oscillatory eeg correlates of episodic trace decay

Oscillatory EEG Correlates of EpisodicTrace Decay

W. Klimesch1, S. Hanslmayr1, P. Sauseng1, W. Gruber1,

C.J. Brozinsky2, N.E.A. Kroll2, A.P. Yonelinas2 and

M. Doppelmayr1

1Department of Physiological Psychology, Institute of

Psychology, University of Salzburg, Hellbrunnerstr. 34, A-5020

Salzburg, Austria and 2University of California, Department of

Psychology, One Shields Avenue, Davis, CA 95616-8686, USA

Recent studies suggest that human theta oscillations appear to befunctionally associated with memory processes. It is less clear,however, to what type of memory sub-processes theta is related.Using a continuous word recognition task with different repetitionlags, we investigate whether theta reflects the strength of anepisodic memory trace or general processing demands, such astask difficulty. The results favor the episodic trace decay hypoth-esis and show that during the access of an episodic trace in a timewindow of ~200--400 ms, theta power decreases with increasinglag (between the first and second presentation of an item). LORETAsource localization of this early theta lag effect indicates thatparietal regions are involved in episodic trace processing, whereasright frontal regions may guide the process of retrieval. Weconclude that episodic encoding can be characterized by twodifferent stages: traces are first processed at parietal sites at ~300ms, then further processing takes place in regions of the medialtemporal lobe at ~500 ms. Only the first stage is related to theta,whereas the second is reflected by a slow wave with a frequencyof ~2.5 Hz.

Keywords: episodic memory, ERD, LORETA, oscillation, theta

Introduction

Inspired from animal research, human theta has been studied

extensively in memory tasks during the last two decades and,

compared with other EEG frequencies, is the best investigated

example of a memory-related brain oscillation (for reviews, see

Klimesch, 1999; Kahana et al., 2001). Theta appears functionally

related primarily to working memory (WM), but it is less clear to

which of the different WM processes or other aspects of

memory theta is associated. We argue in the following that

different types of theta responses can be distinguished that

reflect (i) the maintenance of information in WM; (ii) sustained

attention; and (iii) episodic encoding/retrieval. In the present

study, we test whether theta — or, more precisely, which

aspect of theta — reflects the strength of an episodic trace.

Studies focusing on the maintenance of information in WM

have shown that with increasing load, theta power increases

(Mecklinger et al., 1992; Gevins et al., 1997, 1998; Grunwald

et al., 1999; Klimesch et al., 1999; Gevins and Smith, 2000;

Raghavachari et al., 2001; Fingelkurts et al., 2002; Jensen and

Tesche, 2002). This load-dependent increase in theta power

was found most consistently during the retention interval of

different types of WM tasks, including the Sternberg and n-back

tasks. In addition, studies using subdural electrodes or specific

source analyzing methods have shown that the length of theta

episodes is related to the duration of WM demands in a memory

scanning (Tesche and Karhu, 2000; Raghavachari et al., 2001) or

spatial WM task (Kahana et al., 1999; Araujo et al., 2002). With

respect to the involvement of different brain areas, Sarnthein

et al. (1998) have shown that during the maintenance of

information in WM, theta coherence is significantly increased

between frontal and posterior recording sites (for similar

findings, see von Stein et al., 1999; Weiss and Rappelsberger,

2000; Sauseng et al., 2002, 2004, 2005). These results suggest

that the active maintenance of information in WM (e.g. by

rehearsal or other executive functions) is reflected by a sus-

tained increase in theta power, particularly at frontal sites, and

an increase in coherence between frontal and posterior sites.

Another type of theta response can be observed in tasks

requiring sustained attention. Several studies indicate that

rhythmic theta activity can be observed in the ongoing EEG

over frontal midline sites (‘frontal midline theta’ or ‘Fmh’;Ishihara and Yoshii, 1972; Asada et al., 1999; Ishii et al., 1999;

Aftanas and Golocheikine, 2001; Kubota et al., 2001) if subjects

perform continuously a demanding task for some time period.

A third type of theta response is characterized by a brief

event-related increase in theta power during episodic encoding

and retrieval. Particularly research using the event-related

desynchronization/synchronization (ERD/ERS) approach (a

method measuring the percentage of power change during

task performance with respect to a reference interval; for

reviews, see Pfurtscheller and Lopes da Silva, 1999) indicates

that during encoding and retrieval a pronounced but transient

increase in theta within a time window of ~100--400 ms after

stimulus presentation can be observed (Klimesch et al., 1994,

1996, 1997a,b, 2000, 2001a,b; Krause et al., 1996, 2001; Burgess

and Gruzelier, 1997, 2000; for a review, see Klimesch, 1999). In

all of these studies the list of items presented during encoding

was much too long to be kept in WM. Thus, in these studies,

theta appears to reflect a different function, most likely episodic

encoding and recognition. This conclusion is supported by (i)

studies focusing on a direct comparison between semantic and

episodic processes (Klimesch et al., 1994); (ii) findings that the

extent of theta ERS during encoding predicts later memory

performance (= theta subsequent memory effect; see Klimesch

et al., 1996, 1997b); (iii) findings that during recognition ERS for

old words is larger than for new words (= theta old--new effect;

Klimesch et al., 1997b, 2000); and (iv) the topography of theta

during encoding and retrieval, which is clearly related to

parietal sites (Klimesch, 1999). The theta old--new and sub-

sequent memory effect — characterized by a brief increase in

ERS (~100--400 ms)— indicate that theta ERS reflects successful

encoding and retrieval of episodic information. The similar

parietal topography during encoding and recognition, together

with the brief and early increase in theta ERS, suggest that

parietal areas play an important role for the early processing of

an episodic memory trace. The parietal topography with its

� The Author 2005. Published by Oxford University Press. All rights reserved.

For permissions, please e-mail: [email protected]

Cerebral Cortex February 2006;16:280--290

doi:10.1093/cercor/bhi107

Advance Access publication May 11, 2005

by guest on March 14, 2016

http://cercor.oxfordjournals.org/D

ownloaded from

http://cercor.oxfordjournals.org/

brief increase in theta distinguishes episodic processing from

more general WM demands that are associated with a sustained

increase in theta power, particularly at frontal sites.

In the present study, we focus on the episodic trace strength

hypothesis, i.e. we investigate whether the short and transient

increase in theta (within a time window of ~100--400 ms)

reflects (at least in part) the strength of a memory trace. Our

assumption is that high trace strength is related to large theta

ERS whereas low trace strength is related to small theta ERS.

