Preferred EEG brain states at stimulus onset in a fixed interstimulus interval auditory oddball task, and their effects on ERP components

https://doi.org/10.1016/S0167-8760(02)00151-4Get rights and content

Abstract

Previous work has indicated the importance of ongoing EEG activity in the elicitation of the event-related potential (ERP), supporting the conceptualisation of the ERP in terms of amplification and attenuation of component frequencies in the EEG. We investigated the importance of the phase of narrow-band EEG activity in generating N1 and P2 components in the auditory ERP. An auditory oddball paradigm requiring a button-press response to targets, with fixed interstimulus interval (ISI) and 15% target probability, was utilised. The continuous EEG at Cz was recorded from 16 subjects as the raw data set. Offline digital filtering was used to separate the EEG into 13 narrow bands from 1 to 13 Hz. For each band, the phase at the onset of each non-target stimulus was determined. These were used to sub-average the unfiltered data stream at each of four phases for each of 13 frequencies for each subject. Phase effects were examined in terms of two orthogonal dimensions of electrical brain activity: Cortical negativity and negative driving. Stimulus onset varied as a function of these dimensions in a non-random fashion across frequency, indicating the preferential occurrence of particular phases, interpretable as preferred brain states. Large differential effects were also apparent in N1 and P2 amplitudes. These data indicate important aspects of brain dynamics, suggesting that in a fixed-ISI paradigm the component frequencies of the EEG are dynamically adjusted in order to provide particular brain states at stimulus occurrence to facilitate the brain's processing of the stimulus.

Introduction

The event-related potential (ERP) elicited by a stimulus is actualised as the averaged time-locked electrophysiological response to repeated presentations of that stimulus. Averaging is necessary to increase the signal:noise ratio of the small stimulus-related response hidden in the background EEG activity, by removing the random EEG ‘noise’. There are a number of problems with this simple conceptualisation, which may be overlooked in common discussion. The first of these is with the concept of the ERP itself. This ERP has no concrete existence outside the average which formed it, yet it sometimes appears to be conceptualised as the response to each stimulus. Perhaps the average does not provide a good estimate of the trial-to-trial response, which could well be highly variable (e.g. Anderson et al., 1991, Ford et al., 1994). The second problem is with the concept of the EEG itself. If the ongoing EEG is considered as background ‘noise’, and hence a problem to be eliminated in the averaging process, the possibility that the ongoing EEG plays a fundamental role in the genesis of the ERP may be overlooked.

An alternative conceptualisation is to explicitly recognise that the average ERP is just that—an average—and that this average says little about the brain's response to any individual stimulus. The responses to individual stimuli may vary widely, perhaps largely depending on the nature of the EEG at stimulus onset. At the cellular level, this has been demonstrated by Arieli et al. (1996). Using visual stimulation in cat, these authors concluded that ‘the processing of sensory input in the visual cortex involves the combination of a deterministic response and ongoing network dynamics’ (p. 1870). Further, Makeig et al. (2002) recently described the human ERP to non-target stimuli in a visual selective-attention task in terms of its generation by ‘partial stimulus-induced phase-resetting of multiple electroencephalographic processes’ (p. 690). In this alternative conceptualisation, the ‘background’ EEG activity is seen as exquisitely reflecting the brain's momentary state of activity, and subtly determining the response to the stimulus, as reflected in both behavioural and ERP outcomes (Başar, 1980). Supporting evidence for this approach is provided by studies of the effect of stimulus presentation in particular brain states. For example, there has been a substantial number of studies examining the effect of differences in the amplitude of particular EEG frequencies in the prestimulus period upon ERP components (e.g. Barry et al., 2000, Başar and Stampfer, 1985, Başar et al., 1989, Başar et al., 1998, Brandt, 1997, Brandt and Jansen, 1991, Brandt et al., 1991, Jasiukaitis and Hakerem, 1998, Price, 1997, Rahn and Başar, 1993).

Perhaps the most widely studied frequency range in these reports has been the alpha band. In a recent study, Barry et al. (2000) concluded that amplitudes of both the N1P2 and N2P3 complexes, elicited by targets in an auditory oddball paradigm, were directly proportional to prestimulus RMS alpha amplitude—i.e. these components are larger when stimuli are presented during high-amplitude alpha states. These data supported the previous findings of Brandt's group with N1P2, and Jasiukaitis and Hakerem (1998) with P3. Barry et al. (2000) noted that contradictory reports of enhanced ERP components with lower levels of prestimulus alpha, and vice-versa, often appeared to be associated with paradigms in which stimuli were selectively presented when subjects were in particular states (e.g. Başar et al., 1998, Price, 1997, Rahn and Başar, 1993). Such paradigms are markedly different from common ERP studies, which instead present stimuli on a scheduled basis independently of fluctuations in the momentary state of the subject. It is the effect of prestimulus EEG state in the latter type of experiment which constitutes the present focus of our work.

