The N550 component of the evoked K-complex: A modality non-specific response?


Dr Ian M Colrain, Department of Psychology, The University of Melbourne, Parkville, Victoria 3052, Australia. Tel: 613 93446350; Fax: 613 93476618; E-mail: colrain@psych.


A large amplitude late negative deflection peaking between 500 and 650 ms is observable in the averaged K-complex wave-form. This peak is thus often labelled the N550. ‘N550’ appears during stage 2 and is maintained into slow wave sleep but is not apparent during REM. Most studies have employed auditory stimuli to elicit the K-complex. Two experiments were run to examine the effects of stimulus modality on the topographical distribution of the N550. In the first experiment, the K-complexes were elicited in an auditory oddball procedure. In the second experiment, K-complexes were elicited by respiratory occlusions. Twenty-nine channel recordings were used to increase spatial resolution. N550 was substantially larger in the average of trials containing K-complexes than in trials in which a K-complex could not be identified. N550 varied inversely in amplitude with the probability of accordance of the stimulus. The topographic distribution of the N550 was consistent between experiments. It was bilaterally symmetrical and was maximal over fronto-central regions of the scalp. The results indicate that the N550 reflects the activity of a modality non-specific, sleep dependent generator that responds to both interoceptive and external stimulation.


During stages 2, 3 and 4 NREM sleep, infrequently presented auditory stimuli will often elicit a large amplitude wave-form, the K-complex. Experiments investigating averaged auditory evoked responses during NREM sleep have often identified two long latency negative wave-forms peaking at approximately 300 and 550 ms and a positive wave-form peaking at approximately 900 ms. They are thus labelled ‘N300’ (or N350 in some studies); ‘N550’; and P900. The N550 is usually reported as being substantially larger than N300, and a much more well defined peak than the P900.

The N550 can easily be elicited in NREM sleep, but does not occur in REM sleep (Campbell et al. 1992). The sensitivity of N550 to variations in stimulus properties is unclear. Bastien and Campbell (1992) have reported invariant amplitudes across a range of stimulus intensities, while Salisbury and Squires (1993) reported that its amplitude varies with the magnitude of the pitch difference between target and standard stimuli. Niiyama et al. (1995) reported N550 as being larger to rare than to frequent stimuli, and larger when subjects were asked to detect targets during wakefulness than when no instructions were given. The N550 is clearly related to K-complexes as several studies have indicated that it is absent or greatly diminished in averages of trials in which K-complexes cannot be elicited. (Campbell et al. 1990; Bastien and Campbell 1992; Bastien and Campbell 1994; Harsh et al. 1994; Niiyama et al. 1994; Sallinen et al. 1994; Niiyama et al. 1995).

The majority of studies have employed auditory stimuli to elicit the K-complex. Recently Webster and Colrain (1998) indicated that the K-complex could also be elicited by internal, non-auditory stimuli. They employed brief inspiratory occlusions to elicit the K-complex. They did not however, compare its morphology to that of the auditory elicited K-complex.

There has been little systematic study of the scalp distribution of the N550 wave. While some studies have indicated that it is maximal over frontal areas of the scalp (Niiyama et al. 1995), others have indicated that it is centrally distributed. Differences in studies may be due to the limited number of scalp electrodes (typically from 1 to 3) that are used to record the EEG. Niiyama et al. (1995) used a larger 21-channel array. They reported that the large amplitude negativity of the auditory-evoked K-complex was maximal over fronto-central areas of the scalp. However, as they averaged all data points extending from 300 to 900 ms, the contribution of the actual N550 peak to this fronto-central distribution could not be determined. In the present study, spatial resolution of the EEG was increased by recording from a 29-channel array. Two experiments were run. In the first experiment, an auditory oddball task was employed to compare the effects of frequent and rare stimuli on the N550 component of the K-complex. In the second experiment, the effect of stimulus modality (auditory vs. respiratory somatosensory) on the N550 component was examined. In both studies, single trial K-complexes were averaged. Scalp distribution maps of the peak of the N550 were then constructed.


