Abstract

The large-scale neural dynamics underlying higher cognitive processes are characterized by at least three types of stimulus-response: (i) the resetting of ongoing oscillatory brain activity without concomitant changes in response amplitude (phase alignment response); (ii) the addition of response amplitude to the ongoing brain activity in a time-locked manner (evoked response); and (iii) the addition of response amplitude that is not time-locked (induced response). Recent animal studies identified evoked responses as a characteristic neural response during stimulus perception but leave open the possibility that higher cognition, such as memory, is characterized more predominantly by phase alignment and/or induced responses. Using whole-head single-trial magnetoencephalography data from eight healthy adults, we show that all three types of response are related to the discrimination of old and new stimuli in a visual word recognition memory paradigm. In four subjects, single-trial evoked responses were the single constituents of event-related field old/new differences that have been previously related to familiarity-based and recollection-based recognition memory. While these data show that the oscillatory brain dynamics underlying recognition memory are characterized by a complex mix of three types of stimulus-response, they also clearly implicate evoked responses in higher cognitive processes such as recognition memory.

Introduction

The recent resurgence of interest in the oscillatory behaviour of neural networks has been motivated, in part, by the improved understanding of how single neurons are embedded into neural mass action via oscillatory phenomena (Steriade, 2000; Logothetis, 2003; Buzsaki and Draguhn, 2004). Oscillatory rhythms are thought to coordinate the precise timing of neurons in large-scale neural networks and thereby influence perception, representation and long-term coding of information (Steriade, 2000; Lisman and Otmakhova, 2001; Mehta et al., 2002; Logothetis, 2003; Buzsaki and Draguhn, 2004). Indeed, oscillatory rhythms, such as theta and gamma rhythms have dominated the animal research of large-scale neural responses related to memory (Lisman and Otmakhova, 2001; Mehta et al., 2002) and are viewed as the ‘middle ground’ linking single neuron activity to behaviour (Buzsaki and Draguhn, 2004).

With the increased interest in oscillatory behaviour (see also Penny et al., 2002), the electrophysiological research of human cognition has been somewhat polarized into theories that emphasize the evoked nature of neural responses to cognitive acts such as perception and attention (Jervis et al., 1983; Hillyard, 1985, 1987; Vaughan and Arezzo, 1988; Mangun, 1992; Schroeder et al., 1995; Shah et al., 2004), and theories that favour the dynamic, oscillatory nature of neural responses (Sayers et al., 1974; Makeig et al., 2002; Jansen et al., 2003).

The evoked response view states that a stimulus evokes an additive neural population response in each single trial and, because this response is time-locked to the onset of stimulus presentation, averaging these evoked responses across trials will preserve them in the so-called event-related potential, or ERP (Jervis et al., 1983; Hillyard, 1985, 1987; Vaughan and Arezzo, 1988; Mangun, 1992; Schroeder et al., 1995; Shah et al., 2004). The dynamic response view states that a stimulus induces ‘phase resetting’, or phase alignment of ongoing oscillatory electroencephalographic (EEG) rhythms in each trial, and that averaging these phase-coherent rhythms produces the ERP (Sayers et al., 1974; Makeig et al., 2002; Jansen et al., 2003).

A recent animal study used intracortical recordings to demonstrate that neural activity in early visual areas associated with stimulus perception is clearly dominated by evoked responses (Shah et al., 2004). The same study left open the possibility that in upstream visual areas of the temporal cortex, higher cognitive processes such as attention and memory could be characterized more predominantly by phase alignment of ongoing oscillations and/or by induced responses. A argument in favour of this assumption was an increase in ongoing — that is, pre-stimulus — oscillatory activity in the inferior temporal cortex when compared with the early visual area V1 (Shah et al., 2004).

A cognitive faculty that is critically dependent on the integrity of the anterior inferior temporal cortex is visual recognition memory (Mishkin et al., 1998). We therefore used whole-head single-trial magnetoencephalography (MEG) data from eight healthy adults to study the contribution of evoked responses and phase alignment to visual recognition memory. ERP studies of visual recognition memory have consistently identified three types of differences between recognized ‘old’ stimuli and correctly rejected ‘new’ stimuli (e.g. Duzel et al., 2003): an early, mainly occipital, difference at 200 and 300 ms after stimulus onset; a mid-latency difference at 400 and 500 ms in the anterior temporal lobe; and a late response at 500 and 700 ms, which is more widely distributed across temporal, parietal and frontal cortices. Of these, the early response is most likely related to an implicit form of memory. The mid-latency response, referred to as the N400-effect by some authors (henceforth called the M400 effect), and the late response, referred to as the late positive component, or LPC-effect (henceforth called the late component, or LC, effect), have been related to familiarity-based and recollection-based recognition, respectively, and are most likely those ERP components of recognition memory that critically depend on the integrity of the anterior temporal lobe (Grunwald et al., 1998; Duzel et al., 2003; Rugg and Yonelinas, 2003).

In addition to phase alignment responses and evoked responses, we also assessed the relationship between ERPs and a third type of neural response, termed induced response (Tallon-Baudry and Bertrand, 1999; Duzel et al., 2003). Induced responses are characterized by addition of response amplitude to the ongoing brain rhythm, but this increase is not time-locked to stimulus presentation (Tallon-Baudry and Bertrand, 1999). Recent evidence suggests that the neural populations that show induced responses during recognition memory might differ from those that generate the ERP effects of recognition memory (Klimesch, 2000).

Material and Methods

The MEG data were recorded as part of a previously published combined MEG and electroencephalography (EEG) study of word recognition memory (Duzel et al., 2003). In the previous publication we had reported the brain sources of the event-related field (ERF, based on MEG) and ERP (based on EEG) responses that distinguished hits and correct rejections. We had also reported oscillatory amplitude differences (but not phase alignment differences) between hits and correct rejections in a frequency band ranging from 4.5 to 42 Hz. Here, we have directly assessed the relationship between ERFs and oscillations by analysing amplitude and phase alignment of oscillations as low as 1 Hz, using a frequency analyses with a higher time resolution than before and using new analysis techniques. Data from eight of the 11 subjects (data from the other three subjects were not available) were considered for the present study.

Subjects

Eight healthy native speakers of German (six women, age range 20–31 years) had volunteered for paid participation in the experiment. All gave written informed consent to participate, were right-handed according to self report, and had normal or corrected to normal vision. The experiment was carried out in accordance with the guidelines of the ethics committee of the Faculty of Medicine, University of Magdeburg.

