Respiratory-related evoked potentials during the transition from alpha to theta EEG activity in Stage 1 NREM sleep


ColrainIan M. Department of Psychology, School of Behavioural Science, University of Melbourne, Parkville, Victoria 3052, Australia. Tel: +61 3 9344 6350; fax: +61 3 9347 6618; email:


It has been argued previously that evoked potential components during Stage 1 sleep in response to both auditory and respiratory stimuli are intermediate between those of wakefulness and Stage 2 sleep. However, state fluctuations in the EEG between alpha and theta during Stage 1 sleep have been linked to changes in a number of respiratory functions including ventilation, upper airway resistance and chemical drive. It was therefore hypothesized that if respiratory related evoked potentials (RREP) were averaged separately for alpha and theta EEG periods during Stage 1 sleep, the alpha RREP would resemble wakefulness and the theta RREP would resemble Stage 2 sleep. RREPs were produced by 250 ms occlusions in 10 subjects. EEG was recorded from 29 scalp sites, referenced to linked ears, together with EOG and EMG. The N1 component was not specifically associated with alpha vs. theta activity, but appeared to be sensitive to any decrease in arousal level, suggesting that it was more related to attention than to changes in the EEG. The late N2 and P300 components were present during wake and Stage 1 alpha. However, in Stage 1 theta, different late components emerged (N300 and P450) that differed in latency, amplitude or topographical distribution from those seen in wakefulness. The P2 proved difficult to interpret, whereas the N550 did not appear until Stage 2 sleep, and as such, was not dependent on alpha/theta state. The results indicate that RREP components are differentially affected by the transition into sleep.


Evoked potential (EP) waveforms are affected markedly by sleep as illustrated by studies investigating respiratory-related evoked potentials (RREP; Webster and Colrain 1998a; Wheatley and White 1993) and auditory evoked potentials (AEP; Nielsen-Bohlman et al. 1991; Campbell et al. 1992; Van Sweden et al. 1994 ). The late RREP and AEP components present during wakefulness are the N1, P2, N2 and P300. During Stage 2 sleep, the N1 component is diminished considerably, while the P2 peak may be either slightly augmented or attenuated. The endogenous P300, which is thought to be related to attentional memory processes ( Johnson 1993), can be elicited in wakefulness by both respiratory and auditory stimuli. During Stage 2 sleep, the auditory P300 disappears. However, a late positivity peaking at 450 ms can still be elicited by infrequently occurring stimuli. The P450 appears to have a different latency and topographic distribution than the P300 ( Winter et al. 1995 ). Two major negative components also emerge during sleep, the N300 (replacing the wake N2) and the N550 ( Campbell et al. 1992; Webster and Colrain 1998a ).

The RREP has been reported to occlusion stimuli ( Davenport et al. 1986, 1996; Revelette and Davenport 1990; Wheatley and White 1993; Logie et al. 1998; Webster and Colrain 1998a ), and to resistive loads ( Bloch-Salisbury and Harver 1994; Knafelc and Davenport 1997; Webster and Colrain 1998b). It consists of a series of early components within 100 ms of stimulus delivery. A positive component (P1) is seen over central and postcentral scalp regions, and a negative component (Nf) over frontal scalp regions ( Davenport et al. 1996; Webster and Colrain 1998a ). It has recently been determined that these are generated within primary somatosensory and supplementary motor cortices, respectively ( Logie et al. 1998 ). As would be expected for a primary somatosensory cortex component, the P1 is sensitive to the intensity of the eliciting stimulus, with Knafelc and Davenport (1997) reporting a log–log correlation of 0.98 between P1 amplitude and the intensity of the resistive load used as a stimulus. Components very similar to P1 have been observed in S1 cortex following direct simulation of the contralateral phrenic nerve in the cat ( Davenport et al. 1985 ), mechanical stimulation of the intercostal muscle of the cat ( Davenport et al. 1993 ) and electrical stimulation of the intercostal muscle in humans ( Gandevia and Macefield 1989).

Late components resembling the endogenous evoked potential components reported previously in other stimulus modalities have also been produced by occulsion and resistive load stimuli. Harver et al. (1995) showed an increase in amplitude in the late positivity when the stimuli were attended rather than ignored, while Strobel and Daubenspeck (1993) showed marked attenuation of a late positivity when stimuli were applied to every breath. Webster and Colrain (1998a, b) indicated that the late positivity (P300) has the centro-parietal maximum topographic distribution characteristic of AEPs and visual evoked potentials (VEPs). RREP studies have also reported both augmentation of the P300 with attention ( Webster et al. 1997 ) and increased P300 amplitude with decreased probability of stimulus presentation ( Colrain et al. 1998 ).

The purpose of this study was to investigate the cortical response to a physiological event (respiratory occlusion) at the transition from waking to sleeping. Only a small number of studies have examined EPs during sleep onset. Ogilvie et al. (1991) measured subjects’ EPs in response to 1000 Hz tone pips. Subjects’ sleep/wake state was defined by behavioural response patterns, with the absence of a response to an infrequently occurring auditory stimulus being indicative of sleep. The N1 component diminished with increased reaction time and was virtually absent in definitive sleep. The N2/N300 component increased and the P300 decreased in amplitude with the onset of sleep. Similar results were observed by Harsh et al. (1994) , also using a reaction time/sleep-onset protocol. Winter et al. (1995) applied an auditory oddball paradigm and observed that the large N1 and P300 components that were apparent during wakefulness were difficult to identify during Stage 2 sleep. In Stage 1 (termed drowsiness), an N330 and P420 emerged that were both sensitive to the intensity of the auditory stimulus. de Lugt et al. (1996) recorded AEPs in response to 1000 Hz tone pips. The N1 component was attenuated during Stage 1 sleep. However, the N2 was not found to increase in amplitude during sleep onset. Webster and Colrain (1998a) recorded RREPs during Stage 1 sleep. The respiratory stimulus used was an occlusion that completely interrupted breathing at mid-inspiration for about 250 ms. This stimulus was presented every 3–6 breaths (minimum of 12 s between stimuli), making it analogous to a target stimulus in an oddball paradigm. They reported that the N1, P2 and P300 all showed small decreases in amplitude from wakefulness.