Although the above reviewed findings favor the episodic trace

hypothesis, the following results question its general validity.

During recognition, ERS is generally larger than during encod-

ing and, most importantly, a significant increase in theta ERS can

be observed not only for old items but also for new items

(Klimesch et al., 1996, 1997b, 2001a). In addition, theta ERS for

new items during recognition is also significantly larger than

ERS for items during successful encoding (Klimesch et al.,

2001a). Thus, theta ERS for new items during recognition

cannot be interpreted to reflect retrieval of an existing memory

trace nor can the entire extent of ERS reflect encoding of a new

trace. It appears plausible to assume that at least part of the

increase in theta during recognition is process specific (similar

to Tulving’s description of a ‘retrieval mode’; Tulving, 1983).

In a previous study, we used the remember--know (R--K)

design, which allows the distinction between recollection and

familiarity during episodic retrieval in order to study the

influence of trace strength on theta ERS, but in a rather indirect

way. Subjects were asked to indicate whether they consciously

recollected the event in which a word was earlier presented

(remembering) or whether they recognized it on the basis that

it was familiar in the absence of recollection (knowing). R-

judgements are associated with higher memory confidence

than K-judgements. Trace strength, however, is not the only

factor differentiating between these two types of judgements.

According to Tulving (Tulving, 1985), R- and K-judgements

reflect two different states of conscious awareness. We have

tested the following hypotheses: (i) whether the time course

and extent of theta ERS is different for both types of retrieval

processes; and (ii) whether theta ERS is larger during recol-

lection (Klimesch et al., 2001b). The first hypothesis was

supported, the second not. We have found that an early and

short lasting increase in theta predicted knowing, whereas

a longer lasting increase (with a similar early onset latency)

predicted remembering. This is in agreement with reaction

time experiments which have shown that familiarity informa-

tion is accessible earlier than recollected information (e.g.

Hintzman et al., 1998). However, the absence of a difference

between theta on recollection and familiarity trials suggests that

theta ERS was not directly related to trace strength. Although

recollection and familiarity are intended to gauge different

states of conscious awareness, trials given familiarity responses

are typically recognized with a lower level of confidence than

recollect trials (for a review, see Yonelinas, 2002). If trace

strength were the only relevant factor, a strong increase in theta

would be expected for recollection only, which was not the

case. But in the R--K design, the effects of trace strength and

type of retrieval process (underlying a familiarity and remember

judgement) might very well be confounded. For recollection

the processing of retrieval cues is important, which is a time

consuming process. Thus, retrieval will be comparatively slow,

and this will be reflected be a prolonged increase in theta. In

contrast, a familiarity judgement may simply be based on trace

strength, information that is available very early during retrieval.

Consequently, theta shows a short lasting and early increase in

ERS. Thus, comparing theta ERS for familiarity and remember

judgements may primarily reflect differences in retrieval

strategies rather than differences in trace strength.

In the present study, we use a continuous word recognition

task with different lag conditions to test the trace hypothesis.

This task has two advantages. Subjects are not in a continuous

retrieval mode (as in standard recognition tasks) because they

also have to encode new items and, most importantly, trace

strength can be directly manipulated by presentation lag (i.e.

a word can by repeated after e.g. 2, 8 or 16 intervening words).

On the basis of the trace hypothesis, we assume that theta ERS

decreases with lag. If theta reflects general task demands (of

WM), in the sense that increasing task difficulty increases theta

ERS, we would have to expect that under the more difficult

condition (lag-16) theta ERS should be larger than for the easier

condition (lag-2). Thus, the trace hypothesis and the WM/

cognitive load hypothesis make different predictions. Accord-

ing to the trace hypothesis, accessing a weak trace should be

associated with a small theta ERS, but according to the WM/

cognitive load hypothesis, processing of a weak trace should be

related to a large theta ERS.

For the evaluation of these predictions, time poststimulus is

of critical importance. Accessing a memory trace must occur

before the recognition judgement can be made. In continuous

word recognition tasks requiring old--new judgements, reaction

times (RTs) vary between ~650 and 750 ms depending on lag

(e.g. Friedman, 1990b; Kim et al., 2001). Accordingly, accessing

and processing a trace must occur before ~500--600 ms. For the

evaluation of the episodic trace hypothesis, we will thus focus

on a time period of ~200--600 ms in assuming that sensory

semantic encoding already starts before ~200 ms. The critical

prediction is that theta ERS will be smaller for the long lag

condition (lag-16) than for the short lag condition (lag-2).

Finally, it should be emphasized that we do not assume that

the total extent of theta ERS reflects trace strength. As we know

from the comparisons between ERS during encoding and

retrieval and between ERS for old and new words during

recognition, some extent of theta ERS must be process specific

and not related to trace strength. Consequently, we have to

expect that the effects of trace strength might be quite small

and rather localized in topography. Based on earlier findings we

expect that parietal and parieto-temporal sites will show the

largest lag-related effects in theta (Klimesch et al. 2000,

2001a,b).

Materials and Methods

SubjectsA sample of 28 right-handed students (nine males, mean age ± SD =26.4 ± 4.5 years; 19 females, mean age = 22.8 ± 3.4 years) participated

voluntarily and after informed consent.

Materials and DesignThe present experiment was designed as a parallel study to a recent

fMRI experiment (Brozinsky et al., 2005), and thus type of stimuli and

task procedure were largely equivalent, with the following exceptions:

Brozinsky et al. (2005) used an additional lag condition (lag-32), shorter

trials [500 ms stimulus exposure time and 2000 ms inter-stimulus

interval (ISI)] and English-speaking subjects.

Stimuli consisted of 600 nouns which were selected from the CELEX

database and grouped into eight lists. Each word was presented for 1000

ms with an ISI of 3000 ms. Words that appeared on one list did not

Cerebral Cortex February 2006, V 16 N 2 281



ownloaded from


appear on other lists. Each list consisted of randomly interleaved trials

comprising 75 new words (30 that never reappeared and 45 that were

repeated). From these 45 words (termed first presentation words),

15 words were repeated after two intervening words (lag-2; SOA = 8 s),

15 words after eight intervening words (lag-8; SOA = 32 s) and 15

words after 16 intervening words (lag-16; SOA = 64 s). Thus, there were

30 + 45 + 45 presentations for each list, with a ratio of 75 new to 45 old

items. Stimuli were presented at the center of a computer monitor,

placed 1.2 m in front of the subjects. The words covered a visual angle of

2.233.3–9.5� andwerepresented in light gray on a dark gray background.

