Supplementary Information (doc 218K)

SUPPLEMENTARY MATERIALS
The effects of nicotine replacement on cognitive brain activity during smoking
withdrawal studied with simultaneous fMRI/EEG
John D. Beaver, PhD*1, Christopher J. Long, PhD1, David M. Cole, MSc1,2, Michael J.
Durcan, PhD3, Linda C. Bannon, BSc3, Rajesh G. Mishra, MD, PhD3, Paul M. Matthews,
MD, DPhil1,2
1. GlaxoSmithKline Clinical Imaging Centre, Hammersmith Hospital, London, UK.
2. Department of Clinical Neuroscience, Imperial College, London, UK.
3. GlaxoSmithKline Consumer Healthcare, Weybridge, UK.
SUPPLEMENTARY TABLE 1. Sequence of events at each treatment visit.
Approximate
time of day
06:30
Time
relative to
first dosing
Activity
Food Intake
Enter unit, Smoke cigarette
Drugs of abuse, CO and alcohol
assessments
07:30 - 08:30
At leisure
Breakfast
08:30
RVIP task
traininga/refamiliarisationb
09:30 – 13:30
At leisure
Lunch
13:30
RVIP task practice (inside
inactive scanner)
14:15
Modified Minnestota Withdrawal
Symptoms completed.
14:30
0h
Dose (4mg nicotine lozenge or
placebo)
+1 hr 50 min Modified Minnesota Withdrawal
Symptoms completed.
Adverse Events.
16:30
+2 hr
Dose (4mg nicotine lozenge or
placebo)
+2 hr 30 min Positioned in MR scanner
+3 hr
fMRI/EEG during RVIP task
+3 hr 50 min Removal from MR scanner.
Modified Minnesota Withdrawal
Symptoms completed.
18:30
Discharge from unit
Dinner
a
Training in cognitive tests conducted at treatment visit 1 only.
b
Refamiliarisation with cognitive tests conducted at treatment visit 2 only.
EEG Preprocessing. Acquisition of EEG during fMRI leads to two major types of artefact
(1) gradient induced and (2) Ballistocardiogram (BCG) effects. (1) Removal of the machine
gradient artefact first detects and places a marker to indicate the start acquisition time for
each fMRI volume. Subsequent steps involve the construction and subtraction of a moving
average artefact template that is estimated from progressive windows of the recorded data.
This imaging artefact tends to exhibit stability in time, providing confidence that the average
template estimates are good representations of the artefact present at each slice acquisition.
Template construction was facilitated by synchronisation during acquisition of the EEG with
the high-frequency MR master clock trigger. This phase synchrony ensures that the recorded
scanner artefact does not drift in time with respect to the EEG data. (2) BCG removal: the
ECG signal was processed with the BrainProducts software to identify robust QRS
complexes and then mark the most temporally localised BCG feature to improve delay
estimation between the measured ECG and the BCG. These markers were then manually
inspected and adjusted or removed depending on the accuracy of the first pass detection.
Once the ECG markers were judged satisfactory, a running average BCG template was
constructed for each channel and subtracted off each ECG epoch in a similar manner as the
gradient removal step. The ECG complex tends to vary significantly in time, which can lead
to residual artefact in the data. Therefore, to further mitigate the effect of the BCG in the
EEG data further (manuscript reference: Phillips et al), we used the frontal electrodes to
inspect the efficacy of the removal process and decomposed the contributing sources of
variance using an Independent Component Analysis step to isolate any remaining BCG
artefact. To help identify the key ICA components,we segmented each ICA component using
the QRS marker. Component channels showing a high degree of contamination with respect
to the ECG marker could then be nulled and back-filtered out of the reconstructed EEG data.
Below we show the alpha power spectral density for the FP1 and FP2 channels, we
still see a good correlation of this waveform (albeit less strongly than the occipital channels)
with respect to the cognitive task (Supplemental Figure 1). This substantiates our belief that
the estimated alpha waveform is correlated with task rather than exhibiting BCG related
effects that would be more spurious in nature.
Supplemental Figure 1. Alpha PSD for the FP1 and FP2 channels overlaid on the cognitive
task model.
SUPPLEMENTARY FIGURE 2. Alpha power derived from the EEG recording during the
behavioural task. We decompose the time-course into two components – the first representing
the mean fMRI block design and the second representing neural fluctuation about the mean.
We make the assumption that, in part, electrophysiological activity gives rise to the observed
BOLD signal and we use this knowledge to posit an (individualised) linear model that more
closely reflects the true neuronal activity during the task. For example, simultaneous
knowledge of global electrophysiological activity via EEG alpha power measurements can
provide information about how brain function might attenuate or increase over the course of
the task. In standard fMRI analysis, such fluctuation information is assumed constant across
the experiment. So the standard model can be augmented by partitioning the alpha power into
two pieces – one to capture the mean effect and any departures from this mean effect. Earlier
work investigating subjects at rest directly correlated the spontaneous changes in the EEG
regressor with each fMRI time-course. However, this approach is not optimal for the present
study as it does not permit estimation of the key RVIP contrasts. The orthogonalization
procedure maintains this flexibility by decoupling the main effect from the fluctuations
around the mean and includes both as regressors in the design matrix.
SUPPLEMENTARY FIGURE 3. Example time-course plot depicting the close correlation
of the BOLD fMRI and alpha EEG signals in the dorsal anterior cingulate of a single
participant. White = the general linear model (GLM) of the task block design; Red = the
BOLD fMRI signal intensity estimate; Green = the alpha power spectral density (PSD)
estimate. The alpha PSD has been inverted for illustrative purposes. Note that the actual
relationship between BOLD and alpha is negative.