There has been some inconsistency in the literature regarding the labelling of the N2/N300 and to a lesser extent, the P300/P450 components. Early AEP research did not differentiate between the wake and sleep ‘N2’ ( Ornitz et al. 1967 ). Research since then has either followed this pattern of nomenclature ( Ogilvie et al. 1991; Wheatley and White 1993; Van Sweden et al. 1994; de Lugt et al. 1996 ), or described the sleep N2 in terms of its latency, such as N300 ( Niiyama et al. 1994 ), N340 ( Nielsen-Bohlman et al. 1991 ) or N350 ( Harsh et al. 1994; Winter et al. 1995 ). In general, the majority of studies investigating the late positive component (appearing between 300 and 500 ms) during wakefulness and sleep have classified this component according to latency of appearance ( Nielson-Bohlman et al. 1991; Ogilvie et al. 1991; Salisbury and Squires 1993; Harsh et al. 1994; Winter et al. 1995 ). The waking late positivity is usually labelled ‘P300’, with the sleep late positivity labelled P420, P430 or P450, depending on its latency in different studies. For nomenclature purposes, the present study will refer to the late negative component in wake as an N2 and in sleep as an N300. The late positive component will be classified as a P300 in wake and a P450 in sleep.

A major problem with the above-mentioned sleep studies is that the transition from waking to sleeping states was generally measured conventionally as Stage 1 sleep. This is problematic as Stage 1 represents a period of oscillation between two different frequency bands, EEG dominated by alpha activity and EEG dominated by theta activity. As such, the definition of subjects state of arousal during sleep onset has not been precise.

There are neurophysiological reasons to anticipate that EEG state during sleep onset would effect EP. The pattern of EP activity observed in a given arousal state will depend on the manner in which thalamic afferent signals are transmitted to the cortex. This transmission is critically dependent on the state of thalamocortical relay cells. According to Coenen (1995), the development of sleep is related to thalamic blocking or gating (and its associated reduction in level of consciousness) and is due at least in part to decreased brainstem reticular formation (BRF) input to the thalamus. EPs recorded at the scalp are different in wake compared with sleep because thalamocortical cells are at different levels of polarization, with the wake level being close to action potential generation threshold and the sleep level being relatively hyperpolarized. Variations in the EEG also reflect variations in the polarization of thalamic neurons, with the movement from faster frequency, low-amplitude EEG (alpha) to slower frequency, higher amplitude EEG (theta) occurring in association with thalamocortical hyperpolarization. We are therefore proposing an ‘EEG alpha/theta-state-dependent’ model of activity such that if the polarization of thalamic neurons determines EP waveforms, then EPs should be dependent on EEG-defined state. The prediction of this model is that EP components will oscillate between wake and sleep levels in association with alpha and theta EEG activity oscillations that are characteristic of the sleep onset period.

Evidence for an EEG alpha/theta-state-dependent model comes from the study of respiratory activity during sleep onset. Alpha/theta-state-specific changes have been found to occur in ventilation ( Colrain et al. 1987, 1990; Trinder et al. 1992 ), upper airway resistance ( Kay et al. 1994, 1995 ), and chemical drive ( Dunai et al. 1996 ). State instability has also been associated with changes in upper airway and pump muscle activity ( Worsnop et al. 1998 ) and the reduced ability to compensate for an inspiratory resistive load ( Gora et al. 1998 ). In each case, the change from alpha- to theta-dominated EEG activity marked the transition in respiratory activity. If the nature of the BRF influence on thalamic circuits is the same as for respiratory and upper airway motoneurons, it would be expected that the transfer of RREP components from a wake- to a sleep-type pattern would occur in association with oscillations from alpha to theta EEG activity during sleep onset.

The aim of this study was to assess the effect of state, as defined by alpha and theta EEG activity during sleep onset, on the presence of RREP components evoked by brief inspiratory occlusions.


Subjects and design

The subjects were five males (mean age 21.6±2.51 y) and five females (mean age 21.6±2.70 y). All were healthy nonsmokers who were free from sleep-related respiratory disorders. Subjects were studied over three nonconsecutive nights in the sleep laboratory, with EPs elicited by brief but complete inspiratory occlusions. EP components were compared across sleep/wake states taking scalp distribution into consideration. The study was approved by the University of Melbourne's Human Subjects Ethics Committee, and subjects gave written informed consent prior to participation.


General laboratory procedures. Subjects were asked to refrain from consuming caffeine or alcohol on the day prior to each sleep session and were required to maintain a supine position during data collection. As the study was concerned with the sleep-onset period, each experimental night involved use of a multiple sleep-onset technique. After the attainment of approximately 10 min of stable Stage 2 sleep, subjects were woken and kept awake until alert and then requested to go to sleep again. A 3-h recording session resulted in up to eight sleep onsets.

