Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The brain and behavior of motor learning: the what, how and where
(USC Thesis Other)
The brain and behavior of motor learning: the what, how and where
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THE BRAIN AND BEHAVIOR OF MOTOR LEARNING:
THE WHAT, HOW AND WHERE
by
Andrew Hooyman
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOKINESIOLOGY)
Degree Conferral Date:
August 2019
ii
Dedication
To my wife, daughter and mother
iii
Acknowledgements
I feel immensely grateful to have reached such a point in my life, and it is all thanks to a
great number of individuals who have inspired, mentored and assisted me throughout. Each has
shaped me in a very special way to be the man and scientist I am and still hope to be.
First, I would like to thank my advisor, Carolee. Without whom none of this would be
possible. I remember writing my first e-mail to her as my time at Teachers College came to an
abrupt close with the hope of transfer and she welcomed me with open arms. I will never forget
her most valuable lesson, Words have Meaning.
I will always consider Dr. Fisher as my second advisor who was always willing to spare
time and ENTHUSIASTIC EDITS for a young writer forever in training. Before I took a single
class at USC I participated in her TMS course. From that first impression I knew I would be
happy here.
I would like to thank Dr. Gordon for the several readings and discussions we had on
motor learning. He has made a huge impact on how I think about science. I have the utmost
respect for him and his leadership in this department.
Thank you Dr. Kutch for generously opening your lab to me. He not only shared his lab
equipment but also his time and insight which I have deeply appreciated. During some of our
interactions I know you even made time to meet when your second child was still very young.
Now that I have my daughter, I am deeply thankful for the Herculean effort it took to fit me in
your schedule.
To Dr. Schweighofer I would like to thank for giving me the gift of code and analysis.
From his modeling course to his informatics course to collaborating on the ICARE data together
it has been a pleasure to work with you.
iv
Alex Garbin has become my closest friend at USC. Our mutual love for basketball and
informatics is a weird one but has made all the difference during my time here. Much of this
dissertation is to his credit as he has been my best collaborator.
To my Brother, Chris. My younger brother but my best friend. Whenever we would talk
his positive nature and good humor would always break me out of any stress related to work. I
hope I was able to provide the same as he transitioned through the many phases of his life as
well. I couldn’t be happier that Frankie has you as an uncle or more proud of the man you have
become.
To my Mom, Dr. Nancy Hooyman, I owe everything. She is the strongest woman I know
and probably the kindest in the history of moms. I started this degree for my dad, but I finished it
for her.
To my wife I could not put into words what she means to me. Without her I don’t know
where I would be. She is my true north and she has guided me with her love and kindness and I
could never pay her back for all the support she has given me during my time here. I love you
Sabrina.
Last but never least, I want to thank Francesca Hooyman. You have only been on this
planet for a short time but already you have stolen my heart. I love you Frankie and I hope dad
has made you proud.
v
Table of Contents
Dedication ...................................................................................................................................... ii
Acknowledgements ....................................................................................................................... iii
List of Tables ................................................................................................................................. x
List of Figures ............................................................................................................................... xi
Chapter 1: Introduction to the Dissertation ............................................................................... 1
Chapter 2: .................................................................................................................................. 2
Chapter 3 .................................................................................................................................... 3
Chapter 4 .................................................................................................................................... 4
Chapter 5: .................................................................................................................................. 5
Chapter 6: .................................................................................................................................. 7
Chapter 2: Insight into individual motor learning strategies: Questioning the existence of
the Non-Learner phenomenon ..................................................................................................... 8
Abstract ...................................................................................................................................... 8
Introduction ............................................................................................................................... 9
Methods .................................................................................................................................... 12
Discovery Task Characteristics .......................................................................................... 12
Quantification Metrics......................................................................................................... 15
Statistical Analysis ................................................................................................................... 18
Confirmatory ........................................................................................................................ 18
Group Level .......................................................................................................................... 19
Sub-Group Level .................................................................................................................. 19
Individual Level ................................................................................................................... 19
Exploratory .......................................................................................................................... 20
Results....................................................................................................................................... 20
Relationship between Task Outcomes and Learning: Confirmatory ............................. 20
Sub-group Identification: Group Level ............................................................................. 22
Sub-Group Level: Performance Updating ........................................................................ 26
Discussion ................................................................................................................................. 31
The Process of Rule Discovery ............................................................................................ 33
The Low Performer Phenomenon ...................................................................................... 34
Conclusion ................................................................................................................................ 36
vi
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning ............ 37
Abstract .................................................................................................................................... 37
Introduction ............................................................................................................................. 38
Methods .................................................................................................................................... 42
EEG Cap Orientation and Resting-State Measurement .................................................. 43
Discovery Task Practice and Retention ............................................................................. 45
Performance and Phase Quantification ............................................................................. 49
Intracortical Connectivity from Resting-state Electroencephalography ........................ 50
Statistical Analysis ................................................................................................................... 51
Results....................................................................................................................................... 53
Exploration Phase--Quantification ..................................................................................... 53
Exploitation Phase-Quantification ..................................................................................... 55
Retention Phase-Quantification .......................................................................................... 56
Resting-State EEG Prediction of Individual Levels of Exploration ................................ 56
Resting-State EEG Prediction of Individual Levels of Exploitation ............................... 57
Resting-State EEG Prediction of Individual Levels of Retention .................................... 58
Improved Fit from Previous History and Intracortical Connectivity ............................. 60
Discussion ................................................................................................................................. 61
Cortical Connectivity and the Exploration Phase ................................................................ 62
Cortical Connectivity and the Exploitation Phase ............................................................ 63
Cortical Connectivity and the Retention Phase ................................................................. 64
Inclusion of Previous Experience in the Model ................................................................. 64
Limitations ............................................................................................................................ 65
Behavioral Sub-groups ........................................................................................................ 65
Future Work......................................................................................................................... 66
Conclusion ................................................................................................................................ 66
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity ................................................................................... 68
Abstract .................................................................................................................................... 68
Introduction ............................................................................................................................. 69
Modulation of Intracortical Connectivity .......................................................................... 69
Resting-state Intracortical Connectivity and its Translational Role in the Field of
Motor Behavior .................................................................................................................... 70
vii
Strong Scientific Premise for Paired Associative Stimulation ......................................... 71
How the PAS/EEG Paradigm can Benefit Motor Learning Research ............................ 73
Method ...................................................................................................................................... 74
Control for Variations in Brain State between Repeated Measures ............................... 74
Signal to Noise Problems Inherent to the Technology ...................................................... 74
Informed Consent and Safety Screening ........................................................................... 75
Inclusion/Exclusion Criteria – Unique Considerations for NIBS Research in Motor
Behavior ................................................................................................................................ 75
Safety Precautions ............................................................................................................... 76
Steps to Maintaining Scientific Rigor for Implementation of the PAS/EEG Study
Design Illustrated in Figure 4.2 .......................................................................................... 77
Pre-measurement Procedures ............................................................................................. 79
Baseline Data Acquisition .................................................................................................... 81
PAS ROI Selection: Functional vs. Anatomical ................................................................ 82
PAS Procedure: Conditions and Parameters .................................................................... 83
Paired Associative Stimulation (PAS) Parameters ........................................................... 84
PAS Condition Randomization ........................................................................................... 86
Controlling for Brain State ................................................................................................. 86
Post PAS Resting-state EEG recording ............................................................................. 87
Day 2: Pre-measurement Procedures for PAS Condition 2 ............................................. 88
Second PAS Condition ........................................................................................................ 88
Post resting-state EEG recording for Condition 2 ............................................................ 89
Data Processing and Analysis ............................................................................................. 89
Example Outcome ................................................................................................................... 92
Discussion ................................................................................................................................. 94
Future Work......................................................................................................................... 95
Conflict of Interest ............................................................................................................... 96
Ethical Considerations ........................................................................................................ 96
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study ................................................................. 97
Abstract .................................................................................................................................... 97
Introduction ............................................................................................................................. 98
Methods .................................................................................................................................. 100
viii
Determination of iPAS Intensity ....................................................................................... 101
Baseline Procedure for rs-EEG ........................................................................................ 102
Wash-out Period ................................................................................................................ 106
Second Intracortical PAS Condition ................................................................................ 106
Data Processing ...................................................................................................................... 107
Coherence Calculation....................................................................................................... 107
Statistical Analysis ................................................................................................................. 107
Baseline Test-retest Reliability of rs-EEG ....................................................................... 107
Changes in Connectivity as a Result of iPAS .................................................................. 108
Visualization of iPAS Effects Across the Cortex ............................................................. 108
Order Effect Between iPAS Conditions ........................................................................... 108
Results..................................................................................................................................... 108
Controlling Brain State ..................................................................................................... 108
Test-retest Reliability ......................................................................................................... 109
Changes in Connectivity Due to iPAS .............................................................................. 110
Visual Localization of iPAS Effects across the Cortex ................................................... 111
Order Effects of iPAS ........................................................................................................ 112
Discussion ............................................................................................................................... 113
Limitations .......................................................................................................................... 116
Conclusion .............................................................................................................................. 117
Chapter 6: Summary and Conclusion ..................................................................................... 119
Chapter 2: Insight into individual motor learning strategies: Questioning the existence of
the Non-Learner phenomenon ............................................................................................. 119
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning ...... 120
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity ............................................................................. 122
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study ........................................................... 122
Limitations ............................................................................................................................. 123
Chapter 2 – Discovery Learning ....................................................................................... 123
Chapter 3 – Prediction of Discovery Learning ................................................................ 123
Chapter 4 & 5 – iPAS Methodology and iPAS Feasibility ............................................. 123
Future Direction .................................................................................................................... 124
ix
APPENDIX ................................................................................................................................ 126
Previous Experience Questionnaire ..................................................................................... 126
Stimulation Comfort Survey ................................................................................................. 127
References .................................................................................................................................. 128
x
List of Tables
Table 3.1: R
2
values for LOOCV, VIF Correction without Experience and VIF Correction
with Experience for each Behavioral P h a se … … ……… ……… …… …… ……… ……… ….61
xi
List of Figures
Chapter 2: Insight into individual motor learning strategies: Questioning the existence of
the Non-Learner phenomenon
Figure 2.1: A. Visual display of the discovery task. The task goal is to land the cursor,
depicted as an astronaut, into the target box within 3 seconds and maintain its position
within the target box for 1 continuous second. If the task goal is achieved, then a reward
tone is played. B. The joystick to cursor movement step function representing the rate
control dynamics of the task. The function applies a continuous pulse of virtual force to the
cursor, astronaut, dependent upon the joystick position. Joystick position (x-axis) is
measured on a sensitivity threshold between -1, leftmost, and 1, right most. If the joystick
passes a sensitivity of +0.5 then a constant positive virtual force of 7.5 is applied to the
cursor. If the joystick passes a sensitivity in the opposite x direction of -0.5 then a negative
virtual force of -7.5 is applied to the cursor .............................................................................. 14
Figure 2.2: A. Example data of average acceleration profiles collected from a single
participant. The 49 successful acceleration profiles in blue with the blue shaded area
representing one standard deviation and 151 failed trials with the red shaded area
representing one standard deviation. B. Simulations of successful acceleration profiles.
Over 700,000 possible acceleration profiles were simulated to determine an average
successful acceleration profile. Approximately 9% of all possible acceleration profiles
would be successful. Notably each of these successful profiles resemble each other with only
a slight variance in execution. Shaded blue area is standard deviation of all successful
simulated trials. C. Example of a successful trial visualized through the collected position
and acceleration data. Total trial length is 450 samples which is 4.5 seconds in length. Grey
shaded box with solid black line represents the black box in space and the grey shaded box
with the dashed shaded line represents the target box in space (right x-axis). The blue line
represents the cursor motion through the virtual space where its final position is
maintained in the target box for 1 second resulting in a reward tone. The vertical dashed
line represents the end of the start chime indicating when the participant should not exit
the start box. Although the cursor may begin moving within the start box before the chime
becomes quiet. The red line represents the virtual acceleration of the cursor which is
generated based on the position of the thumb joystick (Figure 2.2C). We calculate scaling
ratio by dividing the negative virtual acceleration over the positive virtual acceleration.
Time outside the start box is determined by summing the total number of samples the
cursor spent outside the start box throughout the trial. Maximum time outside the start
box is 300. We can determine reversal of the joystick based on when the cursor experiences
a negative virtual acceleration .................................................................................................... 17
Figure 2.3: A. Individual participant data demonstrating relationship between mean
scaling ratio and mean time in target on Day 1. B. Individual participant data
demonstrating relationship between time outside start box and average time in target on
Day 1. C. Individual subject data demonstrating relationship between mean modified
scaling ratio and mean time in target on Day 1. The modified scaling ratio represents a
better fit of average time in target for each individual participant. Color coding and
centroids (larger points) in C represents results of k-means clustering with color coding
representing each of three different sub-groups. Shaded area represents standard error of
prediction ..................................................................................................................................... 21
Figure 2.4: Individual subject data for each of the identified sub-groups. Modified scaling
ratio is depicted on the y axis and trial number is on the x axis for Day 1. High performers
in red, moderate performers in blue and low performers in green. The black line within
each plot is a loess fit. .................................................................................................................. 23
Figure 2.5: A. Time course over practice of exploration for each sub-group. B. Average
Time in Target (Day 1) versus Average Day 1 Exploration for each individual. Participants
are color coded to their cluster analysis group. Black line in B is a fitted quadratic based on
the relationship between average exploration and average time in target on Day 1, shaded
area is standard error of relationship ........................................................................................ 24
Figure 2.6: A. Systematic examination of exploration across 10 trial blocks from 1:10 to
1:190 that most predicts future performance (trials 190-200) and retention/recall (trials 1-
50). B. Average exploration by sub-group during the first 80 trials (peak in A) predicts
average time in target during the last 10 trials on Day 1. C. Average exploration by sub-
group during the first 80 trials (peak in A) predicts time in target during retention (Day 2).
Black line is linear regression result and shaded area is standard error of prediction in B
and C. ........................................................................................................................................... 26
Figure 2.7: Relationship between previous task feedback (i.e. time in target), to next trial
exploration on Day 1. A) The Low Performer sub-group demonstrated a positive
exploration/performance relationship, Moderate Performers exhibited no relationship and
High Performers maintained a negative relationship. Black line represents results of
regression and shaded area is standard error of prediction. B) When each sub-group is
plotted on a continuum it creates an inverted U relationship that spans across the sub-
groups. Black line is result of regression and shaded area is standard error of prediction.
C) The relationship between next trial modified scaling ratio (Y-axis) and current Time in
Target (X-axis) plotted across sub-groups which are differentiated by color. Maximum
Time in Target is 100 centiseconds. D) Progression of Exploration (Y-axis) by Average
Time in Target (X-axis) for each sub-group. Point size represents progression of the
exploration/performance relationship across bins of trials (see legend) with smaller points
representing earlier practice trials and larger points reflecting later practice trials. ........... 29
xii
xiii
Figure 2.8: Relationship between Previous experience and Average Time in Target on Day
1. Participants who reported a low number of hours playing video games and physical
activity had the lowest average time in target compared to participants who reported
higher hours of video game play and physical activity throughout their life. Black line
represents regression results shaded area is standard error of prediction ............................ 31
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning
Figure 3.1: Visual depiction of study timeline. On Day 1 Participants are first consented
into the study and then are oriented with an EEG cap specific to individual head
circumference which is then calibrated. Prior to task practice they undergo five minutes of
eyes open resting-state EEG with eyes focused on a fixation cross. Then Participants
practice 200 trials of the discovery task with start chime and reward tone present
throughout. Day 2 begins the day after Day 1 beginning with 50 trials of the discovery task
where no start chime or reward tone is given (i.e., retention/recall phase). After the 50
trials are completed, participants are given a general survey questionnaire about their
previous lifetime experience with video games and/or physical activity reported in total
hours spent. The study concludes with participants being debriefed on the study purpose.
....................................................................................................................................................... 43
Figure 3.2: Eyes-open Resting-State EEG set-up: A) Participants stared at a fixation cross
for 5 minutes prior to task practice while EEG activity was recorded. B) Electrode layout
corresponds to the 10-20 system. Blue electrode (GND) served as the ground and the green
electrode (CPZ) functioned as the reference electrode for all participants. .......................... 44
Figure 3.3: A) Display of discovery learning task. Astronaut represents the cursor which
the participant controls with a joystick designed for the right thumb. Left box is the
START box and right box is the TARGET box. At the start of each trial a 1.5 s start tone
warns the participant to leave the START box, at the off set of the start tone. Beginning
with the off set of the start tone, the participant has 3 seconds to land the cursor in the
TARGET box. If the cursor stays in the TARGET box for one continuous second then a
reward tone sounds and the trial is considered a success. Scripts to recreate this task can be
found at the investigators Github: https://github.com/hooymana/Brooks_Task.git. B)
Example of a single successful trial and all corresponding outcome measures. The grey box
with a solid line represents the start box and the grey box with the dashed line represents
the target box within the virtual space of the task. The blue line represents the cursor
position throughout the trial traveling from the start box to target, measured on the right y
axis, with the dashed vertical line representing the end of the start chime (i.e. the point that
cursor can exit the start box). The red line is the acceleration of the cursor throughout the
trial measured on the left y axis. Here the cursor first experienced a positive acceleration to
begin propulsion through space and then a negative acceleration to bring it to a relative
stop within the target box. C) Example of experiment layout with a participant holding the
xiv
controller performing the discovery task. The cursor is only controlled by the right thumb
stick of the controller which is depicted in the inset image ..................................................... 47
Figure 3.4: A) Binned trials of exploration in progressive blocks of 10 predicting
subsequent blocks of Performance (Time in Target – hundredths of a second). The redder
the color the higher the accuracy of the exploration bin to predict Time in Target during a
specific block of practice. Bin of exploration between trials 1 and 40 (Peak Exploration) has
the highest average predictive accuracy across all blocks of performance. Bin of
exploration between trials 1 and 110 (End of Exploration) shows signs of reduced accuracy
to predict future bins of performance. Reduced prediction accuracy continues after trial
110 indicating that this is the point that exploration ceases on average for the study sample.
B) Average prediction accuracy across all bins of exploration trials, which is determined in
Figure 4A, was used to determine phases of Exploration and Exploitation. The Exploration
phase is blocked in blue. Exploitation was determined as the point during practice when
exploration accuracy fell below all previous levels, yet Time in Target continued to
improve. We indicated the exploitation phase as the trials from 110 to 200 shaded in red.
The Retention phase was the average time in Target during all trials practiced on day 2,
shaded in green ............................................................................................................................ 55
Figure 3.5: A) Comparison of observed exploration values on Y axis versus predicted
exploration values on X axis. If prediction was perfect all points would fall long the blue 1:1
line. Black line represents linear regression line between predicted and observed values. B.
Intracortical connectivity measures that are responsible for predicted exploration value
after multiple comparisons. Predictive Intracortical Connections: Fpz-FC5, Fpz-AF7, Fpz-
F7, Fpz-F5, Fpz-F3, Fpz-FC3, Fpz-C3, F7-C3, F7-C1, F7-F3, AF7-C3, Fpz-FC6, Fpz-FC4,
Fpz-C6, Fpz-C4, Fp2-C6, AF8-C6, AF8-C4, AF8-FC4, CP4-PO4, P4-PO4. Color of line
between electrode pairs signifies the relationship of the connection to predicted behavior in
the SVM ....................................................................................................................................... 57
Figure 3.6: A. Comparison of predicted exploitation against observed exploitation. Black
line represents linear regression line between predicted and observed values. B. Three
measures of intracortical connectivity after multiple comparisons that significantly predict
individual exploitation. The three measures are ipsilateral to the performing hand and
extend from frontal to parietal regions of the cortex. Predictive Intracortical Connections:
AF8-C6, CP4-PO4, P6-PO6. Color of line between electrode pairs signifies the relationship
of connection to the predicted behavior in the SVM ................................................................ 58
Figure 3.7: A. Comparison of predicted mean time in TARGET box during the retention
day versus actual time in TARGET box. Black line represents linear regression line
between predicted and observed values. B. Intracortical connectivity network responsible
for significant prediction of retention phase extends throughout the cortex with prevalent
xv
connections arising from left prefrontal cortex and extending to left motor and right
occipital/parietal. Predictive Intracortical Connections: Fpz-FT8, F7-CP5, AF7-CP5, AF7-
C5, C5-F1, AF7-P2, AF7-PO6, AF7-O2, P6-PO6, P6-PO4, O2-P8, AF8-C6, T7-TP7. Color
of line between electrode pairs signifies the relationship of connection to the predicted
behavior in the SVM ................................................................................................................... 60
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity
Figure 4.1: The change in magnitude of Long Term Potentiation and Long Term
Depression between hypothetical brain areas A and B as a function of interstimulus
interval. If the post-synaptic pulse, pulse delivered to brain area B, occurs before the pre-
synaptic, pulse delivered to brain area A, then long term depression, decreased
connectivity, red curve, will result. Conversely, if the pre-synaptic pulse occurs prior to the
post-synaptic, then long term potentiation, increased connectivity, blue curve, will result.
The closer the coupling of the two pulses, the closer they fire around an interstimulus of 0
ms, then the greater the overall Long Term Potentiation or Long Term Depression.
However, a change in connectivity will still persist, although at a lower magnitude, with a
longer interstimulus interval. ..................................................................................................... 73
Figure 4.2: A complete diagram of all calibration procedures (Pre-measurement
Procedures) and measurement/PAS procedures (Condition 1 and Condition 2). ................. 77
Figure 4.3: A) Depiction of the 10-20 electrode EEG cap layout. Red circles indicate the
electrodes over which a TMS coil is placed while the red arrow indicates the direction of
information transmission between brain areas when PAS is applied. B) The placement of
each TMS coil while a participant wears a 64 electrode EEG cap. The TMS coils are held in
place by moveable arms and the head of the participant is rested on a chin rest. ................ 81
Figure 4.4: Examples of brain (A) and eye (B) components. To improve signal to noise ratio
of the rs-EEG signal, the eye component (B) should be removed from the data set and the
brain component (A) remains. A brain component can be identified from the following
information provided graphically by resulting ICA of a 5 minute resting-state recording.
Brain components can be identified by 1) a peak at 7 – 15 Hz within the Activity power
spectrum 2) The ICA activity matrix, representing the change in frequency of the
component throughout the recording (graph in the top right of A), demonstrates random
spiking of both high and low frequency. An eye component, seen in B, has the following
characteristics 1) eye components lack any peak in the 7 – 15 Hz range and instead has a
peak near 1 Hz 2) IC2 activity matrix, has large and sustained changes in frequency
throughout the recording. These red changes within the matrix are when the EEG system
recorded muscle activity from the eyes. 3) This can be further confirmed from the electrode
xvi
map next to the IC2 activity matrix where the frequency is dominated near the eye region.
These images are generated through the opensource eeglab software and further detail on
how to implement and interpret ICA through EEG lab can be found here (Delorme &
Makeig, 2004). ............................................................................................................................. 91
Figure 4.5: Changes in Area of Significant Coherence across each condition for a single
participant: A) Baseline, B) PAS5 and C) PAS500. A. Baseline changes in intracortical
connectivity decrease within the target circle and fluctuate randomly throughout the
cortex. B. Changes in intracortical connectivity as a result of the PAS5 condition are local
to the target circuit AF8-C6 and do not appear to impact other areas of target circle and
fluctuate randomly throughout the cortex. B. Changes in intracortical connectivity as a
Baseline, B) PAS5 and C) PAS500. A. Baseline changes in intracortical connectivity
decrease within the Figure 4.5: Changes in Area of Significant Coherence across each
condition for a single participant: A) For Peer Review the cortex. C. PAS500 does not
modulate intracortical connectivity within the target circuit. Overall connectivity is
decreased in the left prefrontal and motor cortical areas. ...................................................... 93
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study
Figure 5.1: Study paradigm. Experiment is a within subjects design that commences with
resting motor threshold of the APB determined with the EEG cap already oriented.
Participants undergo a double baseline procedure before either iPAS conditions for
reasons of rs-EEG measurement reliability. Experimental conditions are counterbalanced
across the study sample to control for order effects of iPAS application. EEG =
Electroencephalograpphy, RMT = resting motor threshold, PAS = paired associative
stimulation. (Figure adapted from Figure 4.2 in Hooyman et al., In Review, JMLD) ........ 103
Figure 5.2: iPAS Set up/EEG electrode layout. A. Placement of two TMS coils is identical
between iPAS conditions. The TMS coil for the first pulse over electrode AF8 is oriented at
a 45 degree angle with respect to the midline in a posterior-anterior position. The TMS coil
of the second pulse over electrode C6 is oriented at a 45 degree angle with respect to the
midline in the anterior-posterior position. B. The EEG electrode layout is based on the 10-
20 system. Dark circles represent the electrodes of interest and the white arrow represents
the theoretical flow of information induced from iPAS between the underlying cortical
regions of interest. Ground electrode (GND) is in blue and reference electrode is in green
(CPZ). ......................................................................................................................................... 105
Figure 5.3: Scatterplot of Individual Basdeline Measures. X-axis is the coherence between
AF8 and C6 obtained at the first baseline; Y-axis is the same measure obtained at the
second baseline. Changes in baseline coherence plotted as a scatter plot to demonstrate
xvi
i
reliability of coherence measures for the two resting-state EEG baseline sessions separated
by a 10 minute rest period ........................................................................................................ 110
Figure 5.4: Changes in AF8-C6 Coherence for the three conditions. Comparison of
coherence change across baseline, iPAS5 and iPAS500 conditions. ANOVA revealed that
iPAS5 had a significantly greater increase in high beta coherence between the AF8-C6
electrode pair compared to baseline and iPAS500 conditions. Box and whisker plots with
individual subject data overlying the Box and whisker distribution (N = 10). Dashed line
represents SEM threshold in positive and negative direction ............................................... 111
Figure 5.5: Individual changes (Pre/Post) in circuit coherence between conditions. A.
Changes in coherence between individuals demonstrated that a majority (8/10) of
individuals experienced an increase of high Beta coherence above SEM in the iPAS5
condition compared to B. Only 1/10 participants experienced an increase above SEM in the
iPAS500 condition ..................................................................................................................... 111
Figure 5.6: Change in connectivity for baseline and iPAS conditions within the high Beta
band. Visual results of topographical maps demonstrate an increase of high Beta coherence
above SEM within the AF8-C6 circuit occurred only after receiving iPAS5, but not
Baseline or iPAS500. ................................................................................................................. 112
Figure 5.7: Changes in iPAS across the 5 rs-EEG sessions for both iPAS orders. Visual
inspection indicates that iPAS5 increased high Beta coherence compared to iPAS500
regardless of condition order. Additionally, iPAS500 did not effect a noticeable change in
high Beta coherence. Paired t-tests by each condition by day were not significantly different
(p >.05). N = 5 for each order condition. Two-way perforated arrow pointed at each iPAS5
slope denotes change in coherence as a result of iPAS5 regardless of condition order. Error
bars represent standard error of each resting-state measurement ....................................... 113
1
Chapter 1: Introduction to the Dissertation
The human capacity to interact with the environment stems from the capability of
learning. Whether a question of normal development or neurorehabilitation the brain and body
must overcome redundancy and complexity of actions to achieve an intended goal. Research
focused on motor learning often overlooks how the process of learning occurs and to what
degree each individual is capable of carrying out said process.
Too often scientists make assumptions that every participant learns in an equivalent
manner. Much of this is rooted in the consistent study of motor tasks that may represent a
fraction of the overall motor learning spectrum. Studies in neuroimaging that then investigate the
neural correlates of learning through the lens of these tasks subsequently offers a narrow
viewpoint of how the brain in fact learns.
Although much has been gained from earlier work, it is important to broaden the horizons
of the field by examining motor tasks that may offer new insights on how we learn. This
dissertation has the following three aims: Aim 1) Examine how individual participants learn a
rule-based motor task where success is contingent upon rule discovery. Aim 2) Identify the
cortical substrates from resting-state EEG that predict the processes of discovery learning. Aim
3) Test the feasibility of an intervention that uses non-invasive brain stimulation to potentiate a
putative neural circuit known to be engaged in the discovery learning process. These aims were
operationalized into two different experimental studies and four papers (Chapter 2-5). Together,
this dissertation is divided into 6 Chapters. A brief overview of each Chapter is described next.
The first (Chapter 1—this one) is a brief introduction that is meant to provide the overview and
background for the experimental work that makes up the bulk of the dissertation.