Experimental Design

The experimental design has been described by Duzel et al. (2003). The experiment was divided into 30 blocks, each consisting of a study and a test phase. During each study phase, 12 words (Celex Database; Baayen et al., 1993), mean word frequency 46, 4–9 letters per word) were presented visually with an inter-stimulus interval of 2 s and a stimulus duration of 1.2 s. Subjects were instructed to make a pleasant–unpleasant judgement on each study word. The test lists comprised 12 studied (‘old’) words and 12 unstudied (‘new’) words, presented in random order every 2 s. Subjects were instructed to indicate whether they had seen a word during the study phase or not. They were instructed to press the index finger for old words and the middle finger for new words. Subjects were told to carry out all responses with their right hand during the first 15 study–test blocks and with their left hand during the second 15 blocks.

Data Recording and Analysis

MEG (148-channel BTi Magnes 2500 whole-head magnetometer, Biomagnetic Technologies Inc., San Diego, CA) signals were recorded with a digitization rate of 254 Hz and filtered (IIR Butterworth filter) with a band pass from DC to 50 Hz. Single epochs of MEG data ranged from 200 ms pre-stimulus to 1800 ms post-stimulus. After artefact rejection, the same single-trial epochs for correctly recognized old words (hits) and correctly identified new words (correct rejections) were used to derive ERFs and to conduct single-trial frequency analyses. The mean number of trials per subject was 220 for hits and 230 for correct rejections. ERFs were derived by averaging single trials.

Single-trial Frequency Analyses

Amplitude and phase alignment of designated frequencies were assessed separately for hits and correct rejections. A continuous wavelet transformation was applied to single trials of MEG from each sensor, using Morlet wavelets.

Morlet wavelet defined as 

\[w_{a,s}(t_{k}){=}\sqrt{a^{{-}1}}e^{2{\pi}i\frac{t_{k}}{a}}e^{{-}\frac{1}{2}\left(\frac{t_{k}}{as}\right)^{2}}\]
where

  • \(t_{k}{=}\frac{k}{r},\)
    k = −n, …, 0, …, n timevector

  • n = 6asr (defines length of timevector)

  • r: sampling-rate in Hz

  • s: stretch-factor (increasing s decreases time-resolution)

  • a: scale in 1/Hz

Two types of Morlet wavelets were used (Fig. 1). A ‘standard’ Morlet wavelet with stretch factor ‘stretch = 1’ (Duzel et al., 2003) and a Morlet wavelet was more strongly dampened (stretch = 1/3) in order to improve the time resolution of the wavelet transform (at the sacrifice of some frequency resolution). Taking the frequency resolution of our ‘standard’ Morlet wavelet into account, 27 logarithmically distributed wavelets optimally covered our frequency range of interest from 1.0 to 45.0 Hz. Each of these 27 wavelets was convolved with the MEG signal of each trial, separately for the ‘standard’ and the ‘dampened’ wavelets. During transformation, a normalization factor assured that a signal with a maximum amplitude of 1 resulted in a transform with maximum amplitude of 1. The modulus of the resulting time–frequency coefficient matrix denotes absolute amplitude, whereas the inverse tangent of its imaginary-to-real part ratio denotes phase. Amplitude and phase were averaged separately across trials for each time point. Phase alignment for each time point was measured according to the ‘phase-locking factor’ of Tallon-Baudry et al. (1996) as the length of the unit phase vector across trials divided by the number of trials. A value of 1 would correspond to perfect phase alignment across trials and a value of 0 to random phase variation across trials.

Figure 1.

Event-related field (ERF) differences between hits and correct rejections. The t value maps in the first row display the topography of significant ERF differences between hits and correct rejections in three time windows (early: 290 ms; M400: 410 ms; LC: 510 ms). The grand mean ERF waveforms below the maps are displayed for the sensors where ERF differences were most prominent in the early time window, the M400 time window and the LC time window. The sensors were left occipital sensor 81, the left temporal sensor 114 and the left parietal sensor 19, and their locations are illustrated on the maps as white circles. The sensor locations are displayed on a 3-D drawing of sensor positions. The time–frequency spectra in the third row show the grand mean (eight subjects) frequency composition of the ERF differences between hits and correct rejections. Black lines on the time–frequency spectra illustrate the timing of significant differences (serial t-tests, P < 0.05) between ERFs to hits and correct rejections. The bottom row shows the two types of Morlet wavelets used for frequency analysis.

Figure 1.

Event-related field (ERF) differences between hits and correct rejections. The t value maps in the first row display the topography of significant ERF differences between hits and correct rejections in three time windows (early: 290 ms; M400: 410 ms; LC: 510 ms). The grand mean ERF waveforms below the maps are displayed for the sensors where ERF differences were most prominent in the early time window, the M400 time window and the LC time window. The sensors were left occipital sensor 81, the left temporal sensor 114 and the left parietal sensor 19, and their locations are illustrated on the maps as white circles. The sensor locations are displayed on a 3-D drawing of sensor positions. The time–frequency spectra in the third row show the grand mean (eight subjects) frequency composition of the ERF differences between hits and correct rejections. Black lines on the time–frequency spectra illustrate the timing of significant differences (serial t-tests, P < 0.05) between ERFs to hits and correct rejections. The bottom row shows the two types of Morlet wavelets used for frequency analysis.

ERF Frequency Analyses

We had previously reported that ERF differences between hits and correct rejections were observed in an early time window between 200 and 400 ms, a mid-latency time window between 400 and 500 ms (the M400 time window), and a late time window between 500 and 700 ms (the LC time window) (Duzel et al., 2003). To determine the dominant frequencies of the ERF differences in these time windows, we selected three sensors on the basis of the grand mean ERFs which showed a maximum ERF difference between hits and correct rejections in one the three time windows. Single-subject ERFs from these sensors, and three sensors neighbouring each of them, were then submitted to a continuous wavelet transformation using the 27 ‘dampened’ wavelets.

Relationship between Single Trial Amplitude and Phase Alignment

To visualize the relationship between single trial amplitude and phase alignment across trials, single subject data were visualized by plotting the single trial complex time-frequency coefficients separately for hits and correct rejections. As illustrated schematically in Figure 4, a concentration or clustering of single trial data to certain phase values would indicate phase alignment, while the eccentricity of the data from the origin would indicate amplitude increase. A pure increase in eccentricity without phase concentration would indicate a pure induced response. A pure increase in concentration without increase in eccentricity would indicate a pure phase alignment response. A concurrent, overlapping increase of concentration and eccentricity would indicate an evoked response, because it would show that increase in amplitude occurs where there is phase concentration. If concentration and eccentricity do not overlap, this would indicate that single trial phase alignment and amplitude increase are not related to each other and can occur independently across trials.

Statistical Analyses

To assess the relationship between ERF effects of recognition memory and single trial analyses of oscillations, between-subjects (Figs 2 and 3) and within-subjects (Figs 4 and 5) analyses of single-trial amplitude values and phase alignment values were conducted. These between-subjects and within-subjects analyses were conducted at sensors that showed the maximum left hemispheric ERF differences between hits and correct rejections. The within-subjects analyses were additionally conducted at sensors neighbouring these sensors to assess the reliability of findings.

