Dr K. B. Campbell, School of Psychology, Montpetit Hall, University of Ottawa, Ottawa, K1N 6N5, Canada. Tel.: +1 613 562 5800, ext. 4294; fax: +1 613 562 5150; e-mail: firstname.lastname@example.org.
Sleep spindles are 12–14 Hz oscillations in EEG, which are thought to inhibit or ‘gate’ information processing. Event-related potentials may be employed to probe the extent of information processing during sleep. Previous research indicates that event-related potentials elicited by moderate intensity stimuli show increased positivity (or further removal of negativity) when stimuli are presented concurrent with spindles. However, the effectiveness of spindles to inhibit the processing of much louder stimuli remains unknown. The purpose of the present study was to investigate the extent of this gating, by using a range of stimuli including those that are loud and intrusive. Eight good sleepers were recorded during a single night. Auditory stimuli were delivered randomly at 0, 60, 80 or 100 dB SPL. Trials were sorted off-line by sleep stage, stimulus intensity and spindle characteristic (i.e. spindle absent, spindle present). During the sleep-onset period, the often-reported changes in event-related potentials were observed – N1 decreased and P2 increased in amplitude. In Stage 2 sleep, P2 was affected by the presence of spindles, particularly when stimulus intensity was loud. Its amplitude was greatest when spindles occurred following the onset of the stimulus. Scalp-recorded spindles might, therefore, be a consequence of the prior thalamic inhibition of information processing, especially when confronted by loud, intrusive external stimuli.
Spindle oscillations are phasic events in the electroencephalographic (EEG) recording of non-rapid eye movement (REM) sleep. They were first described by Loomis et al. (1935a , 1935b) as rhythmic 12–14 Hz periodic events lasting from 1 to 1.5 s. Spindle activity is thought to be generated in the thalamus as a result of a network of synaptic interactions involving inhibitory neurons (i.e. GABA) of the reticular thalamic nucleus, thalamocortical cells and cortical pyramidal neurons ( Steriade and Llinas 1988; Steriade et al. 1993 ). Steriade and Amzica (1998) proposed that during sleep, the depolarizing component of a cortically driven slow wave (<1 Hz) serves to trigger thalamic spindles. These rhythmic oscillations are due to the inhibitory postsynaptic potentials (IPSPs) of thalamic reticular neurons. Thus, a possible role of the spindle may be to gate synaptic transmission through the thalamus, thereby allowing sleep maintenance through inhibition of processing sensory information from the external environment.
Event-related potentials (ERPs) offer a method to probe the extent of information processing during human sleep ( Campbell et al. 1992 ). ERPs are changes in electrical activity of the nervous system as elicited by an external physical stimulus or an internal psychological event ( Picton et al. 1995 ). They are often used to investigate information processing associated with changing levels of attention and arousal. ERPs consist of a series of negative and positive deflections or ‘components’. The late ‘vertex’ potential of the auditory ERP consists of ‘N1’ (negative wave peaking at ≈ 80–100 ms) and ‘P2’ (positive wave peaking at ≈ 175–225 ms) deflections. These late components have been shown to be markedly affected by sleep ( Williams et al. 1962 ; Fruhstorfer and Bergström 1969; Noldy et al. 1988 ; Nielsen-Bohlman et al. 1991 ; Ogilvie et al. 1991 ; Campbell et al. 1992 ; Salisbury et al. 1992 ; Harsh et al. 1994 ; Bastuji et al. 1995 ; Winter et al. 1995 ; de Lugt et al. 1996; Elton et al. 1997 ). In general, during non-REM sleep, when a sufficiently long time constant is employed and baseline-to-peak measurements are used, N1 decreases while P2 increases in amplitude ( Campbell et al. 1992 ). Campbell et al. (1992 ) suggested that this increase in amplitude of the positive wave, P2, and decrease in the amplitude of the negative wave, N1, is due to the removal of a long-lasting negative wave during sleep. This slow negative wave, labelled ‘processing negativity’ ( Näätänen 1990) overlaps and summates to the N1 and P2 components during wakefulness. Processing negativity is thought to reflect the additional processing that is received by attended stimuli. The changes in the N1–P2 vertex complex at sleep onset and during non-REM sleep are therefore considered to reflect inhibition of information processing.