Application of respiratory stimuli. Subjects wore a face mask (Hans-Rudolph Series 7940) that was positioned so that subjects could comfortably respire with minimal facial muscle activity. The mask was secured using a head strap to cover the nose and mouth and was attached to a two-way nonrebreathing valve (Hans-Rudolph Series 2600). The mask and breathing valve had a dead space of ≈120 mL. Inspiratory airflow was measured using a heated Fleisch pneumotachograph placed in the inspiratory line, and connected to a differential pressure transducer (Validyne DP45–14), with the output converted to a voltage signal using a carrier demodulator (Validyne CD15). The inspiratory port of the nonrebreathing valve was connected to a pressure-activated occlusion valve (Hans-Rudolph 2100) with a response closure time of 2 ms. The sound generated by valve closure was ≈40 dB. Despite this small noise artefact, the advantage of the pressure-activated occlusion valve over previous methods ( Davenport et al. 1986; Revelette and Davenport 1990 ) was that it allowed for greater precision in the timing of occlusion stimulus presentation. Inspiration was interrupted by manual activation of the occlusion valve during mid-inspiration (as determined from the airflow signal). Occlusions lasted ≈250 ms and were presented randomly every 3–8 breaths. Mask pressure (Pm) was monitored from a tap on the inspiratory port of the breathing mask using a second differential pressure transducer (Validyne DP45–14). Subjects were instructed about possible mask leaks during wakefulness and were asked to alert the experimenter if these occurred. The experimenter also checked the mask for leaks during the period of wakefulness between each sleep onset.

Measurement of EEG, sleep state and digital recording. EEG activity was recorded (using an ECI electrocap) from 29 tin electrodes referenced to linked ears. (FP1, FP2, Fz, F3, F4, F7, F8, FCz, FC3, FC4, FT7, FT8, Cz, C3, C4, T3, T4, CPz, CP3, CP4, TP7, TP8, Pz, P3, P4, T5, T6, O1 and O2), based on the International 10/20 system. A vertical EOG was recorded from electrodes placed at the supra and infra orbital ridges of the right eye. A horizontal EOG was recorded from electrodes placed at the outer canthus of each eye. An electrode placed midway between Fz and Cz served as the ground. Electrode impedance values were tested and maintained below 5 KΩ.

A Neuroscan system recorded all EEG sites, the vertical EOG and Pm at a sampling rate of 1000 Hz. Single trials were recorded over 1500 ms epochs with a baseline of 100 ms. The bandpass filters for EEG and Pm were set to 0.1–100 Hz, while the input range was±225 μV.

The C3 and Oz EEG, EOG, EMG and Pm activity were also recorded continuously on a Compumedics system at a sampling rate of 500 Hz. Bandpass filters were set at 0.1–40 Hz for the EEG (range±125 μV); 0.1–40 Hz for the EOG (range±250 μV) and 3–40 Hz for the EMG (range±125 μV). On the basis of these recordings, each sleep onset was divided into three states; presleep wakefulness, Stage 1 and Stage 2 sleep using standard criteria ( Rechtschaffen and Kales 1968). As described below, Stage 1 sleep was subdivided into periods of alpha and theta EEG activity.

Automated EEG scoring. The EEG was analysed on the basis of 1 s epochs. Thus, a peak-to-peak period analysis of the Oz EEG was conducted to determine the frequency spectrum associated with each second. The ratio of EEG alpha activity (> 8Hz) to theta activity (≤8Hz) was then calculated for each second. This estimated ratio was then compared with a criterion ratio in order to classify each second as being associated with predominantly alpha or theta EEG activity. The criterion ratio was computed separately for each subject due to individual variability of alpha in the waking state. To calculate the criterion ratio for each subject, samples of 20 pages (≈600 s) of both unambiguous alpha (from wakefulness) and theta (from Stage 1 sleep) EEG activity were identified visually identified by an experienced scorer from the Compumedics trace. The alpha/theta ratio for each of these seconds was calculated resulting in two distributions, one for wake and one for Stage 1 sleep. A signal-detection method (equal likelihood ratio) was used to determine the value (criterion) that best discriminated between the two distributions. The resulting criterion ratios were 0.81, 0.80, 0.87, 0.80, 0.86, 0.87, 0.82, 0.83, 0.82, 0.84 for each subject, respectively. These methods have been described in detail in a previous publication ( Kay et al. 1994 ). In previously conducted unpublished work, we established that there is substantial agreement between visual scoring of alpha and theta EEG activity and the present algorithm. Using signal detection theory between visual scoring and automated analysis, d′ values for four subjects were calculated to be 3.54, 2.88, 2.39, 2.72 (perfect agreement d′=4.66). Furthermore, the percentage agreement for the classification of epochs into those occurring in alpha or theta for visual scoring and automated analysis was > 90% for each of the four subjects. In the present study, an occlusion during Stage 1 sleep was classified as having been applied in alpha or theta based on the second preceding the stimulus.