Figure 2.

Single trial analyses of amplitude and phase alignment differences between hits and correct rejections at left occipital sensor 81. The time frequency spectra show the grand mean average (eight subjects) of single trial amplitude differences (in fT) between hits and correct rejections at left occipital sensor 81. Time frequency spectra in the upper row were derived with ‘standard’ wavelets while time frequency spectra in the lower row were derived with a more strongly ‘dampened’ wavelet. The two types of wavelets differ in their time/frequency resolution, with the upper row having a higher frequency resolution and the lower row a higher time resolution. Results from both analyses show a single trial amplitude difference between hits and correct rejections between 200 and 500 ms at 4–5 Hz. In the same time window, there is marked increase in phase alignment for both hits and correct rejections which is tends to be more prominent for hits than for correct rejections. The plots in the bottom row show the grand mean and the standard deviation of amplitude and phase alignment for hits (red lines) and correct rejections (blue lines) in the frequency range marked by the dashed vertical lines in the time frequency spectra. The bottom row displays the time course and the magnitude of t values form serial related-measures t-tests between hits and correct rejections. The two dashed lines corresponds to P values of 0.05 and 0.01 respectively. Amplitude values are in fT.

Figure 2.

Single trial analyses of amplitude and phase alignment differences between hits and correct rejections at left occipital sensor 81. The time frequency spectra show the grand mean average (eight subjects) of single trial amplitude differences (in fT) between hits and correct rejections at left occipital sensor 81. Time frequency spectra in the upper row were derived with ‘standard’ wavelets while time frequency spectra in the lower row were derived with a more strongly ‘dampened’ wavelet. The two types of wavelets differ in their time/frequency resolution, with the upper row having a higher frequency resolution and the lower row a higher time resolution. Results from both analyses show a single trial amplitude difference between hits and correct rejections between 200 and 500 ms at 4–5 Hz. In the same time window, there is marked increase in phase alignment for both hits and correct rejections which is tends to be more prominent for hits than for correct rejections. The plots in the bottom row show the grand mean and the standard deviation of amplitude and phase alignment for hits (red lines) and correct rejections (blue lines) in the frequency range marked by the dashed vertical lines in the time frequency spectra. The bottom row displays the time course and the magnitude of t values form serial related-measures t-tests between hits and correct rejections. The two dashed lines corresponds to P values of 0.05 and 0.01 respectively. Amplitude values are in fT.

Figure 3.

Single trial analyses of amplitude and phase alignment differences between hits and correct rejections at left temporal sensor 114 and left parietal sensor 19. The time frequency spectra in the upper row and the plots in the lower row display the same information as in Figure 2. Amplitude values are in fT.

Figure 3.

Single trial analyses of amplitude and phase alignment differences between hits and correct rejections at left temporal sensor 114 and left parietal sensor 19. The time frequency spectra in the upper row and the plots in the lower row display the same information as in Figure 2. Amplitude values are in fT.

Figure 4.

Schematic illustration of single trial time-frequency coefficients for an induced response, a phase alignment response and an evoked response, for a given time point (e.g. 410 ms), frequency (e.g. 2 Hz) and sensor (e.g. left temporal sensor 114). A stronger clustering of single trial values (red or blue dots) to a quadrant indicates a stronger phase alignment. A stronger eccentricity (distance from the middle) of the single trial values corresponds to a higher amplitude. For each single trial value, the amplitude can be derived by taking the square root of the sum of the squares of x-axis and y-axis values. The four types of response are explained in Materials and Methods. Amplitude values are in fT.

Figure 4.

Schematic illustration of single trial time-frequency coefficients for an induced response, a phase alignment response and an evoked response, for a given time point (e.g. 410 ms), frequency (e.g. 2 Hz) and sensor (e.g. left temporal sensor 114). A stronger clustering of single trial values (red or blue dots) to a quadrant indicates a stronger phase alignment. A stronger eccentricity (distance from the middle) of the single trial values corresponds to a higher amplitude. For each single trial value, the amplitude can be derived by taking the square root of the sum of the squares of x-axis and y-axis values. The four types of response are explained in Materials and Methods. Amplitude values are in fT.

Figure 5.

Evidence for evoked responses from two single subject (S1 and S6) comparisons of ERFs, phase alignment, amplitudes and single trial time–frequency coefficients. Each column shows data for one subject and one sensor. Temporal and parietal refer to the sensors 114 and 19 mentioned in Figures 1–3. The first column shows ERF differences and corresponding oscillatory changes in the M400 time window and the second column shows ERF differences and corresponding oscillatory changes in the LC time window. The first row shows the unfiltered ERFs for hits and correct rejections, that is, the mean of single trial data for the two conditions. The second row displays the time course and the magnitude of t-values form serial related-measures t-tests (as in Fig. 2) between the single trial raw data for hits and correct rejections whose average contributed to the ERFs in raw one. The third row shows the mean of single trial amplitudes in the dominant frequency of the ERF difference between hits and correct rejections displayed in the first columns. The dominant frequency was 2 Hz for the M400 and 1.9 Hz for the LC time windows. The fourth row shows the time course and the magnitude of t values form serial related-measures t-tests (as in Fig. 2) between single trial amplitudes in the dominant frequency for hits and correct rejections. The fifth row shows phase alignment values across single trials in the same frequency bands. Error bars denote SD. The sixths row displays the single trial time–frequency coefficients for each frequency and sensor as in Figure 4. Red line/dots: hits; blue line/dots: correct rejections. Amplitude values are in fT.

Figure 5.

Evidence for evoked responses from two single subject (S1 and S6) comparisons of ERFs, phase alignment, amplitudes and single trial time–frequency coefficients. Each column shows data for one subject and one sensor. Temporal and parietal refer to the sensors 114 and 19 mentioned in Figures 1–3. The first column shows ERF differences and corresponding oscillatory changes in the M400 time window and the second column shows ERF differences and corresponding oscillatory changes in the LC time window. The first row shows the unfiltered ERFs for hits and correct rejections, that is, the mean of single trial data for the two conditions. The second row displays the time course and the magnitude of t-values form serial related-measures t-tests (as in Fig. 2) between the single trial raw data for hits and correct rejections whose average contributed to the ERFs in raw one. The third row shows the mean of single trial amplitudes in the dominant frequency of the ERF difference between hits and correct rejections displayed in the first columns. The dominant frequency was 2 Hz for the M400 and 1.9 Hz for the LC time windows. The fourth row shows the time course and the magnitude of t values form serial related-measures t-tests (as in Fig. 2) between single trial amplitudes in the dominant frequency for hits and correct rejections. The fifth row shows phase alignment values across single trials in the same frequency bands. Error bars denote SD. The sixths row displays the single trial time–frequency coefficients for each frequency and sensor as in Figure 4. Red line/dots: hits; blue line/dots: correct rejections. Amplitude values are in fT.