2
Chapter 2: Insight into Individual Motor Learning Strategies: Questioning the Existence of
the Non-Learner Phenomenon
This chapter describes the behavioral aspects of learning a novel complex skill. Nearly 25
years ago a research study performed by Vernon Brooks and colleagues (Brooks, Hipperath,
Brooks, Ross, & Freund, 1995) used a rule-based task to investigate the role of verbalization on
insight into the process of motor learning (i.e. did participants who talked through their learning
process identify the rule faster/slower than those that did not.) Interestingly, and unanticipated by
Brooks, a significant sub-group (~25%) of Non-learners (7/25) emerged, those who were unable
to discover the rule even after 200 practice trials. They were included in the paper but their data
were not analyzed for the results section. We decided to adapt the Brooks’ task because it
seemed to provide an important opportunity to identify the large variability in performance
among a seemingly homogeneous group of non-disabled individuals. This allowed us to probe
the commonly held assumption that an otherwise non-disabled young adult population were
equally capable of learning such a laboratory-based task. This learning equivalency assumption
is an important aspect of research in the field of motor learning as a relatively large body of work
in the field focuses on how the conditions of practice (e.g., practice structure, provision of
augmented feedback) affect groups of individuals, and ignores or considers the existence of non-
learners as outliers in their sample (Stéphanie Lefebvre et al., 2015; Tzvi et al., 2017; Wulf,
Shea, & Lewthwaite, 2010).
Importantly, the process of learning this discovery task seemed to enlist at least two
different kinds of processes: the first necessitating exploration of movement execution to identify
the task rule and the second to then exploit the rule to continue to refine performance. The
concept of a two-stage learning process is well discussed in theoretical papers about the motor
3
learning process that date back to the 1960s (Gentile, 1972; Miller, Galanter, & Pribram, 1960).
However, the majority of motor learning research has used tasks that engage
refinement/exploitation processes but not exploration (Wulf et al., 2010).
We adapted the Brooks’ task and recruited 32 non-disabled adults to: 1) determine
whether we could reproduce Brooks’ earlier findings and 2) provide a systematic quantitative
analysis of the performance characteristics we observed over a 200-trial practice day and no
augmented feedback, retention/recall session the next day. The quantitative analysis was
motivated, in part by the two-stage learning process model described above.
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning
Prior to discovery task practice described in Chapter 2, each participant undergoes 5
minutes of resting-state electroencephalography (rs-EEG). This Chapter describes the brain
correlates—specifically, the strength of connectivity between cortical areas that predict the
phases of motor learning characterized in Chapter 2. The communication shared within the
resting brain, measured as intracortical connectivity, has been shown to be a robust predictor of
individual motor performance and learning (Wu et al., 2015; Wu, Knapp, Cramer, & Srinivasan,
2018; Wu, Srinivasan, Kaur, & Cramer, 2014). This may be because the activity of the resting
brain represents a close proxy to the actual strength and layout of specific neural networks that
are responsible for execution of various processes such as learning (Lewis, Baldassarre,
Committeri, Romani, & Corbetta, 2009a; Sami, Robertson, & Miall, 2014). Resting-state
Intracortical connectivity strength may provide a link between learning capability and brain
function.
Intracortical connectivity of the rs-EEG data were calculated between every possible
electrode pair to predict the exploration, exploitation and retention phase of discovery task
4
learning. We chose to examine every feasible electrode pair as a means of conducting a whole
brain analysis of which intracortical connectivity can best predict discovery learning. This is
somewhat novel as many previous experiments utilized a seed-based analysis where the
investigator examines how brain connectivity to a specific brain area, e.g. primary motor cortex,
predicts future performance (K. R. Lohse, Wadden, Boyd, & Hodges, 2014). We believe this is
somewhat limiting in that it neglects brain regions not considered part of the well-known motor
learning network, and that may serve as strong predictors of essential processes engaged for
motor learning.
We determined the exploration and exploitation phase of discovery task performance by
systematically mapping which bins of exploration performance provide the highest level of
prediction accuracy for subsequent bins of performance. The exploration bin with the highest
prediction accuracy is the most meaningful among the study sample and is then predicted by rs-
EEG intracortical connectivity. We then determine the exploitation phase at the point where
exploration accuracy drops below all previous levels, yet performance continues to improve.
Finally, we predict the retention/recall phase using the average performance from the second day
of task performance that was without feedback (i.e. retention).
By utilizing a whole brain analysis to predict these distinct learning phases, we aim to
identify distinct intracortical circuits associated with each phase of learning.
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity
This Chapter describes the iPAS paradigm, a novel intracortical paired associative
stimulation paradigm (iPAS) used in the second experiment for this dissertation. With our recent
understanding that behavior is more mediated through networks of brain activity rather than
5
single brain regions, there is an urgent need to develop new methods for neuromodulation
(Hordacre, Rogasch, & Goldsworthy, 2016). Based on a fundamental mechanism of
neuroplasticity, Spike Timing Dependent Plasticity (STDP), we developed the intracortical
Paired Associative Stimulation (iPAS) paradigm (Bi & Poo, 1998). Chapter 4 describes the iPAS
paradigm in detail and Chapter 5 demonstrates that iPAS is a feasible method that can be used to
strengthen resting-state intracortical connectivity (rs-IC).
Paired Associative Stimulation has previously been used to facilitate connectivity
between the central and peripheral nervous system and is based on the STDP mechanism (K.
Stefan, Kunesch, Cohen, Benecke, & Classen, 2000). However, with the advancement in EEG
technology an intracortical PAS paradigm can now safely be applied to humans. In this Chapter,
we not only provide a clear road map of how to appropriately apply iPAS to a non-disabled
human cohort, but we include other measures that address methodological weaknesses identified
in previously used neuromodulatory paradigms.
We propose a within subject counter-balanced study design where participants experience
both a sham and true iPAS condition on the same day. The only difference between the true and
shame condition is the timing between paired pulses (i.e., the inter-stimulus interval, ISI). Proper
execution of this protocol should support a proof of principle that iPAS can be used to modulate
rs-IC.
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study.
To determine feasibility of our proposed iPAS paradigm described in Chapter 4, we
recruited ten non-disabled adults. The purpose of this experiment is not only to demonstrate
feasibility of iPAS to modulate rs-IC, but also to demonstrate how the addition of a double
6
baseline is important for improving the overall quality of the research. No previous cortico-
cortical PAS paradigm has demonstrated the capability to modify rs-IC. Previous work using
iPAS with a simple motor behavior demonstrated mixed results in efficacy to modify a targeted
behavior (Palmer, Wolf, & Borich, 2018).
To prevent future iPAS research from falling down the same rabbit hole as previous
research in the general area of non-invasive brain stimulation (NIBS), (i.e. attempting
immediately to modify motor behavior), we designed an experiment to modulate a specific rs-IC.
The targeted rs-IC was chosen because it is one with high enough connectivity characteristics in
the beta band, to emerge as predictive of performance associated with the exploitation phase of
learning and identified through the experiment described in Chapter 3. We wanted to first,
confirm the mechanism of iPAS, spike-timing dependent plasticity (STDP), and then its
capability to modulate our targeted rs-IC. We believe that the success of future research must
rely more on developing an understanding of the putative brain-behavior relationship being
tested through NIBS rather than the efficacy of the NIBS paradigm. As previous research has
indicated, there is considerable variability in responsiveness using some of the most popular
neuromodulatory paradigms, TDCS and rTMS, to reliably induce a change in corticospinal
excitability. This is clearly a problem and one of the most limiting features of these NIBS
paradigms.
Therefore, we felt it was an important first step to establish feasibility of our iPAS
paradigm on rs-IC before attempting to modify the learning process at the outset, for example
targeting the non-learner/poor learner sub-group described in Chapter 2. With proof of principle
supported (Chapter 5) we can move forward with future work to investigate the precise brain-
7
behavior relationship without threat of serious methodological confounds, that have plagued the
NIBS research community.
Chapter 6: Summary Conclusion.
Overall, this dissertation aims to first quantify distinct behavioral strategies of a rule-
based discovery task, then identify rs-IC networks predictive of specific phases of discovery
learning, and finally establish a proof of principle that predictive rs-IC can be potentiated with
iPAS. Together, results from this dissertation provide new evidence of an understudied area, the
process of motor learning (what is learned?) and not just the outcome; which key cortical
networks are engaged in the learning process (where does the learning occur?), and if a putative
neural correlate (i.e., intracortical circuit) can be effectively modified using a high quality iPAS
paradigm (how can we effect the learning process?). We close by describing some of the
limitations of this work, and then lay out a few ideas for future directions.
8
Chapter 2: Insight into individual motor learning strategies: Questioning the existence of
the Non-Learner phenomenon
Abstract
A seminal paper about learning, the what and how in a human motor task, reportedly found a
small number of non-disabled adults who failed to learn. This failure to learn was attributed to an
inefficient motor system. To reproduce and analyze this non-learner phenomenon, we designed
an experiment that employed a similar complex motor task which required specific force and
timing constraints allowing for only one correct movement pattern, known as the task rule, to
achieve the goal. Like the previous research, participants were naïve to the task rule and had to
discover, through practice, the appropriate movement pattern to be successful. We hypothesized
that specific performance sub-groups would demonstrate a unique relationship between visual
feedback (i.e., time in target) and exploration (i.e., variance in task rule execution) that would be
associated with different levels of success. Out of the 32 non-disabled adults, we identified three
distinct sub-groups: (Low Performer/Non-Learner (LP, N = 9), Moderate Performer (MP, N =
12) and High Performer (HP, N = 11)). An analysis of sub-group behavioral patterns revealed
three important findings. First, levels of exploration in early practice predicted late performance
and next day retention (late performance, R
2
= 0.48; retention, R
2
= 0.45). Second, the
exploration (Y-axis) by visual feedback (X-axis) relationship revealed a performance continuum
whereby the three sub-groups were systematically and serially aligned (i.e. low, moderate, to
high) on an inverted U. Finally, the level of self-reported video game and physical activity
experience was linked to individual performance (R
2
= 0.50). In summary, exploration appears to
be a distinct contributor to varying levels of human learning of a complex motor task for which
9
the solution is not obvious. Evidence of a performance continuum suggests that Non-Learners
are simply inexperienced; they do not necessarily possess an inefficient motor system. This
insight opens up new avenues of research directed towards transforming so-called non-learners
into learners.
Introduction
The predominant theories within the field of behavioral motor learning adhere to a
general framework in which the motor system learns novel skills through the tuning of
parameters within an established motor program (Schmidt, 1975; Wulf et al., 2010). This
framework stimulated researchers to test hypotheses about how augmented feedback or practice
structure affected acquisition and retention of sequence tasks (e.g. finger sequence), tracking
tasks (e.g. pursuit tracking), and postural tasks (e.g., stabilometer) (Kantak & Winstein, 2012; K.
R. Lohse et al., 2014; Wulf & Lewthwaite, 2016). One assumption underlying this research is
that each participant has a somewhat equal capability of learning the practiced motor skill and
that manipulating the feedback or practice structure will benefit retention (learning) even though
it may impede immediate performance during the acquisition phase. However, recent research in
motor behavior has identified “non-learners”, a sub-group that fails to learn the practiced task at
a rate comparable to the so-called norm (Golenia, Schoemaker, Mouton, & Bongers, 2014;
Stéphanie Lefebvre et al., 2015; Stéphanie Lefebvre, Dricot, Gradkowski, Laloux, &
Vandermeeren, 2012; Lissek, Vallana, & Guentuerkuen, 2013; Tzvi et al., 2017).
The presence of a “non-learner” sub-group appears to refute the common assumption that
motor learning in an otherwise non-disabled population is a given. Yet, one could ask, is this
subgroup so uncommon? Frequently studies report the presence of outliers, individuals who
10
demonstrate similar behavior to the non-learner subgroup but are instead removed from analyses
because they do not fit our expected hypotheses (Lee & Genovese, 1988). Within clinical
populations, we suggest that outliers or non-learners/non-responders should not be excluded
from analysis, but rather should be studied more deeply to better understand why they exist in
the first place. This double standard has stifled the field and left important questions unanswered.
One consideration is that this non-learner sub-group uses a learning strategy incapable of
achieving task success (Holland, Codol, & Galea, 2018). If we could identify the non-learner
strategy(s) along with a viable explanation for why non-learners use such strategy(s), this would
constitute a valuable contribution to the motor learning field and at the same time, provide a
useful account of the long ignored non-learner phenomenon. Could it be that within an
otherwise non-disabled adult population there are individuals who simply possess inefficient
motor systems incapable of learning a novel motor skill?
To directly address the non-learner phenomenon we adapted a complex motor task first
used by Vernon Brooks and colleagues that identified a large (~25%) sub-group of non-learners
from the total sample population (7/25) (Brooks et al., 1995). These individuals were labeled by
the researchers as “Failure to Learn”. As described, these individuals could not fathom how to
accomplish the task goal.
The Brooks and colleague’s task is unique in that for success, it required the participant
to learn a well-defined task rule for which the solution was not provided. Due to the obscurity of
the correct solution, the learner would need to discover the task rule through trial and error and
feedback about task success until they could consistently achieve the intended goal. The act of
discovery as part of the learning process aligns well with Reinforcement Learning Theory where
a policy/rule associated with goal achievement is learned through a two-stage process of
11
exploration followed by exploitation (Sutton & Barto, 1998). Recent research has demonstrated
the importance of exploration as a possible predictor for future skill level (Babik, Galloway, &
Lobo, 2017; King, Ranganathan, & Newell, 2012; Ranganathan, 2017; Sidarta, Vahdat,
Bernardi, & Ostry, 2016; Singh, Jana, Ghosal, & Murthy, 2016). However, why some
participants naturally explore more than others is unknown.
For the purposes of this research we move beyond the identification of learner sub-groups
and instead focus on the learning process to better understand which individual strategies are
used to achieve success. In doing so, we extend Brooks and colleagues’ work through a more
formal quantification of the different learning strategies associated with performance of non-
disabled adults attempting to learn a complex motor skill. Therefore, our focus is to understand
how a person learns--to directly understand what is learned, with attention to the process and not
simply the outcome (K M Newell, 1991).
This research has two aims. The first is to develop and implement an adapted discovery
task patterned after that used by Brooks and his colleagues (Brooks et al., 1995) and, the second
is to use the performance data from the motor task developed in Aim 1 to quantify the specific
behavioral strategies that are associated with performance and learning. Through identification of
behavioral strategies, we can define what traits are exemplary of a non-learner compared to a
super learner. We hypothesize that there will be clear quantifiable characteristics associated with
distinct learner sub-groups similar to what Brooks and colleagues observed. Further and more
importantly, we hypothesize that the study cohort will demonstrate various levels of early
exploration which can be used to predict success at the individual participant level. Finally, we
explore the relationship between learning outcomes and previous movement-related life
12
experiences as a possible explanation for the differing levels of task success across the entire
cohort.
Methods
Thirty-two non-disabled young adults (mean age: 25.7, age range: 18 – 35, 14 female)
participated in this study after written informed consent was obtained. The participants were all
right-hand dominant, with no previous history of mental illness or injury of the arm/hand.
Discovery Task Characteristics
Prior to practice participants were introduced to the task and the primary goal to transport
a computer cursor (depicted as an astronaut) from a start box to a target box (Figure 2.1A).
Participants were given the following instructions:
“At the start of each trial a cursor which you will control will appear in the left most
START box. A tone will play for 1.5 seconds indicating not to move the cursor until the tone has
stopped (play wait tone). Once the tone has stopped you will have 3 seconds to move the cursor
from the left START box to the right TARGET box (20 cm away). Once the cursor enters the
TARGET box it must stay within the box for 1 second. If you complete the task based on these
parameters a success tone will play (play success tone). If you move the cursor prior to the tone
extinction the trial will start over. If you are unable to move the cursor into the TARGET box in 3
seconds the trial will start over.”
The cursor is controlled by a handheld joy stick which is moved with the thumb of the
dominant hand (Microsoft, Redmond, WA). At the start of each trial participants hear a
continuous chime for 1.5 seconds that signifies the duration of time when the cursor should not
leave the start box. Once the tone terminates the participant has 3 seconds to transport the
astronaut from the start box to the target box. If the participant exits the start box prior to the start
13
tone extinction, the trial aborts and then restarts. If the participant can maintain the cursor
position within the target box for 1 continuous second, they will receive a reward tone, otherwise
no tone is played. Participants practice this task across two days. The first day consists of 200
practice trials and the second day serves as a retention/recall test day consisting of 50 trials
without the start chime or reward tone. Participants are not given any feedback by the research
team nor do they interact with the research team during practice or retention days.
14
Figure 2.1: A. Visual display of the discovery task. The task goal is to land the cursor,
depicted as an astronaut, into the target box within 3 seconds and maintain its position
within the target box for 1 continuous second. If the task goal is achieved, then a reward
tone is played. B. The joystick to cursor movement step function representing the rate
control dynamics of the task. The function applies a continuous pulse of virtual force to the
cursor, astronaut, dependent upon the joystick position. Joystick position (x-axis) is
measured on a sensitivity threshold between -1, leftmost, and 1, right most. If the joystick
passes a sensitivity of +0.5 then a constant positive virtual force of 7.5 is applied to the
cursor. If the joystick passes a sensitivity in the opposite x direction of -0.5 then a negative
virtual force of -7.5 is applied to the cursor.
The discovery task was developed using Unity software version 5.0.1 and in the spirit of
open and reproducible science all scripts used to create the game can be found on the first
author’s Github (Unity Technologies, San Francisco CA,
https://github.com/hooymana/Brooks_Task.git). The mapping of joy stick movement to cursor
movement was determined by rate control, whereby once the joystick is moved past a specific
positional threshold in the positive x direction a constant virtual force is applied to the cursor in
that direction. Conversely, if the joy stick is moved past the same positional threshold but in the
negative x direction then a constant virtual force is applied to the cursor in that direction. This
rate control requirement creates a step function that describes the mapping between joystick
movement and cursor acceleration (Figure 2.1B). Participants were kept naïve to this cursor to
joy stick mapping until after the conclusion of the experiment.
It is this rate control mapping that the participants must discover and ultimately learn to
control to achieve success and a reward tone. Due to the imposed time and maximum force
15
constraints there is a limited number of possible joy stick movements that allow for task success.
To confirm this limited set of successful movement patterns we simulated over 700,000 possible
cursor acceleration profiles and identified that approximately 9% of all possible acceleration
profiles yield a successful trial (Figure 2B). The percentage of successful trials can be modulated
based on manipulation of task constraints (i.e. larger target area, greater trial time, increased
virtual force). Earlier pilot work demonstrated that the task constraints chosen here were
challenging enough and sufficient to identify individual changes in performance and learning.
How well the cursor position to virtual force step function, or task rule, is learned by each
participant can be measured from trial to trial by a metric known as the scaling ratio.
Quantification Metrics
Scaling ratio is a ratio determined by negative acceleration (numerator) over positive
acceleration (denominator) of the controlled cursor (Brooks et al., 1995). Given the constraints of
the task, a scaling ratio equal to or close to 0 would indicate no movement or only a positive
acceleration of the cursor. A scaling ratio close to or equal to 1 indicates equal levels of
acceleration were applied to the cursor in both the positive and negative direction. Scaling ratio
can easily identify if a participant has accurately executed the task rule, that being a scaling ratio
close to 1. We observed that early in practice, participants perform a movement resulting in a
scaling ratio at or near 0, (i.e. the cursor flies past the target and off the screen with no reversal
taking place). These early attempts suggest that the participant’s initial understanding of joystick
to cursor mapping is based on positional control (i.e. when the joy stick ceases to move, the
cursor should cease to move). However, it is possible to fall short of the target, due to a short
duration of positive and negative virtual force applied to the cursor and reach a scaling ratio near
1. Therefore, it is not only necessary to have equal amounts of positive and negative acceleration
16
but also accurate timing (i.e. length of time applying both positive and negative acceleration) to
be successful. This is illustrated in Figure 2.1.
17
Figure 2.2: A. Example data of average acceleration profiles collected from a single
participant. The 49 successful acceleration profiles in blue with the blue shaded area
representing one standard deviation and 151 failed trials with the red shaded area
representing one standard deviation. B. Simulations of successful acceleration profiles.
Over 700,000 possible acceleration profiles were simulated to determine an average
successful acceleration profile. Approximately 9% of all possible acceleration profiles
would be successful. Notably each of these successful profiles resemble each other with only
a slight variance in execution. Shaded blue area is standard deviation of all successful
simulated trials. C. Example of a successful trial visualized through the collected position
and acceleration data. Total trial length is 450 samples which is 4.5 seconds in length. Grey
shaded box with solid black line represents the black box in space and the grey shaded box
with the dashed shaded line represents the target box in space (right x-axis). The blue line
represents the cursor motion through the virtual space where its final position is
maintained in the target box for 1 second resulting in a reward tone. The vertical dashed
line represents the end of the start chime indicating when the participant should not exit
the start box. Although the cursor may begin moving within the start box before the chime
becomes quiet. The red line represents the virtual acceleration of the cursor which is
generated based on the position of the thumb joystick (Figure 2.2C). We calculate scaling
ratio by dividing the negative virtual acceleration over the positive virtual acceleration.
Time outside the start box is determined by summing the total number of samples the
cursor spent outside the start box throughout the trial. Maximum time outside the start
box is 300. We can determine reversal of the joystick based on when the cursor experiences
a negative virtual acceleration.
18
In addition to scaling ratio, we examined two other critical control features: 1) start chime
anticipation (i.e. reaction time measured as time outside the box) which reflects the ability to
quickly exit the start box after the start que, and 2) reversal timing, the point, if any, that the
participant reverses the joystick direction and applies a negative virtual force to the cursor. Since
the rate control step function is based on a positional threshold there is no advantage to how
quickly the joystick can be moved as maximal force application is limited. Therefore, for this
task, the primary skill required for task success is one of precise timing, first to anticipate when
to move the cursor out of the start box (i.e., time outside the box), second to reverse virtual force
application at the optimal time (i.e. reversal timing), and third to apply positive and negative
acceleration for the appropriate temporal duration (i.e. scaling ratio). To identify these
performance metrics: time outside the start box, reversal timing, and scaling ratio, we acquired
position and acceleration data of cursor movement using a 100 Hz sampling rate. Calculation of
each metric can be visualized in Figure 2.2C.
Statistical Analysis
Our statistical analysis approach is two-fold: 1) to confirm the presence of different
performer sub-groups within our cohort, and 2) to identify and quantify group level and
individual level learning strategies. To accomplish this, we performed four sets of analyses
which fall into four categories: confirmatory, group level, individual level and exploratory.
Confirmatory
To confirm the relationship of the task rule to task success we used scaling ratio, our
constructed task rule, time outside start box and time of joystick reversal to predict successful
trials across all subjects. We used linear and logistic regression for this. First, we examined the
linear relationship between each performance metric to time in target. To confirm the task rule to
19
task success relationship we used logistic regression and leave one out cross validation to
determine overall prediction accuracy across all participants. We also attempted combinations of
performance features to identify an optimal task rule that would best summarize individual trial
performance. We reasoned that identification of a rule would allow for better quantification of
different learning strategies in subsequent analyses.
Group Level
K-means clustering was used to identify sub-groups similar to those previously identified
(Hartigan & Wong, 1979). The number of clusters included in the analysis was determined by
the within sum of squares across all cluster possibilities. Linear regression was used to determine
how sub-groups updated their performance (scaling ratio (i + 1)) in relation to previous task
feedback (Time in Target (i)). Task feedback was defined as the visual information of the cursor
relative to the target box and quantified as Time in Target.
Sub-Group Level
To identify learning strategies at the sub-group and individual level, exploration was
calculated as variability of scaling ratio during a binned series of trials. We reasoned that a large
scaling ratio standard deviation would indicate high levels of exploration compared to a small
standard deviation that would be indicative of a low level of exploration (i.e. the same movement
pattern used repeatedly). Linear regression was used to identify how current task feedback (i.e.,
start timing, cursor position, success tone) predicted future exploration within each sub-group.
Individual Level
Linear regression was used to determine if an individual’s exploration strategy during the
early phase of practice – defined as performance during the first 80 trials - could predict
performance and learning (retention) of the discovery task.
20
Exploratory
Participants were surveyed about how many lifetime hours they would estimate that they
spent engaged in video games and/or physical activity (see Supplement). We used Linear
regression to determine if previous movement-related life experience had any predictive effect
on task performance and learning.
Results
Relationship between Task Outcomes and Learning: Confirmatory
The linear regression model between group-level task rule (scaling ratio) learning and
mean performance (time in target) from Day 1 revealed a strong positive and predictive
relationship (R
2
= .87, p < .05, Figure 2.3A). Further, regression analysis between scaling ratio
and number of successful trials demonstrated a strong positive relationship (R
2
= .65, p < .05)
[not shown]. The association between other quantitative metrics indicative of task rule learning
and time in target box ranged in strength from mean time outside of start box, R
2
= .78, p < .05
(Figure 2.3B), to none for mean reversal time, R
2
= .1, p > .05 [Not Shown].
21
Figure 2.3: A. Individual participant data demonstrating relationship between mean
scaling ratio and mean time in target on Day 1. B. Individual participant data
demonstrating relationship between time outside start box and average time in target on
Day 1. C. Individual subject data demonstrating relationship between mean modified
scaling ratio and mean time in target on Day 1. The modified scaling ratio represents a
better fit of average time in target for each individual participant. Color coding and
centroids (larger points) in C represents results of k-means clustering with color coding
representing each of three different sub-groups. Shaded area represents standard error of
prediction.
To improve predictability between the task rule and time in target we modified the
scaling ratio metric by multiplying it by the average time outside the start box (R
2
= .94, p < .05,
Figure 2.3B). We chose to combine these features due to their inherent relationship with one
another, a source of collinearity in any multivariate prediction. Additionally, the combination of
22
these two features--time outside start box and scaling ratio--created a new scaling ratio measured
between 0 and 300 units. This modified scaling ratio metric provided a means to quantify trial
performance with a single measure. As such, a modified scaling ratio closer to 300 was more
predictive of a longer time in target compared to a modified scaling ratio closer to 0. The
modified scaling ratio was used to predict trial success in a logistic model across all participant
Day 1 trials that was statistically significant (χ 2 (3798,2) = 889, p < .001). Leave one out cross
validation of the logistic model resulted in an overall accuracy of 91% and a calculated ROC
curve with an area under the curve of 0.93, not shown. Therefore, with the modified scaling ratio
we could predict trial success with a high level of accuracy.
Sub-group Identification: Group Level
In accordance with our statistical analysis approach, the first analysis between mean Day
1 modified scaling ratio and time in target revealed three distinct groups (Figure 2.3C). We have
labeled these groups based on their relative performance in the task: High Performers (HP, N =
11), Moderate Performers (MP, N = 12) and Low Performers (LP, N = 9). We use these three
emergent groups for all subsequent analyses to uncover differences in performance strategies,
between sub-groups. Similar to Brooks and colleagues, the LP sub-group (green in Figure 2.3C)
contained 7 out of 9 participants who achieved zero successful trials on both Day 1 and 2 of the
experiment. In the discussion, we explain the rationale for naming this group as Low Performer
instead of Non-learner. Visualization of individual participant time series data of modified
scaling ratio by trial for each of the three groups on Day 1 is illustrated in Figure 2.4.
23
Figure 2.4: Individual subject data for each of the identified sub-groups. Modified scaling
ratio is depicted on the y axis and trial number is on the x axis for Day 1. High performers
in red, moderate performers in blue and low performers in green. The black line within
each plot is a loess fit.
Sub-Group Level: Exploration
For this analysis we used standard deviation of the modified scaling ratio as the
dependent measure. To identify the relationship between individual exploration and time in
target box we first identified how changes in exploration fluctuated as a function of practice. We
examined exploration using 10-trial blocks across Day 1 and observed the changes in magnitude
of exploration across blocks. Examination of the exploration time course for each sub-group
(Figure 2.5A) demonstrated unique differences with the greatest interaction between sub-groups
occurring during the early trials (i.e. first 10 trial blocks). Comparison of average exploration for
24
the entire day versus average performance for Day 1 reveals an inverted U relationship which is
continuous across groups (Figure 2.5B).
Figure 2.5: A. Time course over practice of exploration for each sub-group. B. Average
Time in Target (Day 1) versus Average Day 1 Exploration for each individual. Participants
are color coded to their cluster analysis group. Black line in B is a fitted quadratic based on
the relationship between average exploration and average time in target on Day 1, shaded
area is standard error of relationship.
