NIH MH064498 (B. R. P.), NRSA GM007507 (UW-Madison NTP)

Relating individual differences in short-term memory-derived EEG to cognitive training effectsBornali Kundu, David W. Sutterer, Bradley R. Postle

Neuroscience Training Program, Departments of Psychology and Psychiatry, University of Wisconsin - MadisonNeuroscience Training Program, Departments of Psychology and Psychiatry, University of Wisconsin - Madison

ResultsResultsbornalikundu@gmail.com

Does training on a working memory task show systematic task-related changes in an individualʼs delay-period

activity?

IntroductionIntroduction

Does training on a working memory task show changes in cortical effective connectivity?

+

++

++

++

+

+ +

Fixation

Targets

Delay Probe

1000 ms 3750 ms 2000 ms

ITI

1000 ms

400 ms200 ms

..

..

37500

TMS TMS

2000 ±250

Voltage (i.e., ERP) measures of the delay-period during a working memory task show considerable individual differences that correlate with memory span (Vogel et al., 2004)

MethodsMethods

Spectral measures of delay-period EEG activity show considerable in-dividual differences and are also stable and trait-like (Kundu et al., SFN 2010 poster, shown above)

Training on delayed-recall tasks causes changes in BOLD and FA measures that localize to fronto-parietal brain regions (Olesen et al, 2004; Takeuchi et al., 2010)

Prolonged, adaptive training on WM tasks improves performance on the task itself, as well as on nonmnemonic tests of general fluid intel-ligence (gF), with the largest gains seen in low gF individuals (Jaeggi et al., 2008)

Pre-Post Spatial Delayed Recognition EEG and TMS-EEGPre-Post Spatial Delayed Recognition EEG and TMS-EEG

There is change in neural contralateral delay activity (CDA) between pre- and post-training sessions. Increased separation of load-related CDA (loads 6, 4, and 2) post-training vs pre-training for experi-mental compared to control subjects.

TrainingTrainingERSP(dB)

−5

0

5

0 1000 2000 3000−5

5

Time (ms)

dB

10

20

30

40

50

20 40dB

0 1000 2000 3000−5

5

Time (ms)

10

20

30

40

50

20 40

Freq

uenc

y (H

z)

dB

Subject 1

ERSP(dB)

−2

0

2

0 1000 2000 3000−4

2

Time (ms)dB

10

20

30

40

50

20 40dB

0 1000 2000 3000−42

Time (ms)

10

20

30

40

50

20 40

Freq

uenc

y (H

z)

dB

Subject 2

2 Groups : Dual N-back training 2 Groups : Dual N-back training (n=3; Brain Workshop http://brainworkshop.sourceforge.net/) and control trainingcontrol training (n=3 ; Tetris http://www.gosu.pl/tetris/); randomized. Both groups trained 40 minutes per day, 5 times per week, for 5 weeks. The control task does not have overt memory de-mands. Subjects were assessed pre- and post- training by select measures.

0 5 10 15 20 251.5

2

2.5

3

3.5

4

4.5

5

5.5Dual N-Back Training Performance

0 5 10 15 20 25−2

−1.5

−1

−0.5

0

0.5

1

1.5

2

2.5

3

Subject 3Subject 4Subject 5

Control Task (Tetris) Performance

Sum

Tot

al H

igh

Scor

e (Z

nor

m)

Training DayTraining Day

Mea

n N

Lev

el R

each

ed (r

aw)

Subject 1Subject 2Subject 6

5 10 15 20 25

−3

−2.5

−2

−1.5

−1

−0.5

0

0.5

1

Subject 1 (smooth)Subject 2 (smooth)Subject 6 (smooth) fit y=A*exp^(-b*x)+c

Dual N-Back Training Performance

Mea

n N

Lev

el R

each

ed (Z

nor

m)

slope: “Growth Rate”

Avg train-ing gain

Learning on both tasks well fit by power law. Data were smoothed with moving average window over 5 day span. Comparable growth rates for both tasks implies effective learning. Mean b(nback)b(nback)=0.0012, adj R =0.89; mean b(tetris)b(tetris) = 0.0011, adj R =0.747. Experimental group showed positive correlation be-tween training gains and delay period power in the theta (4-7 Hz) and alpha (8-14 Hz) bands. Negative correlation found between gains and gamma (26-50 Hz) band power. Control group showed positive cor-relations between training gain and alpha band power. No relationship was apparent between gains and beta (15-25 Hz) band power for either group.