In the between-subjects analyses, the phase alignment values and the mean of the single trial amplitude values for hits and correct rejections for each subject were entered into serial related-measures t-tests, and these were compared with serial related-measures t-tests of single-subject ERFs for hits and correct rejections (Figs 2 and 3). For the within-subject analyses, the ERFs for hits and correct rejections of each subject were compared with each subject's single trial amplitude differences between hits and correct rejections, which were assessed using serial related-measures t-tests on single trial amplitude data (Figs 4 and 5).

The serial related-measures t-tests for the between-subjects and within-subjects analyses were conducted every 4 ms at the three sensors (and their three neighbours in the case of within-subjects analyses) that showed the maximum ERF differences between hits and correct rejections and the sensors immediately surrounding these three sensors. The number of successive significant (at P < 0.05 for each time point) differences between hits correct rejections was required to be the duration of a full cycle (at a given frequency) divided by 4 ms (i.e. for a 10 Hz oscillation, 25 successive data points had to show a significant difference) for the ‘standard’ wavelet analysis and the duration of a half cycle divided by 4 ms for the wavelet analysis with the ‘dampened’ wavelets.

To examine the time course and topography of amplitude and phase alignment differences between hits and correct rejections across all 148 sensors, we used Partial Least Squares (PLS), a multivariate statistical technique (for details, see McIntosh et al., 1996; Lobaugh et al., 2001; Duzel et al., 2003). PLS analyses provides so-called latent variables (LVs). Each LV entails three pieces of information: (i) design saliences, which show the pattern of relative contribution of each frequency to the differences between hits or correct rejections; (ii) sensor saliences, which show the time course and topography of the differences between hits and correct rejections identified in the corresponding design salience; and (iii) singular values, which are used to quantify the amount of covariance in the data explained by a given LV. To determine the stability of the saliences identified in the LVs, the standard errors of the saliences were estimated through 500 bootstrap samples using sampling with replacement. The ratio of the salience to the bootstrap standard error is approximately equivalent to a z-score. Data points for which the salience is greater than twice the standard error (P < 0.05) are considered reliable. To test the statistical significance of each LV, each participant's data was randomly reassigned without replacement to different experimental conditions, and the entire PLS procedure was repeated. Following 500 such randomizations, the number of times the singular value from the randomized PLS analysis exceeded the singular value from the original PLS analysis was noted, thus providing an exact probability (McIntosh et al., 1996; Lobaugh et al., 2001; Duzel et al., 2003).

Separate PLS analyses were conducted for amplitude and phase alignment values. A third PLS was performed to assess the relationship of phase alignment of oscillations in the dominant frequency range of the ERF differences between hits and correct rejections and the corresponding amplitude differences in the gamma frequency range. The data matrix in this analysis combined phase alignment values in the delta/lower theta range (1–4.6 Hz) and amplitude values in the gamma range (22–45 Hz) for hits and correct rejections.

Results

Definitions

Our definitions of a phase alignment response, an evoked response and an induced response are compatible with those proposed by Shah et al. (2004). A phase alignment response is defined as an increase in phase alignment of oscillatory brain activity without concomitant increase in amplitude. An evoked response is defined as an increase in phase alignment of oscillatory brain activity with a concomitant increase in amplitude. Across single trials, the phase values at which the amplitude increase occurs overlap with the phase values at which phase alignment increases. An induced response is defined as an increase in amplitude of oscillatory brain activity without concomitant increase in phase alignment.

Behavioural Data

The behavioural data of the eight subjects were very similar to the data reported previously from 11 subjects (Duzel et al., 2003). Mean ± SD hit rate was 88.5 ± 13.9% and mean correct rejection rate was 89.4 ± 14.1%. Reaction times were 715 ± 110 ms for hits and 760 ± 130 ms for correct rejections.

Relationship between ERFs and Oscillatory Responses

t-test maps in Figure 1 show significant left hemispheric ERF differences, with three distinct topographies in three time windows: a mainly left occipital difference (peak difference at sensor A81) between hits and correct rejections in an early time window at 220–360 ms, a mainly left temporal difference (peak difference at sensor 114) in the M400 time window (300–600 ms) and a mainly left parietal difference (peak difference at sensor A19) in the LC time window (470–600 ms). Time-frequency spectra of the grand mean ERFs showed that the ERF difference had a dominant frequency of 4–5 Hz in the early time window (left occipital sensor), of 1.5 and 2 Hz in the M400 time window (left temporal sensor), and of 2.4 Hz in the LC time window (left parietal sensor, Fig. 1).

The between-subjects single trial based amplitude and phase alignment analyses for the three time windows and sensors are shown in Figures 2 and 3. At the left occipital sensor (Fig. 2), there was a significant single trial 4–5 Hz amplitude difference between hits and correct rejections that overlapped with the time window of the early ERF difference. Phase alignment, on the other hand, showed a sharp increase from its pre-stimulus value for both hits and correct rejections, but was not reliably different between the two stimulus types in the early time window. At the left temporal sensor, on the other hand, there was a significant 1.5–2 Hz difference in amplitude as well as phase alignment for hits and correct rejections, which, when using the ‘dampened’ wavelet for amplitude analyses, overlapped with the M400 time window (Fig. 3, upper two rows). In the serial t-tests, both amplitude and phase alignment were higher for correct rejections than for hits, but the overlap in time with the M400 time window was better for the phase alignment difference, suggesting that the higher amplitude of the ERF for correct rejections was due mainly to an increase in phase alignment for correct rejections which, 100 ms later, was followed by an amplitude increase that might not have directly contributed to the ERF increase for correct rejections. Finally, over the left parietal sensor, there was a significant difference in the LC time window during the 2.4–3.1 Hz phase alignment of hits and correct rejections but no reliable amplitude difference. Hits were associated with higher phase alignment than correct rejections, suggesting that the higher amplitude of the corresponding ERF for hits was due to an increase in phase alignment with no reliable increase in single trial amplitude (Fig. 3). Here again, a good correspondence was found between the timing of ERF differences and phase alignment differences when using the ‘dampened’ wavelets for the frequency analyses.

