an accuracy of 100%. After patients successfully performed the learning session, the patients were administered the retrieval task consisting of nine trials. The presence of six landmarks allowed for the assessment of 18 unique trials in which participants were asked to move from one location to another: nine of these trials were randomly selected to be used in the test phase of the study, and the remaining nine were used in the re-test phase. In each trial, the patients started facing a landmark and a sign reporting the landmark they need to reach, following the shortest pathway as quickly as possible. The test session was performed immediately after the learning session in the afternoon before the experimental night sleep (at about 6.30 p.m.), whereas the retest session was performed at least 2 hours after the following morning awakening (at about 9.00 a.m.). For each trial, we measured the time (seconds) that participants spent in reaching a target location in addition to the ideal time calculated to get from starting to target location. This additional time (i.e., time delay) has been shown, in fact, to be representative of spatial ability in the use of a cognitive map (Iaria et al., 2007).
1161
Behavioral Data Analysis The difference in time delay between trials in the test and re-test phases of the study was treated as dependent measure. Given the large inter-individual differences and intra-individual fluctuations in this kind of performance (Ferrara et al., 2008), we used the median time as a more stable index of central tendency. Moreover, to establish a criterion for the exclusion of possible outliers, we calculated the percentage of performance change (i.e. increment or decrement) between test and retest conditions in a sample of 100 subjects enrolled in our laboratory in two previous studies (Ferrara et al., 2008; Tempesta et al., 2012). This large sample showed a mean percentage of test-retest variation of 27.56% and a standard deviation of 29.41%. Since the test-retest retention interval in this sample was variable, a further analysis on a subsample of 20 subjects with the same retention interval of the present study confirmed the above results (mean percentage 5 27.53%; standard deviation 5 31.36%). Thus, patients who had a decrease larger than 3 standard deviations from that of the entire sample were excluded from statistical analyses.
SEEG/EEG Data Analysis Polygraphic recordings were scored by one of the authors (P.P.), who was blind to the experimental conditions. Resting wake with eyes opened and learning, test and retest sessions of the CMT were visually scored in 2 s epochs. Then, occurring artifacts, interictal spikes and pathological EEG signals were manually removed from SEEG traces. Sleep scoring was performed visually on scalp recordings (Fz–Cz, EOG, and EMG) following the standard criteria (Rechtschaffen and Kales, 1968) in 20-s epochs. We are unable to report sleep latency because, in the clinical setting, patient’s sleep was recorded with an ambulatory recording system in his/ her room, and time of "lights out" was not acquired. Thus the following polysomnographic (PSG) parameters were considered as dependent variables: (1) total sleep time (TST), defined as the sum of time spent in stages 1, 2, SWS, and REM; (2) percentage of each sleep stage (time spent in a sleep stage/TST); (3) wakefulness after sleep onset (WASO), expressed as the intrasleep time (minutes) spent awake. After sleep scoring, SEEG/EEG traces were cleared from artifacts, interictal spikes and pathological EEG signals through a manual rejection.
SEEG Power Spectra Analysis Power spectral analysis was conducted on anterior hippocampal derivations and on selected cortical derivations (see Table 1 for contact details). Spectral power for each derivation signal was computed using the Fast Fourier Transform (FFT—Welch method) applied to 2-s segments (Hamming window), with an overlapping period of 1 s, in the frequency range of 0.5–30.0 Hz. As the sleep recordings are concerned, resulting power spectral density was averaged in 20-s epochs. To enable comparison with the data of our previous study (Moroni et al., 2008), Hippocampus
1162
MORONI ET AL.
power spectral analysis was carried out on the first NREM period (stages 2, 3, and 4). The first NREM sleep period was chosen because of its sensitivity to postlearning plasticity processes (e.g., Huber et al., 2004) and of its proximity to the experimental manipulation. Due to a large inter-subject variability on the spike number and distribution and, consequently, on the incidence of rejected epochs across the first NREM period, the power density was divided into quintiles and then averaged. In this way the timecourse of each first sleep period was preserved, making it sure that equivalent sleep periods during the two nights contributed to the NREM power average.