Data reduction and analysis. Periods in which respiration was disturbed or affected by body movements (as exhibited by EEG, EOG and EMG recordings) were not selected for analysis. Furthermore, trials in which eye movements affected the evoked potential waveforms were also discarded from the analysis. EP analysis consisted of measuring the amplitude, latency and topographic distribution of the late latency component waveforms. For wakefulness, these included the N1 (70–130 ms), a P2 component (160–240 ms), an N2 (200–280 ms) and a P300 (250–450 ms). For sleep these included the N1 and P2 as above and an N300 (250–400 ms) and a P450 (350–500 ms). Amplitude and latency data have also been given for the N550 component that was present only during Stage 2 sleep. Amplitude values for component peaks that had very small amplitudes were taken from the largest peak that occurred within the time window specified above for each particular component. All latencies were expressed relative to the start of Pm change, which occurred 36 ms after valve closure due to the elasticity of the air inside the 3 m of tubing between the valve and the face-mask. Latency values were determined at the relevant midline scalp sites for each component by taking the latency value corresponding to the peak amplitude. Topographic maps were constructed for the time slices corresponding to the maximum deflections in the appropriate time window for each component, at Cz for N1, P2 and N2/N300, at Pz for P300/P450 and at Fz for N550.

Statistical analysis. All 29 scalp sites were used for statistical analysis of amplitude and scalp topography. However, only midline sites (Fz, FCz, Cz, CPz, Pz, O1) were used for analysis of latency. This was because in most cases, the more lateral sites had much smaller peak amplitudes and as such, some of the variation in latency at these sites may be misleading.

The amplitude of each component peak (N1, P2, N2/N300 and P300/P450) was assessed using a two-factor repeated-measures ANOVA model. The two factors were EEG state (wake, Stage 1 alpha, Stage 1 theta, Stage 2 sleep) and scalp site (29 sites). Differences in scalp topography between different EEG states could be assessed by evaluating the state-by-site interaction term in the model. To facilitate this comparison, data were scaled using the normalization procedure outlined by McCarthy and Woods (1985). However, it should be noted that a significant state-by-scalp-site interaction effect can occur for two reasons: either a change in topographic distribution, or the appearance or loss of a localized component following a state change. Thus, when interpreting significant interaction effects the graphical representation of the data was taken into consideration. Main effects are reported based on analysis of unscaled data.

Where a significant main or interaction effect was indicated, three pair-wise repeated measures ANOVAs for each component peak were conducted in order to identify at which point during sleep onset any changes were occurring. Each analysis consisted of a 2 (state) by 29 (scalp site) ANOVA, where state refers to wake vs. Stage 1 alpha in the first analysis, Stage 1 alpha vs. Stage 1 theta in the second and Stage 1 theta vs. Stage 2 sleep in the third. All reported probability levels were Greenhouse–Geisser corrected.

Statistical analysis of latency involved the use of a 4 (state) by 6 (scalp site) repeated measures ANOVA conducted for each component. State was the same as above, with scalp site referring to the six midline sites.


The average number of trials obtained for subjects within each state was 194±44.4 in wakefulness, 151±22.9 in Stage 1 alpha, 166±37.7 in Stage 1 theta and 241±50 in Stage 2 sleep.

N1 component

As illustrated in Fig. 1 and Table 1, the N1 component appears to gradually diminish in amplitude from wakefulness through to Stage 2 sleep. Figure 1 depicts the grand mean waveform for all trials within subjects for the four states at the Cz scalp site. There was a significant main effect of state for amplitude (F3,27=42.27, P< 0.001). Pair-wise ANOVAs revealed reductions in amplitude from wake to Stage 1 alpha (F1,9=27.71, P< 0.01), Stage 1 alpha to Stage 1 theta (F1,9=25.78, P< 0.01) and Stage 1 theta to Stage 2 sleep (F1,9=12.82, P< 0.01), indicating a progressive change as sleep developed.

Figure 1.

. Grand mean waveforms during wake, Stage 1 alpha, Stage 2 theta and Stage 2 sleep recorded at F Z, C Z and P Z. N1, P2, N2/N300, P300/P450 and N550 peaks are shown. Waveforms are plotted 140 ms prior to and 1460 ms following the change in Pm (indicated by the vertical line in each panel). Voltage is plotted on the y-axis. Voltage (μV) and latency (ms) values are taken from grand mean evoked potential averages.

Table 1.  . Mean amplitude and latency values for the N1, P2, N2/N300 and P300/P450 components in each arousal state. Values are presented for midline sites ( F Z, FC Z, C Z , CP Z, P Z and O1). Standard deviations (SD) represent between subject variationThumbnail image of

Figure 2 illustrates the topographical distribution of each component in each sleep/wake state. The N1 component was largest fronto-centrally and centrally in wake, Stage 1 alpha and Stage 1 theta, becoming less clearly focused in Stage 2 sleep. There was a significant state-by-site interaction (F4,35=3.82, P< 0.05), although pair-wise ANOVAs revealed nonsignificant state-by-scalp-site interactions for wake/alpha (F4,33=0.97, P> 0.05), alpha/theta (F4,33=2.59, P> 0.05) and theta/Stage 2 (F4,38=2.05, P> 0.05). However, as inspection of Fig. 2 indicates, the overall significant state-by-site interaction was due to the loss of the N1 component in sleep rather than to a change in its topography.

Figure 2.

. Topographic maps of component peaks: N1 (latency at C Z), P2 (latency at C Z), N2/N300 (latency at C Z), P300/P450 (latency at P Z) and N550 (latency at F Z), during wake, Stage 1 alpha, Stage 1 theta and Stage 2 sleep. Voltage (μV) and latency (ms) values are taken from grand mean evoked potential averages. Each map represents a top-down view of the head, with the anterior of the map representing the front of the head. Voltage scales are different for each component but are consistent across arousal states within each component.

