THE BEHAVIOURAL AND NEURAL SCIENCE OF MOTOR SKILL LEARNING IN HEALTHY INDIVIDUALS AND PEOPLE WITH STROKE by Katie P. Wadden BKin, Memorial University of Newfoundland, 2008 MSc (Kin.), Memorial University of Newfoundland, 2010 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Rehabilitation Sciences) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) July 2017 © Katie P. Wadden, 2017 ii Abstract Background Due to a high occurrence of motor impairment following stroke, motor learning is fundamentally important for stroke rehabilitation. Motor learning interventions can be difficult to deliver by clinicians and researchers due to individual differences in motor abilities that are compounded by neural changes associated with an ischemic insult. To enhance motor learning, interventions must be grounded in understanding patterns of performance change, both behaviourally and neurologically. Methods In Chapters 2 and 3, performance data were fitted to exponential curves to measure performance change during skill acquisition. In Chapter 2, curve parameters were used to create an individualized learner-adapted algorithm to manipulate the level of difficulty of motor practice conditions. In Chapter 3, curve parameters were used to measure the rate of skill acquisition in healthy individuals and stroke populations. Chapter 4 explored the functional connectivity of motor-related brain networks before and after skill acquisition in healthy individuals and individuals with stroke. In Chapter 5, residual white matter in a motor-related brain network was used to examine individual responses following an intervention combining non-invasive brain stimulation and motor skill practice in individuals with stroke. Findings In Chapter 2, skill acquisition under a learner-adapted algorithm, developed from curve parameters, showed that higher individualized levels of difficulty in practice were better for skill retention. In Chapter 3, individuals with stroke showed a slower rate of skill acquisition compared to healthy individuals, which was associated with worse motor performance-related iii change at a delayed retention test. In Chapter 4, individuals with stroke did not activate motor learning–related functional brain networks in the same manner as healthy individuals following motor skill practice. In Chapter 5, the integrity of white matter in the motor-related brain network was higher in individuals who positively responded to the intervention. Conclusions The findings from this dissertation highlight the importance of modelling performance data to advance the evaluation of stroke rehabilitative interventions. This dissertation contributes new knowledge of a gray matter motor network associated with motor learning, and a white matter motor network biomarker that characterizes the response to non-invasive brain stimulation paired with motor practice in individuals with stroke. iv Lay Summary Putting a golf ball or shooting a basketball are examples of skills; they involve performing movements to achieve a goal. Following a stroke, damage to the brain causes individuals to lose the ability to perform skills. Learning new skills, or relearning old skills, is the key ingredient to stroke rehabilitation. The present dissertation investigated behavioural and brain measures to characterize how individuals with stroke learn motor skills. First, I investigated the speed at which individuals with stroke learned a new motor skill. Then I evaluated areas of the brain individuals with stroke used to learn a new motor skill. Lastly, I looked at the health of brain connections to determine if they influenced the ability of individuals with stroke to learn a new motor skill. From the results of this dissertation, I believe encouraging clinicians and researchers to evaluate and deliver individualized motor learning interventions will improve brain and hand function in people with stroke. v Preface The work in this dissertation was conceived, conducted, and written by Katie Wadden. The research described in this dissertation was approved by the University of British Columbia’s (UBC) Clinical Research Ethics Board: H13-02436, H06-70351, and H09-00368. Chapters 1 and 6 were written by Katie Wadden. Drs. Lara Boyd, Nicola Hodges, and Todd Woodward assisted in editing these chapters. Chapter 2 is based on work conducted by Katie Wadden, Dr. Lara Boyd, Dr. Nicola Hodges, Kristopher De Asis, and Dr. Jason Neva. Katie Wadden was responsible for study conception and design, data collection, analyses and interpretation, and writing and revising the manuscript. Drs. Lara Boyd and Nicola Hodges assisted in designing the study, interpreting the data, and editing the manuscript. Kristopher De Asis assisted in data analysis and interpretation. Dr. Jason Neva assisted in data interpretation and editing the manuscript. Chapter 3 is based on work conducted by Katie Wadden, Drs. Lara Boyd, Bimal Lakhani, Cameron Mang, Jason Neva, Sean Meehan, and Bubblepreet Randhawa, and Sue Peters, Kristopher De Asis, and Brenda Wessel. Katie Wadden and Dr. Boyd were jointly responsible for study design. In addition, Katie Wadden was responsible for data analyses and interpretation, and writing and revising the manuscript. Dr. Boyd assisted in data analysis and interpretation, and manuscript editing. Kristopher De Asis assisted in data analysis and interpretation. Drs. Lakhani, Neva and Mang, and Sue Peters assisted in data interpretation and editing the manuscript. Drs. Meehan and Randhawa, and Brenda Wessel were responsible for data collection. Chapter 4 is based on work conducted by Katie Wadden, Dr. Lara Boyd, Dr. Todd Woodward, Paul Metzak, Katie Lavigne, Dr. Bimal Lahkani, Dr. Angela Auriat, Brenda Wessel, vi Dr. Bubblepreet Randhawa, and Dr. Sean Meeehan. Katie Wadden and Dr. Boyd were jointly responsible for study design. Katie Wadden was responsible for magnetic resonance imaging data analyses and interpretation, and writing and revising the manuscript. Dr. Boyd was responsible for the study design, and assisted in data collection, analysis and interpretation, and editing the manuscript. Dr. Woodward was responsible for data interpretation and editing the manuscript. Paul Metzak and Katie Lavigne, as well as Drs. Lakhani and Auriat, were responsible for data interpretation and editing the manuscript. Drs. Meehan and Randhawa, and Brenda Wessel were responsible for data collection. Chapter 5 is based on work conducted by Katie Wadden, Drs. Lara Boyd, Sean Meehan, Michael Borich, Cameron Mang, and Kate Hayward; Sue Peters, Katlyn Brown, Nicholas Snow, and Dr. Jason Neva. Katie Wadden, and Drs. Borich, Meehan, and Boyd were jointly responsible for the study design. Katie Wadden was responsible for the data collection, analyses, interpretation, and writing and revising the manuscript. Dr. Boyd was responsible for the study design, and assisted in data interpretation and editing the manuscript. Dr. Sean Meehan was responsible for study design and editing the manuscript. Drs. Borich and Mang, along with Katlyn Brown and Nicholas Snow, were responsible for data collection and editing the manuscript. Drs. Lakhani, Hayward, and Neva were responsible for data interpretation and editing the manuscript. Sue Peters was responsible for data analysis and interpretation, and editing the manuscript. A version of Chapter 3 was published in 2017 as “Predicting motor sequence learning in individuals with chronic stroke.” See the Bibliography for a full citation. vii A version of Chapter 4 was published in 2015 as “Compensatory motor network connectivity is associated with motor sequence learning after subcortical stroke.” See the Bibliography for a full citation. viii Table of Contents Abstract .......................................................................................................................................... ii Lay Summary ............................................................................................................................... iv Preface ............................................................................................................................................ v Table of Contents ....................................................................................................................... viii List of Figures ............................................................................................................................... xi List of Tables .............................................................................................................................. xiv List of Abbreviations .................................................................................................................. xv Acknowledgements ................................................................................................................... xvii Dedication ................................................................................................................................... xix Chapter 1: Introduction ..........................................................................................................................1 Preamble ................................................................................................................................................1 Research objectives ...............................................................................................................................6 Proposed outcomes ................................................................................................................................7 Motor learning .....................................................................................................................................10 Phases of motor learning .....................................................................................................................11 Factors that influence motor learning .................................................................................................14 Stroke and motor rehabilitation ...........................................................................................................22 Behavioural assessments of motor learning ........................................................................................27 Neurophysiology of motor learning ....................................................................................................39 ix Neurophysiology assessments of motor learning ................................................................................47 Adjunct therapies to facilitate motor recovery and learning ...............................................................56 Aims of this thesis ...............................................................................................................................62 Chapter 2: Individualized challenge point practice informs motor sequence learning: More time in an early phase of practice benefits later retention .........................................................................66 Introduction .........................................................................................................................................66 Study 1 — Methods ............................................................................................................................71 Statistical analysis ...............................................................................................................................80 Results .................................................................................................................................................81 Discussion ...........................................................................................................................................88 Study 2 — Methods ............................................................................................................................89 Statistical analysis ...............................................................................................................................93 Results .................................................................................................................................................94 General discussion ............................................................................................................................101 Chapter 3: Predicting motor performance-related change in individuals with chronic stroke ...108 Introduction .......................................................................................................................................108 Methods .............................................................................................................................................112 Statistical analysis .............................................................................................................................119 Results ...............................................................................................................................................123 Discussion .........................................................................................................................................130 Chapter 4: Compensatory motor network connectivity: motor sequence learning after subcortical stroke .................................................................................................................................139 Introduction .......................................................................................................................................139 Methods .............................................................................................................................................143 Imaging protocol ...............................................................................................................................147 x fMRI data analysis ............................................................................................................................148 Results ...............................................................................................................................................153 Discussion .........................................................................................................................................160 Conclusion .........................................................................................................................................168 Chapter 5: White matter biomarkers associated with motor change in individuals with stroke: A continuous theta burst stimulation study ..........................................................................................170 Introduction .......................................................................................................................................170 Methods .............................................................................................................................................175 Statistical analyses ............................................................................................................................186 Results ...............................................................................................................................................192 Discussion .........................................................................................................................................201 Chapter 6: General discussion ...........................................................................................................208 Overview ...........................................................................................................................................208 Challenge in practice relates to enhanced motor learning for young, healthy adults .......................211 Rate of motor skill acquisition is associated with motor performance–related change in individuals with stroke .........................................................................................................................................215 Functionally connected motor learning–related network ..................................................................220 Biomarkers for motor learning in chronic stroke ..............................................................................222 Limitations ........................................................................................................................................225 Implications and future directions .....................................................................................................228 Conclusion .........................................................................................................................................230 Bibliography .........................................................................................................................................232 xi List of Figures Figure 1: Neuroimaging factors. .................................................................................................... 5 Figure 2: General aims of thesis. ................................................................................................... 6 Figure 3: The contextual interference (CI) effect. ....................................................................... 18 Figure 4: Illustration of motor learning tasks. ............................................................................. 29 Figure 5: Phases of motor learning. ............................................................................................. 37 Figure 6: Power versus exponential curves. ................................................................................ 37 Figure 7: Neurophysiology of motor learning. ............................................................................ 42 Figure 8a and b: The discrete pairing task (DPT). ...................................................................... 73 Figure 9: Outline of design for Studies 1 and 2. .......................................................................... 76 Figure 10: The phases of skill acquisition based on performance-resource function. ................. 79 Figure 11: Three examples of performance curves. .................................................................... 82 Figure 12: Scatterplot showing the relation between alpha (α) values extracted from the exponential curve fitting for each participant during practice and absolute retention tmRTT (s).85 Figure 13: Single-subject practice and retention data for Study 1 (baseline). ............................. 87 Figure 14: Single-subject normalized curves for practice. .......................................................... 87 Figure 15: Single-subject practice and retention data for low difficulty condition. .................... 97 Figure 16: Single-subject normalized performance curves for practice for low difficulty condition. ...................................................................................................................................... 97 Figure 17: Single-subject practice and retention data for medium difficulty condition. ............. 98 Figure 18: Single-subject normalized performance curves for practice for medium difficulty condition. ...................................................................................................................................... 98 xii Figure 19: Single-subject practice and retention data for high difficulty condition. ................... 99 Figure 20: Single-subject normalized performance curves for practice for high difficulty condition. ...................................................................................................................................... 99 Figure 21a and b: Practice and retention performance. ............................................................ 100 Figure 22a, b, c, and d: Continuous tracking task (CTT). ........................................................ 116 Figure 23: Skill acquisition follows an exponential decay as performance improves. .............. 118 Figure 24a and b: Healthy control (HC) repeated sequence performance data. ....................... 124 Figure 25a and b: Individuals with stroke repeated sequence performance data. .................... 124 Figure 26: Repeated sequence practice curves. ......................................................................... 126 Figure 27a and b: Repeated sequence performance. ................................................................ 128 Figure 28: T1 weighted images with lesion location. ................................................................ 145 Figure 29: Continuous tracking task (CTT). .............................................................................. 147 Figure 30: Functional magnetic resonance imaging and constrained principal component analysis (fMRI-CPCA) methodology. ........................................................................................ 150 Figure 31a and b: Plots of whole-brain and masked predictor weights on retention, and images for component one networks revealed by functional magnetic resonance imaging and constrained principal component analysis (fMRI-CPCA) at baseline and retention. .................................... 157 Figure 32: Tracking performance. ............................................................................................. 159 Figure 33: Relationship between functional connectivity in the masked motor network and repeated tracking performance for the stroke (ST) group at retention (day 7). .......................... 160 Figure 34: Experiment design and apparatus. ............................................................................ 176 Figure 35a and b: Diffusion tensor imaging (DTI). .................................................................. 185 Figure 36: Constrained motor connectome (CMC). .................................................................. 186 xiii Figure 37a, b, c, and d: Active continuous theta burst stimulation (cTBS) motor practice responders and non-responders. .................................................................................................. 189 Figure 38: Lesion figure for M1c and S1c (contralesional sensorimotor [SMc])-cTBS motor practice responders versus non-responders. ................................................................................ 191 Figure 39a, b, and c: Mean response time total (RTT) for repeated and random sequences. .. 194 Figure 40a, b, and c: Motor skill acquisition curves for contralesional primary somatosensory cortex (S1c), contralesional primary motor cortex (M1c), and sham stimulation groups. ......... 196 Figure 41: Comparison of responder versus non-responder of CMC in the SMc-cTBS group. Error bars represent SD. .............................................................................................................. 199 Figure 42: Subject-specific examples of white matter tractography. ........................................ 199 Figure 43: Statistical study design and results flow chart. ........................................................ 200 xiv List of Tables Table 1: Summary of three examples of functional connectivity analyses .................................. 53 Table 2: Correlations between rate of acquisition, phase ratio, and retention. ............................ 84 Table 3: Individual number of switches for each of the learner-adapted conditions for Study 2 (ascending order). ......................................................................................................................... 95 Table 4: Participant characteristics. ........................................................................................... 114 Table 5: Participant characteristics ............................................................................................ 144 Table 6: Cluster volumes, baseline (day 1) ................................................................................ 155 Table 7: Cluster volumes, retention (day 7) ............................................................................... 156 Table 8: Participant characteristics ............................................................................................ 178 Table 9: Constrained motor connectome (CMC) ....................................................................... 186 Table 10: Motor practice responder descriptive profile ............................................................. 190 Table 11: Comparison (mean and SD) of responder versus non-responder DWI characteristics in SMc-cTBS group. ................................................................................................................... 198 xv List of Abbreviations BOLD blood-oxygen-level-dependent CC corpus callosum CI contextual interference CMC constrained motor connectome CPCA constrained principal component analysis CPF challenge point framework CSD constrained spherical convolution CST corticospinal tract CTT continuous tracking task cTBS continuous theta burst stimulation DCM dynamic causal modeling dHb deoxygenated hemoglobin DLPFC dorsolateral prefrontal cortex DPT discrete pairing task DTI diffusion tensor imaging DWI diffusion-weighted imaging FM Fugl-Meyer assessment fMRI functional magnetic resonance imaging GM gray matter HC healthy control HDR hemodynamic response LI laterality index xvi M1 primary motor cortex MCA middle cerebral artery MEP motor evoked potential MMSE mini mental state examination MT movement time PLIC posterior limb of the internal capsule PFC prefrontal cortex PMC premotor cortex RMSE root mean squared error ROI region of interest RMT resting motor threshold RT reaction time RTS retention test score RTT reaction time total rTMS repetitive transcranial stimulation S1 primary somatosensory cortex SMA supplementary motor area SRT serial reaction time STT serial targeting task TBS theta burst stimulation UE-FM upper -extremity Fugl-Meyer assessment WB whole-brain WM white matter WMFT Wolf motor function test xvii Acknowledgements First, I would like to express my deepest gratitude to my doctoral supervisor, Dr. Lara Boyd. Lara, you supported and encouraged me, providing me with a sense of empowerment and confidence as a young researcher. I want to sincerely thank you for always taking time to speak with me, and providing guidance throughout my time in the BBL. I could not have imagined completing my PhD with any other supervisor, or in any other lab. The BBL family is as productive, supportive, and close as we are because of the positive, fun-loving environment that you have created. Thank you for not only being a supportive supervisor and mentor, but for always balancing our research discussions with our in-depth running conversations. Besides my supervisor, I would like to thank the rest of my thesis committee: Dr. Nicola Hodges, and Dr. Todd Woodward. Nikki, thank you for your patience, attention to detail, and infinite help. Above and beyond, thank you for instilling in me the importance of approaching research with an open and inquisitive nature. Todd, thank you for providing me with an opportunity to join your lab and learn your methodological approaches, and for your thoughtful insight into my research. I’d like to thank the entire BBL family, past and present, for making my time in Vancouver, and at UBC, an incredible, life-changing experience. Thank you all for being there to support me in both my academic and personal life. Special thanks to my Vancouver sister, Kate Brown, for your kindness, friendship and encouragement throughout my PhD journey. There are no words to describe how much your friendship means to me, and I can’t imagine completing this without you. To my parents: Mom, none of this would have been possible without you. You believed in me from the beginning, and never let me forget my potential. You are the strongest influence xviii in my life, and I’m so grateful and proud to call you my mother and friend. Dad, you taught me how to think outside the box, dream big, and follow my heart. Throughout this journey, I knew I could always rely on you to offer me the best advice, and encouraging words of wisdom. To my sisters, Nancy and Jennie, my living angels: I’m so thankful for free long-distance calling. You’re my biggest supporters, and my heart and soul. To my roommate, Claire Witt: thank you for your support through every single step of my PhD. To Danny: thank you for being such a positive influence in my life. Your words of encouragement and kindness have helped me persevere until the end. “By endurance we conquer.” — Ernest Shackleton xix Dedication I dedicate this dissertation to my grandmother, Pauline Alma Parsons. Forever, and always in my heart, this is for you, Grams. 