Statistical Analysis Since SEEG power spectra show a large inter-subject variability, for each frequency bin EEG power density is expressed as a percentage of the total power corresponding to the entire frequency range (0.5–30 Hz). After percentage normalization, the resulting power density values were grouped in the following frequency bands: (a) low delta: 0.5–2.0 Hz; high delta: 2.1–4.0 Hz; Theta: 4.1–8.0 Hz: alpha: 8.1–12.0 Hz: beta1: 12.1–16.0 Hz: beta2: 16.1–30.0 Hz for waking recordings, and (b) low delta: 0.5–2.0 Hz; high delta: 2.1–4.0 Hz; theta: 4.1–8.0 Hz: alpha: 8.1–12.0 Hz: sigma: 12.1–16.0 Hz: beta: 16.1–30.0 Hz for sleep recordings. For each frequency band, waking SEEG power spectra were submitted to one-way repeated measure ANOVAs with Condition (resting wakefulness, learning session, test session, and retest session) as factor. Similarly, for each frequency band, sleep SEEG power spectra were submitted to a one-way repeated measure ANOVA with Night (baseline, postlearning) as factor, separately for hippocampal, and cortical derivations. Moreover, we performed one-way repeated measure ANOVAs to ascertain whether baseline and postlearning nights differed with respect to one or more PSG measures. Paired Student’s t test (two-tailed) was performed on behavioral data comparing task performance scores at presleep test and postsleep retest.
RESULTS Virtual Navigation Task Performance A first data inspection confirmed that the task is characterized by a large performance variability. In particular, two patients drastically worsened their performance at retest showing, respectively, a 153.3 and 172.5% increase of time delay to reach the target locations at retest. As their performance had a decrease larger than 3 standard deviations from the mean of the above reported normative sample, these two patients were excluded from all the successive analyses. No significant difference was observed between presleep test and postsleep retest performance in the sample of six patients (t 5 20.41; P 5 0.69, see Fig. 2 for patients performance scores). Hippocampus
FIGURE 2. Task performance during test and retest sessions for each of the six patients. Task performance was calculated as the difference between the time (s) spent to reach the target locations (landmarks) and an ideal minimum time for each of the nine test and retest trials. *Patients who were administered the task during the day after adaptation night, having the order of condition inverted.
Polysomnographic (PSG) Measures Repeated measure ANOVAs comparing the PSG variables in the baseline and postlearning nights did not show significant changes for any of the measures assessed (see Table 2), indicating that there were no substantial differences in the sleep macro-architecture of the two nights.
SEEG Power Spectra During the First NREM Sleep Period The power spectra analysis was carried out on equivalent NREM periods in the two nights (mean 6 SE: baseline: 3,583 s 6 335.5; postlearning: 4,290 s 6 837.2; t 5 20.73; P 5 0.50). The ANOVA indicated a significant difference between baseline and postlearning night for the high-delta (2.1–4.0 Hz) band (F1,5 5 14.21; P 5 0.013). In particular, a high-delta power increase was observed during the postlearning night compared to baseline (mean 6 SE: baseline: 13.51 6 1.13; postlearning: 14.3 6 1.15, see Fig. 3). Higher levels of highdelta power during the postlearning night as compared to baseline were observed in all of the six patients, comprising the two patients (#4 and 5) who had an inverted order of nights. Moreover, the high-delta power during postlearning night was strongly correlated with the performance difference between postsleep retest and presleep test (r 5 0.92; P 5 0.009). In
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY TABLE 2. Means and SEs of the PSG Variables of the Baseline and Postleaning Night Baseline
Postlearning
Variables
Mean
SE
Mean
SE
F(1,5)
P
Stage 1 (%) Stage 2 (%) SWS (%) NREM (%) REM (%) WASO (min) TST (min)
5.44 67.34 9.49 74 20.55 54.83 397.89
0.72 3.09 4.64 1.49 1.64 17.4 29.88
7.76 64.24 8.24 69.74 22.49 55.16 327.11
2.03 4.47 3.25 4.65 5.38 19.42 59.59
2.91 0.46 0.49 0.8 0.13 0.0002 1.15
0.15 0.53 0.53 0.41 0.73 0.99 0.33
1163
derivation (F1,5 5 2.60; P 5 0.17), indicating that the observed effects on percentage values are not a spurious effect of total power changes. Finally, to evaluate the robustness of our results, we performed the same ANOVAs with a Bonferroni correction. Since it is too conservative in the case of correlated outcome variables, the alpha level was adjusted by taking into account the mean intercorrelation (r 5 0.47) between the dependent variables (Sankoh et al., 1997; Perneger, 1998) and the P value after Bonferroni’s adjustment was set to 0.0134. The effect observed on hippocampal high-delta band is still significant (P 5 0.0130) after the Bonferroni’s adjustment.