Single-subject, Single-trial Relationship between ERF and Oscillatory Responses

To further assess the relationship between ERFs, single trial amplitude differences and phase alignment, we analysed individual subject data in more detail, by assessing the existence of evoked, induced and phase alignment responses across single trials (as schematically illustrated in Fig. 4) and by using within-subjects statistics. Figure 5 gives examples of single subject ERFs (unfiltered data), single trial amplitudes and phase alignment data from the left temporal and left parietal sensors. These examples illustrate instances of parallel differences between hits and correct rejections in ERFs, single trial amplitudes and phase alignment. Subject S1, for instance, shows a larger (more negative, blue line) ERF amplitude at ∼400 ms (M400 time window) for correct rejections at the left temporal sensor. In the single trial analyses, the same sensor shows higher single trial amplitudes at 2 Hz and a higher phase alignment for correct rejections in the same time window (Fig. 5, first column). Figure 5 also shows the single trial complex time–frequency coefficients separately for hits and correct rejections for this sensor, and for the 2 Hz wavelet at 410 ms. As outlined in Figure 4, the stronger clustering of single trial values to one of the four quadrants indicates a stronger phase alignment for correct rejections (blue dots) than for hits (red dots). Higher amplitude values for correct rejections than hits are concentrated where there is a clustering of phase. In Figure 5, this is indicated by the visible overlap between the concentration of blue dots and their eccentricity. Taken together, the parallel findings concerning ERF amplitudes, single trial amplitudes, phase alignment and phase overlap between single trial amplitudes and phase alignment shows that the larger M400 for correct rejections in S1 is caused by a larger evoked response for correct rejections than for hits. Figure 5 shows a similar type of correspondence indicative of an evoked response in the LC time window. In the early time window, evidence for evoked responses were found at 300 ms and 5 Hz over the left occipital sensor also in four subjects (data not shown).

Evoked responses in relation to the ERF effects in the M400 and LC time windows were found only in four of the eight subjects. We further assessed the reliability of these findings in these four subjects by studying the relationship outlined in Figure 4, in three sensors neighbouring each of the three selected ones (A19, A114, A81). In the four subjects that showed an evoked response, there also were evoked responses in 66% of the neighbouring sensors. This is exemplified for subject S1 in the M400 time window (Fig. 6). Two sensors neighbouring the left temporal sensor A114 also show clear indices of an evoked response as underlying the ERF difference between hits and correct rejections. In 33% of the neighbouring sensors there was a phase alignment response only, but no concomitant difference in single trial amplitude.

Figure 6.

Evidence for evoked responses in two neighbouring sensors of left temporal sensor 114 in one example subject (S1) taken from Figure 4. The rows of the figure display the same information as in Figures 4 and 5. In two sensors neighbouring sensor 114, a similar evoked response is evident as in sensor 114, displayed in Figure 5.

Figure 6.

Evidence for evoked responses in two neighbouring sensors of left temporal sensor 114 in one example subject (S1) taken from Figure 4. The rows of the figure display the same information as in Figures 4 and 5. In two sensors neighbouring sensor 114, a similar evoked response is evident as in sensor 114, displayed in Figure 5.

Why the other four subjects did not show an evoked response difference between hits and correct rejections in sensors A19, A114, A81 (and in three sensors neighbouring each of these three) is illustrated for the M400 time window in Figure 7. Two of the subjects showed a phase alignment response for correct rejections but no amplitude increase relative to hits, and the two other subjects showed no difference in amplitude, phase alignment or ERFs between hits and correct rejections, but the same evoked response for both stimulus classes (Fig. 7). These findings were consistent across neighbouring sensors. There were no differences in the level of baseline amplitude between these four subjects and the four who showed an evoked response difference between hits and correct rejections.

Figure 7.

Four subjects who did not show evidence of an evoked response difference between hits and correct rejections in the M400 time window. Data are from left temporal sensor 114. The rows of the figure display the same information as in Figures 5 and 6. Subjects S8 and S3 show a phase alignment response only. Subjects S4 and S2 show the same evoked response for hits and correct rejections and no ERF difference.

Figure 7.

Four subjects who did not show evidence of an evoked response difference between hits and correct rejections in the M400 time window. Data are from left temporal sensor 114. The rows of the figure display the same information as in Figures 5 and 6. Subjects S8 and S3 show a phase alignment response only. Subjects S4 and S2 show the same evoked response for hits and correct rejections and no ERF difference.

Thus, to summarize, of the eight subjects, six showed a significant left temporal single trial ERF difference between hits and correct rejections in the M400 time window. In two of the six subjects, this ERF difference was due to a phase alignment response only, and in four it was due to an evoked response. The same pattern was observed over left parietal sensors in the LC time window.

Partial Least Squares Analyses of Amplitude and Phase Alignment

PLS analyses of amplitude or phase alignment values for frequencies within the delta and theta range (1–7 Hz) identified four significant LVs (all Ps < 0.05) for each analysis. Figure 8A shows the first LV, which explained the highest amount of variance of the data for amplitudes (23.5% of variance explained, P < 0.001), and the first two LVs related to phase alignment, which together explained a similar degree of variance (26% of variance explained together, P < 0.001) as the first LV in the amplitude PLS. The other LVs identified interactions between low and high frequenciesm but did not differentiate hits and correct rejections and will therefore not be reported further.

Figure 8.

(A) PLS analysis of phase alignment and amplitude (first three rows) and ERF differences between hits and correct rejections (fourth row). The design saliences in the left column display how frequencies ranging from 1 to 7 Hz distinguish hits (grey bars) and correct rejections (black bars). The topographic maps on the left display the topography of the corresponding frequency patterns at early, M400 and LC time windows. The values depicted in the topographic maps in the first three rows are bootstrap ratios for the corresponding sensor saliences to their standard error which are similar to z-values. Values >1.96 or <–1.96 are statistically significant (P < 0.05). Red indicates a positive and blue indicates a negative correlation with the corresponding frequency pattern. For instance, a blue area over the left anterior temporal sensors (black arrow) in the first row indicates a stronger phase alignment for correct rejections than for hits, whereas, in the same map, red areas indicate a stronger phase alignment for hits than for correct rejections. For comparison, the topography of the ERF differences between hits and correct rejections is displayed at respective time points in the bottom row (units are in fT). (B) PLS analysis of the covariance between phase alignment in the delta and theta range and amplitude in the beta and gamma range. The design saliences in the left column display how phase alignment of frequencies ranging from 1 to 4.6 Hz and amplitude of frequencies ranging from 22 to 45 Hz jointly distinguish hits (grey bars) and correct rejections (black bars). The topographic maps on the left display the topography of the corresponding frequency patterns at early, M400 and LC time windows. The values depicted in the topographic maps are the same as in Figure 6. For instance, a blue area over the left anterior temporal sensors (black arrow) indicates a stronger phase alignment for correct rejections than for hits, and a concomitantly higher amplitude of oscillations between 22 and 34 Hz for correct rejections than for hits.

Figure 8.