1 Chapter 1: Introduction Preamble Motor learning occurs with repeated practice of skilled movements (Willingham, 1998). During motor skill practice, individuals participate in problem-solving processes to execute goal-oriented motor actions (Gallahue & Ozmun, 2012). Repetitive movement patterns are a form of problem solving, whereby individuals use trial-and-error to improve the performance of their motor actions (Adams, 1984). Over time, individuals’ movement patterns gradually increase in accuracy and efficiency (Cohen & Sternad, 2009). Motor performance, which is transient in nature, is assessed during practice and can be influenced by factors such as variability of practice (i.e., task order of practice conditions). In contrast, motor learning is a relatively stable and permanent improvement in a skill, and is assessed during a delayed retention test (Salmoni, Schmidt, & Walter, 1984; Schmidt & Lee, 2011). Factors that influence performance in practice can differentially affect the retention of a motor skill. Furthermore, individual differences in motor and cognitive abilities impact the manner in which motor skills are acquired and retained (Guadagnoli & Lee, 2004). However, few studies have investigated the differential effect of individualized practice conditions (i.e., individualized variability of practice) on motor performance and learning. The investigation into the effect of learner-adapted practice conditions on skill acquisition and retention will inform the design of interventions to enhance motor learning. Following an incident of stroke, approximately 55%─75% of afflicted individuals have chronic impairments in upper extremity function (Nichols-Larsen, Clark, Zeringue, Greenspan, & Blanton, 2005). A plethora of work has shown that individuals with stroke preserve their ability to learn or relearn motor skills to promote functional recovery, even during the chronic 2 stages of stroke long after the original ischemic insult (Boyd, Vidoni, & Wessel, 2010; Boyd & Winstein, 2001, 2003, 2004a; Boyd & Winstein, 2004b). To help clinicians increase effectiveness of motor learning interventions, practice parameters known to relate to long-term retention of motor skills should be identified. Practice parameters derived from the performance curve, a well-established function to describe processes of learning (Adams, 1987; Book, 1908; Heathcote, Brown, & Mewhort, 2000; Newell & Rosenbloom, 1981), may be of clinical importance. Understanding performance curve parameters that enhance learning has potential implications for motor learning interventions in stroke rehabilitation. Specifically, determining the relationship between the performance curve parameter, alpha, which represents the rate of skill acquisition, and the retention of a motor skill can inform the design of motor learning interventions. At present, few measures are known to predict, and optimize, an individual’s capacity for motor learning following a stroke (Reinkensmeyer et al., 2016). To fully exploit the motor performance and learning relationship, identifying parameters in performance curves that can be modified (i.e., rate of skill acquisition) to enhance the retention of motor skills could bring us closer to individualized motor learning interventions post-stroke. Neurophysiological changes that have been attributed to learning processes during motor skill acquisition are well documented in healthy populations (Doyon et al., 2009; Doyon & Benali, 2005; Doyon, Penhune, & Ungerleider, 2003; Doyon, Ungerleider, Squire, & Schacter, 2002; Reiss et al., 2005; Seidler, 2010). However, following neurologic injury, such as ischemic stroke, learning processes do not typically follow patterns of neurophysiological change as observed in healthy individuals (Meehan, Randhawa, Wessel, & Boyd, 2011). Individuals are considered to enter the chronic phase at six months post-stroke (Bernhardt et al., 2017), and traditionally, in this phase, it was thought that motor recovery plateaus (Dobkin, 2004). 3 However, based on evidence from neuroimaging (Johansen-Berg et al., 2002; Luft et al., 2004), and behavioural outcomes following upper extremity rehabilitation interventions (Langhorne, Coupar, & Pollock, 2009; Richards, Stewart, Woodbury, Senesac, & Cauraugh, 2008; Stewart, Cauraugh, & Summers, 2006), researchers suggest that this assumption no longer remain. Individuals with ischemic stroke in the chronic phase (> 6 months) have been found to demonstrate improvements in functional recovery of the upper extremity and the ability to learn new motor skills (Borich, Brown, & Boyd, 2014; Boyd & Winstein, 2006; Boyd et al., 2009; Boyd & Winstein, 2001, 2003, 2004a). While improvements in motor recovery emerge in this phase, post-stroke motor impairments change the way motor learning, and re-learning, occur (Boyd & Winstein, 2006; Boyd & Winstein, 2001, 2003, 2004a; Hanlon, 1996; Lefebvre et al., 2015; Meehan, Randhawa, et al., 2011; Winstein, Merians, & Sullivan, 1999). However, few studies have directly investigated the patterns of neurophysiological change during the performance of a newly learned motor skill in individuals with chronic stroke. Currently, prognostic models that include baseline clinical and neuroimaging parameters to predict levels of motor function and impairment in stroke are proving to be of clinical use (Stinear, Barber, Petoe, Anwar, & Byblow, 2012; Stinear et al., 2007) (Figure 1; adapted from review by Auriat et al. 2015). Building upon the use of prognostic models to stratify individuals based on potential for functional recovery, identifying baseline clinical and neuroimaging parameters that determine the response to motor leaning-based interventions is of importance (Reinkensmeyer et al., 2016). To augment motor learning processes in individuals with upper extremity impairment post-stroke, researchers are investigating the use of pairing neuromodulators, such as repetitive non-invasive brain stimulation, with motor skill practice (Brodie, Meehan, Borich, & Boyd, 2014; Meehan, Dao, Linsdell, & Boyd, 2011). While this 4 pairing has shown promising results, the response to non-invasive brain stimulation is highly variable (Brodie, Borich, & Boyd, 2014). More research exploring the identification of biological markers, known as “biomarkers”, that determine the response to non-invasive brain stimulation is warranted. In an era of personalized medical care and rehabilitation, there is limited research that has individualized interventions for motor learning following stroke. The present dissertation offers behavioural and neurophysiological methods, which can be used to establish personalized rehabilitation approaches in a clinical setting post-stroke. 5 Figure 1: Neuroimaging factors. Neuroimaging factors that may predict motor function, impairment, and learning, in individuals with chronic stroke (adapted from Auriat et al., 2015). a. Three important components of motor recovery: (1) motor function, as measured by Wolf Motor Function Test (WMFT; demonstrated in picture), (2) motor impairment as measured by Fugl-Meyer assessment (FM; demonstrated in picture), and (3) motor learning as measured by serial reaction time task (SRT; demonstrated in picture). b. Examples of neuroimaging modalities that account for portions of variability in motor recovery, in chronic stroke: (1) level of activation and/or connectivity of the paretic and non-paretic hand, as measured by functional magnetic resonance imaging (fMRI); gray and white matter volumetrics as measured by FreeSurfer automated or manual delineation software; corticospinal and intracotrical excitability measures, as measured by transcranical magnetic stimulation (TMS); white matter integrity of the corticospinal tract (CST) and/or corpus callosum (CC) as measured by diffusion tensor imaging (DTI); levels of glutamate, and/or N-acetyl-aspartate (NAA) as measured by magnetic resonance spectroscopy (MRS). fMRIDTIVolumetricsMRSTMSMOTOR FUNCTION MOTOR IMPAIRMENT MOTOR LEARNINGPARAMETERS TO PREDICT MOTOR RECOVERYMOTOR RECOVERYA.B. 6 Research objectives The following four studies were designed to evaluate behavioural measures and neurophysiological correlates and determinants of motor learning in healthy individuals and individuals post-ischemic stroke. Specifically, this thesis sought to: (1) evaluate motor practice and learning in healthy individuals as well as those with ischemic chronic stroke; (2) assess the efficacy of individualized motor practice; (3) evaluate motor learning neurophysiology; and (4) identify new biological markers (“biomarkers”) of motor learning. Figure 2 provides a schematic representation of the overall dissertation design. Figure 2: General aims of thesis. Outline of general aims and objectives of this thesis. Overall Thesis Aim: To advance stroke rehabilitation interventions to help facilitate motor learningSpecific Aim 1:To assess individual changes in motor practice and learning(Chapter 2, 3, 5)Specific Aim 2:To evaluate individualized practice paradigms to promote long-term retention(Chapter 2, 3, 5)Specific Aim 3:To accurately identify motor networks important for motor learning(Chapter 4, 5)Specific Aim 4:To identify biomarkers of motor learning (Chapter 5)Evaluation of motor behaviourEvaluation of motor neurophysiologyBiomarkers of motor learningEfficacy of individualized motor practiceGeneral Aim 1 General Aim 2 General Aim 3 General Aim 4 7 Proposed outcomes A full description of the experimental design, including the conditions and measures of each experiment outlined in the dissertation, will be described in Chapters 2, 3, 4 and 5. In brief, this dissertation is divided into two distinct, yet complementary sections, comprised of two studies in each section. In the first section, the use of performance curves to quantify motor skill acquisition and assess relationships between rate of motor skill acquisition and motor learning is investigated (Chapter 2, 3). Moving beyond the behavioural quantification of motor learning, the second section of this dissertation utilizes neuroimaging techniques to assess the underlying neural networks associated with motor learning in healthy individuals and individuals with chronic ischemic stroke (Chapter 4, 5). The specific research objectives and hypotheses for chapters are as follows. Chapter 2 Aim: To investigate individualized variability of practice conditions (i.e., the manipulation of task order), determined from baseline practice performance and a computer controlled learner-adapted algorithm, for motor sequence learning in young, healthy people. Hypotheses: Practice parameters derived from performance curves in a baseline practice condition inputted into a computer controlled learner-adapted algorithm will create individualized practice conditions of varying levels of difficulty, and differentially influence motor sequence learning. Practice conditions with a high level of individualized difficulty (high variability of practice) will result in superior motor sequence learning compared to medium and low levels of individualized difficulty. 8 Chapter 3 Aim: To investigate the relationship between rates of motor skill acquisition and performance at a retention test in individuals with chronic stroke and matched healthy controls (Wadden et al., 2017). Hypotheses: There will be a positive relationship between rate of motor skill acquisition and reduced loss of motor performance during the retention interval in individuals with stroke and healthy individuals. Healthy individuals will demonstrate a faster rate of motor skill acquisition in repeated sequence practice, compared to individuals with stroke. Chapter 4 Aim: To employ a functional magnetic resonance imaging (fMRI)-based multivariate analysis to assess the functional connectivity of a network associated with motor learning in healthy individuals and individuals with stroke (Wadden et al., 2015). Hypotheses: Healthy individuals will demonstrate greater functional connectivity of a motor learning–related network compared to individuals with stroke during repeated sequence performance on a delayed retention test. Chapter 5 Aim: To investigate the microstructural properties of pre-existing white matter tracts in a motor network as a biomarker for motor learning–related change, following repetitive transcranial magnetic stimulation (rTMS) over the sensorimotor region of the contralesional hemisphere, when paired with motor skill practice in individuals with chronic stroke. Hypotheses: The integrity of the white matter motor network will be higher in individuals who demonstrate a positive capacity for motor learning–related change following rTMS over the 9 sensorimotor region of the contralesional hemisphere, when paired with motor skill practice, compared to those who do not respond. Chapter 3 was based on a secondary analysis of a sample of healthy individuals and individuals with stroke from a study funded by the National Institutes of Health (NIH: R03 NS051714). Chapter 4 was based on a re-analysis of magnetic resonance imaging data for a question-focused dataset of healthy individuals and individuals with right sided, ischemic stroke confined to the subcortex of the previously mentioned study, funded by the NIH (R03 NS051714). The primary results from the datasets in Chapter 3 and 4 have been published elsewhere (Meehan, Randhawa, et al., 2011). Chapter 5 formed a subcomponent of a planned randomized controlled trial that was funded by Canadian Institutes for Health Research (CIHR) and registered with Clinical Trials (NCT01371409). 10 Motor learning Most of human behaviour involves the performance of learned motor skills (Fitts & Melton, 1964). Largely, motor learning is a form of procedural learning that involves the processing of information beyond that of a simple stimulus-response association (Reber, 1976, 1989; Seger, 1994). Motor skills can range from simple to complex, requiring short or long time scales of practice, respectively, until a level of expertise is achieved (Newell, Liu, & Mayer-Kress, 2001). Due to the range of behaviours associated with the execution of motor skills, cognitive, perceptual, and motor processes are all engaged; each process has a distinct contribution depending on the nature of the task (Edwards et al., 2010). On this basis, two forms of laboratory-based motor learning commonly evaluated are: (1) repeated performance of a specific series of movements known as motor sequence learning (Grafton, Hazeltine, & Ivry, 2002), and (2) compensatory movements in response to environmental changes or perturbations known as motor adaptation learning (Shadmehr & Krakauer, 2008). In the forthcoming sections, this literature review will discuss general principles of motor performance and learning, with specific emphasis on motor sequence learning. Motor sequence learning Motor sequence learning involves the acquisition of knowledge, often implicitly, of task regularities within a sequence of stimuli. Individuals learn task regularities by repeatedly performing movements that are sequenced together (Nissen & Bullemer, 1987). Continued practice of the repeated movements permits the automatization of the sequence of stimuli into a single entity to be executed (Frensch, Lin, & Buchner, 1998). Learning sequences of movements evokes functional remodeling and reorganization of neural circuits (Karni, Meyer, Jezzard, & Adams, 1995), and is therefore essential in the rehabilitative processes following 11 neurophysiological trauma (Boyd & Winstein, 2003). Within the work of this dissertation, there is a dedicated focus on the study of motor sequence learning in healthy individuals and individuals with chronic ischemic stroke (see Chapter 2, 3, 4, and 5). Phases of motor learning During the acquisition of new motor skills, behavioural changes appear to represent the sequential phases of motor learning. Fitts and Posner (1967) were the first to outline three distinct phases of skill acquisition: (1) the cognitive stage, (2) the associative stage, and (3) the autonomous stage (Fitts & Posner, 1967). The cognitive stage is a time when individuals use problem-solving processes to learn the goals of a task. Individuals learn the instructions and objectives and determine the appropriate sequence of motor actions to achieve their desired goal (Fitts & Posner, 1967). During this phase, skill acquisition is labile and susceptible to influence by external factors such as the environment or other concurrent tasks. The associative stage commences when the sequence of movements is determined and attention can be directed to precise details of the movements. Here, there is a refinement in the general movement strategy to achieve the desired goal; error gradually decreases as individuals hone the timing of their movements and bodily orientation (Fitts & Posner, 1967). In the autonomous stage, the sequence of movements used to achieve a desired goal becomes automatic. At this point, little cognitive effort is required to perform the motor actions, and external factors such as the environment or other concurrent tasks minimally influence performance (Fitts & Posner, 1967; Taylor & Ivry, 2012). The model of skill acquisition proposed by Fitts and Posner (1967) follows a performance curve that shows gradual decreases in movement time and/or error across a wide range of tasks (Taylor & Ivry, 2012). The use of performance curves has been used to support theories of motor 12 learning and prescribe practice interventions based on various characteristics of the performance curve (Lane, 1987). Following the inception of Fitts and Posner’s (1967) model, other theorists developed and refined frameworks to describe the phases of skill acquisition (Anderson, 1982, 1993; Shiffrin & Schneider, 1977). Ackerman (1988) developed an information processing framework to explain individual differences during skill acquisition against the backdrop of changing skill and task demands. In Ackerman’s framework, there is a functional similarity between three classes of cognitive and motor abilities (general intelligence, perceptual speed ability, psychomotor ability) and the three phases of skill acquisition. The early practice phase is thought to require high amounts of cognitive resources and results in slow and inaccurate performance. Performance in this phase is said to be associated with general intelligence and broad-content abilities, such as spatial ability for spatial tasks, and verbal ability for oral or written tasks. In the middle phase, greater competence is achieved, and stimulus-response connections strengthen. Performance in this phase is associated with an individual’s perceptual-speed abilities (speed of processing information). In the last phase, the task can be performed with little cognitive or attentional effort, and is executed with precision and speed. In this phase, performance is associated with psychomotor abilities (speed and accuracy of motor responding) (Ackerman, 1986, 1987, 1988). In Ackerman’s framework, individuals transition through the phases of motor skill acquisition at learner-dependent rates, as information-processing capabilities rely on the skill level of the learner and the difficulty of the task. Despite this knowledge, little has been done to quantify the duration of time spent in each of the phases of motor skill acquisition, and the effect of learner-dependent rates and phase duration on the retention of motor skills (Chapter 2). 13 Phases of motor sequence learning An elaboration to the early Fitts and Posner (1967) research was developed to provide a framework specific to the phases of motor sequence learning for a discrete sequence task (discrete sequence production task) (Abrahamse, Ruitenberg, de Kleine, & Verwey, 2013). In this framework, cognitive and motor processors operate as a dual processor to both acquire and execute discrete movement sequences. In this dual processor model, during the early phase of acquisition the cognitive processor decodes external cues discretely, and precipitates the motor processor to execute responses independently (Abrahamse et al., 2013). In the middle phase, explicit sequence knowledge develops, which increases the execution speed of the motor sequence. Cognitive and motor processors work in parallel to promptly initiate successive movements as the preceding responses in the motor sequence increase in familiarity. With practice, the cognitive processor creates unified collections of elemental information within the sequences. This strategy reduces the constraints of our finite information-processing capabilities as multiple items are coded into a single representation. In motor sequence learning, these unified collections of information are known as motor chunks (Book, 1908) and are loaded in advance into a motor buffer, which is a limited-capacity form of working memory, to be executed by the motor processor. In the last phase, the contribution of the cognitive processor is significantly reduced and the sequence is represented in large motor chunks to be executed by the motor system (motor buffer and motor processor) in an autonomous process (Abrahamse et al., 2013). Historically, the performance curve has been used to describe theories of general motor skill acquisition (Fitts & Melton, 1964; Snoddy, 1926), however, few studies have used performance curves to examine the conceptual framework of motor sequence learning in healthy individuals and individuals with stroke, which will be described in more detail in Chapters 2 and 14 3. Additionally, motor sequence learning and, potentially, motor sequence performance curves, can be influenced by a number of factors including the conditions of practice (intensity and variability) and individual ability, both of which will now be discussed. Factors that influence motor learning Due to the considerable effects of the structure of practice on motor performance and long-term retention, researchers have focused their attention on identifying factors that differentially influence motor performance and learning processes (Lee & Magill, 1983; Magill & Hall, 1990; Shea & Morgan, 1979). When comparing different practice methods, it is critical to consider the objective of motor practice. For example, in rehabilitation settings, the clinicians’ goal is not the facilitation of motor performance during practice, but rather the enhancement of motor learning and long-term retention of motor skills (Boyd et al., 2010). Furthermore, it is important to consider individual differences as motor performance and learning may be dependent on the skill level of the individual, in combination with the factors under which the motor skill is practiced (Guadagnoli & Lee, 2004). A common factor that has been studied, and that is known to influence the learning process is the intensity of practice, which can be manipulated through the variability of practice, each of which is described in greater detail below. Intensity of practice Intensity of task practice is defined as the amount of physical and/or psychological effort in a single movement or series of movements during a defined period (Page, Schmid, & Harris, 2012). Intensity of task practice is an important factor in exploiting the “performance versus learning distinction,” as practice conditions with greater physical and/or psychological effort often produce slower rates of learning during practice, but enhance the long-term retention of 15 motor skills (Salmoni et al., 1984; Schmidt & Lee, 2011). Bjork (1994) was the first to discuss the importance of creating “desirable difficulties” during practice conditions, which initiate elaborate encoding and retrieval processes to support learning. Furthering Bjork’s theories of desirable difficulties, the challenge point framework (CPF) presented by Guadagnoli and Lee (2004) has been advanced to explain the interactive effect of task difficulty and individual skill level on performance during motor skill acquisition. In the CPF, there is an optimal “challenge point,” which is a theoretical level of difficulty that yields the highest level of interpretable information that can be effectively extracted during motor skill practice (Guadagnoli & Lee, 2004). Practically, practicing at an optimal challenge point yields the greatest learning effect; this only exists when information does not exceed beyond, nor drop below, an individual’s processing abilities. The optimal challenge point is dependent upon task difficulty, which can be divided into two categories: (1) nominal task difficulty, a constant level of difficulty that is not influenced by the skill level of the learner or the conditions under which the task is performed; and (2) functional task difficulty, which is dependent upon the skill level of the learner and the conditions under which the task is practiced (Guadagnoli & Lee, 2004). During motor learning, individuals problem-solve using differing amounts of available and interpretable information. The potential amount of information available is dependent on the nominal level of task difficulty, while the amount of interpretable information is dependent on the functional level of difficulty (Guadagnoli & Lee, 2004). Functional difficulty results from the conditions under which the task is practiced, as well as the skill level of the learner. Practice variables can change the functional task difficulty, resulting in increases or decreases to the optimal challenge point of the task; in turn, these directly affect an individual’s potential to learn (Guadagnoli & Lee, 2004). 16 As a supplement to the CPF from a cognitive psychology perspective, Sweller (1988) developed the cognitive load theory that speaks to intensity of practice and achieving desirable difficulty during learning through the interaction of three “load” factors, which include: (1) intrinsic cognitive load; (2) extraneous cognitive load; and (3) germane cognitive load. Intrinsic cognitive load is a fixed component, defined as the natural complexity of information that must be learned (Sweller, 1988). The intrinsic cognitive load can only be altered by changing the basic task or changing the individual’s knowledge level (Sweller, 2010). If the goal of the task is to learn a repeating sequence, the intrinsic load is comprised of the elements that create the sequence. Extraneous load is related to any irrelevant aspects of the task, which, once removed, do not alter the goal of the task. In a rehabilitation setting, irrelevant cues or distractors could come from other patients in the practice environment, and do not contribute to the goal of the task. Germane load refers to working memory resources allocated to processing the intrinsic cognitive load of the task. In comparison to the CPF, cognitive load theory states that the optimal challenge point as well as the greatest potential for learning occurs when the extraneous load is minimal, such that an individual’s working memory resources can be directed towards acquiring the intrinsic cognitive load and therefore maximizing the germane load (Sweller, 2010). As presented above, the intensity of practice is an important factor to influence an individual’s ability to acquire motor skills. However, with the proposed theoretical importance of practicing at an optimal challenge point, more investigation into the practicality and efficacy of individualized intensity of practice conditions is required (Chapter 2). Variability of practice A component of intensity of practice that has been shown to influence motor learning is variability of practice. Task order is one characteristic of variability of practice that differentially 17 influences short and long–term motor performance; it is known as contextual interference (CI) (Shea & Morgan, 1979). Over the course of a practice session, the repeated performance of one task before the performance of a subsequent task is termed blocked practice, or “low CI.” In comparison, when multiple tasks are performed in a random order, this is known as random practice, or “high CI” (Shea & Morgan, 1979). When the nominal difficulty of motor tasks is relatively low, blocking the order of the practice trials (low CI) compared to the random ordering of trials (high CI) produces practice conditions of low and high functional task difficulty, respectively. Practicing under conditions of low CI, and thus low functional task difficulty, results in superior practice performance as compared to high CI, and high functional task difficulty. However, a delayed retention test reveals the reverse effect: high CI produces superior long-term performance at a delayed retention test, compared to low CI. Thus, inferior performance in practice during a high CI practice schedule typically produces enhanced motor learning (Guadagnoli, Holcomb, & Weber, 1999; Lee & Magill, 1983; Magill & Hall, 1990; Rey, Wughalter, & Carnes, 1987; Shea & Morgan, 1979). Figure 3 illustrates such performance-learning dissociations as a result of practice in random and blocked conditions (left side) and in retention testing, under blocked and random conditions (right side) (Shea & Morgan, 1979). 