For the lag conditions, the range of possible list positions was

identified, and broken into 15 equally spaced bins. Test items were

then placed at a random location within each of these lag-specific bins.

This procedure was used to prevent autocorrelation between adjacent

trials, and also to reduce the confounding of lag and serial list position

(i.e. lag-16 items cannot be presented until the seventeenth trial,

whereas lag-2 items can begin after the third trial). Test lists were also

balanced for word frequency (10--299 per million; Baayen et al., 1995),

concreteness (204--593), imageability (275--638), number of letters

(3--8) and number of syllables (1--3). Repeated and non-repeated words

had the same linguistic properties within and across lists.

The experimental session started and ended with a resting period.

Subjects were asked to relax and the resting EEG was recorded first for

2 min with open eyes and then for 2 min with closed eyes. Prior to the

recognition task, a training block was run to ensure that the subjects

were able to respond fast and accurately. The recognition task consisted

of eight blocks with 120 trials each. Subjects were asked to make

a recognition judgement in each trial by rating the confidence of their

judgement. They did so by pressing one of six buttons (1 = very certain

old, 2 = certain old, 3 = uncertain old, 4 = uncertain new, 5 = certain new,

6 = very certain new).

ApparatusEEG-signals were amplified by a 32-channel Neuroscan Synamps

amplifier (frequency response: 0.15--45 Hz) and were then converted

to a digital format. The sampling rate was 250 Hz.

Electrophysiological RecordingsA set of 30 silver electrodes were placed, according to the International

Electrode (10--20) Placement system, at Fp1, Fp2, F7, F3, Fz, F4, F8, FC5,

FC1, FC2, FC6, T3, C3, Cz, C4, T4, CP5, CP1, CP2, CP6, T5, P3, Pz, P4, T6,

PO3, PO4, O1, Oz and O2. In addition, two earlobe electrodes were

attached to the left and right ear. Data were recorded against a linked

earlobe reference. For data analysis only the following ten electrodes

were selected: F3, Fz, F4, T5, T6, Pz, PO3, PO4, O1 and O2. The

electrooculogram (EOG) was recorded from two pairs of leads in order

to register horizontal and vertical eye movements.

All epochs were carefully checked individually for artifacts and were

categorized with respect to items judged correctly as new for first

presentation words and as old at lag-2, lag-8 and lag-16. After rejecting

artifacts and erroneous trials, an average of 76.5 epochs remained for

first presentation words, 81.5 epochs for lag-2, 82.1 epochs for lag-8 and

83.6 epochs for lag-16.

Electrophysiological Variables and Data AnalysisFour different electrophysiological variables were calculated: (i) ERD/

ERS; (ii) spectral estimates for whole power; (iii) latencies; and (iv)

amplitudes of the N2, P2 and P3 of event-related potentials (ERPs).

Spectral estimates were calculated in steps of 0.25 Hz for the resting

period (of 3 min) with eyes closed. Statistical analyses are based on EEG

data for correctly recognized old items. ERD/ERS analysis is based on

a reference interval of 200 ms duration preceding (–600 to –400 ms)

stimulus presentation.

Latencies and Amplitudes of ERP ComponentsThe time windows for determining the N2, P2 and P3 component were

92--192, 192--352 and 352--660 ms, respectively. Components were

determined with the aid of a computer (by searching for the maximal

amplitude in the respective time windows) and then visually inspected

for artificial results.

Source Analyzing Methods: LORETALow-resolution electromagnetic tomography (LORETA; Pascual-Marqui

et al., 1994) was applied to individual ERPs to assess the underlying brain

electric sources of the scalp potentials. LORETA is an inverse solution

method that computes the three-dimensional distribution of neuronal

generators in the brain as a current density value (A/m2) for a total of

2394 voxels, with the constraint that neighboring voxels show maximal

similarity. All 30 electrodes were used for LORETA analyses.

Two time windows, one for the P2 (252--352 ms poststimulus)

and another for the P3 component (400--600 ms), were analyzed. The

same time windows were also used to assess the neural sources of

evoked delta and theta. For this analysis the raw data were bandpass-

filtered between 2 and 4 Hz for delta and between 4 and 6 Hz for theta.

After averaging over all respective trials for each subject, LORETA was

applied for the same time windows as for the LORETA analysis with

unfiltered data.

Statistical Data AnalysesAnalyses of variance (ANOVAs) were calculated to assess the influence

of lag on behavioral data, ERP components and ERD/ERS. To test the

influence of lag on the percentage of correct responses a two-way

ANOVA with the factors lag (2,8, 16) and judgement (very certain old,

certain old, uncertain old) was used. For ERP components, two-way

ANOVAs with the factors lag (2,8, 16) and component (N2, P2 and P3)

were calculated for each of the 10 recording sites. For ERD/ERS as

dependent measure, two-factorial ANOVAs with the factors lag (2,8, 16)

and time (t1 = 0--200, t2 = 200--400, t3 = 400--600, t4 = 600--800) were

calculated for each recording site and frequency band. Greenhouse

Geisser corrected results are reported where appropriate. The signif-

icance level for Scheffe tests was 5%. The use of electrode-specific

analyses is motivated by theoretical considerations and experimental

findings suggesting that the effects of trace strength are small and rather

localized in topography. We expect that parieto-temporal sites will play

a specifically important role.

For LORETA, the lag effect was analyzed within the time interval

for the P2 and P3 by a pairwise voxel-by-voxel comparison between

lag 2, lag 8 and lag 16. Statistical significance was assessed using a

nonparametric randomization test (Nichols and Holms, 2002). Results

were corrected for multiple comparisons on the cluster level.

Results

Behavioral Data

Overall recognition performance (percentage of hits) was

91.5%. The false alarm rate was at 6.94%. Collapsed over the

three ‘old’ judgement conditions, the percentage of correct

responses decreased from 95% for lag-2 to 93% for lag-8 down

to 85% for lag-16. The distribution of confidence judgements for

the lag-conditions is plotted in Figure 1 and shows that with

increasing lag, very certain old judgements decrease whereas

certain old and uncertain old judgements increase. These

findings are reflected by the ANOVA results with percent

correct answers as dependent measure showing highly sig-

nificant effects for the factors judgement [F (2,54) = 603.21;

P < 0.001], lag [F (2,54) = 37.76; P < 0.001] and the interaction

judgement 3 lag [F (4,108) = 51.54; P < 0.001]. Scheffe tests

calculated for each of the levels of factor judgement indicate

that for very certain old and certain old judgements, all

comparisons between the three lag conditions are significant

at an alpha level of 0.05. For uncertain old judgements, only the

comparison between lag-2 and lag-16 reached significance.