General procedure

Multi-channel EEG and polysomnography recording

EEG was recorded from 29 equi-distant scalp sites referenced to linked ears using an ECI electrocap, Neuroscan Synamps, and Neuroscan Scan software. The sites were FP1, FP2, F7, F3, Fz, F4, F8, FT7, FC3, FCz, FC4, FT8, T3, C3, Cz, C4, T4, TP7, CP3, CPz, CP4, TP8, T5, P3, Pz, P4, T6, O1 and O2. (Note that in some electrode labelling conventions, FT7/FT8, T3/T4, TP7/TP8, and T5/T6 are referred to as FC7/FC8, T7/T8, CP7/CP8 and P7/P8, respectively). A vertical electro-oculography (EOG) was recorded from supra- and infra-orbital ridges of the left eye. An electrode placed between Fz and FCz served as the earth. Inter-electrode impedance was maintained below 5 kOhms. The analogue filter bandpass was 0.01–500 Hz. A 16-bit A/D converter sampled the physiological signals beginning 100 ms prior to stimulus onset and continuing for another 1000 ms following it (i.e. 900 ms post stimulus). The sampling rate was 1000 Hz per channel. The single trials were stored on hard disk for later off-line analysis. The C3-A2 EEG, a horizontal EOG (recorded from the outer canthi of each eye) and a submentalis electromyography (EMG) signal were continuously recorded by a Compumedics automated sleep acquisition system, with a sampling rate of 500 Hz. These latter data were used to ascertain sleep stage using the criteria of Rechtschaffen and Kales (1968).

The EEG was recorded during the first two hours of night sleep. Within periods classified as being stage 2 NREM sleep, the response to each stimulus was categorised as to whether or not it included a K-complex. A K-complex was defined using a modification of the Rechtschaffen and Kales criteria applied at the C3 electrode site. That is, ‘a waveform having a well delineated negative sharp wave which is immediately followed by a positive component. The total duration of the complex should exceed 0.5 s.’ (Rechtschaffen and Kales 1968, page 6). An amplitude criterion was also implemented defining that the negative component had to be at least 75 μV at C3. Trials associated with such responses were labelled as ‘KC+’.

A K-complex absent category (‘KC−’) was also defined to include trials in which no visually discernible response to the stimulus was present. In addition, trials in which some other slow or fast wave activity was present, but without a K-complex, were classified as KC− trials. Since recordings were made during the early part of sleep, some vertex sharp waves were observed. These were excluded from the present analysis.

Scalp distribution maps

The peak of the N550 was identified in the grand-averaged KC+ trials (the average of each subject's averaged KC+ trials). It was identified at Fz as the maximum negative peak in the 450–650 ms range. Its amplitude was measured at all other electrode sites at this latency. A scalp distribution map was then computed. The maps were made continuous by using a four-nearest neighbour interpolation in the regions between the actual electrode sites. The map was scaled using 16 different colours.



Eight young males aged between 20 and 25 years were used in the study. All were free from auditory or sleep pathology and were healthy non-smokers.

AEP protocol and experimental procedure

Stimuli consisted of 50 dB SPL(A) 500 and 1000 Hz pure tones, filtered using a cosine envelope. Stimulus duration was 250 ms. They were presented with a constant interstimulus interval of 5 s through a single loudspeaker, situated 30 cm above the subject's head. The intensity of the sounds was calibrated with a Dawes Instruments Ltd type 1419 sound level meter placed at the approximate location of the subjects head.

Each subject spent two consecutive nights in the laboratory. On one night, the 1000 Hz tone was presented with a probability of 20% (rare) and the 500 Hz tone with a probability of 80% (common). On the other night the probabilities were reversed. For half the subjects, the 500 Hz tone served as the rare stimulus on the first night while for the other half it was the 1000 Hz tone. In each case, the rare tone was designated as a target that the subject was to count during the wake period preceding the sleep recordings. Preliminary analysis indicated that there were no significant effects of tonal pitch on the probability of eliciting a K-complex or the amplitude of N550. The data were therefore combined across both nights for all subsequent analyses.

Statistical analysis

N550 amplitude values were subjected to a three-way repeated measures analysis of variance. The three factors were stimulus type (frequent and rare), response type (KC+ and KC−), and electrode site. Homogeneity of variance was assessed for the electrode site factor. Huynh–Feldt Epsilon values were determined, and used to correct the degrees of freedom.

The topographic distributions of N550 for the K-complex averages for each stimulus type were compared using two-way repeated-measures analysis of variance. Comparison of different scalp distributions is methodologicaly complex. This is because the usual ANOVA procedure assumes that experimental effects are additive. It is assumed that the experimental effect results in a constant being added to (or subtracted from) a baseline condition. With EEG data, experimental effects usually result in a constant being multiplied to the scalp distribution data. McCarthy and Wood (1985) have developed a normalisation procedure to correct for these non-additive effects. Amplitude values were therefore scaled within each response type using the McCarthy & Wood normalization method. Differences between distributions are indicated by significant interaction terms between the site and stimulus type factors.