18 Figure 3: The contextual interference (CI) effect. The data from this graph has been approximated from Shea and Morgan (1979) to demonstrate the effect of blocked and random practice scheduling (i.e., CI) on performance and delayed retention tests (10 minutes delayed and 10 days delayed). The random and blocked practice groups performed the delayed retention test in both a blocked and random order. Individuals in the blocked practice group (gray line, triangle markers) performed better (lower movement time) than the random practice group (black line, circle markers) during practice (higher movement time); however, the reverse effect was noted at both retention tests (post-10 minutes and post-10 days). During retention, individuals performed the tasks in a random (R) or blocked (B) order. In the above figure, the first letter indicates the group condition in practice (R, B) and the second letter indicates the condition order (R, B) in retention. Descriptively, during the 10-minute retention test, the random practice group was faster than the blocked practice group under both the random (R-R > B-R) and blocked (R-B > B-B) condition orders, but not when comparing conditions specific to those practiced (i.e., B-B > R-R). Descriptively, at the 10-day retention test, these results were largely maintained, except now the random practice group was slower than the blocked practice group performed in a blocked order (B-B > R-R). Statistically, there were no significant differences between 10-minute and 10-day retention intervals (p > 0.05), and post hoc analysis demonstrated all comparisons were significant (p < 0.05), except for between blocked-blocked (B-B) and random-random (R-R) conditions, which were not significant (p> 0.05). 00.511.522.531 2 3 4 5 6Mean Total Movement Time (s)PracticeRandomBlocked10 min 10 dayB-RB-BR-RR-B 19 Interestingly, largely as a result of individual differences observed during motor skill acquisition, the CI effect does not always hold true (Guadagnoli et al., 1999; Rey et al., 1987). An individual with a low skill level must divide their cognitive and attentional resources between learning the task and the demands of task switching (McLaughlin, Rogers, & Fisk, 2006). This can negatively impact delayed retention performance following high CI practice conditions. Conversely, an individual with a high skill level requires less cognitive and attentional resources to devote to motor skill acquisition and, therefore, is more likely to experience the performance-learning benefits of high CI practice conditions (Guadagnoli & Lee, 2004). Thus, this evidence highlights the importance of creating a match between the constraints of the individual and the characteristics of the task, in order to optimize learning conditions. The CI effect on motor learning has been explained by two opposing, yet somewhat complimentary, theoretical views: (1) the elaboration and distinctiveness view; and (2) the forgetting and reconstruction view (Lee, 2012). In the former, there is an emphasis on an individual’s ability to compare tasks, creating elaborate and distinctive memories (Battig, Thompson, & Voss, 1972; Battig, 1979). Higher levels of CI during practice allow for the greatest opportunity to compare tasks. This taxes short-term memory processes during practice and creates long-term memories that lead to superior retention (Lee, 2012). The latter view, put forth by Lee and Magill (1983), emphasizes the use of working memory. During high CI practice, different action plans are continuously executed and then unloaded from ones working memory in order to prepare for the next task (Lee & Magill, 1983). This unloading from working memory draws on long-term memory processes to construct action plans from one trial to the next. Although these two theoretical views have opposing memory-processing mechanisms, both may work concurrently during high CI practice (Lee & Simon, 2004). 20 Individualization of variability of practice: Learner-adapted practice is a systematic approach to designing practice conditions based on principles espoused in the CPF (Guadagnoli & Lee, 2004). However, few studies have investigated learner-adapted CI during motor skill acquisition and retention. The motivation for learner-adapted practice is to move beyond extreme forms of practice scheduling (i.e., blocked and random) and instead create more degrees of challenge to allow individuals to practice closer to their own optimal challenge point. Currently, there are minimal studies evaluating the practical applications of designing and implementing learner-adapted practice conditions (Choi, Gordon, Park, & Schweighofer, 2011; Choi, Qi, Gordon, & Schweighofer, 2008; Pollock, Boyd, Hunt, & Garland, 2014). Thus, more work investigating the use of learner-adapted CI during motor skill acquisition to enhanced motor learning-related change is warranted (Chapter 2). Learner-controlled versus computer-controlled practice: There are two methodological classifications of learner-adapted practice: learner-controlled and computer-controlled (Choi et al., 2008). Learner-controlled practice, which allows individuals to choose their desired practice schedule, has proven operationally advantageous for learning. During learner-controlled practice, when individuals decide to switch to a new task, their performance throughout skill acquisition typically becomes comparable to blocked practice scheduling (Titzer, Shea, & Romack, 1993). However, rather than the decrease in retention performance that is typically associated with blocked practice conditions, learner-controlled practice yields learning benefits, at a delayed retention test, comparable to random practice scheduling (Titzer et al., 1993). Expanding upon these findings, Wu and Magill (2011) studied a yoked group that was forced to switch to a new task on the same trial as a matched individual in the learner-controlled group whose members could choose when to switch tasks. Learner-controlled practice was superior to yoked practice 21 and a significant positive relationship between the number of switches (i.e., amount of CI) and retention performance was demonstrated (Wu & Magill, 2011). Furthermore, in this study, individuals continuously increased the number of switches following a successful trial. However, limitations in learner-controlled practice may exist. For instance, some individuals may never reach their optimal challenge point due to a false sense of competence during practice, which could limit their ability to achieve maximum skill retention (Simon & Bjork, 2001). An alternative to learner-controlled practice is computer-controlled practice. Choi et al. (2008) developed and implemented a computer-controlled, learner-adapted algorithm during motor skill acquisition of a visuomotor transformation task. The visuomotor transformation task involved participants learning four visuomotor transformation by manipulating a force-feedback joystick to move a cursor on a computer screen. Participants were instructed to move their cursor from an initial position to a color-coded target that was assigned an associated angular relationship between the joystick movement and cursor movement (-30°, 60°, -90°, and 120°) (Choi et al., 2008). Two learner-adapted algorithms were developed to manipulate the level of difficulty and the number of practice trials. Both algorithms effectively challenged individuals and led to participants in these groups outperforming fixed number of trials and random scheduling conditions on a delayed retention test (Choi et al., 2008). A computationally simplistic alternative to the learner-adapted algorithm is “win-shift, lose-stay,” which showed similar learning benefits to random practice scheduling (Simon, Cullen, & Lee, 2002). The limitation of this latter method is that task switching is solely based on performance during practice, and does not consider performance at the delayed retention test. The amount of task switching should be based on maintaining an individual’s rate of skill acquisition that positively relates to long-term retention of motor skills (Choi et al., 2008). To further maximize 22 individualized practice, algorithms need to incorporate the individual skill level (i.e., a determinant of prior experience). To date, the implementation of computer-controlled, learner-adapted practice during motor sequence learning has not been tested (Chapter 2). Stroke and motor rehabilitation In Canada, at any point in time, over 315,000 individuals are living in the community with the effects of stroke, and with every increasing year there as many as 60,000 new incidences of stroke (HSF, 2015). While some degree of recovery occurs in individuals who survive, globally, stroke is a leading cause of disability (Mendis, 2013). In the acute/subacute phases of stroke (less than or equal to six months after stroke), approximately 85% of individuals experience upper extremity impairments (Nichols-Larsen et al., 2005). Following the initial damage, a large amount of spontaneous recovery occurs and the presence of impairment in upper extremity function is reduced to approximately 50% to 60% (Lai, Studenski, Duncan, & Perera, 2002). However, as individuals enter the chronic phase (greater than or equal to six months after stroke), further decline in the level of impairment in upper extremity function occurs (Ward, 2017). To prevent additional loss of function, it is imperative that individuals relearn movement patterns to execute and control motor actions of their hemiparetic limb (Lefebvre et al., 2015). Motor recovery of the hemiparetic limb has been considered a form of motor learning and is central to stroke rehabilitation (Kitago & Krakauer, 2013). Ischemic stroke is a category of stroke that accounts for 87% of all cases (Lloyd-Jones et al., 2009). The pathology of an ischemic stroke involves the occlusion of a brain blood vessel that results in a deficiency of oxygen supplied to neuronal tissue (Woodruff et al., 2011). Clinical symptoms develop rapidly as the consequences of a blockage causes neuronal cellular injury, inflammatory responses, and neuronal death (Woodruff et al., 2011). Motor impairment is the 23 most common deficit after ischemic stroke (Wade, 1992). While there have been significant amounts of research focusing on functional motor recovery in the chronic phase post-stroke, effects of stroke rehabilitation interventions have been moderate (Ward, 2017). Currently, the best practice guidelines for stroke rehabilitation are not sufficient to effectively enhance recovery, and factors that influence motor re-learning post-stroke, such as optimal dose of treatment, are largely unknown (Ward, 2017). Furthermore, there is speculation that true motor recovery after stroke is problematic, as a portion of the cortical tissue lost due to cell death may not be spared beyond several months post-infarct (Krakauer, 2006; Levin, Kleim, & Wolf, 2009). Thus, recovery of motor function is desirable and pragmatic, relying on the recruitment of compensatory brain networks that recruit nervous tissue minimally impacted by, or recovered shortly following, the ischemic insult (Krakauer, 2006; Levin et al., 2009). Findings from stroke rehabilitation research demonstrate large variability in recovery and difficulties to predict response to interventions (Reinkensmeyer et al., 2016). Therefore, there are many important factors to consider when conducting interventions in stroke rehabilitation. Specifically, due to the heterogeneity of stroke pathology, higher inter-individual variance can result in a reduction in study power (Burke Quinlan et al., 2015). However, gathering a cohort of individuals with similar lesions is difficult. Conclusive findings directly related to the primary lesion location must be interpreted with caution, as multiple strokes and small white matter lesions in individuals with stroke are common (Vermeer et al., 2003). Nevertheless, to address the important issue of the effectiveness of stroke interventions, researchers are beginning to investigate methods to optimize motor learning therapies through manipulating the dosage and intensity of practice, as well as the use of adjunct therapies, such as repetitive forms of non-invasive brain stimulation which are thought to have therapeutic effects on stroke recovery 24 (Ward, 2015b; Ward, 2017). Furthermore, due to the variable outcomes following the delivery of stroke intervention, researchers are focusing on the identification of behavioural and biological predictors of motor recovery (Reinkensmeyer et al., 2016). Motor learning in individuals with stroke In the chronic stage of stroke, the residual brain preserves the ability to acquire new motor skills (Warraich & Kleim, 2010). Following a stroke, individuals use error-based learning to train the motor system to adjust movements to the new characteristics of the hemiparetic limb (Seidler, Kwak, Fling, & Bernard, 2013). Problem-solving processes are heavily relied upon, as impairments in the sensory system can lead to large errors in motor actions during motor skill re-learning (Seidler et al., 2013). Despite these deficits, the ability to learn new motor skills is not lost after stroke (Boyd & Winstein, 2001; Hanlon, 1996; Meehan, Randhawa, et al., 2011; Platz, Denzler, Kaden, & Mauritz, 1994; Winstein et al., 1999). In the chronic phase post-stroke, the overall magnitude of motor learning–related change is comparable to that of age-matched counterparts when assessed using delayed retention tests (Platz et al., 1994; Winstein et al., 1999). While comparable improvements are noted, deficits showed by individuals with stroke include greater variability in movement patterns, an overall slowing of movements, and an increase in corrective actions (Platz et al., 1994; Winstein et al., 1999). The heterogeneity of damage and thus impairments following stroke has led to variations in findings that complicate our understanding of the effects of neurological injury on motor learning processes (Kitago & Krakauer, 2013). When individuals with stroke relearn a motor skill, two mechanisms are believed to be at play: true recovery mechanisms and compensation in movement patterns (Krakauer, 2006; Levin et al., 2009). Both are related to motor learning. “True recovery” refers to the recruitment of 25 undamaged regions via preexisting brain connections to engage muscle groups used to execute the skill prior to stroke (Krakauer, 2006; Levin et al., 2009). Individuals will reestablish muscle synergies that were utilized before the lesion to perform motor skills (Nudo, 2013). “Compensation” refers to the recruitment of alternative brain regions and muscle groups to complete the movement (Krakauer, 2006; Levin et al., 2009). Through compensation, individuals engage in movement patterns that did not exist before the lesion (Nudo, 2013). In this sense, some individuals may gain improvements in motor function (i.e., the end-product of a goal-directed movement), while seeing little change in motor impairment (i.e., the kinematics underlying the process of movement execution) (Levin et al., 2009). In the chronic phase of stroke, it is believed that much of the recovery is due to this compensation, which is enabled by motor learning (Reinkensmeyer et al., 2016). Factors that influence motor learning in stroke rehabilitation: In both acute and chronic phase of stroke, to promote motor recovery, optimal dose of practice is considered. In stroke rehabilitation, the intensity of practice is defined as the amount of therapy delivered, and is therefore synonymous to dose (Langhorne, Bernhardt, & Kwakkel, 2011). This is currently a limitation in the field of stroke rehabilitation as greater consideration of the level of difficulty of practice conditions is merited. With regards to dose of practice, authors conducted a meta-analysis that investigated the dose-response relationship for motor rehabilitation post-stroke and reported that a larger dose of practice results in greater improvements in motor function post-stroke (Lohse, Lang, & Boyd, 2014). Additionally, this finding was independent of time following stroke – and showed no definitive ceiling effect on improvements – highlighting the important need to increase time in therapy across all phases of stroke recovery (Lohse, Lang, et al., 2014). While present findings from the literature indicate greater dose of practice is 26 beneficial for stroke recovery, a review from Lang et al. (2015) on the dose and timing relationship in neurorehabilitation concluded that the ‘more is better’ may be too vague as it is largely based on studies with arbitrarily-set doses. The authors are currently interested in answering the questions, “how much more is better?” and “better for whom?”, through a dose-response trial investigating four different doses of task-specific training in the chronic phase of stroke (Lang, Lohse, & Birkenmeier, 2015). While dose of practice appears to be important in stroke rehabilitation, we must consider turning to the motor learning literature in healthy individuals, to optimize motor practice paradigms to the skill level of individual (i.e., level of impairment) (Guadagnoli & Lee, 2004). For example, the Extremity Constraint Induced Evaluation (EXCITE) trial investigated Constraint-Induced Movement Therapy (CIMT), which is the restraint of the less affected (non-hemiparetic) upper extremity using a padded mitt to restrict hand usage, paired with repetitive and adaptive task practice of the more affected (hemiparetic) hand, on improvements in motor impairment and function (Wolf et al., 2010). Specific to optimal delivery of practice, similar effects from the EXCITE trial, which involved 10 days of six hours of training, were observed with a modified, adapted CIMT protocol that involved shorter training periods for an elderly population of stroke patients (Wu, Chen, Tsai, Lin, & Chou, 2007). Thus, the authors of the EXCITE trial concluded that an individualized approach based on the clinical profile of the individual (i.e., motor and cognitive abilities), or sub-group of individuals, may be a more appropriate practice protocol to optimize motor rehabilitation post-stroke (Wolf et al., 2010). In a review of stroke care by Langhorne, Bernhardt & Kwakkel (2011), the authors highlighted the importance of properly characterizing the target population (i.e., skill level of individual, level of impairment), because, depending on the factors that comprise the 27 rehabilitation intervention, individuals’ responses may vary greatly (Langhorne et al., 2011). In stroke rehabilitation, the importance of identifying determinants of motor learning is an essential next step in improving the effect of interventions and increasing prognostic capabilities for post-stroke motor recovery (Chapter 3, 5). As such, in addition to identifying a need for individualized motor rehabilitation strategies, it is crucial to discern which individual-specific factors can be exploited to enhance the success of these interventions. Behavioural assessments of motor learning Change in behaviour can be inferred from measuring changes in performance during practice and in a delayed retention test (Shea & Morgan, 1979). Since motor learning is comprised of a broad range of behaviours, when developing motor learning paradigms and interpreting subsequent behavioural changes, it is important to consider the nature of the task being used. In this dissertation, to investigate the nature of a task on motor learning–related changes in healthy individuals and individuals with stroke, three variants of the serial reaction time (SRT) task to assess motor sequence learning were employed: (1) discrete pairing task (DPT); (2) continuous tracking task (CTT); and (3) serial targeting task (STT), each of which will be discussed after reviewing the standard SRT. Motor sequence tasks One method for assessing motor sequence learning in a laboratory setting is with SRT tasks, which require participants to follow visual cues to perform a series of motor actions (Nissen & Bullemer, 1987). An SRT task allows for assessment of procedural motor sequence learning, as discrete movements are repeatedly performed in fixed combinations (Nissen & Bullemer, 1987). In an SRT task, adequate motor sequence learning is demonstrated when an individual is able to perform a series of repeated motor actions more quickly (i.e., they have a 28 faster reaction time) as compared to performing random motor sequences (Schendan, Searl, Melrose, & Stern, 2003). In the SRT, individuals respond to a visual cue that appears in one of a number of (typically horizontal) locations on a computer screen by pressing a key (Keele, Ivry, Mayr, Hazeltine, & Heuer, 2003; Robertson, 2007). Emphasis is placed on movement speed, as individuals are instructed to respond as quickly as possible when the correct matched stimuli appears (Nissen & Bullemer, 1987). During the SRT task, improved performance is observed for both the random and repeated sequences, reflecting advances in general motor control. Importantly, however, more change noted for repeated sequences, which can be learned, reflects motor sequence learning. Differences between performance of repeated and random sequences when assessed during a motor learning retention test are interpreted as reflecting motor sequence learning (Boyd et al., 2010; Boyd & Winstein, 2001; Meehan, Randhawa, et al., 2011; Schendan et al., 2003). Over the years, variants of the SRT have been developed to study the specific components of motor learning, such as motor, spatial, and temporal elements of learned movement sequences (Boyd, Vidoni, & Siengsukon, 2008; Schwarb & Schumacher, 2012). In this dissertation, I employed three motor skill tasks to investigate motor skill acquisition and learning of motor behaviours. The use of three motor skill tasks that were fundamentally similar (i.e., motor sequence tasks), yet that possessed different task-specific characteristics (i.e., key-press versus joystick versus computer mouse; explicit knowledge versus implicit learning) allowed for the interpretation of task-specific data versus generalization of findings. These three tasks are illustrated in Figure 4, which follows. 29 Figure 4: Illustration of motor learning tasks. Left. Discrete pairing task (DPT; Chapter 2). Centre. Continuous tracking task (CTT; Chapter 3, 4). Right. Serial targeting task (STT; Chapter 5). Discrete pairing task: Developed by our laboratory specifically for this dissertation, this novel discrete pairing task (DPT) incorporates rule-based associations into each stimulus-response trial. First, individuals learn the rule-based associations within each trial (i.e., positioning two repeating shapes side-by-side) and second, sequential learning is achieved as individuals procedurally acquire the repeating pattern of the stimuli across trials. Participants are explicitly told prior to practice that there are three repeating sequences; thus, the DPT task relies upon both the explicit and implicit motor systems. Early in practice, explicit knowledge of the sequences may help to facilitate the acquisition of the motor sequence. Later in practice, with increased expertise and automation of performance, individuals may rely on the implicit motor sequence to execute the sequence from memory. In the traditional SRT task, it is hypothesized that the execution of the sequence is based on linking perceptual and motor components, which require cognitive representations for appropriate motor responses (Schwarb & Schumacher, Mang, et al. 2014 Continuous Tracking Task (CTT)Discrete Pairing Task (DPT) Serial Tracking Task (STT) 30 2012). In the DPT task, individuals must first learn the paired association rule, which requires positioning two repeating shapes side-by-side. A single trial involves the identification of a repeating shape at two of the four horizontal spatial locations on a computer screen. Subsequently, individuals must initiate two responses by pressing a key: (1) the first press highlights the shape to be moved; and (2) the second press moves the highlighted shape to a new location. Identifying and pairing the repeating shapes adds a layer of cognitive complexity. Each trial is composed of two-phase motor units, which are known to be the foundational building blocks of motor sequences (Fitts, 1966). The entire motor sequence is comprised of five trials of two-phase movements, termed “polyphase motor units” (Melton, 2014). Individuals must learn three polyphase motor unit sequences in a practice session. With practice, individuals decrease their response times for the two-phase movements and then the polyphase motor units (see Figure 4 above for a visual representation of DPT; Chapter 2). Continuous tracking task: Many real-world tasks demand the continuous integration of visuospatial properties. The continuous tracking task (CTT) requires continuous movements and enables the study of specific kinematic patterns at a predefined movement velocity (Boyd & Winstein, 2006; Pew, 1974; Wulf & Schmidt, 1997). The CTT is an extension of Pew’s original pursuit task and uses waveforms based on a Gaussian distribution comprised of three 20-second segments; two-thirds of the waveforms are random and unpredictable, and one-third repeats a sequential pattern (see Figure 4 above for a visual representation of CTT; Chapter 3 and 4). In contrast to the DPT, the CTT exploits implicit motor systems in two ways: 1) participants rarely gain awareness of the presence of the repeating sequence; and 2) explicit knowledge does not facilitate motor sequence learning (Boyd & Winstein, 2004a; Green & Flowers, 1991). During performance of the CTT, participants’ transition from feedback control to feed-forward memory- 31 based control as early acquisition relies on visual and proprioceptive sensory information to facilitate corrective action plans (Meehan, Randhawa, et al., 2011). Root mean square error (RMSE) is measured during tracking, and motor performance can be decomposed into spatial and temporal error scores, which provides specific data on the elements of movement that were learned (i.e., spatial accuracy and temporal precision, respectively). Using this CTT in a sample of healthy young adults, Mang et al. (2014) demonstrated that an acute bout of aerobic exercise, performed prior to skill acquisition, impacts implicit sequence-specific learning of the temporal elements of the motor task at a delayed retention test, without impacting spatial accuracy (Mang, Snow, Campbell, Ross, & Boyd, 2014). Separation of the spatial and temporal components of this task allows for a sensitive detection of the specific behavioural impact of an adjunct therapy, in this case aerobic exercise, on motor skill acquisition (Mang et al., 2014). Serial targeting task: To incorporate both discrete and continuous dimensions of motor skill learning, Meehan et al. (2011) developed an SRT variant termed the serial targeting task (STT) to evaluate task-specific learning. This task allows for the amalgamation of discrete and continuous movement goals that are typically distinct in nature (Boyd & Winstein, 2006). For discrete movements, as compared to continuous movements, added motor control may be required to transition between on and off motor actions (Boyd & Winstein, 2006). Continuous movements have no recognizable beginning or end, as individuals perform fluid motor actions until a predetermined end point (Boyd & Winstein, 2006). Discrete movements have been shown to be acquired by explicit processes, and conversely, continuous movements are controlled by implicit strategies (Zelaznik, Spencer, & Ivry, 2002). To evaluate discrete and continuous movement goals in an experimental setting, the STT requires individuals to control a computer mouse to manipulate a cursor into a circular stimulus located at one of nine locations on 32 computer screen. Participants watch a computer screen for the appearance of the circular stimulus (target) and respond as quickly and as accurately as possible by moving the cursor inside the target. Individuals must keep the cursor inside the target for 500 ms before the presentation of the next