No differences are observed between the two nights for each considered PSG variable.
SEEG Power Spectra During Resting Wakefulness and Spatial Navigation Memory Task
particular, the more was the power density in the high-delta band during postlearning night, the larger was the performance increase at retest (see Fig. 4). No other frequency band showed significant difference between conditions (see Table 3). With respect to the cortical leads, no significant difference was observed between baseline and postlearning night for all the frequency bands (see Table 3). Control analyses (ANOVAs) carried out on total EEG power (0.5–30 Hz) did not show any significant difference between nights for hippocampal (F1,5 5 0.20; P 5 0.67) and cortical
ANOVA carried out on SEEG power spectra density of the four waking sessions showed no significant effects for the hippocampal derivations. However, it is of note that high-delta power was tendentially higher in the learning phase and positively correlated (r 5 0.60, ns) with learning performance (i.e., the higher the delta power, the shorter the learning phase). ANOVA also showed a significant main effect of Condition in the low-delta band (0.5–2.0 Hz) for neocortical derivations (F3,15 5 3.53; P 5 0.04). Post hoc comparisons showed that low-delta power was significantly higher during the Resting Wakefulness condition as compared to the other conditions
FIGURE 3. Hippocampal relative SEEG power for low-delta (0.5–2.0 Hz), high-delta (2.1– 4.0 Hz), theta (4.1–8.0 Hz), alpha (8.1–12.0 Hz), sigma (12.1–16.0 Hz), and beta (16.1–30.0 Hz) frequency bands of the first NREM episode of baseline and postlearning nights. Highdelta activity during postlearning night is significantly higher than during baseline night (*P 5 0.013). Panels are scaled differently for each band for clarity purposes. Hippocampus
1164
MORONI ET AL.
DISCUSSION In the present study we aimed at evaluating for the first time the effects of a hippocampus-dependent spatial memory task on the ensuing hippocampal sleep SEEG activity. Our findings showed that hippocampal SEEG power in the highdelta range (2.1–4 Hz) increases during the first NREM period after task training compared to baseline; such a delta power increase is related with task performance improvement at retest. These findings support the hypothesis that, in humans, hippocampal EEG activity in the delta band is involved in spatial memory formation processes during sleep.
Hippocampal SEEG Power in the High-Delta Range (2.1–4 Hz) Increases During the First NREM Period After Training on a Navigation Task FIGURE 4. Scatterplot of the individual correlations between hippocampal stereo-EEG power in the high-delta during postlearning night and CMT retest-test performance difference (negative values indicate a performance improvement).
(see Table 4 for details), presumably due to the fact that patients were more relaxed and/or behaviorally less active. Also in this case, control analyses (ANOVAs) carried out on total EEG power (0.5–30 Hz) did not show any significant difference between the four waking conditions for hippocampal (F3,15 5 0.50; P 5 0.69) and cortical derivation (F3,15 5 0.71; P 5 0.56), indicating that the observed effects on percentage values are not a spurious effect of total power changes.