When using reaction time (RT) as the dependent measure,

the one-way ANOVA with factor LAG showed highly significant

effects [F (2,53) = 46.1; P < 0.01]. The respective means were

948.7, 1000.0 and 1083.2 ms for the lag-2, lag-8 and lag-16

conditions, respectively. Scheffe tests revealed that all lag

conditions differed significantly from each other.

282 Theta and Episodic Trace d Klimesch et al.



ownloaded from


ERP Data

As illustrated by Figure 2, the lag effect (comparison of lag-8 and

lag-16 with lag-2) is characterized by a more negative going ERP.

Differences between the lag conditions emerge around the P2,

but are most prominent during the positive going slope of the

P3 at ~500 ms. Inspection of Figure 2 indicates that at lag-16,

ERPs become almost as negative going as during the first

presentation.

ANOVAs reflect this finding and show that for the interaction

lag 3 component (which reached significance at all sites but

F4), lag-related differences in ERP components were significant

only for the P3 (tested with Scheffe’s test; see Table 1). Note

that the difference between lag-8 and lag-16 did not reach

significance in any of the comparisons. The only site showing no

lag-related differences at all was F4.

For the latencies of the ERP components, no significant main

effect for factor lag or the interaction lag 3 component were

obtained.

Selection of Frequency Bands

The ERD/ERS results at T5 and PO3 (depicted in Fig. 3),

averaged over the entire sample of subjects, were used in this

study to select those frequency bands between 2 and 14 Hz that

are most reactive. As Figure 3 illustrates, two particularly

reactive frequency regions can be observed, one between ~2and 6 Hz, showing pronounced and prolonged ERS (from

stimulus onset over the entire poststimulus interval, up to

1000 ms), and another between ~8 and 12 Hz, exhibiting

pronounced ERD over most of the post-stimulus interval. Delta

and theta were defined as bands with a width of 2 Hz, from 2 to

4 Hz and from 4 to 6 Hz respectively. Alpha was determined as

the broad band (around peak frequency of 10 Hz at PO3) from

8 to 12 Hz. Inspection of power (Fig. 3, lower row) demon-

strates nicely the drop of alpha power (i.e. alpha suppression or

desynchronization) in response to the presentation of a word in

a frequency range of 8--12 Hz.

Figure 2. Event-related potentials (ERPs) for first presentation, lag-2, lag-8, and lag-16 items. Note that with increasing lag, ERPs become more negative, particularly around thetime window of the P3. First presentation items were not subjected to statistical analysis.

Figure 1. Confidence judgements for correctly identified items (hits). Very certain oldjudgements decrease with lag, whereas certain and uncertain old judgements increasewith lag. We assume that the decrease in confidence with lag reflects the decay of anepisodic memory trace. New items (5 first presentation items) are plotted todocument the ERP old--new effect. Statistical analysis, however, was carried out forcorrect old items (with lag-2, lag-8 and lag-16) only.




ownloaded from


ERD/ERS

Statistical findings for ERD/ERS are summarized in Table 2 and

illustrated graphically in Figures 4 and 5. As already is evident

from Figure 3, the overall pattern is that delta and theta show

ERS but alpha shows ERD. This is reflected by factor TIMEwhich

is significant for all but two cases (T6 for delta and PO4 for theta;

see Table 2). The time course of ERS is different for delta

and theta. As exemplified in Figures 4 and 5, delta exhibits

Table 1ERP amplitudes

F3 Fz F4 T5 Pz T6 PO3 PO4 O1 O2

Lag, F (2,56) 5 4.34* 4.3* 3.5*Component, F (2,66) 5 57.6** 54.2** 71.0** 144.5** 100.2** 104.5** 122.0** 111.3** 85.2** 110.3**Lag 3 component, F (4,112) 5 4.8** 2.8* 7.5** 11.9** 6.1** 13.0** 8.8** 4.5** 5.9**Scheffe: significant for P3 only Lag-2[8 Lag-2[8;

lag-2[16Lag-2[8;lag-2[16

Lag-2[8;lag-2[16

Lag-2[8;lag-2[16

Lag-2[8;lag-2[16

Lag-2[8;lag-2[16

Lag-2[16 Lag-2[16

*Significant at the 5% level; **significant at the 1% level.

Figure 3. Time--frequency plots of ERD/ERS and power.

Table 2ERD/ERS

F3 Fz F4 T5 Pz T6 PO3 PO4 O1 O2

Delta ERSTime, F (3,81) 5 21.5** 28.1** 11.8** 4.1* 10.2** 9.3** 7.5** 6.2* 7.9**Lag, F (2,54) 5 3.3* 6.4** 4.7*Time 3 lag, F (6,162) 5 4.8**Scheffe t3: lag-2[8; lag-2[16

Theta ERSTime, F (3,81) 5 31.9** 39.9** 24.0** 5.1* 7.5** 24.6 5.0* 8.6** 6.7**Time 3lag, F (6,162) 5 3.4* 3.5* 3.0* 3.1* 4.0* 3.1*Scheffe t2: lag-2[16; t4: lag-16[8 t4: lag-16[8 n.s t4: lag-16[8 n.s. t4: lag516[8

Alpha ERDTime, F (3,81) 5 56.2** 48.1** 44.9** 94.5** 66.1** 150.9** 93.9** 114.9** 90.9** 103.9**Lag, F (2,54) 5 3.7*Time 3 lag, F (6,162) 5 3.7*Scheffe n.s.

t1, t2, t3, t4 are levels of factor time representing consecutive intervals of 200 ms each.

*Significant at the 5% level; **significant at the 1% level.




ownloaded from


a continuous increase up to the latest time interval (t4), whereas

theta shows maximal values for t2 (200--400 ms). Alpha ERD

shows a pronounced increase up to t3 (400--600 ms), with only

a marginal further increase up to t4 (600--800 ms).