Recently, Elton et al. (1997 ) provided ERP evidence for the inhibitory effects of sleep spindles during Stage 2 of non-REM sleep. ERPs to auditory stimuli were analysed over a 1-s epoch or ‘sweep time’. Trials were sorted into those in which stimuli were present during spindle activity (SP, or spindle present trials) and those in which spindles were absent during stimulus presentation (SA, or spindle absent trials). During SA trials, the usual effect of sleep was noted – N1 decreased in amplitude while P2 increased compared with the waking period. During SP trials, P2 increased further in amplitude. This increased positivity in the ERP for trials in which spindles were present was interpreted as reflecting additional inhibition of information processing over and above that usually noted during Stage 2 sleep.
The purpose of the present study was to extend the methodology employed by Elton et al. (1997 ) in investigating the role of the sleep spindle on human information processing during sleep. Elton et al. (1997 ) presented a moderate intensity, 65 dB SPL binaural auditory stimulus. The extent to which spindles can protect the sleeper against the intrusive effects of much louder stimuli has yet to be determined. Auditory tone pips ranging from low (60 dB) to high (100 dB) intensity were therefore presented in this study. The extent of information processing may well be different for trials in which the spindle occurs at the same time as stimulus presentation, compared with when the spindle occurs after the stimulus has already been delivered. In the present study, trials in which spindles were identified were further separated into those in which the spindle occurred concurrently with or following stimulus presentation. Furthermore, Elton et al. (1997 ) presented stimuli at a constant, fixed rate. It is possible, although unlikely, that the sleeping participants might have anticipated the stimulus. The anticipation of the stimulus, rather than the processing of the stimulus per se, might have caused the changes observed in the ERP. To control for this possibility, stimuli were occasionally omitted (i.e. not presented). If changes in the ERP were due to anticipation of the stimulus, then ERPs should also be apparent in the no stimulus condition.
Ten, healthy, good sleepers volunteered to spend one night in the sleep laboratory. Data were later rejected for two participants because of insufficient sleep during the night. Statistical analyses are therefore based on the data from eight participants (four females) aged 19–30 y (mean age=22.3, SD=4.0 y). All participants were right-handed, nonsmokers and were free from medication at the time of study. None reported a history of neurological disorder. None reported disordered sleep as indicated by the Yoshitoke Fatigue Questionnaire ( Yoshitoke 1978) and a sleep/wake history questionnaire. Hearing was verified using an audiometer to be within 15 dB ISO at 500, 1000, 1500 and 2000 Hz frequencies. Prior to the recording sessions participants were instructed to abstain from naps, alcohol and caffeine. Participants signed informed consent and received a $25 honorarium for their participation in the study. This study was conducted according to the guidelines for ethical principles of the Medical Research Council of Canada.
The EEG was recorded from gold electrodes placed at midline (Fz, Cz, Pz) and temporal (T7, T8) sites, and referenced to the tip of the nose. A vertical electro-oculogram (EOG) was recorded from electrodes placed at the supra- and infra-orbital ridges of the right eye. A horizontal EOG was recorded from the outer canthi of each eye. The physiological signals were amplified using a Nihon Kohden model 4314B polygraph, with the high frequency filter set at 35 Hz, and a time constant of 1 s. EEG and EOG were digitized using a 12-bit analogue-to-digital (A/D) converter with a sampling rate of 256 Hz per channel.
Auditory stimuli were 1000 Hz tone pips, having a total duration of 55 ms and a rise-and-fall time of 5 ms. Stimulus intensity was set at either 0, 60, 80 or 100 dB SPL. Each intensity was delivered at random with an equal probability (P=0.25) of occurrence. Stimuli were presented at a fixed interstimulus interval (ISI) of 2000 ms. The stimuli were presented in blocks of 480 trials. Each intensity was therefore presented 120 times within each block. The order of presentation of the various intensities was randomized. All auditory stimuli were synthesized using an InstEP Systems 16-bit waveform generator card and were presented monaurally to the right ear via a modified hearing aid device. The hearing aid system assured constancy of auditory input in spite of possible head movements. A Bruel and Kjaer 2209 sound-level meter equipped with a 2-cm3 coupler was used to calibrate the auditory intensities at the beginning and end of each night.