We systematically explored which bin of exploration would be most predictive of future
performance and learning. Each bin of exploration, ranging from trial 1 to 190, spanned by
increments of 10, could significantly predict time in target during the last ten trials on Day one,
trials 190 to 200, and average retention of Day 2, average time in target for all 50 trials (Figure
2.6A). Trials 1 to 80 were most predictive of time in target for trials 190 to 200 (R
2
= .48, p <
.05, Figure 2.6B) and retention – Trials 1 to 50 (R
2
= .45, p < .05, Figure 2.6C). A closer
examination of Figure 2.6A illustrates a sharp rise with a bimodal distribution followed by a
25
more gradual fall of explanatory power of our exploration metric as practice progresses. This
fluctuating dynamic of the predictive power of exploration supports the existence of an
exploration phase as one part of goal achievement, but clearly is not the entire story. In Figures
6B and C, color coded by sub-group demonstrates that low performers exhibit the least amount
of exploration and ergo the lowest time in target during late performance and retention thereby
highlighting the importance of exploration for task learning. However, one explanation for why
exploration is not more predictive of future performance may be due to the varying individual
capability to exploit the task rule and improve future performance. We elaborate on this idea in
the discussion.
26
Figure 2.6: A. Systematic examination of exploration across 10 trial blocks from 1:10 to
1:190 that most predicts future performance (trials 190-200) and retention/recall (trials 1-
50). B. Average exploration by sub-group during the first 80 trials (peak in A) predicts
average time in target during the last 10 trials on Day 1. C. Average exploration by sub-
group during the first 80 trials (peak in A) predicts time in target during retention (Day 2).
Black line is linear regression result and shaded area is standard error of prediction in B
and C.
Sub-Group Level: Performance Updating
Based on the finding that the exploration metric predicts later performance (Figure 2.6)
we asked does previous performance influence future exploration? If previous experience does
hold a relationship to subsequent exploration, then we may observe differences in this
relationship at the sub-group level. Our objective was to better understand if the relationship
between current task feedback and future exploration would vary at the sub-group level. If a sub-
group demonstrated the canonical exploration/exploitation relationship, then we would expect to
see a negative relationship between current feedback and future exploration where increases in
time in target would lead to decreases in exploration. Our approach consisted of binning both
time in target and exploration by blocks of 2 trials and then performing linear regression to
analyze how previous feedback (i.e. time in target at block i) would predict an increase or
decrease in future exploration (i.e. exploration at block i+1). A linear regression model was used
to determine the updating relationship for each sub-group with results demonstrating a positive
relationship between feedback (i.e. time in target) and future exploration for low performers (LP:
Beta = .13, R
2
= .287, p < .05), no relationship with moderate performers (MP: Beta = .03, R
2
= -
.009, p > .05) and a negative relationship for high performers (HP: Beta = -.46, R
2
= .241, p <
27
.05; Figure 2.7A). Overall, the results of the updating analyses indicate that different performers
use different initial exploration strategies to accomplish the task goal but also refine their
exploration strategy based on performance feedback (Figure 2.7A). Interestingly, when we plot
each sub-group together there appears to be a continuous relationship in the form of an inverted
U (Figure 2.7B). Plotting the data in this way suggests that performance of each group is not a
unique function of their performance but a continuous function of skill learning.
We then analyzed the relationship between task feedback at the ith trial relative to future
modified scaling ratio of the next trial (i +1) and found that sub-groups fell on a continuum
similar to Figure 2.7B but rather with a logarithmic relationship compared to the quadratic
inverted U (Figure 2.7C). The logarithmic relationship fit the data significantly better than a
linear model providing evidence that as a performer improves on the task their modified scaling
ratio stabilizes and reaches a plateau. Additionally, color coding of sub-groups along this
logarithm demonstrates that high performers falling on the right most portion were more capable
of exploiting the rule compared to moderate performers. As demonstrated in Figure 2.6B and
2.6C high performers were also greater explorers compared to the other sub-groups indicating
this group is efficient at both exploration and exploitation thus leading to their greater success on
the task. To determine if the inverted U relationship found in Figure 2.7B is the result of a skill
continuum we looked at the progression of data points from one bin to the next to see if it shifted
with practice (Figure 2.7D). Data point size in Figure 2.7D demonstrates the progression of the
exploration/performance relationship with each group shifting from left to right along the
inverted U over practice.
28
29
Figure 2.7: Relationship between previous task feedback (i.e. time in target), to next trial
exploration on Day 1. A) The Low Performer sub-group demonstrated a positive
exploration/performance relationship, Moderate Performers exhibited no relationship and
High Performers maintained a negative relationship. Black line represents results of
regression and shaded area is standard error of prediction. B) When each sub-group is
plotted on a continuum it creates an inverted U relationship that spans across the sub-
groups. Black line is result of regression and shaded area is standard error of prediction.
C) The relationship between next trial modified scaling ratio (Y-axis) and current Time in
Target (X-axis) plotted across sub-groups which are differentiated by color. Maximum
Time in Target is 100 centiseconds. D) Progression of Exploration (Y-axis) by Average
Time in Target (X-axis) for each sub-group. Point size represents progression of the
exploration/performance relationship across bins of trials (see legend) with smaller points
representing earlier practice trials and larger points reflecting later practice trials.
Animation of Fig 2.7B demonstrating the change in scaling ratio variance in relationship to
previous time in target with bin progression can be viewed here: https://youtu.be/-
fjdyDfl9SY.
Exploratory Analysis: Previous Experience
We understand the challenge that comes with creating a truly novel motor task for adult
learners. Our task, adapted from Brooks and colleagues’ original work, reproduced sub-groups
who demonstrated similar performance strategies to what he and his colleagues described over
two decades ago in their 1995 paper. One possible explanation for these individual performance
differences could be differences in previous life experience with a similar class of actions. To
test this assumption, we administered an open-ended survey after task practice to analyze how
30
previous movement-related life experience would predict time in target box (i.e. success). The
two following questions were asked: 1) How many hours of Video Games have you played in
your life? 2) How many hours of sport/physical activity have you practiced/participated in your
life? Linear regression found the more hours a person accumulated in both video games and
sport had a greater mean time in the target box on Day 1 (R
2
= .509, p < .001, Figure 2.8).
Additionally, self-reported experience was also positively associated with individual exploration
– trials 1 to 80, (r = .41, p < .05, not shown). We interpret these results as a strong indicator that
skill acquisition of a novel motor task depends on prior experience with a similar class of
actions. For those individuals with low prior experience, who also represent the Low Performer
sub-group, we see lower time in target.
31
Figure 2.8: Relationship between Previous experience and Average Time in Target on Day
1. Participants who reported a low number of hours playing video games and physical
activity had the lowest average time in target compared to participants who reported
higher hours of video game play and physical activity throughout their life. Black line
represents regression results shaded area is standard error of prediction.
Discussion
To our knowledge, this is the first study that used rigorous quantification metrics of task
performance to demonstrate specific behavioral learning processes and to map individual-level
learning processes onto sub-groups of performers attempting to learn a novel motor task. This is
also the first time, to our knowledge that anyone has shown the predictive power of self-reported
previous experience with a similar class of actions to individual performance with a novel motor
task. Our first aim was to develop an adapted motor task previously used by Brooks and
colleagues. Having that motor task allowed us to conduct a set of confirmatory and follow-up
group-level and individual-level analyses to better understand what and how a supposedly
homogeneous group of non-disabled young adults acquired skill through execution of a novel
motor task.
Akin to Brooks and colleagues, we found a similar frequency (and %) of non-
learners/Low Performers within the entire sample; Brooks et al., 7/25 (28%) (Brooks et al.,
1995), Hooyman et al., 9/32 (28%). Another parallel with Brooks and colleagues was the
emergence of two other sub-groups with similar performance curves within the two studies (i.e.
“Non-Searchers”/Brooks to High Performers/Hooyman and “Searchers”/Brooks to Moderate
Performers/Hooyman). In contrast to Brooks and colleagues’ approach, our study goes beyond
32
the descriptive label and for the first time, accurately quantifies the exploration
1
strategies used
by young non-disabled adults while acquiring the task rule governing a complex motor task.
Our primary aim was to identify and then quantify strategies used to acquire skill of a
motor task for which the solution is not given. Complex motor tasks that require discovery of a
rule have rarely been used in typical laboratory-based motor learning studies that routinely draft
young non-disabled undergraduate students (Karl M. Newell, Liu, & Mayer-Kress, 2009). In a
rare exception, focused on the dynamics of motor learning, Liu and colleagues reported three
different performance curves among individual participants who practiced a novel roller ball
task: power curve - High Performer, s-shaped - Moderate Performer, and linear/flat - Low
Performer (similar to our data, Figure 2.4) (Liu, Mayer-Kress, & Newell, 2006a). Unfortunately,
Liu and colleagues did not pursue the question of how and why these different patterns emerged.
To the best of our knowledge, we are the first group to offer an explanation for how and why
these different performance patterns emerge. In contrast to the learning that follows the well-
known power law, our results suggest that success in learning a complex motor task relies on: 1)
high levels of exploration early in practice (see Figure 2.5 and 2.6), aimed to uncover the task
rule and 2) accurate exploitation of that acquired rule later in practice (see Figure 2.7C). Further,
our findings support the idea that the probability for implementation of such a two-step strategy
is predicated on the amount of previous experience or exposure to a similar class of actions.
Our focus on “learning strategy” instead of task success allowed us to determine that
there was no evidence to support the idea of “non-learners” among those who achieved zero
successful trials—instead we labelled this sub-group, “low performer”. Further, we showed that
1
*Brooks et al. (1995) identified sub-groups within their experiment as searchers, non-searchers and
failure to learn. He used the term search instead of explore although they did not provide quantitative
evidence that participants indeed searched for the task rule.
33
their performance and learning was strongly predicted by the level of task exploration they
exhibited in the first 80 practice trials and their relatively low level of prior experience engaged
in similar motor-related activities (e.g. video games, physical activity). This finding provides a
testable hypothesis for future work that could potentially elevate the low performer to a moderate
or even high performer; something the non-learner label would not suggest. Further, we
speculate that this perspective is relevant to the implementation of rehabilitation strategies where
focus should be on the precise development and then application of a novel motor program
especially in the case of erroneously labelled non-learners/non-responders (see Low Performer
Phenomenon below).
The Process of Rule Discovery
We hypothesized that an early level of exploration would predict performance and
learning across the two days of the experiment. Our findings do support our hypothesis that
exploration is a crucial type of meta-skill necessary to acquire a novel motor skill. Although, as
our data indicates only a portion of the total sample can implement this strategy optimally – High
Performers (see Figure 2.5A & 2.5B). Those who achieve task success, moderate and high
performers, demonstrated a capability to utilize previous performance feedback to then explore
and update the future motor command (see Figure 2.7). Interestingly the moderate performers
were able to explore and discover the task rule but had difficulty switching to an exploitation
strategy as evidenced by their flat exploration to time in target relationship (see Figure 2.7A) and
concomitant high levels of exploration exhibited throughout the experiment (see Figure 2.5A).
Our data indicate that skill on the task exists on a continuum that utilizes both an exploration and
exploitation process (Sutton & Barto, 1998). As such, the provision of an intervention to speed
up the rule discovery process would need to be customized to the specific sub-group deficit.
34
Among the sub-groups the Low Performers may benefit from a practice intervention or priming
that is focused on exploration; Moderate Performers may benefit from a practice intervention
focused on exploitation of the task rule; whereas High Performers may benefit from doing what
they are already doing and require no intervention. Future research designed to test these
customized intervention strategies is needed to advance our understanding of the motor skill
learning process, the remediation of ineffective strategies and its application to the aging and
rehabilitation context.
The Low Performer Phenomenon
The Low Performer group demonstrated a drastically different performance strategy than
the Moderate- and High-Performer groups. Despite 200 practice trials, compared to the moderate
and high performers, the low performers demonstrated limited exploration of the critical
modified scaling ratio and were therefore unable to reach a comparable level of task success (see
Figure 2.7). In fact, the low performers appeared to perseverate on a strategy which led to little
change in time in target. Interestingly, recent research has identified a similar sub-group of non-
learners when engaged in a reinforcement learning task (Holland et al., 2018). Though not
causal, one explanation for this perseverative behavior is offered by the strong correlation
between our success metric (i.e., average time in target) and prior movement-related experience
(see Figure 2.8). Without the rich repertoire allowed through prior experience, the low
performers appear to not know how to change their outcome. While a nurture perspective (i.e.,
prior experience) offers one explanation for perseveration of an unsuccessful strategy, a nature
perspective offers another viable candidate. The genetic variant such as the single nucleotide
polymorphism, BDNF val(66)met has been shown to influence motor system function including
experience-dependent plasticity (Kleim et al., 2006; McHughen et al., 2010). Thirty percent of
35
the population have this genetic variant and interestingly a similar percentage of our cohort 9/32
(28%) and Brooks and colleagues’ cohort 7/25 (28%) were identified as low performers/non-
learners. Unfortunately, we did not consider a genetic hypothesis in the design of our study and
therefore do not have the genetic information on our sample needed to directly test it. This is an
important hypothesis for future testing.
Overall, the ability to explore an unknown dynamical mapping can be considered a skill
unto itself. The ability to explore and then identify aspects of the task which provide further
insight on how to improve performance could be considered a type of meta learning skill (Braun,
Aertsen, Wolpert, & Mehring, 2009; Green & Bavelier, 2015; Krakauer & Mazzoni, 2011). It
could be that these capabilities of exploration and updating may allow for learning and transfer
across a spectrum of motor skills in a more generalized way (R. D. Seidler, 2007). Using a clever
design, Tim Lee and colleagues demonstrated that the contextual interference effect was not due
simply to practice structure but to how practice structure engages the learning process (Lee,
Wulf, & Schmidt, 1992). It is possible that the High Performer group had a plethora of prior
opportunities to engage an explore and exploit strategy that enabled them to more quickly
transfer that strategy and identify essential task control features needed to solve the movement
problem (Braun et al., 2009).
These sub-group differences in learning strategy for skill acquisition may also be linked
to individual differences in brain activity or connectivity. Recent work has demonstrated that
active and resting state brain connectivity are predictive of individual performance improvement
of a practiced motor skill (Stéphanie Lefebvre et al., 2012; R. Seidler et al., 2015; R. D. Seidler,
2010; Wu et al., 2014). Our ongoing research is focused on the identification of neural correlates
of individual learning strategies used for this discovery motor task. New insights into this
36
understudied area of motor learning are likely once the brain-behavior relationship is better
understood.
Together, we discussed three possible explanations for the low performer phenomenon—
experience, genetics, and brain connectivity—given that the brain is a learning machine, it is
unlikely that these domains are completely independent of one another. Future work including
metrics across these domains will undoubtedly provide important new knowledge and direction
moving forward.
Conclusion
We developed an adapted discovery task patterned after that used by Brooks and
colleagues in their now seminal 1995 paper, “Learning what and how in a human motor task”.
With our task we successfully reproduced similar behavioral results to Brooks. We then
quantified what was learned, and how it was learned differently among three different sub-groups
identified in an otherwise homogeneous group of non-disabled young adults. Our findings
highlight the importance of an exploration strategy for future performance and learning of a
complex rule-based motor skill. We provide evidence that the unique learning strategies among
the three sub-groups align along a performance continuum suggesting that supposed non-learners
fall within the early stage of skill learning. This position is further supported by a strong positive
relationship between individual self-reported previous experience and overall performance.
Overall, our results indicate that low performers are simply inexperienced with such rule-based
tasks and are not doomed to a life of non-learning due to a faulty motor system. Forthcoming
research is ongoing to investigate the neural correlates of these behavioral differences in motor
performance and more specifically those associated with exploration of the task space.
37
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning
Abstract
Intracortical Resting-state electroencephalography (rs-EEG) can serve as a robust biomarker to
predict motor skill acquisition, cognitive capability and stroke recovery. The aim of this study is
to identify the intracortical connectivity from rs-EEG signals that accurately predicts
performance at each phase of discovery motor learning (i.e., exploration, exploitation and
retention). Discovery learning is a process used to acquire complex motor skills when the task
rule(s) for success are unknown. It requires the learner to engage in a two-phase process. First is
exploration where the learner must explore the task space using a variety of movement patterns
to find an unknown rule that is linked to task success. Second is exploitation that includes
refinement of the recently found task rule to enable consistent success and continued
performance improvements. We recruited 32 right hand dominant, non-disabled adults to
participate in a two-day study. On day 1, participants underwent 5 minutes of eyes open rs-EEG
and then practiced for 200 trials to achieve the task goal. Participants returned the next day for a
retention test where they practiced for another 50 trials without feedback on task success.
Support Vector Machine regression was used to identify connectivity pairs predictive of each
phase of discovery learning, and false discovery rate correction was used to identify the most
significant pairs. The exploration phase was predicted by a bilateral prefrontal network with
leave one out cross validated (LOOCV) R
2
= 0.42, p < .05. The exploitation phase was predicted
by an ipsilateral prefrontal/motor network with a LOOCV R
2
= 0.46, p < .05. The retention phase
was predicted by a bilateral prefrontal, motor and parietal network with a LOOCV R
2
= 0.26, p <
.05. These findings are generally comparable to previous resting-state research in motor learning.
38
However, the novelty of this experiment is how the different learning processes within a
discovery learning task: exploration, exploitation and retention are each predicted by a distinct
set of resting-state intracortical connectivity.
Introduction
Identifying the neural substrates associated with motor skill learning is critical to
understanding brain correlates of effective physical interaction with the environment. A recent
meta-analysis aimed at identifying brain activity linked to motor learning found the primary
motor, premotor and prefrontal cortex of the hemisphere contralateral to the hand performing the
task to be most associated with motor skill acquisition (K. R. Lohse et al., 2014). However, the
experiments included in this analysis used a relatively narrow set of typical motor tasks including
motor sequence and rotor pursuit tasks—tasks that Newell (K. M. Newell, 1991) refers to as
parameterization type tasks—requiring the learner to adjust a few parameters of an already
known task rule, but not to discover the task rule necessary for success. The study of motor
learning that is provided by such a limited set of tasks may therefore represent a narrow view of
how humans learn (K. M. Newell, 1991). Further, this type of learning may not be representative
of how people with motor impairments, either neurological or musculoskeletal in origin, re-learn
motor skills (Karl M. Newell & Verhoeven, 2017). This idea challenges the usefulness of current
brain-behavior relationships, i.e. motor cortex activity associated with motor learning, for future
developments in application such as neurorehabilitation. This is especially relevant, given that
the outcome of recent clinical trials demonstrate little to no effect of neuromodulation on primary
motor cortex to promote motor recovery after stroke (Harvey et al., 2018; Levy et al., 2016).
39
To address the task problem, work in our lab uses a motor task for which the task rule
must be discovered through practice. The process of motor learning that this task enables is
known as discovery learning (Hooyman et al., In Preparation – Chapter 2). Adapted from
previous research, the process of learning the discovery task is characterized by the necessary
implementation of a search strategy that uncovers the task rule mediating success (Brooks et al.,
1995). The process of learning this type of task is known to engage a two phase process (similar
to reinforcement learning) where first the learner must explore the task space to identify the rule
and then exploit the rule to consistently achieve the task goal (Sutton & Barto, 1998). This two
phase learning process is a profoundly understudied area of motor learning and yet, it may
provide the most ecologically valid information about the motor learning process (Liu et al.,
2006a; Sailer, Flanagan, & Johanssen, 2005). Indeed there is long standing theoretical support
for the exploration phase as being integral to novel motor skill learning (Gentile, 1972). With
each phase representing a distinct behavior it is reasonable to assume that each phase is carried
out by a different area or network of the brain. However, the resting-state neural correlates
associated with each phase (i.e. exploration, exploitation and retention) have not until now been
investigated.
Additionally, up until recently, research into the neural correlates of motor learning have
used contrasts between active and rest periods during performance to uncover separate brain
regions associated with motor learning. This approach may be another factor limiting our
understanding of how the brain learns. We view these older methods, which have previously
dominated neuroimaging research in motor learning, as limited in scope and not particularly
relevant for understanding how information exchange within the brain functions during a
dynamic learning process (Lewis et al., 2009a; Sami et al., 2014). In contrast, recent research
40
focused on the identification of essential intracortical connectivity--a measure of shared
communication, between distinct brain regions--may better reflect how the brain functions to
learn a novel motor skill (Carter, Shulman, & Corbetta, 2012; Hardwick et al., 2015). Moving
forward toward more meaningful neural correlates of motor learning, it will be important to not
only investigate other forms of motor learning such as discovery learning, but to also analyze
neural data in terms of multi-modal connectivity, such as non-motor mediated predictors of
motor skill learning (Deco, Jirsa, & McIntosh, 2011a).
One neuroimaging modality that has been used to identify robust predictors of motor skill
acquisition, motor skill learning and recovery from neurological injury (i.e. stroke) is resting-
state electroencephalography (rs-EEG) (Wu et al., 2015, 2018, 2014). Importantly, over the last
decade the resting brain has demonstrated a unique ability to predict a variety of cognitive, motor
and clinical outcomes (Mary et al., 2017; Siegel et al., 2016; Zou et al., 2013). Further evidence
suggests that the spontaneous activity of the resting brain corresponds to the underlying
anatomical connectivity of an individual’s brain (Deco, Jirsa, & McIntosh, 2011b; Jarbo &
Verstynen, 2015; Tomassini et al., 2011). Resting brain activity has been found to be stable
overtime and related to brain regions that are active during task execution, making it a strong and
reliable proxy for how the brain functions while learning (Shah, Cramer, Ferguson, Birn, &
Anderson, 2016).
Previous research has used rs-EEG to predict acquisition and retention of motor skills
such as motor sequence learning and visual adaptation (Manuel, Guggisberg, Thézé, Turri, &
Schnider, 2018; Wu et al., 2018, 2014). Each of these experiments found that prefrontal, motor
and parietal areas of the dominant hemisphere could predict changes in performance, similar to
work demonstrating the association of active brain regions to task practice (K. R. Lohse et al.,
41
2014). However, limitations of these studies stem from the question of what was learned?
Practice of the typical laboratory-based motor tasks for which the task rule is known to the
learner from the outset, may not enlist the same kind of problem solving operations that would
be used for practice of more complex tasks where the task rule is unknown at the outset. As such,
what is learned from these contrasting motor task practice paradigms is likely very different. We
and others suggest that the discovery task paradigm may provide a more ecologically valid
experimental paradigm for understanding what was learned across a continuum of real life
scenarios from sports performance to neurorehbilitation (Karl M. Newell & Verhoeven, 2017).
Previous studies to determine the neural correlates of motor learning implemented a type
of seed analysis to identify intracortical connectivity predictive of learning. The seeds were
either placed in the primary motor or parietal cortex contralateral to the hand performing the task
and thus negated any potential connectivity patterns that might exist outside of the canonical
motor system. In contrast, recent work using resting-state functional Magnetic Resonance
Imaging to predict stroke impairment used a whole brain analysis and found that predictive
functional connectivity was widely dispersed throughout the entire brain (Siegel et al., 2016).
The purpose of this experiment is to identify beta band intracortical connectivity that best
predicts the different phases of discovery motor learning using rs-EEG. We hypothesize that
each phase of discovery learning--exploration and exploitation--can be predicted by rs-EEG and
that intracortical connectivity predictive of each phase will be distinct from one another.
Additionally, we hypothesize that intracortical connectivity will predict next day retention--
performance without feedback--which will also be distinct from the exploration and exploitation
practice phases. Further, we expect that each performance-learning phase will be predicted by a
connectivity network that incorporates both motor and non-motor cortical regions.
42
Methods
Thirty-two non-disabled adults (mean age: 25.7, 14 female) were recruited and consented
into the study. Each participant was assessed for hand dominance using the Edinberg
Handedness questionnaire (Oldfield, 1971). Any participant who was not right hand dominant, or
had any previous orthopedic injury or surgery to the right hand, or had a previous history of self-
declared mental illness was excluded from the study.
43
Figure 3.1: Visual depiction of study timeline. On Day 1 Participants are first consented
into the study and then are oriented with an EEG cap specific to individual head
circumference which is then calibrated. Prior to task practice they undergo five minutes of
eyes open resting-state EEG with eyes focused on a fixation cross. Then Participants
practice 200 trials of the discovery task with start chime and reward tone present
throughout. Day 2 begins the day after Day 1 beginning with 50 trials of the discovery task
where no start chime or reward tone is given (i.e., retention/recall phase). After the 50
trials are completed, participants are given a general survey questionnaire about their
previous lifetime experience with video games and/or physical activity reported in total
hours spent. The study concludes with participants being debriefed on the study purpose.
EEG Cap Orientation and Resting-State Measurement
Each participant was fitted with a 64 EEG cap specific to their head circumference. The
cap is oriented to the 10-20 EEG electrode system (Figure 3.2B). This electrode layout assumes
that specific electrodes overlay specific areas of the cortex. Given that we did not use individual
structural MRI images to confirm electrode placement, the cortical area to electrode mapping
remains an assumption of the study.
44
Figure 3.2: Eyes-open Resting-State EEG set-up: A) Participants stared at a fixation cross
for 5 minutes prior to task practice while EEG activity was recorded. B) Electrode layout
corresponds to the 10-20 system. Blue electrode (GND) served as the ground and the green
electrode (CPZ) functioned as the reference electrode for all participants.
With the cap oriented, each electrode is calibrated with a saline gel to adjust for
impedance, <= 20 kOhms. Electrode signals are compared to a single reference electrode, CPZ
(Green electrode, REF, in Figure 3.2). Once the cap is calibrated, participants undergo a single 5-
minute session of rs-EEG. During the rs-EEG session participants remain seated with eyes
focused on a fixation cross. They are given specific instruction to keep their face muscles relaxed
for the 5 minute session; a procedure to reduce electrical activity from eye and facial muscles
that could be detected by the EEG electrodes. To encourage a state of mind wandering,
participants are not given any specific instructions about what brain state to assume (Deco et al.,
45
2011a). Once the resting state session is complete the EEG cap is removed, and the task practice
session begins.
Discovery Task Practice and Retention
The discovery learning task (Figure 3.3A) was created using Unity software version
5.0.1f1 (Unity Technologies, San Francisco, CA). Cursor control is performed through
manipulation of a joystick designed to capture movements of the right thumb (Microsoft,
Redmond, WA). Participants practice the discovery learning task on a laptop computer (screen
dimensions: 16” wide x 12” long, Figure 3.3C). Prior to task practice, participants are given the
following instruction about the goal and time constraints of the task.
“At the start of each trial a cursor which you will control will appear in the left most START box.
A tone will play for 1.5 seconds indicating not to move the cursor until the tone has stopped
(play wait tone). If you move the cursor prior to the tone extinction the trial will start over. Once
the tone has stopped you will have 3 seconds to move the cursor from the left START box to the
right TARGET box (20 cm away). Once the cursor enters the TARGET box it must stay within the
box for 1 second. If you are able to complete the task based on these parameters a success tone
will play (play success tone). If you are unable to move the cursor into the TARGET box in 3
seconds the trial will start over.” (See Figure 3.3 for visual depiction of task screen and outcome
measures)
46
47
Figure 3.3: A) Display of discovery learning task. Astronaut represents the cursor which
the participant controls with a joystick designed for the right thumb. Left box is the
START box and right box is the TARGET box. At the start of each trial a 1.5 s start tone
warns the participant to leave the START box, at the off set of the start tone. Beginning
with the off set of the start tone, the participant has 3 seconds to land the cursor in the
TARGET box. If the cursor stays in the TARGET box for one continuous second then a
reward tone sounds and the trial is considered a success. Scripts to recreate this task can be
found at the investigators Github: https://github.com/hooymana/Brooks_Task.git. B)
Example of a single successful trial and all corresponding outcome measures. The grey box
with a solid line represents the start box and the grey box with the dashed line represents
the target box within the virtual space of the task. The blue line represents the cursor
position throughout the trial traveling from the start box to target, measured on the right y
axis, with the dashed vertical line representing the end of the start chime (i.e. the point that
cursor can exit the start box). The red line is the acceleration of the cursor throughout the
trial measured on the left y axis. Here the cursor first experienced a positive acceleration to
begin propulsion through space and then a negative acceleration to bring it to a relative
stop within the target box. C) Example of experiment layout with a participant holding the
controller performing the discovery task. The cursor is only controlled by the right thumb
stick of the controller which is depicted in the inset image.