(A) PLS analysis of phase alignment and amplitude (first three rows) and ERF differences between hits and correct rejections (fourth row). The design saliences in the left column display how frequencies ranging from 1 to 7 Hz distinguish hits (grey bars) and correct rejections (black bars). The topographic maps on the left display the topography of the corresponding frequency patterns at early, M400 and LC time windows. The values depicted in the topographic maps in the first three rows are bootstrap ratios for the corresponding sensor saliences to their standard error which are similar to z-values. Values >1.96 or <–1.96 are statistically significant (P < 0.05). Red indicates a positive and blue indicates a negative correlation with the corresponding frequency pattern. For instance, a blue area over the left anterior temporal sensors (black arrow) in the first row indicates a stronger phase alignment for correct rejections than for hits, whereas, in the same map, red areas indicate a stronger phase alignment for hits than for correct rejections. For comparison, the topography of the ERF differences between hits and correct rejections is displayed at respective time points in the bottom row (units are in fT). (B) PLS analysis of the covariance between phase alignment in the delta and theta range and amplitude in the beta and gamma range. The design saliences in the left column display how phase alignment of frequencies ranging from 1 to 4.6 Hz and amplitude of frequencies ranging from 22 to 45 Hz jointly distinguish hits (grey bars) and correct rejections (black bars). The topographic maps on the left display the topography of the corresponding frequency patterns at early, M400 and LC time windows. The values depicted in the topographic maps are the same as in Figure 6. For instance, a blue area over the left anterior temporal sensors (black arrow) indicates a stronger phase alignment for correct rejections than for hits, and a concomitantly higher amplitude of oscillations between 22 and 34 Hz for correct rejections than for hits.

The LVs for phase alignment showed a bell-shaped distribution of frequencies within the delta range, with a maximum at 3.00–3.46 Hz dissociated hits (grey bars in Fig. 8A) from correct rejections in the first LV and a slightly slower set of frequencies in the second LV (black bars in Fig. 8A). The LV for amplitudes had a different frequency distribution, which was shifted towards higher frequencies (maximum difference between hits and correct rejections at 4.6–5.31 Hz, Fig. 8A). Figure 8A also shows topographies of the frequency patterns (bootstrap ratios) from all three LVs at 290, 400 and 510 ms, corresponding to the early, M400 and LC time windows, respectively.

The main results of this PLS analysis was that the patterns of phase alignment and amplitude which explained the largest amount of variance in the data differed from each other in terms of their frequency composition and in terms of their topography. The topographical distribution of phase alignment differences between hits and correct rejections resemble much more the left occipital, parietal and temporal topography of the ERF differences between hits and correct rejections (see also bottom row of Fig. 8A) than the topography of amplitude differences which have a prominent fronto-midline distribution. In fact, the finding that the fronto-midline amplitude differences between hits and correct rejections were not accompanied by concomitant phase alignment differences and differed in topography from the ERF differences between hits and correct rejections clearly indicates that these amplitude differences stem from induced responses. Thus, data patterns that contributed to the greatest amount of variance of phase alignment and amplitude values did not stem from evoked responses only, but rather from phase alignment responses and from induced responses, respectively.

Covariance of Slow and Fast Oscillations

Results from the third PLS shed light on the covariance between delta/theta phase alignment and the amplitude of beta and gamma oscillations. One LV clearly separated hits and correct rejections (P < 0.01, 12% of variance explained), and showed covariance between delta/theta phase alignment and the amplitude of beta and gamma oscillations (Fig. 8B). The pattern of the design salience in Figure 8B consists of a positive covariation of delta/theta phase alignment at 2.9 and 3.4 Hz with beta and gamma power (from 29.3 and 33.8 Hz) and a negative correlation with gamma power at 45 Hz. The topography of this pattern partly overlaps with the topography of the ERF difference between hits and correct rejections. For instance, in the M400 time window (410 ms), there is a left temporal higher phase alignment for correct rejections than for hits (Fig. 8B; see also Figs 4–7). This phase alignment difference is accompanied by higher left temporal beta and gamma amplitudes (from 29.3 and 33.8 Hz) for correct rejections than for hits.

Discussion

As expected, the dominant frequency of the event-related field (ERF) differences between hits and correct rejections was in the delta range in the M400 and LC time windows, and in the lower theta frequency range in the early time window. If hits and correct rejections differed from each other in evoked neural activity within this frequency range, there should have been parallel differences in the phase alignment and amplitude of delta/theta oscillations between the two stimulus classes. However, our multivariate analyses revealed a dissociation between phase alignment and amplitude, both in terms of frequency composition and in terms of topography (Fig. 8A). As shown in Figure 8A, amplitude differences between hits and correct rejections were predominantly in the mid-theta range, thereby faster than the dominant frequency of the ERF differences in the M400 and LC time windows, whereas the phase alignment differences were predominantly in the delta range, thereby overlapping with the dominant frequency of the ERF differences in the M400 and LC time windows. Furthermore, the topography of phase alignment differences resembled the topography of the ERF effects, whereas the topography of the amplitude differences were quite distinct, showing a predominantly fronto-midline distribution.

A recent animal study of intracortical oscillatory activity related to visual perception showed that, unlike early visual areas such as V1, the inferior temporal cortex has a considerable amount of ongoing oscillatory activity in the theta and delta frequency range even prior to stimulus onset (Shah et al., 2004). It is likely that ERP/ERF effects of recognition memory in the M400 and LC time windows are generated or modulated by neural populations in upstream visual areas of the temporal lobes (Grunwald et al., 1998; Duzel et al., 2001, 2003). As can be seen in Figures 2 and 3, the sensors which showed the maximal ERF effects in the M400 and LC time windows also showed a considerable amount of slow activity in the baseline period preceding stimulus onset. Shah et al. (2004) suggested that the existence of pre-stimulus oscillatory activity in regions upstream of early visual areas could indicate that cognitive processes which depend on these higher visual areas are likely to rely less on evoked mechanisms and, instead, possibly show a dissociation of phase alignment and amplitude changes. Our data therefore support the notion that upstream visual areas show a considerable amount of ongoing slow oscillatory activity and also support the notion that phase alignment responses are the most reliable finding related to recognition memory. Phase alignment differences between hits and correct rejections were the most robust finding in the M400 and LC time windows in the between-subjects analyses (Fig. 3) and in the multivariate analyses (Fig. 8A). In the within-subject analyses, phase alignment differences were found in all six subjects who showed a significant ERF difference between hits and correct rejections, whereas evoked differences were found in only four of the six. Furthermore, phase alignment differences were found in all neighbouring sensors of the three sensors of interest, whereas evoked differences were found in 66% of neighbouring sensors.