All participants underwent a screening procedure in which they were tested for normal hearing and completed sleep questionnaires. Upon arrival at the laboratory, electrodes were affixed, a hearing aid device was fitted and a presleep questionnaire was completed. EEG was recorded and stimuli were delivered during relaxed wakefulness while the subjects read a book. Subjects were then permitted to fall asleep. Presentation of auditory stimuli began after consolidated sleep onset latency (i.e. 5 min following continuous Stage 2 sleep). A minimum of two blocks of 480 trials were presented in Stages wake, 2 and slow-wave sleep (SWS).
An on-line spindle detector was used to identify 11–15 Hz activity in the EEG. The spindle analyser consisted of sharp analogue filters having a bandpass of 11–15 Hz. Spindles were later verified by visual inspection and included only 11–15 Hz activity that exceeded 25 μV and had a duration of 0.5 s. The continuous physiological signals were sorted into sleep stages by an experienced rater who used standard Rechtschaffen and Kales (1968) procedures. In cases of stage ambiguity, the epochs were excluded from further analysis. Stage 2 sleep was separated into first and second halves of the night in order to determine time of night effects on information processing.
The continuous EEG and EOG were reconstructed into discrete epochs (‘sweeps’) off-line. A sweep consisted of 256 data points beginning 100 ms prior to stimulus presentation and continued for 900 ms following it. Trials in which the EEG exceeded ± 150 μV were rejected from further analysis. This effectively removed those trials in which K-complexes were elicited. During the waking state, trials in which the EOG exceeded ± 100 μV were rejected. Single trials were stored on disk for subsequent off-line analysis. ERP waveforms were later filtered digitally in the frequency domain (employing an inverse FFT algorithm) using a low pass filter of 15 Hz.
The data were sorted on the basis of stimulus intensity, stage of sleep and the presence or absence of a spindle activity. Trials were initially sorted into those in which a spindle was present during the sweep and those in which it was absent (SA category). When spindles were identified during the sweep, they were further sorted according to those that occurred concurrently with stimulus presentation (SC category), and those that occurred following stimulus presentation (SF category). Figure 1 illustrates the detection of spindle activity (11–15 Hz) by the on-line spindle detector (filter).
The N1 peak is often difficult to detect during sleep because its amplitude is attenuated to near baseline level. For this reason, the amplitude of N1 was measured using a data point averaging method. N1 was defined as the average of all data points from 75 to 125 ms following stimulus onset. A distinctive P2 is however, usually visible in both waking and sleeping states. The peak amplitude of P2 was therefore measured relative to the average of all data points in the prestimulus interval (the ‘baseline’). P2 was initially measured at Cz as the maximum positive peak between 175 and 250 ms. Its amplitude was subsequently measured at all other sites at this latency.
The amplitudes of N1 and P2 did not differ between the first and second halves of the night (F < 1 in all cases). Data from both early and late Stage 2 sleep were, therefore, collapsed. This increased the number of trials available for sorting and averaging according to the different spindle categories, thereby further reducing background EEG noise. Spindles were present on approximately 37% of trials in Stage 2 sleep. For each intensity, an average of 80 and 72 trials were identified per subject as SC and SF categories, respectively. There were insufficient data from all participants during SWS to permit reliable sorting and averaging of the different categories of spindle activity. The effects of stimulus intensity and sleep spindle activity on ERPs are, therefore, reported for Stage 2 sleep across the entire night.
Wakefulness and Stage 2 sleep
The grand average ERP waveforms in waking and Stage 2 states are shown in Fig. 2. The latencies of N1 and P2 were somewhat delayed during Stage 2 (peaking at 110 and 205 ms, respectively), compared with wakefulness (peaking at 93 and 192 ms, respectively), but this difference was not significant (F < 1).
Differences between the waking and Stage 2 sleep data were determined using a two-way ANOVA with repeated measures on sleep stage (wake and Stage 2) and intensity (0, 60, 80, 100 dB SPL). Data from the Cz site, in which N1 and P2 were largest, were used for the analysis. A main effect of sleep stage was found for the amplitude of N1, F1,7=18.16, P < 0.05. N1 was significantly larger (i.e. more negative) during wakefulness and became markedly attenuated to near baseline levels during Stage 2 sleep. An interaction between intensity and sleep stage was also found for the amplitude of N1, F3,21=12.54, P < 0.05. Simple main effects testing indicated that during wakefulness, N1 decreased significantly in amplitude from 100 dB to 60 dB to 0 dB SPL. Stimulus intensity had no significant effect on N1 amplitude during sleep.