Although the relationships between prestimulus alpha and the N1 and P3 ERP peaks have been widely explored, a few other EEG frequency/ERP component linkages have also been examined. For example, Stampfer and Başar (1985) concluded that the ‘N100 peak is formed mainly by 3.5–8 Hz, and 8–13 Hz activity; the P200 peak is formed by the superposition of the first peak of 1–3.5 Hz activity and the first positive peak of 3.5–8 Hz activity’ (p. 187). Başar et al. (1984) concluded that ‘the endogenous N100 wave is related to the alpha activity, and… waves with latencies higher than 300 ms are related to activities in the theta and delta frequency range’ (p. 19). Kolev and Schürmann (1992) confirmed the important role of theta and alpha oscillations in the formation of ERPs in a non-response single-stimulus paradigm, essentially confirming the dependence of the components of the N1P2 complex on modulation of these frequencies. This appears to be compatible with a study by Brandt et al. (1991), which reported that N1P2 amplitudes produced by visual stimuli presented with a quasi-random interstimulus interval (ISI) were directly related to prestimulus alpha amplitude, and inversely related to prestimulus delta and theta amplitudes. There was no relation with beta activity. Over all bands, prestimulus power predicted some 45% of the total variance in N1P2 amplitude. Başar-Eroglu et al. (1992) and Demiralp and Başar (1992) explored this further, noting that delta and theta activity seemed important in relation to late ERP components in the oddball task (presumably the P3) while theta activity appeared important in relation to early components (presumably the N1P2 complex) in a stimulus omission study. Başar et al. (1998) discussed earlier in relation to alpha activity, also explored the impact of prestimulus theta in elicitation of the visual evoked potential. Yordanova and Kolev, 1996, Yordanova and Kolev, 1997 and Yordanova et al. (1996) have reported further work on the development of the theta-P3 relationship with children and adults. That is, studies to date have found that activity in the delta, theta, and alpha EEG frequency ranges affect ERP components, but frequencies beyond the alpha range appear to have little direct involvement in the generation of the major ERP components.

Embedded in many of the reports cited above, particularly those from the Başar group, is the suggestion that with regular stimuli, phase adjustments occur in the prestimulus EEG. For example, Başar and Stampfer (1985) reported that, in a fixed-ISI paradigm with alternating target and non-target stimuli, the delta activity before target stimuli shows ‘evidence of a stable phase-reordering or ‘preferred phase angle’, at the point of stimulation. In the examples shown, there is evidence of a maximum negativity at the time of stimulation’ (p. 167). In the same study, alpha activity was also ‘associated with increasing phase alignment at the point of stimulation’ (p. 170), and a figure suggests that this also involved a preferential occurrence of EEG negativity; further, the subsequent alpha peak appears to largely determine the timing of the N1 component. Such observations led them to conclude that ‘a regular pattern of stimulation can induce a ‘preferred’ phase angle, which appears to facilitate an optimal brain response to the sensory input’ (p. 175). Similar findings were presented by Başar et al. (1984) in a study which presented tones at 1 s ISIs and omitted every third or fourth tone, leading them to state that ‘phase-reordering at the time of stimulus omission is due to the behavioural state resulting from the application of repetitive stimuli’ (p. 9).

The importance of such a preferred phase appears to arise from its impact on functional aspects of brain state and behaviour. For example, Rémond (1969) demonstrated that the phase of alpha activity at stimulus onset was important in determining the amplitude and topography of the subsequent alpha activity. Other early work (e.g. Callaway and Yeager, 1960, Dustman and Beck, 1965, Trimble and Potts, 1975) concentrated on the differential effects of stimulation at negative and positive peaks of the alpha cycle. Generally, maximum excitability occurred when stimuli were presented at negative peaks, an effect apparent in both ERP components and reaction time measures. Such findings appear to interleave with other work which links cortical negativity with increased brain excitability (e.g. Pleydell-Pearce, 1994, Rockstroh et al., 1989). Together they suggest that, at least within fixed-ISI paradigms, the brain reorders the phase of its alpha activity, and perhaps other frequencies, to preferentially display negativity at stimulus onset, and that this optimises its excitability and hence affects behavioural and ERP outcomes.