target. The inter-target interval is 500 ms. The STT has been used to evaluate task-specific learning as indexed by improvements in movement times and kinematics measures (Meehan, Dao, et al., 2011). However, due to the sequential nature of the STT, a sequence can be embedded into the series of movements to investigate implicit motor sequence learning. The STT permits the analysis of movement times as well as kinematic measures such as reaction time (RT), peak velocity, peak acceleration, movement onset time, movement time, and endpoint deviation (Meehan, Dao, et al., 2011). As discrete and continuous task are known to engage different memory processes (Chambaron, Berberian, Delbecque, Ginhac, & Cleeremans, 2009) and neural structures (Boyd & Winstein, 2006), the practice parameters that predict motor learning of STT may be unique due to the nature of the task (see Figure 4 above for a visual representation of STT; Chapter 5). Moreover, the success of different adjunct therapies could similarly depend on the characteristics and performance measures of the motor task employed (Mang et al., 2014; Mang, Snow, Wadden, Campbell, & Boyd, 2016). Mathematical functions of motor performance Motor performance measurements, such as mean reaction time, movement time, and/or accuracy scores within a block, session, or day of practice are often evaluated to assess behavioural changes in practice and delayed retention tests. However, as discussed, motor learning has been described as occurring in distinct phases that are associated with characteristic cognitive and motor processes. Across a wide range of tasks, transitioning through the phases of motor skill acquisition follows a performance curve (Taylor & Ivry, 2012). To further investigate 33 the influence of factors on motor performance and learning, researchers have begun to use performance curves (Lakhani et al., 2016; Mang et al., 2016). Performance curves, which are the focus of several experiments in this dissertation, help to determine practice parameters across multiple points of performance and capture the phases of motor learning. Performance curves: The performance curve can be used to illustrates change in motor performance with practice. Performance curves have been investigated for over 100 years (Bahrick, Fitts, & Briggs, 1957; Heathcote et al., 2000; Lane, 1987; Newell & Rosenbloom, 1981; Snoddy, 1926; William & Harter, 1899). Discrete measures of performance, such as change scores (i.e., the last block of practice minus first block of practice) or indices of the final level of performance (e.g., mean response time, standard deviation, percent correct, etc.), have dominated research reporting in the field of motor learning. However, these measures reflect motor performance at single time points and as such do not capture the evolution of change in motor ability across practice. To better understand learning, it is important to statistically capture the dynamic evolution of motor performance throughout motor skill practice (Anderson, 1993; Logan, 1988; Palmeri, 1997). Individual motor performance across multiple trials follows a nonlinear form for most complex tasks and skills (Harbourne & Stergiou, 2009; Newell et al., 2001). In the early phase, rapid improvements in performance are typically observed, followed by slow decreases in the late phase of skill acquisition (Heathcote & Brown, 2004). Nonlinear changes in performance are ignored in conventional analyses that average data over multiple practice trials within and between individuals (Newell et al., 2001). Calculating an average performance from small windows of data (or blocks) can provide inaccurate information due to the natural fluctuations in performance known to occur (e.g., due to inattention, fatigue, lack of motivation, etc.) (Muratori, 34 Lamberg, Quinn, & Duff, 2013). In comparison, curve fitting uses all data points to capture an overall trend. Assessing trends in performance over time is advantageous when utilized in populations where the performance of movements can be highly variable (Lang & Bastian, 1999, 2001). Curve fitting (i.e., exponential or power functions) exploits all information by fitting a curve through the middle of all data. This fitted line represents the “mean performance scores” established from the overall trend of the data across practice (typically on an individual level, but this may also be across participants). Depending on the motor task performed and determining the line of best fit, curve fitting to mean performance scores calculated from each block, trial, or epoch of practice data uses the entire data set to model mean scores as a continuous curve. Implementing a mathematical function to describe the learning process allows for the computation of parameters related to practice (Cousineau, Helie, & Lefebvre, 2003). The basic description of the curve yields three important learning parameters. Firstly, the initial phase of acquisition/performance shows the greatest amount of change. The rate of motor skill acquisition change is a performance parameter that captures the speed at which performance reaches an asymptote. Second, the asymptotic value is a practice parameter variable that indicates where performance plateaus. Third, the change in performance, or amplitude, is calculated by subtracting the performance plateau from the initial starting performance value (Cousineau & Lacroix, 2006). These three parameters provide an overall performance curve profile, which complements information about change in performance that cannot be reviewed when comparing the mean performance at discrete points across practice days. Nonlinear skill acquisition parameters may exhibit distinct relationships with long-term motor learning and neurophysiological measures that are dependent on the task and practice conditions. For these reasons, in addition to evaluating conventional performance and learning measures, performance 35 curve parameters offer more detailed information on motor sequence learning. Conventionally, the shape of performance curves are negatively accelerated, which depicts improved performance over time, but with lessening returns (Ritter & Schooler, 2002). However, smooth monotonic, S-shaped and stepped curves have all been used to describe the learning process (Glautier, 2013). While the heterogeneity of performance curves may contribute to their underutilization, applying a priori theoretical understanding increases the accuracy of learning-curve predictions (Glautier, 2013). For example, exponential functions have proven to be more suitable than the power function when investigating individual skill acquisition (Heathcote et al., 2000). Before this finding, the “Power Law of Practice” dominated the field of skill acquisition, which modelled learning according to a power function. Newell and Rosenbloom (1981) analyzed the relationship between practice and task completion time in existing data sets across a range of tasks at the group level (Newell & Rosenbloom, 1981). However, Heathcote et al. (2000), modelled data at the individual level across a range of tasks and concluded that the exponential function was a more accurate depiction of the data and provided a superior fit. The three-parameter-exponential function better describes learning than the three-parameter-power function when modelling individual skill acquisition. Examples of single-subject performance data fitted to power and exponential function (two parameter functions) are illustrated in Figure 6 (below). However, the parameters that result from the fitting of exponential functions to individual datasets must be interpreted with caution, as a general rate of learning does not exist and any relationship between the components of change is both individual and task-specific (Newell et al., 2001). In the motor learning literature, performance curves are used to: (1) compare the effect of practice manipulations on motor performance (Lang & Bastian, 2001); and (2) infer learning (Mang et al., 2016). An area that is less understood is 36 how the shape of performance curves during practice and the extracted parameters reflect and relate to long-term learning during a delayed retention test in healthy individuals and individuals with chronic stroke. Quantifying the phases of motor learning: The performance curve is described as an evaluation of an individual’s ability by plotting their scores across the phases of practice to measure improvements (Cronbach, 1963). While performance curves provide a reliable pattern of improvement, few researchers have utilized mathematics to characterize this pattern across the phases of practice. In addition to the practice parameters, to date, attempts to quantitatively define the phases of motor learning have been limited (Kleim et al., 2004). The most common approach to performance curves is an arbitrary division of practice into halves or thirds, and the comparison of outcomes between earlier and later periods, as illustrated in Figure 5 below (Feldman, Cao, Andalib, Fraser, & Fried, 2009). Documentation of progression through the phases may highlight individual differences in performance that are related to cognitive and motor abilities, and to long-term retention of motor skills (Muratori et al., 2013). However, the approaches above are not individualized approaches and it is important to develop a mathematical method to quantitatively individualize the practice phases based on performance curves. A novel approach to quantify individual phases of motor skill acquisition to address this gap in the literature is outlined in Chapter 2. 37 Figure 5: Phases of motor learning. The phases of motor skill acquisition are conventionally divided into thirds based on the number of trials. This figure plots individual performance curves for response time(s) across 180 trials. Therefore, each phase of acquisition would be 60 trials. However, because individuals learn at different rates, it is possible that the duration of their acquisition phases may differ. Figure 6: Power versus exponential curves. This is an example of single-participant, motor practice performance data from the discrete pairing task (DPT). The response time data was fit to both an exponential (dashed line) and power function (dotted line) to visually demonstrate differences between these approaches. Power Function: y = 1.1581x-0.208R² = 0.35838Exponential Function: y = 0.5554e-0.001xR² = 0.3502800.20.40.60.811.21.41 51 101 151 201 251 301 351 401 451Response Time (s)Trials, nSingle Subject Motor Perfermance Data 38 Performance curves in rehabilitation: The use of performance curves to assess changes in motor behaviour is underutilized in rehabilitation (Cousineau & Lacroix, 2006). Yet understanding the parameters of performance curves (rate of skill acquisition, asymptotic and amplitude values) has implications for motor learning in rehabilitation settings with patient populations. For example, in the early stages of recovery after a stroke, the amount of neurologic and functional recovery is described as following a curved shape, with a gradual decay in returns over time. Indeed, the amount of eventual clinical functional recovery evolves over time and can plateau at any stage (Kaplan, Cailliet, & Kaplan, 2003); however, the point at which this plateau occurs is highly variable across individuals. In the chronic phase of stroke, the use of the performance curve has predominantly been used to assess motor skill acquisition and compare rates of improvement during practice. The comparison is typically between different practice conditions (Lang & Bastian, 1999, 2001), between individuals with stroke and healthy controls (Deuschl, Toro, Zeffiro, Massaquoi, & Hallett, 1996; Martin, Keating, Goodkin, Bastian, & Thach, 1996), or among different neurological populations (Ioffe, Ustinova, Chernikova, & Kulikov, 2006). The use of performance curves in laboratory settings offers insight into how patients relearn motor skills following stroke. However, to meaningfully interpret performance curve parameters, and increase their clinical exposure, we must establish the relationship between the rate of skill acquisition in practice and long-term retention across conditions, skills, and populations. Critical questions need to be answered for the implementation of performance curves in rehabilitation settings, including the following. How do different practice conditions, tasks, and skill levels affect the performance curve? How do practice variables relate to performance at a delayed retention test? Can we predict performance curves in individuals with stroke based on 39 residual brain structures? Can we develop individualized practice conditions based on curve parameters? For example, performance curves could be used to inform the magnitude of change in motor behaviour, as the “cost” of doing repetitive tasks generally decreases while experience is gained (Kaplan et al., 2003). Thus, using performance curves to characterize motor behaviour in rehabilitation, or motor behaviour associated with motor learning, may offer a method by which change can be robustly characterized, modelled, and tracked. Such an approach could increase the efficiency of existing rehabilitation practices, enhance the rate and degree of motor recovery in certain individuals, and improve the utilization of resources in clinics and the affordability of motor rehabilitation for outpatients. Neurophysiology of motor learning The compilation of decades of neuroimaging from motor learning experiments led to the development of several neurophysiological theoretical models of motor learning (Dayan & Cohen, 2011; Doyon, Song, et al., 2002; Hikosaka, Nakamura, Sakai, & Nakahara, 2002). Two distinct neural networks that likely interact to support motor learning are the cortico-cerebellar and cortico-striatal networks (Doyon, Song, et al., 2002; Doyon, Ungerleider, et al., 2002; Ungerleider, Doyon, & Karni, 2002). These two networks are differentially recruited throughout phases of the motor learning process, and the level of engagement of these systems is dependent upon the nature of the task (i.e., motor sequence versus motor adaptation). In the early phase of motor learning, both the acquisition of motor sequence and motor adaptation skills recruit similar regions within the cortico-cerebellar and cortico-striatal systems, which include the frontal associative regions, motor cortical regions, parietal cortices, medial temporal lobe, striatum and cerebellar cortices (Doyon & Benali, 2005; Doyon, Song, et al., 40 2002). The involvement of this large network of regions is attributed to the establishment of movement patterns important for the acquisition of a new motor skill (Doyon & Benali, 2005). As learning progresses into the later phases, there is a divergence of cortico-cerebellar and cortico-striatal system involvement, depending on whether the task involves motor adaptation or motor sequence learning. During motor adaptation learning, with prolonged practice, within the cortico-cerebellar loop, when individuals enter the late phase of learning, a transfer of activity from the cerebellar cortex to the dentate nucleus is observed. Furthermore, when performance plateaus and automatization of a motor adaptation task is achieved, the striatum is no longer active (Doyon & Benali, 2005). Conversely, during motor sequence learning, within the cortico-striatal system, performance in the early phase of learning involves activation of the rostral associative striatum. In the late phase of learning, activation transfers to the caudal sensorimotor regions of the striatum. This shift, from the associative to sensorimotor striatum, is believed to reflect the spatial and motor representations of the sequence created in each area (Doyon et al., 2009; Doyon & Benali, 2005). Additionally, when performance plateaus and automatization of a motor sequence is achieved, activation of the cerebellum is no longer required. For motor sequence learning, retention of the sequence occurs strictly in the striatum (Doyon & Benali, 2005). A recent meta-analysis of functional magnetic resonance imaging (fMRI) studies systematically quantified time-dependent motor learning networks to provide an evidence-based framework of motor learning across time scales (Lohse, Wadden, Boyd, & Hodges, 2014). This approach effectively allowed for the evaluation of a shared network across tasks, experiments, and studies, increasing statistical power. In line with theoretical models of motor learning discussed here (Doyon & Benali, 2005; Doyon, Song, et al., 2002), for combined motor sequence 41 and adaptation tasks, findings from the meta-analysis quantitatively confirmed an early reliance on associative areas that included cortical regions of prefrontal and parietal areas, the cerebellar cortex, and the rostral areas of striatum. Later in the learning process, greater activity was observed in the cortical regions of the sensorimotor cortex, the deep nuclei of the cerebellum (dentate nucleus), and caudal areas of the striatum (Lohse, Wadden, et al., 2014). A schematic of such regions is shown in Figure 7, below. 42 Figure 7: Neurophysiology of motor learning. Early motor skill acquisition relies on the associative network compromised of the prefrontal, premotor, and parietal cortices, and associative areas of the basal ganglia and cerebellar cortex (black circles). In the later phases of motor learning, there is a shift towards activation of the sensorimotor network, including regions of the striatum and deep nuclei of the cerebellum (gray circles). Circles are scaled to represent an approximation of anatomical size of regions. Adapted from Lohse et al. (2014). Neurophysiology of motor learning in individuals with stroke Motor learning plays a central role in post-stroke motor recovery, and understanding how individuals with stroke learn to recruit neuronal resources and reconfigure motor networks is important for improving rehabilitation methods (Lefebvre et al., 2015). In the early (acute/subacute) phase of stroke recovery, movement of the hemiparetic limb results in compensatory patterns of cortical activation often characterized by the additional recruitment of the contralesional hemisphere (Foltys et al., 2003), and increased reliance on ipsilesional hemisphere associative regions, such as the somatosensory and premotor areas (Tombari et al., 2004; Ward et al., 2006). In the contralesional hemisphere, increased activation of motor and premotor areas is shown to support motor recovery (Calautti & Baron, 2003; Johansen-Berg et al., 2002). Also in the early phase of stroke recovery, depending on the degree of damage to the ipsilesional motor cortex and associative areas, inter-hemispheric motor cortical activity may be important for supporting movement through functionally compensatory activation (Jones, 2017). 43 However, with increasing recovery time and arm use following a stroke, individuals can alter their patterns of brain activation to facilitate improvements in motor function. From neuroimaging findings, following stroke rehabilitation, the lesioned brain preserves the capacity for plasticity (Jones, 2017). The unmasking of latent motor pathways, and/or reorganization of secondary motor networks of new compensatory cortical connections, are believed to be key mechanisms underlying improvements in motor function during post-stroke rehabilitation (Calautti & Baron, 2003). In the chronic phase of stroke, the aim of motor re-learning interventions is the reorganization of a sensorimotor network that results in a more balanced network of activity between contralesional and ipsilesional hemispheres (Johansson, 2000). Specifically, during movement of the hemiparetic limb, there is a transcallosal imbalance of activation between motor cortices; the contralesional hemisphere sends greater inhibitory signaling to the ipsilesional hemisphere. Disruption in transcallosal pathways resulting in an interhemispheric imbalance is associated with reduced motor function, and disinhibition of the contralesional hemisphere is believed to be an important mechanism underlying motor recovery (Johansson, 2000). This finding, that damage to the ipsilesional hemisphere disrupts inhibitory signaling between hemispheres, has been given rise to the development of the “interhemispheric competition model”, which describes the ipsilesional hemisphere as “double-disabled”, due to the excess inhibition from the contralesional hemisphere that suppresses activity in an already inhibited ipsilesional cortex (Di Pino et al., 2014). Non-invasive neuromodulatory techniques (i.e., repetitive non-invasive brain stimulation) are being developed to test the efficacy of this model in clinical laboratory settings on post-stroke motor recovery (Chapter 5). 44 Stroke and motor sequence learning: Brain reorganization likely supports improved hemiparetic arm motor function following stroke (Nudo, Wise, SiFuentes, & Milliken, 1996), as opposed to true recovery of lost or damaged neural tissue (Krakauer, Carmichael, Corbett, & Wittenberg, 2012). However, few studies have investigated neuroplastic changes underlying motor sequence learning following stroke in humans (Boyd et al., 2010; Meehan, Randhawa, et al., 2011). Based on fMRI findings from human neuroimaging research, two studies investigated the neural correlates of motor sequence learning in individuals with chronic stroke. First, in an fMRI study, Boyd et al. (2010) investigated whether motor sequence learning could promote functional reorganization of the contralesional and ipsilesional motor cortices (Boyd et al., 2010). Individuals with middle cerebral artery stroke were randomized into one of two groups: (1) a motor sequence learning group that practiced repeated and random sequences embedded in the STT; or (2) a general arm-use group that performed a non-specific training task delivered by a physical therapist. Both groups had three sessions of practice, with the number of movements equated for both groups. Activation within the primary motor cortex of both hemispheres was analyzed 24 hours prior to the start of practice (baseline) and 24 hours after the last day of practice (delayed retention test). A laterality index (LI), which represents the ratio of the volume or intensity of an activity in the motor cortices, was calculated. Individuals in the motor sequence learning group demonstrated more change in motor behaviour and neurophysiology, as shown by differences in response times during repeated versus random sequences in the STT during the delayed retention test, and a reduced volume of contralesional primary motor cortex (M1) cortical activity (signified by a shift in the LI during repeated sequence performance), respectively (Boyd et al., 2010). There was no shift in LI during the random sequence performance. Furthermore, the general arm-use group showed no change in the intercortical 45 relationship, as shown by the lack of LI shift (Boyd et al., 2010). These data suggest that motor sequence learning, rather than increased hemiparetic arm use, shifts patterns of brain activity after stroke. To obtain further insight into the reorganization of the brain following motor sequence learning post-stroke, the second study, performed by Meehan et al. (2011), utilized an fMRI study aimed at comparing whole-brain activation between healthy individuals and individuals with a subcortical stroke during performance of the CTT. Individuals with subcortical stroke demonstrated the capacity for motor sequence learning; however, the brain activation patterns between groups differed. Healthy individuals showed increased activation of the dorsolateral prefrontal cortex (DLPFC) in the early phase of learning, which decreased with practice and improvements in performance (Meehan, Randhawa, et al., 2011). Comparatively, individuals with stroke did not demonstrate this decrease in DLPFC activation, and continued to activate prefrontal areas following seven days of motor sequence practice. These results suggest that, following extensive practice, the CTT continued to be highly attention-demanding for individuals with stroke (Meehan, Randhawa, et al., 2011). In addition, based on models of motor learning, it is possible that the individuals with stroke remained in an earlier phase of learning, even at the end of the study, and required additional practice to decrease their reliance on prefrontal areas (Meehan et al., 2011). Linking behaviour and neurophysiology, as well as the use of performance curves instead of providing simple “pre– to post–” assessments of motor behaviour, would have shown whether performance in the stroke group had plateaued or not, and would have provided evidence as to when to stop or continue practice. White matter structure predicts motor recovery and learning: Following stroke, damage to white matter (WM) pathways of the descending corticospinal tract (CST) (Borich, Mang, & 46 Boyd, 2012; Mang et al., 2015; Schaechter et al., 2009; Stinear et al., 2007) and transcallosal pathways (Borich, Mang, et al., 2012; Gupta et al., 2006; Jang, 2009; Lindenberg & Seitz, 2012; Radlinska et al., 2012) have been observed. There are also numerous studies demonstrating associations between the degree of WM tractography disruption and motor function and impairment outcomes in individuals with stroke (Auriat, Neva, Peters, Ferris, & Boyd, 2015; Borich, Mang, et al., 2012; Gillard et al., 2001; Jang, Cho, et al., 2005; Jang, You, et al., 2005; Kusano et al., 2009; Maeda, Ishizaki, & Yura, 2005; Mang et al., 2015; Yang et al., 1999; Yoshioka et al., 2008). Similarly, there is increasing evidence that characterization of WM motor pathways, through diffusion-weighted imaging (DWI), with a specific concentration on the CST, is a valuable tool to predict long-term outcomes in individuals with stroke (Ward, 2015a). For example, moving towards the use of neuroimaging parameters for prognostic predictions of recovery, Stinear et al. (2007, 2012) developed the predicting recovery potential (PREP) algorithm to stratify individuals in the acute phase of stroke into one of four levels (“complete”, “notable”, “limited”, “none”) of recovery in a stepwise fashion. The intention of the PREP algorithm was to help clinicians predict further improvements in motor function or impairment in chronic stage of stroke. In this algorithm, the first step is the assessment of upper extremity impairment by means of the sum of the shoulder abduction and finger extension (SAFE) test in the first 72 hours after stroke (Stinear et al., 2012). If a score of eight or above (maximum 10) is achieved on SAFE, then the individual is stratified into the “complete” recovery level. If the individual scores below eight, further assessment is required to detect the presence of motor evoked potentials (MEPs) from the upper extremity in response to single-pulse transcranial magnetic stimulation (TMS). If MEPs are present, the individual is stratified into the “notable” recovery level. If no MEP is present, an asymmetry index of fractional anisotropy (FA) of the 47 ipsilesional and contralesional posterior limbs of the internal capsules (PLICs) is derived from DWI (Stinear et al., 2012). If the asymmetry index is less than 0.15, then individuals fall into the “limited” level of recovery; if the asymmetry index is greater than 0.15 individuals fall into the “none” level of recovery, and hence are believed to have little capacity for recovery (Stinear et al., 2007). In real world clinical settings, the use of the PREP algorithm to determine individual rehabilitation potential has been shown to correctly predict 85% of patients’ upper extremity function at 12 weeks post-stroke (Stinear et al., 2016). Combining behavioural and neurological measures to predict prognosis of upper extremity recovery thus has significant implications in the effectiveness of individualized motor learning interventions following stroke. To date there is only one study associating WM tract integrity and changes in motor sequence learning following a stroke (Borich et al., 2014). Borich et al. (2014) investigated the relationships between post-training WM integrity of the ipsilesional PLIC and learning of a motor sequence task in 13 individuals with chronic stroke. Individuals completed five sessions of CTT practice and changes in motor behaviour were calculated by comparing the baseline performance with the results of a delayed retention test (Borich et al., 2014). Using hierarchical multiple linear regression analyses, the authors showed that age, time post-stroke, and FA from the ipsilesional PLIC significantly predicted the magnitude of change associated with motor learning (Borich et al., 2014). These findings