While these findings clearly show that in the dominant frequency range of the event-related responses phase alignment differences are a reliable phenomenon in both the M400 and LC time windows, a closer within-subjects inspection of single trial data nevertheless revealed clear indices of evoked responses. Of the six subjects who showed significant ERF differences between hits and correct rejections over left temporal sensors (the two subjects that did not show a significant ERF difference are shown in Figure 7), four showed parallel differences in phase alignment and amplitude between hits and correct rejections (Figs 5 and 6). More importantly, in these four subjects, the single-trial amplitude differences appeared to be largest at the phase values that showed the largest differences in phase alignment. This is a clear indication that the ERF difference in these subjects can be attributed to an evoked response (Figs 4–6).

An influential hypothesis regarding the functional interpretation of the ERP/ERF old/new differences in the M400 and LC time windows has been that they reflect the activation of a widespread network controlled by the MTL (Halgren and Smith, 1987). Our findings confirm the idea that, in principle, these ERF effects can reflect local ‘activation’ because, in addition to phase alignment, they can show an evoked component. The control of large-scale integration of distributed information by the MTL (Rolls, 1996; Vargha-Khadem et al., 1997), on the other hand, could be reflected in the fronto-midline induced theta oscillations that are elicited more strongly by old than by new stimuli. Induced fronto-midline theta oscillations have previously been associated with limbic theta oscillations originating in the anterior cingulated cortex (Kirk and Mackay, 2003), and it has been suggested that in humans fronto-midline theta is associated with functional connectivity among distributed brain regions (Mizuhara et al., 2004). Alternatively, this fronto-midline could be functionally related to sensorimotor integration between medio-temporal processing and the motor output required to make the old/new response (Bland and Oddie, 2001; Luu et al., 2004). Finally, an index of local integration of distributed information might be the covarying fast (beta and gamma) oscillations that co-occur in time windows of elevated phase alignment (Fig. 8B) and complements a number of earlier findings concerning relationships between slow and fast oscillations in animals (Basar-Eroglu and Basar, 1991; von Stein et al., 2000; Engel et al., 2001) and humans (Burgess and Ali, 2002; Schack et al., 2002; Duzel et al., 2003; Fell et al., 2003).

The question to what extent evoked, induced and phase alignment responses contribute to recognition memory is also relevant for the relationship between EEG/MEG and functional magnetic resonance imaging (fMRI) data, that is for multimodal imaging. Logothetis and co-authors (Logothetis et al., 2001; Logothetis, 2003; Logothetis and Wandell, 2004) recently argued that the amplitude of fast local field potential (LFP) oscillations are more strongly correlated with the fMRI BOLD signal than the amplitude of slow oscillations. Although the physiological underpinnings of LFPs and of the BOLD signal are not entirely understood (Logothetis and Wandell, 2004), it is clear that discrepancies between EEG/MEG data with BOLD measurements can occur to the extent that evoked, induced and phase alignment responses are associated with different BOLD responses, and to the extent that these responses do not covary with responses of fast oscillations (Duzel et al., 2003; Foucher et al., 2003). The present data support the notion that such covariation of fast and slow oscillations exist and can be non-invasively measured in humans. While, according to our knowledge, not much is known about the relationship between phase alignment response and the BOLD response, a covariance between phase alignment response and fast oscillatory amplitude changes could be a basis for an indirect relationship. A systematic investigation of the relationship between phase of background oscillations and amplitude of fast oscillations will certainly help to improve our understanding and utilization of multimodal imaging.

It should be noted that we have outlined three concepts of large-scale neural responses, but these might not fully account for all possible forms of event-related dynamics, such as cases in which a mild increase in phase alignment is accompanied by a mild increase in amplitude. In this case, our term ‘evoked’ would not be precise enough since the peak latency of the ‘evoked’ oscillation will vary from trial to trial. Makeig et al. (2004) have recently argued that event-related phenomena may occur anywhere in the whole space of amplitude, phase alignment, frequency possibilities. Whether there is such a continuum or whether there are classes of neural response patterns as proposed here awaits further studies. However, for the purpose of the current study, our definition of the term ‘evoked’ was adequate, because of interest were the neural responses underlying the ERP/ERF effects of recognition memory which, by definition, occur in designated time windows.

To summarize, even higher cognitive processes that are critically dependent on and partly originate in the anterior portions of the ventral visual stream, and that occur relatively late after stimulus onset, can be partly based on evoked neural mechanisms. Although evoked responses were clearly implicated in recognition memory, our data also show that differential neural responses to hits and correct rejections cannot be entirely accounted for by evoked mechanisms. In fact, prominent amplitude differences between hits and correct rejections were induced responses, that is, they were not accompanied by corresponding increases in phase alignment, and the most reliable concomitants underlying ERF effects of recognition appeared to be phase alignment responses. It remains to be established how the three different neural response types — evoked responses, induced responses and phase alignment responses — are functionally related to each other and to recognition memory.

We thank M. Scholz and B.H. Schott for assistance with data recording and analysis. This study was supported by grants from the Deutsche Forschungsgemeinschaft (DFG/SFB 426, TP D3) and BMBF (01GO0202, CAI).

References

Baayen RL, Piepenbrock R, von Rijn H (
1993
) The celex lexical database. Philadelphia: University of Pennsylvania, Linguistic Data Consortium.
Basar-Eroglu C, Basar E (
1991
) A compound P300–40 Hz response of the cat hippocampus.
Int J Neurosci
 
60
:
227
–237.
Bland BH, Oddie SD (
2001
) Theta band oscillation and synchrony in the hippocampal formation and associated structures: the case for its role in sensorimotor integration.
Behav Brain Res
 
127
:
119
–136.
Burgess AP, Ali L (
2002
) Functional connectivity of gamma EEG activity is modulated at low frequency during conscious recollection.
Int J Psychophysiol
 
46
:
91
–100.
Buzsaki G, Draguhn A (
2004
) Neuronal oscillations in cortical networks.
Science
 
304
:
1926
–1929.
Duzel E, Vargha-Khadem F, Heinze HJ, Mishkin M (
2001
) Brain activity evidence for recognition without recollection after early hippocampal damage.
Proc Natl Acad Sci USA
 
98
:
8101
–8106.
Duzel E, Habib R, Schott B, Schoenfeld A, Lobaugh N, McIntosh AR, Scholz M, Heinze HJ (
2003
) A multivariate, spatiotemporal analysis of electromagnetic time-frequency data of recognition memory.
Neuroimage
 
18
:
185
–197.
Engel AK, Fries P, Singer W (
2001
) Dynamic predictions: oscillations and synchrony in top-down processing.
Nat Rev Neurosci
 
2
:
704
–716.
Fell J, Klaver P, Elfadil H, Schaller C, Elger CE, Fernandez G (
2003
) Rhinal–hippocampal theta coherence during declarative memory formation: interaction with gamma synchronization?
Eur J Neurosci
 