Seven young males aged between 19 and 22 years were used in the study. All were free from respiratory or sleep pathology and were healthy non-smokers.

RREP protocol

A face mask (Hans Rudolf series 7940), connected to a nonrebreathing valve (Hans Rudolf series 2600), supported by a suspending cable, was positioned so that the subject could comfortably and quietly respire with minimal facial muscle activity. The inspiratory port of the nonrebreathing valve was connected to an occlusion valve (Hans Rudolf series 2100). Airflow was measured for the inspiratory circuit using a Fleisch Pneumotacograph connected to a Validyne differential pressure transducer (model CD15). Mouth pressure was sampled at a port in the face mask, and measured using a second a Validyne differential pressure transducer.

Inspiration was interrupted for 250 ms by manual activation of the occlusion valve after the onset of the inspiration as indicated by the airflow signal. The interruption was in the form of a total occlusion that occurred randomly every second to fifth breath. All evoked potential latencies were adjusted to be expressed relative to the start of the change in mouth pressure. This occurred 40 ms after valve closure, and thus after the triggering of each data collection trial. The delay is due to the elasticity of the air inside the 3 m of tubing between the occlusion valve and the face mask. Such an adjustment is standard in the respiratory evoked potential literature (Davenport et al. 1986; Davenport et al. 1996; Logie et al. 1998; Webster and Colrain 1998).

Statistical analysis

N550 values were subjected to two-way repeated-measures analysis of variance, the two factors being response type (KC+ and KC−) and electrode site. As with experiment 1, homogeneity of variance was assessed for the electrode site factor, and Huynh–Feldt Epsilon values were determined, and used to correct the degrees of freedom.


Experiment 1

K-complexes were identified on 37% of the trials when a rare auditory stimulus was presented and on 17% of the trials when a frequent stimulus was presented. Figure 1 presents the averaged wave-forms at all scalp sites for rare stimuli on both KC+ and KC− trials. Figure 2 displays the data at Fz for both KC+ and KC− trials to both rare and common stimuli. As can be observed, the KC+ wave-forms contained a small amplitude N300 and a much larger and prominent N550 component. On KC− trials, the N300 was again elicited, although its amplitude was smaller than on KC+ trials. The N550 was difficult to identify.

Figure 1.

Averaged wave-forms of all trials for all subjects (i.e. grand averages) for KC+ and KC− responses following rare auditory stimuli. The darker wave-forms represent KC+ and the lighter wave-forms the KC−. Negative polarity in this and all other figures is represented by an upward deflection.

Figure 2.

Fz averaged wave-forms of all trials for all subjects for KC+ and KC− responses to auditory stimuli. Darker wave-forms represent the average of KC+ trials and lighter wave-forms KC− trials. Thicker lines represent rare responses; thinner lines represent common responses.

On KC+ trials, N550 peaked at 644 and 611 ms following the rare and frequent stimuli, respectively. The amplitude of N550 at Fz for individual subjects is presented in Table 1. Analysis of variance revealed significant effects of stimulus type (F1,7=44.1, P < 0.001); response type (F1,7=375.2, P < 0.001); and electrode site [F28, 196=23.1, P < 0.01, ε=0.109]. These indicated that the N550 was larger to the rare stimuli, in the KC+ trials, and at Fz.

Table 1.  N550 amplitude values (in μV) for each subject. KC+ and KC− data are presented separately for averages of rare and common tone responses Thumbnail image of

The scalp topography of N550 in the KC+ trials was compared between the rare and common trials using a two-way ANOVA, using scaled data. The interaction term was not significant (F28, 196=2.8, P > 0.01, =0.137), indicating no differences in the overall distribution of the component across the scalp between the common and rare stimuli. The topo- graphic representations of the KC+N550 to the average of all auditory stimuli are shown in Figure 3.

Figure 3.

Topographic distribution maps of the maxima of N550 in the average of KC+ trials to all auditory trials and to respiratory occlusion stimuli. Note that the voltage scales in the two maps vary in order to highlight the spatial distribution of each component.

Experiment 2

K-complexes were elicited on 28% of trials in which respiration was occluded. Figure 4 presents the averaged wave-forms at all sites following occlusion trials in which a K-complex was elicited (KC+ trials) and in which it was not elicited (KC− trials). Figure 5 provides a zoom of the Fz data. Table 2 presents the N550 amplitude on KC+ and KC− trials for individual subjects at Fz. There was a significant effect of electrode site [F28, 140=22.1, P < 0.01, =0.172], with the largest values occurring at Fz in all cases. As can be observed, N550 was significantly larger on the KC+ averaged trials (−88.1 μV at Fz), than on the KC− trials (−18.1 μV at Fz) (F1,5=47.1, P < 0.001). The peak latency of the KC+N550 to respiratory stimuli was slightly earlier (604 ms) than for auditory stimuli.