indicate that the FA of the ipsilesional motor tracts is associated with response to motor learning and suggest that this may be an important biomarker with which to predict individual responses to motor rehabilitation interventions. Neurophysiology assessments of motor learning A major aim of this dissertation is to investigate reorganization of gray matter (GM) and WM following motor skill practice in healthy individuals and individuals with stroke. The field 48 of neurophysiology is rapidly advancing, with marked improvements in non-invasive techniques for acquiring information on GM and WM microstructural properties of the human brain. This section will introduce neurophysiology assessments used to investigate neural properties associated with motor learning–related change. Specifically, we will discuss methods used to: (1) evaluate GM functional activity of brain regions and networks (fMRI); and (2) determine the integrity of WM tracts and networks (DWI). Functional magnetic resonance imaging Motor learning is moderated by changes in the response of an ensemble of neurons stemming from activity-dependent input to the nervous system, known as neuroplasticity (Ma et al., 2010). Understanding neuroplasticity in humans has been significantly advanced by the non-invasive measurement of oxygenation in the brain using magnetic resonance imaging (MRI). Magnetic susceptibility increases between blood vessels and surrounding tissues when greater amounts of paramagnetic deoxygenated hemoglobin (dHb) molecules are present (Vazquez & Noll, 1998). This can be converted into an interpretable signal from the blood-oxygen-level-dependent (BOLD) contrast via a radio frequency transmitter and receiver coil. This process forms the basis of fMRI, which is the focus of Chapter 4 of this dissertation. Interestingly, the BOLD signal increases following neuronal activation, likely because of a large compensatory increase in cerebral blood flow that oversupplies oxygenated blood to the region. The BOLD response has been closely associated with neuronal activity, though with a delay of approximately three to six seconds for peak signal change (Logothetis, 2003). Most fMRI-based motor learning research applies a univariate analysis approach that is a voxel-wise technique used to assess independent activity in regions of the brain. This method, known as functional segregation, assesses specialized regions of the brain that are clustered 49 together to infer common functional properties (Friston, 1997). Each voxel is analyzed as a separate entity of the brain, and findings are reported as activation within anatomically distinct regions associated with the task (Dayan & Cohen, 2011). However, due to advances in fMRI statistical analysis methods, there has been an increase in multivariate analysis approaches, which are beneficial for determining patterns of activation within a functional network of spatially discrete but anatomically connected regions associated with a task. This will be a focus of Chapter 4, which assesses activation patterns between multiple brain regions (known as functional integration). Functional connectivity: Functional connectivity is an fMRI multivariate method of analysis that investigates patterns of inter-regional interactions. Two methodological approaches are typically employed: (1) model–driven models; and (2) data–driven models. The model-driven models used to assess functional connectivity employ a regions of interest (ROIs)-based analysis, which involves the delineation of a cluster of voxels from previous fMRI univariate coordinates or theoretical models associated with the behavioural processes of interest (i.e., motor learning) (Coynel et al., 2010; Sun, Miller, Rao, & D'Esposito, 2007; Zhang et al., 2012). Data-driven models do not rely on prior knowledge of activation, and do not constrain the analysis to a specific set of voxels, but rather use a whole-brain approach to assess functional integration (Tamas Kincses et al., 2008). Both model– and data–driven models of functional connectivity analysis have been used to investigate patterns of connectivity associated with resting-state and task-based networks (Coynel et al., 2010; Tamas Kincses et al., 2008). The focus of Chapter 4 is to investigate task-based, motor learning–related functional brain networks in healthy individuals and individuals with stroke, through both a model- and data-driven functional connectivity analysis known as “constrained principal component analysis” (CPCA) (Woodward et al., 2006). 50 The continuous development of multivariate techniques, such as CPCA, allows for a greater understanding of functionally integrated neurophysiological systems that operate during motor sequence learning. Constrained principal component analysis: CPCA (or fMRI-CPCA) uses a statistical analysis method that amalgamates multivariate multiple regression and principal component analysis (PCA) techniques in a unified framework (Woodward et al., 2006). fMRI-CPCA differs from other approaches that examine activation correlations among brain regions, in that it identifies functional networks that are based on task-related covariance/correlation in the BOLD signal (Woodward et al., 2006). Previous studies using these standard methods, such as independent component analysis (ICA) and PCA, have had two main difficulties: (1) relating brain networks to the timing of behavioural tasks carried out while subjects are in the scanner; and (2) simultaneously comparing activity between groups or conditions using one task-related network (Woodward et al., 2006). When applied to fMRI, CPCA isolates relevant BOLD signal fluctuations to derive the degree to which one or more related functional networks are involved in each experimental condition for each subject (Woodward et al., 2006; Woodward, Feredoes, Metzak, Takane, & Manoach, 2013). During fMRI-CPCA, the BOLD signal for each voxel is modelled to the hemodynamic response (HDR) shape. This method isolates patterns in the signal that summarize the data and extract networks that underlie motor and cognitive processes (Woodward et al., 2013). fMRI-CPCA has been used in past work to extract functional networks associated with a variety of cognitive-based tasks (Lavigne, Metzak, & Woodward, 2015; Metzak et al., 2011; Metzak et al., 2012; Woodward et al., 2006). Due to the importance of functionally integrated brain networks during motor sequence learning, fMRI-CPCA may 51 significantly contribute to our neurophysiological understanding of the networks that support motor behaviour in healthy individuals and individuals with stroke (Chapter 4). Functional connectivity methods in healthy individuals: The evaluation of the functional connectivity of motor learning–related networks through multivariate, compared to univariate, analyses, provides a differential and complimentary depiction of activation within the brain. For example, Sun et al. (2007) investigated task-based, time-series frequency coherence of the BOLD signal between multiple ROIs, including the M1, DLPFC, primary somatosensory cortex (S1), premotor cortex (PMC), parietal cortex (PC), and supplementary motor area (SMA) to determine functionally connected regions for within-session early compared to late motor learning of a bimanual variant of the SRT. Compared to later performance of the sequence task, significantly greater interhemispheric connectivity between M1 and bilateral PMC, SMA, and PC, and between the PMC and DLPFC, was observed in the early phase of motor sequence learning. Interestingly, there were no regions that demonstrated significantly greater connectivity for late compared to early learning of a novel task (Sun et al., 2007). The authors concluded that changes in cortico-cortical connectivity are more pronounced in the early phases of learning of a motor sequence task (Sun et al., 2007). Tamas Kincses and colleagues (2008) used an alternative task-based, functional connectivity analysis known as ICA to investigate whole-brain connectivity networks, rather than connectivity between a priori ROIs. In this study, authors observed an M1-premotor-parietal-cerebellar circuit to be involved in early learning of the repeated sequence during SRT performance. Additionally, improved performance of the repeated sequence was related to decreased functional connectivity of this network, which was interpreted as an improvement in network efficiency associated with motor learning (Tamas Kincses et al., 2008). A second 52 network, the posterior parietal-premotor circuit, demonstrated increases in functional connectivity throughout the early learning phase, during repeated versus random sequence performance. This network also correlated with improvements in reaction times, proposed to reflect the engagement of spatial attention, working memory, and processing required for motor sequence tasks (Tamas Kincses et al., 2008). In a combined methodological approach with which to investigate the long-term activity-dependent plasticity of networks, Coynel et al. (2010) used a task-based, ROI-based ICA functional connectivity analysis to examine changes in networks over four weeks of motor sequence learning. The authors delineated ROIs belonging to two sub-networks defined by a previous model (Hikosaka et al., 2002), which included the associative/premotor and sensorimotor networks. During performance of a variant of the SRT, increases in connectivity within regions of the associative/premotor network were observed during early motor learning, followed by within-network decreases in late motor learning. This decrease in network functional connectivity over time was accompanied by increases in connectivity between the associative/premotor and sensorimotor networks. For the sensorimotor network, the within-region connectivity remained unchanged throughout learning (Coynel et al., 2010). This study is of interest because the findings contradict theorized motor learning models that hypothesized an increase in the sensorimotor network in the later phase of motor learning (Hikosaka et al., 2002). Taken together, these studies reveal that motor learning–related functional brain networks provide an additional degree of understanding of the communicative complexity between regions that support motor skill acquisition and learning (see Table 1, which follows, for summary of the three studies detailed above). 53 Table 1: Summary of three examples of functional connectivity analyses DLPFC = dorsolateral prefrontal cortex; M1 = primary motor cortex; PC = parietal cortex; PMC = premotor cortex, primary somatosensory cortex (S1). ⬇ = decrease in functional connectivity ⬆ = increase in functional connectivity Functional connectivity methods in stroke: To date, there has been minimal task-based functional connectivity analysis associated with motor learning following stroke. Task-based functional connectivity analysis of motor learning-related neural networks requires individuals with stroke to perform a motor task inside a magnetic resonance (MR) scanner, which results in many challenges (due to motor synergies, mirror movement, and head motion), which tend to reduce signal-to-noise ratios. Nonetheless, analyzing patterns of network activity is required to determine the effect of a lesion on brain activity, and to illustrate optimal compensatory network patterns (Auriat et al., 2015). Functional Connectivity StudiesExamples of MultivariateAnalysisRegion-wise analysisTask/Resting StateFindingsSun et al. 2007 Coherence Maps DLPFC, S1, M1, PMC, PC Motor Sequence Task⬆ Inter and intrahemisphericconnectivity between ROIs(early versus late)Tamas Kincses et al. 2008 Independent Component Analysis (ICA)Whole Brain Motor sequencetask⬇M1-premotor-parietal-cerebellar network connectivity (late versus early); ⬆posteriorparietal-premotor circuit connectivity (early versus late)Coynel et al. 2010 Independent Component Analysis (ICA)AssociativeNetwork; Sensorimotor NetworkMotor sequence task⬇ Associative network (lateversus early) ⬆Associative and sensorimotor between network(late versus early) 54 Grefkes et al. (2008) was one of the only teams of researchers to investigate functionally connected networks during motor sequence learning following stroke. To do this, they applied a directional form of functional connectivity analysis known as dynamic causal modeling (DCM). DCM is a model-driven approach that estimates interactions in a pre-defined network of brain regions as well as the influence that one area exerts over another (i.e., the specific type of functional connectivity, known as “effective connectivity”) (Grefkes et al., 2008). Changes in the motor network during the rhythmic opening and closing of the left or right fist following subcortical stroke were studied. In the sub-acute phase of stroke (approximately 10 weeks), during unimanual hand movements, inhibitory influences from contralesional to ipsilesional M1 were observed, and this additional inhibition was positively correlated with motor dysfunction when individuals with stroke moved their hemiparetic hand (Grefkes et al., 2008). During bimanual hand movements, there was an increase in interhemispheric connectivity from SMA to M1 in both hemispheres. Findings from this study not only demonstrate the relationship between altered connections within and between hemispheres and motor recovery post-stroke, but also the directional influence of the contralesional hemisphere on the ipsilesional hemisphere during movement (effective connectivity). The investigation into the motor network associated with motor sequence learning (as opposed to generalized movement) in healthy individuals and individuals with stroke is an important next step to understand the patterns of brain activation that support learning (Chapter 4). Diffusion-weighted imaging DWI is an imaging modality that quantifies the integrity (anisotropy and diffusivity) of WM structures in the brain following stroke (Auriat et al., 2015; Borich et al., 2014; Mang et al., 2015). The pathways of water molecules in the brain are restricted by the structure of GM and 55 WM. DWI is sensitive to signals from water molecules and captures different structures of the tissues where water molecules diffuse based on the directionality of water diffusion (Basser & Jones, 2002). In individuals with no neurological pathology, WM pathways are highly organized, with tightly packed axons that allow water molecules to diffuse parallel to the long axis of the fibre bundles (Thiel & Vahdat, 2015). These highly organized bundles create a barrier for isotropic diffusion, and water movement along these bundles can be measured in amplitude and direction to provide an indication of the diffusivity properties of the pathway of interest (Basser & Jones, 2002). Based on the unrestricted movement of water molecules in all three spatial directions (x, y, z), DWI uses diffusivity information to measure microstructural integrity and create a three-dimensional model of WM tracts, often characterized by a diffusion tensor model (Thiel & Vahdat, 2015). The most frequently reported diffusion measure is FA, which specifies the degree of water-movement directionality within the tissue, as determined by tissue features such as axons, myelin, and microtubules (Auriat et al., 2015). FA is a unitless measure that can range from 0 (isotropic) to 1 (anisotropic), with a higher FA value representing a greater structural alignment of fibre tracts and a lesser FA value suggesting aberrant tissue organization, tissue damage, or inflammation (Alexander, Lee, Lazar, & Field, 2007; Mang et al., 2015). Diffusion-weighted imaging methods in stroke: Motor recovery after stroke is related to the structural remodeling of WM (Schaechter et al., 2009). In the acute phase after stroke, FA of the CST is decreased (Puig et al., 2013; Puig et al., 2011; Radlinska et al., 2010). These decreases are hypothesized to result from primary damage from the ischemic insult, and secondary damage from apoptosis, inflammation, diaschisis, and neurodegeneration. Microstructural properties of brain structure after stroke are impacted by Wallerian degeneration (Thomalla et al., 2004), which is the anterograde degeneration of axons and associated myelin 56 sheaths. Compared to healthy individuals, reduced FA and tract volume in individuals with chronic stroke is thought to be the result of WM atrophy that is secondary to the occurrence of Wallerian degeneration and distal to the original locus of the infarct (Thiel & Vahdat, 2015). The ipsilesional M1 sends efferent signals to the periphery via the CST via direct pyramidal synapses, and is likely influenced by secondary motor areas and cerebellum regions via descending motor pathways such as the reticulospinal or rubrospinal tracts (Isa, Ohki, Alstermark, Pettersson, & Sasaki, 2007). However, few researchers have evaluated WM pathways outside the CST or CC to assess the influence of structural damage caused by stroke on predictors of motor recovery, despite the converging evidence for complex movement and motor learning-related functional brain networks (Kumar, Kathuria, Nair, & Prasad, 2016; Li, Wu, Liang, & Huang, 2015; Mang et al., 2015; Puig et al., 2013; Stewart et al., 2017; Ward, 2015a). Even fewer studies have adopted a multimodal approach to characterize brain recovery after stroke by combining information from both DWI and fMRI measures, with two attempts existing only as case studies (Caria et al., 2011; Jang, You, et al., 2005). It is important that a more comprehensive understanding of the complex network reorganization that includes interactions between GM and WM after stroke occurs. Integrative methodological approaches that combine neuroimaging approaches, such as fMRI and DWI, may be a promising, logical next advancement to the field (Chapter 5). Adjunct therapies to facilitate motor recovery and learning First incepted for human research in 1985 (Barker, Jalinous, & Freeston, 1985), TMS is a noninvasive brain stimulation technique used to artificially and extrinsically stimulate neural tissue (Rossi, Hallett, Rossini, & Pascual-Leone, 2009). Based on Faraday’s electromagnetic induction principle, TMS involves the use of rapidly switching magnetic fields to elicit electrical 57 current in cortical and (depending on stimulus intensity) corticospinal neurons (Rossi, Hallett, Rossini, & Pascual-Leone, 2009). TMS can also be delivered in trains, known as repetitive TMS (rTMS). Delivering repetitive bouts of TMS pulses (rTMS) at predetermined intensity with a specific frequency and pattern can excite or inhibit a local cortical region to influence network activity and even behaviour (Ridding & Rothwell, 2007). There are two primary methods of rTMS-based neuromodulation proposed to help facilitate recovery following stroke. The first is aimed at increasing cortical excitation of the ipsilesional hemisphere by delivering high-frequency rTMS over an ipsilesional brain region, to counteract the excessive inhibitory transcallosal signaling from the contralesional hemisphere. Multiple TMS pulses are delivered at a high frequency, typically 2 Hz or more (often up to 20 Hz), over a cortical site of interest (Pascual-Leone, Valls-Solé, Wassermann, & Hallett, 1994). The second proposed method of stimulation to help facilitate recovery following a stroke is low-frequency (typically 1 Hz or less) inhibitory rTMS over the contralesional hemisphere to suppress contralesional excitation and excessive inhibition signaling to the ipsilesional hemisphere (Takeuchi, Chuma, Matsuo, Watanabe, & Ikoma, 2005). In order to induce transient cortical inhibition a low frequency stimulation of 1 Hz or less is delivered (Chen et al., 1997; Fitzgerald, Brown, Daskalakis, Chen, & Kulkarni, 2002). A relatively novel rTMS protocol that is much shorter, yet just as effective, involves patterned stimulation; theta burst stimulation (TBS) also inhibits or facilitates cortical excitability, but requires substantially less time to deliver the same number of pulses. For both inhibitory and excitatory neuromodulation effects, pulses are delivered in 5-Hz stimulation patterns with triplets of 50 Hz stimulation, though intermittently (2 s on, 10 s off) for facilitation (facilitatory, intermittent TBS [iTBS]) and 58 continuously for inhibition (inhibitory, continuous TBS [cTBS]) (Huang, Edwards, Rounis, Bhatia, & Rothwell, 2005). Functional activation of motor networks is important for motor skill acquisition following stroke (Wadden et al., 2015). The rationale for applying repetitive, non-invasive brain stimulation (rTMS, TBS) over the ipsilesional or contralesional hemisphere is to stabilize the neural activity of lesioned brain to augment motor recovery processes when paired with motor practice (Meehan, Linsdell, Handy, & Boyd, 2011), providing grounds for a promising adjunct therapy in motor rehabilitation. Increasing cortical excitation of the ipsilesional hemisphere or decreasing cortical excitation of contralesional hemisphere by delivering differing rTMS protocols has shown positive improvements in motor recovery following a stroke (Hsu, Cheng, Liao, Lee, & Lin, 2012). Facilitatory rTMS To investigate the effects of facilitatory rTMS on motor learning–related changes, in a crossover design Kim et al. (2006) delivered either 10 Hz or sham rTMS over the ipsilesional M1 in 15 individuals in the chronic phase of stroke. The intervention consisted of eight bouts of high frequency (or sham) rTMS, followed by motor sequence practice with the paretic hand (repetitively pressing buttons in response to a seven-digit number stimulus), and a short period of rest (Kim et al., 2006). Ten MEPs were measured before and immediately after each bout of rTMS stimulation. Compared to sham stimulation, high-frequency rTMS produced larger increases in corticospinal excitability as measured by MEP amplitude, and enhanced motor skill performance was demonstrated for both repeated sequence movement time and accuracy with the high-frequency rTMS having larger increases in both movement scores than sham rTMS (Kim et al., 2006). The authors hypothesized that pairing rTMS with motor practice amplified the 59 remodeling or unmasking of dynamic neural motor substrates associated with repeated motor practice. More importantly for stroke, rTMS paired with motor practice indirectly affected cortical excitability through inhibitory transcallosal pathways to the contralesional hemisphere, aiding in enhanced motor performance (Kim et al., 2006). Brodie, Meehan et al. (2014), observed similar improvements following 5 Hz rTMS delivered over ipsilesional S1 when paired with motor skill practice in the chronic phase following stroke. In a partial crossover design, 15 individuals with chronic stroke received 5 Hz rTMS and 11 individuals with chronic stroke received sham TMS, followed by six blocks of STT practice with the hemiparetic upper extremity. In this study, motor performance was assessed during the five-day stimulation intervention, and was separately evaluated on day one and seven (pre– and post–tests) without stimulation (Brodie, Meehan, et al., 2014). The delayed retention test was performed 24 hours following the last stimulation session (post-test), and in addition to evaluating motor learning, cutaneous somatosensation, motor function (Wolf Motor Function Test [WMFT] and Box and Block Test [BBT]), and corticospinal excitability (resting motor threshold [RMT]) were assessed (Brodie, Meehan, et al., 2014). Individuals who received 5 Hz rTMS over ipsilateral S1 showed greater improvements in STT performance, as demonstrated by decreases in response time, increases in peak velocity, and decreases in cumulative distance across all sessions. Specific to motor learning, increases were observed in response time on the delayed retention test during repeated sequence performance compared to the sham rTMS group (Brodie, Meehan, et al., 2014). However, there was no transfer to paretic arm motor function, as demonstrated by the null difference between groups for the WMFT and BBT. Due to the complex interaction between the sensory and motor systems during motor learning, stimulation over ipsilesional S1 may aid in motor skill retention in individuals with chronic stroke, possibly 60 due to increased functional connectivity between the ipsilesional S1 and M1 during sensory guided movements (Brodie, Meehan, et al., 2014). Inhibitory rTMS To investigate the effects of inhibitory rTMS on motor learning–related changes in motor skill performance, Meehan et al. (2011) delivered three days of cTBS over the S1 or M1 of the contralesional hemisphere, compared to sham stimulation, paired with STT practice. Twelve individuals with chronic stroke were randomly assigned to one of three groups (contralesional S1 [S1c], contralesional M1 [M1c], or sham stimulation) (Meehan, Dao, et al., 2011). Pre– and post– tests were delivered on separate days with stimulation paired with motor skill practice to assess motor learning–related changes on STT performance. There were observed between-groups differences: M1c cTBS resulted in significantly greater decreases in peak velocity and peak acceleration, compared to both sham atimulation and S1c cTBS. Interestingly, S1c cTBS yielded larger reductions in time to initiate movement and time for completing the WMFT, compared to sham stimulation and M1c cTBS. The authors hypothesized these differences to be the result of the differential contribution of M1 and S1 during motor skill learning (Meehan, Dao, et al., 2011). To build on the study above (Meehan, Dao, et al., 2011), Chapter 5 will investigate the long-term effect of M1c and S1c cTBS paired with motor skill practice over 5 days of practice. The goal of which is to determine if these differential effects continue to be observed over time using a composite measure of performance (i.e., response time total [RTT]). Responders and non-responders Neuromodulatory rTMS parameters have been established from observed changes in single– and paired–pulse TMS (Huang et al., 2005; Huang & Rothwell, 2004) as well as MRI activation of brain regions (Bestmann et al., 2004). However, the neural mechanisms underlying 61 these indirect observations are not clearly understood (Chen, Yung, & Li, 2003). While pairing motor skill practice with rTMS protocols has shown promising results (Auriat et al., 2015), significant inter-individual variability in responsiveness to rTMS exists (Maeda, Keenan, Tormos, Topka, & Pascual-Leone, 2000). Owing to the lack of mechanistic understanding, in combination with extensive variability in response, there are challenges when interpreting study findings, which has led to a delay in the advancement of optimal rTMS stimulation protocols for individuals with stroke (Brodie, Borich, et al., 2014; Carey et al., 2014; Meehan, Dao, et al., 2011). As much of this variability cannot be explained by simple demographics such as age, sex, time post-stroke, or stroke severity (Brodie, Borich, et al., 2014; Carey et al., 2014; Mang et al., 2015), the field has begun to move in the direction of understanding the varied responses to rTMS through neurological assessments of GM and WM volumetric and WM structural integrity (Brodie, Borich, et al., 2014; Carey et al., 2014; Demirtas-Tatlidede et al., 2015). Recent findings demonstrate that the residual architecture of the post-stroke brain may predict individuals’ responsiveness to rTMS paired with skilled practice (Ward, 2015b). Despite the group-level differences observed by Brodie, Meehan, et al. (2014), considerable variability in response to rTMS was noted, that led the authors to investigate neurological measures characterizing responders and non-responders (i.e., individuals who demonstrated improved motor skill learning compared to those with no improvements, respectively). Brodie and Borich et al. (2014) examined sensorimotor cortex morphology using MRI T1-weighted anatomical images by parceling the GM and WM volumes of the ipsilesional S1 and M1 cortices. Exclusively in the group that received 5 Hz high-frequency rTMS over S1 paired with motor skill practice (n = 11), the authors observed a significant positive association between the volume of WM in the ipsilesional S1, and motor learning–related change in skill 62 performance (Brodie, Borich, et al., 2014). In addition, age, as well as ipsilesional S1 GM, and WM volumes significantly predicted rTMS-induced motor learning–related changes in a regression model. Based on these findings, within-group variation may be the result of differences in underlying brain morphology, and WM volume in the cortex near the stimulation site of interest may be an important predictor for determining responders and non-responders to high-frequency 5 Hz rTMS paired with motor skill practice (Brodie, Borich, et al., 2014). Additionally, Carey et al. (2014) observed, in a group of individuals that positively responded (i.e., had improved motor function of the hemiparetic hand) to 6 Hz rTMS over the M1c, greater WM volume of the PLIC of the ipsilesional hemisphere, compared to those individuals that did not improve function. In addition to anatomical volumetrics, the integrity of WM tracts may provide valuable information about the response profile following rTMS paired with motor skill practice. Findings from a 1 Hz rTMS study over the M1c in individuals with mild-to-moderate stroke demonstrated a positive relationship between the residual structural integrity of the corpus callosum (CC) and improved motor impairment (Fugl-Meyer score) (Demirtas-Tatlidede et al., 2015) Findings from these studies highlight residual brain structure as a potential biomarker for rTMS protocols and suggest it may help to predict individual responsiveness to an intervention (Auriat et al., 2015; Lindenberg & Seitz, 2012). The use of advanced neuroimaging techniques that quantify the residual functional architecture of the brain may be useful in pre-stratifying individuals into a sub-group prior to an intervention, or retrospectively explaining variability in outcomes known to exist following an intervention (Chapter 5) (Ward, 2017). Aims of this thesis The overall objective of this dissertation is to examine novel behavioural and 63 neurophysiological measures associated with motor performance and learning in healthy individuals and individuals in the chronic phase of stroke. This was largely motivated by small-to-moderate effects of motor recovery observed following rehabilitation interventions in the chronic phase of stroke (Ward, 2017). Following stroke, the effects of motor learning interventions on recovery may be improved when we move away from the substandard “one-size-fits-all” approach. Thus, the purpose of the first two studies of the presented work was to examine and establish methods for predicting individualized intensity and dose of motor practice, to better facilitate motor sequence learning. Furthermore, building upon behavioural factors that influence motor learning, this dissertation aims to predict individual response to motor learning interventions in individuals with stroke. In the second two studies of this dissertation, a whole-brain GM motor network associated with motor skill learning was examined. Identifying the GM motor network permitted me to examine the corresponding residual WM motor network, used to determine response to cTBS over the contralesional hemisphere (S1c, M1c) paired with motor skill practice. To achieve this main objective, there are four general aims: (1) evaluating motor performance and learning; (2) measuring the efficacy of individualized motor practice; (3) evaluating motor learning neurophysiology; (4) identifying biomarkers of motor learning (see Figure 1 in the Preamble to this thesis). 