Facial EMG Reveals Top-Down and Bottom

Reward vs. Punishment: An fMRI Analysis Approach to Identifying
the Neural Substrates of Motivation and Cognitive Control
Ya’el C. Courtney, Debbie Yee, Marie Krug, Jo Etzel & Todd S. Braver
Methods: Liquid Incentives
Background
• Motivation increases cognitive control and task performance (Botvinick and
Braver 2015)
• Incentives used as rewards and punishments both improve performance
(Wachter et al 2009)
• Impairments in cognitive control (and particularly an abnormal response to
motivation) underlie disorders such as schizophrenia, anxiety, depression,
eating disorders, and addictions.
• Why liquid incentives?
– It has been suggested that primary incentives offer distinct conceptual and methodological advantages
over secondary incentives for several reasons. Primarily, a liquid can be immediately perceived as “good”
or “bad”, and is more effective in this way than gaining or “losing” money in an experimental setup.
– Liquids used for each participant were selected from among several options based upon Session 1 ratings
on Liking and Intensity (1-7 Likert scale)
• Reward: 4 different fruit juices
• Punishment: saltwater, quinine
• Neutral: water, various dilutions of “saliva” (KCl/NaCO3) solution
•
• Research has made great progress in discovering the behavioral and neural
mechanisms that underlie motivation and cognitive control. However, a significant
question that remains to be addressed is:
7
6
5
Liking
4
Intensity
3
HYPOTHESIS: Reward and punishment will result in comparable
2
task performance, but utilize distinct neural substrates.
1
Experiment Overview:
Session 1: Behavioral Testing
*Practice Face-Word Task Switching
*Taste and rate reward, punishment, and neutral liquids
*Individual differences questionnaires
Session 2: fMRI 1, Reward
Session 3: fMRI 2, Punishment
*For fMRI sessions, subjects practiced the Face-Word task outside the
scanner, then performed 3 baseline runs and 6 incentive runs of the FaceWord task inside the scanner.
*Reward/Punishment order of fMRI testing was counterbalanced between
subjects.
Trial Structure
+
300 ms
Attend Face
400 ms
CTI
+
4300 ms
Target
HEALT
H
Up to
2500 ms
Reward: 4 different fruit juices
Reward (reward obtained): M = 59.1%
Punishment (punishment avoided): M = 57.9%
ns (T = .446, p = .659)
Reward
Punishment
Liking:
*Reward Liquid liked more than Punishment Liquid (T = 36.045, p < .001)
and Neutral Liquid (T = 14.872, p < .001)
*Punishment Liquid liked less than Neutral Liquid (T = 14.684, p < .001)
Intensity:
*Reward Liquid (T = 20.119, p < .001) and Punishment Liquid (T =
19.304, p < .001) higher in intensity than Neutral Liquid
Neutral
Data pre-processed using Analysis of Functional Neuroimaging (AFNI)
software ver. AFNI_16.2.06
Participants: N=33, healthy adults from WashU and the Saint Louis
community (19-39 years; mean = 25.9; 16 male)
Cue
As hypothesized, reward and punishment
incentives resulted in comparable behavior task
performance.
Processing Pipeline
Methods: Participants and Design
Flicker
Task Performance - Incentive Success Rate:
Ratings of Selected Liquids
Rating
Do rewards and punishments utilize the same or
different neural substrates to yield motivational
effects?
Results: fMRI
TFI
+
RT - 2500 +
500/3000ms
Feedback
!
300 ms
Face-Word Task-Switching paradigm:
Cue: Indicates which task to perform and incentive value of trial
• Attend Face: respond whether face is male or female in gender
• Attend Word: respond whether word is 2 syllables or not (1 or 3 syllables)
• Baseline runs: cue color meaningless
• Incentive runs: cue color indicated non-incentive (NI) (no liquid rewards or
punishments possible) or incentive (I) trials
Feedback: Indicates trial outcome
• For baseline and NI trials- no liquids received
• For I trials:
• If fast (< than 40th percentile baseline RT) & accurate:
• Reward Condition: rewarding liquid (juice)
• Punishment Condition: neutral liquid
• If too slow or inaccurate:
• Reward Condition: neutral liquid
• Punishment Condition: punishment liquid (saltwater or quinine)
Results of significance testing on contrasts between Punish-Baseline and
Reward-Baseline conditions yield interesting regions of activation unique
to each condition.
Conclusions & Future Directions
Quality Control: Motion, Standard Deviations, Means
Data for this experiment were collected in
2011, so a major part of the analysis
incorporated
thorough
quality
control
measures to ensure accurate results of GLM
contrasts and later MVPA work.
Motion: At right: motion in six directions (x, y,
z, roll, pitch, and yaw) for a good (S033) and
a bad subject (S001). Each graph displays
two days of scans, with three baseline and six
incentive trials per day. A vertical black line
along the bottom of a graph represents a
frame that was discarded due to excess
motion. Motion censoring yielded only 12 high
quality subjects out of the original 33.
1. First study to isolate primary reward vs. punishment incentives
in a challenging cognitive environment
2. Reward and punishment incentives result in comparable
behavioral task performance.
3. Preliminary evidence that regions associated with reward
processing show activation where hypothesized (insula for
punishment condition, medial PFC).
4. Data are now pre-processed, filtered to the highest quality, and
these positive contrast results give sufficient cause to move
forward with Multi-voxel pattern analysis.
Multi-Voxel Pattern Analysis:
• GLM analysis focuses on the relationship between cognitive brain variable
and individual voxel activation
• MVPA is able to analyze multi-voxel patterns of activity, which may be
important if reward vs. punishment pathways utilize similar magnitudes of
activation, but different patterns.
NOTE: Excessive motion may be due
in part to the nature of liquid incentive
delivery in the scanner.
Voxel-wise mean and standard deviation:
For each run for each subject, voxel-wise
mean and standard deviation images were
generated. These standard deviation images
are an additional quality control: brighter
areas indicate either CSF and blood (present
in every subject), or excessive motion (clearly
brighter brains). Mean images allow visual
checks of orientation and displacement. At
left: images from S033 (good) and S001
(unusable) demonstrate the utility of this
quality control measure.
Acknowledgements
This work was funded by the BP-ENDURE Neuroscience Pipeline Program, supported by NINDS
grant R25NS090985 through Dr. Erik Herzog. Much gratitude to program co-director Dr. Diana JoseEdwards as well.