| Article Access Statistics|
| Viewed||5333 |
| Printed||293 |
| Emailed||7 |
| PDF Downloaded||50 |
| Comments ||[Add] |
| Cited by others ||9 |
|Year : 2010
: 12 | Issue : 47 | Page
|Event-related potentials as a measure of sleep disturbance: A tutorial review
School of Psychology, University of Ottawa, Ottawa, ON, K1N 6N5, Canada
Click here for correspondence address
|Date of Web Publication||14-May-2010|
This article reviews event-related potentials (ERPs) the minute responses of the human brain that are elicited by external auditory stimuli and how the ERPs can be used to measure sleep disturbance. ERPs consist of a series of negative- and positive-going components. A negative component peaking at about 100 ms, N1, is thought to reflect the outcome of a transient detector system, activated by change in the transient energy in an acoustic stimulus. Its output and thus the amplitude of N1 increases as the intensity level of the stimulus is increased and when the rate of presentation is slowed. When the output reaches a certain critical level, operations of the central executive are interrupted and attention is switched to the auditory channel. This switching of attention is thought to be indexed by a later positivity, P3a, peaking between 250 and 300 ms. In order to sleep, consciousness for all but the most relevant of stimuli must be prevented. Thus, during sleep onset and definitive non-rapid eye movement (NREM) sleep, the amplitude of N1 diminishes to near-baseline level. The amplitude of P2, peaking from 180 to 200 ms, is however larger in NREM sleep than in wakefulness. P2 is thought to reflect an inhibitory process protecting sleep from irrelevant disturbance. As stimulus input becomes increasingly obtrusive, the amplitude of P2 also increases. With increasing obtrusiveness particularly when stimuli are presented slowly, a later large negativity, peaking at about 350 ms, N350, becomes apparent. N350 is unique to sleep, its amplitude also increasing as the stimulus becomes more obtrusive. Many authors postulate that when the N350 reaches a critical amplitude, a very large amplitude N550, a component of the K-Complex is elicited. The K-Complex can only be elicited during NREM sleep. The P2, N350 and N550 processes are thus conceived as sleep protective mechanisms, activated sequentially as the risk for disturbance increases. During REM sleep, the transient detector system again becomes somewhat activated, the amplitude of N1 reaching from 15 to 40% of its waking level. Very intense and/or very infrequently presented stimuli might elicit a P3-like deflection suggesting an intrusion into some aspect of consciousness. The types of experimental paradigms used in most ERP studies are quite different from those used in the study of noise and its effects on sleep. ERP studies will need to employ procedures that have greater ecological generalization; stimulus intensity needs to be lower, less abrupt, with much longer durations, and importantly, stimuli should be presented much less often.
Keywords: Event-related potentials, ERPs, sleep, disturbance, obtrusiveness, sleep protection
|How to cite this article:|
Campbell K. Event-related potentials as a measure of sleep disturbance: A tutorial review. Noise Health 2010;12:137-53
| Introduction|| |
This article will review the use of the so-called "event-related potentials" (ERPs) as a measure of the extent of information processing in the sleeping human brain. This Special Issue is devoted to the study of how external environmental noise can affect the quality of sleep. A major issue is this field is a concern for identifying the types of environmental noise that will affect sleep. Most environmental sounds do not, in fact, disrupt sleep. ERPs provide an exquisite ms-by-ms means to trace the extent of processing of different external environment sounds, some of which may lead to the disruption of sleep, some of which may not. They also have the advantage that they provide a means of monitoring the extent of processing of a potentially disruptive environmental source prior to the time of the actual disturbance, during the disturbance and as a consequence of it. It is realized that many readers may not be familiar with ERP methods and thus this article is intended more as a tutorial rather than a review. More detailed reviews of ERP techniques and the specific effects of sleep can be found in other articles from my lab and those of my colleague, Ian Colrain, discussing ERPs in general [1-3] and ERPs that are unique to sleep, such as the K-Complex. ,
| The Theory of Signal Averaging|| |
ERPs are the minute changes in the ongoing electrical activity of the brain (i.e., the EEG), that are elicited by an external stimulus or an internal psychological event. A major problem in the use of ERPs for the study of environmental noise is that ERPs are usually so small that are they are buried in the background, ongoing EEG. A single noise event may well cause a change in the ongoing EEG but this will often be so small that the change cannot be observed, or at least be distinguished from the ongoing EEG. Signal averaging techniques can however be used to extract the ERP from the ongoing EEG. But, to do so requires that the stimulus/psychological event be repeatedly presented. An assumption of signal averaging is that the signal (the ERP) that is elicited by the stimulus is constant (it does not vary upon repetition of the stimulus) while the background noise (the ongoing EEG) is random (it can be negative- or positive-going following the presentation of the stimulus, its amplitude varying according to a Gaussian distribution). The average of a constant is, of course, the constant. On the other hand, if there are a sufficiently large number of stimulus repetitions, the average amplitude of the random, background EEG should tend toward zero, allowing the invariant, constant amplitude ERP to emerge. The averaging procedure does reduce the background noise, but it never eliminates it. Thus what is displayed as the ERP is in fact a summation of the true ERP and the remaining background noise. How many trials (stimulus presentations) are required to allow the signal to emerge? Phrased differently, what is the optimal signal-to-noise (S/N) ratio? The S/N ratio should be at least 2:1 (the ERP is twice as large as the background EEG), and ideally 5:1, although many labs insist that a 10:1 ratio is required. Mathematically, the amplitude of the background EEG noise decreases as a square-root function of the number of trials. Therefore, to reduce the amplitude of the background EEG 2 times, requires not 2 but 4 trials. In the waking state, ERPs vary in amplitude from less than 1 μV to perhaps as large as 10-20 μV, whereas the background EEG may vary in amplitude from 20 to 100 μV. Let us suppose that the ERP measures 5 μV and the background EEG measures 20 μV; the S/N ratio is thus 1:4 (0.25). To double this to 1:2 (0.5) would require the averaging of 4 stimulus presentations. To quadruple it (i.e., 1:1, the signal and noise are the same amplitude) would require 16 trials. To achieve a S/N ratio of 2:1 would require the presentation of 64 trials and to achieve the 5:1 ratio, 400 stimulus presentations are needed. In most ERP studies, stimuli can be presented rapidly, as often as several times a second, and perhaps as slowly as every 2 s. In the example above, to achieve an S/N of 5:1, if stimuli were presented every 2 s, a total testing time of 800 s, about 14 min, would be required. Environmental noise rarely occurs as often as every 2 s. To make matters worse, during sleep, the EEG may range from 100 to 400 μV within slow wave sleep (SWS) and often, the ERPs are much attenuated relative to the waking state. Fortunately, the background noise can be reduced in other ways. The easiest means is to filter unwanted frequencies in the EEG. Thus, delta activity accounting for the slow wave activity can be filtered reducing its amplitude significantly. Of course the frequency spectra of the actual ERP cannot overlap that of the EEG or else the filtering technique would also reduce the amplitude of the true signal. Moreover, as will be pointed out later in this article, the largest ERPs that can be recorded on the scalp of human subjects occur during non-rapid eye movement (NREM) sleep. A particularly obtrusive stimulus might elicit a very large (100 μV+) K-Complex. The background EEG during the long-lasting stage 2 of sleep might measure 50 μV. Thus, the S/N ratio is at least 2:1. The K-Complex may thus be readily observable on a single trial.
The results of the averaging process are displayed in [Figure 1]. In the left portion of the figure, five channels of continuously recorded eye movement activity (EOG) and EEG from a single subject are displayed. As in any physiological recording, a calibration signal providing a measure of time and the amplitude of the physiological signals is included. The total "sweep" time or "epoch" is 16 s. A brief 60 dB SPL pure tone auditory stimulus is presented on average every 2000 ms (ranging between 1800 and 2200 ms). The subject was asked to read a book and ignore the auditory stimuli. The first two channels display horizontal and vertical eye movements, respectively. Because the subject's task was to read a book, the horizontal eye movements are apparent as the eyes move from left to right (this causes a downward deflection in the display). When the eyes reach the end of a line, they jump back to the left edge of the page resulting in a large upward deflection in the display. Vertical eye movements are displayed in the second channel. As the subject moves his/her eyes from the right to the left edge of the page and down to the next line, a downward movement is visible. The remaining channels display EEG activity in the midline frontal, central and parietal (Fz, Cz and Pz) channels, respectively. Note that the amplitude calibration signal on the y-axis also includes an indication of size and the polarity of the signal. In this example, in the EEG channels, an upward deflection of the magnitude in the display represents a 40 μV positivity at the scalp (Fz, Cz or Pz electrode) relative to the reference (the tip of the nose in this example) and thus a downward deflection would represent a negative-going change. Most clinical EEG labs record "negative up." Most sciences however record "positive up." Unfortunately, there is no standard in ERP research. Many labs follow the clinical standard but many others follow the rest of science and thus display positive up. In this article, a positive-going deflection is indicated as an upward deflection. In [Figure 1], the time of a stimulus presentation is indicated by a large vertical tick mark in the display. The ERP in response to the stimuli is very small (less than 10 μV) and cannot be observed in the background noise of the "raw" EEG in the left portion of the figure. The right portion of the Figure represents the averaged EEG following 200 stimulus repetitions. The "sweep time" is 500 ms, beginning 50 ms before stimulus presentation. There should be no change in the ongoing EEG prior to the presentation of the stimulus. Thus, the average of this pre-stimulus period serves as a zero voltage baseline from which all peak deflections are measured. The calibration signal for the ERPs is much smaller, only 1 μV because the averaging process has reduced the background noise considerably allowing the constant amplitude ERP to emerge. A negative peak (the large downward deflection) is visible at about 100 ms and measures from 5 to 10 μV depending on the electrode site.