There was a significant main effect of state for latency (F3,27=6.76, P< 0.01). However, pair-wise ANOVAs also revealed no latency difference between wake/alpha (F1,9=0.91, P> 0.05), alpha/theta (F1,9=5.03, P> 0.05) or theta/Stage 2 (F1,9=1.60, P> 0.05). This revealed there was no evidence of a specific difference between alpha and theta ( Table 1). There was no main effect of site (F5,45=3.46, P> 0.05). However, there was a state-by-site interaction (F15,135=3.46, P< 0.001) indicating some variation in latency for this component at different sites across different states ( Table 1).

Thus for the N1, amplitude decreased markedly and progressively, while latency decreased only slightly.

P2 component

The P2 component was found to vary in amplitude, but not in topographical distribution or latency during sleep onset ( Table 1 and Fig. 1). There was a significant main effect of state for amplitude (F3,27=7.32, P< 0.01), with pair-wise ANOVAs revealing a larger P2 in wake compared with alpha (F1,9=25.07, P< 0.01), but a smaller P2 in theta compared with Stage 2 (F1,9=11.61, P< 0.01; Table 1). No amplitude difference was observed between Stage 1 alpha and theta (F1,9=0.19, P> 0.05).

The topographical distribution of the P2 revealed a fronto-central and central distribution in all states. However, the degree of localization varied between states as illustrated in Fig. 2. The distribution of the P2 was prominent in wakefulness, less clearly localized in Stage 1 alpha and Stage 1 theta, and prominent again in Stage 2 sleep. There was no significant state-by-scalp-site interaction (F4,37=1.83, P> 0.05). As such, pair-wise ANOVAs were not performed.

The latency of the P2 did not vary as a function of state as indicated by a lack of a significant main effect (F3,27=1.56, P> 0.05). There was also no main effect of site (F5,45=0.45, P> 0.05) or state-by-site interaction (F15 136=0.61, P> 0.05).


As shown in Fig. 1 and Table 1, the N2/N300 component increased in amplitude from wake to Stage 2 sleep. This was confirmed by a significant main effect of state (F3,27=61.30, P< 0.001). The change in amplitude from wake to alpha was very small and nonsignificant (F1,9=2.17, P> 0.05). However, there were large increases in amplitude from alpha to theta (F1,9=53.17, P< 0.001) and from theta to Stage 2 sleep (F1,9=63.31, P< 0.001).

As the amplitude of the late negative N2 component was very small in wake and Stage 1 alpha, no discernible topographical focus could be determined during these states ( Fig. 2). However, a prominent vertex distribution was observed during Stage 1 theta and Stage 2 sleep for the late negative N300. The appearance of this component was indicated by a significant state-by-site interaction across the four states (F4,35=7.76, P< 0.001). Pair-wise ANOVAs revealed no wake/alpha (F3,25=0.4, P> 0.05) or theta/Stage 2 (F4,27=1.21, P> 0.05) difference in scalp topography. However, the state-by-scalp-site interaction for alpha/theta (F4,27=5.78, P< 0.01) was significant. This indicated that during Stage 1 theta, a late negative N300 component appeared (that was not present during Stage 1 alpha) that was prominent at the vertex during both Stage 1 theta and Stage 2 sleep ( Fig. 2 and Table 1).

As illustrated in Figs 1 and 2, and Table 1, the latency of the N2/N300 component differed according to sleep/wake state. This was confirmed by a significant main effect of state for latency (F3,27=9.25, P< 0.001). This component appeared later in both Stage 2 sleep compared with theta (F1,9=18.49, P< 0.01) and in theta compared with alpha (F1,9=5.91, P< 0.05). No latency difference occurred between wake and alpha (F1,9=4.60, P> 0.05). The magnitude of the alpha–theta and theta–Stage 2 latency difference at Cz was 20 ms at the central site in each case.

A main effect of site for latency (F5,45=8.44, P< 0.05), but no state-by-site interaction (F15 135=0.91, P> 0.05) further revealed the N2/N350 varied in latency at different scalp locations.


The late positive P300/P450 appeared to decrease in amplitude and increase in latency from wake to Stage 2 sleep ( Fig. 1 and Table 1). The P300 showed a significant main effect of state for amplitude (F3,27=8.9, P< 0.001). Pair-wise ANOVAs revealed no significant difference in amplitude between wake and alpha (F1,9=0.06, P> 0.05). The P300 appeared to be smaller in amplitude with a shift from alpha to theta ( Fig. 1), although this was not found to be significant (F1,9=3.48, P> 0.05). There was a significant reduction in amplitude from theta to Stage 2 sleep (F1,9=8.01, P< 0.05).

A significant state-by-scalp-site interaction (F4,33=7.18, P< 0.001) for the P300/P450 suggested that scalp topography was altered between different sleep/wake states. The P300/P450 component had a maximal centro-parietal distribution in wake and Stage 1 alpha ( Fig. 2 and Table 1). This similarity in scalp distribution was confirmed statistically (F2,15=0.78, P> 0.05). The prominence of this component shifted parietally when moving from alpha to theta (F3,25=7.54, P< 0.001), and shifted occipitally when moving from theta to Stage 2 sleep (F4,34=14.05, P< 0.001; Fig. 2 and Table 1).