These instructions give information on the goal and timing constraints of the task but do
not provide any detail about how to accomplish the task goal (i.e. the task rule). This is the
unique feature of this task and what makes it a “discovery” task. In essence, the task goal can
only be accomplished through the execution of a specific motor sequence. The transformation of
48
joystick motion to cursor motion is determined through rate control—a critical feature that the
learner is naïve to at the start of practice. Motion of the joystick to the right applies a constant
force to the cursor in that direction and vice versa. The cursor, astronaut, travels in a frictionless
environment so the cursor will maintain a constant acceleration, and therefore accumulate greater
velocity, until the joystick comes back to center or has a force applied to it in the opposite
direction. Due to the time constraint within each practice trial the participant must discover the
precise timing and duration to apply a positive then negative force to the cursor for it to stay in
the TARGET box for one continuous second (Figure 3.3B). After task instruction is given the
participant is told they will not be given any verbal feedback, neither informational nor
motivational, by the researcher during practice. This encourages participants to use the inherent
task feedback (i.e., start and reward tones) and to explore the available joystick to cursor motion
space to achieve the task goal. Participants complete 200 practice trials on day 1 and then return
the next day for a retention/recall test.
On the retention day participants complete 50 trials of the discovery task but without the
start and reward tones. This tests how well each participant can effectively reproduce the
necessary movement pattern for task success without any augmented feedback. After completion
of the retention test, participants were given a brief survey questionnaire to obtain information
about previous experience with video games and/or physical activity. Participants were asked to
estimate how many total hours of video games and/or physical activity they have participated
during their lifetime. Once the participant completes the survey, they are debriefed about the
study and their participation is concluded.
49
Performance and Phase Quantification
Due to the tight task constraints, maximum possible acceleration of the cursor and set
time per trial, the participant must apply a temporally coordinated bout of positive and then
negative force to the cursor to achieve success. The positive/negative force application can be
calculated as a ratio, total negative force/total positive force, which accurately quantifies
execution of the task rule and individual trial performance. This metric is called the scaling ratio
and its calculation is based on its original use from the foundational research that first studied
discovery learning (Brooks et al., 1995). We found that we could improve the predictive
accuracy of the scaling ratio by multiplying the ratio by the total time the cursor spent outside the
start box during the trial. Maximum possible time the cursor could exist outside the start box was
three hundredths of a second. Here, a modified scaling ratio close to 300 would be interpreted as
accurate execution of the rule and highly predictive of success. Conversely, a scaling ratio at or
near 0 would be interpreted as poor rule execution and would be very unlikely to yield a
successful trial. Details of the scaling ratio are provided in Hooyman et al., In Preparation –
Chapter 2.
Quantification of exploration.
Through the process of binning trials into blocks of ten we can calculate the standard
deviation of modified scaling ratio in each block to determine the level of exploration (i.e. high
standard deviation represents high exploration and vice versa). We chose to bin the data by
progressive blocks of 10 because our previous research demonstrated: 1) robust group
differences in exploration and 2) the magnitude of early exploration was a strong predictor of
subsequent performance (Hooyman et al., In Preparation – Chapter 2). We identified the
exploration phase by first systematically calculating the prediction accuracy, R
2
, of every
50
possible trial bin of exploration (1 to X, with X increasing by 10 with each iteration) to all
possible future performance bins (10 to 250 – including retention trials). This allows us to
visualize how informative each bin of exploration is across both short and long time scales of
practice (Figure 3.4A). We then calculated the average R
2
for each exploration bin and identified
the peak of the curve to be the most meaningful exploration phase (blue curve in results Figure
3.4B).
Quantification of exploitation.
To identify the beginning of the exploitation phase we binned performance data using the
10-trial time in target (measured as time in TARGET box) mean and identified an intersection
point when time in target continued to increase while exploration had fallen below all previous
levels (results Figure 3.4B – red shaded area). We rationalized that the exploitation phase persists
till the end of Day 1 as this strategy is optimal for continued progress on the task after the task
rule is discovered and the start chime, and reward tone are still available.
Quantification of retention.
For the retention phase we averaged the time in TARGET box across day 2 by 10 trial
bins for each participant (results Figure 3.4B – green shaded area). Each of these phases,
exploration, exploitation and retention, are then used in the SVMr analysis described in the
Statistical Analysis section.
Intracortical Connectivity from Resting-state Electroencephalography
Individual rs-EEG data is first subjected to Independent Component Analysis (ICA) for
purposes of removing face muscle and eye artifacts which may contaminate the true EEG signal.
Prior to ICA the data is filtered at 1 and 60 Hz to remove head movement and device noise. The
ICA then segregates the EEG data into 10 components where the lead author visually inspects
51
each component, removing only those components that demonstrate features of eye/blink
movement and muscle activation (Delorme & Makeig, 2004). For each participant only a
maximum of 2 components are removed from each rs-EEG dataset. Each participant’s rs-EEG
had at least the eye component removed (mean group component removal = 1.3).
After ICA the artifact reduced rs-EEG data is zero-phase band passed filtered with cut off
frequencies of 10 to 40 Hz. Previous research has shown that the high beta band (20 – 32 Hz) is
the primary frequency of the motor system and predictive of motor skill acquisition and learning
(Deeny, Haufler, Saffer, & Hatfield, 2009; Krause, Meier, Dinkelbach, & Pollok, 2016;
Pfurtscheller, Stancák, & Neuper, 1996; Roopun et al., 2006; Wu et al., 2018, 2014). For this
reason and the purposes of this study we examined intracortical connectivity within the high beta
frequency band. Intracortical connectivity of each electrode pair is calculated using the mscohere
function in MATLAB (Version 2013a) using a one second Hanning window with no overlap.
This function outputs a coherence spectrum within the frequency range of interest. We record the
peak coherence value within the frequency band to signify the connectivity between the two
EEG signals. This peak value is representative of functional connectivity and lies between 0 and
1, similar to a Pearson correlation. A coherence value close to 0 means low connectivity and a
value close to 1 means high connectivity between electrodes. In total each participant’s rs-EEG
dataset produces 1830 individual coherence values, representing all possible electrode pairs
among the 61 recording electrodes.
Statistical Analysis
Individual rs-EEG coherence arrays are concatenated to create a single intracortical
connectivity matrix (32 x 1830). We chose Support Vector Machine regression (SVMr) analysis,
in part, because it had recently been used successfully to identify key resting-state brain
52
measures predictive of clinical outcomes in chronic pain (Kutch et al., 2017). The independent
variable in the SVMr are the participants rs-EEG coherence data collected prior to practice to
predict the three phases of motor learning (i.e. exploration, exploitation and retention).
To confirm the accuracy of the SVMr results we performed leave-one-out cross
validation (LOOCV). The LOOCV results are regressed against the true results identified from
each behavioral phase to determine overall prediction strength. To determine if the mean square
error (MSE) of LOOCV is significant we compared this error versus a sample of random MSE
created through 1000 bootstrap models generated with resampling. If the LOOCV error is
significantly lower than the bootstrap models (alpha = .05) we consider the true SVMr model
predictions to be accurate and significant.
The SVM analysis also gives each coherence measure an individual beta weight, an
indicator of its relative relationship to the predicted behavioral phase. We compared the beta
weights from our SVMr model to the beta weights from the 1000 bootstrap models. This allowed
us to identify which coherence measures were significantly predictive of exploration,
exploitation and retention compared to models with random weights, alpha = .05. To correct for
multiple comparisons, we used a false discovery rate correction to create a correction adjusted p
value which could then be interpreted as being meaningfully predictive (Storey, 2002).
Once the corrected intracortical connectivity is established for each behavioral phase we
executed separate regression models for each phase which included the composite score of the
participant responses from the previous physical and video experience survey. This was to
determine what interaction, if any, existed between previous life experience (hours playing video
games and physical activity), rs-EEG and current results of each learning phase.
53
With the intracortical connectivity measures determined from each phase, after multiple
comparison correction we performed a multivariate regression using the individual connectivity
measures combined with the pervious history survey results for each phase. To account for the
possibility of variance inflation due to correlation between connectivity measures, we removed
highly correlated features until the variance inflation factor for each independent variable in the
analysis was below 4. The remaining variables of intracortical connectivity and previous
experience results that corresponded with each phase were then regressed against each phase to
determine overall predictability given intracortical connectivity and previous experience.
Results
Exploration Phase--Quantification
The 10-trial bin analysis revealed the optimal exploration phase to be over the first 40
practice trials (Figure 3.4B, shaded blue area). We determined that the most informative phase of
exploration (i.e. trials 1 to 40), signified as a peak in average R
2
value for all possible exploration
bins. This time appears to be critical for future performance and learning of the task as the first
40 trials is more predictive of early and late performance than any other point in practice (Figure
3.4A). We chose to not include exploration between bins 40 and 110, white area between
exploration and exploitation (Figure 3.4B), because although it is possible exploration is still
ongoing, it is not as informative as the time between trials 1 and 40 for future success on the
task.
However, we also used the rs-EEG data to predict the potential exploration period
between trials 1 and 80 and found overall LOOCV prediction results of the SVMr to be reduced
compared with that for trials 1 and 40, R
2
= .31, although the overall network connectivity
remains largely unchanged [not shown]. Based on these results we believe this to be
54
confirmatory evidence that our defined exploration phase between trials 1 and 40 is an accurate
indicator of group level behavior for SVMr compared to an alternative exploration phase range
(Figure 3.4).
55
Figure 3.4: A) Binned trials of exploration in progressive blocks of 10 predicting
subsequent blocks of Performance (Time in Target – hundredths of a second). The redder
the color the higher the accuracy of the exploration bin to predict Time in Target during a
specific block of practice. Bin of exploration between trials 1 and 40 (Peak Exploration) has
the highest average predictive accuracy across all blocks of performance. Bin of
exploration between trials 1 and 110 (End of Exploration) shows signs of reduced accuracy
to predict future bins of performance. Reduced prediction accuracy continues after trial
110 indicating that this is the point that exploration ceases on average for the study sample.
B) Average prediction accuracy across all bins of exploration trials, which is determined in
Figure 3.4A, was used to determine phases of Exploration and Exploitation. The
Exploration phase is blocked in blue. Exploitation was determined as the point during
practice when exploration accuracy fell below all previous levels, yet Time in Target
continued to improve. We indicated the exploitation phase as the trials from 110 to 200
shaded in red. The Retention phase was the average time in Target during all trials
practiced on day 2, shaded in green.
Exploitation Phase-Quantification
To determine the transition from the exploration phase to the exploitation phase within
the group we aimed to identify a point where exploration had declined, blue line in Figure 3.4B,
yet average time in target, red line in Figure 3.4B, continued to rise. Thus, the driver of future
performance gains was not based on task rule identification through exploration but rather
through a strategy of error reduction to continue to improve performance (i.e. increased mean
time in target). The exploration bin between trials 1 and 110 had the lowest predictive quality of
all previous exploration bins, mean exploration R
2
, Figure 3.4B – blue line. This indicated to us
56
that participants were no longer exploring the task space during this time. Therefore, we
reasoned, using this conservative criterion, that the start of the exploitation phase began at or
near trial 110 and ended at or near trial 200 (Figure 3.4B, red shaded area). We reasoned the
exploitation phase would persist throughout day 1 as the participant would continuously utilize
the start chime and/or reward tone as augmented feedback to optimize their performance.
Retention Phase-Quantification
The retention phase was identified as the average time in target for all trials during Day 2
where no augmented sound feedback was provided (Trials 1 to 50). Participants maintained (if
not improved) the end of practice performance during the no feedback recall test. This was an
indication that a relatively permanent capability to perform the complex task had been acquired
through practice on Day 1. (Figure 3.4B – green shaded area).
Resting-State EEG Prediction of Individual Levels of Exploration
Individual rs-EEG intracortical connectivity was used to predict individual levels of
exploration recorded during the above identified exploration phase as the standard deviation of
modified scaling ratio between trials 1 and 40. Regression of the LOOCV results, predicted
exploration, against actual exploration returned an R-squared = 0.42 (Figure 3.5A). This
demonstrates that intracortical connectivity explains 42% percent of the variance among
individual levels of exploration. After correcting for multiple comparisons, the SVMr analysis
identified a bilateral network of frontal, premotor and motor areas that predict exploration
beyond chance (corrected p-value < .05) (Figure 3.5B).
57
Figure 3.5: A) Comparison of observed exploration values on Y axis versus predicted
exploration values on X axis. If prediction was perfect all points would fall long the blue 1:1
line. Black line represents linear regression line between predicted and observed values. B.
Intracortical connectivity measures that are responsible for predicted exploration value
after multiple comparisons. Predictive Intracortical Connections: Fpz-FC5, Fpz-AF7, Fpz-
F7, Fpz-F5, Fpz-F3, Fpz-FC3, Fpz-C3, F7-C3, F7-C1, F7-F3, AF7-C3, Fpz-FC6, Fpz-FC4,
Fpz-C6, Fpz-C4, Fp2-C6, AF8-C6, AF8-C4, AF8-FC4, CP4-PO4, P4-PO4. Color of line
between electrode pairs signifies the relationship of the connection to predicted behavior in
the SVM.
Resting-State EEG Prediction of Individual Levels of Exploitation
The exploitation phase, average time in target for each participant between trials 110 and
200 on day 1, was predicted using individual rs-EEG intracortical connectivity. Variance
explained was determined through regression of the observed individual exploitation phase
against the predicted exploitation phase resulting from the LOOCV. The regression analysis was
significant with an R
2
of 0.47 demonstrating that the rs-EEG data explains 47% of the variance
of the observed exploitation phase (Figure 3.6A). To determine the intracortical connectivity
58
responsible for this prediction multiple comparisons were applied to the total of significant
bootstrap samples. The resulting network was within the right hemisphere with distinct
connections between the frontal and motor cortex (AF8 to C6), and within the parietal cortex (P6
to P06, CP4 to P04) (Figure 3.6B).
Figure 3.6: A. Comparison of predicted exploitation values, time in TARGET box from
trials 120 to 200 (milliseconds), against observed exploitation values. Black line represents
linear regression line between predicted and observed values. B. Three measures of
intracortical connectivity after multiple comparisons that significantly predict individual
exploitation. The three measures are ipsilateral to the performing hand and extend from
frontal to parietal regions of the cortex. Predictive Intracortical Connections: AF8-C6,
CP4-PO4, P6-PO6. Color of line between electrode pairs signifies the relationship of
connection to the predicted behavior in the SVM.
Resting-State EEG Prediction of Individual Levels of Retention
The retention phase metric of performance was the average time in target over the 50
trials on Day 2. The rs-EEG intracortical connectivity recorded prior to Day 1 practice was used
to predict individual retention scores. The LOOCV regression of predicted retention versus
59
observed retention was significant with a R
2
value of 0.26 (Figure 3.7A). After bootstrap analysis
and correcting for multiple comparisons the resulting network of predictive intracortical
connectivity consisted of a bilateral network extending throughout the frontal, motor and parietal
cortices. Primarily there appeared to be a distinct network between the left prefrontal (AF7), left
motor (C5, CP5) and right parietal and occipital areas (P2, P6, O2, PO8) (Figure 3.7B).
60
Figure 3.7: A. Comparison of predicted mean time in TARGET box during the retention
day versus actual time in TARGET box (milliseconds). Black line represents linear
regression line between predicted and observed values. B. Intracortical connectivity
network responsible for significant prediction of retention phase extends throughout the
cortex with prevalent connections arising from left prefrontal cortex and extending to left
motor and right occipital/parietal. Predictive Intracortical Connections: Fpz-FT8, F7-CP5,
AF7-CP5, AF7-C5, C5-F1, AF7-P2, AF7-PO6, AF7-O2, P6-PO6, P6-PO4, O2-P8, AF8-C6,
T7-TP7. Color of line between electrode pairs signifies the relationship of connection to the
predicted behavior in the SVM.
Improved Fit from Previous History and Intracortical Connectivity
After correction for multiple comparisons and the removal of highly correlated
intracortical variables (i.e., VIF > 4), results of the multivariate regression analysis now
including previous experience was: Exploration (21 intracortical connections in SVM, Figure
3.5, to 10 intracortical connections VIF corrected): without previous history, adj R
2
= 0.57 p <.05
vs with previous history, p<.05, adj R
2
= 0.64, Exploitation (3 intracortical connections, no
change between SVM, Figure 3.6, and VIF corrected): without previous history, adj R
2
= 0.48 p
<.05 vs with previous history, adj R
2
= 0.58, p <.05 and Retention (13 intracortical connections
in SVM, Figure 3.7, to 9 intracortical connections VIF corrected): without previous history, adj
R
2
= 0.73 p <.05 vs with previous history, adj R
2
= 0.82, p <.05. Comparison of each model with
previous history against the model without previous history using analysis of variance
demonstrated that each model with previous history was better associated with their respective
learning phase, p <.05. Overall, these results indicate that the addition of previous history has an
additive effect on explanatory power when coupled with resting-state intracortical connectivity.
61
Table 3.1: R
2
values for LOOCV, VIF Correction without Experience and VIF Correction
with Experience for each Behavioral Phase.
R
2 LOOCV VIF no Experience VIF with
Experience
Exploration 0.42 0.57 0.64
Exploitation 0.46 0.48 0.58
Retention 0.26 0.73 0.82
Discussion
Results of this study indicate that distinct phases of motor learning that uses a discovery
task can be predicted from rs-EEG (LOOCV R
2
: Exploration = .42, Exploitation = .46, Retention
= .26). Additionally, when the identified rs-EEG measures are combined with self-reported
experience with video games and physical activity then overall prediction accuracy of behavioral
phase is boosted significantly (intracortical connectivity only R
2
vs. intracortical connectivity +
experience R
2
): exploration: 0.57 vs 0.64 = +7%, exploitation: 0.48 vs 0.58 = +10%, retention:
0.73 vs 0.82 = +9%. In support of our hypothesis each phase of motor learning, exploration,
exploitation and retention, are predicted by distinct pairs of functional connectivity that span
throughout motor and non-motor regions of the cortex. This result is informative on how
differential networks of connectivity distributed throughout specific cortical areas of the brain
mediate distinct behavioral aspects of motor learning, particularly when the task rule is not
provided. To the best of our knowledge there has been no previously published research
identifying these unique networks predictive of distinct phases of learning a novel discovery
task.
62
Here, we used a novel motor task that requires participants to actively explore the task
space to be successful. The process of exploration potentially engages learning processes that
might be expected from any beginning learner (e.g., budding athlete to recovering stroke victim)
seeking to solve the movement problem. Exploration capability has been shown to be a
predictive behavior for motor learning and transfer in non-disabled and disabled populations
(Pacheco & Newell, 2015; Singh et al., 2016; Therrien, Wolpert, & Bastian, 2016). We also used
a rigorous whole brain analysis approach that diverged from the previous seed-based analysis
approach. With the inclusion of connectivity measured throughout the entire cortex, we have no
prior assumptions going in about which specific neural substrates mediate the motor learning
process. Overall our results support our hypothesis that high beta band intracortical connectivity
recorded during the resting state can be used to predict distinct phases (i.e. exploration,
exploitation and retention) of a complex motor learning task. Additionally, when combined with
self-reported measures of previous relevant experience increased variance explained with each
learning phase between 7 and 10%.
Cortical Connectivity and the Exploration Phase
The rs-EEG network results that best predicted performance during the exploration phase
comprised a bilateral frontal network primarily consisting of electrodes associated with the
ventro-medial prefrontal cortex (VMF), the dorsal lateral prefrontal cortex (DLPFC) and
premotor cortex (PMC). Previous research found this functional network also has structural
underpinnings as well as additional connections to the striatum (Jarbo & Verstynen, 2015).
Specifically, the VMF has been associated with decision making during the exploration process
(Blanchard & Gershman, 2018; Guillot et al., 2008). Our previous research focused on individual
strategies that dictated success on this task revealed the importance of exploration and those
63
participants who did not explore failed to achieve task success (Hooyman et al., In Preparation –
Chapter 2). This suggests that individuals with weaker connectivity in this putative frontal
network did not explore to a level that was required for goal achievement.
Cortical Connectivity and the Exploitation Phase
Exploitation as defined here was the point during practice when exploration had fallen
below all previous levels, yet performance continued to improve (Figure 3.4B). Using a whole
brain analysis approach, we found a network of three significant intracortical connections within
the right hemisphere, specifically between the frontal and motor cortex and the parietal cortex.
All individuals of this study were right hand dominant so we cannot confirm if the right
hemisphere network identified here is specific to the right hemisphere or specific to the hand
performing the task (Serrien, Ivry, & Swinnen, 2006). Recent research provides beginning
evidence that points to the importance of the right motor cortex in motor learning and control
(Kobayashi, Hutchinson, Theoret, Schlaug, & Pascual-Leone, 2004; Pruitt et al., 2016; Waters,
Wiestler, & Diedrichsen, 2017). Specifically, research by Wischnewski and colleagues
demonstrated that task accuracy demand modulated the level of inhibition from the right primary
motor cortex to left primary motor cortex (Wischnewski et al., 2016). Although our results do
not demonstrate bilateral connectivity between the primary motor cortex associated with time in
target during the exploitation phase, the task does require a high level of movement timing
accuracy for which the connectivity between ipsilateral frontal and motor cortices may be
important. This together with our previous work (Hooyman et al., In Preparation – Chapter 2)
provides that individuals with weaker connectivity among all three of these intracortical circuits
will be more likely to exhibit poorer performance (i.e. lower time in target) during the
exploitation phase.
64
Cortical Connectivity and the Retention Phase
Prediction of retention performance as average time in target for Day 2, was primarily
predicted by connections between the contralateral left prefrontal cortex, electrode AF7 and the
ipsilateral parietal/occipital cortex, electrodes P2, O2, PO8. This electrode broadly corresponds
to the DLPFC and it appears to be a hub sharing connections with the contralateral motor cortex
and the ipsilateral parietal cortex. This result is consistent with previous motor learning studies
looking at the contextual interference effect (Lin et al., 2013). Resting state connectivity between
the frontal and parietal cortices has been referred to as the “executive-control network” (Seeley
et al., 2007). Strong connectivity within this network may indicate engagement of processes,
such as reliance upon an internal reference of correctness that allows accurate recall of the
specific motor sequence without reference to the augmented feedback provided during Day 1
practice (Kantak & Winstein, 2012).
Inclusion of Previous Experience in the Model
Prediction of individual exploration, exploitation and retention is significantly improved
when previous experience, a self-reported measure of hours engaged in video games and
physical activity across the lifetime, is included in the final correlation analysis. In essence, the
overall fit of the specific network that predicted each learning phase improved by a non-
negligible degree. Inclusion of this simple survey measure demonstrates how, at the very least,
perception of, if not actual life experience with motor skills (video or actual) can contribute to
our understanding of the process of motor learning. We agree with others who have shown that it
is not the brain alone that can fully explain the motor processes we hope to understand (R.
Seidler et al., 2015). For the adult learner, we must also understand the individual and how the
context of their environment and life experience impacts their current behavior.
65
Limitations
This research is limited to intracortical connectivity measured at the cortex. Therefore,
connectivity arising from subcortical and cerebellar brain regions may also contribute to
prediction, but our chosen methodology limited us to cortical substrates. Additionally,
connectivity between cortical brain areas may be due to a tertiary source such as the basal
ganglia that can only be inferred but not directly measured. Our methodology and analysis
approach are unable to confirm if indeed specific intracortical connectivity is truly responsible
for prediction of motor learning processes. Furthermore, we cannot confirm the direction of
information flow between cortical regions nor if the connectivity between regions is excitatory or
inhibitory in nature. Our survey of previous relevant experience is a relatively gross measure of
individual experience with video games and physical activity. We cannot confirm if indeed these
numbers are accurate or if they are confounded by the participant’s perception or ability on the
task.
Behavioral Sub-groups
Previous work in our lab examined the individual learning strategies of our study sample
revealing three distinct sub-groups of learners, each demonstrating a different performance
strategy during task practice and retention (Hooyman et al., In Preparation – Chapter 2). One
could argue that by averaging performance across these sub-groups we overly simplify the story
and end up with a prediction that misrepresents the true underlying behavior. However, our
previous work demonstrated that these sub-groups were well aligned along a performance
continuum that persisted throughout practice and retention phases. Therefore, prediction of
average individual performance, as we have done here, is still a valid approach to capture true
differences in behavior between individuals. Arguably, it is more informative to predict based on
66
individual performance compared to sub-groups as it allows for a more granular investigation of
the distinct behavioral stages identified here.
Future Work
This work may inspire future research to better understand how these brain areas affect
motor learning behavior and how that knowledge can be translated for the betterment of
neurorehabilitation. Current work in our lab is focused on modulating resting-state connectivity
within specific neural circuits. If one goal of neurorehabilitation is to improve function through
brain repair associated with motor learning, then accurate modulation of intracortical
connectivity will be necessary. Current neuromodulatory methods, repetitive Transcranial
Magnetic Stimulation and Transcranial Direct Current Stimulation, are limited to the modulation
of excitatory threshold of a single cortical region. As more research provides evidence that the
brain operates through a series of functionally connected networks, there will be an urgent need
for neuromodulatory methods aimed to directly increase specific networks of connectivity
between cortical regions (Buckner, Andrews-Hanna, & Schacter, 2008; van den Heuvel, Mandl,
Kahn, & Hulshoff Pol, 2009).
Conclusion
To the best of our knowledge this is the first study to demonstrate differential resting-
state connectivity networks that predict the different motor learning phases of a novel complex
motor task. Although this work did not consider subcortical brain areas which most certainly
play a role in the motor learning process, it does provide beginning evidence for a complex
connectivity network including motor and non-motor areas that mediate the process of motor
learning. Inclusion of self-reported motor experience improved the prediction and provided
additional evidence of the interaction between brain and behavior, both current and prior, that
67
most certainly impacts motor learning capability. Future work should focus on modulation of
intracortical connectivity to better understand the brain/behavior relationship that governs more
ecologically relevant motor learning processes.
68
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity
Abstract
Recent neuroimaging research has demonstrated that resting-state intracortical connectivity, i.e.
the shared communication between two brain regions, can serve as a robust predictor of motor
performance and learning. Theoretically, direct modulation of resting-state intracortical
connectivity within the motor system could then improve motor performance and learning.
However, previous neuromodulation techniques such as repetitive Transcranial Magnetic
Stimulation may be limited in the capacity to modulate targeted intracortical connectivity. Paired
Associative Stimulation (PAS) has shown efficacy in facilitating connectivity primarily between
the central and peripheral nervous system based on the neuroplasticity mechanism of Spike
Timing Dependent Plasticity (STDP). It may therefore be plausible for a reconfigured
corticocortical PAS paradigm to modulate resting-state intracortical connectivity using a dual
stimulator methodology over specific cortical nodes. However, potential theoretical and
technological considerations of such a paradigm first need to be addressed prior to application
for the purposes of manipulating motor behavior. We posit a corticocortical PAS paradigm used
in conjunction with resting-state electroencephalography (rs-EEG) to demonstrate efficacy of
potentiating motor learning associated resting-state intracortical connectivity within the human
brain. Here we provide a precise PAS/EEG experimental design, details on data analysis,
recommendations for maintaining scientific rigor, and preliminary proof of principle within a
single-subject.
69
Introduction
Modulation of Intracortical Connectivity
Current non-invasive brain stimulation (NIBS) paradigms, rTMS and tDCS, focus on
changing behaviors through up or down regulation of a single cortical region which is confirmed
by changes in Motor Evoked Potentials (MEPs) (Klomjai, Katz, & Lackmy-Vallee, 2015;
Rossini et al., 2015). However, instead of single brain regions, evidence from recent
neuroimaging research demonstrates that complex motor behavior is mediated by interconnected
networks of brain activity (Carter, Shulman, & Corbetta, 2012; Hardwick et al., 2015).
Therefore, intracortical connectivity may be better than MEPs for assessing meaningful change
in the motor system if the goal is to assess the brain-behavior relationship involved in motor
learning. Hypothetically, if communication within a specific intracortical circuit could be
strengthened in a reliable and persistent manner, one could potentially treat dysfunctional brain
circuits and thereby improve motor performance and learning. Additionally, such a dual site
modulatory paradigm could also be used in a top-down approach to investigate neural substrates
that mediate motor learning (Hordacre, Rogasch, & Goldsworthy, 2016).