The averaging technique provides an exceedingly powerful means to allow the researcher to visualize minute responses in the large amplitude background, ongoing EEG. Nevertheless, it should be stressed that while averaging reduces the amplitude of the background noise, it does not eliminate it. Over a sufficiently large number of trials, the average of the background EEG should tend to zero but there is always at least some noise superimposed on the remaining ERP average. How can one know how much of the displayed response is a true ERP and how much is, in fact, residual noise? It is mathematically possible to compute the amplitude of the noise. A simple graphical means is however illustrated in the figure. The pre-stimulus baseline period should be "flat" because a stimulus has yet to be presented. In this example, there is some small fluctuation in its amplitude. This is probably residual noise. A more powerful means is however to run the experiment a second time and then display the two results. Random noise, by definition, should not replicate itself. On the other hand, the ERP should be the same across replications, because its amplitude is constant. [Figure 1] superimposes the average of the two replications. The two traces in the first 300 ms following stimulus presentation are very similar, although not identical. There is thus some noise remaining in the averages, although it is small
| Component Structure of ERPs|| |
ERPs consist of a series of negative- and positive-going components. One method of categorizing the different components employs the time (the "latency") at which a specific component reaches its maximum amplitude. Thus, one might observe a series of short, middle or long latency "components."
Short- and mid-latency evoked potentials
Short-latency brainstem auditory evoked potentials (BAEPS) peak within the first 10 ms of stimulus onset and reflect activity in the peripheral auditory nerve and the brainstem relay centers. In the waking state, the BAEP is not affected by attention. Indeed, it is not even affected by the deepest levels of sleep. , Only the physical aspects of the stimulus will affect the BAEP; the psychological salience will not affect it. Thus, the BAEP following presentation of a stimulus that disturbs sleep at one moment in time and following presentation of the identical stimulus but which fails to result in disturbance, should not differ. The mid-latency auditory evoked potentials peak between 10 and 50 ms. They appear to reflect processing in the thalamus and primary auditory cortex. The effects of sleep are somewhat equivocal, some studies failing to observe an effect while others demonstrating a reduction in the amplitude of various components of the mid-latency auditory response, particularly if stimuli are presented rapidly. ,,
Long-latency evoked potentials
The mid-latency auditory evoked potentials peaks are followed by a complex series of components that overlap both spatially and temporally, making their interpretation exceedingly difficult. The long-latency components occurring between 50 and 300 ms are markedly altered during the sleep onset period and by definitive non-REM sleep itself. The ERPs displayed in [Figure 1] would thus be categorized as long-latency ERPs. The most frequently studied of these components is N1, a negative wave peaking between 80 and 150 ms and a later positive component, P2, peaking between 180 and 225 ms. A great deal of attention has also been paid to a later positive component, P3, associated with the active discrimination of a rare stimulus event. In very easy discrimination tasks, P3 peaks at about 300 ms, thus its alternative label, P300. In some situations, a P3-like wave can be elicited even if the subject is not engaged in an active task (i.e., not actively attending to the stimulus channel). This passively elicited positive wave occurs from 250 to 300 ms. The P3 category of ERP components is thus subdivided into an earlier passively elicited P3a and a later P3b (P3b being equivalent to P300). A number of very long evoked potentials appear to be unique to sleep. They first begin to emerge in the sleep onset period and peak from 300 to 900 ms following stimulus onset.
This tutorial review will mainly focus on the late components of the auditory ERP. This is because factors such as stimulus relevance and "disturbance" have minimal effect on the short- and middle latency components but may have large effects on the late components.
Components can also be categorized on the basis of their functional significance, according to the experimental manipulations that might affect them. Components that are mainly affected by the physical aspects of the stimulus (e.g., its location, frequency, intensity or its duration in the auditory domain) are also called sensory or exogenous (exo outside) evoked potentials (EPs). Exogenous ERPs are minimally affected by psychological processes such as attention and consciousness. Components that are mainly affected by the psychological relevance of a stimulus are also called cognitive or endogenous (endo inside) EPs. Endogenous ERPs are minimally affected by the physical features of the stimulus per se. In general, during the waking state, the shorter latency components mainly reflect exogenous processes while the longer latency components reflect endogenous processes.
Because exogenous components are mainly affected by the physical features of the stimulus, it is critical that care be taken to assure that the stimulus that is presented to the ear does not vary during the course of the experiment. For this reason, during sleep, the use of loudspeakers should be avoided. Tight-fitting earphone inserts can assure constancy of stimulus input in spite of movement during the night.
Many authors restrict the usage of the label "ERPs" to cognitive EPs but this was not how the original usage was intended. Many clever experimental designs have been employed to determine whether an ERP component reflects the more sensory or the more psychological aspects of processing. Thus, if an experimenter increases the intensity of the stimulus (a manipulation of the physical aspect of the stimulus) and the morphology (shape) of an ERP subsequently varies, then it can be concluded that this is a sensory ERP. Let us suppose, the experimenter now presents an identical stimulus in two different conditions, an attend and an ignore condition. The ERP varies between the two conditions. This would be an example of a cognitive ERP because an internal (endogenous) psychological state, attention, has been manipulated while the physical stimulus has been held constant. Although the sensory and cognitive division is convenient, in reality, it may not nearly be this simple. Varying the intensity of the stimulus is indeed often used to study sensory ERPs. Because these ERPs are not affected by attention, subjects are often asked to carry out an incidental task, such as reading a book. They thus ignore the auditory stimuli. Let us suppose that the ERP in question does not actually vary until very high intensities are presented. Is this because of the physical manipulation of the stimulus or because of the fact that the subjects can no longer ignore the stimulus, in spite of instructions to do so?
ERPs are almost always recorded from the scalp of normal subjects. The scalp-recorded components represent the volume-conducted activity of different intracranial (within the brain) sources (or generators). ERP components are thus also differentiated on the basis of their intracranial sources. Indeed, in their classic review, NĠĠtĠnen and Picton define a component as "the contribution to the recorded waveform of a particular generator process, such as the activation of a localized area of cerebral cortex by a specific pattern of input Whereas the peaks and deflections (of the scalp-recorded potential) can be directly measured from the average waveform, the components contributing to these peaks can only usually be inferred from the results of the experimental manipulation" (p.376). The scalp distribution of a component can be used to mathematically model the underlying intracranial sources. Importantly, two ERP peaks that have different scalp distributions must have different intracranial sources. As an example, P3a and P3b, mentioned above, can also be differentiated on the basis of their scalp distribution. P3a is maximum over centro-frontal areas of the scalp, whereas P3b has a more posterior parietal maximum scalp distribution. Their underlying generator sources must therefore be different.
This special issue examines the effects of environmental noise on the disruption of sleep. In order to sleep, the processing of all but the most highly relevant of stimulus input must be inhibited. Sleep is called an unconscious state because the sleeper is almost entirely unconscious of external environmental stimuli, as a result of this inhibitory process. Although there are other unconscious states (e.g., coma, general anesthesia), a major difference between natural sleep and these other unconscious states is that the sleep state is rapidly reversible. The sleeper must be made aware of highly relevant biological or personal information (a very intense stimulus, the interruption of breathing, a baby's cry), thus permitting the reversal of sleep and in the subsequent waking to a conscious state, to take appropriate action. Yet for sleep to be of benefit, it needs to remain as undisturbed as possible, and only interrupted when absolutely necessary. Unfortunately, sleep may be disturbed and waking may occur as a result of what turns out to be an irrelevant external noise (a passing train). Even if the incoming stimulus does not awaken the sleeper, it may produce a brief-lasting microarousal. Somewhere in the brain, a decision must be made whether incoming stimulus input is relevant enough to disrupt sleep or to allow sleep to continue undisturbed. In the accompanying Muller-Gass and Campbell article in this issue,  a series of ERP components are discussed, each of which play a significant role in making a decision about the nature of the incoming, encoded stimulus, thus protecting sleep against unwanted disturbance. These will be discussed in more detail in this tutorial.
Sleep is not a uniform state. Rather it is best conceptualized as a series of different states (REM and non-REM) where non-REM (NREM) can be broken down into a continuum of stages (1 to 3 in the newer standardized sleep scoring system) reflecting depth of sleep. The processing of external stimulus input is quite different between NREM and REM sleep.
Many models of auditory processing have been proposed. This tutorial will describe an elaborate model of auditory processing, proposed by R. NĠĠtĠnen, a well-known Finnish cognitive neuroscientist. The NĠĠtĠnen model  is unique in that it provides a means of quantifying the extent of processing even if the subject is not actively attending to the auditory stimulus. Changes in various ERP components are used to provide evidence of processing.
The vertex potential, N1-P2
NĠĠtĠnen discusses two routes through which an otherwise unattended auditory stimulus can result in attention capture, a forced interruption of the central executive and subsequent possible switching of attention to the auditory channel. The first route involves activation of a transient detector system by an obtrusive stimulus. This tutorial will mainly be concerned with this system. A stimulus does not, however, need to be obtrusive to result in interruption of the central executive, if the stimulus signals a change from the acoustic past and the extent of change is particularly large. The changed stimulus results in activation of a second system, appropriately named, the change detection system. For example, a change in the frequency, duration, location or intensity of a stimulus will activate the change detection system. This system could therefore result in attention capture even if the incoming stimulus signals a decrease in the intensity of the stimulus, provided the extent of change is large enough. Modern alarm systems tend to incorporate features that would activate both systems. An alarm is loud and obtrusive, thus resulting in considerable activation of the transient detector system. The features of the alarm however change; it is not a homogenous sound. The pitch of the alarm might vary; it might pulsate at irregular times. It thus also activates the change detection system.