For delta, a significant main effect for LAG was found at Fz, Pz

and PO4. Inspection of respective means reveals an almost

linear decrease in ERS from lag-2 to lag-16 over all time intervals.

The interaction time 3 lag at T5 reveals a decrease in delta ERS

with lag that is significant only at t3 between lag-2, lag-8 and

lag-16.

For theta, significant interactions between TIME and LAG

were found at T5, Pz, PO3, PO4, O1 and O2. Inspection of the

respective means indicate that during the early time intervals

(t1--t3) there is a tendency that theta ERS decreases with

increasing lag. The results of Scheffe comparisons indicate

that the lag-related decrease in ERS is significant already at t2between lag-2 and lag-16 at T5. It should be noted that in

contrast to lag-2 and lag-8, theta fails to drop with time for lag-16

during later time intervals. This difference is significant between

lag-8 and lag-16 at T5, Pz, PO4 and O2, as Scheffe comparisons

reveal.

For alpha, a significant lag effect was found for Pz. The

respective means indicate that ERD is slightly larger for lag-8

and lag-16 as compared with lag-2. Inspection of the respective

means for the significant interaction time 3 lag, obtained at T5,

also reveals a slight increase in ERD with increasing lag but only

for early time intervals, whereas at the late interval t4 this effect

reverses. Scheffe comparisons show that for none of the time

intervals, the differences in lag reach significance.

Lag-related Decrease in Delta ERS and P3

The fact that the lag-related decrease in delta ERS and P3 occur

in the same time interval (t3: 400--600 ms) and overlap at site T5

has led us to calculate correlations between the two measures.

Significant correlations were obtained between the decrease in

delta (measured as difference between lag 2 and lag 16) and P3

(amplitude difference between lag 2 and lag 16) at electrodes

T5 (r = 0.53; P < 0.01), T6 (r = 0.44; P < 0.05), PO3 (r = 0.40;

P < 0.05) and O2 (r = 0.49; P < 0.49).

LORETA Source Analysis

For the unfiltered P2 component, LORETA detected only one

small source in the left inferior parietal lobe. The strength of this

source showed a slight increase in strength for lag-16 as

compared with lag-2 (t > 4.13, P < 0.05 corrected). Differences

between lag-2 and lag-8 as well as lag-8 and lag-16 were not

significant.

However, when evoked theta in the P2 time window was

compared between lag-2 and lag-16, significantly higher current

density was found for lag-2 in bilateral superior parietal lobes,

left inferior parietal lobe, and right superior and middle frontal

Figure 4. ERS in the delta band at PO4 and T5. Delta ERS exhibits an increase at t1 of ~10% and at t3 -- t4 of ~20--25% (as compared with a reference preceding p1). In general,delta ERS decreases with lag. Only at T5 this decrease is restricted to t3.




ownloaded from


gyrus (t > 2.12; P < 0.05 corrected). The right frontal and

asymmetric parietal activation (favoring the left hemisphere)

are depicted in Figure 6. Stronger activation for lag-2 in right

superior parietal and right superior frontal gyrus was also found

when compared with lag-8. No significant lag effects regarding

the neural sources were obtained for delta.

The P3 component elicited stronger activity for lag-2 than lag-

16 in left superior temporal gyrus and middle temporal gyrus,

posterior cingulate gyrus, bilateral lingual gyrus and, most

interestingly, in right hippocampus and parahippocampal gyrus

(t > 3.77, P < 0.05 corrected). The respective findings are

summarized in Figure 7. There were no significant differences

between lag-2 and lag-8, or between lag-8 and lag-16.

In the time window of the P3, there was only slightly more

evoked theta activity for lag-2 than lag-16 in the right middle

frontal gyrus (t > 2.59, P < 0.05 corrected). No significant

differences were obtained for delta.

Discussion

ERPs show the expected typical pattern of results known from

other studies (e.g. Friedman, 1990a; Nielsen-Bohlman and

Knight, 1994; Chao et al., 1995; Guillem et al., 1999; Kim

et al., 2001). The general finding is that the P2 component is

sensitive to old--new differences and the P3 component is

sensitive to lag effects. Specifically, lag-related differences in

ERP amplitudes reached significance only for the comparatively

late P3 component with a latency of ~500 ms.

Evaluation of the episodic trace decay hypothesis is based on

the idea that during the access of an episodic trace in a time

window of ~200--600 ms, theta ERS decreases with lag. In good

agreement with the proposed hypothesis, theta ERS shows

a consistent decrease in ERS with lag during an early time

interval (t2 and/or t3) over most recording sites (cf. Fig. 5a),

which is statistically most reliable at T5 for interval t2 (200--400

ms; cf. Fig. 4b and Table 2). During the late interval t4 (600--800

ms), however, an increase in ERS can be observed for lag-16

(cf. Fig. 5). Because mean RT is ~1000 ms and includes not only

the old--new but also the confidence judgement and motor

response, it appears reasonable to assume that trace processing

is completed (at the latest) at ~500 ms. This conclusion is based

on the fact that typical reaction times vary between ~650 and

750 ms, depending on lag (e.g. Friedman, 1990b; Kim et al.,

2001), and on the assumption that execution of a motor

response requires ~200 ms. Consequently, the late increase in

theta ERS (during 600--800 ms) for lag-16 items might reflect

a decision related process. The interesting point here is that

theta exhibits a dissociation between task difficulty and time.

Whereas in the early time window t2 theta ERS for lag 16 is

significantly smaller than for lag 2, in the latest time interval t4ERS is larger for lag-16 than for lag-8 (cf. Fig. 5). The fact that the

Figure 5. ERS in the theta band exhibits a maximal increase already around t2. This is evident not only at T5 but also in the average over all electrodes. At T5 and time window t1theta ERS already is ~20% reaching a maximum of ~30% during t2. The lag-related decrease in ERS is restricted to t2 (cf. Table 2).




ownloaded from


difference between lag-16 and lag-2 did not reach signifi-

cance at t4 may be due to the generally higher level of ERS

for lag-2.

This dissociation suggests that theta reflects different stages

during the processing of an episodic trace. In an early time

window, theta reflects encoding (of new items) or access of an

existing trace (for old items), depending on the type of task. We

know from previous studies that successful (as compared with

not successful) encoding is reflected by a large theta ERS (e.g.