The aim of this paper is to describe a neuromodulation paradigm, based on a strong
scientific premise, that can be used to modulate a specific brain circuit associated with motor
learning. Recent research with advanced EEG technology has already shown promise with a
combined PAS/EEG approach by demonstrating stimulation-induced changes in brain activity
(Casula, Pellicciari, Picazio, Caltagirone, & Koch, 2016; Rajji et al., 2011; Veniero, Ponzo, &
Koch, 2013). We describe the combined PAS/EEG paradigm in the context of prior research
which has determined that resting-state intracortical connectivity between the dorsolateral
prefrontal cortex (DLPFC) and primary motor cortex (M1) is predictive of motor learning
70
capability (Wu et al., 2014). Previous corticocortical PAS experiments have not specifically
attempted to modulate resting-state intracortical connectivity. As such, this novel brain-behavior
paradigm is designed to directly measure the effects of PAS on this specific resting-state cortical
motor circuit measured via electroencephalography. Theoretically then, successful modulation of
M1-DLPFC resting-state circuitry could lead to major changes in motor behavior and learning.
Further, we provide recommendations for maintaining scientific rigor that are meant to
generalize beyond the single human subject presented here.
Resting-state Intracortical Connectivity and its Translational Role in the Field of Motor
Behavior
The ability to predict an individual’s motor performance and learning can give
researchers and ultimately coaches and clinicians the opportunity to intervene early for those that
need it most. Information within the resting brain has been shown to predict future motor
performance in non-disabled individuals and recovery in those who have sustained a disabling
stroke (Wu et al., 2015, 2016; Wu, Knapp, Cramer, & Srinivasan, 2018; Wu, Srinivasan, Kaur,
& Cramer, 2014). Resting-state intracortical connectivity is a measure of the relative strength of
the communication between two distinct brain areas while a person is at wakeful rest (i.e. brain
activity is measured while a person is awake but not focused on any specific task) (Srinivasan,
Winter, Ding, & Nunez, 2007). Specifically, intracortical connectivity of the resting brain
measured by electroencephalography (rs-EEG) has been found to be a very robust predictor of an
individual’s acquisition of a motor skill, R
2
> 0.9. (Wu et al., 2014; Wu et al., 2016; Wu et al.,
2018). One potential intervention to improve future motor performance/learning is to increase or
decrease the strength of the predictive rs-EEG circuit through neuromodulation (Ros &
Gruzelier, 2011). However, a non-invasive neuromodulatory intervention capable of
71
strengthening connectivity between two cortical areas would require different methodology than
the established forms of brain stimulation (e.g. repetitive Transcranial Magnetic Stimulation
(rTMS) and Transcranial Direct Current Stimulation (tDCS)).
Strong Scientific Premise for Paired Associative Stimulation
A stimulation paradigm capable of facilitating or inhibiting neural communication has
been in existence since 2000. It is called Paired Associative Stimulation (PAS) (Stefan, Kunesch,
Cohen, Benecke, & Classen, 2000). Originally performed between the peripheral and central
nervous system, PAS uses a temporally coordinated delivery of electrical impulses to increase or
decrease cortical excitability (e.g. increase MEP amplitude). The observed increase in
connectivity along a circuit is achieved through a Hebbian mechanism of plasticity known as
Spike Timing Dependent Plasticity (STDP) (Bi & Poo, 1998). In vivo studies of associative
stimulation within the brain determined that the precise interstimulus interval (ISI) between
paired pulses is what modulates the connectivity strength on the targeted neural circuit (Huang,
Pittenger, & Kandel, 2004). To strengthen connectivity between targeted brain regions A and B,
the pre-synaptic pulse delivered at brain area A, must arrive at brain area B immediately before
the post-synaptic pulse delivered to brain area B. An accumulation of paired pulses will
modulate the functional plasticity between the two brain areas resulting in a higher likelihood
that brain area B will depolarize because of brain area A’s depolarization (Long Term
Potentiation – LTP). Conversely, if the post-synaptic pulse is delivered prior to the firing of the
pre-synaptic pulse then there is a decreased likelihood that brain area B will depolarize because
of brain area A’s depolarization (Long Term Depression – LTD). Interestingly, the magnitude of
the LTP or LTD between the brain areas falls along a continuum dependent upon the
interstimulus interval (i.e. a longer ISI leads to decreased LTP or LTD like effects – Figure 4.1).
72
Previously, measurement of the PAS effect in humans was only possible between the peripheral
and central nervous system, however, advances in EEG technologies, (e.g. TMS tolerant
electrodes) now make it feasible to apply PAS and measure its effect on resting brain
intracortical connectivity.
73
Figure 4.1: The change in magnitude of Long Term Potentiation and Long Term
Depression between hypothetical brain areas A and B as a function of interstimulus
interval. If the post-synaptic pulse, pulse delivered to brain area B, occurs before the pre-
synaptic, pulse delivered to brain area A, then long term depression, decreased
connectivity, red curve, will result. Conversely, if the pre-synaptic pulse occurs prior to the
post-synaptic, then long term potentiation, increased connectivity, blue curve, will result.
The closer the coupling of the two pulses, the closer they fire around an interstimulus of 0
ms, then the greater the overall Long Term Potentiation or Long Term Depression.
However, a change in connectivity will still persist, although at a lower magnitude, with a
longer interstimulus interval.
How the PAS/EEG Paradigm can Benefit Motor Learning Research
We propose that the PAS/EEG paradigm is optimal for motor learning research for the
following reasons: 1) EEG can be used to acquire data at a high temporal resolution which is
required to not only capture PAS effects but benefits many motor learning paradigms in which
task execution occurs at a speed that functional Magnetic Resonance Imaging would not be able
to capture (Mary et al., 2017; Muraskin et al., 2016). 2) EEG is a relatively easy and cost-
effective neuroimaging modality, after initial investment has been made. 3) EEG is more
maneuverable than other neuroimaging modalities and can therefore be easily applied to research
paradigms that engage complex motor skills. However, prior to application of PAS/EEG to
manipulate motor behavior an important first step will be to outline a rigorous methodology to
first demonstrate that PAS can potentiate target resting-state intracortical connectivity. The
purpose of this paper is to provide important and thorough details on the PAS/EEG methodology
74
outlined below, that can be used to inform future motor learning researchers who are interested
in the application of neuromodulation.
Method
To maintain scientific rigor, a combined PAS/EEG experiment must implement several
important components to control for prevalent behavioral and technological confounds.
Control for Variations in Brain State between Repeated Measures
First, we suggest collection of at least two baseline rs-EEG measures to account for
inherent changes in rs-EEG for each participant and to establish reliability of our primary
outcome measure. This allows for better comparison of what changes occur due to PAS
compared to changes due to normal fluctuations in rs-EEG. Second, to maintain consistent
recording between multiple rs-EEG recordings and PAS conditions, it is important for the
participant to maintain a consistent brain state throughout the experiment. Intracortical
connectivity measured during rs-EEG assumes a default brain state (i.e. mind wandering with no
specific focus). It is unknown how an inconsistent brain state may alter the effect of PAS on the
brain between separate conditions. Therefore, to achieve a consistent brain state across different
PAS conditions, participants are given a simple counting task and fixation cross to sustain their
focus.
Signal to Noise Problems Inherent to the Technology
Second, we advocate conservative EEG processing methods to identify changes in rs-
EEG that rise above a specific statistical threshold. Rs-EEG, like all neuroimaging modalities,
has a certain level of inherent noise which may interfere with accurate signal measurement
resulting in a false negative or false positive change. By applying a statistical constraint on rs-
EEG there is less risk of incorrectly inferring changes in connectivity that may be a result of
75
noise. We discuss this statistical procedure in detail within the Data Processing and Analysis
section.
Informed Consent and Safety Screening
After informed consent is obtained from prospective participants, a TMS safety screen to
confirm no health risk to brain stimulation is completed. Additionally, to ensure the safety and
comfort of each participant we suggest adherence to published recommended safety standards
(Farzan et al., 2016; Rossi, Hallett, Rossini, & Pascual-Leone, 2009; Rossini et al., 2015).
Inclusion/Exclusion Criteria – Unique Considerations for NIBS Research in Motor
Behavior
Thus far in non-invasive neuromodulation research, there is high variability when it
comes to responsiveness to modulation and its effect on behavior (M. Hamada, Murase, Hasan,
Balaratnam, & Rothwell, 2013; Masashi Hamada et al., 2014a; López-Alonso, Cheeran, &
Fernández-del-Olmo; Rajji et al., 2013). Some researchers have suggested employing pre-
screening measures to ensure that recruited participants meet a specified criterion that would
yield a high likelihood of responsiveness. Below we detail these predetermined criteria.
1) No consumption of nicotine, caffeine, alcohol, or exercise participation prior to
experimentation. Each of these activities can have direct and meaningful effects on brain
chemistry. Therefore, to control for these effects one should ask participants to abstain until after
the experiment.
2) Data collection to occur at the same time of day for each participant. It is well documented
that diurnal changes in an individual's circadian rhythm has effects on hormone release related to
arousal (Cortisol, HPA Axis, circadian rhythm) (de Weerth, Zijl, & Buitelaar, 2003). Therefore,
one participant’s brain may be more plastic in the morning compared to another’s who is tested
76
in the afternoon with responsiveness varying as a function of the time of day the experiment was
conducted.
3) Ensuring that all participants have the same hand dominance if the targeted brain circuit has
only been found in a cohort with the same handedness. Future studies that wish to explore the
effects of intracortical PAS on motor behavior should avoid recruiting participants of different
handedness as application of PAS may have differential effects on the underlying behavior due
to group sample heterogeneity. For our experiment we recruit participants who are right hand
dominant as previous research in our lab has only identified the predictive cortical circuit,
DLPFC and M1, in right hand individuals.
Safety Precautions
As an extra precaution for each participant’s safety a Stimulation Comfort Survey (see
supplement for details) is completed by the participant before and after all bouts of brain
stimulation. This is to acknowledge and attend to any unintended changes in the participant’s
comfort, be it physical or mental. This also provides the investigator with a written record of the
participants’ physical and mental status throughout the experiment.
77
Steps to Maintaining Scientific Rigor for Implementation of the PAS/EEG Study Design
Illustrated in Figure 4.2
Figure 4.2: A complete diagram of all calibration procedures (Pre-measurement
Procedures) and measurement/PAS procedures (Condition 1 and Condition 2).
Pre-measurement procedures.
Prior to any EEG collection, the participant will wear the EEG cap from the start of the
experiment. This ensures the distance between the TMS coil and scalp is consistent throughout
each epoch of data acquisition. The strength of a TMS pulse can be attenuated by small changes
in the distance between the TMS coil and the scalp. Therefore, to maintain consistent space
between the scalp and the TMS coil the EEG cap is applied at the very onset of the experiment.
Baseline.
We employ a double baseline procedure prior to the application of PAS to account for
normal fluctuations in cortical interconnectivity. As discussed above, one should not assume that
that rs-EEG is stable over time given the high noise associated with EEG.
78
PAS condition ordering: rationale for within subject design.
This experiment utilizes a within subject study design to account for individual
differences in responsiveness for each condition (no response to real PAS, response to sham
PAS, etc.) between subjects. To account for any order effect between the real and sham
conditions we recommend counterbalancing the condition presentation (real then sham versus
sham then real) between participants. Therefore, an equal number of participants will receive the
sham and real conditions first.
Accounting for after effects between PAS conditions.
To attenuate any after effects of the first PAS condition on the second PAS condition,
(Figure 4.2) we recommend each PAS procedure be performed on separate days, at least 24
hours apart. Although longer time periods between conditions, several days to a week, may be
necessary if PAS procedures of greater stimulation intensity or total pulse number are utilized.
The time between and the counterbalancing of PAS conditions will account for any residual after
effects on intracortical connectivity as a result of PAS. However, conditions performed across
many days may introduce additional confounds related to individual differences in sleep,
exercise, stress, nutrition, etc. between PAS conditions. Therefore, participants will have to be
given strict instruction to maintain their current lifestyle with special restrictions on things such
as caffeine consumption and exercise participation prior to the second PAS procedure of the
study protocol (Farzan et al., 2016). Given the experience dependent nature of the resting brain
and the future application of PAS aimed at modifying motor learning it will also be important for
participants to not practice any task involved in future experiments (Lewis, Baldassarre,
Committeri, Romani, & Corbetta, 2009b)
79
Pre-measurement Procedures
EEG cap fitting and orientation.
First the participant is fitted with a 64 electrode EEG cap (ANT Neuro, Enschede,
Netherlands) based on their head circumference. When cap size has been determined the CPZ
electrode is aligned with the vertex of head. Vertex is determined for everyone using
anterior/posterior (i.e. the nasion and inion) and medial/lateral (i.e., tragus of each ear) skull
markers. After the EEG cap is fitted one may proceed with determination of the participant’s
resting motor threshold (RMT).
Resting motor threshold procedure/determination of PAS intensity.
Resting motor threshold is the process of identifying the minimum stimulator output that
can achieve a 50 μV peak to peak Motor Evoked Potential (MEP) amplitude in at least 5 out of
10 consecutive trials in the desired muscle using Transcranial Magnetic Stimulation (Hallett,
2007) (Magstim 200, Morrisville, NC). An MEP is a measurement of motor cortical excitability
that results from a TMS pulse delivered to a specific muscle representational area within the
primary motor cortex. This procedure is common place in TMS research and is utilized here as a
method for identifying the stimulation intensity at which to deliver PAS. Determination of RMT
is a simple way of confirming a stimulation intensity capable of depolarizing cortical neurons
associated with a muscle used during a motor task.
For this experiment the target muscle is the abductor pollicis brevis (APB) of the right
hand. The APB served as the prime mover of a motor learning task during which predictive rs-
EEG ROIs, right prefrontal to right motor, had previously been determined (Hooyman, Babikian,
Kutch, & Winstein, 2017). A surface EMG electrode is secured to the skin over the right APB
muscle belly. Future research that uses PAS/EEG to modulate ROIs governing different motor
80
tasks should determine RMT for muscles specific to those motor tasks. For example, postural
control research focused on anterior/posterior sway about the ankle may want to use RMT of the
Tibialis Anterior to determine stimulation intensity of the PAS procedure.
Since PAS is applied through the EEG cap the participant should wear the cap during the
RMT procedure. This is to account for the additional distance the cap and EEG electrodes creates
between the scalp and the TMS coil. Our pilot research has found that stimulating a muscle at
RMT determined without the EEG cap is insufficient to produce consistent MEPs once the cap is
applied. Without accounting for the EEG cap distance any previously determined RMT without
the cap may have little to no effect on the targeted cortical area. Therefore, we encourage
researchers to have the cap worn by participants prior to the RMT procedure and continue to
wear it throughout the experiment. Once RMT has been determined the final stimulator intensity
should be recorded as a baseline for the PAS paradigm. Then begins the first resting-state EEG
collection with calibration of the EEG electrodes.
EEG electrode calibration.
Calibration of the EEG requires reducing impedance of each electrode to the desired level
of 20 kOhms. We use a 64 electrode EEG system that is compatible with TMS (ANT Neuro,
Enschede, Netherlands) although larger electrode arrays have also been using in rs-EEG (We et
al., 2014 & Wu et al., 2018). The impedance of each electrode is reduced through the injection of
a saline gel. The saline gel is hypoallergenic and does not contain any chemicals known to cause
scalp irritation or hair discoloration. This EEG system uses one electrode of the cap as a ground
(Figure 4.3A – blue electrode) and one as a reference electrode (Figure 4.3A – green electrode)
to reduce noise and generate the EEG signal. The reference electrode is placed far from the area
of stimulation to ensure that noise from the TMS does not interfere with the reference. The
81
reference electrode for this experiment is CPZ. With a low level of impedance of the EEG
electrodes achieved and the reference set, the first resting-state collection begins.
Figure 4.3: A) Depiction of the 10-20 electrode EEG cap layout. Red circles indicate the
electrodes over which a TMS coil is placed while the red arrow indicates the direction of
information transmission between brain areas when PAS is applied. B) The placement of
each TMS coil while a participant wears a 64 electrode EEG cap. The TMS coils are held in
place by moveable arms and the head of the participant is rested on a chin rest.
Baseline Data Acquisition
EEG resting-state acquisition – double baseline collection.
For each resting-state collection, the participant is asked to stare at a fixation cross
displayed on a laptop for 5 minutes. While maintaining their gaze on the fixation cross the
participant keeps a quiet posture while in a comfortable seated position. Data from each EEG
82
electrode is recorded at a sampling rate of 2048 Hz. Once 5 minutes of resting-state has been
recorded the participant rests for 10 minutes. Following that 10-minute rest interval, another 5-
minute resting-state measurement is completed. This procedure serves as a natural baseline
collection of a participant’s normal resting-state activity. The time interval of the double baseline
collection is identical to that of the PAS procedures described below. Therefore, one can control
for any normal changes with the PAS conditions by subtracting from the double baseline
collection. The PAS procedure begins immediately after baseline collection.
PAS ROI Selection: Functional vs. Anatomical
This protocol operates with the assumption that the EEG electrodes now overlay specific
areas of the cortex that are the target of stimulation. For example, with the EEG electrode array
configured in the 10-20 system, electrode F3 represents the assumed position of the left dorsal
lateral prefrontal cortex (DLPFC). However, without a structural MR image of the participant’s
brain there is no guarantee that the F3 electrode is in fact measuring DLPFC brain activity. We
argue that this may only be a limitation for prospective research that aims to increase
connectivity between regions of interest (ROIs) based purely on anatomical connectivity. In this
case an MR image is necessary to ensure the TMS coils are placed over the precise anatomical
brain regions. For this proposed protocol, the ROIs are based on previous research based on
functional connectivity. Here the brain activity between electrodes is what guides where the
TMS coils are placed and not necessarily the exact position of the participant’s cortex. This is the
difference between using PAS based on anatomical connectivity versus functional connectivity.
Researcher’s must be aware that applying PAS/EEG to ROIs based solely on anatomical
connectivity without MR imaging can be a serious limitation resulting in inaccurate stimulation
of intended ROIs.
83
PAS Procedure: Conditions and Parameters
With the double baseline complete, participants remain connected to the EEG computer
and amplifier and move to the TMS chair (Rogue Research, Montreal, Quebec, CA). The TMS
chair has two mobile TMS coil mount (Figure 4.3B). Each mount has multiple degrees of
freedom, thus enabling placement of each TMS coil over precise positions on the EEG cap
during the PAS procedure.
For our experiment, one TMS coil is placed over the prefrontal cortex, electrode AF8,
and a second TMS coil is placed over the right motor cortex, electrode C6 (Figure 4.3). The
BiStim TMS system is used to deliver paired pulses to these electrodes (Morrisville, NC).
Previous rs-EEG research conducted in our lab established that connectivity between AF8 and
C6 could predict performance and learning of a complex motor task (Hooyman et al., 2017). The
choice to deliver the lead pulse over AF8 and the proceeding pulse over C6 is due to the
hypothesized information flow between the two cortical areas. Previous research has shown that
information flow between the frontal and motor cortex is bidirectional but many studies both
functionally and structurally demonstrate information flow from the prefrontal cortex to the
motor cortex (Anwar et al., 2016). Therefore, the lead pulse over electrode AF8 depolarizes the
prefrontal cortical cells and axons which then travels along the projections arriving at motor
cortical neurons under electrode C6 (Figure 4.3A). Prior to the lead pulse’s arrival to the motor
cortex, the proceeding pulse is delivered at C6. Based on STDP mechanism, continuous volleys
of these paired pulses should increase the measured functional connectivity between electrodes
AF8 and C6.
To maintain head position throughout the PAS procedure a chin rest is used. Maintenance
of head position is critical for consistent delivery of the TMS pulse to the targeted electrodes.
84
The PAS procedure begins once the TMS coils are appropriately placed and the head is
positioned.
Paired Associative Stimulation (PAS) Parameters
PAS is a neuromodulation procedure in which paired single pulses of TMS are applied to
different areas of the nervous system in a temporally precise manner. We have modified the
traditional PAS procedure to modulate connectivity between distinct cortical areas. The
interstimulus interval (ISI) between lead pulse, first pulse, and the proceeding pulse, second
pulse, determines the effect of PAS on connectivity (Stefan et al., 2000). This experiment utilizes
two conditions, real and sham PAS, that only vary in ISI, 5 ms and 500 ms respectively. The key
features within each condition are the timing of the paired pulses controlled by ISI, total number
of pulses delivered and TMS intensity. Previous research has found that shorter ISI’s have larger
effects in terms of facilitation (Civardi, Cantello, Asselman, & Rothwell, 2001). Specifically,
connectivity between frontal and motor areas that are approximately 6 cm apart have been
inhibited with an ISI as short as 4 ms. In order to gain the greatest effect of potentiation within
our circuit, whose brain areas are approximately 7 cm in distance, we chose our real PAS
condition to have an ISI of 5 ms. It is important to apply shorter ISI’s for the purposes of larger
effects that can rise above the noise level of the EEG system. It could be that long ISIs, for
example 10 ms, will also have an effect, albeit reduced in size, due to the relationship between
spike timing and potentiation strength of STDP (Figure 4.1). While generating smaller effects
with longer ISIs may ensure potentiation, this approach may come at the cost of a change in
signal that is below the noise threshold determined by the double baseline.
85
Timing.
The real PAS condition, 5 ms ISI (PAS5), should increase connectivity between the
targeted rs-EEG ROIs while the sham PAS condition, 500 ms ISI (PAS500), should have no
effect on rs-EEG ROI connectivity. The 5 ms ISI for the real PAS condition was selected based
on the estimated time it would take for the lead pulse over AF8 to reach the assumed motor
neurons under the C6 electrode (Casula et al., 2016).
Number of Pulses.
Apart from the timing between pulses the number of paired pulses is also important.
Previous research has found that an application of 100 paired pulses can reliably increase cortical
activity between targeted regions. 100 paired pulses are delivered for each condition, sham and
real PAS. Between each paired pulse there is a 5 second interval. Using Signal software
(Cambridge Electronic Design Ltd., Cambridge, UK), precise timing for each condition can be
automated. With 100 paired pulses per condition the total length of each PAS procedure is ~8
minutes 20 seconds. No previous PAS research has demonstrated any negative side effects after
receiving real or sham PAS with the proposed number of pulses (Casula et al., 2016; Rajji et al.,
2011).
Intensity.
The stimulator intensity of each paired pulse is based on the participants RMT which was
measured at the start of the experiment. The stimulator intensity for the lead and proceeding
pulse for each condition is set at 120% of each participant’s RMT of the dominant APB muscle
(See Resting Motor Threshold Procedure/Determination of PAS Intensity Section for details).
The stimulator intensity of 120% RMT was chosen based on previous PAS research
demonstrating consistent depolarization of cortical neurons that can be measured by EEG while
86
still being safe for human application (Casula et al., 2016; Rajji et al., 2011, 2013). The most
important parameter for the successful application of the PAS paradigm is the intensity of the
TMS stimulation. If the TMS pulses are not capable of consistent depolarization, then the PAS
paradigm may be ineffective.
PAS Condition Randomization
Participants are pseudo-randomly assigned to first receive either the PAS5 or PAS500
condition. To control for any after effects between conditions, it is important to counterbalance
the conditions across subjects. Half of the participants receive the PAS5 condition first and, after
a 24-hour delay, the PAS500 condition. The other half receive the opposite condition order.
Counterbalancing allows for comparison between the PAS5 and PAS500 conditions even though
the methodology utilizes a within subject design. This maximizes statistical power of the
paradigm while requiring fewer participants to be recruited compared to a between subjects
design. Due to known variability in individual responsiveness to neuromodulation, it is beneficial
to understand if a sham and real condition have similar or differential effects for an individual. A
between subject design where groups either receive only PAS5 or PAS500 would leave questions
unanswered as far as if each paradigm has the same effect within each individual.
Controlling for Brain State
While the participants are receiving either PAS5 or PAS500 conditions they are
instructed to mentally count each pair of pulses as they are being delivered. Participants are not
told beforehand how many total pulses they receive for each condition. After each condition
participants are asked how many paired pulses they received to verify they were counting. This is
an important factor for controlling the participant’s mental state during each condition as it has
87
been shown that different mental states (i.e. performing a working memory task, resting-state, or
movement) can impact the effect of PAS on brain activity (Farzan et al., 2016).
While PAS is applied, the participants brain activity should be simultaneously recorded
with the EEG system. This allows one to examine the effect of PAS on intracortical connectivity
while the 100 pulses are being delivered, thus potentially providing insight on the optimal
amount of pulses to modulate connectivity. Once all 100 pulses are completed, the post resting-
state EEG measurement is performed.
Post PAS Resting-state EEG recording
The post PAS resting-state EEG recording is performed immediately after the first PAS
procedure and is identical to the first resting-state EEG recording (i.e. 5 minutes of recording
while looking at a fixation cross in a seated position). One should ensure that the impedance
measures for the EEG cap are maintained at the appropriate threshold with any necessary
adjustments made prior to recording. This is to maintain consistent signal to noise ratio for each
electrode throughout the course of the experiment. Once this recording is complete participants
will return for the seconds PAS condition after a 24-hour delay.
Inter-condition Delay
The 24-hour time period between conditions is necessary for preventing after effects from
the first condition to the second. Previous research has shown that the majority of rTMS after
effects, e.g. changes in MEP amplitude, return to baseline levels between 20 to 60 minutes
(Klomjai et al., 2015). Although, there is variability in how long neuromodulatory effects persist
among individuals. Ideally, the second PAS condition would commence once resting-state
intracortical connectivity has returned to baseline. This would require individual wash-out
periods for each participant based on their response to PAS which would not be feasible as
88
confirmation that whole scalp after effects have dissipated takes considerable time to compute.
Further, some individuals may require hours, if not days, of wash-out before they return to
baseline. This was the experience in the original peripheral PAS work by Stefan and colleagues
(ref). As mentioned earlier, individual wash-out periods may be impractical, but a 24-hour delay
between conditions should provide a practical solution for the general population that reduces the
probability of other confounds related to the participant’s routine habits of lifestyle (Discussed in
section, Accounting for After Effects Between Conditions).
Day 2: Pre-measurement Procedures for PAS Condition 2
Participants should return for Day 2 of the experiment at approximately the same time of
day they experienced Day 1. Prior to the application of the second PAS condition, participants
should be re-oriented with the same EEG cap and the cap should be calibrated to the same
impedance levels as Day 1. The stimulator intensity for the second condition should be the same
as the PAS condition on Day 1 (120% RMT). Prior to the second PAS condition the participant
will undergo a 5-minute resting-state measure with eyes-open focused on the same fixation cross
as was used for Day 1.
Second PAS Condition
Once the inital resting-state EEG measurement has been completed the participant
undergoes another PAS procedure. If the participant first underwent the PAS500 condition, they
would then receive the PAS5 condition and vice versa. Each participant is asked to keep count of
the number of paired pulses, 100, throughout the procedure to ensure the same brain state for
both PAS conditions. Once the 100 pulses are completed a final resting-state EEG measurement
is completed.
89
Post resting-state EEG recording for Condition 2
With each PAS condition concluded, the participant immediately has their resting-state
brain activity recorded for a final 5-minute collection. Once the collection is complete the
participant can remove the EEG cap.
Data Processing and Analysis
Resting-state analysis - artifact removal.
Analysis of resting-state data must first undergo removal of muscle and eye artifacts that
can interfere with accurate analysis of neural data. For these purposes Independent Component
Analysis (ICA) is used as a means of segmenting out muscle artifacts from the EEG recording.
For this experiment, we recommend the open source software EEGlab and its embedded ICA
analysis for EEG artifact removal (Delorme & Makeig, 2004; Delorme, Sejnowski, & Makeig,
2007). As part of ICA, components identified by the analysis have to be individually removed by
the researcher. If ICA is used to de-noise, we highly recommend prospective researchers first
review the EEGLAB materials and practice artifact removal before experiment commencement.
We provide guidelines on how to visually identify ICA components that should be removed as
follows (Figure 4.4).
90
91
Figure 4.4: Examples of brain (A) and eye (B) components. To improve signal to noise ratio
of the rs-EEG signal, the eye component (B) should be removed from the data set and the
brain component (A) remains. A brain component can be identified from the following
information provided graphically by resulting ICA of a 5 minute resting-state recording.
Brain components can be identified by: 1) a peak at 7 – 15 Hz within the Activity power
spectrum, 2) The ICA activity matrix, representing the change in frequency of the
component throughout the recording (graph in the top right of A), demonstrates random
spiking of both high and low frequency. An eye component, seen in B, has the following
characteristics 1) eye components lack any peak in the 7 – 15 Hz range and instead has a
peak near 1 Hz 2) IC2 activity matrix, has large and sustained changes in frequency
throughout the recording. These red changes within the matrix are when the EEG system
recorded muscle activity from the eyes. 3) This can be further confirmed from the electrode
map next to the IC2 activity matrix where the frequency is dominated near the eye region.