1. Specific aim one: Evaluating motor performance and learning The first aim of this dissertation is the evaluation of motor practice and learning, with the specific objective of assessing individual rates of motor skill acquisition and their relation to learning. In rehabilitation settings, decisions about individualizing the dose and intensity of practice are often subjective in nature rather than data-driven (Pollock et al., 2014). In the 64 research presented in Chapters 2 and 3 of this dissertation, performance curves were used to model performance data during motor practice for two different motor sequence tasks. To prescribe individualized practice paradigms, due to the complex relationship between motor performance and learning, it was important to establish the task-specific relationship between curve parameters and motor learning–related change in task performance. Exponential curve functions were fitted to performance data over a set number of trials of the DPT in healthy individuals (Chapter 2), and the CTT in healthy individuals and individuals with stroke (Chapter 3), to establish the relationship between rate of motor skill acquisition and motor sequence learning. 2. Specific aim two: Measuring the efficacy of individualized motor practice The second aim of this thesis is to measure the efficacy of individualized motor practice, with the specific objective of prospectively prescribing individualized levels of difficulty and retrospectively determining individualized doses of practice, in order to promote motor sequence learning. As discussed, presently, level of difficulty and dose of practice are often generalized across population groups. When implementing motor learning-based rehabilitation, which is a specialized subcomponent of recovery, individualized motor learning paradigms should be considered in order to improve motor recovery. In Chapter 2, we used a learner-adapted algorithm to prospectively individualize the intensity of task practice in three practice conditions (low, moderate, and high difficulty) of the DPT, in order to examine the effect of difficulty (challenge point and CI) on motor performance in a delayed retention test. In Chapter 3, we retrospectively determined the individualized dose of practice of the CTT based on healthy individuals and individuals with stroke, and predicted performance plateaus derived from performance curves. 65 3. Specific aim three: Evaluating motor learning neurophysiology The third aim of this thesis is the evaluation of motor learning neurophysiology, with the specific objective of identifying whole-brain GM motor learning networks. Motor learning post-stroke has been predominately assessed using fMRI univariate analysis, or motor recovery has been assessed using multivariate analysis at rest. In Chapter 4, using a task-based fMRI multivariate analysis, I investigated GM motor learning-based functional brain networks in healthy individuals and individuals with chronic stroke, following motor skill practice of the CTT. In a secondary analysis, for individuals with stroke, I determined the relationship between brain activity in a constrained motor learning network and motor performance in a delayed retention test. In Chapter 5, using a multimodal neuroimaging approach, I applied the GM motor learning network (“constrained motor connectome” [CMC]) to evaluate the integrity of underlying residual WM motor learning network in individuals with chronic stroke. 4. Specific aim four: Identifying biomarkers of motor learning The fourth and final aim of this thesis is the identification of biomarkers of motor learning, with the specific objective of assessing the relationship between individuals’ capacity for motor learning and the integrity of their WM structural motor network. Presently, single WM pathways (i.e., CST, CC) are used to predict motor recovery in chronic stroke. In Chapter 5, I used performance curves to assess capacity for motor learning–related change, and to determine motor practice responders of cTBS over the contralesional hemisphere paired with motor skill practice. I used the integrity of a WM motor network (CMC) to determine the responder profile following cTBS paired with motor skill practice in individuals with stroke. 66 Chapter 2: Individualized challenge point practice informs motor sequence learning: More time in an early phase of practice benefits later retention Introduction Motor sequence learning occurs with repeated practice. Behaviour associated with repeated practice follows a nonlinear rate of improvement (Brown & Heathcote, 2003). In speed/accuracy–based tasks, individual movement times decrease and accuracy increases with practice. An improvement in motor performance reflects changes in task demands as individuals’ skill proficiency increases (Wright et al., 2016). The order in which motor skills are practiced affects how they are acquired and, importantly, how well they are retained (Cross, Schmitt, & Grafton, 2007; Shea & Morgan, 1979; Wright et al., 2016). These practice order effects have been captured through the concept of contextual interference (CI), which relates to the amount of task switching/interference associated with the practice of different motor skills. CI offers one method by which motor task difficulty can be manipulated. However, it has been suggested that task difficulty should be tailored to each individual, based on an optimal “challenge point” (Guadagnoli & Lee, 2004). Individuals have different cognitive and motor abilities that, in conjunction with the intrinsic demands of the task, uniquely influence motor skill acquisition (Guadagnoli & Lee, 2004). Our aim in the current studies was to operationalize and test the concept of an individual challenge point as a tool to study, and to potentially facilitate, motor sequence learning. 67 To accomplish this aim, we first examined individual performance curves during an initial acquisition session (Study 1). This allowed us to assess how the rate of skill acquisition, and the duration that individuals spend in various phases of practice, relate to measures of motor learning (i.e., retention). In a second study, we used practice parameters from Study 1 to devise individualized practice schedules with respect to the amount of practice within each phase (Study 2). We predicted that a slower rate of acquisition (i.e., more time in a cognitive phase of practice) would lead to better retention. Further, we expected that optimal challenge would be related to a moderate or high degree of task switching (keeping individuals in a cognitive phase of practice for longer). The challenge point framework (CPF) is a conceptual framework that describes motor learning–related changes in behaviour (Guadagnoli & Lee, 2004). The theoretical point at which an individual reaches a maximum potential for motor learning during practice is known as the “optimal challenge point” (Guadagnoli & Lee, 2004). The challenge point reflects an interaction between the skill level of the individual and task difficulty. Hypothetically, the challenge point can be modified to the skill level of the individual by manipulating task constraints. For example, challenge point can be influenced by the amount of switching between tasks within a practice session (so termed CI). To date, the notion of individualized optimal challenge points has been largely understood at the conceptual level, but has not been specifically studied in human behaviour (cf., Choi, Gordon, Park, & Schweighofer, 2011; Choi et al., 2008). Theoretically, individualizing practice to an optimal challenge point ensures that task processing demands do not exceed, or fall below, cognitive and motor abilities (Guadagnoli & Lee, 2004). One task constraint on learning that is related to optimal challenge is the degree of CI experienced during practice of two or more different skills. High CI occurs when multiple 68 variations of a task are practiced under high levels of variability (Magill & Hall, 1990). CI is most easily induced by altering the order in which motor tasks are practiced which increases the amount of switching between tasks (Lee & Magill, 1983). Randomizing the practice order of motor tasks constitutes the highest form of CI. The frequent switching in high CI is thought to affect the cognitive operations involved in evaluating, planning and assembling movements (Lee & Magill, 1983; Lee & Simon, 2004; Shea & Morgan, 1979; Shea & Zimny, 1983, 1988). Although the information processing demands associated with frequent switching typically impairs initial performance and slows the rate of skill acquisition, increased processing demands enhance performance at a retention test. This is in comparison to low CI conditions, where tasks are practiced in a repetitive, blocked order, which requires a small amount of task switching. Blocked practice often speeds the rate of skill acquisition and initial performance, but at a cost of poorer performance at a retention test. These blocked/random dissociations between practice and retention have been termed the CI effect. Random practice paradigms are considered to have “desirable difficulties”, as the high amount of switching increases information processing demands and facilitates retention of motor skills (Schmidt & Bjork, 1992). However, because individuals possess different cognitive and motor abilities, this type of randomized practice can create a level of challenge outside the optimal range for a particular individual (Guadagnoli & Lee, 2004). There is evidence that age, level of expertise, and task complexity moderate the effects of random practice (Albaret & Thon, 1998; French, Rink, & Werner, 1990; Hebert, Landin, & Solmon, 1996; Shea, Park, & Braden, 2006). Learner-adapted practice exploits the concept of tailoring the demands of the task to the skill of an individual. By accounting for each individual’s current performance, it is possible to dynamically modulate task difficulty (e.g., task-switching) only after a predetermined 69 performance criteria is met (Choi et al., 2008). There are two methodological classifications of learner-adapted practice: learner-controlled and computer-controlled (Choi et al., 2008). Learner-controlled practice allows individuals to choose their own practice schedule and tailor the difficulty to an individual’s perceived ability/challenge. When individuals choose how to order practice, rarely do they engage in as much task-switching as would be experienced with a random schedule, yet they show comparable retention benefits (Hodges, Lohse, Wilson, Lim, & Mulligan, 2014; Keetch & Lee, 2007). These findings have been demonstrated in keyboard- (Hodges et al., 2014) and mouse-controlled (Keetch & Lee, 2007) sequencing tasks. Alternatively, computer-controlled learning uses an algorithm to adjust difficulty during practice. Choi et al. (2008) used a visuomotor adaptation task and showed that those who practiced with computer-controlled difficulty conditions outperformed those who practiced randomly at a delayed retention test (Choi et al., 2008). Despite the success of this approach, the computer-controlled method has not received further study. Our current work was designed to test the efficacy of a computer-controlled, learner-adapted practice schedule on the acquisition of a motor sequence task. In contrast to Choi et al. (2008), who used a group mean reference value to adjust difficulty during practice, we employed individual-specific reference values determined from earlier practice at a related task. In Study 1 we quantitatively distinguished individual phases of learning to provide an indication of the cognitive and motor demands of the task (Ackerman, 1986; Hirano, Kubota, Tanabe, Koizume, & Funase, 2015). The information processing demands associated with each phase of skill acquisition are contingent on task performance, which is known as the “performance-resource function” (Ackerman, 1986). Thus, in Study 1, we extended previous studies that classified individualized practice performance curves with respect to the phases of 70 learning (Hirano et al., 2015), to identify three phases of practice. This resulted in an exponential description of performance during one massed practice session. We then quantified motor performance (response time) and trials spent in each of the three phases to determine their relationship to performance at a delayed retention test, a key index of motor learning (Schmidt & Lee, 1988). In Study 2, individual practice metrics from Study 1 were used to create three computer-controlled, learner-adapted conditions. The algorithm integrated the learner’s rate of motor skill acquisition and, based on the performance-resource function, the mean response time from each of the three phases of motor skill acquisition. This resulted in three individualized practice schedules of differing challenge (low, medium, and high difficulty). The three practice conditions acted to maintain individual performances near an individualized mean response time extracted from the three phases of practice in Study 1. We were able to use data from Study 1 to account for individual differences in task performance to test optimal challenge points. Thus, in Study 2, individuals were forced to stay near an individualized reference value through manipulation of task difficulty. Therefore, in our two studies, we evaluated the relationship between rate of skill acquisition and time spent in each practice phase and retention performance (Study 1) and used this individualized information to inform a learner-adapted practice algorithm for the learning of motor sequences, based on individually determined indices of challenge (low, medium, and high difficulty; Study 2). We tested two hypotheses. In Study 1, we hypothesized that a slower rate of motor skill acquisition and a significant amount of practice spent within the early cognitive phase of practice constituted “optimal” challenge and would result in improved scores on a retention test. This hypothesis was based on the idea that high amounts of switching keep individuals at a 71 cognitively demanding stage of practice (Cross, Schmitt & Grafton, 2007) and enhance long-term retention (Lee & Magill, 1983; Shea & Zimny, 1988; Wright et al., 2016). Indeed, in related work in our laboratory, we showed that individuals with the slowest rates of motor improvements at a semi-immersive virtual reality task that required skilled movements were the ones who performed best in later tests of delayed retention (Lakhani et al., 2016). In Study 2, we hypothesized that the individualized practice condition, which required participants to perform near a reference value that would keep response times slow and hence on a moderate or high switching schedule, would enhance motor learning. This hypothesis is constructed from CI literature that shows a beneficial learning outcome following more (rather than less) erroneous performance during practice (Magill & Hall, 1990). Study 1 — Methods Participants Fourteen healthy young adults, self-reported as right-handed, participated (mean [M] age = 24.35; standard deviation [SD] = 4.40 years; maximum [max.] = 34 years; 6 = female [F]). Participants were recruited from the University of British Columbia (UBC) campus. All participants provided written informed consent to the experimental procedures that was approved by the UBC Office of Research Ethics under the Clinical Research Ethics’ Board. Discrete pairing task (DPT) We developed a motor task that required participants to execute multiple motor sequences, with each being of equal difficulty. We designed the discrete pairing task (DPT) to draw upon information processing demands within stages of the motor sequence learning framework (Abrahamse et al., 2013). In the DPT, sequences were comprised of three alternating shapes 72 (triangle, circle, and square) displayed at four spatial locations within the centre of the screen (see Figure 8). The type of shape at the four locations differed between sets of stimuli within a sequence. The keyboard letters — V, B, N, M — corresponded to each of the four spatial locations in an ordinal manner from left to right. Within each set of stimuli, one of the shapes appears twice. The participant’s task was to make two key presses to move the repeating shapes together, beginning with the most leftward press. The first key press highlighted the location of the leftward repeating shape (Response 1; Figure 8a2). The second key press moved the first highlighted shape to the location next to the second repeating shape (Response 2; Figure 8a3). Once the pairing was correctly completed, four new stimuli immediately appeared on the screen and the participant was again required to identify the pair and move the shape to a new location. Each repeating sequence consisted of five sets of stimuli (Figure 8b). Since two key presses were executed during performance of one set of stimuli, 10 key presses were executed during the performance of one full sequence. Participants were instructed to complete each sequence as quickly and accurately as possible. They were also informed that there were three sequences to learn. 73 Figure 8a and b: The discrete pairing task (DPT). The goal was to spatially pair two repeating shapes. On a computer screen, four white shapes appeared on a black background following the cue “Ready, next sequence!” a. Two of the four shapes repeated. Hence the participant must identify the repeating shape and then select that shape (by pressing the spatially compatible key) and then move it next to its partner by selecting the key corresponding to this new location. An example is shown in a1, where the circle shape repeats. The participant presses the spatially-corresponding key (a2) to highlight the most leftward circle (“V”), and then presses a second key, indicating the new location to which the leftward circle will move, in this case switching with the triangle by pressing “N” (a3). Correct pairing of the two repeated shapes will trigger the next set of stimuli. b. Each sequence consisted of five sets of stimuli. Because two key presses were executed during performance of one set of stimuli, a total of 10 key presses were executed during the performance of one sequence. In the example shown the sequence order is V, N, V, N, B, N, B, N, V, B. 1231 2 3 4 5b.a. 74 The addition of a paired association rule in the DPT, which required individuals to position two repeating shapes side-by-side, added a layer of cognitive complexity beyond that typically present in simple, motor sequence learning, tasks (i.e., the serial reaction time task [SRT]). The rule-based associations place additional demands on cognition (associated with movement decisions and motor planning), allowing us to better manipulate processing demands associated with the task. The DPT requires explicit strategizing early in practice, but with mastery, demands on attention decrease and motor proficiency develops, enabling fast and accurate execution. This expected progression provided our rationale for delineating the three phases of motor sequence learning. Due to the relatively simplistic nature of the task and the constraints on the types of shapes and location, we were able to create comparable sequences of equal difficulty. Difficulty was controlled by the location of the matched shapes and hence the key response. Sequences containing the same set of locations had a similar difficulty level. Since each set of stimuli was comprised of four shapes, where one shape occurred twice, but never side-by-side to start, there were three possible ways to place the matched shape. Assuming a “1” represents the matched shape and a “0” is any other shape, the possible shape placements were defined as: Placement 1 — 1010 Placement 2 — 0101 Placement 3 — 1001 In the DPT, every sequence had two occurrences of shape placement 1, two occurrences of shape placement 2, and one occurrence of shape placement 3, in differing orders (e.g., 11223, 12312, etc.). Via manipulation of the order of placement, it was possible to create the three sets of sequences (i.e., sequence A, B, and C). Therefore, a sequence — for example, sequence A — 75 might have the following order of locations: 11223. This would require the learner to press the following keys: V, B, V, B, B, N, B, N, V, N. The DPT was developed and executed in a custom Microsoft Visual Studio 4.0 XNA game studio program (© 2014 Microsoft Corporation, Redmond, WA, USA). Study protocol In the task-familiarization session, participants completed 30 minutes of DPT practice, but there were no repeating sequences. This session was designed to familiarize individuals with the task goal and procedures, such that any later learning effects would be a result of the practice conditions and repeating sequences and not of general task understanding and motor control. One week later, participants completed the experimental phase, practicing three different motor sequences (consisting of 10 key-presses each) in a serial, repeating order (e.g., A, B, C, A, B, C, etc.), during a single practice session. During this session, participants performed 30 blocks of the DPT task and each block contained six trials of each 10-movement motor sequence (e.g., sequence A, B, C, A, B, C, etc.). Each sequence (A, B, C) was performed 180 times across the entire practice session (i.e., 540 practice trials which lasted approximately 60 minutes). Participants returned to the laboratory 24 hours after the practice session to complete a delayed retention test that involved them performing one block of the DPT task that contained 10 trials per sequence (A, B, C), again completed in a serial order. Retention tests were completed in approximately 15 minutes. 76 Figure 9: Outline of design for Studies 1 and 2. Study 1 was conducted across two weeks, involving serially-ordered practice of the different sequences and a 24-hour delayed retention test. Study 2 involved three additional weeks of testing under three counterbalanced conditions (low, medium, and high difficulty). Each condition was followed by a 24-hour delayed retention test. Practice performance curve The DPT is based on discrete sequencing skill acquisition (Abrahamse et al., 2013), which permits individuals to use explicit trial-and-error discovery. Participants are unable to advance to the next trial until the targets are positioned correctly. Thus, the primary dependent measure was response time total (RTT) (response 1 + response 2) for each set of stimuli. For each sequence, the mean RTT (mRTT) across the five sets of stimuli was calculated. The total Randomization of conditions across weeksSerial RetentionSerial PracticeWeek 2Medium PracticeMedium RetentionLow PracticeLow RetentionHigh RetentionHigh PracticeStudy 1Week 3Week 4Week 5Day 2Day 1Day 1Day 2Day 2Day 1Task FamiliarizationWeek 1Study 2Day 1Day 2 77 mean RTT (tmRTT) is the average mRTT for all three sequences calculated for each respective trial. To determine the individual rate of motor skill acquisition, an exponential function (performance curve) was fitted to the tmRTT across the practice session using the following equation (equation 1) (Brown & Heathcote, 2003): 1. 𝑬(𝒕𝒎𝑹𝑻𝑻𝑵) = 𝑨 + 𝑩𝒆 /𝜶∗𝑵 E(tmRTTN) is the expected value of RTT on practice trial N; A is the expected values of RTT after the practice has been completed (asymptote parameter); B is the change in the expected value of RTT from the beginning of the practice to the end of practice (change score parameter); Alpha (α) is the exponential learning rate parameter. Instead of relying on arbitrary divisions as has occurred in the past (Kleim et al., 2004), we employed slope calculations to delineate three separate phases of practice from each individual’s exponential function. First, we calculated the mean slope (equation 2) of individuals’ exponential function (see above, equation 1): 2. 𝒓𝒊𝒔𝒆𝒓𝒖𝒏 = 𝒇 89 / 𝒇 8:𝒙𝒏/𝒙𝟎 Next, the running average for the slope of the exponential function was determined along the intervals [0, x1] and [x2, n]. The first interval, running left to right, delineated the first phase of practice ([0, x1]; Phase I). The second interval, running right to left, delineated the third phase of practice ([x2, n]; Phase III). The running average for the slope of the exponential function across these two intervals, [0, x1] (equation 3) and [x2, n] (equation 4), are detailed, respectively, below: 78 3. 𝒇 𝒙 / 𝒇 𝟎𝒙/𝟎 = 𝜜>𝜝𝒆@𝜶𝔁/ 𝑨/𝑩𝒙 = 𝜝(𝒆@𝜶𝔁/ 𝟏)𝒙 4. 𝒇 𝒏 / 𝒇(𝒙)𝒏/𝒙 = 𝜜>𝜝𝒆@𝜶𝒏/ 𝑨/𝜝𝒆@𝜶𝔁𝒏/𝒙 = 𝜝(𝒆@𝜶𝒏/ 𝒆@𝜶𝔁)𝒏/𝒙 The points x1 and x2 were determined based on the SD of the running average for each interval from the mean slope of the exponential function. Specifically, beginning left to right for the interval [0, x1], the first point x1 was determined where the average value of the running slope is no longer greater than three SDs of the mean slope (DeLeeuw & Mayer, 2008; Kirsch & Hoffmann, 2012). The second point, x2, was determined in the same manner but beginning from right to left for the interval [x2, n]. This bidirectional approach was adopted to enable identification of the extremes (i.e., Phase I and III), based on procedures adopted in similar work (DeLeeuw & Mayer, 2008; Kirsch & Hoffmann, 2012). After determining the two division points x1 and x2, three phases of practice were identified: Phase I [0, x1], Phase II [x1, x2], and Phase III [x2, n]. Phase II was represented by trials between Phase I and III (see Figure 10). 