The functioning of the transient detector system is described in more detail in the Muller-Gass and Campbell article in this issue. The transient detector system, as the name implies, is responsible for the detection of changes in the transient energy of a stimulus, most often, the onset of a brief-lasting stimulus. The output of this system is proportional to the change in energy and the rareness of stimulus presentation. Thus, activity in the transient detector system dramatically increases following the presentation of a particularly intense stimulus even if it is presented relatively frequently, or alternatively a relatively less intense stimulus, if it is presented quite infrequently. The output of this system can be monitored through a small amplitude, negative ERP component, N1, peaking at about 100 ms. When stimuli are presented relatively rapidly (every 0.5-4 s), the peak of N1 is maximum over fronto-central areas of the scalp and inverts in polarity (i.e., is recorded as a positive potential) at lateral, inferior regions below the Sylvian fissure. The output of the transient detector varies directly as a function of the energy in the acoustic stimulus. Thus, the amplitude of N1 increases with increases in the transient energy in the brief-duration stimulus. This is illustrated in [Figure 2]. This Figure is adapted from Cote et al.  A 55-ms, 1000-Hz auditory stimulus was presented every 2 s. Within each condition, the intensity of the stimulus was set at random to be either 60, 80 or 100 dB SPL. ERPs were recorded in both the waking and sleeping states. In the waking state (left-hand column), the amplitudes of N1 and a later P2 gradually increase as the intensity of the stimulus increases. In the NĠĠtĠnen model, the operations of the transient detector system are claimed to function relatively independent of attention, consciousness and ongoing cognitive activity. It thus does not matter what the observer "is doing." Because the operations of this system occur at a pre-conscious level, at this point in time, the observer would not be aware that the auditory stimulus has been presented. The output of the transient detector system is, however, forwarded to the central executive, controlling the allocation of attentional resources needed for the demands of ongoing cognitive tasks. When this output reaches a certain critical threshold, ongoing cognitive activity is interrupted and attention is then switched to the (task-irrelevant) auditory channel. Attention capture thus provides a means by which the observer is forced to become aware of information in the previously unattended auditory channel. Although the operations of the transient detection system function outside consciousness (at a pre-conscious level) and independent of ongoing cognitive demands, NĠĠtĠnen concedes that the threshold for interrupting the central executive is flexible. It might, for example, be set to be quite high during particularly demanding cognitive tasks, but could be lowered when the demands are much reduced. There may also be considerable individual differences. This switching of attention may also cause deterioration in performance on the relevant, ongoing cognitive task. A very low threshold would lead to frequent and inappropriate interruption of the central executive, resulting in high levels of stress and anxiety. This is thus one route through which particularly obtrusive environmental noise can lead to sleep disturbance. Further, the fact that the threshold for the interruption of the central executive is variable across conditions and individuals might also explain why external environmental noise might be quite disruptive in one situation, yet not in another. It can also explain the considerable variability that is observed among individuals when exposed to seemingly identical levels of noise.
The switching of attention to the formerly irrelevant auditory channel allows its contents to become available to consciousness. The contents are however general or what we  call "fuzzy." Thus, the observer may be aware that some sort of auditory stimulus has been presented but may not be aware of its specific features. The switching of attention is thought to be reflected by a later positive component, the P3a, peaking 250-300 ms after stimulus onset. In [Figure 2], in the waking state, the amplitude of N1 gradually increases as the intensity of the stimulus increases. However, a P3a is only observed only when the very loud 100 dB SPL stimulus is presented. The output from the transient detector system following presentation of the lower intensity level 80 and 60 dB SPL stimuli would presumably not have been large enough to reach the critical threshold necessary for the forcible interruption of the central executive. Note, however, that stimuli in the Cote et al. study were presented rapidly (every 2 s). In the accompanying Muller-Gass and Campbell article (this issue), lower intensity stimuli were presented at about the same rate (on average every 2 s) or very slowly (on average, every 10 s). Stimuli that are presented slowly are obtrusive and, thus, difficult to ignore. This is because the output of the transient detector system also increases as the rate of stimulus presentation is slowed. While Cote et al. were able to record a P3a only when stimuli were very loud (100 dB), Muller-Gass and Campbell noted that a P3a could be elicited by a 60-dB stimulus if it is presented very slowly. This is not incidental. The types of environmental noise that most often disturb sleep are typically not loud; they do however occur very infrequently.
Sleep, to be of benefit, must be undisturbed by external stimulus input. Somehow the processing of all but the most salient of auditory input must be inhibited. ERPs provide an exquisite means to quantify the massive inhibition of processing, necessary to protect sleep from unwanted disturbance. Nevertheless, highly salient information must be capable of disrupting sleep and perhaps awakening the sleeper. It is the auditory modality that is best suited for this purpose. While we can close our eyes to gate visual input, we cannot close our ears. The auditory system provides a means of alerting the sleeper to highly salient acoustic stimuli. Almost all alarm systems function by emitting high-intensity auditory stimuli. Again, ERPs provide an exquisite means to quantify the processing of stimuli that potentially may disrupt sleep.
The middle and right portions of [Figure 2] illustrate the effects of NREM and REM sleep, respectively, on the output of the transient detector system. Remarkably, during NREM sleep, N1 is much attenuated and does not exceed the zero voltage baseline even if the intensity of the stimulus is 100 dB SPL! It would appear that the operations of transient detector system essentially cease during NREM sleep. The failure to record an N1 during NREM sleep is a highly consistent finding, observed by many labs using many different experimental procedures. A number of studies have also examined the sleep onset period, marking the transition from a waking and conscious state to one of sleep and unconsciousness. In some of these studies, the loss of consciousness is defined on the basis of a behavioral response. Subjects are asked to button press upon detection of an external auditory stimulus. When they are unable to do so, the auditory stimulus N1 is much attenuated. ,, The failure to detect external stimulus input coincides with the appearance of theta activity in the EEG during stage 1 of sleep. N1 is also much reduced in amplitude upon the appearance of theta activity, although, as expected, this effect may be mitigated when stimuli are presented slowly.  Recall that in order for the central executive to be interrupted and for the observer to become conscious of the unattended auditory stimulus requires that the output of the transient detector system reach a critical threshold. The ERP data provide strong evidence that the central executive cannot be interrupted during NREM sleep, at least through the transient detector route.
Obviously, acoustic stimuli can disrupt NREM sleep and awakenings do occur. How is this possible if the operations of the transient detector system cease? While averaging does permit the extraction of small-amplitude ERPs such as the N1 from the ongoing background EEG, this can only be accomplished by the repeated presentation of the stimulus. Almost all ERP studies examine processing in homogeneous stages of sleep. The highly consistent finding that N1 is at or near baseline level during NREM sleep is based on data from undisturbed sleep. The times at which the stimulus does disrupt sleep are usually rejected from the averaging process (see, e.g., Muller-Gass and Campbell in this special issue). One might speculate that N1 would not be at baseline level if the stimulus does result in an arousal or an awakening. The disruption of sleep is, however quite rare, so rare that there are an insufficient number of sleep-disrupted trials to permit the extraction of the what might be a small-amplitude N1 from the large background EEG of NREM sleep through the averaging process.
The absence of an N1 does not imply that processing ceases altogether. If this were the case, the ERP waveform following the N1 component would essentially resemble the pre-stimulus period; in other words, it would appear as a straight line. As is evident in [Figure 2], this is far from the case. A series of later components emerge during the sleep period. One of these is a later positivity, P2, peaking at about 180-200 ms. The P2 component is also apparent in the waking state. Other later negativities, such as a large amplitude N350 (visible during NREM sleep in the figure) and still later but much larger, N550 (not visible in [Figure 2]) appear to be unique to sleep. These components will be discussed later in this review.
A frequent finding is that the amplitude of P2 is often much larger during NREM sleep than in the waking state.  The decrease in the amplitude of N1 and increase in the amplitude of P2 is often also seen in the waking state during studies of selective attention. When subjects are asked to attend to stimuli in one auditory channel (perhaps the left ear) and ignore another auditory channel (perhaps the left ear), the amplitude of the scalp-recorded N1 may increase but the amplitude of P2 may however decrease. NĠĠtĠnen explains that this is because of the superimposition and summation of another independent attention-related component that he labels as the Processing Negativity (PN), on both N1 and P2. The purpose of selective attention is to allow relevant, attended information to attain consciousness. At the same time, information that is irrelevant and thus to-be-ignored, must be prevented from doing so. Processing of the to-be-ignored information must be inhibited. PN provides a means of monitoring the extent of processing of information in the unattended channel and attended channels. PN is thus an endogenous and relatively long-lasting component reflecting the additional processing that an attended channel receives. Because it is a negative-going waveform and overlaps temporally (in time) and spatially (sharing a similar scalp distribution) with N1 and P2, it causes N1 to appear to increase but P2 to decrease in amplitude. Importantly, the PN is present following presentation of stimuli in the to-be-ignored channel. This is consistent with the fact that the observer must process enough information extracted from stimuli in the unattended channel until a decision can be made that it is indeed irrelevant, and subsequent processing is no longer warranted. Further, attention-related processing then continues in the attended channel. If the to-be-attended and to-be-ignored channels are easily distinguished, the PN to the stimuli in the unattended channel will cease early, at or before the time of N1. If, on the other hand, the two channels cannot easily be distinguished, then the PN to the to-be-ignored channel will continue for some time, well after the peak of N1. Thus, N1 (or in reality, the composite N1 + PN) for the stimuli in both the to-be-attended and to-be-ignored channels may be identical.
