The Impact of Norepinephrine on the Fidelity of Neural Representations

The locus coeruleus (LC) is the brainstem neuromodulatory nucleus responsible
for most of the norepinephrine (NE) released in the brain. It has widespread
projections throughout the neocortex. When an animal is actively engaged in
performing a task, LC neurons exhibit a rapid, phasic increase in discharge
rate to task-relevant and otherwise motivationally salient stimuli. The ensuing
release of NE in cortical areas temporarily increases the responsivity (or
gain) of these areas to their afferent input (Berridge & Waterhouse, 2003),
selectively potentiating any activity present concurrent with LC activation.
Some researchers have suggested that this transient, LC-induced increase in
responsivity serves to optimize simple decision making (i.e., selecting a
response based on perceptual evidence) by increasing the signal-to-noise ratio
(SNR) of signal processing in the cortex (e.g. Usher, Cohen, Servan-Schreiber,
Rajkowski & Aston-Jones, 1999; Aston-Jones & Cohen, 2005). However, other
researchers have suggested that increases in NE levels serve to facilitate a
network reset that allows a complete reorganization of neural activity to
facilitate a new set of responses (e.g. Bouret & Sara, 2005; Yu & Dayan 2005;
Dayan & Yu, 2006). Still others have focused on the role of NE in long-term
potentiation (e.g. Harley, 2007). To date there has been scarce research
directly investigating the impact of NE on human cognition. We propose a
psychopharmacological study that uses cutting edge fMRI analysis techniques to
directly test if NE increases the SNR of cortical processing. In addition, to
fully take advantage of the long duration of our psychopharmacological
manipulation, we will have subjects perform a few additional tasks aimed at
testing other proposed effects of NE.

1.1 Norepinephrine and Signal-to-Noise Ratio
Norepinephrine enhances the responsivity of neurons to both excitatory and
inhibitory inputs, effectively increasing the gain of the neuron
(Serban-Schreiber, Prinz, & Cohen, 1990). Serban-Schreiber and colleagues
demonstrated that when a homogenous increase in gain is applied to all
composite neurons in a neural network, the SNR of the network is improved, such
that signals are enhanced and noise is decreased. In target detection tasks,
this means there are fewer misses (strong signals) and fewer false alarms (less
noise). It follows from this that if the neural network is representing
information, it should represent that information with greater precision, or
fidelity. That is, there should be less variability (noise) in the way the
network represents the same information from trial to trial than when the SNR
is lower. A new technique of analyzing fMRI data, called multivariate pattern
analysis (MVPA) offers a direct way of testing this hypothesis.

1.2 MVPA and Representational Space
MVPA is a method of extracting more information from fMRI data than traditional
analyses, by taking into account the relative activity of voxels in an area of
interest (e.g. Mur, Bandettini, and Kriegeskorte, 2009; Pereira, Mitchell, and
Botvinick, 2009). By plotting the relative activation in each voxel of interest
relative to others, a neural network state can be represented as a single point
in multidimensional space (this representational space will have as many
dimensions as there are voxels in the analysis). In this multi-dimensional,
representational space, network states provoked by the same category of event
will tend to cluster together, whereas network states elicited by different
categories will tend to be distal from one another. Haxby and colleagues (2001)
used this powerful technique to reliably discriminate between the brain
activity elicited by eight categories of visual stimuli: houses, faces, cats,
chairs, scissors, shoes, jugs, and nonsense pictures. That is, they could
reliably predict from the brain activity which category of stimulus the
participants viewed on any given trial. As an even greater demonstration of the
information that can be obtained with MVPA, Haynes and colleagues (2007) were
able to reliably distinguish between trials where subjects chose to add two
numbers together, versus trials where the subject chose to subtract one number
from the other. These examples of classification involved determining which
category a trial belonged to by determining which category cluster the network
state was closest to: Distance in multi-dimensional space is a direct measure
of the similarity between representations.