79 Figure 10: The phases of skill acquisition based on performance-resource function. Total mean response time total (tmRTT) was fit to an exponential function: 𝑬(𝒕𝒎𝑹𝑻𝑻𝑵) = 𝑨 + 𝑩 ∗ 𝒆 /𝜶∗𝑵. E(tmRTTN) is the expected value of tmRTT on practice trial N; A is the expected value of tmRTT after practice (asymptote parameter); B is the change in the expected value of tmRTT from the beginning to the end of practice (change score parameter); alpha (α) is the rate parameter (Heathcote et al., 2000). The three phases (I to III) were identified based on calculation of the change in the slope of curve. To determine the amount of motor sequence learning at retention, both absolute and relative retention performance indices were used. Absolute retention is not affected by temporary performance factors in practice, such as fatigue or effort; it has been defined as the simplest and most scientifically justifiable measure of learning (Schmidt & Lee, 2011). Relative retention offers some insight into memory and forgetting. Relative retention was the degree of gain or loss in the retention interval as a function of the amount of improvement in practice. Relative 00.20.40.60.8126.96.36.199 21 41 61 81 101 121 141 161Response Time (s)Trials (n)!("#$%%&)= ' +() (−*&)Phase II Phase IIIPhase IBA 80 retention was calculated by subtracting mean absolute retention from the last block of practice, which was then divided by the performance on the last block of practice minus the performance on the first block of practice (i.e., tmRTT of Block 30 – Retention / tmRTT of Block 30 – tmRTT Block 1) (Haibach, Reid, & Collier, 2011). For both indices (i.e., absolute retention, relative retention), a lower value represents greater retention. Statistical analysis Primary analyses: A repeated measures analysis of variance (ANOVA) was performed on tmRTT across the three phases (I, II, III) to determine if the proposed theoretical phases produced significantly different performance and learning outcomes. These phase values were later used in Study 2 as individual-specific reference values for the computer-controlled, learner-adapted algorithm to create the three practice difficulty levels. The relationship between the motor skill rate of acquisition (α) and the absolute and relative retention tmRTT values were tested using Pearson’s correlation analyses (r). Pearson’s correlation analyses were also performed on the proportion of trials spent in the various practice phases (I, II, III) as a function of the overall amount of practice trials and both absolute and relative retention. Secondary analyses: Subsequently, to ensure rate of skill acquisition was the primary predictor of absolute retention, a hierarchical multiple regression was performed to predict the absolute retention based on the tmRTT for block 1 and 30, and the rate of skill acquisition (α). tmRTT for block 1 and 30 were first entered in step 1, followed by rate of skill acquisition in step 2. The final model for the hierarchical regression analysis was evaluated for overall significance, and the significance level of the change in R2 when adding rate of skill acquisition to the model (p ≤ 0.05). 81 To further assess the association between practice parameters and performance at retention, and relationships within practice parameters, the relationship between the absolute tmRTT values and tmRTT in the practice phases (I, II, III) were tested using Pearson’s correlation analyses (r). Additionally, the relationship between the ratio of trials in the practice phase (I, II, III) and the rate of skill acquisition (α) were tested using Pearson’s correlation analyses. All data were visually inspected for skewness and kurtosis and objectively tested for normality with the Shapiro-Wilk test with a significance level set at p < 0.001 (Gamst, Meyers, & Guarino, 2008). Post hoc pairwise comparisons were performed following significant effects. To control for familywise errors associated with multiple statistical tests (i.e., to guard against Type-I errors), Bonferroni corrections were applied (pcorrected). Pending the significance of the Mauchly’s test of sphericity, a Greenhouse-Geisser correction was used to adjust degrees of freedom. Effect sizes were reported as partial eta-squared (ηρ2), where 0.01 is considered a relatively small effect, 0.06 moderate and more than 0.14, a large effect (Gray & Kinnear, 2012). The 95% confidence intervals (CIs) of the mean difference (MD) were used to describe the effect of phase on the motor performance. Data are presented in the text as mean (M) plus or minus standard deviation (SD) or standard error (SE). For all statistical tests, significance was set at probability value (p) ≤ 0.05. SPSS 22.0 (SPSS Inc., Chicago, IL, USA) statistical software was used for analyses. Results Primary analyses: One subject was excluded from subsequent analyses for reporting they did not hear the pre-practice instruction that explicitly informed individuals of the presence of three repeating sequences. We first determined goodness-of-fit of the data based on the 82 exponential learning rate parameter (α). The average R squared (R2) for all participants (n = 13) was 0.54 (SD = 0.15; min. = 0.22, max. = 0.70). We considered these values to provide a moderate to good representation of the data (Lang & Bastian, 1999, 2001). Exemplary data from three participants who showed either a fast, slow, or relatively variable rate of acquisition is shown in Figure 11. Single-subject practice and retention data, and normalized performance curves for Study 1 are displayed in Figures 13 and 14, respectively. Curves were divided by a normalization factor of A + B; A = asymptote value; B = change score. Figure 11: Three examples of performance curves. Individuals showed a fast (alpha [α] = 0.038; dashed line), slow (alpha [α] = 0.0098; solid line), and variable (alpha [α] = 0.016; dotted line) rate of skill acquisition across 30 blocks of practice. To confirm that the practice phases were distinguishable with respect to tmRTT, a repeated measures ANOVA yielded a main effect of phase, F(1.33, 16.01) = 49.35, p < 0.001, ηρ² = 0.80, large effect size. Post hoc comparisons confirmed that Phase I response times (M = 0.65 seconds [s], SD = 0.09) were on average slower than Phase II (M = 0.47 s, SD = 0.12) (MD = 00.20.40.60.811.21 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30tmRTT(s)BlockVariableFastSlow 83 0.18, 95% CI [0.09, 0.26]) and Phase III (M = 0.39 s, SD = 0.13) (MD = 0.26, 95% CI [0.18, 0.35]). Phase II was also significantly slower than Phase III (MD = 0.083, 95% CI [0.042, 0.12]) (all ps ≤ 0.001). For comparison, tmRTT in retention was 0.41 s (SD = 0.13). Following Bonferroni correction for multiple comparisons (pcorrected = 0.025), the relationship between rate of acquisition (α) and absolute but not relative retention was significant (see Table 2a). As shown by the significant positive relationship, individuals whose performance was worse in retention (i.e., longer response time) more quickly achieved their asymptote value (α) (see also Figure 12). Similar trends were seen for relative retention, but these correlations were not significant (see Table 2b). 84 Table 2: Correlations between rate of acquisition, phase ratio, and retention. Shows correlations between rate of acquisition, phase ratio (proportion of trials in this practice phase), and absolute retention(panel a) and relative retention (panel b). * represents statistical significance at p ≤ 0.05. ________________________________________________________________ Correlation (r) with absolute retention p-value ________________________________________________________________ Practice parameter Alpha (α) 0.77 0.018* Practice phase ratio I -0.76 0.002* II 0.75 0.003* III 0.76 0.002* ________________________________________________________________ Correlation (r) with relative retention p-value ________________________________________________________________ Practice parameter Alpha (α) 0.60 0.029 Practice phase ratio I -0.61 0.027 II 0.58 0.040 III 0.61 0.026 a. b. 85 Figure 12: Scatterplot showing the relation between alpha (α) values extracted from the exponential curve fitting for each participant during practice and absolute retention tmRTT (s). Higher alpha (α) indicates faster rate of skill acquisition. We also examined whether the proportion of trials spent in the various practice phases correlated with absolute and relative retention (Table 2b). The average number of trials for each phase were: Phase I (M = 48.23, SD = 13.45), Phase II (M = 25.69, SD = 2.59), and Phase III (M = 105.15, SD = 11.62). All practice phase durations showed significant correlations with absolute retention (pcorrected = 0.017). Importantly, the number of trials spent in Phase I was negatively related to absolute retention (i.e., relatively more time in Phase I was related to faster and lower tmRTT), whereas the reverse was true for Phases II and III. For relative retention, the data mirrored that seen for absolute retention, although the correlations were not significant following Bonferroni corrections. 00.10.20.30.188.8.131.52 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04tmRTT(s)Rate of Acquisition (α) 86 Secondary analyses: The overall regression model, with tmRTT for block 1 and 30 entered in step one, rate of skill acquisition in step two, achieved significance (F(3,9) = 6.13, p = 0.015), with an R2 of 0.67. When only tmRTTb1 and tmRTTb30 were included, the model was not significant (F(2,10) = 0.09, p = 0.92), with an R2 of 0.02. The addition of α in the second block significantly improved the model: ∆R2 = 0.65, ∆F(1, 9) = 17.92, ∆p = 0.002. Thus, start-point and floor effects did not influence the absolute retention performance. Absolute retention also did not significantly correlate with tmRTT in each phase of practice (p > 0.05), although all correlations were small and positive (r = 0.29, r = 0.24, and r = 0.32 for Phases 1 to III, respectively). Ratio of trials in each practice phase was highly correlated with rate of skill acquisition (all rs > ± 0.97). Greater time spent in Phase I was associated with a slower rate of acquisition, while more time spent in Phases II and III was associated with a faster rate of acquisition. 87 Figure 13: Single-subject practice and retention data for Study 1 (baseline). Additional analysis: A repeated measures analysis of variance (ANOVA) yielded a significant main effect of practice block on mRTT: F(29, 319) = 27.46, p < 0.001, ηρ² = 0.71, large effect size. Figure 14: Single-subject normalized curves for practice. For each participant, curves were divided by a normalization factor of A(i) + B(i); A = asymptote value; B = change score; i = participant number. 00.20.40.60.811.20 5 10 15 20 25 30mRTT(s)Practice BlocksStudy 1 s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13Retention0.00.20.40.60.81.01.21 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171 181Normalized tmRTTTrialsStudy 1s1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13 88 Discussion The goal of Study 1 was to develop a methodological approach to assess individual differences in motor learning. To accomplish this, we developed the DPT and quantified practice metrics with an exponential curve fitting method. Exponential curve fitting characterized individuals’ performance during skill acquisition of multi-task practice in a massed single session. This allowed for the calculation for the rate of motor skill acquisition (α) and the proportion of trials in each phase of practice, based on systematic changes in the slope of the performance curve. In accordance with previous data-driven methods that dissociate potential phases of learning (Hirano et al., 2015), we showed that skill acquisition could be meaningfully differentiated into three phases. Moreover, in support of findings in the CI literature (e.g., Schmidt & Bjork, 1992), a slower rate of motor skill acquisition during practice was associated with better delayed retention (i.e., faster response times). The amount of interpretable information is a factor in determining an optimal challenge point (Guadagnoli & Lee, 2004). However, this concept is difficult to operationalize and potentially difficult to measure during practice. We hypothesize that a slower rate of motor skill acquisition, as well as more practice time spent within the early cognitive phase of practice, would be associated increased performance at retention. In Study 1, individuals practiced three sequences in a serial, repeating order. Although the practice schedule task difficulty was consistent, the skill level of the individual increased with each practice trial. Individuals transitioned through the phases at learner-dependent rates (α), signifying their overall motor skill acquisition capability. In Study 2, we used the performance metrics from Study 1 to manipulate task difficulty in the same individuals using a computer-controlled, learner-adapted practice 89 schedule. This allowed us to gain insights about how best to operationalize and optimize individual challenge in practice to facilitate learning. Study 2 — Methods CI effects can be influenced by task difficulty, which can be divided into two categories: (1) nominal difficulty; and (2) functional difficulty. Nominal task difficulty relates to the conditions under which the tasks are performed. Functional task difficulty is dependent upon the skill level of the learner (Guadagnoli & Lee, 2004). When the nominal difficulty of tasks is low, blocking the order of the practice trials so that all trials for one task are performed before the next, leads to superior short-term performance compared to random ordering of tasks. However, delayed retention testing reveals the reverse effect: random (high CI) practice produces superior retention. Under conditions where the nominal difficulty of the task is relatively high, the benefits of random practice on long-term retention can be negated (Del Rey, Wughalter, & Whitehurst, 1982). Further, experienced performers benefit more from high CI practice than do novices (Guadagnoli et al., 1999). In Study 2, we created three practice difficulties (low, medium, and high difficulty), based on each individual’s practice performance curve from Study 1, to assess the effectiveness of learner-adapted practice schedules in a task that had constant nominal difficulty and relatively easy task difficulty. A learner-adapted algorithm was used to systematically manipulate task difficulty/cognitive demands during motor sequence learning of multiple sequences. Individual performance curves from Study 1 informed the computer-controlled, learner-adapted practice schedule. Rather than using a generalized challenge point (i.e., mean reference value) across all individuals (Choi et al., 2008), we used participant-specific, individual challenge points. This 90 was based on motor skill acquisition under the serial practice condition from Study 1, allowing us to determine (and manipulate) motor practice difficulty. Based on results from Study 1, a slower rate of motor skill acquisition (α), with more time spent in Phase I of practice, was expected to relate to superior performance at retention. Therefore, we hypothesized that a high difficulty practice algorithm, which requires individuals to practice near their Phase I (tmRTT reference value), would result in the best performance at a delayed retention test. Thus, in Phase I, faster response times should promote constant switching between trials because performance quickly surpasses the reference value. This contrasts with a low difficulty practice condition, which would encourage individuals to practice near their Phase III reference value (i.e., low switching between trials). A medium practice difficulty (related to Phase II of practice) and a moderate amount of switching between trials were expected to result in an intermediate level of retention. We implemented a “within-subjects” design that allowed us to compare the efficacy of these three types of practice based on individual performance curves. We did not explicitly compare the effectiveness of adaptive practice schedules with non-adaptive schedules. Rather, our aim was to test the potential usefulness and optimality of practice schedules (comparing low, medium and high challenge) based on individualized reference values extrapolated from a different, yet related practice task. Participants Study 2 had the same participants as Study 1 (n = 13). Practice conditions Study 2 involved three weeks of learner-adapted practice and delayed retention sessions (see Figure 9). After a one-week washout period from Study 1, participants were randomly assigned to take part in the three learner-adapted practice conditions in counterbalanced order. 91 Individual-specific reference values from Study 1 (tmRTT) were generated from Phases I, II, and III to create practice conditions of high, medium, and low difficulty, respectively. Each learner-adapted condition was separated by a one-week wash-out period. Each week was comprised of one massed practice session and a 24-hour delayed retention test. Nine sequences were created and randomized across the three practice difficulty conditions. During each practice session, participants practiced three sequences resulting in 30 blocks of the DPT. Each block contained 18 trials of a 10-movement motor sequence. Each sequence was performed 180 times across the entire practice session. Participants returned to the laboratory 24 hours following the practice session to complete a delayed retention test. To systematically manipulate task difficulty, the mean tmRTT within each of the three phases of testing from Study 1 provided the individual-specific reference values for each of three levels of learner-adapted task difficulty; low (Phase III), medium (Phase II), and high (Phase I). The individualized reference values for Phase I ranged from 0.51 to 0.80 s, for Phase II, from 0.24 to 0.68 s, and for Phase III, from 0.24 s to 0.65 s. These were used to dictate when an individual switched to practice a new sequence. In addition to these values, switching was also based on the rate of motor skill acquisition rate (α) extracted from the exponential function of each individual’s practice performance curve. Therefore, these values (i.e., tmRTT_ref and α) were inputted in the equation (equation 5) below: 5. 𝑺(𝒕) = 𝑺(𝒕 − 𝟏) ∗ (𝟏 + 𝜶(𝒎𝑹𝑻𝑻(𝒕) − 𝒕𝒎𝑹𝑻𝑻_𝒓𝒆𝒇 )) The switching function (S(t)), which determined if the participant performed a new sequence or continued with the present sequence, equaled the switching function of the previous trial, S(t-1), multiplied by a number based on the tmRTT of the current trial in comparison with the corresponding reference RTT value (RTT_ref). Before the start of a block, and when 92 switching to a new sequence occurred, the value of S(0), the switching function at trial 0, was set to 1. Sequence switching occurred when the switching function’s value went below a specified threshold (based on pilot testing a threshold of 0.90 was set). If the participant had a faster RTT relative to the reference value, the switching function decreased and vice versa. The individualized motor skill acquisition rate (α, calculated from Study 1) affected how much the switching function changed for a given difference in RTT to the reference tmRTT value. Although this value differed between individuals, it was constant between the three learner-adapted conditions. Thus, if the rate of acquisition was fast in Study 1, the switching function would increase. Conversely, if the rate of acquisition was slow in this previous study, the switching function would decrease (i.e., there was less switching). The average motor skill acquisition rate calculated from Study 1 for all participants was α = 0.02 (SD = 0.01; min. = 0.010; max. = 0.034). The order of each of the three sequences in the block was random. The maximum number of switches between sequences in each block was 17 and the minimum was two. Hence, a hypothetical, fully random condition would result in 510 switches in total. Due to the known impact of prior explicit knowledge on motor sequence learning, explicit recognition testing was performed at the end of retention (Boyd & Winstein, 2004b; Wulf & Weigelt, 1997). Participants were asked if they recognized a sequence after watching it played on a screen. For each sequence, the five sets of stimuli flashed on the screen for 1.5 s, in the sequential order performed during practice. After the completion of each sequence, participants were asked to indicate recognition by pressing “1” for “Yes” or “2” for “No.” Eighteen sequences were randomly “played” during the recognition test. Each of the three sequences performed during practice appeared twice, the remaining were novel. Percentage 93 correct was calculated. If participants’ success rate was greater than chance, they were considered to have explicitly recognized the presented sequences (i.e., implicit learning did not occur). Performance measures The tmRTTs for each of three practice phases (I, II, III) for each condition were calculated (see Study 1). Both absolute and relative retention measures were used (see Study 1). Statistical analysis To provide validation for the learner-adapted algorithm, a repeated measures ANOVA was performed on the number of switches for CONDITION (low, medium, high difficulty). To ensure there was no difference in initial performance across conditions, a repeated measures ANOVA was performed on tmRTT for each sequence during the first block of practice. To evaluate the effect of CONDITION on response time across practice, a two-way repeated-measures ANOVA was performed with CONDITION (low, medium, high difficulty) and PRACTICE PHASE (I, II, III) as variables. For retention and recognition tests, a one-way, repeated measures ANOVA was performed to test for CONDITION effects (low, medium, high difficulty). Post hoc pairwise comparisons were performed following significant effects. As in Study 1, Bonferroni corrections were applied to correct for multiple comparisons (pcorrected). All data were visually inspected for skewness and kurtosis and objectively tested for normality with the Shapiro-Wilk test with a significance level set at p < 0.001 (Gamst et al., 2008). The number of switches for low and medium conditions were non-normal (W(13) ≤ 0.57, p < 0.001). Following log10 transformation, all switch data were found to be normal (p > 0.001). For statistical analyses, log10 transformations were applied to the number of switches for all conditions. Pending the 94 significance of the Mauchly’s test of sphericity, Greenhouse-Geisser corrections were implemented to adjust degrees of freedom. Effect sizes were reported as partial eta-squared (ηρ²) where 0.01 is considered a relatively small effect, 0.06 moderate and more than 0.14, a large effect (Gray & Kinnear, 2012). The 95% CIs of the MD were used to describe the effect of learner-adapted practice on the motor performance. Data are presented in the text as M ± SD or SE. For all statistical tests, significance was set at p ≤ 0.05. SPSS 22.0 (SPSS Inc., Chicago, IL, USA) statistical software was used for statistical analyses. Results Practice The total number of switches between sequences differed statistically across conditions, despite the large between subject variability, F(2, 24) = 32.15, p < 0.001, ηρ² = 0.728, large effect size. Post hoc tests revealed that the high condition (M = 225.2 switches, SD =108.83, min. =113, max. = 439) had significantly more switches than the medium (M = 141.6 switches, SD = 85.9, min. = 96, max. = 417, pcorrected = 0.005) and low (M = 94.8 switches, SD = 76.41, min. = 60, max. = 338, pcorrected < 0.001) conditions. There was also a significant difference between the low and medium conditions (pcorrected < 0.001) (see Table 3 for individual number of switches). 95 Table 3: Individual number of switches for each of the learner-adapted conditions for Study 2 (ascending order). Low difficulty Medium difficulty High difficulty 60 96 123 60 118 167 61 107 113 61 127 117 62 104 201 65 120 177 66 99 183 79 79 374 79 146 263 96 168 231 65 139 156 140 121 384 338 417 439 In accordance with predictions, mean response times co-varied with condition such that the low difficulty condition had the shortest duration (M = 0.38 s; SE = 0.02), followed by medium (M = 0.39 s, SE = 0.02) and high (M = 0.43 s, SE = 0.02). Despite differences between conditions, the main effect of condition was not significant: F(2,24) = 2.55, p = 0.10, ηρ² = 0.18, large effect size. There was a practice phase effect, F(1.1,13.16) = 6.63, p = 0.02, ηρ² = 0.36, large effect size, but not a significant CONDITION × PRACTICE PHASE interaction, F(1.2,14.0) = 3.04, p = 0.10, ηρ² = 0.20, large effect size. Phase I practice (M = 0.46 s, SE = 0.03) had a significantly longer mean response times than Phase II (M = 0.37 s, SE = 0.02) (MD = 0.095, pcorrected = 96 0.000003, 95% CI [0.065, 0.12]), but not Phase III (M = 0.38 s, SE = 0.02) (MD = 0.083, pcorrected = 0.131, 95% CI [-0.019, 0.19]). Phase II and Phase III did not significantly differ (MD = -0.12, pcorrected = 1.00, 95% CI [ -0.098, 0.074]) (see Figure 21a). Individuals started each condition at a similar level of performance between practice weeks, confirmed by no effect of condition during Block 1, F < 1, p > 0.05. Single-subject practice and retention data for low, medium, and high difficulty conditions are displayed in Figures 15, 17, and 19, respectively. Normalized performance curves for low, medium, and high difficulty conditions are displayed in Figures 16, 18, and 20, respectively. 97 Figure 15: Single-subject practice and retention data for low difficulty condition. Additional analysis: a repeated measures analysis of variance (ANOVA) yielded a significant main effect of practice block: F(29, 319) = 11.48, p < 0.001, ηρ² = 0.56, large effect size. Figure 16: Single-subject normalized performance curves for practice for low difficulty condition. 00.20.40.60.811.20 5 10 15 20 25 30mRTT(s)Practice BlocksLow Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13Retention00.20.40.60.811.21 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113117121125129133137141145149153157161165169173177Normalized tmRTTTrialsLow Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13 98 Figure 17: Single-subject practice and retention data for medium difficulty condition. Additional analysis: a repeated measures analysis of variance (ANOVA) yielded a significant main effect of practice block: F(29, 319) = 16.21, p < 0.001, ηρ² = 0.60, large effect size. Figure 18: Single-subject normalized performance curves for practice for medium difficulty condition. 00.20.40.60.811.20 5 10 15 20 25 30mRTT(s)Practice BlocksMedium Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13Retention00.20.40.60.811.21 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113117121125129133137141145149153157161165169173177Normalized tmRTTTrialsMedium Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13 99 Figure 19: Single-subject practice and retention data for high difficulty condition. Additional analysis: A repeated measures analysis of variance (ANOVA) yielded a significant main effect of practice block: F(29, 319) = 15.92, p < 0.001, ηρ² = 0.59, large effect size. Figure 20: Single-subject normalized performance curves for practice for high difficulty condition. 00.20.40.60.811.20 5 10 15 20 25 30mRTT(s)Practice BlocksHigh Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13Retention00.20.40.60.811.21 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 101105109113117121125129133137141145149153157161165169173177Normalized tmRTTTrialsHigh Conditions1 s2 s3 s4 s5 s6 s7 s8 s9 s10 s11 s12 s13 100 Retention Practice condition significantly affected absolute retention: F(1.29, 15.47) = 13.16, p = 0.001, ηρ² = 0.52, large effect size. As shown in Figure 21b, the high difficulty condition (M = 0.32, SD = 0.811) resulted in significantly faster response times than the low (M = 0.47, SD = 0.128) (MD = .15 s, pcorrected = 0.005, 95% CI [0.046, 0.249]) and medium conditions (M = 0.40, SD = 0.090) (MD = 0.081 s, pcorrected = 0.001, 95% CI [0.035, 0.127]). Low and medium did not differ from each other (p > 0.05). There was also a CONDITION main effect for relative retention: F(2, 24) = 3.93, p = 0.033, ηρ² = 0.25, large effect. A significant difference was only observed between the high (M = –0.05, SD = 0.264) and low (M = 0.50, SD = 0.708) (MD = 0.54, pcorrected = 0.02, 95% CI [0.085, 1.00]) conditions. There was no significant difference between the high and medium (M = 0.32, SD = 0.564) conditions (MD = -0.32, pcorrected = 0.13, 95% CI [–0.086, 0.818]). Figure 21a and b: Practice and retention performance. a. Practice performance across the three phases of practice (Phase I, II, III) collapsed across learner-adapted difficulty conditions. * = significance at p corrected < 0.05 (error bars show standard deviation of the mean). b. Retention performance (absolute tmRTT) for each of the three learner-adapted difficulty conditions. * = significance at pcorrected < 0.0167 (error bars show standard deviation of the mean). 