Klimesch et al., 1996, 1997a,b). Successful encoding is associ-

ated with a strong trace, and thus a large ERS reflects a strong

trace. The findings of the present study allow us to arrive at

a similar conclusion. With increasing lag, subjects show a de-

crease in recognition confidence (cf. Fig. 1), increased RTs and

a decrease in ERS. During a late interval (beyond ~500 ms), the

accessed trace will be evaluated by the subject. Evaluating

a weak trace is more difficult (e.g. in terms of spreading of

attentional resources) than a strong trace. Obviously, the late

theta ERS reflects the difficulty of this evaluation process. Thus,

theta reflects two different aspects of episodic retrieval,

episodic trace strength during an early time window and

evaluation of an episodic trace during a late time window.

We consider recognition confidence a behavioral variable

reflecting trace strength. In general, we assume that high

confidence is related to recollection and low confidence to

familiarity. Even for lag-16, certain and very certain judgements

account for ~80% of correct old responses (cf. Fig. 1). Thus, the

assumption that in the present experiment confidence reflects

episodic trace strength appears plausible. Variations in confi-

dence (within a lag condition), however, may also be due to

other variables that influence in trace strength, such as specific

encoding strategies or item difficulty. These variables are less

likely to play an important role in the present study because

items were presented at a fast rate (with an ISI of 3 s), subjects

could not know in advance whether an item would be repeated,

and items were controlled for word frequency, concreteness,

imageability, number of letters and number of syllables.

Most remarkably, the predictions of the trace decay hypoth-

esis were also confirmed by findings about delta. A lag-related

decrease in ERS was found at Fz, Pz, PO4 and T5, but at the latter

site for t3 only. The lack of a significant lag 3 time interaction at

Fz, Pz and PO3 seems to imply that already at the earliest time,

lag-related differences are significant. Inspection of Figure 4,

however, reveals that delta shows a rather continuous increase

from a pre-stimulus interval (–200 to 0ms) up to t4. A time smear

(in the range of half a theta cycle of ~150 ms) and between-

subject variability may have prevented the lag 3 time interaction

from reaching significance. Although not significant, lag-related

differences are more pronounced in late as compared with early

time windows (cf. Fig. 4).

In contrast to delta and theta, alpha shows a weak increase

in ERD with lag (at Pz and T5). The magnitude of this effect is

small — in the range of a few percentage points.

An increase in ERD or ERS may be interpreted in terms of

increased cortical activation (see e.g. Klimesch, 1999). Thus, the

expected decrease in cortical activation with lag — possibly

Figure 6. Current source density for evoked theta. Comparison between lag-2 and lag-16 items in the P2 time window (252--352 ms). Cortical regions marked in black indicatestronger current density for lag-2 than for lag-16. The LORETA image shows decreasing current source density with increasing lag bilaterally in the superior parietal lobes and overright dorsolateral prefrontal cortex.




ownloaded from


reflecting trace decay — can be observed only for theta and

delta. Given that the lag-related decrease in theta ERS is

significant for t2 (200--400 ms), we assume that the episodic

memory trace is accessed at ~300 ms. In agreement with this

assumption, the overall time course of theta ERS exhibits

a maximum at ~300 ms. Because delta ERS shows a gradual

increase over time with a maximum at t4 (600--800 ms), we

assume that delta reflects processes that are different from

those reflected by theta. It is possible that delta reflects the

evaluation of a memory trace in the sense of a ‘target detection/

evaluation’ process. Consistent with this interpretation is the

finding that the lag-related decrease in delta ERS (in the time

window of the P3) is significantly correlated with the lag-related

decrease in P3. This is also in agreement with the finding that

the ERP old--new effect (a P3 like component is larger for

correctly recognized old than new items) is generated by

frequencies in the delta range (Klimesch et al., 2000).

Most interestingly, LORETA revealed bilateral parietal and right

frontal sources for evoked theta — but not delta — at ~300 ms,

a time window related to the appearance of the P2 component.

These sources were weaker for lag-16 as compared with lag-2.

This is further evidence that theta plays an important role in

a time period of ~300 ms post-stimulus. For the unfiltered P2

component, LORETA detected a small parietal source that

increased with lag. In the time window of the P3, LORETA

reveals only marginally significant sources for evoked theta and

delta but large sources for the unfiltered P3. The P3 component

showed a lag-related decrease in temporal areas (including the

right hippocampal and parahippocampal regions), in the poste-

rior cingulate gyrus and the right and left bilateral lingual gyrus.

Figure 7. LORETA image for the (unfiltered) P3 component. Between 400 and 600 ms after stimulus onset stronger current source density can be obtained in lag-2 than in lag-16trials over the left superior temporal lobe, bilateral lingual gyrus, right parahippocampal gyrus and hippocampus, and the posterior cingulate gyrus.




ownloaded from


The parallel fMRI study by Brozinsky et al. (2005) showed

that, compared with lag-2 items, lag-16 items are associated

with an increase in the BOLD signal in different regions of the

temporal lobe, including the left rhinal cortex, and the left

posterior and right posterior hippocampus. Consistent with

single-unit studies demonstrating repetition suppression

(Brown and Xiang, 1998), lag-2 and lag-8 items exhibited

a decrease in the BOLD signal when compared with new items.

No such repetition suppression was found for lag-16 and lag-32

items. With respect to topography, there is some overlap with

the sources of the P3 in the present study within regions of the

temporal lobe. However, the direction of lag-related changes in

signal strength are different. The BOLD signal increases with lag,

but the P3 amplitude decreases with lag. Inspection of the ERPs

depicted in Figure 2 reveals that the P3 is characterized by

a slow component between ~400 and 600 ms (that may be

interpreted as a half period of a slow wave with a frequency

close to 2.5 Hz). In assuming that a positive going wave reflects

inhibitory processes whereas a negative going wave indicates

excitatory processes (Elbert et al., 1981), the significant de-

crease in P3 amplitude with lag (cf. Fig. 2) suggests that

excitation increases with lag. Thus, in theory, both the lag-

related increase in the BOLD signal and the lag-related increase

in (relative) negativity (decrease in the P3 amplitude) may

reflect similar processes that are related to the evaluation and/

or further processing of the accessed trace which becomemore

difficult when the trace decays.

It should also be noted that the comparatively fast and

transient decrease in theta during the access of an

episodic trace most likely cannot be detected with fMRI.

We assume that a parietal network found for evoked theta

(cf. Fig. 6) is related to the access of an episodic trace.