These images are generated through the opensource eeglab software and further detail on
how to implement and interpret ICA through EEG lab can be found here (Delorme &
Makeig, 2004).
After the noise has been removed the data is zero phase, band-passed filtered with cutoff
frequencies between 10 and 60 Hz (Gustafsson, 1996). We recommend this frequency range due
to previous research investigating the association of the beta band (13 – 32 Hz) with the motor
system (Wu et al, 2014; Wu et al., 2018). Each electrode pair (1830 combinations total) is then
measured for their specific intracortical connectivity using the mscohere function from the Signal
Processing Toolbox within MATLAB (version 2013a) with a 1 second Hanning window. This
function outputs a coherence spectrum, Cxy, within the frequency range of interest which is
calculated from equation (1). Pxx and Pyy are the power spectrum densities of electrodes x and y
92
and Pxy is the cross spectral density of x and y.
𝐶 𝑥𝑦
=
|𝑃 𝑥𝑦
(𝑓 )|
2
𝑃 𝑥𝑥
(𝑓 )𝑃 𝑦𝑦
(𝑓 )
(1)
We record the peak coherence value within the frequency band to signify the connectivity
between the two EEG signals (Srinivasan et al., 2007) and compare the change between
conditions. This peak value is representative of functional connectivity and lies between 0 and 1,
similar to a correlation coefficient. A coherence value close to 0 means low connectivity and a
value close to 1 means high connectivity. One can then compare this coherence value between
our hypothesized electrodes, in this example AF8 and C6, and the other 1829 electrode pair
combinations to determine if the effect of PAS is selective or global.
Example Outcome
Once the rs-EEG data has been preprocessed, subtraction of resting-state collections for
analysis of intracortical connectivity potentiation can commence. Here we provide an example
participant demonstrating modulation in the circuit between electrodes AF8-C6. It is important to
note that this participant data should be viewed as a relative proof of principle. We provide this
example to help prospective researchers visualize expected locality of changes in intracortical
connectivity as a result of PAS delivered to the specific circuit of interest. First, we subtract the
first baseline condition from the second baseline condition to observe normal changes in brain
activity between each baseline (Figure 4.5A). Then we subtract the pre-PAS5 condition from the
post-PAS5 condition to measure changes in connectivity specific to PAS5 (Figure 4.5B). The
same post-pre subtraction is performed on the PAS500 condition (Figure 4.5C). Each of the head
maps in Figure 4.5 are referenced to electrode AF8. Therefore, all changes in coherence are
directly linked to changes in intracortical connectivity that occurred between that cortical area
93
and AF8. As shown in Figure 4.5, the single participant depicted here experienced a localized
increase of intracortical connectivity underlying C6 as a result of the PAS5 condition which did
not occur in either the baseline nor PAS500 condition.
Figure 4.5: Changes in Area of Significant Coherence across each condition for a single
participant: A) Baseline, B) PAS5 and C) PAS500. A. Baseline changes in intracortical
connectivity decrease within the target circle and fluctuate randomly throughout the
cortex. B. Changes in intracortical connectivity as a result of the PAS5 condition are local
to the target circuit AF8-C6 and do not appear to impact other areas of target circle and
fluctuate randomly throughout the cortex. B. Changes in intracortical connectivity as a
Baseline, B) PAS5 and C) PAS500. A. Baseline changes in intracortical connectivity
decrease within the Figure 4.5: Changes in Area of Significant Coherence across each
condition for a single participant: A) For Peer Review the cortex. C. PAS500 does not
modulate intracortical connectivity within the target circuit. Overall connectivity is
decreased in the left prefrontal and motor cortical areas.
Secondary Analysis to Account for Responsiveness and Real Time PAS effect
Under a well powered study we would expect the PAS5 condition to significantly
increase intra-cortical connectivity between the AF8-C6 electrode pair compared to the PAS500
94
condition. With subtraction of the baseline collection from each PAS condition, we can more
clearly delineate what changes in connectivity occur due to the PAS condition versus normal
fluctuations in brain activity. Given the variability of responsiveness in previous
neuromodulatory paradigms (M. Hamada et al., 2013; Masashi Hamada et al., 2014b; Wiethoff,
Hamada, & Rothwell, 2014), there is a reasonable likelihood that intracortical connectivity will
be modulated in some individuals but not others. Beneficially, EEG affords the possibility of
predicting responders vs. non-responders given the plethora of EEG data that can be derived
from a single subject. A well powered study would be able to analyze which EEG measures can
predict PAS responsiveness for purposes of pre-screening viable applicants for an intracortical
PAS paradigm. For example, one may find that individuals with a baseline coherence near 1
(high connectivity) are less likely to experience coherence changes secondary to PAS.
As a secondary analysis one can analyze the EEG data collected during the delivery of
PAS to examine the effects of each condition on intracortical connectivity over the course of the
100 pulses. To process this data, one would remove the data points immediately before the pulse
and 1 second after the pulse. This would result in 100 four-second EEG epochs, one for each pair
of pulses. Coherence would be calculated as described above for the rs-EEG data and compared
between epochs. This measurement provides a real-time determination of how intracortical
connectivity is modulated overtime.
Discussion
Within the described methodology, we have highlighted a rationale for how to control for
individual variability and technological limitations in the application of PAS/EEG. Additionally,
we have provided steps for a rigorous analysis of PAS/EEG data with code that can be found on
the researchers Github repository https://github.com/hooymana/PAS-EEG.git . In an effort to
95
demonstrate reproducibility, openness and propagation of this method, we are providing our code
so the outlined PAS/EEG method can be used by others and improved upon for future studies.
We believe this methodological paradigm is a necessary launching point for future
investigations using PAS to not only manipulate EEG-based neural substrates but also the
associated motor behavior. As stated above, current neuromodulatory paradigms, rTMS and
tDCS, have demonstrated limited capability to modulate a specific neural circuit (Johnson,
Hamidi, & Postle, 2010; Rothwell, 2012; Woźniak-Kwaśniewska, Szekely, Aussedat, Bougerol,
& David, 2014). Furthermore, the incorporation of EEG to assess the efficacy of PAS allows for
more sensitive measurement of brain changes than previously used methods, such as MEP
amplitude. By developing and testing a PAS/EEG paradigm, we are providing a novel method to
investigate the relationship between brain function (i.e. specific cortical circuits) and motor
performance and learning.
Future Work
Brain-behavior research.
Anchored by the reliable and accurate application of intracortical PAS to modulate
targeted intracortical connectivity, the primary question is can these improvements in targeted
connectivity benefit motor behavior? We recommend that any PAS study with the aim of
modulating motor behavior always include a double baseline rs-EEG collection prior to PAS to
ensure reliability. Thus far in motor learning/neuromodulation research, changes in behavior
secondary to changes in connectivity have been tenuous (López-Alonso, Cheeran, & Fernández-
del-Olmo, 2015). To assure scientific rigor, it is important for future brain-behavior studies to
have a strong scientific rationale and convincing preliminary data for why a specific intracortical
circuit was chosen. One can easily fall into the trap that because a neural circuit is predictive of
96
motor learning, modulation of said circuit will directly change behavior. It may be that the circuit
indirectly impacts behavior and at different intensities for different people. Therefore,
researchers performing PAS brain-behavior studies should power the study to accommodate a
big enough sample to allow sub-group analyses clustered by a range of responses to the PAS
intervention (e.g. low/none, moderate, high).
PAS research and motor behavior.
Here we have focused on a PAS method to induce Long Term Potentiation. Future work
of intracortical PAS, if scientifically warranted, may want to explore the effects of Long Term
Depression of a targeted circuit through reversal of the timing between pulses (see Fig 1), post-
synapse before pre-synapse. We have demonstrated preliminary evidence, as a proof of principle,
that PAS can modulate connectivity immediately after application. However, another important
line of future research concerns the persistence of PAS after effects both in the functional and
structural plasticity of the brain. These ideas represent what is likely just the tip of the iceberg for
this new and exciting brain-behavior research paradigm.
Conflict of Interest
None of the researchers have any conflicts of interest as part of this research.
Ethical Considerations
All research was conducted with integrity. Participant informed consent was obtained
prior to study and identity is kept confident and anonymous. All efforts were carried out to
prevent harm to the research participant. Methods and results reported here were performed
independently and impartial to any outside party.
97
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study
Abstract
Introduction: Intracortical connectivity measured with whole brain resting-state
electroencephalography (rs-EEG) has been shown to be a robust and highly accurate predictor of
motor skill learning and stroke recovery. However, dominant methods of non-invasive brain
stimulation (NIBS) are plagued by inconsistencies in effecting reliable neuromodulation, and are
restricted to a single cortical region. Previous work has used Paired Associative Stimulation
(PAS) to modulate peripheral and central nervous system connectivity through a mechanism of
Spike Timing Dependent Plasticity (STDP). With recent advances in EEG technology it is now
possible to configure an intracortical PAS (iPAS) paradigm with the aim to improve resting-state
intracortical connectivity (rs-IC) between two cortical regions.
Purpose: The overall purpose of this study is to establish feasibility of using iPAS to improve rs-
IC between targeted cortical regions measured by rs-EEG.
Methods: Eleven right hand dominant non-disabled adults were recruited (mean age 26.4, sd
5.6, 5 female). Each participant underwent a double baseline measurement, and then a true- and
sham-iPAS condition in a counter-balanced order. For both conditions iPAS was delivered at
120% of Resting Motor Threshold of the Abductor Pollicus muscle of the dominant hand.
Participants received either 100 pulses of true iPAS with a 5 ms inter-stimulus delay (iPAS5) or
100 pulses of sham iPAS with a 500 ms inter-stimulus delay (iPAS500), counter-balanced for
order, between the prefrontal and motor cortex area of the right hemisphere. To determine the
effect of iPAS, whole brain rs-EEG was acquired pre- and post each iPAS condition and then
98
subjected to a series of analyses including: coherence calculation using ICA (Independent
Component Analysis); baseline reliability using ICC (intraclass correlation coefficient), and
determination of inherent baseline noise using standard error of measurement (SEM). The
primary analysis compared the change in coherence under the three conditions (i.e., Baseline,
iPAS5, iPAS500) using a one-way ANOVA.
Results: iPAS5 induced a significant increase in rs-IC, measured as coherence between
electrodes AF8-C6, compared to iPAS500 and baseline (p < .05); this provided a demonstration
of feasibility for iPAS5 to effectively modulate intracortical connectivity.
Conclusion: Stimulation order did not impact the efficacy of iPAS5 to modulate rs-EEG
between the left prefrontal and primary motor cortex regions. The efficacy of iPAS5 over
iPAS500 provides preliminary support for a mechanism of action common to STDP for which
the timing between paired pulses is the critical ingredient for effective iPAS neuromodulation.
Introduction
The use of functional connectivity, the shared communication between distinct brain
areas, in neuroimaging has evolved our understanding of how the brain learns motor skills
(Mehrkanoon, Boonstra, Breakspear, Hinder, & Summers, 2016a; Tomassini et al., 2011; van
den Heuvel et al., 2009). There is now strong evidence that information exchange among distinct
networks within the brain, and not simply activity from a single brain region, is strongly
associated with skill learning (Mehrkanoon, Boonstra, Breakspear, Hinder, & Summers, 2016b).
This is most evident in the examination of how resting-state brain activity, measured as
correlated oscillations of activity between cortical regions during wakeful rest, is capable of
99
highly accurate prediction of future motor skill performance and learning (Wu et al., 2018,
2014).
Theoretically, to improve the learning capabilities of those with low resting-state
intracortical connectivity, a predictor of poor motor learning, may require Non-Invasive Brain
Stimulation (NIBS) (Hordacre et al., 2016). Ideally, stimulation would be specific to the
identified rs-EEG circuit (e.g. between the prefrontal and motor cortex regions) to boost
information exchange and thus facilitate the learning process. However, current NIBS
techniques, such as repetitive Transcranial Magnetic Stimulation (rTMS) and Transcranial Direct
Current Stimulation (TDCS) have been used primarily to modulate a single brain region of motor
origin (Kantak, Sullivan, Fisher, Knowlton, & Winstein, 2010; Rroji, van Kuyck, Nuttin, &
Wenderoth, 2015). Therefore, there is an urgent need for a new NIBS method capable of
modulating/strengthening intracortical connectivity as measured by rs-EEG.
To modulate a specific circuit we propose a reconfiguration of the traditional Paired
Association Stimulation (PAS) paradigm (K. Stefan et al., 2000). Previously, PAS has been
shown to strengthen communication between the peripheral and central nervous system using
coordinated bouts of paired pulses between the two sites of activation, primarily the median
nerve and motor cortex (Kamke, Nydam, Sale, & Mattingley, 2016). Communication strength is
enhanced by PAS through the Hebbian mechanism of Spike Timing Dependent Plasticity
(STDP) and has been shown to effect more reliable modulation compared to rTMS and TDCS
(Strube, Bunse, Malchow, & Hasan, 2015; Wiethoff, Hamada, & Rothwell, 2014). To target a
specific rs-EEG circuit we have effectively created an intracortical PAS (iPAS) paradigm by
moving the once peripheral stimulation site to a targeted region of the cortex (Hooyman et al., in
review, JMLD).
100
Due to advances in EEG technology the effects of an iPAS paradigm can now be
captured without risk to human subjects (Rizzo et al., 2009). Therefore, we could use an iPAS
paradigm to modulate previously identified rs-EEG circuits known to be functional predictors of
motor learning. The purpose of this experiment is to demonstrate a proof of concept that iPAS
can strengthen a specific rs-EEG circuit and that the interstimulus interval (ISI) between paired
pulses, the deterministic factor of STDP efficacy, is the mechanism of action for the change in
communication strength. Confirmation that iPAS can be used to strengthen a predictive
biomarker of motor learning could serve as a critical foundation for future research that aims to
determine the brain-behavior relationship between intracortical functional connectivity and
motor skill learning.
Methods
Detailed methods of this study’s protocol have been published elsewhere for purposes of
replication and may be viewed there (Hooyman et al., in review, JMLD). For brevity, the
methods described here are abbreviated. Eleven non-disabled adults (5 female, mean age 28.2)
consented to participate. All participants were right hand dominant as measured by the
Edinburgh Handedness questionnaire (Oldfield, 1971). Each participant also received a TMS
safety screen to ensure that their participation would be of low risk to health and well-being
(Rossini et al., 2015). One participant had to be excluded from future analysis because
comparison of baseline measures was too variable and indicated unreliability of signal for future
comparison and inclusion. This resulted in 10 participants total, five in iPAS5-iPAS500 and five
in iPAS500-iPAS5 order groups.
This experiment utilizes a within subject design with each participant undergoing both a
true and sham iPAS condition. To account for any order effect of condition, participants were
101
counterbalanced across conditions pseudo-randomly so an equal number of participants receive
true and then sham iPAS and vice versa. Prior to and after each iPAS condition participants
answer a stimulator comfort survey to ensure they are not experiencing any negative side effects,
physical or otherwise, as a result of either of the two iPAS procedures.
Determination of iPAS Intensity
Prior to iPAS application, resting motor threshold (RMT) of the right abductor brevis
pollicus (APB) is determined. It is critical that before iPAS intensity is determined, a 64
electrode EEG cap (ANT Neuro, Netherlands) is fitted to individual head circumference. Single
pulses of TMS are delivered by a 70mm figure-8 coil connected to a Magstim stimulator
(Norrisville, NC). The RMT for each participant is determined as the lowest stimulator output
intensity capable of producing five out of 10 Motor Evoked Potentials (MEPs) with amplitude >
0.5 volts (Hallett, 2007). The decision to measure RMT of the right APB is based on previous
research in our lab where APB served as the prime mover during practice of a novel motor task
where performance on the task was predicted by rs-EEG (Hooyman et al., submitted). Here, we
chose RMT of the same prime mover as an appropriate proxy for this and future work that would
use this novel iPAS methodology.
Determining RMT after the fitting the EEG cap is notable because it allows us to account
for the additional space the EEG cap creates between the TMS coil and the scalp. Thus, we can
be confident that the RMT identified with the EEG cap is capable of reliable depolarization of
cortical neurons involved with the motor learning process. The EEG cap uses ferromagnetic
electrodes that prevents increased heat of the electrode that might occur in concert with the TMS
pulse. Therefore, application of TMS is safe while a participant wears the EEG cap. For this
experiment we use 120% of RMT as the iPAS intensity; an intensity that has been shown to be
102
effective in previous iPAS studies examining other neuronal mechanisms (Koch, Ponzo, Di
Lorenzo, Caltagirone, & Veniero, 2013; Veniero, Ponzo, & Koch, 2013).
Baseline Procedure for rs-EEG
Once iPAS intensity is established, the EEG cap is calibrated with each electrode set to
20 kOhm or lower. Participants undergo the first 5-minute resting-state measurement where they
are instructed to assume a comfortable seated posture, maintain relaxed facial musculature and
keep their eyes fixed on a red fixation cross displayed on the computer monitor 36 inches in front
of them. Once the first rs-EEG measurement is completed the participant will undergo a 10-
minute rest period and then receive a second 5-minute rs-EEG measure. This serves as a repeated
baseline procedure (Figure 5.1) to control for changes in rs-EEG that may occur as a function of
the passage of time, equivalent to each iPAS procedure. With the two baseline rs-EEG measures
we can determine reliability of measurement and determine noise levels of the EEG system
above which a meaningful signal can be detected. This double baseline procedure generates an
individual resting change that is independent of any stimulation and therefore provides a robust
control condition to compare changes induced by the iPAS conditions (Casula, Pellicciari,
Picazio, Caltagirone, & Koch, 2016).
103
Figure 5.1: Study paradigm. Experiment is a within subjects design that commences with
resting motor threshold of the APB determined with the EEG cap already oriented.
Participants undergo a double baseline procedure before either iPAS conditions for
reasons of rs-EEG measurement reliability. Experimental conditions are counterbalanced
across the study sample to control for order effects of iPAS application. EEG =
Electroencephalograpphy, RMT = resting motor threshold, PAS = paired associative
stimulation. (Figure adapted from Figure 4.2 in Hooyman et al., In Review, JMLD)
Intracortical PAS Conditions, Coil Orientation
Immediately after the second rs-EEG procedure participants undergo either the true of
sham iPAS condition. Each condition is performed with two 70 mm figure-8 coils connected to a
Magstim bistim system (Morrisville, NC). Each iPAS condition consists of 100 paired pulses at
120% RMT. The only difference between conditions is the interstimulus interval (ISI) between
each paired pulse. In the true condition the ISI is 5 ms (iPAS5) and in the sham condition the ISI
104
is 500 ms (iPAS500). Based on previous iPAS studies and the STDP mechanism the 5 ms ISI is
capable of direct cortical modulation, whereas the 500 ms ISI should yield no effect (Johnen et
al., 2015).
The TMS coil on the AF8 electrode was positioned at 45 degrees with respect to the
midline with the coil facing in the posterior-anterior direction. The coil on the C6 electrode is
positioned at 45 degrees to midline in the anterior-posterior direction (Figure 5.2). This is a
similar configuration to previous research modulating resting-state networks using iPAS and
functional Magnetic Resonance Imaging (fMRI) (Santarnecchi et al., 2018). Throughout each
condition the stimulation was set to monophasic.
105
Figure 5.2: iPAS Set up/EEG electrode layout. A. Placement of two TMS coils is identical
between iPAS conditions. The TMS coil for the first pulse over electrode AF8 is oriented at
a 45 degree angle with respect to the midline in a posterior-anterior position. The TMS coil
of the second pulse over electrode C6 is oriented at a 45 degree angle with respect to the
midline in the anterior-posterior position. B. The EEG electrode layout is based on the 10-
20 system. Dark circles represent the electrodes of interest and the white arrow represents
the theoretical flow of information induced from iPAS between the underlying cortical
regions of interest. Ground electrode (GND) is in blue and reference electrode is in green
(CPZ).
After the second round of rs-EEG the participant is fitted with the TMS coils with one
coil placed on either electrode on the EEG cap that composes the rs-EEG circuit of interest (AF8
and C6, Figure 5.2). As the order of the PAS pulses determines the flow of information along a
circuit the first, lead, pulse is AF8 and the second, proceeding, pulse is C6. Once the TMS coils
are in place the participant will either receive 100 pulses of iPAS5 or iPAS500 paired pulses
delivered at 120% RMT of the right APB with a 5 second pause between each set of paired
pulses. To control for brain state, each participant is asked to count the total number of paired
pulses delivered during each condition while maintaining gaze on the rs-EEG fixation cross
(Farzan et al., 2016). Participants were not told the total number of paired pulses they would
receive for each condition. After each condition we asked participants how many pulses they
counted as a measure of accuracy. Overall, each iPAS condition takes approximately 9 minutes
to complete.
106
Wash-out Period
Conclusion of the first iPAS condition leads to a follow-up 5 minute rs-EEG
measurement that is the same as the previous two administered before iPAS. Once the first post
iPAS condition and 5 minute rs-EEG is finished the participant undergoes a 30-minute wash-out
period. It is important to note that the participant continues to wear the EEG cap throughout the
experiment. As a result the impedance levels prior to each rs-EEG measurement are checked for
any changes in impedance above 20 kOhms. This wash-out period is used to mitigate any after
effects that may carry over to the second iPAS condition. However, we cannot be certain in real
time that any effects of the first iPAS condition have completely dissipated. Because of the
computation time required to measure rs-EEG, this determination must be done post hoc and
reported as such. This limitation is what led us to use a counterbalanced design so that if there
are any after effects that persist, they are subjected to cross cancelation.
Second Intracortical PAS Condition
At the end of the wash-out period the participant undergoes a 5-minute rs-EEG
collection. Then the participant undergoes the second iPAS condition. If the participant first
received iPAS5 then they will receive iPAS500 and vice versa. With the second iPAS condition
complete the participant will undergo a final 5-minute rs-EEG measurement. After the 5
th
rs-
EEG measurement, the participant is debriefed and their participation officially ends. If they
report on the stimulation comfort survey that they are feeling any side effects as a result of either
iPAS condition, then we ask them to remain in the laboratory until their symptoms have
normalized.
107
Data Processing
Coherence Calculation
Each rs-EEG collection is converted to a MATLAB (Version 2013a) compatible file type
and imported for artifact removal, filtering and coherence calculation. First, each measure
undergoes an independent component analysis (ICA) to remove any eye and/or face muscle
artifacts (Delorme & Makeig, 2004; Delorme, Sejnowski, & Makeig, 2007). To maintain
blinding and prevent bias in later stages of statistical analysis, we used an algorithm that jumbled
and then coded the data. This was to ensure that the experimenter would remain blinded to the
iPAS condition undergoing processing and until a later stage of statistical analysis.
Once eye and muscle artifacts are removed then the data are band-passed filtered with
cut-off frequencies of 5 and 60 Hz. Each channel has its coherence spectrum calculated,
referenced to AF8, the site of the lead TMS pulse. For each electrode pairing the peak coherence
value is recorded in the alpha (8-12 Hz), low beta (13-20 Hz), high beta (21-30 Hz), and gamma
band (31-50 Hz) (Srinivasan, Winter, Ding, & Nunez, 2007).
Statistical Analysis
Baseline Test-retest Reliability of rs-EEG
We analyzed the test-retest reliability of the two baseline rs-EEG measures given a
change in time equivalent to each iPAS condition (i.e. 10-minutes). We used the two baseline
measures to determine a standard error of measurement (SEM) in order to differentiate a change
in connectivity due to measurement noise from a true change in connectivity from iPAS. Finally,
we compared the change between the two baseline conditions using a paired t-test.
108
Changes in Connectivity as a Result of iPAS
Previous research in our lab specifically looked at how high beta coherence could predict
motor skill learning. Given that beta coherence has been used in other research to predict motor
skills we chose to evaluate modulation using only this bandwidth (Wu et al., 2018, 2014). To
determine if changes in resting-state intracortical connectivity were significantly different
between conditions, a one-way analysis of variance (ANOVA) was used to compare changes in
connectivity between baseline, iPAS5 and iPAS500.
Visualization of iPAS Effects Across the Cortex
To observe the effects of each iPAS condition across the cortex we calculated coherence
for each electrode pair across the scalp with reference to AF8. We identified only those electrode
pairs that exhibited change greater than SEM and generated a topographical map of the results.
This allowed us to identify the magnitude and specificity of coherence change between iPAS and
baseline conditions.
Order Effect Between iPAS Conditions
We visually inspected for the presence of an order effect between iPAS conditions and
formally tested this using a paired t-tests between each iPAS condition by day. Specifically we
looked for any differential response in coherence changes that depended on the initial condition
(i.e., iPAS5 or iPAS500). All statistical tests were done with MATLAB and R Version 3.5.1.
Results
Controlling Brain State
Each participant demonstrated a near accurate count of total paired pulse delivery during
each condition: iPAS5 mean pulse count = 99.6, range = 98 – 101; iPAS500 mean pulse count
100.1, range 99 – 101.
109
Test-retest Reliability
The calculated intraclass correlation coefficient of the double baseline between AF8 and
C6 was 0.76 (CI = 0.24 - 0.94) and statistically significant (2, 9, <.01). According to Koo and Li
(Koo & Li, 2016) an ICC of 0.76 is considered to be good quality. The standard error of
measurement (SEM) derived from the resulting ICC was .096. The SEM is used to determine if
changes in coherence among each participant were above this noise level. A scatterplot of
baseline measures is shown in Figure 5.3. A paired t-test confirmed that the two baseline
measures were not statistically different from each other (p > .05).
110
Figure 5.3: Scatterplot of Individual Basdeline Measures. X-axis is the coherence between
AF8 and C6 obtained at the first baseline; Y-axis is the same measure obtained at the
second baseline. Changes in baseline coherence plotted as a scatter plot to demonstrate
reliability of coherence measures for the two resting-state EEG baseline sessions separated
by a 10 minute rest period.
Changes in Connectivity Due to iPAS
The iPAS5 condition had the greatest increase in coherence compared to baseline and
iPAS500. The ANOVA that compared change in coherence between baseline, iPAS5 and
iPAS500 conditions was statistically significant (F = 14.23, p<.01, Figure 5.4). Post-hoc analysis
revealed that the locus of the significant effect was the iPAS5 condition (p < .05). Within the
iPAS5 condition we compared each participant’s change to that of SEM (i.e. .096) and found that
8 out of 10 individuals had increases in coherence greater than SEM. In the iPAS500 condition
only 1 out of 10 individuals exceeded the SEM threshold (Figure 5.4 and 5.5).
111
Figure 5.4: Changes in AF8-C6 Coherence for the three conditions. Comparison of
coherence change across baseline, iPAS5 and iPAS500 conditions. ANOVA revealed that
iPAS5 had a significantly greater increase in high beta coherence between the AF8-C6
electrode pair compared to baseline and iPAS500 conditions. Box and whisker plots with
individual subject data overlying the Box and whisker distribution (N = 10). Dashed line
represents SEM threshold in positive and negative direction.
Figure 5.5: Individual changes (Pre/Post) in circuit coherence between conditions. A.
Changes in coherence between individuals demonstrated that a majority (8/10) of
individuals experienced an increase of high Beta coherence above SEM in the iPAS5
condition compared to B. Only 1/10 participants experienced an increase above SEM in the
iPAS500 condition.
Visual Localization of iPAS Effects across the Cortex
Topographical maps, referenced to AF8, of changes in coherence above SEM
demonstrated no change in connectivity between baseline conditions; iPAS5 had local effects
112
near the targeted electrode C6, while iPAS500 had no effect. The specificity of the iPAS5
condition appears to be local to the site of interest (i.e., Electrode C6, Figure 5.6)
Figure 5.6: Change in connectivity for baseline and iPAS conditions within the high Beta
band. Visual results of topographical maps demonstrate an increase of high Beta coherence
above SEM within the AF8-C6 circuit occurred only after receiving iPAS5, but not
Baseline or iPAS500.
Order Effects of iPAS
Visual inspection of data did not demonstrate an order effect between individuals who
initially received iPAS5 or iPAS500. Statistically there was no difference in AF8-C6
connectivity change between the iPAS5 conditions (p > .05) or the iPAS500 conditions (p > .05)
dependent on condition order (Figure 5.7).