1.3 The Fidelity of Neural Representations
Schurger, Pereira, Treisman, and Cohen (2010) used distance in representational
space as a measure of the fidelity of a neural representation. Instead of
examining the separation between clusters of network states, they examined
separation within clusters. A cluster of network states can be diffuse and
spaced out widely, or it can be tight and compact. A tightly spaced cluster
indicates that the neural representation varied very little across multiple
trials of the same category, rather like the tightly spaced arrows of an expert
archer. This measure of the fidelity of the neural representation (they used
the term reproducibility) was the measure that significantly differed between
consciously and sub-consciously perceived visually degraded stimuli. Schurger
and company offered the conjecture that the tighter clusters meant the neural
representation was more robust. We interpret the effect as a stronger, more
prominent signal and a decreased impact of noise. As such, an increase in SNR
should have the effect of producing higher precision in neural representation.

1.4 The Effect of NE on Neural Representation
The spacing of clusters of network states representing a category of everyday
objects, such as houses, should be influenced by both real variation in the
representation of different pictures of houses, and by variation due to
unrelated processing (noise). If NE enhances the SNR (reducing the influence of
noise), then the cluster should get tighter as cortical levels of NE increase.
Thus, we will manipulate NE levels pharmacologically with atomoxetine, with the
hypothesis that the increased levels of cortical NE produced by atomoxetine
(e.g. Bari & Aston-Jones, 2012) will be associated with more tightly spaced
category clusters.
In addition, we will record pupillometry as a secondary measure of NE activity
employed within each level of the psychopharmacological manipulation. Pupil
diameter has been shown to closely track NE levels in the brain (Aston-Jones &
Cohen, 2005), whereby smaller pupil diameter indicates lowers levels of tonic
NE. Within each drug condition, we will rank trials according to average
pre-stimulus pupil diameter, and divide them at the median. We expect that
trials with greater pre-stimulus pupil diameter (associated with higher tonic
NE) will have more tightly spaced category clusters than trials with small
pre-stimulus pupil diameter.

1.5 The Effect of NE on Exploratory vs. Exploitative Behavior
According to a recently proposed theory (Aston-Jones & Cohen, 2005;
Nieuwenhuis, Aston-Jones, & Cohen, 2005), the different modes of LC activity
serve to regulate a fundamental tradeoff between two behavioral strategies:
exploitation vs. exploration. Individuals must continually decide whether it
would be better to pursue known sources of reward (exploitation), or whether
there is more to be gained by searching for new strategies or opportunities
(exploration). This dilemma in how we invest our time and effort is a
well-known problem in the reinforcement literature.
High phasic/low tonic LC activity promotes exploitative behavior by
facilitating processing of task-relevant information (via the phasic response),
while filtering out irrelevant stimuli (through low tonic responsivity). By
increasing the phasic response of the LC, the cognitive system is better able
to engage in the task at hand, and maximize rewards harvested from this task.
In contrast, low phasic/high tonic LC activity promotes behavioral
disengagement by producing a more enduring and less discriminative increase in
responsivity. Although this degrades performance within the current task, it
facilitates the disengagement of attention from this task, thus allowing
potentially new and more rewarding behaviors to be emitted. Thus, the
transition between the two LC modes can serve to optimize the trade-off between
exploitation and exploration of opportunities for reward, and thereby maximizes
utility.
Our proposed psychopharmacological increase in NE should be
approximately equivalent to raising tonic levels of NE, thus inducing the low
phasic/high tonic LC mode and promoting exploratory behavior. Our collaborators
at Princeton University have designed two gambling tasks that elicit both
exploratory and exploitative behavior in participants, and that allow
quantitative assessment of instances of each type of behavior. We will have our
participants complete these tasks outside of the fMRI scanner, after the
primary (in-scanner) task has been completed. We will assess the degree to
which subjects engaged in exploratory behavior under the influence of
atomoxetine and compare that to the control (placebo) condition. We predict
that subjects will exhibit significantly more exploratory behavior when under
the influence of atomoxetine.