The functional meaning of the early right frontal source is

difficult to interpret, but might reflect (consistent with the

HERA model; Tulving et al., 1994) a control process

related to episodic retrieval. Sources in medial temporal

areas, found in a later time window, may mediate the

further processing of the retrieved trace (cf. Fig. 7).

We hypothesize that the sequence of encoding can be

described as follows. Up to the time window of the N1

(~160 ms) a sensory code is established, then at ~300 ms

(in the time window of the P2) the episodic trace is

processed. The parietal--frontal network detected by LOR-

ETA for evoked theta most likely reflects two different

processes. Whereas the right frontal source may be related

to the process of retrieval (this is consistent with Tulving’s

HERA model; Tulving et al., 1994), early episodic trace

processing takes place at parietal areas. Then, during the

time window of the P3, the episodic trace is processed

further in areas of the temporal lobe (including the

hippocampal formation). The proposed hypothesis is dif-

ferent from the traditional view with respect to the early

time that the episodic trace can be accessed. ERP studies

have demonstrated that episodic processes are related to

the comparatively late window of a P3-like component

with a latency between 400 and 800 ms (e.g. Allan et al.,

1998; Fernandez et al., 1999; Mecklinger, 2000). Our

conclusion, however, is based on findings about theta

which are consistent with data from other laboratories.

As an example of a recent study, Duzel et al. (2003) found

an early parietal theta old--new effect in a time window

around 300 ms.

Notes

This research was supported by the Austrian Science Fund (FWF),

P-16849.

Address correspondence to Univ. Prof. Dr Wolfgang Klimesch,

Department of Physiological Psychology, Institute of Psychology,

University of Salzburg, Hellbrunnerstr. 34, A-5020 Salzburg, Austria.

Email: [email protected].

References

Aftanas L, Golocheikine S (2001) Human anterior frontal midline theta

and lower alpha reflect emotionally positive state and internalized

attention: high-resolution EEG investigation of meditation. Neurosci

Lett 310:57--60.

Allan K, Wilding EL, Rugg MD (1998) Electrophysiological evidence for

dissociable processes contributing to recollection. Acta Psychol

98:231--252.

Araujo D, Baffa O, Wakai R (2002) Theta oscillations and human

navigation: a magnetoencephalography study. J Cogn Neurosci

14:70--78.

Asada H, Fukuda Y, Tsunoda S, Yamaguchi M, Tonoike M (1999) Frontal

midline theta rhythms reflect alternative activation of prefrontal

cortex and anterior cingulated cortex in human. Neurosci Lett

274:29--32.

Baayen RH, Piepenbrock R, Gulikers L (1995) The CELEX lexical

database. Philadelphia, PA: Linguistic Data Consortium University

of Pennsylvania.

Brown MW, Xiang JZ (1998) Recognition memory: neuronal substrates

of the judgement of prior occurrence. Prog Neurobiol 55:149--189.

Brozinsky CJ, Kroll NEA, Ranganath C, Yonelinas AP (2005) Short

lived repetition effects throughout the medial temporal lobes.

Hippocampus (in press).

Burgess A, Gruzelier JH (1997) Short duration synchronization of

human theta rhythm during recognition memory. Neuroreport

8:1039--1042.

Burgess A, Gruzelier JH (2000) Short duration power changes in the EEG

during recognition memory for words and faces. Psychophysiology

37:596--606.

Chao LL, Nielsen-Bohlman L, Knight RT (1995) Auditory event-related

potentials dissociate early and late memory processes. Electro-

encephalogr Clin Neurophysiol 96:157--168.

Duzel E, Habib R, Schott B, Schoenfeld A, Lobaugh N, McIntosh AR,

Scholz M, Heinze H J (2003) A multivariate, spatiotemporal analysis

of electromagnetic time-frequency data of recognition memory.

Neuroimage 18:185--197.

Elbert T, Lutzenberger W, Rockstroh B, Birbaumer N (1981) The

influence of low-level transcortical DC-currents on response speed

in humans. Int J Neurosci 14:101--114.

Fernandez G, Effern A, Grunwald T, Pezer N, Lehnertz K, Dumpelmann

M, Van Roost D, Elger CE (1999) Real-time tracking of memory

formation in the human rhinal cortex and hippocampus. Science

285:1582--1585.

Fingelkurts A, Fingelkurts A, Krause C, Sams M (2002) Probability

interrelations between pre-/post-stimulus intervals and ERD/ERS

during a memory task. Clin Neurophysiol 113:826--843.

Friedman D (1990a) Cognitive event-related potential components

during continuous recognition memory for pictures. Psychophysi-

ology 27:136--148.

Friedman D (1990b) ERPs during continuous recognition for words. Biol

Psychol 30:61--87.

Gevins A, Smith ME, McEvoy L, Yu D (1997) High-resolution EEG

mapping of cortical activation related to working memory: effects of

task difficulty, type of processing, and practice. Cereb Cortex

7:374--385.

Gevins A, Smith ME, Leong H, McEvoy L, Whitfield S, Du R, Rush G

(1998) Monitoring working memory load during computer-based

tasks with EEG pattern recognition methods. Hum Factors 40:79--91.

Gevins A, Smith M (2000) Neurophysiological measures of working

memory and individual differences in cognitive ability and cognitive

style. Cereb Cortex 10:829--839.




ownloaded from


Grunwald M, Weiss Th, Krause W, Beyer L, Rost R, Gutberlet I, Gertz H

(1999) Power of theta waves in the EEG of human subjects increases

during recall of haptic information. Neurosci Lett 260:189--192.

Guillem F, Rougier A, Claverie B (1999) Short- and long-delay in-

tracranial ERP repetition effects dissociate memory systems in the

human brain. J Cogn Neurosci 11:437--458.

Hintzman DL, Caulton DA, Levition DJ (1998) Retrieval dynamics in

recognition and list discrimination: further evidence of separate

processes of familiarity in recall. Mem Recogn 26:449--462.

Ishii R, Shinosaki K, Ukai S, Inouye T, Ishihara T, Yoshimine T, Hirabuki

N, Asada H, Kihara T, Robinson S, Takeda M (1999) Medial prefrontal

cortex generates frontal midline theta rhythm. Neuroreport

10:675--679.

Ishihara T, Yoshii N (1972) Multivariate analytic study of EEG andmental

activity in juvenile delinquents. Electroencephalogr Clin Neuro-

physiol 33:71--80.