Campbell and Colrain  suggest that attention-related processing must cease in order for sleep to occur. Importantly, they emphasize the NĠĠtĠnen claim that all stimulus inputs, whether attended or not, must receive a certain amount of attention-related processing, as reflected by the amplitude of PN. In order to protect sleep from unwanted and irrelevant disturbance, it is this waking Processing Negativity (wPN) that is removed during the sleep onset process and during definitive NREM sleep. The removal of a long-lasting summating and overlapping scalp causes the amplitude of N1 to be reduced (N1 is less negative-going), but on the other hand, causes the amplitude of P2 to increase (P2 is also less negative- or more positive-going). The large reduction of the amplitude of N1 but the increase in the amplitude of P2 thus reflects the massive inhibition of auditory information processing during the NREM period. Crowley et al. thus indicate that the large-amplitude P2 may serve as a means to quantify the extent of inhibition of auditory input during sleep. Note that in [Figure 2], during NREM sleep, P2 is especially large following presentation of the 100-dB SPL stimulus, reflecting the difficulty in the processing of this very obtrusive stimulus. In addition to the large P2, a very large negativity peaking at about 350 ms (thus N350) is apparent.
ERPs during REM sleep are very different from those during NREM sleep. Again, many studies have indicated that N1 can be elicited during REM sleep, although it is much attenuated. Nevertheless, this suggests that the transient detector system remains functional in this stage of sleep but its output is limited and as such, only very obtrusive stimuli will be capable of interrupting ongoing cognitive processing (perhaps engaged in internal mentation). A number of studies have now reported P3-like components following presentation of particularly obtrusive stimuli during REM sleep. There is some debate whether these reflect processing associated with P3a, peaking from 250 to 300 ms or P3b, peaking from 300 to 450 ms activity. As mentioned, P3a and P3b are also distinguished on the basis of their functional significance and their scalp distribution. The late positivities share a role in consciousness. They are both elicited by rare and relevant stimulus events. However, in the case of P3b, the subject attends to the channel and must actively detect the rare stimulus event, thus becoming conscious of it. In the case of the earlier P3a, the subject is engaged in another task and thus it (the P3a) can only be elicited passively, though the attention capture process. It is the features associated with a rare, but highly relevant stimulus that forces the subject to switch attention to the stimulus channel and only then become conscious of its contents. P3b is maximum over parietal areas of the scalp although and is reduced in amplitude over more anterior regions. P3a is more anteriorally distributed being maximum over centro-frontal areas and is reduced in amplitude over posterior regions. Some authors (see, e.g., Cote et al. , ) have noted that a particularly loud and obtrusive stimulus will elicit a parietal maximum P3, peaking at about 300 ms. This is typical of the P3b. Others (see, e.g., Macdonald et al.  Muller-Gass and Campbell)  also report a P3 to less obtrusive stimuli. This P3 also peaks at about 280-300 ms following stimulus presentation but takes on a more central scalp distribution, typical of a P3a. The procedures in the various studies differ at times widely, and it is possible that these differences can explain why a P3a is elicited in one study and a P3b in another. There are also more subtle explanations. It is possible that the P3a is delayed during REM sleep, thus it peaks at a latency more typical of the P3b. Nevertheless, two distinctly different scalp distributions of the late positivities have been reported. Obtrusive stimuli will elicit a later very large amplitude N350 during NREM sleep and may continue to do so in REM sleep. This N350 takes on a very distinctive central-maximum scalp distribution, similar to that of the P3a. Because the scalp-recorded P3a and N350 may overlap temporally and spatially, the very large amplitude N350 may summate with and thus appear to cancel the appearance of the smaller amplitude P3a over central and frontal regions, leaving only its parietal dispersion to be apparent. Recent and complex analyses procedures such as those employing independent component analyses (IAT)  might be employed to separate a scalp deflection into additive subcomponents assuming the independent contributions of single source components such as the P3a and N350. This will, however, require extensive multichannel recordings from as many as 32 to 128 different scalp sites.
Regardless of whether the late positivity reflects a P3a or P3b source, its presence provides very strong evidence that the stimulus was so obtrusive as to disrupt REM sleep forcing the subject to become conscious that it was presented. The nature of this consciousness is also debated. For example, Cote et al.  interpreted their parietal maximum P3 and the absence of dispersion to more frontal sites as a reflection that the sleeper was in fact conscious (aware) of the stimulus but that another aspect of consciousness (perhaps affect, a role of the frontal lobes) was not present. This may well explain why such highly obtrusive stimuli do not awaken the sleeper. The sleeper may be conscious of the external stimulus, but in the absence of affect (perhaps fear following a particularly loud stimulus), the decision is made to allow REM sleep to continue undisturbed. There is also evidence that the external stimulus may be incorporated into current mentation, typically dreaming. Schredl et al.  presented olfactory stimuli to subjects during REM sleep. They reasoned that olfactory stimuli, because they are so biologically relevant, might be incorporated in dream content. They thus presented subjects with either hydrogen sulfide (described as the smell of rotten eggs) or phenyl ethyl alcohol (smell of roses) intranasal chemosensory stimulation during REM sleep. Upon awakening the subject from REM sleep, subjects reported more positively-toned dreams following presentation of the pleasant stimulus and more negatively-toned dreams following presentation of the unpleasant stimulus.
ERPs that are unique to sleep, the N350
In [Figure 2], during NREM sleep, a large-amplitude negativity, peaking at about 350 ms (thus, the N350), is apparent following presentation of the 100 dB stimulus and it is still apparent but reduced in amplitude following presentation of the less intense, 80 dB stimulus. In the accompanying Muller-Gass and Campbell article,  a large N350 is also visible following presentation of the 60 dB stimulus, if it is presented slowly. Picton et al.  first noted that a "sleep N2" (to distinguish from an N2 observed in the waking state), maximum over central areas of the scalp, can be elicited about 350 ms after presentation of an auditory stimulus in subjects who became drowsy and fell asleep during a long-duration vigilance task. Many labs have confirmed that this sleep N2 can be elicited by relatively obtrusive auditory stimuli, its latency ranging from 300 to 350 ms. The label "N350" has gained currency over the past 20 years in spite of the fact that its latency, typical of many endogenous components, can be somewhat variable. It cannot be recorded at all during the waking period. Colrain et al.  indicate that the N350 may be identical to the classic vertex sharp wave, so-named because it is maximum over central areas of the scalp (the "vertex"). The vertex sharp wave is commonly used as a marker of the sleep onset process.
N1 and P2 are elicited by any auditory stimulus, even those that are only slightly above auditory threshold. N350, on the other hand, is only elicited by fairly intense auditory stimuli or those that are presented very infrequently. The effects of the rate of stimulus presentation are dramatically illustrated in [Figure 3]. This figure is taken from Campbell et al.,  a collaborative effort between his lab and that of David Michaud, examining the disruptive effects vehicular back-up alarms during sleep. An acoustic back-up alarm consists of a series of repetitive on-off tones that are emitted while the vehicle is reversing. When the vehicle is advancing, the alarm is not sounded. In countries having severe winters, snow is often removed from roads during the night. The back-up alarms inadvertently act as a considerable environmental noise hazard and may disrupt sleep. Indeed, during the infamous construction of the Central Artery Tunnel in Boston (the Big Dig project), the nighttime use of back-up alarms was the major complaint of residents living in the vicinity. We designed a lab study to determine why these alarms are so disruptive. The back-up alarm is experienced as a train of repetitive tones as the vehicle reverses followed by a long period of silence as the vehicles now moves forward after which the train is again sounded as the vehicle again reverses. The time between the presentation of the initial and the subsequent acoustic signals is thus very short. However, the delay between the presentation of the final stimulus in the train and the subsequent initial stimulus is very long. Campbell et al.  recorded ERPs following presentation of these back-up alarms in the sleep lab. The back-up alarm was modeled by presenting a sequence of five auditory 1000 Hz tones having a duration of 500 ms and an offset time of 500 ms (thus imitating the sound of the reversing vehicle). Stimulus presentation was paused for between 10 and 15 s following the presentation of the fifth stimulus in the sequence (thus imitating a vehicle now going forward) and then the alarm was reinitiated. In different conditions, the intensity of these tones was set at either 60 or 80 dB SPL.
The ERPs following the processing of the five stimuli in the sequence during stage 2 of NREM sleep are illustrated in the figure. The initial stimulus in the sequence at times elicited a large amplitude K-Complex and at times, the same identical initial stimulus did not. The stimuli in positions 2-5 never elicited a K-Complex. Sequences were thus sorted and averaged on the basis of whether the initial stimulus in the sequence elicited a K-Complex. In [Figure 3], the average to the five different stimuli in the sequence is presented, when the initial stimulus did not elicit a K-Complex (upper portion) and when it did (lower portion). A large amplitude (about 25 μV) negative deflection, peaking at about 350 ms is apparent following the presentation of the initial stimulus when this stimulus did not elicit a K-Complex. This negativity is the N350 and it is maximum in amplitude at the central electrode site (Cz). It is difficult to discern any response following presentation of the remaining four stimuli in the sequence. The large N350 that occurs to the initial stimulus is an effect of the time between stimulus presentations. The time between the offset of the last stimulus in the sequence and the subsequent onset of the first stimulus is more than 10 s, while the time between the offset and onset of the remaining four stimuli is only 500 ms. On more than one-third of the sequences, the first stimulus elicited a very large amplitude K-Complex (lower portion of the figure), maximum at the frontal (Fz) electrode site. The K-Complex may measure more than 100 μV, many times larger than the 1-10 μV amplitude N1, apparent in [Figure 2].