1.6 The Effect of NE on Biasing Neural Activity
The brain-wide increase in neural responsivity produced by increases in NE
levels should have the impact of enhancing differences in strength between
competing neural representations: Already strong representations should become
stronger, while weaker, competing representations should become even weaker.
It follows from this that biases in processing should be enhanced as NE levels
increase. For example, Eldar and Niv recently demonstrated that when measures
of pupil diameter suggest high levels of cortical NE, subjects demonstrate a
stronger bias toward predisposed learning styles (2012) and a stronger bias
toward attended versus unattended stimuli (manuscript in preparation); while
Proulx and colleagues (Proulx & Heine, 2008; Proulx, Heine, & Vohs, 2010) have
demonstrated that uncertainty, a factor thought to increase NE levels (e.g. Yu
& Dayan, 2005), causes a stronger affirmation of a subject*s central, moral
beliefs. We will investigate two of these ways in which NE effects should
manifest in behavior.
First, subjects will read several short descriptions of crimes and
heroic acts, and will be required to either set the monetary punishment for the
crime (ticket amount) or the monetary reward for the heroic act. We expect
these opportunities to affirm one*s moral beliefs will be influenced by NE
levels whereby high NE under the influence of atomoxetine will result in more
extreme values to be set as punishments and rewards compared to the placebo
condition.
Second, subjects will have to identify an ambiguous letter embedded
within a three-letter string that forms a word when the ambiguous letter is
interpreted as one of two possible letters, but does not form a word when the
ambiguous letter is interpreted as the other possible letter. Eldar and Niv
found subjects are less likely to be influenced by the irrelevant outer letters
when pupil diameter suggests NE levels are high. We expect to observe this same
effect when comparing subjects under the influence of atomoxetine versus
placebo.

The primary objective of this study is to measure the effect of a
psychopharmacological manipulation of NE on fMRI measures of the fidelity of
cortical representation.

The secondary objective is to determine if increased levels of cortical
norepinephrine will be associated with more exploratory behaviour in a gambling
task, as predicted by the adaptive gain theory of the LC-NE system.

The third objective is to determine if atomoxetine-enhanced noradrenergic
activity will strengthen predispositions that may be associated with habitual
and/or entrenched patterns of neural activity, as manifested by the
interpretation of letters within three-letter strings and in answering
questions related to the participant*s personal beliefs.

Design

The proposed study will use a double-blind, placebo-controlled, cross-over
design.

General procedure
The proposed study will consist of two sessions of event-related fMRI data
collection. Each subject will perform the task in the MRI scanner under the
influence of atomoxetine in one session, and under the influence of a placebo
in the other. The study will start in the EEG lab on the ninth floor of the
LUMC (Psychiatry department), and move down to the fMRI room on the first floor
about 112 minutes after the subject first arrives. Total scanning time will
constitute approximately 41 minutes in each session, and the subject will
return to the EEG room after being scanned. There are an additional 65 minutes
of experiment tasks to perform outside of the scanner (in the EEG lab before
and after the fMRI scanning). With breaks, time for explanations, for the drug
to take effect and for moving between locations, each session will last
approximately four hours. Participants will be administered the drug 75-80
minutes before the first experiment task to ensure that tasks are performed
during peak blood levels (Chamberlain, Muller, Blackwell, Robbins, et al.,
2006; Graf, et al., 2011), and the final task will be completed 210 minutes (3
hours, 30 minutes) after taking the drug, within the window of time when the
drug should still be having an effect on cognition (Sauer, Ring, & Witcher,
2005).

Drug Intervention
Participants will receive on one occasion 40 mg of the selective norepinephrine
reuptake inhibitor atomoxetine (Navarra, et al., 2008), orally administered.
Although other recent studies have used dosages of 80 mg (Graf, et al., 2011),
here we opt for the typical starting dose used in clinical practice, 40 mg, to
avoid the reported side effects of increased heart rate at high atomoxetine
doses (Heil, et al., 2002).

In the other session, either one week earlier or one week later, participants
will receive a placebo pill.