Figure 4.

Grand averaged wave-forms at all sites for responses to respiratory stimuli. The darker wave-form represents KC+ and the lighter wave-form KC− averages.

Figure 5.

Fz grand averaged to respiratory stimuli. The darker wave-forms represent the average of KC+ trials and the lighter wave-forms KC− trials.

Table 2.  N550 amplitude values (in μV) for each subject for responses to respiratory occlusions. KC+ and KC− data are presented separately Thumbnail image of

The scalp distribution of N550 for the KC+ trials was compared to that of the auditory target KC+ trials using a two factor mixed model ANOVA. The within subject factor being electrode site and the between subjects factor being stimulus modality. The interaction term (based on scaled data) was also not significant [F28, 336=2.1, P > 0.05, =0.152], indicating that the topographic distributions were not significantly different as a function of stimulus modality.


The data illustrate that the N550 component of the averaged K-complex has the same scalp topography regardless of whether it is elicited by externally produced auditory stimuli or internally produced respiratory stimuli. Furthermore, the amplitude of the response seems to relate to the probability of stimulus occurrence, with rarer stimuli producing greater N550 amplitude values than more common stimuli.

The probability of eliciting a K-complex was slightly lower than in the Cotéet al. (submitted) and Bastien and Campbell (1992) articles. This is probably because auditory stimuli were presented at a lower intensity level in the present study. Bastien and Campbell (1992) have noted that the probability of eliciting a K-complex varies directly with the intensity of the auditory stimulus. Rare auditory stimuli elicited more K-complexes than the frequently occurring standards. This is in agreement with a number of other studies (Salisbury and Squires 1993; Sallinen et al. 1994; Niiyama et al. 1995).

The results also confirm the previously described relationship between K-complexes and the N550 component of the auditory evoked potential (Campbell et al. 1990; Bastien and Campbell 1992; Bastien and Campbell 1994; Harsh et al. 1994; Niiyama et al. 1994; Sallinen et al. 1994; Niiyama et al. 1995; Cotéet al. submitted). Furthermore, the similarity of the N550 wave-form to auditory and respiratory somatosensory stimuli following the averaging of K-complexes, confirm the K-complex as a modality-independent response, with a topographic distribution very similar to that reported by Cotéet al. (submitted). Their use of a nose reference compared to the linked ear reference of the present study had little effect on this overall distribution.

The present study employed an equi-distant electrode array. Cotéet al. employed a more widely distributed but unequal placement to include inferior frontal and parietal sites. They did note an inversion of N550 at sites inferior to the Sylvian fissure (i.e. the mastoids and inferior parietal regions). Nevertheless, the fact that their array included holes in the electrode array does not appear to have affected the overall fronto-central distribution of N550.

Importantly, in this and other studies (Niiyama et al. 1995; Cotéet al. submitted) in which multiple channel recordings have been employed, the N550 component of the K-complex has been consistently observed to have a distinct frontal maximum distribution. This finding needs to be emphasised due to the persistence in the literature of the notion that K-complexes are maximally recorded at the vertex. This statement appears in texts (Carskadon and Rechtshaffen 1994, p. 950) guides for sleep scoring (Rechtschaffen and Kales 1968, p. 6) and even in the most recently published guidelines for the interpretation of phasic events during sleep (Terzano, Parrino and Mennuni 1997, p. 57).

It could be argued that the above statements are not refuted by topographic data that only investigate one of the components that have traditionally been identified as being part of the K-complex. What then of the N300 and P900 components? The N300 component is not uniquely seen within the context of the presence of K-complexes, and had been reported in the average of KC− trials in a number of studies (Campbell et al. 1990; Bastien and Campbell 1992; Bastien and Campbell 1994; Harsh et al. 1994; Niiyama et al. 1994; Sallinen et al. 1994; Niiyama et al. 1995). It is also not visible in the raw EEG when it is part of a K-complex and thus cannot contribute to the judgement of scalp topography. Analysis of the N300 is the subject of a separate paper currently submitted. The time window used for data collection in the present study did not permit analysis of P900. However, Cotéet al. (submitted) report that it also displays a frontal maximum.