The amplitude of P2 showed a small increase in Stage 2 of sleep. However, neither the main effect of sleep stage nor the sleep stage by intensity interaction reached significance, F < 1 in both cases. A statistically significant main effect of intensity was found for the amplitude of P2, F3,21=22.08, P < 0.05. P2 amplitude was significantly larger (i.e. more positive) for the 100 dB SPL intensity compared with the 0, 60 and 80 dB SPL intensities.
Effects of spindle activity in Stage 2 sleep
No ERPs were visible following the 0 dB ‘stimulus’ in either waking or sleeping states. All subsequent statistical analyses were therefore based on when a stimulus was actually presented (i.e. following 60, 80 or 100 dB SPL stimuli). Figure 3 illustrates the grand averages when trials were sorted according to spindle absent (SA), spindle concurrent (SC) and spindle following (SF) categories.
Elton et al. (1997 ) indicated that spindle activity has its largest effects at centro-parietal regions. For this reason, separate two-way ANOVAs with repeated measures on spindle category and stimulus intensity were run at each electrode site. No significant main effects or interactions were found at any electrode site for the amplitude of N1. The means and standard deviations for P2 amplitude are presented in Table 1. There were no significant main effects or interactions for the amplitude of P2 at the temporal (T7, T8) sites. At the midline frontal (Fz) site, there was a main effect for intensity, F2,14=11.16, P < 0.01, but the effect of spindle category was not significant. At Cz, the spindle by intensity interaction was significant, F4,28=3.33, P < 0.05. Simple main effects testing was employed to isolate the source of this interaction. For the 100 dB intensity, P2 was larger when a spindle followed stimulus presentation (SF category) than when it occurred concurrently with the stimulus (SC category) or when the spindles were absent (SA category). P2 amplitude did not vary significantly between SC and SA categories. For both the 60 and 80 dB stimuli, differences were much smaller (and not significant), although the effects were consistent with those in the louder intensity category. A similar trend was noted for the Pz data, but the interaction did not reach significance. The main effect for intensity at Pz was significant, F2,14=28.70, P < 0.0001, and the main effect for spindle category tended toward statistical significance, F2,14=3.52, P < 0.06. Again, P2 was largest when spindles followed stimulus presentation.
Table 1. Mean amplitude (μV) and standard error (in parentheses) for P2 at Fz, Cz, Pz during Stage 2 sleep across spindle absent (SA), spindle concurrent (SC) and spindle following (SF) trials
In the waking state, clear N1 (peaking at 110 ms) and P2 (peaking at 205 ms) waveforms were observed. The amplitudes of N1 and P2 increased with increasing stimulus intensity. This is consistent with many previous studies (see Näätänen and Picton 1987). No auditory evoked potential was visible following stimulus omission (i.e. when stimulus intensity was set to 0 dB) during either waking or sleeping states. There is thus no evidence that the N1–P2 waveform can be explained by anticipation of the stimulus.
Näätänen and Picton (1987) have indicated that N1 and P2 are affected by both ‘exogenous’ and ‘endogenous’ factors. On the one hand, manipulation of the exogenous, physical characteristics of the stimulus will affect these components. On the other hand, manipulation of endogenous, psychological constructs will also affect N1 and P2. For example, directing the subject’s attention may cause N1 to increase and P2 to decrease in amplitude, because of the presumed overlapping effects of the PN ( Näätänen 1990). In the present study, while awake, subjects ignored the auditory stimulus and read a book. Since subjects were inattentive, the PN should have been minimal. The amplitude of N1–P2 should have been determined mainly by the physical attributes of the stimulus. However, Campbell et al. (1992 ) suggested that the apparently exogenous effects on N1 may be explained by an attentional confound. Awake and alert subjects are probably never able to completely ignore an auditory stimulus. Indeed, Campbell et al. suggest that the waking N1 is largely, if not completely, endogenous in nature. Its amplitude is determined by the extent to which subjects attend to the auditory stimulus, even if they are instructed to ignore it. In the example of manipulation of stimulus intensity, as the stimulus becomes increasingly louder, the subject may no longer be able to ignore it. Increases in stimulus intensity cause N1 to increase in amplitude. This could be due to due to a psychological effect – the subject is unable to ignore the stimulus. The same argument cannot account for the P2 data. P2 increases in amplitude as stimulus intensity is increased, but decreases in amplitude when the subject is more attentive. P2 therefore appears to show both exogenous and endogenous (attentional) effects.