[Figure 4] presents a zoom of the ERPs elicited by the first stimulus in the sequence, again single trials being sorted and averaged according to whether this stimulus elicited a K-Complex or not. The N350 is also influenced by the obtrusiveness of the stimulus. Trials are thus also sorted and averaged according to the intensity of the stimulus, either 60 or 80 dB SPL. In the left portion of the figure, the average of all initial stimulus presentations in which a K-Complex was not elicited is displayed. Neither an N1 nor a P3a was elicited by the initial stimulus regardless of whether a K-Complex was elicited or not. However, on these trials, the stimulus did elicit an N350, maximum over central areas of the scalp, which was much larger when the intensity of the initial stimulus was 80 than 60 dB SPL. This is also consistent with other sleep studies (see the classic Bastien and Campbell  ). The amplitude of N350 thus increases as the intensity of the stimulus increases or when the rate of stimulus presentation is slowed. Thus, even when particularly obtrusive stimuli fail to elicit either an N1 or a P3a, the N350 may still be elicited.
How can this be if the transient detector system has not been activated? The most parsimonious answer to this question is that the N350 generation system is independent of the functioning of the transient detector system. There is now good evidence that the sources of the N1 auditory ERP component are located in the auditory cortex.  The operations of the transient detector system are specific to the auditory modality. The vertex sharp wave, analogous if not identical to the N350, on the other hand, appears spontaneously at sleep onset in the absence of any apparent stimulus. Of course, this does not mean that an uncontrolled and unidentified stimulus source, a muscle twitch, movement, touch, extraneous acoustic noise or perhaps even internal thought might be responsible for the elicitation of the vertex sharp wave. The long latency of the stimulus-elicited N350, 300-350 ms after stimulus onset, is consistent with an endogenous, modality nonspecific functional role. The large majority of studies have however employed auditory stimuli. Colrain et al.  were the first to indicate that vertex sharp waves could be systematically elicited by stimuli in different modalities, elicited by both auditory stimuli and occlusions of inspiratory respiration (a highly biologically relevant "stimulus" event), a finding later replicated by Gora et al.  The actual intracranial sources of the N350 are still not known. Nevertheless, the conditions under which it is elicited, including the fact that it can be elicited by stimuli in different modalities and perhaps may even occur spontaneously, strongly suggest that the source is not in or around the auditory cortex, the likely site of activity in the transient detector system. A major purpose of the transient detector system is to provide a means by which the observer can potentially become conscious of highly relevant stimulus input to an otherwise unattended auditory channel. During NREM sleep, it would appear that even the most obtrusive of stimuli cannot intrude into consciousness (but again, caution should be heeded; ERPs studies have not examined the processing of stimuli that do result in awakenings). This suggests that forced awakenings may not be a consequence of an intrusion into consciousness during NREM sleep. Rather consciousness may be a consequence of the awakening. The sleeper may be awakened by a particularly obtrusive stimulus but may not know why. It is only upon the reversal of sleep and a subsequent return to a conscious, waking state that the observer becomes aware of the cause of the forced awakening, perhaps the sound of a fire alarm.
The definition of what constitutes and defines an arousal during sleep has been the subject of much discussion. The data that have accumulated about the circumstances under which the N350 is elicited have led the recent Bonnet et al.  task force to include the N350 ERP as a possible measure of arousal during sleep. In actual fact, there is little evidence to suggest that the sleeper is actually aroused. The sleeper rarely awakens following an N350 nor is there a dramatic change in the characteristics of the EEG or in the stage of sleep. Perhaps the N350 may be better understood as a protective response, inhibiting possible awakening and sleep disruption. The amplitude of N350 can thus be used as an indication of potential sleep disruption. Nevertheless, there is still another protection mechanism that may occur following the N350. This is the K-Complex.
The N550 and the K-Complex
The Bastien and Campbell , studies found that the amplitude of the N350 increases as stimuli are presented increasingly slowly and with increases in the intensity of the stimulus, findings now replicated in many different studies. Interestingly, Bastien and Campbell also noted that at times the same physically identical stimulus would also elicit the very large amplitude K-Complex. When the K-Complex was elicited, the amplitude of N350 was larger compared to trials in which the same stimulus did not elicit the K-Complex. They also noted that the probability of eliciting a K-Complex was much larger for higher intensity stimuli and when the stimulus was presented slowly, the same factors that will cause the amplitude of the N350 to increase. Thus, a model emerged in which the N350 gradually increased in amplitude with increasing obtrusiveness of the stimulus and when it reached a critical threshold level, it could only continue to increase in amplitude if the very large amplitude N550 was also elicited. The N350 was thus conceived to act as a trigger for the K-Complex.
The K-Complex is so large that it was easily observed in the background EEG of the earliest sleep recordings. Colrain  provides an interesting, detailed and at times entertaining historical review of the K-Complex. Loomis et al.  reported that the presentation of an external stimulus to the sleeping subject could elicit a very large amplitude waveform, which they called the "K-Complex." In the same year, Davis et al.  reported that the K-complex might also appear spontaneously without any apparent external stimulation. [Figure 5] presents two 16 s "epochs" of raw, unaveraged EEG recorded during stage 2 of sleep in a single subject. An 80 dB SPL brief duration tone was presented after 8 s. In the figure, the EEG is divided into 8 s pre- and 8 s post-stimulus periods, the onset of the stimulus occurring at time 0. In the upper portion of the figure, the stimulus elicits a negativity peaking at about 350 ms. This is the N350. After a delay of about 30 s, the stimulus is again presented. It now elicits a large amplitude K-Complex (lower portion of figure). The K-Complex consists of a series of components including the N350, maximum over central areas of the scalp and a much larger negativity, the N550, maximum over frontal areas of the scalp. Some authors also include a later positivity, the P900, also apparent in the figure, but it is poorly understood. On a later trial, the same stimulus might now elicit another K-Complex or it might not.
It has long been known that the K-Complex can be elicited by a wide range of stimuli in different modalities  and is also thus not specific to the auditory. The N550 is maximum over fronto-central areas of the scalp. The generator sources of the N550 also remain unknown, although Colrain and Crowley  suggest that it may be modulated by a brainstem reticular system-inferior frontal cortex circuit, thus having no direct input from the auditory cortex in the temporal lobe.
Although many studies have now indicated that the N350 is indeed larger on trials in which the N550 is also elicited, some studies have indicated that this is not always the case. Gora et al.  and Crowley et al.  observed that it was not larger when stimuli elicited both an N350 and an N550 compared to when they stimuli elicited only the N350. This may be explained in part by the fact that the N350 is often difficult to discern and is buried in the much larger N550 waveform. In the right portion of [Figure 4], the averaged K-Complex is displayed, elicited by either a 60 or an 80 dB SPL brief duration tone. The large amplitude N550 is readily visible but the N350 is not. Again, the application of sophisticated mathematical techniques such as ICA may be required to separate the two components.
While the P2 and N350 can both be elicited in NREM and are attenuated during REM sleep, the K-Complex is exclusive to NREM sleep. It cannot be elicited during REM sleep. The probability of eliciting a K-Complex is much dependent on the physical features of the stimulus, its abruptness (rise-time), intensity and rate of presentation. A high-intensity stimulus is much more likely to elicit a K-Complex than a low-intensity stimulus, rates varying from about 25% of trials for a 60-dB SPL stimulus to 50-60% of trials for a higher 80 dB stimulus. Critical to the elicitation of the K-Complex is the rate of stimulus presentation. Although both P2 and N350 can be elicited by stimuli presented relatively rapidly, a K-Complex cannot be elicited if the time between stimuli is less than 5 s. On the other hand, if the stimuli are presented as slowly as every 30 s, a K-Complex might be elicited on as many as 60-70% of trials. The probability of eliciting a K-Complex is also dependent on circadian influences. Remarkably, Nicholas et al.  reported that a very slowly presented 80 dB SPL stimulus could elicit a K-Complex on about 90% of trials if presented during stage 2 in the 10-min period just prior to the onset of stage 3/4 sleep but on only 63% of trials during stage 2 in the 10-min period just prior to stage REM. This might be related to the fact that the amplitude of both P2 and N350 are also larger prior to SWS than REM sleep. Because SWS occurs mainly in the first half of the night and stage REM in the second half of the night, the probability of eliciting a K-Complex is much higher during stage 2 in the first half than during stage 2 in the second half of the night. While the probability of eliciting a K-Complex varies with the stimulus features, and between and within the stage of sleep influences, an unusual aspect of the N550 is that on trials in which it is elicited, its average amplitude is invariant. Regardless of the intensity of the stimulus or the rate of presentation, the amplitude of the N550, unlike that of the P2 or N350 follows an all-or-none law; it is either elicited with no variation in its amplitude or it is not elicited. This is also illustrated in the right portion of [Figure 4], displaying the average of all trials when a K-Complex was elicited by either a 60 or an 80 dB tone. Although, the 80 dB tone did elicit more K-Complexes, when it was elicited by the 60 dB tone, the amplitude of the large N550 did not differ from that elicited by the 80 dB tone.