Jensen O, Tesche C (2002) Frontal theta activity in humans increases

with memory load in a working memory task. Eur J Neurosci

15:1395--1400.

Kahana MJ, Sekuler R, Caplan JB, Kirschen M, Madsen JR (1999) Human

theta oscillations exhibit task dependence during virtual maze

navigation. Nature 399:781--784.

Kahana MJ, Seelig D, Madsen JR (2001) Theta returns. Curr Opin

Neurobiol 11:739--744.

Kim MS, Kim JJ, Kwon, JS (2001) The effect of immediate and delayed

word repetition on event-related potential in a continuous recogni-

tion task. Cogn Brain Res 11:387--396.

Klimesch W (1999) EEG alpha and theta oscillations reflect cognitive

and memory performance: a review and analysis. Brain Res Brain Res

Rev 29:169--195.

Klimesch W, Schimke H, Schwaiger J (1994) Episodic and semantic

memory: an analysis in the EEG-theta and alpha band. Electro-

encephalogr Clin Neurophysiol 91:428--441.

Klimesch W, Doppelmayr M, Russegger H, Pachinger T (1996) Theta

band power in the human scalp EEG and the encoding of new

information. Neuroreport 7:1235--1240.

Klimesch W, Doppelmayr M, Pachinger T, Ripper B (1997a) Brain

oscillations and human memory performance: EEG correlates in the

upper alpha and theta bands. Neurosci Lett 238:9--12.

Klimesch W, Doppelmayr M, Schimke H, Ripper B (1997b) Theta

synchronization in a memory task. Psychophysiology 34:169--176.

Klimesch W, Doppelmayr M, Schwaiger J, Auinger P, Winkler T (1999)

‘Paradoxical’ alpha synchronization in a memory task. Cogn Brain Res

7:493--501.

Klimesch W, Doppelmayr M, Schwaiger J, Winkler T, Gruber W (2000)

Theta oscillations and the ERP old/new effect: independent phe-

nomena? Clin Neurophysiol 111:781--793.

Klimesch W, Doppelmayr M, Stadler W, Poellhuber D, Sauseng P,

Roehm D (2001a) Episodic retrieval is reflected by a process

specific increase in human electroencepaholographic theta activity.

Neurosci Lett 302:49--52.

Klimesch W, Doppelmayr M, Yonelinas A, Kroll NEA, Lazzara M, Roehm

D, Gruber W (2001b) Theta synchronization during episodic re-

trieval: neural correlates of conscious awareness. Cogn Brain Res

12:33--38.

Krause CM, Lang HA, Laine M, Kuusisto MJ, Poern B (1996) Event-related

EEG desynchronization and synchronization during an auditory

memory task. Electroencephalogr Clin Neurophysiol 98:319--326.

Krause CM, Salminen P, Sillanmaeki L, Holopainen I (2001) Event-related

desynchronization and synchronization during a memory task in

children. Clin Neurophysiol 112:2233--2240.

Kubota Y, Sato W, Toichi M, Murai T, Okada T, Hayashi A, Sengoku A

(2001) Frontal midline theta rhythm is correlated with cardiac

autonomic activities during the performance of an attention

demanding meditation procedure. Cogn Brain Res 11:281--287.

Mecklinger A (2000) Interfacing mind and brain: a neurocognitive

model of recognition memory. Psychophysiology 37:565--582.

Mecklinger A, Kramer AF, Strayer D (1992) Event-related potentials and

EEG components in a semantic memory search task. Psychophysi-

ology 29:104--119.

Nielsen-Bohlman L, Knight RT (1994) Electrophysiological dissociation

of rapid memory mechanisms in humans. Neuroreport 5:1517--1521.

Nichols TE, Holmes AP (2002) Nonparametric permutation tests for

functional neuroimaging:a primer with examples. Hum Brain Map

15:1--25.

Pasqual-Marqui RD, Michel CM, Lehmann D (1994) Low resolution

electromagnetic tomography: a new method for localizing electrical

activity in the brain. Int J Psychophysiol 18:49--65.

Pfurtscheller G, Lopes da Silva FH (1999) Handbook of EEG and clinical

neurophysiology: event-related desynchronization. Amsterdam:

Elsevier.

Raghavachari S, Kahana M, Rizzuto D, Caplan J, Kirschen M, Burgeois B,

Madsen J, Lisman J (2001) Gating of human theta oscillations by

a working memory task. J Neurosci 21:3175--3183.

Sarnthein J, Petsche H, Rappelsberger P, Shaw GL, von Stein A (1998)

Synchronization between prefrontal and posterior association

cortex during human working memory. Proc Natl Acad Sci USA

95:7092--7096.

Sauseng P, KlimeschW, Gruber W, Doppelmayr M, Stadler W, Schabus M

(2002) The interplay between theta and alpha oscillations in the

human electroencephalogram reflects the transfer of information

between memory systems. Neurosci Lett 324:121--124.

Sauseng P, Klimesch W, Doppelmayr M, Hanslmayr S, Schabus M, Gruber

W (2004) Theta coupling in the human electroencephalogram

during a working memory task. Neurosci Lett 354:123--126.

Sauseng P, Klimesch W, Schabus M, Doppelmayr M (2005) Fronto-

parietal EEG coherence in theta and upper alpha reflect central

executive functions of working memory. Int J Psychophysiol

(in press).

Tesche C, Karhu J (2000) Theta oscillations index human hippocampal

activation during a working memory task. Proc Natl Acad Sci USA

97:919--924.

Tulving E (1983) Elements of episodic memory. New York: Oxford

University Press.

Tulving E (1985) Memory and consciousness. Can Psychologist 25:1--12.

Tulving E, Kapur S, Craik FI, Moscovitch M, Houle S (1994) Hemispheric

encoding/retrieval asymmetry in episodic memory: positron emis-

sion tomography findings. Proc Natl Acad Sci USA 91:2016--2020.

von Stein A, Rappelsberger P, Sarnthein J, Petsche H (1999) Synchro-

nization between temporal and parietal cortex during multimodal

object processing in man. Cereb Cortex 9:137--150.

Weiss S, Rappelsberger P (2000) Long-range EEG synchronization

during word encoding correlates with successful memory perfor-

mance. Cogn Brain Res 9:299--312.

Yonelinas A (2002) The nature of recollection and familiarity: a review

of 30 years of research. J Mem Lang 46:441--517.




ownloaded from


oscillatory eeg correlates of episodic trace decay

Documents