Task 1
In the scanner, subjects view a series of isoluminant, monochromatic images of
faces, houses, and cats, and are required to categorize each picture
accordingly using one of three right-hand button-press responses. These
categories of stimuli have been used successfully before in MVPA studies of
visual categorization (e.g. Haxby et al., 2001; Schurger et al., 2010). Images
will be presented in random order, and will be shown on screen for 1500 ms
each, followed by a response screen displayed for two seconds asking the
subject to indicate the category to which the picture belonged. After the
two-second response screen, a fixation cross will be presented for an
inter-stimulus interval varied randomly between 7500 ms and 9500 ms, averaging
8500 ms. This slow design is conducive to both our fMRI methods, and our
pupillometry methods. Subjects will complete two blocks of 60 trials (120 in
total), with a one-minute break between blocks. Each block will be 12 minutes
in duration.

Task 2
Participants will play a series of gambling games where they select from one of
two slot machines, with the goal of maximizing winnings. Each game involves two
*new* slot machines with unknown pay-off probabilities, and starts with four
*forced* trials where the program selects which slot machine to pull, with the
subject accumulating winnings, and more importantly, accumulating information
about the pay-off probabilities of the two slot machines based on those first
four outcomes. After the forced pulls, the subject will get to make a variable
number of additional free-choice pulls: Either one free-choice pull, six, or
eleven, depending on the game. Subjects will complete 150 games, with each
taking an average of 12 seconds to complete, for a total duration of 30
minutes. Participants will then complete a second, continuous-game version of
this task that is not divided between different games but instead contains
random *change points* where the underlying probabilities of the slot machine
changes without any cue to the participant. This second version takes
approximately 15 minutes to complete. Subjects score points during these tasks
and will receive a monetary bonus at the end of the tasks that varies linearly
with the amount of points scored, ranging from 0 Euros to 6 Euros.

Task 3
Participants will be instructed to identify an ambiguous central letter
embedded within a three-letter string, while ignoring the outside letters. The
identity of the central letter will be ambiguous, created by morphing one
letter to look like another letter. For example, the letter *H* can be made to
look more like the letter *A* by angling the top vertical lines of the *H*
toward the center. Critically, on each trial, one of the two letters used to
create the morph will fit with the outer letters to make a common three-letter
word, whereas the other letter would not form a real word. For example, the A/H
morph could appear with outer letters to seemingly form the word *THE* or the
word *DAY*, whereas in contrast *TAE* and *DHY* are not real English words.
These letter strings are preceded by a subliminal prime (33 ms) which is either
semantically related or unrelated to the real-word interpretation of the letter
string. The letter string is then presented for 225 ms, followed by a response
screen presented for 5000 ms. Subjects are instructed to concentrate only on
the shape of the central letter and ignore the influence of the outer letters.
They indicate which letter the central letter looks most similar to by choosing
from a list of four letters, two of which are the two letters used to create
the morph. Subjects will perform 88 trials of this task, for a total duration
of approximately eight minutes.

Side effects of atomoxetine
A single dose of atomoxetine has not been reported to have long-lasting
effects, either adverse or beneficial. Short-term side effects of the drug can
include fatigue, increased heart rate, akathisia and dry mouth, which have been
shown to disappear around 2 hours after drug ingestion (Chamberlain, Muller,
Blackwell, Clark, et al., 2006; Chamberlain, Muller, Blackwell, Robbins, et
al., 2006). For some groups, atomoxetine does carry more serious effects:
individuals with glaucoma, with heart disease, or taking monoamine oxidase
inhibitors (MAO inhibitors). These groups will be excluded from participation.

fMRI
There are no known risks associated with participating in an fMRI study. This
is a noninvasive technique involving no catheterizations or introduction of
exogenous tracers. Numerous human subjects have undergone magnetic resonance
studies without apparent harmful consequences. Radiofrequency power levels and
gradient switching times used in these studies are within the FDA approved
ranges. Some people become claustrophobic while inside the magnet and in these
cases the study will be terminated immediately at the subject's request.

Pupillometry
The eye-tracker system uses detailed analysis of high-definition video to
record pupil diameter at any given time during the experiment. The subjects do
not have to wear any special apparatus for the eye-tracker to work, and are at
no significant risk of any type of injury or discomfort due to this aspect of
the experiment.