Jansen and Brandt (1991) investigated the effects of alpha phase using visual stimulation presented at irregular intervals, with no task requirements. They used 1000 stimuli with ISIs varying from 2 to 6 s, and divided the prestimulus alpha cycle into eight phases. Generally, alpha activity was found to continue into the poststimulus period, and the first or second negative peak was identified with the N1. Its amplitude and latency varied substantially with alpha phase at stimulus onset. N1 was found to be maximum when stimulus onset was near a positive-going zero crossing. Smaller effects were obtained in relation to the P2, which was considered to be related to partial blocking of the alpha activity. Unfortunately, this interesting study presented data from only 5 subjects, and it is unclear how these individual results can be generalised. Also, the study employed an eyes-closed condition, which, together with the absence of a task, would be expected to result in relatively high alpha amplitudes. Phase determination was then restricted to selected 1 s prestimulus epochs with high alpha level (>100 μV). This high level of alpha activity further limits the generalisability to common ERP paradigms. More importantly, their phase determination was based on peak and trough identification in the EEG (bandwidth 0.032–35 Hz, which was ‘smoothed’ with an unspecified filter), and trials were accepted if the period of the cycle used to determine phase corresponded to a frequency between 5 and 20 Hz. This extension beyond the normal alpha limits makes interpretation of the results extremely problematic.

More recently, Brandt (1997) confirmed that poststimulus filtered alpha was enhanced when stimuli were presented in the half-cycle associated with a positive rather than a negative zero-crossing. This means that stimuli presented in the half cycle between a prestimulus negative peak and the following positive peak were most influential in generating an enhanced negativity approximately 100 ms poststimulus onset, activity presumed to underly the N1. Unfortunately, Brandt examined these poststimulus effects only in terms of the alpha frequency range rather than the broad-band ERP itself. Thus, while these studies have served to illustrate that EEG phase at stimulus onset is an important characteristic influencing the brain's response, a systematic exploration of the effects of phase is lacking.

One of the impediments to research investigating phase effects is the difficulty of conceptualising, measuring, and analysing the phase of a frequency component as an independent variable in the experimental paradigm. For example, Brandt (1997) used the approach of Winfree (1987) to investigate the occurrence of phase-resetting at stimulus onset. This is a highly mathematical approach which appears useful in the detection and description of the phenomenon of phase-resetting rather than an understanding of its functional role in ERP and behavioural outcomes. In another approach, Haig and Gordon (1998b) introduced the use of angular or circular statistics to handle phase measures, and used this approach to relate ERP outcomes to prestimulus alpha phase synchronicity, defined in terms of reduced phase variability across five occipito-parietal sites. Larger N100s and shorter reaction times were obtained when prestimulus alpha phase synchronicity was high. Unfortunately, this measure appears to be confounded with alpha amplitude, in that highly synchronised alpha in the occipito-parietal region is likely to be of high amplitude. It is thus unclear whether a measure of phase synchronicity has potential value independently of prestimulus amplitude or power measures. In a later study further exploring the use of circular statistics in this field, Haig and Gordon (1998a) established, in a fixed-ISI auditory oddball paradigm, that single-trial target ERPs which had a detectable P3 component differed from those which did not, in terms of the mean alpha phase at stimulus onset. Some three in four trials had P3s, and their mean phase at stimulus onset was late in the period Brandt (1997) referred to as ‘positive zero crossing’, while the one in four trials without P3s had mean phase at stimulus onset earlier in the same period. However, their topographic analysis used averages with and without P3s, and their corresponding mean alpha phases at stimulus onset, which were defined independently at each site, making it difficult to speculate on any underlying mechanisms.

The current work adopted a different approach to the measurement of phase, in an effort to facilitate the relating of phase measures to a wider range of concepts which appear useful in this arena. Fig. 1 shows a simple sine wave, drawn with negative up in order to relate to common ERP conventions, with 4 phase divisions at intervals of π/2. Phases B and C subdivide Brandt's (1997) ‘positive zero crossing’ period, while D and A correspond to his ‘negative zero crossing’ period. Haig and Gordon's (1998a) phase measure was defined in terms of 0° at a positive peak and 180° at a negative peak, so that Phase A corresponds to their range of 90–180°. Rather than work with such mathematically defined divisions as independent variables, an attempt was made here to relate them to more intuitively meaningful physical concepts, so that their effects on ERP measures could be better explored. Thus, this study investigated two easily conceptualised physical variables. Cortical negativity can be examined in the comparison of the effects of (A+B) vs. (C+D). This represents perhaps the earliest aspect of research into phase effects, investigated by presenting stimuli at the negative or positive peaks of alpha waves. A second physical variable, negative driving, can be examined in the comparison of (A+D) vs. (B+C). This relates to the nature of the changing of the cortical negativity variable—is it increasing (phases A and D) or decreasing (phases B and C)? Negative driving covers Brandt's ‘negative zero crossing’ vs. ‘positive zero crossing’ effects. Note that both (A+D) and (B+C) cover negative and positive phases, and hence effects of negative driving are independent of the effects of cortical negativity.