There is much debate about the functional role of the K-Complex. The early literature hypothesized that the K-Complex reflected an arousal response, elicited by the types of stimuli that are most likely to result in attention capture during the waking state. Thus, during NREM sleep, the K-Complex provided the critical means to allow the sleeper to awaken following presentation of highly salient stimulus events. In actual fact, awakenings are exceedingly rare following the K-Complex. Indeed, spectral analyses following trials in which a K-Complex is elicited or not elicited reveal remarkable little change and, if any, there might be a slight increase in delta activity contrary to what the arousal theory would predict.  An alternative "protective" theory has gained foothold (but see Halasz  for another perspective). While the K-Complex is certainly elicited by salient and biologically relevant information, the analysis of their features reveals that they are not so biologically urgent that the sleeper need awaken or even alter their current depth of sleep. To test this hypothesis, Peszka and Harsh  studied the effects of sleep deprivation and Nicholas et al.  studied the effects of sleep fragmentation (in which the sleeper is frequently awakened but otherwise has normal sleep) on the probability of eliciting a K-Complex. Sleep is typically "deeper" (more SWS) following these manipulations. The arousal hypothesis would thus predict that fewer K-Complexes should be elicited by the external stimuli because of the increased depth of sleep. The protective hypothesis would however predict that more K-Complexes should be elicited because of the need to protect sleep on the recovery night. The probability of eliciting a K-Complex was higher on the night following the experimental manipulations than in the control night of sleep.
The Colrain group have also studied individual differences in the probability and characteristics of the K-Complex. For example, in the elderly, a group marked by decreasing amounts of SWS, fewer K-Complexes were elicited by external stimuli and when they were, the amplitude of the N550 was much attenuated.  This might offer an explanation of why the sleep of the elderly is so often disturbed and fragmented. The proportion of SWS is also much reduced in alcoholics. The amplitude of the N550 of the evoked K-Complex has also been found to be much reduced in this group.  The production of delta waves and the K-Complex appear to be related to the volume of cortical populations in the frontal lobe. The frontal lobe does not fully mature until late adolescence. Melendres et al.  have noted that although respiratory occlusions will elicit a large amplitude K-Complex in 5-12 yr old children, the N550 is maximum over central areas of the scalp rather than the frontal maximum seen in young adults.
Limitations of ERP studies
ERPs are best elicited by abrupt, brief-duration stimuli. In addition, most studies employ relatively moderate to high intensity level stimuli (to maximize the amplitude of the ERP signal), presented at relatively rapid rates (to allow for the presentation of many stimuli and reducing the amplitude of the background noise through averaging). The features of environmental noise observed in aircraft, highway and rail traffic field and lab studies are very different. Their intensity as measured in the home setting is less intense. Their acoustic frequency spectrum is much more complex than the acoustically simple pure tones often used in ERP studies. Environmental stimuli generally have a much longer duration, their onset and offset are less abrupt (long rise- and fall-times) and they occur very rarely. A major limitation of ERP methodology is that a large number of stimuli need be presented in order for small amplitude components such as the N1-P2 to emerge from the background, ongoing EEG. It may not be possible to monitor the operations of the transient detector system as reflected by the N1 following presentation of environmental noise because of the relatively rare occurrence of these stimuli. This is not the case for the much larger amplitude N350, which can be observed after the averaging of only a few stimulus repetitions and of course, the K-Complex which can be observed following a single stimulus presentation, at least during low-amplitude stage 2 NREM sleep.
Some advances have been made in the study of ERP following the presentation of spectrally complex stimuli including human speech and "novel" environmental sounds (e.g., whistles, dog barks, squeaks) but their duration is typically less than 500 ms. Little is known about the processing of the many other types of environmental sounds employed in several of the studies in this special issue, even in the waking state, let alone during sleep. Recently, my lab has begun to systematically do so, examining the influence of the various features of long duration stimuli. We begin these studies by examining ERPs in the waking state. We employ a rationale similar to that of Bruck et al.  who studied the effectiveness of various types of fire alarms in arousing subjects from sleep. The physical features of the different alarms are first tested in the waking state to determine the threshold at which they can be heard. The Bruck lab notes that the waking threshold is very predictive of the threshold for inducing an awakening during sleep. Thus, responses in the waking state can be used to predict sleep disturbance and eventual awakenings.
My lab has run initial pilot studies to examine the processing of longer duration stimuli. It has now been well-documented that during the waking state, the onset and the offset of a long duration stimulus elicit transient negative-going ERPs, the N1-on and N1-off, each peaking at about 100 ms following the change in transient energy. The continued presentation of the stimulus also elicits a negative sustained potential (SP), maximum over fronto-central areas of the scalp, lasting for the entire duration of the stimulus.  The initial pilot study employed a 70-dB (average) SPL white noise stimulus having a total duration of 1.3 s. Four subjects were run in this study that were tested while they were awake and during a brief 30 min nap. Attention to the stimulus will affect the appearance of the SP because subjects mentally develop an expectancy that the stimulus will terminate. This expectancy is associated with a different and an independent long-lasting negativity called the (CNV). The long-lasting CNV and the long-lasting SP will summate.  During sleep, it is highly unlikely that subjects will actively maintain a mental expectation about the duration of a stimulus and thus the CNV influence should be minimal. We tried to minimize the effects of attention in the waking state by including a condition in which the subject read a book, thus ignoring the auditory stimuli. The nap consisted of mainly stage 2 NREM sleep. No subjects entered stage REM.
The grand averages of these four subjects are presented in [Figure 6]. During wakefulness, the usual N1/P2-on and small N1/P2-off transient components are clearly visible. During sleep, the N1-on did not exceed the zero baseline level but a large P2-on was elicited, findings quite consistent with the effects of NREM sleep on these components discussed earlier in this article. It is not easy to discern an N1/P2- off at the offset of the stimulus during sleep. A long-lasting SP is also apparent at the frontal electrode site during wakefulness. Surprisingly, the SP also appears to be quite large during the unconscious sleep state. A reasonable interpretation of this finding is that the SP reflects pre-conscious processing, occurring independent of consciousness. Unfortunately, there are methodological confounds in this pilot study. The onset of the stimulus also elicits a large N350, typical of NREM sleep. The N350 was not, of course, elicited in the waking condition. This negativity probably overlapped and summated with the SP. Indeed, it could be argued that the SP was not elicited at all. Rather, what appears to be the SP is actually a very slow return to baseline of the large amplitude N350. This problem will need to be overcome before the use of long duration environmental-type noise stimuli can be implemented in the ERP lab. Again, more complex ICA-type algorithms may need to be employed to disentangle these overlapping and summating potentials.
Another possible means to also overcome the problem of the N350 summation with the SP is to employ even longer duration stimuli. It is possible that the SP will continue for the entire duration of the stimulus, whereas the N350 should return to baseline level well before the offset of the stimulus. The inclusion of very long duration stimuli also has the advantage of offering considerable ecological generalization to the types of stimuli used in field studies. Little is however known about the effects of very long duration stimuli on the SP in the waking state. A good deal of basic research will thus need to be carried out before the large costs associated with sleep studies can be justified.
Another prominent feature of stimuli observed in field studies is that their intensity is rarely constant; the intensity usually gradually rises to a maximum level (as the vehicle approaches) and then gradually declines in amplitude (after it has passed). We have thus examined the processing of a stimulus that slowly increases in intensity and compared these results to those obtained when the stimulus slowly decreases in intensity. Again, there is little existing ERP literature to act as a guide for the setting of even basic stimulus parameters. Our lab has now begun pilot studies, but this time only in awake and alert subjects, examining the effects of very long rise- and very long fall-times. Again, four young adults were tested while awake. They were presented with various stimuli whose features were systematically manipulated. [Figure 7] presents the initial results from one of the conditions in which another 1.3 s duration white noise stimulus was employed. In the left portion of the figure, the initial intensity of the stimulus was set to 70 dB SPL (following a 5-ms rise-time) and was linearly ramped down to 0 dB after 1.3 s. The intensity of this stimulus thus slowly decayed (similar to what might be experienced as a vehicle passes). In the right portion, the initial intensity was 0 dB and then rose to 70 dB after 1.3 s before abruptly (5 ms fall-time) terminating (similar to what would be experienced as a vehicle approaches). Subjects either attended or ignored the auditory stimuli. The abrupt onset of the 70-dB stimulus was associated with the usual N1-P2 transient components. They were followed by a waking N2, peaking at about 300 ms that was much attenuated in the ignore condition. The SP rapidly declines in amplitude as the intensity if the stimulus decreased to 0 dB. The processing of the slowly rising stimulus was very different. A smaller amplitude, N1-P2 was visible but the peaks of both were delayed by about 50 ms. This is rather surprising given the very low intensity of the stimulus at its onset. A large amplitude waking N2 was also apparent. A long-lasting SP was elicited and lasted the entire duration of the stimulus. Some authors attribute the differential processing of stimuli that gradually increase or decrease in intensity to be a result of salience. A stimulus that slowly decreases in intensity is perceived as "fading" or "retreating" while one that slowly increases in intensity is perceived as "looming" or "advancing
These types of stimuli still remain far-removed from those employed in field studies. For example, the duration of vehicular noise may last from a few seconds in the case of construction noise and rapidly passing vehicles to as long as 30 s in the case of aircraft and freight trains. The intensity of the stimulus is perhaps the most important feature that leads to sleep disturbance and ERPs studies offer an important, in depth insight into why this is. Further, a critical factor affecting microarousals appears to be the rise-time of the stimulus rather than its total duration.  However, the interaction of both the intensity and the duration of the stimulus may also play an important role in awakening the sleeper.  Unfortunately, ERP research has not examined these critical features in detail. Limitations of ERP methodology, importantly the need for a large number of stimulus presentations in order to obtain a sufficiently large signal-to-noise in the average, will make the use of slowly presented stimuli technically difficult. Advanced modeling algorithms have however become available recently, making single trial extraction of the ERP signal feasible. Over the past decade, the use of modern high-resolution digital DC amplifiers also permits the recording of very long-lasting slow potentials in the brain. It should thus be possible to present stimuli having durations from 10 to 30 s to determine the morphology of the SP over this very long period of time. A critical gap in the ERP literature is that we know very little about the processing of external stimuli that lead either to an arousal or an awakening. Indeed most ERP studies reject trials in which an arousal or an awakening occur. Averages will need to be computed on trials in which a stimulus does elicit these marked changes in the EEG and those in which it does not. In at least the lab setting, a sufficient number of such disruptive events could be presented, similar to what is currently being done in the study of sleep fragmentation. As an example, Bangash et al.  examined the changes in a number of physiological measures caused by an auditory stimulus that resulted in sleep fragmentation. These effects are illustrated in [Figure 8] [[Figure 1] from Bangash et al., 2008]. The auditory stimulus caused changes in the EEG and these changes are clearly visible in the ongoing, raw EEG. Because of their large size, averaging of even a few trials would be feasible. Note that the auditory stimulus has different effects on blood flow and blood pressure during REM and NREM sleep. Again trials could be separated and averaged according to the different effects on the peripheral measures. The resulting muscle and eye movements associated with awakenings will certainly cause artifact in the EEG but this artifact will probably occur sufficiently long after the disrupting stimulus, thus allowing a "clean" ERP signal to appear. In [Figure 8], the changes in the peripheral measures occur sometime after those observed in the EEG.