The latency of the P300/P450 component was also affected by sleep/wake state as indicated by a significant main effect of state (F3,27=55.29, P< 0.001; Fig. 2 and Table 1). Pair-wise ANOVAs revealed no wake/alpha difference in latency (F1,9=2.95, P> 0.05). However, this component appeared later in theta than in alpha (F1,9=50.81, P< 0.001) and later again in Stage 2 when compared with theta (F1,9=7.99, P< 0.05). The magnitude of the alpha–theta latency difference at Pz was 100 ms, with the theta–Stage 2 difference 29 ms at parietal sites.

There was no main effect of site (F5,45=1.80, P> 0.05) or state-by-site interaction (F15 135=1.30, P> 0.05) for the P300/P450.


An N550 component was also observed, appearing during Stage 2 sleep but not during wake or Stage 1 sleep ( Fig. 1). This component had a latency of ≈566 ms and a fronto-central prominence, with a mean amplitude of –47 μV at Fz.


The waking and Stage 2 RREPs in the present study were similar to that reported by Webster and Colrain (1998a), indicating that the sleep effect on RREPs is relatively robust.

The results of this study indicate that the N1 (and perhaps P2) are sensitive to any change in EEG-defined arousal and are not specifically related to alpha vs. theta EEG state. However, alpha/theta state had a marked effect on later waveforms. A large negativity peaking at ≈300 ms (N300), which was not prominent in wakefulness or Stage 1 alpha, developed in Stage 1 theta. Furthermore, a late positivity, which peaked at ≈300 ms (P300) in wakefulness and Stage 1 alpha, disappeared in Stage 1 theta and Stage 2 sleep, with a later and more posterior P450 emerging.

Previous EP research has indicated that Stage 1 represents an intermediate stage between wake and Stage 2 sleep ( Ogilvie et al. 1991; Winter et al. 1995 ; Webster and Colrain 1998a). It should be noted that this previous research did not divide Stage 1 into alpha and theta EEG activity and therefore Stage 1 in these studies reflected a combination of trials conducted during both alpha and theta EEG activity. The present study thus provides evidence of a state-dependent difference in later EP waveforms between alpha and theta states.


The N1 was attenuated in amplitude by ≈50% from wakefulness to Stage 1 alpha. As both wakefulness and Stage 1 alpha, as defined in this study, were characterized by relatively continuous EEG alpha activity, the N1 does not appear to be closely related to changes in the EEG. An alternative possibility is that N1 is related to changes in attention or awareness. Other sleep-onset studies employing auditory stimuli have reported large reductions in N1 amplitude from wakefulness to Stage 1 sleep ( Ogilvie et al. 1991; de Lugt et al. 1996 ). Campbell et al. (1992) interpreted this decrease in negativity to be a result of decreasing attentional ‘resources’. The N1 (or an overlapping slow negative wave) can be altered greatly by manipulation of the subjects level of attention ( Naatanen et al. 1992 ). Both Harsh et al. (1994) and Winter et al. (1995) have noted that during the sleep-onset period, N1 is larger when subjects are awake and attentive than when they are awake but inattentive. The N1 was near baseline level during Stage 2 in the present study. This replicates other studies employing auditory stimuli ( Harsh et al. 1994; Niiyama et al. 1994; Loewy et al. 1996 ). Thus the gradual decrease in the respiratory N1 from wakefulness to Stage 1 and finally to Stage 2 provides strong evidence for cross-modality consistency.

A small but significant decrease in N1 amplitude was observed during the transition from alpha to theta. This reduction in amplitude may indicate that the N1 shows some alpha/theta-state-dependent shift that is independent of the earlier attention related reduction from wake to Stage 1 alpha. Alternatively, the alpha to theta reduction in amplitude may also be interpreted as a further reduction in the level of awareness. The observation that the alpha to theta change was less than from waking to Stage 1 alpha or from Stage 1 theta to Stage 2 sleep is perhaps more consistent with the latter interpretation.

AEP studies have determined that the N1 is affected by selective attention, and is susceptible to both the physical characteristics of the stimulus and other nonspecific influences associated with the general state of the individual, such as arousal changes associated with sleep/wake state and predictability factors ( Naatanen and Picton 1987). As such, the N1 has both exogenous and endogenous characteristics. Naatanen (1990) suggests that the N1 may act as a transient detector system that triggers an internal attention system and may subserve conscious perception of a stimulus, irrespective of stimulus modality. If the N1 does indeed play such a role, then the large reduction in N1 seen from wakefulness to Stage 1 alpha, compared with the smaller state-dependent or further awareness related reduction seen from Stage 1 alpha to theta, indicates that attention deteriorates rapidly with the onset of drowsiness.


The results for P2 are more difficult to interpret. The P2 in wakefulness was smaller in amplitude than during Stage 2. Other studies have also reported this change. AEP studies have generally observed the amplitude of the P2 during Stage 1 and 2 to be similar, with both being larger than the wake P2 ( Winter et al. 1995; de Lugt et al. 1996 ). However, Ogilvie et al. (1991) found the P2 to gradually increase in amplitude as sleep onset developed. During Stage 2 sleep, the P2 amplitude more than doubled in size when compared with sleep onset. Furthermore, during Stage 1 and 2 sleep, the P2 has been shown to discriminate between both deviant and standard tones ( Winter et al. 1995 ). Campbell et al. (1992) interpreted the decrease in N1 and increase in P2 during definitive sleep as being due to the removal of a slow negative wave that overlaps and summates to N1 and P2 in the waking state.