113
Figure 5.7: Changes in iPAS across the 5 rs-EEG sessions for both iPAS orders. Visual
inspection indicates that iPAS5 increased high Beta coherence compared to iPAS500
regardless of condition order. Additionally, iPAS500 did not effect a noticeable change in
high Beta coherence. Paired t-tests by each condition by day were not significantly different
(p >.05). N = 5 for each order condition. Two-way perforated arrow pointed at each iPAS5
slope denotes change in coherence as a result of iPAS5 regardless of condition order. Error
bars represent standard error of each resting-state measurement.
Discussion
Our findings support the hypothesis that iPAS is a feasible tool to use to modulate
functional connectivity between two cortical regions. Additionally, a change in coherence can be
driven uniquely by specific inter-pulse timing as evidenced by that fact that within our targeted
circuit, iPAS5 generated a greater improvement in coherence than did iPAS500. This provides
114
evidence that the putative mechanism responsible for change in rs-EEG is most probably Spike
Timing Dependent Plasticity (STDP). A targeted intervention to increase rs-IC between distinct
brain regions using non-invasive dual stimulation sites has yet to be described in the literature.
Previous research used either a single stimulation site to modulate resting-state brain networks or
looked at the change in correlation between networks (Gratton, Lee, Nomura, & D’Esposito,
2013; Halko, Farzan, Eldaief, Schmahmann, & Pascual-Leone, 2014; Santarnecchi et al., 2018;
Wang et al., 2014). The capability to target a specific intracortical circuit that has previously
been shown to predict motor learning behavior may in the future, advance our understanding of
the causal relationship between brain and behavior in motor neuroscience.
Our double baseline measure allowed us to accomplish two things: 1) determine
reliability of our outcome measure, rs-EEG and 2) Calculate a noise level, SEM, to compare to
any changes within either iPAS condition thus allowing for a more rigorous identification of
responders/non-responders. Previous NIBS work has not always established reliability for
measures such as MEP amplitude to determine LTP like effects on connectivity (Müller-
Dahlhaus, Orekhov, Liu, & Ziemann, 2008). Commonly any change in amplitude is determined
statistically or arbitrarily pre/post NIBS intervention. This means that changes in MEP that occur
irrespective of NIBS cannot be controlled (Nettekoven et al., 2015).
We demonstrated that iPAS5 had a superior effect on rs-IC compared to iPAS500.
Importantly, the only difference between these conditions was the inter-stimulus interval (ISI, 5
ms compared to 500 ms) between sets of paired pulses. Each condition was equivalent in number
(100) and intensity (120% RMT) of delivered pulses. As such, we conclude that the
improvement in resting-state functional connectivity is driven by an STDP-like mechanism.
These results are consistent with previous research showing how even small changes in the Inter-
115
stimulus interval between pulses can have strong effects on spontaneous inter-regional activity
(Casula et al., 2016; Rizzo et al., 2009; Veniero et al., 2013).
The 5 ms ISI used here was chosen based on previous research and hypothesized
conduction velocity of cortico-cortical association fibers (Massimini, Huber, Ferrarelli, Hill, &
Tononi, 2004; Nunez & Srinivasan, 2014). However, other ISI delays may be more successful
than what was used here. More importantly, there is need for a systematic investigation of a full
range of ISIs and their capability to modulate rs-IC with durable after effects. This is especially
relevant using current methods that allow for individualized ISI measurements taken with
TMS/EEG recordings (Casali et al., 2013). Further, improved modulation effects may be
manipulated as a function of brain state (Farzan et al., 2016; Johnen et al., 2015). In this
experiment each participant was given the same task, pulse counting, during each iPAS condition
in order to control brain state across individuals. We believe brain state for each participant was
maintained because each individual reported an accurate total count of pulses received for each
iPAS condition. However, an application of iPAS while participants are practicing or performing
mental imagery of the task may enhance the effect on rs-EEG, similar to that seen in rTMS
research (Cona, Marino, & Semenza, 2017).
Based on the estimated rs-EEG SEM (.096), 80% (8/10) of participants responded to the
iPAS5 condition compared to 10% (1/10) in the iPAS500 condition. This response rate is similar
to the canonical PAS paradigm used to affect a change in connectivity between the peripheral
and central nervous system (Strube et al., 2015). The non-responders in the study could possibly
be explained by neurotransmitter release, such as serotonin. Previous work found that the
administration of citalopram, a selective serotonin reuptake inhibitor, facilitated LTP like effects
of PAS compared to a placebo (Batsikadze, Paulus, Kuo, & Nitsche, 2013). We cannot confirm
116
that diminished serotonin levels might have caused non-response, but it is important to note that
neural plasticity relies strongly on the coupling of both chemical and electrical transmission to
effect change (Cahill, McGaugh, & Weinberger, 2001).
We utilized a counterbalanced within subjects study design to identify the unique effects
of iPAS5 and iPAS500 within an individual. We believe this within-subject approach is superior
to a between-subject approach in light of the high variability in responsiveness seen in previous
single site neuromodulatory methods (Hamada et al., 2014). A within-subjects design allows
better control of the variability between individuals than variability between experimental
conditions. Although our wash out period is only 30 minutes in length, by counter-balancing
participants we can control for possible after effects that might contaminate the next condition. It
would be unrealistic to personalize wash-out periods given the time it takes to process rs-EEG
and the fact that some individuals may have after effects that persist for hours after the
intervention (Katja Stefan, Kunesch, Cohen, Benecke, & Classen, 2000). In fact, we did not
detect an order effect of modulation condition, thus, supporting the effectiveness of our 30
minute wash-out period.
Limitations
The 10-20 EEG electrode layout overlays specific electrodes on assumed brain areas. We
examined changes in coherence between AF8 and C6, electrodes that are assumed to align with
the right DLPFC and M1, respectively (Klem, Lüders, Jasper, & Elger, 1999). However, without
structural MRI to confirm this alignment, we cannot be certain that for the 10 subjects who
participated in this study, that these electrodes were truly aligned with the anatomical regions of
interest for each participant. Importantly, the aim of this study targeted electrodes based on their
previously identified functional relationship with motor learning and not the underlying link with
117
brain structure (Hooyman et al., In Preparation – Chapter 3). Interestingly, recent research using
high definition TDCS found that the identified TMS hotspot of first dorsal interosseous did not
always overlap with M1 (Stephanie Lefebvre et al., 2018). This provides clear evidence that the
function of a specific motorneuron pool does not always originate in the assumed brain area. It
may be that this experiment could be attempted with structural instead of functional biomarkers
and fail replication because the identified brain structures do not mediate the desired behavior
(e.g. motor learning) targeted for modulation. A more precise way to perform iPAS would be to
identify individual functional biomarkers for each participant based on either seed-based resting-
state functional Magnetic Resonance Imaging or TMS/EEG recordings (Santarnecchi et al.,
2018). This may enhance group average modulation and reduce the number of non-responders.
Our sample size is small, however, based on our results we did achieve a power greater
near 0.8. However, we are aware that small sample studies can still achieve a statistically
significant result out of pure luck; further, we know that multiple experiments with small sample
size can have highly variable effect sizes (K. Lohse, Buchanan, & Miller, 2016). However, in the
spirit of transparency, reproducibility and open science we uploaded all our code on a github
repository (https://github.com/hooymana/PAS-EEG) for others to use. Additionally, we have a
complete methods paper currently under review, that details the entire experimental protocol
used here to encourage replication (Hooyman et al., In Review, JMLD).
Conclusion
Can Intracortical Paired Associative Stimulation be used to improve resting-state
functional connectivity? Yes, our findings show that iPAS is capable of modulating a specific rs-
IC circuit, especially if you choose an appropriate inter-stimulus interval (i.e. 5 ms). Further we
provide evidence that the mechanism by which modulation occurs appears to be through Spike
118
Timing Dependent Plasticity (STDP). These findings, once replicated provide the foundation for
future iPAS studies that aim to identify the casual relationship between rs-EEG connectivity and
motor behavior.
119
Chapter 6: Summary and Conclusion
The aim of this dissertation was to provide new evidence about the motor learning
processes engaged to acquire skill for a rule-based discovery task; where those processes are
mediated in the brain, and if neural correlates associated with these learning strategies could be
modulated. Below is a summary of the new evidence provided by chapter, the new questions that
arise and where these ideas fit within the broader field of motor skill learning.
Chapter 2: Insight into individual motor learning strategies: Questioning the existence of
the Non-Learner phenomenon.
Similar to the original discovery task research by Brooks and colleagues (Brooks et al.,
1995) we identified three sub-groups that we categorized as high, moderate and low performers.
Each group demonstrated a unique variation of the theoretical exploration to exploitation model
of learning: Low performers demonstrated low capability to explore the task space and identify
the task rule, moderate performers could identify the rule but had difficulty advancing forward
through exploitation to refine their performance, and high performers were capable of both
exploration and exploitation.
Detailed analysis demonstrated that although these groups appeared distinct, when
plotted together their individual results generated an inverted U continuum along the exploration
dimension. This indicates that perhaps with more practice each group could progress across the
continuum and eventually all become high performers. Interestingly, a post hoc survey of self-
reported experience in video games and physical activity revealed a strong positive correlation
with task performance. Therefore, performance on this de novo discovery task may not be
attributed to faults within the motor system but merely a product of limited experience with a
specific class of actions (Flanagan, Bowman, & Johansson, 2006; Karl M. Newell et al., 2009).
120
This study concludes with the idea that performance among a seemingly homogeneous
group of non-disabled adults can be highly variable between individuals. This between-subject
variability seems to be the result of a reduced capability to either properly explore or exploit the
task space for optimal learning. However, one’s capability to effectively use these movement
strategies may be linked to previous experience with tasks similar to this one. This study
provides a strong follow-up to the original Brooks’ and colleagues’ work and provides insight
into the original Failure to Learn group (Brooks et al., 1995). We provide evidence that the
alleged “non-learners” are simply amateurs with a complex motor task and should not be written
off as outliers because they do not support the original assumptions about learning.
Chapter 3: Resting Brain Connectivity Can Predict the Phases of Motor Learning
The neural substrates of discovery learning have not been well investigated. Further,
identification of the underlying neural circuitry may improve our understanding of how the brain
mediates the learning process. From our behavioral analysis described in Chapter 2, we
identified three important phases to put into our prediction models: exploration, exploitation and
retention. The exploration phase represents the early time in practice when the participants are
searching for the task rule. The exploitation phase follows exploration and signals that the task
rule is identified, and with continued refinement of the parameters of the rule (i.e., timing)
performance continues to improve. The retention phase is a test of recall of the learned motor
skill after a 24-hour delay and under no augmented feedback conditions.
The exploration phase was predicted by a bilateral cortical network predominantly
connecting frontal and motor areas. Specifically, an apparent network hub was focused on the
EEG electrode overlaying the area of the brain assumed to be the ventromedial prefrontal cortex.
121
This area has been previously associated with exploration in other motor tasks but none, to the
best of our knowledge, like that of the discovery learning task used here.
The exploitation phase was predicted by a small region of connectivity in the left
hemisphere (ipsilateral to the moving limb) between the prefrontal and motor cortex areas. This
single circuit has been predictive of motor performance in previous rs-EEG studies looking at
tasks that functioned within the parameterization/exploitation stage of motor learning. This result
supports that our whole brain analysis has external validity given its generalization with previous
neuroimaging research.
The retention phase was predicted by a bilateral network with the majority of connections
located in the hemisphere contralateral to the hand performing the task. Connectivity in this
hemisphere was between the prefrontal, motor and parietal areas of the brain. This result is
consistent with previous research that examined the neural correlates of durable learning as
measured by retention or recall tests (refs?).
Additionally, when we included self-reported experience with rs-IC to predict each
behavioral phase we found that the variance explained for each learning phase increased by
approximately 10%. This indicates that prediction of learning process can be further
strengthened through additional self-report measures that rs-IC, alone may not capture. We
recommend that future prediction studies include other relevant measures (e.g. genetic markers)
along with the brain measures.
A future direction of this research may be to incorporate genetic biomarkers as a
predictor for learning phase. Previous research has found that the Valmet66, a gene linked to
brain derived neurotrophic factor – a key protein responsible for nerve growth, is related to
motor learning capability (Kleim et al., 2006). A study that combines rs-IC, individual Valmet66
122
polymorphism and additional subject measures may not only boost prediction but also provide
viable links between genetics, brain function and behavior that have not yet been investigated.
Chapter 4: Paired Associative Stimulation Rewired: A Novel Paradigm to Modulate
Resting-state Intracortical Connectivity
The iPAS paradigm we detail here combines an innovative NIBS method to potentiate rs-
IC, and a rigorous study design with embedded controls against the confounds that previous
designs are vulnerable too (i.e. double baseline collection). The aim of this methods paper is to
provide a roadmap for implementation of the iPAS paradigm for researchers in the field of motor
learning (Hooyman et al., In Review - JMLD).
Chapter 5: Can Intracortical Paired Associative Stimulation be used to Improve Resting-
State Functional Connectivity? A Feasibility Study
The application of true iPAS demonstrated a significantly positive effect on rs-IC specific
to the targeted circuit compared to sham iPAS and baseline. Additionally, the sham iPAS did not
elicit any changes to circuit strength that was above the double baseline condition. Finally, the
study supported the STDP mechanism as only the iPAS5 condition had a positive effect on rs-IC.
This preliminary evidence indicates that iPAS can be used to strengthen rs-IC and may be an
effective targeted intervention to enhance motor learning, especially for the Failure to learn
subgroup.
However, before moving forward to a fully developed intervention study, using iPAS, it
will be important to further optimize the paradigm by establishing the best dosing parameters
including the total number of paired pulses delivered and stimulator intensity. Furthermore,
future research is needed to establish the optimal parameters of iPAS (i.e. number of pulses and
length of ISI) and whether multiple sessions of iPAS delivered overtime, and how these different
123
schedules influences brain structure (i.e. possible increases in fractional anisotropy along the
targeted circuit). This work would further confirm the effect of iPAS at both the functional and
structural level. If iPAS can be shown to effect both brain function and structure, but behavior
does not change in the expected direction, such a result would inform researchers about how the
brain learns and at the same time, significantly reshape our thinking about and approach to NIBS
interventions.
Limitations
Chapter 2 – Discovery Learning
Ideally, participant experience with video games and physical activity would be collected
prior to task performance. With the self-report data collected after the Day 2 retention,
participants may have an inaccurate recollection of their experience. However, this data was
primarily collected to explore a possible interaction between experience and task performance.
Chapter 3 – Prediction of Discovery Learning
Resting-state EEG is incapable of recording subcortical or cerebellar brain areas. With
only cortical activity measured we cannot confirm if rs-IC is truly due to the cortex
independently or if the measured connectivity is derived from a tertiary source. Additionally, we
cannot confirm the direction of information flow between the identified rs-IC although some
research supports that many connections are bi-directional (Anwar et al., 2016).
Chapter 4 & 5 – iPAS Methodology and iPAS Feasibility
Due to the lack of structural MRI we cannot confirm that the electrodes AF8 and C6
overlap brain areas of the DLPFC and M1, respectively. However, this experiment was designed
to test functional markers of discovery learning and not necessarily modulation between
predefined cortical areas. It may even be more beneficial to target functional predictors of
124
learning versus structure due to recent research demonstrating that specific brain areas do not
always underlie the specific behavior targeted for modulation (Stephanie Lefebvre et al., 2018).
The wash-out period between iPAS conditions may too short to prevent after effects from
the 1
st
condition from spreading to the next. Therefore, we counterbalanced the presentation of
the iPAS conditions among the study sample to effectively cross cancel the after effect confound.
The overall study sample was small and potentially under-powered. This increases the
chance of false positives as small studies have a greater likelihood of simply being lucky (K.
Lohse et al., 2016). We controlled this by establishing a standard error of measurement to
differentiate if changes in connectivity were due to the iPAS condition or a result of
measurement noise.
Future Direction
Discovery Learning and other forms of learning like it have gone under investigated even
though the theoretical underpinnings have existed for decades (Braun, Mehring, & Wolpert,
2010; Gentile, 1972; Miller et al., 1960; Schmidt, 1975; Wolpert, Diedrichsen, & Flanagan,
2011). Ultimately the future of discovery learning relies on future researchers to reproduce the
findings here and investigate other tasks that demonstrate similar learning processes (i.e.
exploration/exploitation) (Liu, Mayer-Kress, & Newell, 2006b; Sailer et al., 2005). Specifically,
future research should confirm if the discovery learning process resembles that of clinical
rehabilitation (Karl M. Newell & Verhoeven, 2017). This would allow for greater translation
between research done in the lab and practice performed in the clinic.
Additionally, the presence of sub-groups in Chapter 2, analysis of discovery behavior, is
a distinct yet unique feature of this experiment and the original Brooks’ work. Typically,
previous motor learning research utilizing already known tasks do not identify sub-groups before
125
or after task practice. It may be the difficulty of the discovery task that allows for the emergence
of these sub-groups, evidence by the low average success rate across all 3 sub-groups on Day 1:
High Performers = 50, Moderate Performers = 8, Low Performers = 0. Future research can
quantitatively modulate discovery task difficulty based on our previous simulation results where
we demonstrated that only 9% of all possible acceleration profiles would yield a reward tone.
Different variations of the discovery task could be utilized, with either higher or lower
percentages of simulated success, to determine if more or fewer sub-groups emerge with changes
in task difficulty. Results of this work may demonstrate how variation in task difficulty reveals
the hidden group heterogeneity within a seemingly homogenous population.
The application of PAS for the purposes of increasing communication strength between
two brain areas opens a realm of possibility on how researchers can now artificially modulate
brain dynamics. For example, one future study could look at how the consistent application of
iPAS over time changes the structure between targeted brain areas. This would further
demonstrate the effectiveness of iPAS to change the brain at both the functional and structural
level. Furthermore, it would be of great interest to test if successful changes in long term
connectivity due to iPAS cause a change in underlying motor learning. If the answer is yes, then
we have a strong neuromodulatory intervention, if the answer is no then we have new questions
that will help move the field forward.
126
Previous Experience Questionnaire
APPENDIX
Question 1: How many hours of video game experience have you had in your life? Provide a
general estimate.
Question 2: How many hours of physical activity have you participated in your lifetime? Provide
a general estimate.
127
Stimulation Comfort Survey
Participant Code
Date
Study Personnel who gave survey:
Condition
Measurement #
Pre/Post
Please rank if you are currently experiencing any of the following (Check the Box Below)
Symptom None Weak Strong
Headache
Head Pain
Neck Pain
Mood Swing
Dizziness
Blurred Vision
Other:
128
References
Anwar, A. R., Muthalib, M., Perrey, S., Galka, A., Granert, O., Wolff, S., … Methuraman, M.
(2016). Effective Connectivity of Cortical Sensorimotor Networks During Finger
Movement Tasks: A Simultaneous fNIRS, fMRI, EEG Study. Brain Topography, 29, 645–
660. http://doi.org/10.1007/s10548-016-0507-1
Babik, I., Galloway, J. C., & Lobo, M. A. (2017). Infants Born Preterm Demonstrate Impaired
Exploration of Their Bodies and Surfaces Throughout the First 2 Years of Life. Physical
Therapy, 97(9), 915–925. http://doi.org/10.1093/ptj/pzx064
Batsikadze, G., Paulus, W., Kuo, M.-F., & Nitsche, M. A. (2013). Effect of Serotonin on Paired
Associative Stimulation-Induced Plasticity in the Human Motor Cortex.
Neuropsychopharmacology, 38, 2260–2267. http://doi.org/10.1038/npp.2013.127
Bi, G. Q., & Poo, M. M. (1998). Synaptic modifications in cultured hippocampal neurons:
dependence on spike timing, synaptic strength, and postsynaptic cell type. The Journal of
Neuroscience : The Official Journal of the Society for Neuroscience, 18(24), 10464–72.
http://doi.org/10.1523/JNEUROSCI.18-24-10464.1998
Blanchard, T. C., & Gershman, S. J. (2018). Pure correlates of exploration and exploitation in the
human brain. Cognitive, Affective, & Behavioral Neuroscience, 18(1), 117–126.
http://doi.org/10.3758/s13415-017-0556-2
Braun, D. A., Aertsen, A., Wolpert, D. M., & Mehring, C. (2009). Motor Task Variation Induces
Structural Learning. Current Biology, 19(4), 352–357.
http://doi.org/10.1016/j.cub.2009.01.036
Braun, D. A., Mehring, C., & Wolpert, D. M. (2010). Structure learning in action. Behavioural
Brain Research, 206(2), 157–165. http://doi.org/10.1016/j.bbr.2009.08.031
Brooks, V., Hipperath, F., Brooks, M., Ross, H., & Freund, H. (1995). Learning What and How
in a Human Motor Task. Learning & Memory, 2(5), 225–243.
http://doi.org/10.1101/lm.2.5.225
Buckner, R. L., Andrews-Hanna, J. R., & Schacter, D. L. (2008). The brain’s default network:
anatomy, function, and relevance to disease. Annals of the New York Academy of Sciences,
1124, 1–38. http://doi.org/10.1196/annals.1440.011
Cahill, L., McGaugh, J. L., & Weinberger, N. M. (2001). The neurobiology of learning and
memory: some reminders to remember. Trends in Neurosciences, 24(10), 578–581.
http://doi.org/10.1016/S0166-2236(00)01885-3
Carter, A. R., Shulman, G. L., & Corbetta, M. (2012). Why use a connectivity-based approach to
study stroke and recovery of function? NeuroImage, 62(4), 2271–2280.
http://doi.org/10.1016/j.neuroimage.2012.02.070
Casali, A. G., Gosseries, O., Rosanova, M., Boly, M., Sarasso, S., Casali, K. R., … Massimini,
M. (2013). A Theoretically Based Index of Consciousness Independent of Sensory
Processing and Behavior. Science Translational Medicine, 5(198), 198ra105-198ra105.
129
http://doi.org/10.1126/scitranslmed.3006294
Casula, E. P., Pellicciari, M. C., Picazio, S., Caltagirone, C., & Koch, G. (2016). Spike-timing-
dependent plasticity in the human dorso-lateral prefrontal cortex. NeuroImage, 143, 204–
213. http://doi.org/10.1016/j.neuroimage.2016.08.060
Cona, G., Marino, G., & Semenza, C. (2017). TMS of supplementary motor area (SMA)
facilitates mental rotation performance: Evidence for sequence processing in SMA.
NeuroImage, 146, 770–777. http://doi.org/10.1016/J.NEUROIMAGE.2016.10.032
Deco, G., Jirsa, V. K., & McIntosh, A. R. (2011a). Emerging concepts for the dynamical
organization of resting-state activity in the brain. Nature Reviews Neuroscience, 12(1), 43–
56. http://doi.org/10.1038/nrn2961
Deco, G., Jirsa, V. K., & McIntosh, A. R. (2011b). Emerging concepts for the dynamical
organization of resting-state activity in the brain. Nature Reviews Neuroscience, 12(1), 43–
56. http://doi.org/10.1038/nrn2961
Deeny, S. P., Haufler, A. J., Saffer, M., & Hatfield, B. D. (2009). Electroencephalographic
coherence during visuomotor performance: a comparison of cortico-cortical communication
in experts and novices. Journal of Motor Behavior, 41(2), 106–116.
http://doi.org/10.3200/JMBR.41.2.106-116
Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial
EEG dynamics including independent component analysis. Journal of Neuroscience
Methods, 134(1), 9–21. http://doi.org/10.1016/J.JNEUMETH.2003.10.009
Delorme, A., Sejnowski, T., & Makeig, S. (2007). Enhanced detection of artifacts in EEG data
using higher-order statistics and independent component analysis. NeuroImage, 34(4),
1443–1449. http://doi.org/10.1016/J.NEUROIMAGE.2006.11.004
Farzan, F., Vernet, M., Shafi, M. M. D., Rotenberg, A., Daskalakis, Z. J., & Pascual-Leone, A.
(2016). Characterizing and Modulating Brain Circuitry through Transcranial Magnetic
Stimulation Combined with Electroencephalography. Frontiers in Neural Circuits, 10, 1–
24. http://doi.org/10.3389/fncir.2016.00073
Flanagan, J. R., Bowman, M. C., & Johansson, R. S. (2006). Control strategies in object
manipulation tasks. Current Opinion in Neurobiology, 16(6), 650–659.
http://doi.org/10.1016/j.conb.2006.10.005
Gentile, A. (1972). A Working Model of Skill Acquisition with Application to Teaching. Quest,
17, 3–23.
Golenia, L., Schoemaker, M. M., Mouton, L. J., & Bongers, R. M. (2014). Individual
Differences in Learning a Novel Discrete Motor Task. PLoS ONE, 9(11), 1–12.
http://doi.org/10.1371/journal.pone.0112806
Gratton, C., Lee, T. G., Nomura, E. M., & D’Esposito, M. (2013). The effect of theta-burst TMS
on cognitive control networks measured with resting state fMRI. Frontiers in Systems
Neuroscience, 7, 124. http://doi.org/10.3389/fnsys.2013.00124
Green, C. S., & Bavelier, D. (2015). Action video game training for cognitive enhancement.
130
Current Opinion in Behavioral Sciences, 4, 103–108.
http://doi.org/10.1016/J.COBEHA.2015.04.012
Guillot, A., Collet, C., Nguyen, V. A., Malouin, F., Richards, C., & Doyon, J. (2008). Functional
neuroanatomical networks associated with expertise in motor imagery. NeuroImage, 41(4),
1471–83. http://doi.org/10.1016/j.neuroimage.2008.03.042
Gustafsson, F. (1996). Determining the initial states in forward-backward filtering. IEEE
Transactions on Signal Processing, 44(4), 988–992. http://doi.org/10.1109/78.492552
Halko, M. A., Farzan, F., Eldaief, M. C., Schmahmann, J. D., & Pascual-Leone, A. (2014).
Intermittent theta-burst stimulation of the lateral cerebellum increases functional
connectivity of the default network. The Journal of Neuroscience : The Official Journal of
the Society for Neuroscience, 34(36), 12049–56. http://doi.org/10.1523/JNEUROSCI.1776-
14.2014
Hallett, M. (2007). Transcranial Magnetic Stimulation: A Primer. Neuron, 55(2), 187–199.
http://doi.org/10.1016/j.neuron.2007.06.026
Hamada, M., Galea, J. M., Di Lazzaro, V., Mazzone, P., Ziemann, U., & Rothwell, J. C. (2014).
Two distinct interneuron circuits in human motor cortex are linked to different subsets of
physiological and behavioral plasticity. The Journal of Neuroscience : The Official Journal
of the Society for Neuroscience, 34(38), 12837–49.
http://doi.org/10.1523/JNEUROSCI.1960-14.2014
Hardwick, R. M., Lesage, E., Eickhoff, C. R., Clos, M., Fox, P., & Eickhoff, S. B. (2015).
Multimodal connectivity of motor learning-related dorsal premotor cortex. NeuroImage,
123, 114–28. http://doi.org/10.1016/j.neuroimage.2015.08.024
Hartigan, J. A., & Wong, M. A. (1979). Algorithm AS 136: A K-Means Clustering Algorithm.
Applied Statistics, 28(1), 100. http://doi.org/10.2307/2346830
Harvey, R. L., Edwards, D., Dunning, K., Fregni, F., Stein, J., Laine, J., … Liu, C. Y. (2018).
Randomized Sham-Controlled Trial of Navigated Repetitive Transcranial Magnetic
Stimulation for Motor Recovery in Stroke. Stroke, 49(9), 2138–2146.
http://doi.org/10.1161/STROKEAHA.117.020607
Holland, P., Codol, O., & Galea, J. M. (2018). Contribution of explicit processes to
reinforcement-based motor learning. Journal of Neurophysiology, 119(6), 2241–2255.
http://doi.org/10.1152/jn.00901.2017
Hooyman, A., Babikian, S., Kutch, J., & Winstein, C. (2017). Is Learning Encoded in the
Resting Brain? In American Society of Neurorehabilitation.
Hordacre, B., Rogasch, N. C., & Goldsworthy, M. R. (2016). Commentary: Utility of EEG
measures of brain function in patients with acute stroke. Frontiers in Human Neuroscience,
10, 621. http://doi.org/10.3389/fnhum.2016.00621
Jarbo, K., & Verstynen, T. D. (2015). Converging structural and functional connectivity of
orbitofrontal, dorsolateral prefrontal, and posterior parietal cortex in the human striatum.