P2 increased in amplitude from wakefulness to Stage 2 of sleep, although the difference did not attain statistical significance. This increase in amplitude has also been reported in other studies ( Noldy et al. 1988 ; Ogilvie et al. 1991 ; Harsh et al. 1994 ), although not all studies report this finding (see Campbell et al. 1992 ). The decrease in amplitude of N1, but increase in amplitude of P2, has been attributed to the removal of the long-lasting and summating effects of the waking PN. The removal of an overlapping negative wave will, of course, cause negative waves to decrease in amplitude (i.e. become less negative) but it will also cause positive waves to increase in amplitude (i.e. also become less negative or more positive). The decrease in the amplitude of N1 and the increase in amplitude of P2 was precisely what was observed during Stage 2 sleep.
Spindle activity will further modulate the N1–P2 effect. Elton et al. (1997 ) observed a significant increase in P2 amplitude when spindles occurred simultaneously with presentation of moderate intensity, binaural 65 dB SPL stimuli. In the present study spindles did not interact with the processing of the 60 or 80 dB stimulus. Several explanations of the apparent discrepancy between the two studies could account for these differences. The present study employed monaural stimuli. It is now well established that binaural stimuli are perceived as being louder than monaural stimuli ( Pollack 1948). Moreover, the amplitude of short-, mid- and long-latency evoked potentials is larger for binaural than monaural stimuli ( Debruyne 1984; Polyakov and Pratt 1994). N1–P2 increases in amplitude when stimuli are presented binaurally compared with when they are presented monaurally. The 65 dB binaural stimulus employed by Elton et al. (1997 ) would probably have been perceived as being as loud as the 80 dB monaural tone pip. Further evidence for this claim is that during Stage 2, the ERP to our 80 dB stimulus was quite similar to that elicited by the 65 dB stimulus employed by Elton et al. (1997 ).
As is apparent in the grand averages, the ERP to the 60 dB stimulus was noisy. A distinctive peak was not visible in the background noise. In contrast, a distinctive P2 was apparent following the 80 dB stimulus. While it was slightly larger when spindles followed the stimulus, there were no P2 differences when spindles were concurrent with the stimulus compared with when they were absent. This finding fails to replicate that observed by Elton et al. (1997 ). They employed an odd-ball task in which a standard stimulus was presented on 80% of trials and a deviant (varying in pitch) on 20% of trials. It is possible that the rare deviant stimulus might have interacted with spindle occurrence differently to the equal probability stimuli employed in this study. Alternatively, it is possible that the 65 dB standard delivered on 80% of trials ( Elton et al. 1997 ) compared with the 80 dB stimulus delivered on 33% of trials in the present study, might also explain differences in spindle effect.
The present study indicated that spindles will affect the processing of very loud, 100 dB, stimuli to a much greater extent than moderate or low-intensity stimuli. The increase in the amplitude of P2 for high-intensity stimuli in the presence of spindles is consistent with further inhibition of information processing during Stage 2 sleep. The present study and that of Elton et al. (1997 ) indicate that spindle activity will have its greatest effects over centro-parietal areas of the scalp. Importantly, the increase in P2 amplitude was largest when spindles occurred following stimulus presentation. In these trials, therefore, inhibition appears to occur before the actual onset of the scalp-recorded spindle. In this interpretation, it may be that thalamic inhibition of information processing must occur initially, prior to the onset of the scalp-recorded spindle. These spindles may, therefore, be a consequence of the inhibition of information processing rather than the cause of it. Depth electrode recordings of thalamic and cortical cell activity offer support of this hypothesis. Some time ago, Steriade et al. (1971 ) noted that antidromic invasion of thalamocortical cells is diminished or abolished well before overt EEG spindles are observed in the cortex.
This research was supported by a grant from the Natural Science and Engineering Research Council (NSERC) of Canada. K. Cote was supported by a predoctoral fellowship from NSERC and an Individual National Research Service Award (NRSA) from the National Institute of Mental Health (NIMH) of the USA. The authors wish to thank Herman van der Bergen for the design and development of the spindle detection equipment employed in this study. We also wish to thank the anonymous reviewers for their helpful comments on the manuscript.