Although some of these studies have used Fast Fourier Transforms (FFTs) to obtain spectral information regarding prestimulus EEG (e.g. Brandt and Jansen, 1991), many have employed band-pass filtering of the EEG on a trial-by-trial basis. These have used either traditional band definitions (e.g. 8–13 Hz alpha: Barry et al., 2000, Brandt, 1997, Haig and Gordon, 1998a, Haig and Gordon, 1998b) or closely related bands derived from peaks in the individual's amplitude–frequency characteristic curve (e.g. Başar et al., 1984). The present study aimed to pursue this issue with very narrow-band (1 Hz bandwidth) digital filtering. This was an attempt to pass beyond the limits of the traditional EEG bands, which may constrain the interpretation of the EEG/ERP relationship. One potential problem with any narrow-band filtering approach is ‘ringing’ in the filter. This refers to an unplanned and unwanted high level of amplification which may occur with frequencies near the turning points of the filter. An ideal filter has an amplification of 1 in the pass band, and 0 beyond. Real filters have amplification ripples in the pass band, which may be relatively large near the cut-off frequencies, leading to ‘ringing’. It was suggested by Brandt and Jansen (1991) and Barry et al. (2000) that some of the conflicting data apparent with the examination of alpha frequency effects using digital filtering may have been caused by ‘ringing’. Barry et al. (2000) specifically checked their alpha filter characteristics with white noise to ensure that ‘ringing’ was not a problem in that study.

Thus the present study investigated the occurrence of different phases of EEG activity at stimulus onset in a fixed-ISI auditory oddball task, at frequencies up to, and including, the alpha range. We also examined their effects on the amplitude of ERP components. Because we sub-averaged ERPs in terms of phase at stimulus onset, only the responses to the more-numerous background stimuli in the oddball task could be examined. This effectively limited analysis to the N1 and P2 components. Effects in these components were derived by examining narrow-band filtered EEG, in 1 Hz bands, in terms of the four phase periods defined above, and analysing the occurrence of these brain states and the subsequent ERP outcomes in terms of the different combinations of phases representing variation in the variables of cortical negativity and negative driving defined above.

Section snippets

Subjects

Sixteen undergraduate subjects (11 females, 5 males), aged between 19 and 47 years, participated in this study as one means of satisfying a course requirement in Psychology. All claimed normal hearing, and gave written informed consent in accordance with a protocol approved by the University of Wollongong Human Research Ethics Committee.

Procedure

Continuous EEG was recorded from 17 scalp sites using an electrode cap with tin electrodes, referenced to linked ears, with a gain of 50 000, a time constant of

Results

The number of ERP responses accepted from the 204 background stimuli ranged across the 16 subjects from 60 to 140, with a mean of 102.2 (S.D.=18.6). The mean ERP to these background stimuli, averaged across subjects, is shown in Fig. 2. Prior to stimulus onset, a CNV of some −1.7 μV is apparent, developing substantially in the 200 ms immediately before stimulus onset. There is a marked N1, with amplitude of −6.3 μV and mean latency approximately 90 ms, followed by a clear P2 with amplitude 2.2

Preferential brain states

Across the frequency range explored, there were marked differences in the occurrence of different phases at stimulus onset. Cortical negativity occurred preferentially in both the delta (from 1 to 3 Hz, but reaching significance at 2 Hz) and alpha (at 10–11 Hz) bands, with the opposite effect apparent in the 4–6 Hz theta range. The size of this last effect is particularly substantial, with cortical negativity occurring in only 1/3 of the epochs. The preferential occurrence of cortical negativity

References (33)

  • J. Yordanova et al.

    Developmental changes in the event-related EEG theta response and P300

    EEG Clin. Neurophysiol.

    (1997)
  • A. Arieli et al.

    Dynamics of ongoing activity: explanation of the large variability in evoked cortical responses

    Science

    (1996)
  • E. Başar

    EEG-Brain Dynamics. Relation between EEG and Brain Evoked Potentials

    (1980)
  • E. Başar et al.

    Important associations among EEG-dynamics, event-related potentials, short-term memory and learning

    Int. J. Neurosci.

    (1985)
  • E. Başar et al.

    A new approach to endogenous event-related potentials in man: relation between EEG and P300-wave

    Int. J. Neurosci.

    (1984)
  • E. Başar et al.

    The EEG is a quasi-deterministic signal anticipating sensory-cognitive tasks

  • Cited by (0)

    View full text