In a previous study, we reported the first direct evidence that slow hippocampal oscillations are involved in human memory consolidation. In that study, in a sample of epileptic patients, we showed that a short but intensive training on a sequential finger tapping task (SFTT) is followed by an increase in the amount of hippocampal SEEG power in the very low frequency range (0.5–1.0 Hz) during the first postlearning NREM period (Moroni et al., 2008). Here we found a similar postlearning power increase, but in the upper part of the delta frequency range (2.1–4 Hz). Such a dissociation can be interpreted in terms of the features of the memory tasks adopted in the two studies. Although the SFTT is a motor task, neuroimaging studies have demonstrated that the hippocampus is
TABLE 3. Repeated Measure ANOVAS Comparing Baseline and Postleaning Night for Hippocampal and Cortical Derivations Baseline
Postlearning
Mean
SE
Mean
SE
F(1,5)
P
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Sigma (12.1–16.0 Hz) Beta (16.1–30.0 Hz)
27.22 13.51 2.75 0.94 0.57 0.11
1.04 1.13 0.23 0.19 0.15 0.02
26.77 14.3 2.68 0.85 0.56 0.11
1.12 1.15 0.22 0.17 0.17 0.02
1.32 14.21 2.54 1.78 0.07 0.002
0.3 0.013 0.17 0.23 0.8 0.96
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Sigma (12.1–16.0 Hz) Beta (16.1–30.0 Hz)
39.62 5 1.2 0.82 0.43 0.06
1.03 0.42 0.06 0.24 0.16 0.02
39.3 5.24 1.29 0.79 0.43 0.06
0.69 0.21 0.11 0.19 0.09 0.02
Hippocampus
Cortex
Relative (percentage) power spectra for each frequency band were calculated on the first NREM period.
Hippocampus
0.63 1.15 1.09 0.1 0.0009 0.22
0.46 0.33 0.34 0.75 0.98 0.66
HIPPOCAMPAL SLOW EEG ACTIVITY AND SPATIAL MEMORY
1165
TABLE 4. Means and SE of the Hippocampal and Cortical SEEG Power Spectra for Each Band During the Four Waking Conditions (Resting Wakefulness (RW), Learning (Lr), Test (Ts), and Retest (Rt) RW
Lr
Ts
Rts
Mean
SE
Mean
SE
Mean
SE
Mean
SE
F(3,15)
P
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Beta1 (12.1–16.0 Hz) Beta2 (16.1–30.0 Hz)
26.47 10.94 3.16 1.36 0.8 0.27
1.11 0.65 0.22 0.28 0.19 0.04
26.32 12.16 2.88 1.17 0.78 0.26
1.39 0.92 0.18 0.23 0.21 0.04
25.89 11.79 2.92 1.21 0.96 0.3
1.21 1.03 0.16 0.19 0.25 0.04
28.27 10.35 2.56 1.16 0.96 0.28
1.83 0.56 0.28 0.22 0.27 0.04
1.27 1.95 2.58 1.27 2.33 1.42
0.32 0.16 0.09 0.32 0.11 0.27
Low_delta (0.5–2.0 Hz) High_delta (2.1–4.0 Hz) Theta (4.1–8.0 Hz) Alpha (8.1–12.0 Hz) Beta1 (12.1–16.0 Hz) Beta2 (16.1–30.0 Hz)
18.52 8.36 5.4 1.63 1.38 0.9
3.26 1.17 1.19 0.28 0.26 0.33
15.37 8.37 6.14 1.88 1.46 1.04
3.36 0.97 1.85 0.4 0.29 0.45
14.76 7.26 5.8 2.04 1.56 1.31
3.04 0.85 1.57 0.42 0.28 0.48
14.64 7.49 6.45 1.98 1.36 1.18
2.5 0.79 138 0.27 0.22 0.34
3.53 0.89 1.07 0.97 0.33 1.46
0.04 0.46 0.39 0.43 0.8 0.26
Hippocampus
Cortex
Repeated measure ANOVAs results are also reported. Comparisons were conducted on relative (percentage) power spectra for each frequency band.
involved also in the consolidation of procedural memories, and of motor sequence-based skills in particular (Walker at al., 2005; Albouy et al., 2013). In the present study we used a truly hippocampusdependent spatial memory task, that has been shown to rely on both proper functioning and structural integrity of the hippocampus (Iaria et al., 2007, 2008), showing that a training on this task modifies hippocampal activity during the ensuing sleep period. Moreover, patients who showed more high-delta power during postlearning NREM sleep had better performance at task retest than patient who showed less high-delta power. This effect was localized at the anterior hippocampus and no differences were observed at neocortical regions. Moreover, postlearning and control nights did not differ on PSG measures, suggesting that the hippocampal high-delta power increase was strictly local and not dependent on macrostructural sleep aspects. We could hypothesize that the different delta ranges involved in sleep-related memory consolidation in our previous study (0.5–1.0 Hz) and in the present one (2.1–4.0 Hz), could be ascribed to the involvement of different and specific consolidation mechanisms. It is plausible that synaptic downscaling, typically conveyed by low-delta activity and in particular by the