17
:
1082
–1088.
Foucher JR, Otzenberger H, Gounot D (
2003
) The BOLD response and the gamma oscillations respond differently than evoked potentials: an interleaved EEG–fMRI study.
BMC Neurosci
 
4
:
22
.
Grunwald T, Lehnertz K, Heinze HJ, Helmstaedter C, Elger CE (
1998
) Verbal novelty detection within the human hippocampus proper.
Proc Natl Acad Sci USA
 
95
:
3193
–3197.
Halgren E, Smith ME (
1987
) Cognitive evoked potentials as modulatory processes in human memory formation and retrieval.
Hum Neurobiol
 
6
:
129
–139.
Hillyard SA (
1985
) Electrophysiology of human selective attention.
Trends Neurosci
 
8
:
400
–405.
Hillyard SA, Picton TW (
1987
) Electrophysiology of cognition. In: Handbook of physiology (F P, ed.), pp. 519–584. Oxford: Oxford University Press.
Jansen BH, Agarwal G, Hegde A, Boutros NN (
2003
) Phase synchronization of the ongoing EEG and auditory EP generation.
Clin Neurophysiol
 
114
:
79
–85.
Jervis BW, Nichols MJ, Johnson TE, Allen E, Hudson NR (
1983
) A fundamental investigation of the composition of auditory evoked potentials.
IEEE Trans Biomed Eng
 
30
:
43
–50.
Kirk IJ, Mackay JC (
2003
) The role of theta-range oscillations in synchronising and integrating activity in distributed mnemonic networks.
Cortex
 
39
:
993
–1008.
Klimesch W, Doppelmayr M, Schwaiger J, Winkler T, Gruber W (
2000
) Theta oscillations and ERP old/new effect: independent phenomena?
Clin Neurophysiol
 
111
:
781
–793.
Lisman JE, Otmakhova NA (
2001
) Storage, recall, and novelty detection of sequences by the hippocampus: elaborating on the SOCRATIC model to account for normal and aberrant effects of dopamine.
Hippocampus
 
11
:
551
–568.
Lobaugh NJ, West R, McIntosh AR (
2001
) Spatiotemporal analysis of experimental differences in event-related potential data with partial least squares.
Psychophysiology
 
38
:
517
–530.
Logothetis NK (
2003
) The underpinnings of the BOLD functional magnetic resonance imaging signal.
J Neurosci
 
23
:
3963
–3971.
Logothetis NK, Wandell BA (
2004
) Interpreting the BOLD signal.
Annu Rev Physiol
 
66
:
735
–769.
Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A (
2001
) Neurophysiological investigation of the basis of the fMRI signal.
Nature
 
412
:
150
–157.
Luu P, Tucker DM, Makeig S (
2004
) Frontal midline theta and the error-related negativity: neurophysiological mechanisms of action regulation.
Clin Neurophysiol
 
115
:
1821
–1835.
Makeig S, Westerfield M, Jung TP, Enghoff S, Townsend J, Courchesne E, Sejnowski TJ (
2002
) Dynamic brain sources of visual evoked responses.
Science
 
295
:
690
–694.
Makeig S, Debener S, Onton J, Delorme A (
2004
) Mining event-related brain dynamics.
Trends Cogn Sci
 
8
:
204
–210.
Mangun G (
1992
) Human brain potentials evoked by visual stimuli:induced rhythms or time-locked components? In: Induced rhythms in the brain (Basar EBT, ed.). Boston:Birkhauser.
McIntosh AR, Bookstein FL, Haxby JV, Grady CL (
1996
) Spatial pattern analysis of functional brain images using partial least squares.
Neuroimage
 
3
:
143
–157.
Mehta MR, Lee AK, Wilson MA (
2002
) Role of experience and oscillations in transforming a rate code into a temporal code.
Nature
 
417
:
741
–746.
Mishkin M, Vargha-Khadem F, Gadian DG (
1998
) Amnesia and the organization of the hippocampal system.
Hippocampus
 
8
:
212
–216.
Mizuhara H, Wang LQ, Kobayashi K, Yamaguchi Y (
2004
) A long-range cortical network emerging with theta oscillation in a mental task.
Neuroreport
 
15
:
1233
–1238.
Penny WD, Kiebel SJ, Kilner JM, Rugg MD (
2002
) Event-related brain dynamics.
Trends Neurosci
 
25
:
387
–389.
Rolls ET (
1996
) A theory of hippocampal function in memory.
Hippocampus
 
6
:
601
–620.
Rugg MD, Yonelinas AP (
2003
) Human recognition memory: a cognitive neuroscience perspective.
Trends Cogn Sci
 
7
:
313
–319.
Sayers BM, Beagley HA, Henshall WR (
1974
) The mechansim of auditory evoked EEG responses.
Nature
 
247
:
481
–483.
Schack B, Vath N, Petsche H, Geissler HG, Moller E (
2002
) Phase-coupling of theta-gamma EEG rhythms during short-term memory processing.
Int J Psychophysiol
 
44
:
143
–163.
Schroeder CE, Steinschneider M, Javitt DC, Tenke CE, Givre SJ, Mehta AD, Simpson GV, Arezzo JC, Vaughan HG Jr (
1995
) Localization of ERP generators and identification of underlying neural processes.
Electroencephalogr Clin Neurophysiol Suppl
 
44
:
55
–75.
Shah AS, Bressler SL, Knuth KH, Ding M, Mehta AD, Ulbert I, Schroeder CE (
2004
) Neural dynamics and the fundamental mechanisms of event-related brain potentials.
Cereb Cortex
 
14
:
476
–483.
Steriade M (
2000
) Corticothalamic resonance, states of vigilance and mentation.
Neuroscience
 
101
:
243
–276.
Tallon-Baudry C, Bertrand O (
1999
) Oscillatory gamma activity in humans and its role in object representation.
Trends Cogn Sci
 
3
:
151
–162.
Tallon-Baudry C, Bertrand O, Delpuech C, Pernier J (
1996
) Stimulus specificity of phase-locked and non-phase-locked 40 Hz visual responses in human.
J Neurosci
 
16
:
4240
–4249.
Vargha-Khadem F, Gadian DG, Watkins KE, Connelly A, Van Paesschen W, Mishkin M (
1997
) Differential effects of early hippocampal pathology on episodic and semantic memory.
Science
 
277
:
376
–380.
Vaughan HG, Arezzo JC (
1988
) The neural basis of event-related potentials. In: Human event-related potentials (Picton TW, ed.), pp. 45–94. New York: Elsevier Science Publishers.
von Stein A, Chiang C, Konig P (
2000
) Top-down processing mediated by interareal synchronization.
Proc Natl Acad Sci USA
 
97
:
14748
–14753.