| Conclusions|| |
ERPs provide an exquisite means to determine with millisecond precision the extent of processing of stimuli that might potentially result in microarousals or the disruption of sleep. They thus provide enormous power to predict the types of autonomic, peripheral slow changes that are a consequence of this processing. At present, ERPs are used to indicate the means by which the processing of acoustic input is inhibited and thus prevented from disrupting. This tutorial review describes three different components of the auditory ERP that may be associated with this inhibition of processing. In the waking state, the purpose of a transient detector system is to allow potentially highly relevant, obtrusive auditory stimuli arriving in a channel outside the focus of attention to interrupt the central executive controlling ongoing, current cognitive activity, forcing the observer to switch attention to this channel. The output of the transient detector system can be monitored by a negative component, N1, peaking at about 100 ms. In order to sleep, the interruption of the central executive must cease or else be prevented. Thus, at the time of definitive sleep onset and during NREM sleep, the amplitude of N1 is much reduced to near baseline level. However, the amplitude of a later positive component, P2, peaking at about 180 ms, is often much larger in NREM sleep than in the waking state. The amplitude of P2 increases as the intensity of the stimulus is increased. P2 has been associated with the need for the initial inhibition of processing thus protecting sleep from unwanted disturbance. If stimulus intensity continues to increase, two additional late negativities may be elicited during NREM sleep. These are the N350 and N550 components. The amplitude of N350 is maximum over the vertex and appears to be quite similar to the classic vertex sharp wave, traditionally a marker of sleep onset and the gating of waking consciousness. The N350 is exclusive to sleep and cannot be elicited in the waking state. N350 is elicited only for moderate and high intensity stimuli even if they are presented relatively rapidly, N350 increasing in amplitude as the intensity of the stimulus increases. The N350 is not easily elicited when the intensity of rapidly presented stimuli is quite low and thus unlikely to disturb sleep. In field settings, however, external stimuli, as measured in the home setting, are rarely loud and they occur relatively rarely. The N350 is however large for low and moderate intensity stimuli if they are presented very slowly. Highly obtrusive stimuli may also elicit the K-Complex, a phenomenon that was first described in the earliest recordings of NREM sleep. Unlike the N1-P2 and N350, the K-Complex can only be recorded in NREM sleep (and not in REM sleep and certainly not in the waking state). Most authors agree that the K-Complex includes the N350, and a later and much larger amplitude N550. The K-Complex cannot be elicited on every presentation of a physically identical stimulus. The probability of eliciting a K-Complex does however vary directly with the obtrusiveness of the stimulus and its rate of presentation. Interestingly, once the K-Complex is elicited, the amplitude of its major component, the N550 is invariant. It is thus elicited or not elicited. Many researchers view both the N350 and N550 as protective mechanisms guarding sleep against disruption by highly obtrusive, but irrelevant, stimulus input. In the absence of the K-Complex, sleep might well be vulnerable to frequent disruption.
A different picture emerges during REM sleep. An N1 component can be elicited although its amplitude is much attenuated. Nevertheless, this provides strong evidence that the transient detector system remains active and with sufficient activation, the potential to disrupt REM sleep. In the waking state, an obtrusive stimulus can interrupt the central executive resulting in a switch of attention. It is thought that a later positive component, the P3a, signals this switching of attention. During REM sleep, a relatively large number of studies have now reported a late P3-like deflection. It can however only be elicited by highly intense stimuli, but even a moderately intense stimulus can do so if it is presented very infrequently. This very slow rate of presentation however corresponds to a rate somewhat similar to that employed in field and lab studies examining the effects of environmental noise. The P3-like deflection may reflect an intrusion into consciousness during REM sleep, although the precise nature of this consciousness during REM sleep has not been determined. Very little research has been carried out with the types of stimuli that are usually observed in field studies of the effects of environmental noise. In particular, studies are needed that examine the processing of very long duration stimuli. Similarly, the effects of stimuli that have very slow rise-and-fall times need to be studied in greater detail. Modern lab and field studies are currently investigating the types of stimuli that might result in brief-lasting microarousals on one hand and in actual awakenings on the other. ERP studies are needed that investigate the prior processing associated with these different outcomes. External stimuli may also elicit different types of arousals associated with different peripheral autonomic responses. Again, ERPs could be used to determine why an external stimulus may result in, for example, a change in heart beat but little change in the EEG or alternatively a large change in the EEG. The amplitude of earlier ERP components such as the N1 and P2 are usually very small compared to the ongoing background EEG of sleep. This is improved by signal averaging techniques but the need for many repeated stimulus presentations imposes a serious limitation to its utility in the study of rarely occurring environmental events. However, the amplitude of the longer latency N350 and N550 components is very large and thus, only a few stimulus repetitions are required to allow the ERP signal to emerge from the background EEG.
| Acknowledgments|| |
Funding for this research has been provided over many years by the Natural Sciences and Engineering Research Council (NSERC) of Canada. The studies have been carried out in collaboration with many former and present students including Cιlyne Bastien, Kim Cote, Merav Sabri, Lauren Sculthorpe, Margaret Macdonald and Parastoo Jamshidi. Alexandra Muller-Gass kindly commented on an earlier draft of this article.