The consistency in the response between the external auditory and internal respiratory stimuli is clearly relevant to the discussion of the comparability of evoked and ‘spontaneous’ K-complexes. This is based on the suggestion that ‘spontaneous’ complexes are in fact elicited by stimuli ‘inside’ the body (Niiyama et al. 1996). External auditory stimuli have classically been used to elicit K-complexes. On the other hand, while the exact location of the afferent source of the respiratory evoked potential is not known (Davenport et al. 1996), it is likely to relate to the activity of muscle spindle receptors in the diaphragm (Davenport et al. 1985) or intercostal muscles (Davenport et al. 1993). Early components of the respiratory evoked potential that are known to persist into NREM sleep (Webster and Colrain 1998) have been localised to generators in the primary somatosensory and supplementary motor cortices (Logie et al. 1998). Thus, while the inspiratory airway is occluded outside of the subject, the ‘stimulus’ is effectively internal. The response to the inspiratory occlusion reflects some interoceptive process. Arguably this could be viewed as similar to airway obstruction, pain or any other internally produced afferent signal to the cortex that might lead to apparently spontaneous K-complex activity (see Niiyama et al. 1996 for a further discussion of apparently spontaneous K-complexes).

The judgement of modality non-specificity is clearly made largely on the basis of the similarity in the topographic distributions rather than using amplitude latency or general morphology. This is despite these other variables being very similar in the auditory and respiratory responses. Amplitude and latency do however, vary to a small extent within the auditory modality as a function of stimulus probability. Thus, differences in amplitude or latency of responses to an auditory and a respiratory stimulus would be impossible to interpret as being evidence for modality specificity of the response. The utility of topographic distribution in this context is that it reflects the state of underlying generator mechanisms. It is generally accepted that two components with the same scalp distribution are more likely to have been produced by the same underlying neural tissue (although it is technically possible for different generators to produce the same scalp distribution). Indeed, Naatanen and Picton (1987) maintain that in order for different components of the evoked potential to be considered to be independent, they must have different intracranial generators.

The similarity in topographic distribution between the different modalities is particularly impressive given the lack of total overlap in the subjects used in the two experiments. Five subjects did indeed complete both experiments. Due to logistical difficulties, it was necessary to recruit another three subjects for experiment 1 and an additional two for experiment 2. Inspection of the data from the five subjects common to both experiments revealed the same topographic distributions as those seen when using all of the recorded data.

Confirmation that stimulus probability is inversely related to N550 amplitude is provided by the comparison of auditory and respiratory responses. Approximately two rare auditory, three to four occlusion, and eight frequent auditory stimuli were presented per minute. The amplitude of the N550 is largest to rare auditory and smallest to frequent auditory stimuli, with the occlusion response intermediate between the two. These results are in agreement with those of Bastien and Campbell (1994), who indicated that tone pips presented every 30 s produced larger averaged K-complex responses than those presented every 10 s or every 5 s. They thus claimed that the refractory period for the K-complex was very long (between 10 s and 30 s). In the present study, the rare auditory stimuli were presented on average every 30 s and the frequent stimulus was presented on average every 7.5 s. Presentation of the frequent stimulus thus falls within the refractory period of the K-complex, and the N550 was larger to the rare than to the frequent auditory stimulus. The respiratory occlusions occurred on average every 15–20 s (depending on the respiratory rate of individual subjects) and the amplitude of the N550 was between that of the rare and frequent auditory stimuli. This argument relies on the refractory periods to the rare and frequent stimuli displaying frequency specificity. That is, that different populations of neurones are activated by the two stimuli (Butler 1968, 1972; Picton et al. 1978; Naatanen et al. 1988). This view is consistent with the data presented by Salisbury and Squires (1993) who reported that N550 amplitude varied with the magnitude of the pitch difference between target and standard stimuli.

The possibility obviously remains that the reason for the rare responses having a larger N550 is due to them having been given a ‘target’ status during wakefulness (Niiyama et al. 1995). Recent work from our laboratory has indicated however, that the rarity of a stimulus plays more of a role in enhancing N550 amplitude than its relevance based on presleep instructions (Colrain, Cardamone and Gora 1998).

The functional significance of K-complexes remains unknown. They are viewed by some as arousal responses associated with an inhibition of sleep spindles and a physically improved thalamo-cortical sensory inflow (Halasz 1993) and by others as a sleep maintenance response (Walter 1953; Wauquier 1995). The fact that two very different types of stimuli produce an evoked K-complex response with a very similar morphology and scalp distribution would suggest that whatever their function, evoked K-complexes can represent a modality-independent, sleep-specific response to stimulation.

Accepted in revised form 10 May 1999; received 10 August 1998