−200 0 200 400 600 800

−2

−1

0

1

2P5/P6

time (ms)

uV

−200 0 200 400 600

800−2

−1

0

1

2P5/P6

time (ms)

uV

−200 0 200 400 800−3−2−10123 avg

600

Control

time (ms)−200 0 200 400 600 800

−3−2−10123 avg

Experimental

time (ms)

Pre-Post Psychometric MeasuresPre-Post Psychometric Measures

Pre-Post “Change Detection” ERP MeasuresPre-Post “Change Detection” ERP Measures

Post-TrainingPre-Training

Load 2 Load 4Load 6

Comparison of CDA amplitude for Load 4 minus Load 2 during the pre- and post- training sessions

−2000 −1000 0 1000 2000 3000

5

Time (ms)

dB

10

20

30

40

50

20 40

Freq

uenc

y (H

z)

dB

ERSP(dB)

−5

0

5

−2000 −1000 0 1000 2000 3000−5

5

Time (ms)

dB

10

20

30

40

50

20 60

Freq

uenc

y (H

z)

dB

Pre-training DPSP Post-training DPSP

−2000 −1000 0 1000 2000 3000 −5

5

Time (ms)

dB

10

20

30

40

50

20 60

Freq

uenc

y (H

z)

dB

Subject 6

Subject 4

Subject 6

Subject 4

Pre-trainingPost-training

uA/mmSCD

SCS mm

bPL

Geodesic Distance toTMS mm

2

2

Measure

K

RAPM

OSPAN

Delay period event related spectral per-turbation (DPSP) plots for Load 4 Le� target condi�on, chan-nel P6, for 2 sub-jects. Subject 6 (top) is experimental. Subejct 4 is control. Subject-specific spectral patterns remain stable even with training pertur-bation.

Left: TMS evoked response for same 2 subejcts is also stable and trait-like.

Below: Absolute change in power between pre- and post-training delay period EEG divided into theta (4-7Hz), alpha (8-14Hz), beta (15-25Hz) and gamma (26-50Hz) bands. The fronto-parieto-occipital topographic dis-tribution of absolute power change is reminis-cent of results of past studies. Spefically, there are changes in frontal mid-line theta and posterior alpha power.

L target

R target

Experimental ControlTheta Alpha Beta Gamma Theta Alpha Beta Gamma

Global SCD

Global SCS

Global bPL

Expt

Expt

Expt

Cntrl

Cntrl

Cntrl

Pre-trainingPost-training

Global measures of effective con-nectivity (Significant Current Density (SCD); Significant Current Scatter (SCS); and broadband Phase Locking (bPL) (Casali et al.) Experimental subjects show smaller fluctuations across measures and less variation within pre- and post-training ses-sions.

Above: Effec�ve conec�vity measures for a single subject.

Pre-training Post-training

ConclusionsConclusionsTraining on a demanding working memory task :

1. Gains predicted by pre-training EEG (low-freq power).2. Generalizes to other mea-sures, including K and gF.3. ERP correlate of K, the CDA, sharpens with training.4. EEG suggests fronto- (theta) parietooccipital (alpha) strengthening.5. TMS suggests systematic changes in the cortical effective connectivity.

Subject 6 (Expt) Pre-training Subject 6 (Expt) Post-training

uV

Session

uV

ERSP(dB)

−5

0

5

−2000 −1000 0 1000 2000 3000−5

5

Time (ms)

dB

10

20

30

40

50

20 60

Freq

uenc

y (H

z)

dB

−5

10

0uV

uV

-10

-10

10

0

-100 0 100 200 300 400 500 600�me(ms)

-100 0 100 200 300 400 500 600 -100 0 100 200 300 400 500 600

-100 0 100 200 300 400 500 600

�me(ms) �me(ms)

�me(ms)

10

0uV

-10

10

0uV

-10

uV

Pre- and post- training measuresPre- and post- training measures

Electrophysiological measures: Electrophysiological measures: 1. Spatial delayed-recognition task: Serial presentation of 2 or 4 identical square stimuli in different locations with lateralized presentation (randomized). Subjects were instructed to remember the locations marked by each stimulus in the cued hemifield. 160 trials per session. TMS was delivered to left SPL for 50% of trials (randomized). 2. Change-detection task: Stimuli were presented simultaneously. Load 2, 4, and 6 were tested (randomized). Lateralized display. Subjects were instructed to remember color and location of colored square stimuli in the cued hemifield. 200 trials per con-dition. Task parameters replicated from Vogel et al., 2004.

Psychometric Measures:Psychometric Measures:1. Short term memory capacity (K value) derived from change detection task. K = S(H-FA)2. Ravenʼs Advanced Progressive Matrices (RAPM; Raven, 1990)3. Operation Span (OPAN; Turner & Engle, 1989)

TMS/EEG:TMS/EEG:Recorded with a 60-channel TMS-compatible amplifier (Nexstim, Helenski, Finland).Sample-and-hold circuit holds amplifier output constant from 100 us to 2 ms post-stimulus. Data were acquired at 1450 Hz, downsampled to 500 Hz and filtered (0.1-80 Hz) offline. All data processing was done with a combination of MATLAB (Mathworks, Inc), EEGLAB and ERPtoolbox (USCD) and, Fieldtrip (Donders Institute, Ni-jmegen). Effective connectivity analysis methods follow Casali et al. 2010. Data-driven mea-sures include Significant Current Density (SCD; local regional measure); Significant Current Scatter (SCS; long range distance measure), and broadband phase locking (bPL; temporal measure). Mathematically orthogonal.

22