| References|| |
|1.||Campbell K, Bell I, Bastien C, Evoked potential measures of information processing during natural sleep. In: Broughton RJ, Ogilvie RD, editors. Sleep, arousal, and performance. Boston (MA): Birkhauser Press; 1992. p. 88-116. |
|2.||Colrain IM, Campbell KB. The use of evoked potentials in sleep research. Sleep Med Rev 2007;11:277-93. |
|3.||Campbell KB, Colrain IM. Event-related potential measures of the inhibition of information processing: II. The sleep onset period. Int J Psychophysiol 2002;46:197-214. |
|4.||Colrain IM. The K-complex: a 7-decade history. Sleep 2005;28:255-73. |
|5.||Bastien CH, Crowley KE, Colrain IM. Evoked potential components unique to non-REM sleep: relationship to evoked K-complexes and vertex sharp waves. Int J Psychophysiol 2002;46:257-74. |
|6.||Campbell KB, Bartoli EA. Human auditory evoked potentials during natural sleep: the early components. EEG Clin Neurophys 1986;65:142-9. |
|7.||Bastuji H, Garcia Larrea L, Bertrand O, Mauguiθre F. BAEP latency changes during nocturnal sleep are not correlated with sleep stages but with body temperature variations. EEG Clin Neurophys 1988;70:9-15. |
|8.||Linden RD, Campbell KB, Hamel G, Picton TW. Human auditory steady state evoked potentials during sleep. Ear Hear 1985;6:167-74. |
|9.||Erwin R, Buchwald JS. Midlatency auditory evoked responses: differential effects of sleep in the human. EEG Clin Neurophysiol 1986;65:383-92. |
|10.||Jones LA, Baxter RJ. Changes in the auditory middle latency responses during all-night sleep recording. Br J Audiol 1988;33:279-85. |
|11.||Nδδtδnen R, Picton T. The N1 wave of the human electric and magnetic response to sound: a review and an analysis of the component structure. Psychophysiology 1987;24:425-75. |
|12.||Muller-Gass A, Campbell K. Event-related potential measures of the extent of processing of infrequently presented low intensity stimuli during natural sleep. Noise Health; 2010. |
|13.||Nδδtδnen R. Attention and Brain Function. Hillsdale (NJ): Erlbaum Press; 1992. |
|14.||Cote KA, Campbell KB, 1999. P300 to high intensity stimuli during REM sleep. Clin Neurophysiol 1999;110:1345-50. |
|15.||Ogilvie RD, Simons IA, Kuderian RH, MacDonald T, Rustenburg J. Behavioral, event-related potential, and EEG/FFT changes at sleep onset. Psychophysiology 1991;28:54-64. |
|16.||Harsh J, Voss U, Hull J, Schrepfer S, Badia P. ERP and behavioral changes during the wake/sleep transition. Psychophysiology 1994;31:244-52. |
|17.||Cote KA, De Lugt DR, Campbell KB. Changes in the scalp topography of event-related potentials and behavioral responses during the sleep onset period. Psychophysiology 2002;39:29-37. |
|18.||Colrain IM, Di Parsia P, Gora J. The impact of prestimulus EEG frequency on auditory evoked potentials during sleep onset. Can J Exp Psychol 2000;54:243-54. |
|19.||Crowley KE, Colrain IM. A review of the evidence for P2 being an independent component process: age, sleep and modality. Clin Neurophysiol 2004;115:732-44. |
|20.||Cote KA, Etienne L, Campbell KB. Neurophysiological evidence for the detection of external stimuli during sleep. Sleep 2001;24:791-803. |
|21.||Macdonald M, Jamshidi P, Campbell K. Infrequent increases in stimulus intensity may interrupt central executive functioning during Rapid Eye Movement sleep. Neuroreport 2008;19:309-13. |
|22.||Makeig S, Onton J. A trial-by-trial pattern approach to event-related EEG analysis: ERP features and EEG dynamics: An ICA perspective. In Luck S, Kappenman E, editors. Oxford handbook of event-related potential components, Oxford: Oxford University Press; 2009. |
|23.||Schredl M, Atanasova D, Hφrmann K, Maurer JT, Hummel T, Stuck BA. Information processing during sleep: the effect of olfactory stimuli on dream content and dream emotions. J Sleep Res 2009;18:285-90. |
|24.||Picton TW, Hillyard SA, Krausz HI, Galambos R. Human auditory evoked potentials. I. Evaluation of components. EEG Clin Neurophysiol 1974;36:179-90. |
|25.||Colrain IM, Webster KE, Hirst G, Campbell KB. The roles of vertex sharp waves and K-complexes in the generation of N300 in auditory and respiratory-related evoked potentials during early stage 2 NREM sleep. Sleep 2000;23:97-106. |
|26.||Campbell K, Michaud DS, Keith SE, Muller-Gass A, Wiebe S. Event-related potential measures of the disruptive effects of trains of auditory stimuli during waking and sleeping states. J Sleep Res 2005;14:347-57. |
|27.||Bastien C, Campbell K. The evoked K-complex: all-or-none phenomenon? Sleep 1992;15:236-45. |
|28.||Scherg M, Vajsar J, Picton TW. A source analysis of the late human auditory evoked potentials. J Cogn Neurosci 1989;1:336-55. |
|29.||Gora J, Colrain IM, Trinder J. The investigation of K-complex and vertex sharp wave activity in response to mid-inspiratory occlusions and complete obstructions to breathing during NREM sleep. Sleep 2001;24:81-9. |
|30.||Bonnet MH, Doghramji K, Roehrs T, Stepanski EJ, Sheldon SH, Walters AS, et al. The scoring of arousal in sleep: reliability, validity, and alternatives. J Clin Sleep Med 2007;3:133-45. |
|31.||Bastien C, Campbell K. Effects of rate of tone-pip stimulation on the evoked K-Complex. J Sleep Res 1994;3:65-72. |
|32.||Loomis AL, Harvey EN, Hobart GA. Distribution of disturbance patterns in the human encephalogram, with special reference to sleep. J Neurophysiology 1939;2:413-30. |
|33.||Davis HP, Davis A, Loomis AL Harvey EN, Hobart G. Electrical reactions of the human brain to auditory stimulation during sleep. J Neurophysiology 1939;2:500-14. |
|34.||Roth M, Shaw J, Green J. The form voltage distribution and physiological significance of the K-complex. EEG Clin Neurophysiol 1956;8:385-402. |
|35.||Colrain IM, Crowley K. Evoked potentials during non-REM sleep: utility and functional significance. In Guilleminault C, editor, Clinical neurophysiology of sleep disorders. Elsevier: Amsterdam; 2005. p. 125-135. |
|36.||Crowley K, Trinder J, Colrain IM. An examination of evoked K-complex amplitude and frequency of occurrence in the elderly. J Sleep Res 2002;11:129-40. |
|37.||Nicholas, CL, Trinder J, Colrain IM. The impact of slow wave sleep proximity on evoked K-complex generation. Neuroscience Letters 2006;104:127-31. |
|38.||Bastien CH, Ladouceur C, Campbell KB. EEG characteristics prior to and following the evoked K-Complex. Can J Exp Psychol 2000;54:255-65. |
|39.||Halαsz P. K-complex, a reactive EEG graphoelement of NREM sleep: an old chap in a new garment. Sleep Med Rev 2005;9:391-412. |
|40.||Peszka J, Harsh J. Effect of sleep deprivation on NREM sleep ERPs and related activity at sleep onset. Int J Psychophysiol 2002;46:275-86. |
|41.||Nicholas CL, Sullivan EV, Pfefferbaum A, Trinder J, Colrain IM. The effects of alcoholism on auditory evoked potentials during sleep. J Sleep Res 2002;11:247-53. |
|42.||Melendres MC, Marcus CL, Abi-Raad RF, Trescher WH, Lutz JM, Colrain IM. Respiratory-related evoked potentials during sleep in children. Sleep 2008;31:55-61. |
|43.||Bruck D, Ball M, Thomas I, Rouillard V. How does the pitch and pattern of a signal affect auditory arousal thresholds? J Sleep Res 2009;18:196-203. |
|44.||Picton TW, Woods DL, Proulx GB. Human auditory sustained potentials. EEG Clin Neurophysiol 1978;197:186-97. |
|45.||Campbell K, Herzig A, Jashmidi P. The extent of active processing of a long-duration stimulus modulates the scalp-recorded sustained potential. Brain Cogn 2009;69:170-5. |
|46.||Bach DR, Schδchinger H, Neuhoff JG, Esposito F, Di Salle F, Lehmann C, et al. Rising sound intensity: an intrinsic warning cue activating the amygdala. Cereb Cortex 2008;18:145-50. |
|47.||Saremi M, Grenθche J, Bonnefond A, Rohmer O, Eschenlauer A, Tassi P. Effects of nocturnal railway noise on sleep fragmentation in young and middle-aged subjects as a function of type of train and sound level. Int J Psychophysiol 2008;70:184-91. |
|48.||Bangash MF, Xie A, Skatrud JB, Reichmuth KJ, Barczi SR, Morgan BJ. Cerebrovascular response to arousal from NREM and REM sleep. Sleep 2008;31:321-7. |
School of Psychology, University of Ottawa, Ottawa, ON, K1N 6N5
Source of Support: Research Grants from Natural Sciences and Engineering Research Council of Canada (NSERC)., Conflict of Interest: None
[Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6], [Figure 7], [Figure 8]
|This article has been cited by|
||Discriminating neurological from psychiatric hypersomnia using the forced awakening test
| ||L. Peter-Derex,F. Perrin,T. Petitjean,L. Garcia-Larrea,H. Bastuji |
| ||Neurophysiologie Clinique/Clinical Neurophysiology. 2013; 43(3): 171 |
|[Pubmed] | [DOI]|
||Discriminating neurological from psychiatric hypersomnia using the forced awakening test
| ||Peter-Derex, L. and Perrin, F. and Petitjean, T. and Garcia-Larrea, L. and Bastuji, H. |
| ||Neurophysiologie Clinique. 2013; 43(3): 171-179 |
||Working Memory Is Partially Preserved during Sleep
| ||Daltrozzo, J. and Claude, L. and Tillmann, B. and Bastuji, H. and Perrin, F. |
| ||PLoS ONE. 2012; 7(12) |
||Modelling of auditory evoked potentials of human sleep-wake states
| ||Coenen, A. |
| ||International Journal of Psychophysiology. 2012; 85(1): 37-40 |
||To wake or not to wake? The two-sided nature of the human K-complex
| ||Jahnke, K. and von Wegner, F. and Morzelewski, A. and Borisov, S. and Maischein, M. and Steinmetz, H. and Laufs, H. |
| ||NeuroImage. 2012; 59(2): 1631-1638 |
||To wake or not to wake? The two-sided nature of the human K-complex
| ||Kolja Jahnke,Frederic von Wegner,Astrid Morzelewski,Sergey Borisov,Marcella Maischein,Helmuth Steinmetz,Helmut Laufs |
| ||NeuroImage. 2012; 59(2): 1631 |
|[Pubmed] | [DOI]|
||Modelling of auditory evoked potentials of human sleep–wake states
| ||Anton Coenen |
| ||International Journal of Psychophysiology. 2012; 85(1): 37 |
|[Pubmed] | [DOI]|
||Towards gene therapy for deafness
| ||Marina Di Domenico,Carmela Ricciardi,Tiziana Martone,Nicoletta Mazzarella,Claudia Cassandro,Giuseppe Chiarella,Luigi DæAngelo,Ettore Cassandro |
| ||Journal of Cellular Physiology. 2011; 226(10): 2494 |
|[Pubmed] | [DOI]|
||Towards gene therapy for deafness
| ||Di Domenico, M. and Ricciardi, C. and Martone, T. and Mazzarella, N. and Cassandro, C. and Chiarella, G. and DæAngelo, L. and Cassandro, E. |
| ||Journal of Cellular Physiology. 2011; 226(10): 2494-2499 |