The Stage 1 alpha and theta peaks of this component were both smaller in amplitude than either wakefulness or Stage 2. Although Fig. 2 illustrates the P2 to be slightly larger in Stage 1 theta compared with Stage 1 alpha, this difference was not significant. Furthermore, no latency differences were observed across arousal states. Webster and Colrain (1998a) also reported the respiratory P2 to decrease in amplitude from wake to Stage 1 sleep, but did not report a further increase in amplitude during Stage 2 sleep. The lack of consistency in the amplitude of P2 during the sleep-onset period has been reported in reviews by Broughton (1989) and Campbell et al. (1992) . They noted that different studies report no change, increases or decreases in P2 amplitude between wake and Stage 2 sleep.


A very small N2 was just discernible in wake and was only slightly larger in Stage 1 alpha. This component appeared to change dramatically in appearance during theta as compared with the alpha state. It showed a large increase in amplitude, a longer latency and a prominent vertex scalp distribution during Stage 1 theta. There was a further increase in amplitude and latency during Stage 2 sleep. These results suggest that with the appearance of Stage 1 theta, an N300 component emerges that is not present during wake and Stage 1 alpha. The presence of the N300 during the onset of sleep confirms previous RREP and AEP studies. AEP research indicates that the emergence of the N300 is related to a slowing and eventual failure of behavioural responsiveness. This may occur relatively early during Stage 1 sleep ( Ogilvie et al. 1991; Harsh et al. 1994 ). The results of the present study would suggest that the appearance of the N300 during Stage 1 sleep in previous AEP studies was predominantly due to the presence of theta activity in the prestimulus EEG in a proportion of the responses.

The emergence of a large N300 component with a vertex-focused topographic distribution is indicative of the activation of one or more intracranial generators. The lack of any difference in topographic distribution between Stage 1 theta and Stage 2 sleep is consistent with the same generators being active in both states ( Picton 1995). The increase in amplitude from Stage 1 theta to Stage 2 is thus best interpreted as an increase in generator strength.

The results of the present study assist in resolving the issue as to whether the N300 is a sleep-specific component or a late N2. It has been argued previously that the auditory N2 in drowsiness may reflect an attenuated waking N2 ( Van Sweden et al. 1994 ) and that it is part of an information processing system that is comparable with the N2 in the waking state ( Salisbury and Squires 1993). However, Ogilvie et al. (1991) could not readily detect an N2 in the waking state but noted an increase in amplitude at the onset of behaviourally defined sleep. These authors suggest that the sleep N2 could be used as an index of the presence of sleep. The present study supports this contention. A prominent N300 was seen in Stage 1 theta (sleep as defined by EEG criteria) but not in Stage 1 alpha, indicating that this component peak is distinctly a sleep phenomenon.

The data also address the issue of whether the N300 in averaged waveform is due to the presence of K-complexes, and thus should be viewed with the N550 as a K-complex-related waveform. The clear presence of N300 in the context of sleep-like EEG prior to the appearance of K-complexes indicates that N300 should be viewed as a separate entity. Harsh et al. (1994) suggested that the N300 may relate to vertex sharp waves. The present data are not inconsistent with this interpretation, as vertex sharp waves are known to appear in the context of theta within Stage 1. Further evaluation of this hypothesis is clearly warranted.


AEP studies have demonstrated P300 to be a memory comparison process ( Johnson 1993). This component has been shown to be differentially responsive, being larger to stimuli that are task relevant and occur relatively infrequently. The P300 has been reported in single stimulus auditory paradigms ( Polich et al. 1994; Polich and Margala 1997 ). Harver et al. (1995) have argued that the occlusion of random breaths in fact represents an oddball task where unoccluded breaths represent standard, and occluded breaths the ‘oddball’ stimuli. The presence of the P300 in single stimulus RREP studies ( Webster et al. 1997, 1998b; Colrain et al. 1998 ) and the similarity of its scalp distribution to the auditory P300 ( Colrain et al. 1996 ) both suggest that brief occlusions of the airway are processed in a similar manner to both the detection of rare auditory targets and single stimulus auditory paradigms. This is clearly sensible given models of P300 amplitude ( Johnson 1993) that, for example, emphasize some aspect of stimulus relevance as being important in the generation of P300. Presumably an indication that the upper airway has become blocked has a high level of intrinsic biological relevance.

The P300 in the present study appeared functionally similar during the waking state and Stage 1 alpha, but was dramatically altered in terms of latency and topography during the transition from Stage 1 alpha to Stage 1 theta. A longer latency positive wave was observed, peaking at about 420 ms in Stage 1 theta and 450 ms in Stage 2 sleep. In addition, the scalp topography of the P450 was more posterior (i.e. maximum over occipital regions) than for the P300. This P300/P450 has previously been found to decrease in amplitude and increase in latency from wakefulness to Stage 2 sleep when elicited by respiratory occlusions ( Webster and Colrain 1998a).

Interpretation of the change in P450 from Stage 1 theta to Stage 2 is complicated by the presence of the K-complex-related N550 component in Stage 2. The reduction in P450 amplitude may be due to the summation of the positive output of the P450 generator and the negative output of the N550 generator. Likewise the occipital shift in topographic focus may be due to the summation resulting in displacement of the P450 by the larger fronto-central N550 negativity. This is clearly a more parsimonious explanation than the alternative hypothesis of separate Stage 1 theta and Stage 2 P450 generators resulting in different scalp distributions.