The Journal of Neuroscience : The Official Journal of the Society for Neuroscience, 35(9),
3865–78. http://doi.org/10.1523/JNEUROSCI.2636-14.2015
131
Johnen, V. M., Neubert, F.-X., Buch, E. R., Verhagen, L., O’Reilly, J. X., Mars, R. B., &
Rushworth, M. F. S. (2015). Causal manipulation of functional connectivity in a specific
neural pathway during behaviour and at rest. ELife, 4. http://doi.org/10.7554/eLife.04585
Kamke, M. R., Nydam, A. S., Sale, M. V., & Mattingley, J. B. (2016). Associative plasticity in
the human motor cortex is enhanced by concurrently targeting separate muscle
representations with excitatory and inhibitory protocols. Journal of Neurophysiology,
115(4). Retrieved from http://jn.physiology.org/content/115/4/2191
Kantak, S. S., Sullivan, K. J., Fisher, B. E., Knowlton, B. J., & Winstein, C. J. (2010). Neural
substrates of motor memory consolidation depend on practice structure. Nature
Neuroscience, 13(8), 923–925. http://doi.org/10.1038/nn.2596
Kantak, S. S., & Winstein, C. J. (2012). Learning-performance distinction and memory processes
for motor skills: a focused review and perspective. Behavioural Brain Research, 228(1),
219–31. http://doi.org/10.1016/j.bbr.2011.11.028
King, A. C., Ranganathan, R., & Newell, K. M. (2012). Individual differences in the exploration
of a redundant space-time motor task. Neuroscience Letters, 529(2), 144–9.
http://doi.org/10.1016/j.neulet.2012.08.014
Kleim, J. A., Chan, S., Pringle, E., Schallert, K., Procaccio, V., Jimenez, R., & Cramer, S. C.
(2006). BDNF val66met polymorphism is associated with modified experience-dependent
plasticity in human motor cortex. Nature Neuroscience, 9(6), 735–737.
http://doi.org/10.1038/nn1699
Klem, G. H., Lüders, H. O., Jasper, H. H., & Elger, C. (1999). The ten-twenty electrode system
of the International Federation. The International Federation of Clinical Neurophysiology.
Electroencephalography and Clinical Neurophysiology. Supplement, 52, 3–6. Retrieved
from http://www.ncbi.nlm.nih.gov/pubmed/10590970
Kobayashi, M., Hutchinson, S., Theoret, H., Schlaug, G., & Pascual-Leone, A. (2004).
Repetitive TMS of the motor cortex improves ipsilateral sequential simple finger
movements. NEUROLOGY, 62(1), 91–98. Retrieved from
http://apps.webofknowledge.com.libproxy2.usc.edu/full_record.do?product=UA&search_m
ode=OneClickSearch&qid=120&SID=Y2zyzF8siGazIpSmEg3&page=1&doc=48
Koch, G., Ponzo, V., Di Lorenzo, F., Caltagirone, C., & Veniero, D. (2013). Hebbian and Anti-
Hebbian Spike-Timing-Dependent Plasticity of Human Cortico-Cortical Connections.
Journal of Neuroscience, 33(23). Retrieved from
http://www.jneurosci.org.libproxy2.usc.edu/content/33/23/9725/tab-figures-data
Koo, T. K., & Li, M. Y. (2016). A Guideline of Selecting and Reporting Intraclass Correlation
Coefficients for Reliability Research. Journal of Chiropractic Medicine, 15(2), 155–63.
http://doi.org/10.1016/j.jcm.2016.02.012
Krakauer, J. W., & Mazzoni, P. (2011). Human sensorimotor learning: adaptation, skill, and
beyond. Current Opinion in Neurobiology, 21(4), 636–644.
http://doi.org/10.1016/J.CONB.2011.06.012
Krause, V., Meier, A., Dinkelbach, L., & Pollok, B. (2016). Beta Band Transcranial Alternating
132
(tACS) and Direct Current Stimulation (tDCS) Applied After Initial Learning Facilitate
Retrieval of a Motor Sequence. Frontiers in Behavioral Neuroscience, 10, 4.
http://doi.org/10.3389/fnbeh.2016.00004
Kutch, J. J., Labus, J. S., Harris, R. E., Martucci, K. T., Farmer, M. A., Fenske, S., … Mayer, E.
A. (2017). Resting-state functional connectivity predicts longitudinal pain symptom change
in urologic chronic pelvic pain syndrome. PAIN, 158(6), 1069–1082.
http://doi.org/10.1097/j.pain.0000000000000886
Lee, T. D., & Genovese, E. D. (1988). Distribution of Practice in Motor Skill Acquisition:
Learning and Performance Effects Reconsidered. Research Quarterly for Exercise and
Sport, 59(4), 277–287. http://doi.org/10.1080/02701367.1988.10609373
Lee, T. D., Wulf, G., & Schmidt, R. A. (1992). Contextual Interference in Motor Learning:
Dissociated Effects Due to the Nature of Task Variations. The Quarterly Journal of
Experimental Psychology Section A, 44(4), 627–644.
http://doi.org/10.1080/14640749208401303
Lefebvre, S., Dricot, L., Gradkowski, W., Laloux, P., & Vandermeeren, Y. (2012). Brain
activations underlying different patterns of performance improvement during early motor
skill learning. NeuroImage, 62(1), 290–9. http://doi.org/10.1016/j.neuroimage.2012.04.052
Lefebvre, S., Dricot, L., Laloux, P., Gradkowski, W., Desfontaines, P., Evrard, F., …
Vandermeeren, Y. (2015). Neural substrates underlying motor skill learning in chronic
hemiparetic stroke patients. Frontiers in Human Neuroscience, 9, 320.
http://doi.org/10.3389/fnhum.2015.00320
Lefebvre, S., Jann, K., Schmiesing, A., Ito, K., Jog, M., Qiao, Y., … Liew, S.-L. (2018). High-
definition transcranial direct current stimulation over the dorsal premotor cortex reliably
modulates motor network excitability. BioRxiv, 487488. http://doi.org/10.1101/487488
Levy, R. M., Harvey, R. L., Kissela, B. M., Winstein, C. J., Lutsep, H. L., Parrish, T. B., …
Venkatesan, L. (2016). Epidural Electrical Stimulation for Stroke Rehabilitation: Results of
the Prospective, Multicenter, Randomized, Single-Blinded Everest Trial.
Neurorehabilitation and Neural Repair, 30(2), 107–119.
http://doi.org/10.1177/1545968315575613
Lewis, C. M., Baldassarre, A., Committeri, G., Romani, G. L., & Corbetta, M. (2009a). Learning
sculpts the spontaneous activity of the resting human brain. Proceedings of the National
Academy of Sciences, 106(41), 17558–17563. http://doi.org/10.1073/pnas.0902455106
Lewis, C. M., Baldassarre, A., Committeri, G., Romani, G. L., & Corbetta, M. (2009b). Learning
sculpts the spontaneous activity of the resting human brain. Proceedings of the National
Academy of Sciences, 106(41), 17558–17563. http://doi.org/10.1073/pnas.0902455106
Lin, C.-H. J., Chiang, M.-C., Knowlton, B. J., Iacoboni, M., Udompholkul, P., & Wu, A. D.
(2013). Interleaved practice enhances skill learning and the functional connectivity of
fronto-parietal networks. Human Brain Mapping, 34(7), 1542–58.
http://doi.org/10.1002/hbm.22009
Lissek, S., Vallana, G. S., & Guentuerkuen, O. (2013). Brain Activation in Motor Sequence
133
Learning Is Related to the Level of Native Cortical Excitability. PloS One, 8(4), e61863.
Retrieved from
http://apps.isiknowledge.com/full_record.do?product=UA&search_mode=GeneralSearch&
qid=4&SID=3BsXRSOh87V3OXMBeqE&page=1&doc=50
Liu, Y.-T., Mayer-Kress, G., & Newell, K. M. (2006a). Qualitative and quantitative change in
the dynamics of motor learning. Journal of Experimental Psychology: Human Perception
and Performance, 32(2), 380–393. http://doi.org/10.1037/0096-1523.32.2.380
Liu, Y.-T., Mayer-Kress, G., & Newell, K. M. (2006b). Qualitative and quantitative change in
the dynamics of motor learning. Journal of Experimental Psychology: Human Perception
and Performance, 32(2), 380–393. http://doi.org/10.1037/0096-1523.32.2.380
Lohse, K., Buchanan, T., & Miller, M. (2016). Underpowered and Overworked: Problems With
Data Analysis in Motor Learning Studies. Journal of Motor Learning and Development,
4(1), 37–58. http://doi.org/10.1123/jmld.2015-0010
Lohse, K. R., Wadden, K., Boyd, L. a., & Hodges, N. J. (2014). Motor skill acquisition across
short and long time scales: A meta-analysis of neuroimaging data. Neuropsychologia, 59(1),
130–141. http://doi.org/10.1016/j.neuropsychologia.2014.05.001
Manuel, A. L., Guggisberg, A. G., Thézé, R., Turri, F., & Schnider, A. (2018). Resting-state
connectivity predicts visuo-motor skill learning. NeuroImage, 176, 446–453.
http://doi.org/10.1016/j.neuroimage.2018.05.003
Mary, A., Wens, V., Op de Beeck, M., Leproult, R., De Tiège, X., & Peigneux, P. (2017). Age-
related differences in practice-dependent resting-state functional connectivity related to
motor sequence learning. Human Brain Mapping, 38(2), 923–937.
http://doi.org/10.1002/hbm.23428
Massimini, M., Huber, R., Ferrarelli, F., Hill, S., & Tononi, G. (2004). The Sleep Slow
Oscillation as a Traveling Wave. Journal of Neuroscience, 24(31), 6862–6870.
http://doi.org/10.1523/JNEUROSCI.1318-04.2004
McHughen, S. A., Rodriguez, P. F., Kleim, J. A., Kleim, E. D., Marchal Crespo, L., Procaccio,
V., & Cramer, S. C. (2010). BDNF val66met polymorphism influences motor system
function in the human brain. Cerebral Cortex (New York, N.Y. : 1991), 20(5), 1254–62.
http://doi.org/10.1093/cercor/bhp189
Mehrkanoon, S., Boonstra, T. W., Breakspear, M., Hinder, M., & Summers, J. J. (2016a).
Upregulation of cortico-cerebellar functional connectivity after motor learning.
NeuroImage, 128, 252–263. http://doi.org/10.1016/j.neuroimage.2015.12.052
Mehrkanoon, S., Boonstra, T. W., Breakspear, M., Hinder, M., & Summers, J. J. (2016b).
Upregulation of cortico-cerebellar functional connectivity after motor learning.
NeuroImage, 128, 252–263. http://doi.org/10.1016/j.neuroimage.2015.12.052
Miller, G., Galanter, E., & Pribram, K. (1960). Plans and the Structure of Behavior. New York:
Hold, Rinehart and Winston.
Müller-Dahlhaus, J. F. M., Orekhov, Y., Liu, Y., & Ziemann, U. (2008). Interindividual
variability and age-dependency of motor cortical plasticity induced by paired associative
134
stimulation. Experimental Brain Research, 187(3), 467–475. http://doi.org/10.1007/s00221-
008-1319-7
Nettekoven, C., Volz, L. J., Leimbach, M., Pool, E.-M., Rehme, A. K., Eickhoff, S. B., …
Grefkes, C. (2015). Inter-individual variability in cortical excitability and motor network
connectivity following multiple blocks of rTMS. NeuroImage, 118, 209–18.
http://doi.org/10.1016/j.neuroimage.2015.06.004
Newell, K. M. (1991). Motor skill acquisition. Annual Review of Psychology, 42, 213–37.
http://doi.org/10.1146/annurev.ps.42.020191.001241
Newell, K. M. (1991). Motor Skill Acquisition. Annual Review of Psychology, 42(1), 213–237.
http://doi.org/10.1146/annurev.ps.42.020191.001241
Newell, K. M., Liu, Y.-T., & Mayer-Kress, G. (2009). Time Scales, Difficulty/Skill Duality, and
the Dynamics of Motor Learning. Advances in Experimental Medicine and Biology, 629,
457–476. http://doi.org/10.1007/978-0-387-77064-2_24
Newell, K. M., & Verhoeven, F. M. (2017). Movement rehabilitation: are the principles of re-
learning in the recovery of function the same as those of original learning? Disability and
Rehabilitation, 39(2), 121–126. http://doi.org/10.3109/09638288.2016.1170895
Nunez, P. L., & Srinivasan, R. (2014). Neocortical dynamics due to axon propagation delays in
cortico-cortical fibers: EEG traveling and standing waves with implications for top-down
influences on local networks and white matter disease. Brain Research, 1542, 138–166.
http://doi.org/10.1016/j.brainres.2013.10.036
Oldfield, R. C. (1971). The assessment and analysis of handedness: The Edinburgh inventory.
Neuropsychologia, 9(1), 97–113. http://doi.org/10.1016/0028-3932(71)90067-4
Pacheco, M. M., & Newell, K. M. (2015). Transfer as a function of exploration and stabilization
in original practice. Human Movement Science, 44, 258–269.
http://doi.org/10.1016/j.humov.2015.09.009
Palmer, J. A., Wolf, S. L., & Borich, M. R. (2018). Paired associative stimulation modulates
corticomotor excitability in chronic stroke: A preliminary investigation. Restorative
Neurology and Neuroscience, 36(2), 183–194. http://doi.org/10.3233/RNN-170785
Pfurtscheller, G., Stancák, A., & Neuper, C. (1996). Post-movement beta synchronization. A
correlate of an idling motor area? Electroencephalography and Clinical Neurophysiology,
98(4), 281–293. http://doi.org/10.1016/0013-4694(95)00258-8
Pruitt, D. T., Schmid, A. N., Danaphongse, T. T., Flanagan, K. E., Morrison, R. A., Kilgard, M.
P., … Hays, S. A. (2016). Forelimb training drives transient map reorganization in
ipsilateral motor cortex. Behavioural Brain Research, 313, 10–16.
http://doi.org/10.1016/j.bbr.2016.07.005
Ranganathan, R. (2017). Reorganization of finger coordination patterns through motor
exploration in individuals after stroke. Journal of NeuroEngineering and Rehabilitation,
14(1), 90. http://doi.org/10.1186/s12984-017-0300-8
Rizzo, V., Siebner, H. S., Morgante, F., Mastroeni, C., Girlanda, P., & Quartarone, A. (2009).
135
Paired Associative Stimulation of Left and Right Human Motor Cortex Shapes
Interhemispheric Motor Inhibition based on a Hebbian Mechanism. Cerebral Cortex, 19(4),
907–915. http://doi.org/10.1093/cercor/bhn144
Roopun, A. K., Middleton, S. J., Cunningham, M. O., LeBeau, F. E. N., Bibbig, A., Whittington,
M. A., & Traub, R. D. (2006). A beta2-frequency (20-30 Hz) oscillation in nonsynaptic
networks of somatosensory cortex. Proceedings of the National Academy of Sciences of the
United States of America, 103(42), 15646–50. http://doi.org/10.1073/pnas.0607443103
Rossini, P. M., Burke, D., Chen, R., Cohen, L. G., Daskalakis, Z., Di Iorio, R., … Ziemann, U.
(2015). Non-invasive electrical and magnetic stimulation of the brain, spinal cord, roots and
peripheral nerves: Basic principles and procedures for routine clinical and research
application: An updated report from an I.F.C.N. Committee. Clinical Neurophysiology,
126(6), 1071–1107. http://doi.org/10.1016/j.clinph.2015.02.001
Rroji, O., van Kuyck, K., Nuttin, B., & Wenderoth, N. (2015). Anodal tDCS over the Primary
Motor Cortex Facilitates Long-Term Memory Formation Reflecting Use-Dependent
Plasticity. PloS One, 10(5), e0127270. http://doi.org/10.1371/journal.pone.0127270
Sailer, U., Flanagan, S., & Johanssen, S. (2005). Eye-Hand Coordination during Learning of a
Novel Visuomotor Task. Journal of Neuroscience, 25(39), 8833–8842.
http://doi.org/10.1523/JNEUROSCI.2658-05.2005
Sami, S., Robertson, E. M., & Miall, R. C. (2014). The time course of task-specific memory
consolidation effects in resting state networks. J Neurosci, 34(11).
http://doi.org/10.1523/JNEUROSCI.4341-13.2014
Santarnecchi, E., Momi, D., Sprugnoli, G., Neri, F., Pascual-Leone, A., Rossi, A., & Rossi, S.
(2018). Modulation of network-to-network connectivity via spike-timing-dependent
noninvasive brain stimulation. Human Brain Mapping, 39(12), 4870–4883.
http://doi.org/10.1002/hbm.24329
Schmidt, R. A. (1975). A Schema Theory of Discrete Motor Skill Learning. Psychological
Review, 82(4), 225–260. Retrieved from http://www.hp-
research.com/sites/default/files/publications/Schmidt (Psych Review, 1975).pdf
Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., … Greicius,
M. D. (2007). Dissociable intrinsic connectivity networks for salience processing and
executive control. The Journal of Neuroscience : The Official Journal of the Society for
Neuroscience, 27(9), 2349–56. http://doi.org/10.1523/JNEUROSCI.5587-06.2007
Seidler, R. D. (2007). Older adults can learn to learn new motor skills. Behavioural Brain
Research, 183(1), 118–122. http://doi.org/10.1016/j.bbr.2007.05.024
Seidler, R. D. (2010). Neural Correlates of Motor Learning, Transfer of Learning, and Learning
to Learn. Exercise and Sport Sciences Reviews, 38(1), 3–9.
http://doi.org/10.1097/JES.0b013e3181c5cce7
Seidler, R., Erdeniz, B., Koppelmans, V., Hirsiger, S., Mérillat, S., & Jäncke, L. (2015).
Associations between age, motor function, and resting state sensorimotor network
connectivity in healthy older adults. NeuroImage, 108, 47–59.
136
http://doi.org/10.1016/j.neuroimage.2014.12.023
Serrien, D. J., Ivry, R. B., & Swinnen, S. P. (2006). Dynamics of hemispheric specialization and
integration in the context of motor control. Nature Reviews Neuroscience, 7(2), 160–166.
http://doi.org/10.1038/nrn1849
Shah, L. M., Cramer, J. A., Ferguson, M. A., Birn, R. M., & Anderson, J. S. (2016). Reliability
and reproducibility of individual differences in functional connectivity acquired during task
and resting state. Brain and Behavior, 6(5). http://doi.org/10.1002/brb3.456
Sidarta, A., Vahdat, S., Bernardi, N. F., & Ostry, D. J. (2016). Somatic and Reinforcement-Based
Plasticity in the Initial Stages of Human Motor Learning. Journal of Neuroscience, 36(46),
11682–11692. http://doi.org/10.1523/JNEUROSCI.1767-16.2016
Siegel, J. S., Ramsey, L. E., Snyder, A. Z., Metcalf, N. V, Chacko, R. V, Weinberger, K., …
Corbetta, M. (2016). Disruptions of network connectivity predict impairment in multiple
behavioral domains after stroke. Proceedings of the National Academy of Sciences of the
United States of America, 113(30), E4367-76. http://doi.org/10.1073/pnas.1521083113
Singh, P., Jana, S., Ghosal, A., & Murthy, A. (2016). Exploration of joint redundancy but not
task space variability facilitates supervised motor learning. Proceedings of the National
Academy of Sciences, 113(50), 14414–14419. http://doi.org/10.1073/pnas.1613383113
Srinivasan, R., Winter, W. R., Ding, J., & Nunez, P. L. (2007). EEG and MEG coherence:
Measures of functional connectivity at distinct spatial scales of neocortical dynamics.
Journal of Neuroscience Methods, 166(1), 41–52.
http://doi.org/10.1016/j.jneumeth.2007.06.026
Stefan, K., Kunesch, E., Cohen, L., Benecke, R., & Classen, J. (2000). Induction of plasticity in
the human motor cortex by paired associative stimulation. Brain, 123(3), 572–584.
http://doi.org/10.1093/brain/123.3.572
Stefan, K., Kunesch, E., Cohen, L. G., Benecke, R., & Classen, J. (2000). Induction of plasticity
in the human motor cortex by paired associative stimulation. Brain, 123, 572–584.
Retrieved from http://www.tmslab.org/tms articles research/022.pdf
Storey, J. D. (2002). A direct approach to false discovery rates. Journal of the Royal Statistical
Society, 64(3), 479–498. Retrieved from
http://genomics.princeton.edu/storeylab/papers/directfdr.pdf
Strube, W., Bunse, T., Malchow, B., & Hasan, A. (2015). Efficacy and Interindividual
Variability in Motor-Cortex Plasticity following Anodal tDCS and Paired-Associative
Stimulation. Neural Plasticity, 2015, 1–10. http://doi.org/10.1155/2015/530423
Sutton, R. S., & Barto, A. G. (1998). Reinforcement Learning: An Introduction (1st Editio).
Cambridge: Spring Science & Business Media. Retrieved from
http://people.inf.elte.hu/lorincz/Files/RL_2006/SuttonBook.pdf
Therrien, A. S., Wolpert, D. M., & Bastian, A. J. (2016). Effective reinforcement learning
following cerebellar damage requires a balance between exploration and motor noise.
Brain : A Journal of Neurology, 139(Pt 1), 101–14. http://doi.org/10.1093/brain/awv329
137
Tomassini, V., Jbabdi, S., Kincses, Z. T., Bosnell, R., Douaud, G., Pozzilli, C., … Johansen-
Berg, H. (2011). Structural and functional bases for individual differences in motor
learning. Human Brain Mapping, 32(3), 494–508. http://doi.org/10.1002/hbm.21037
Tzvi, E., Zimmermann, C., Bey, R., Münte, T. F., Nitschke, M., & Krämer, U. M. (2017).
Cerebellar degeneration affects cortico-cortical connectivity in motor learning networks.
NeuroImage: Clinical, 16, 66–78. http://doi.org/10.1016/j.nicl.2017.07.012
van den Heuvel, M. P., Mandl, R. C. W., Kahn, R. S., & Hulshoff Pol, H. E. (2009). Functionally
linked resting-state networks reflect the underlying structural connectivity architecture of
the human brain. Human Brain Mapping, 30(10), 3127–3141.
http://doi.org/10.1002/hbm.20737
Veniero, D., Ponzo, V., & Koch, G. (2013). Paired Associative Stimulation Enforces the
Communication between Interconnected Areas. Journal of Neuroscience, 33(34), 13773–
13783. http://doi.org/10.1523/JNEUROSCI.1777-13.2013
Wang, J. X., Rogers, L. M., Gross, E. Z., Ryals, A. J., Dokucu, M. E., Brandstatt, K. L., … Voss,
J. L. (2014). Targeted enhancement of cortical-hippocampal brain networks and associative
memory. Science (New York, N.Y.), 345(6200), 1054–7.
http://doi.org/10.1126/science.1252900
Waters, S., Wiestler, T., & Diedrichsen, J. (2017). Cooperation Not Competition: Bihemispheric
tDCS and fMRI Show Role for Ipsilateral Hemisphere in Motor Learning. The Journal of
Neuroscience, 37(31), 7500–7512. http://doi.org/10.1523/JNEUROSCI.3414-16.2017
Wiethoff, S., Hamada, M., & Rothwell, J. C. (2014). Variability in Response to Transcranial
Direct Current Stimulation of the Motor Cortex. Brain Stimulation, 7(3), 468–475.
http://doi.org/10.1016/j.brs.2014.02.003
Wischnewski, M., Kowalski, G. M., Rink, F., Belagaje, S. R., Haut, M. W., Hobbs, G., &
Buetefisch, C. M. (2016). Demand on skillfulness modulates interhemispheric inhibition of
motor cortices. Journal of Neurophysiology, 115(6), 2803–13.
http://doi.org/10.1152/jn.01076.2015
Wolpert, D. M., Diedrichsen, J., & Flanagan, J. R. (2011). Principles of sensorimotor learning.
Nature Reviews Neuroscience, 12(12), 739. http://doi.org/10.1038/nrn3112
Wu, J., Knapp, F., Cramer, S. C., & Srinivasan, R. (2018). Electroencephalographic connectivity
measures predict learning of a motor sequencing task. Journal of Neurophysiology, 119(2),
490–498. http://doi.org/10.1152/jn.00580.2017
Wu, J., Quinlan, E. B., Dodakian, L., McKenzie, A., Kathuria, N., Zhou, R. J., … Cramer, S. C.
(2015). Connectivity measures are robust biomarkers of cortical function and plasticity after
stroke. Brain, 138(8), 2359–2369. http://doi.org/10.1093/brain/awv156
Wu, J., Srinivasan, R., Kaur, A., & Cramer, S. C. (2014). Resting-state cortical connectivity
predicts motor skill acquisition. NeuroImage, 91, 84–90.
http://doi.org/10.1016/j.neuroimage.2014.01.026
Wulf, G., & Lewthwaite, R. (2016). Optimizing performance through intrinsic motivation and
attention for learning: The OPTIMAL theory of motor learning. Psychonomic Bulletin &
138
Review. http://doi.org/10.3758/s13423-015-0999-9
Wulf, G., Shea, C., & Lewthwaite, R. (2010). Motor skill learning and performance: a review of
influential factors. Medical Education, 44(1), 75–84. http://doi.org/10.1111/j.1365-
2923.2009.03421.x
Zou, Q., Ross, T. J., Gu, H., Geng, X., Zuo, X.-N., Hong, L. E., … Yang, Y. (2013). Intrinsic
resting-state activity predicts working memory brain activation and behavioral performance.
Human Brain Mapping, 34(12), 3204–3215. http://doi.org/10.1002/hbm.22136
Abstract (if available)
Abstract
This dissertation has the following three aims: Aim 1) Examine how individual participants learn a rule-based motor task where success is contingent upon rule discovery. Aim 2) Identify the cortical substrates from resting-state EEG that predict the processes of discovery learning. Aim 3) Test the feasibility of an intervention that uses non-invasive brain stimulation to potentiate a putative neural circuit known to be engaged in the discovery learning process. Together, results from this dissertation provide new evidence of an understudied area, the process of motor learning (what is learned?) and not just the outcome
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Relationship between brain structure and motor behavior in chronic stroke survivors
PDF
Neural substrates associated with context-dependent learning
PDF
Contextual interference in motor skill learning: an investigation of the practice schedule effect using transcranial magnetic stimulation (TMS)
PDF
Relationship between interhemispheric inhibition and bimanual coordination in musicians
PDF
Brain and behavior correlates of intrinsic motivation and skill learning
PDF
Impact of enhanced efficacy expectation on motor learning in individuals with Parkinson’s disease
PDF
Learning self-agency during a contingency paradigm in infants at commnity and elevated risk of autism
PDF
Understanding associations between whole-body dynamic balance and spatiotemporal asymmetry during healthy and pathological gait
PDF
Corticomotor excitability of gluteus maximus: influence on hip extensor strength and hip mechanics
PDF
Hemisphere-specific deficits in the control of bimanual movements after stroke
PDF
Locomotor skill learning in virtual reality in healthy adults and people with Parkinson disease
PDF
Learning reaching skills in non-disabled and post-stroke individuals
PDF
Neural substrates of motor memory consolidation: a double dissociation of primary motor cortex and dorsolateral prefrontal cortex induced by practice structure
PDF
Virtual surgeries as a tool for studying motor learning
PDF
Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke
PDF
Motor cortical representations of sensorimotor information during skill learning
PDF
Computational model of stroke therapy and long term recovery
PDF
Cortical inputs to leg muscles in increasingly difficult standing tasks
PDF
Computational principles in human motor adaptation: sources, memories, and variability
PDF
Behavioral and neurophysiological studies of hand dexterity in health and Parkinson's disease
Asset Metadata
Creator
Hooyman, Andrew McMahon
(author)
Core Title
The brain and behavior of motor learning: the what, how and where
School
School of Dentistry
Degree
Doctor of Philosophy
Degree Program
Biokinesiology
Publication Date
06/28/2019
Defense Date
05/01/2019
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
discovery learning,electroencephalography,exploration,machine learning,motor learning,OAI-PMH Harvest,paired associative stimulation,transcranial magnetic stimulation
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Winstein, Carolee (
committee chair
), Fisher, Beth (
committee member
), Gordon, James (
committee member
), Kutch, Jason (
committee member
), Schweighofer, Nicolas (
committee member
)
Creator Email
hooyman.andrew@gmail.com,hooyman@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-179037
Unique identifier
UC11660423
Identifier
etd-HooymanAnd-7522.pdf (filename),usctheses-c89-179037 (legacy record id)
Legacy Identifier
etd-HooymanAnd-7522.pdf
Dmrecord
179037
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Hooyman, Andrew McMahon
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
discovery learning
electroencephalography
exploration
machine learning
motor learning
paired associative stimulation
transcranial magnetic stimulation