The results of the present study, with respect to the P300/P450, are generally consistent with previous AEP studies that have attempted to distinguish arousal states within sleep onset. Harsh et al. (1994) observed the P300 in response to targets increased in latency from wake to Stage 1 A (defined as the loss of alpha rhythm just prior to traditional Stage 1 sleep). Furthermore, the P300 decreased in amplitude from Stage 1 A to traditional Stage 1 and disappeared in Stage 2 sleep. Faster reaction times were associated with a short latency and large amplitude P300, whereas slower reaction times were associated with a delayed and attenuated P300. When no behavioural response was given, no P300 was visible. Ogilvie et al. (1991) found similar results across the sleep-onset period.

It has been argued that the sleep P450 may simply be a delayed P300 ( Nielson-Bohlman et al. 1991; Salisbury et al. 1992; Harsh et al. 1994; Winter et al. 1995 ). These studies showed the P450 to be differentially sensitive to deviant stimuli with a somewhat similar scalp distribution to the waking P300 (although occipital scalp sites were not recorded). However, the presence of a P450 to frequently occurring standard tones (unlike the P300), would argue against this.


The N550 component was clearly present during Stage 2 sleep, but not during other states, and was most prominent fronto-centrally. This is consistent with RREP ( Webster and Colrain 1998a) and AEP studies ( Harsh et al. 1994; Niiyama et al. 1994 ). The specific appearance of the N550 during Stage 2 sleep is also consistent with other studies that have identified a relationship between K-complexes, which are specific to Stage 2, 3 and 4 sleep, and the N550 ( Campbell et al. 1990; Bastien and Campbell 1992, 1994; Harsh et al. 1994; Niiyama et al. 1994, 1995; Sallinen et al. 1994 ). These studies have reported that the N550 is greatly diminished when K-complexes are excluded from the responses to be averaged. The absence of an N550 in Stage 1 theta indicates this component is not dependent on alpha/theta state. However, the conditions necessary for its appearance cannot be determined by the current experimental paradigm.

From a methodological perspective, the slight sound associated with use of the gas-activated valve raises the issue of whether the responses reflect processing of respiratory or auditory information. There are a number of pieces of evidence to indicate that the occlusion stimulus activates respiratory somatosensory pathways, and that it is the somatosensory rather than auditory activation that contains the salient aspects of the stimulus. First, source localization data for the early RREP components, obtained using the same respiratory apparatus and occlusion type as that used in the current study, show dipole generators in somatomotor regions rather than in auditory cortex ( Logie et al. 1998 ). Secondly, the P300 during wakefulness is larger in response to higher levels of inspiratory resistive loading than to loading levels around sensory threshold ( Bloch-Salisbury and Harver 1994; Webster and Colrain 1998b). This occurred despite the fact that the small noise produced by valve closure was identical for all resistive load sizes. Thirdly, Colrain et al. (1996) reported the wake P300 to AEP target stimuli (374 ms) to be significantly longer in latency than the P300 to respiratory occlusion stimuli (284 ms), for the same subjects. This P300 RREP latency is very similar to that of the present study (296 ms), with identical occlusion stimuli used in both studies. Fourthly, the latency of the Stage 2 sleep N300 in response to an auditory targets (333 ms) is also longer than the N300 in response to occlusion stimuli (284 ms) ( Colrain et al. 1996 ). Again, the N300 RREP latency here is very similar to that reported in the present study (274 ms).

Latency values for all RREP components in response to the occlusion stimulus in all of the above-mentioned studies have been taken from the onset of Pm change. In order to set the valve closure sound as the possible eliciting stimulus, 24 ms must be added to the RREP latency values given above. This 24 ms takes into consideration both the valve closure time and the time for the sound of valve closure to travel down the 3 m length of tubing to reach the subject. Even when adjusting the RREP wake P300 and sleep N300 latencies given above, the P300 in response to auditory targets is still 66 ms longer than for occlusion stimuli and the N300 is still 25 ms longer for auditory than occlusion stimuli. We would therefore argue that the eliciting stimulus in this study is in fact the occlusion stimulus and not the small auditory sound produced by valve closure.


The data suggest that different RREP components show different relationships to arousal state. Consistent with respiratory activity, there are significant changes in certain RREP components between alpha and theta EEG states within the sleep-onset period. The N2/N300 and P300/P450 appear to follow the state-dependent model. They are both similar in appearance during wake and alpha, but change markedly in latency, amplitude or topographical distribution when moving from Stage 1 alpha to Stage 1 theta. This indicates that different components emerge during sleep in both cases. Based on its topographical consistency and reduction in amplitude, the N1 appears to represent some form of attention-related component that is not exclusively associated with EEG alpha to theta state changes within Stage 1. The P2 has proven difficult to interpret, while the N550 is not dependent on alpha/theta state.

As mentioned earlier, it has been argued previously that during the period of drowsiness between wake and Stage 2 sleep there is a gradual shift from wake- to sleep-type auditory ( Winter et al. 1995 ) and respiratory ( Webster and Colrain 1998a) information processing. The results of the present study indicate that the morphology of two late RREP components change dramatically at the alpha-to-theta transition that occurs in Stage 1 sleep. Furthermore, the physiological data are consistent with a marked change in awareness. The transition from Stage 1 alpha to theta appears to be the threshold point from a state of awareness of the external environment (reflected by the presence of P300) to one of unawareness (reflected by the absence of P300). The present data also indicate that only some components reflect the alpha/theta-state-dependent model of thalamic polarization in EP elicitation, while others appear to reflect different processes.


The authors would like to thank Dr Kenneth Campbell for his assistance in the preparation of this manuscript.

Accepted in revised form 20 November 1998; received 15 June 1998