00.10.20.30.184.108.40.206Low Medium HightmRTT(s)Practice ConditionsLow Moderate High* p = 0.005* p = 0.001a. b.* p = 0.02220.127.116.11.18.104.22.168Phase I Phase II Phase IIItmRTT(s)Practice Phases RetentionPractice 101 Accuracy was generally high on the explicit recognition test (> 90%), but there was no significant difference between conditions (F < 1, p > 0.05); high (M = 92.3%, SD = 8.93, min. = 72.2%), medium (M = 93.1%, SD = 8.22, min. = 77.8%) and low (M = 96.2%, SD = 5.24, min. = 88.9%). General discussion We used a within-subjects experimental design to investigate individualized-adapted practice schedules and to test whether a high or moderate degree of task challenge is most beneficial for motor learning. The learner-adapted algorithm produced significant differences in the number of total switches between the low and medium, low and high, and medium and high conditions, showing that our algorithm was effective in altering the amount of CI. Differences were noted both on a between-subject level as well as within participants, given that switching was dependent on an assumed performance-resource function based on an earlier practice episode (Study 1). For some individuals, what would be seen as a moderate amount of switching would be classed as “high” switching for others and similarly “low” switching would be classed as medium for others (see Table 3). Indeed, the minimum number of switches for the high condition (113) was less than the average for the medium difficulty condition (142). Despite the individualized nature of the CI conditions, there were significant differences across the three conditions of practice at retention. In accordance with previous CI studies where more random (high CI) conditions of practice resulted in superior retention, we showed that the high condition resulted in faster response times in retention (Magill & Hall, 1990; Shea & Morgan, 1979; Wright et al., 2016). Keeping individuals in a practice phase representative of the early practice 102 stage, which is thought to be high in cognitive effort, controlled the high level of difficulty during practice, which subsequently benefited retention. Our high difficulty condition would be considered a type of “hybrid” schedule, characterized by short periods of blocked practice before a switch (in comparison to medium and low CI conditions, which would have medium and long periods of blocked practice, respectively). That is, the algorithm we employed ensured that individuals practiced the same motor sequence until they had achieved a specific level of performance (tmRTT) before moving to the next motor task. Potentially, this hybrid CI scheduling in the high condition, which alternated between blocked and random practice, produced an environment that was relatively high in attentional demands and cognitive effort, thereby enhancing performance compared to the medium and low difficulty conditions. This supports previous studies showing that hybrid practice schedules produced learning effects comparable to random practice, while maintaining a level of performance during practice comparable to blocked practice (Al-Ameer & Toole, 1993; Simon, Lee, & Cullen, 2008). However, because we did not test yoked practice conditions (where these individualized switching schedules are imposed on a partner), we are unable to make direct comparisons with past work. Nevertheless, it is positive that our method resulted in long-term learning effectiveness, without apparent costs in practice (i.e., performance differences in acquisition). In our studies, practice parameters from an earlier learning phase, including rate of motor skill acquisition and average response time in a practice phase, were effective in generating high levels of individualized challenge, which positively impacted delayed retention. However, given the relatively low nominal task difficulty of the current sequence-learning paradigm, task-specific effects should be further probed. When the motor response or task-demands are more 103 complex, a high degree of challenge is unlikely to be the best condition for learning. For example, blocked practice conditions were better for retention of novel key-press sequences when a task-irrelevant focus (involving verbal identification of the pitch of an auditory tone) was additionally required. This was compared to conditions when a task-relevant focus was required (Raisbeck, Regal, Diekfuss, Rhea, & Ward, 2015). Similarly, when the motor task involved sensorimotor integration and timing, in comparison to acquisition of stimulus-response associations as required in our task, low-challenging, blocked practice was preferable to high challenge conditions, at least in the initial phase of practice (Savion-Lemieux & Penhune, 2010). These task-specific effects in the CI literature marry with the recommendations based on the challenge point framework; that challenge is both a function of the individual’s skills (functional difficulty) and the task-demands (nominal difficulty, Guadagnoli & Lee, 2004). It is important to investigate, and to begin to quantify, the evolving processes underlying changes in motor performance across practice, as well as how these phenomena relate to or explain longer-term retention and transfer to similar yet novel skills. Our chosen method for doing this was based on deriving an exponential function from three parameters based on an individual’s performance curve: A, B, and α. If individuals do not demonstrate a large change in performance (y values) across their entire practice (x values), the rate of motor skill acquisition (α) will be high because performance quickly plateaus. Conversely, if individuals do demonstrate a large change in performance (y values) across a small number of early practice trials (x values), the rate of motor skill acquisition (α) will also be high because performance quickly plateaus. Both situations may indicate that the practice paradigm was either too difficult or too easy, respectively, for individuals (i.e., not at an optimal challenge point), leaving them with no useful information to extract. Inspection of individual performance curves alerted us to both these 104 situations in Study 1. For example, subject 4 and subject 11 had comparable rates (α = 0.036 and α = 0.038, respectively; Figure 14); however, subject 11 had a change in performance (B value) of 0.12, while subject 4 had a change in performance of 0.85. In addition, there was no association between performance in the phases of practice and retention. In the present sample, both subjects demonstrated faster rates of motor skill acquisition, compared to other individuals, which translated into poorer retention test performance in Study 1. These two subjects show different degrees of improvement in performance with similar rates of skill acquisition, supporting the notion that information processing demands were above or below optimal. However, for subject 4, based on the larger number of switches (low difficulty: 338; medium difficulty: 417; high difficulty: 439) in the learner adaptive practice conditions, that indicates frequently surpassing baseline reference values, external factors, such as motivation and attentional, may have contributed to lack of change in baseline performance (Lewthwaite & Wulf, 2017). There may be other methods that are equally or more useful in determining task difficulty, based on performance-curve fitting to dictate optimal methods of practice for either multiple or single tasks. This will probably be task-dependent, practice amount-dependent, and of course dependent on the similarity between the initial task where curve fitting was performed and the transfer task. We chose to fit our data to an exponential function for two reasons: (1) exponential functions are preferred over power functions for individual performance data, whereas the latter provides a better fit for group data (Heathcote & Brown, 2000); and (2) the learner-adapted algorithm in the present paper was based on a model derived from Choi et al. (2008), where it was assumed that motor learning related performance curves are relatively well modelled with exponential functions. Despite these reasons, we were only able to capture just over 30% of the 105 variance in our data through this method, and it may be that there are better model fits than exponential curve fitting which can help describe learning and which can be used to derive parameters for altering learning in future tasks. In Study 1, participants acquired three different motor tasks in a highly predictable environment (i.e., serial order). In this type of practice, sequence knowledge develops and presumably motor chunks are created in a sequential fashion, progressing from an individual element level to multiple elements within a sequence (Wright et al., 2016). We showed advantages associated with a slower rate of motor skill acquisition for later retention. Retention advantages that result from a slower rate of motor skill acquisition presumably stem from to time spent interpreting information and maintaining cognitive effort during practice (Boyd & Winstein, 2004b). This is supported by an fMRI study showing greater activity in motor regions associated with movement planning and response selection during slowed initial performance in random practice (Cross et al., 2007). Individuals in the random practice group took more time to plan their movements when acquiring a motor skill, yet showed an enhanced learning effect at a delayed retention test (Cross et al., 2007). In our study, it is likely that enhanced retention performance was related to time spent interpreting information and cognitive effort involved in learning the sequence of movements through planning, problem solving, and strategizing. The results of Study 2 support our expectations that learning could be manipulated through an individualized, computer-controlled, learner-adapted algorithm, and that a condition that necessitated high cognitive effort would best promote learning. The high difficulty condition, which required individuals to be near the performance-resource demands of Phase I (cognitive phase), generated a level of challenge that was superior for learning in comparison to the low and medium difficulty conditions (corresponding to Phases III and II, respectively). In 106 addition, benefits for the high difficulty condition were shown regardless of the method used to assess retention (absolute or relative retention), which suggests this condition was not only effective in producing an enhanced response time at the delayed retention test, but also in maintaining and/or increasing performance at retention. Although we did not make comparisons across generalized or experimenter-determined algorithms in this study (e.g., pure random or blocked practice), nor did we compare learner-adapted to yoked conditions, we did satisfy our aim of showing that our method of determining individual-specific difficulty, based on performance curve analysis and optimizing challenge, can lead to within-subject learning benefits compared to learner-adapted low CI practice. In future work, it will be necessary to probe the effectiveness of this computer-controlled protocol in comparison to practice schedule protocols that are based on a generally high (or low) degree of task difficulty (irrespective of the individual or task). Also, comparisons between the individual challenge points used in our learner-adapted algorithm to the more general challenge points used by Choi et al. (2008) would be helpful. Performance curves are not new to the field of motor performance and learning (Snoddy, 1926); however, the present studies highlight a novel use of performance curves in generalized (Study 1) and learner-adapted (Study 2) practice conditions. While there are known difficulties in the application of performance curves, and potential errors that can be made in their interpretation (Schmidt & Lee, 2013), our study demonstrates that when used methodically, valuable information, such as the rate of skill acquisition, and performance in practice phases, can be obtained to design learner-adapted practice paradigms. Specifically, entry-level abilities and/or prior experience can provide useful information that is often unexamined when assessing and using performance curves to inform future training paradigms (Lane, 1987). This is a first 107 step towards the use of prior knowledge of individual skill level to improve response to motor training through individualized practice schedules based on rate of skill acquisition. In conclusion, we provide evidence that a high-difficulty practice condition is optimal for learning of a discrete-pairing, sequence-learning task (DPT). Although this high difficulty condition was characterized by more switching between sequences than the other conditions, it differed from typical high-CI schedules in that small blocks of practice before a switch to a new sequence characterized the schedule of practice, and there was also considerable variability of switching between participants. Moreover, this high-CI schedule did not have the typically shown practice-performance costs seen with “random” schedules when compared to blocked practice (low CI). This was the case even though we used a within-subjects design to assess the effectiveness of these learner-adapted practice schedules (which is an atypical method for assessing and comparing across practice schedule manipulations). We relate our data to the concept of optimal challenge, as introduced by Guadagnoli and Lee (2004). However, the theoretical concept and identification of an “optimal” challenge point requires further investigation. Overall, our results illustrate that performance curve metrics (based on a previous practice episode of a similar task) may facilitate the assignment of a numerical value to the concept of optimal challenge. 108 Chapter 3: Predicting motor performance-related change in individuals with chronic stroke Introduction Following a stroke, motor outcomes reflect the capacity for learning as individuals work to relearn old, and acquire new, motor skills (Muratori et al., 2013). To relearn, or acquire, a motor skill, individuals learn basic movement patterns through a series of trial-and-error movement strategies (Doyon et al., 2009). If sufficient practice is provided, large improvements in motor performance occur in the early phase of skill acquisition, followed by smaller improvements when approaching a plateau in performance in the late phase of skill acquisition (Muratori et al., 2013). Motor learning, which refers to “relatively permanent changes in the capability for skilled behavior” (Schmidt & Lee, 2011), is assessed during a delayed retention test following practice. Conventionally, change in motor performance over the practice period has been quantified with discrete measures of behavioural change taken at pre- to post-practice time points (Boyd & Winstein, 2006; Boyd et al., 2009; Boyd & Winstein, 2001; Boyd & Winstein, 2004b; Vidoni & Boyd, 2008). However, discrete measures provide minimal information about patterns of change in motor performance that occur during motor skill acquisition, and the influence of these patterns on motor learning-related change (Whitall, 2004). Pre- and post-testing of motor performance lacks the sensitivity needed to delineate information regarding how quickly individuals improve as well as whether, and when, performance plateaus (Whitall, 2004). Motor outcomes in clinical populations have typically been described with summary values, such as mean response time or percentage of correct responses, taken at discrete points in 109 time to produce pre-post change scores (Boyd & Winstein, 2006; Boyd et al., 2009; Boyd & Winstein, 2001; Boyd & Winstein, 2004b). However, behavioural changes can be characterized using models that encompass the evolution of performance changes over time (Lane, 1987), to elucidate information about the process of skill acquisition throughout the practice period. For example, curve fitting uses all data points, as opposed to small blocks, and as such captures the overall trend of performance during practice. Individual performance data is fitted to a curvilinear function (i.e., power or exponential function) to quantify how performance evolves over time (Cousineau & Lacroix, 2006). From performance curves, the amount, or “dose”, of practice required to achieve a plateau, or an asymptote, in performance can be determined (Lang & Bastian, 1999). The amount or dose required to stimulate learning after stroke is an important topic among clinicians and researchers alike (Lohse, Lang, et al., 2014). In the motor performance literature, curve fitting approaches have been considered in healthy adults (Newell, 1991; Sampaio-Baptista et al., 2015; Sampaio-Baptista et al., 2013; Sampaio-Baptista et al., 2014), and used to detect differences in motor skill acquisition in healthy people and individuals with stroke alike (Ioffe et al., 2006; Lang & Bastian, 1999, 2001). Yet the clinical research application of these principles in motor sequence learning post-stroke has lagged. Previous researchers’ work investigating effects of brain lesions on changes in motor skill acquisition have modelled motor performance using a curvilinear function in individuals with stroke (Ioffe et al., 2006; Lang & Bastian, 1999, 2001). Lang & Bastian (1999, 2001) modeled the rate of motor adaptation using an exponential function during the performance of an upper extremity ball catching task. The decay constant, or the rate of adaptation, represented an estimation of the number of trials required for a subject to proceed approximately two thirds of the way through the adaptation process (Lang & Bastian, 1999). Based on differences in rate of 110 adaptation during the catching task, between healthy controls and individuals with cerebellar stroke, the authors concluded that the cerebellum plays an important role in motor adaptation, which cannot be assumed by the undamaged central nervous system (Lang & Bastian, 1999, 2001). Ioffe et al. (2006) compared performance curves between individuals with lesions of the motor cortex, nigro-striatal system, and cerebellum, and to healthy controls during supervised learning of postural tasks. Based on differences in the rate of improvement between lesion groups, the authors dissociated the distinct role of each neural structure (motor cortex, basal ganglia, cerebellum) in supervised learning of the aforementioned postural tasks (Ioffe et al., 2006). In addition to understanding mechanisms of recovery through differences in performance curves between healthy controls and individuals with stroke, we hypothesize that the rate parameter may be useful to predict motor performance-related change at retention post-stroke. In a recent review discussing the neural mechanisms supporting stroke rehabilitation, pre- and-post testing was posed as a hindrance to our understanding of how an intervention works (Ward, 2015a). Exploiting the relationship between rate of motor skill acquisition, during motor skill practice, and motor performance-related change at retention may enhance the efficacy of motor skill rehabilitation interventions after stroke. Some researchers have used post hoc analyses to quantify the dose of practice needed to induce clinically meaningful change (Lohse, Lang, et al., 2014), or dose escalation methods to identify the maximum tolerated dose (Dite et al., 2015). However, a methodology that uses the rate of skill acquisition to predict retention performance, and the dose of practice necessary to achieve performance plateau during motor skill acquisition, does not yet exist. To improve the long-term retention of motor skills post-stroke, characteristics of the task, such as the level of task difficulty, must be modified to individuals’ own capabilities (Pollock et al., 2014), as 111 opposed to using a “one-size-fits-all” approach. Performance curves are influenced by the nature of a task, with task difficulty or intensity heavily influencing the rate of skill acquisition. Therefore, performance curves may be used to retrospectively quantify an individual’s ability to successfully acquire a given task, and their capacity for motor learning-related change at retention (Lane, 1987). For implicit sequence-specific learning in individuals with stroke, determining the relationship between the rate of skill acquisition and motor learning-related change can provide insight to the appropriateness of task difficulty to enhance learning. Additionally, this approach could enable the individualized prediction of the dose of practice required for rehabilitation of impaired motor skills after stroke. Depending on the nature and level of difficulty of the task, individuals with stroke may require a large dose of practice before an appropriate motor strategy, and movement proficiency is achieved (Lane, 1987). The following example can illustrate the plausible efficacy of such an approach. In the medical field (i.e., during laparoscopic surgery training), the understanding of how skills are acquired on an individual level, through changes in performance over a set number of practice trials, has been called into question (Feldman et al., 2009). Due to individual differences in skill acquisition, rather than using fixed numbers of procedures for credentialing and licensing, parameters from performance curves are being recognized as important measures to identify where individuals are on their learning curves (i.e., early versus late phase) (Feldman et al., 2009). Thus, in a population of individuals with stroke with various levels of residual motor function, the rate of skill acquisition and numbers of trials (i.e., the dose) necessary for motor learning may vary from person to person. Understanding the progression of learning through individual performance curves can help in the development of efficient practice paradigms in rehabilitation settings. 112 In the current work, we used an exponential function to generate a curve of improvement in implicit motor skill performance (during skill acquisition) of a continuous tracking task (CTT) as a function of time (Boyd & Winstein, 2004b; Heathcote et al., 2000). The use of a performance curve is advantageous in the investigation of the dynamic nature of performance, as skill acquisition for the CTT involves known transitions throughout the learning process (Meehan, Randhawa, et al., 2011). Modeling mean scores as a continuous curve so that neighbouring blocks have some influence on the succeeding block accurately depicts the CTT learning process. Rather than calculating mean scores from each block individually, curve fitting incorporates more information to make a prediction, by using all available data points; performance in the initial blocks sequentially influences performance in the latter blocks. Thus, the purpose of the present study was two-fold: (1) to derive and compare practice parameters estimated from an exponential function fitted to an individual’s acquisition curve, between individuals with and without stroke during implicit motor skill acquisition; and (2) to determine the relationship between estimated practice parameters within groups and subsequent retention of motor performance following practice (i.e., motor learning). We hypothesized that exponential curve fitting of implicit sequence-specific motor performance across six days of motor practice would result in group differences in the predictor values that would be associated with the motor learning-related change in performance following the retention interval. Methods Participants The stroke (ST) group included 14 right-handed individuals, with subcortical infarcts in the chronic phase of stroke (greater than 6-months post-stroke) (mean [M] age = 64.7; standard deviation [SD] = 7.24 years; 4 females [F]). Of the 14 individuals, 12 individuals had right 113 hemispheric subcortical lesions, one individual had a left cerebellar lesion, and one individual with a right hemispheric subcortical lesion showed the presence of a second subclinical left hemispheric subcortical lesion; however, all participants reported predominant left arm hemiparesis. Physical impairment level was assessed using the Fugl-Meyer assessment (FM), upper extremity portion (UE-FM) (M = 52.7; SD = 13.0; maximal score 66) (Fugl-Meyer, Jaasko, Leyman, Olsson, & Steglind, 1975). Eleven right-handed, older healthy individuals made up the control group (HC; M age = 64.8; SD = 8.5 years; 7 F) (Table 4). Individuals were recruited from the University of British Columbia (UBC) and surrounding local communities. Each participant’s consent was obtained according to the Declaration of Helsinki; the Research Ethics Boards at UBC approved all aspects of this work. To evaluate the use performance curves during implicit sequence-specific learning, participants from this study were drawn from a dataset of healthy individuals and individuals with stroke where the primary results of a smaller cohort of individuals have been published elsewhere (Meehan, Randhawa, et al., 2011). 114 Table 4: Participant characteristics. Group n Age Mean (SD) Gender MMSE Mean (SD) UE-FM Mean (SD) HC 11 65.1 (8.11) 7 F; 4 M 29.9 (0.326) N/A ST 14 64.7 (7.24) 4 F; 10 M 29.2 (0.974) 52.7 (13.0) Table shows data for healthy controls (HC) group, stroke (ST) group, mini-mental state examination (MMSE), and Fugl-Meyer assessment (FM). Data are presented in the text as mean (M) and standard deviation (SD). F, female; M, male; SD, standard deviation. Exclusion criteria Exclusion criteria included any psychiatric diagnosis, history of substance abuse, clinically evident signs of neurological impairment or disease (other than stroke), inability to perform the experimental task, and taking drugs known to hamper neuroplastic change (e.g., anticholinergics, GABAergics, or NMDA blockers). No one presented with dementia (≥ 26 or on the Mini-Mental State exam [MMSE]) (Folstein, Folstein, & McHugh, 1975). Experimental design Individuals participated on seven separate days over a two-week span, with no more than five days between practice sessions. On the first day, the ST group completed the Wolf Motor Function Test (WMFT) and both groups (ST, HC) performed four baseline training blocks (30 trials per block) of the CTT. Days 2 to 6 involved CTT practice, and on day 7 a 24-hour retention test consisting of four blocks was performed (30 trials per block). On each practice day (i.e., days 115 2 to 6) participants practiced 100 trials (five blocks; 20 trials per block); completion of each practice session took approximately 30 minutes (Figure 22a). Motor function assessment On day 1, the portion of the Wolf Motor Function Test (WMFT) examining upper extremity function was assessed by a registered physical therapist. Mean performance time to complete 15 items of the WMFT with the paretic and non-paretic arms was determined. Participants’ WMFT rate was then calculated to determine how many times an individual could complete the task continuously for 60 seconds (60 seconds divided by the mean performance time); if an individual could not perform the task in 120 seconds a score of 0 was given (Hodics et al., 2012). The WMFT mean rate has been shown to be a more sensitive method to characterize motor function of the upper extremity in individuals with stroke, compared to the time to complete each task(Hodics et al., 2012). Behavioural task During CTT performance, the ST group used their paretic left arm and the HC group used their non-dominant left1 arm to track the vertical path of a target with wrist movements that controlled a joystick (Current Designs, Philadelphia, PA, USA). Participants’ movements were represented as a red filled circle and the target circle was outlined in white; both objects were visible on a black background (Figure 22b, c). Each 20-s trial consisted of two 10-s tracking segments. Unknown to the participants, a predefined tracking pattern was embedded in the 10-s segment of each trial. This repeated sequence remained identical across practice and retention blocks (Wulf & Schmidt, 1997). A different random sequence was used for every trial; however, 1 One participant in the HC group performed the tracking task (CTT) with their dominant right arm. Statistical analyses were performed with and without this dataset; however, the results remained unchanged. The subsequent analyses included this individual (HC; n = 10). 116 to ensure uniformity, all participants practiced the same random tracking patterns. The random and repeated segments were linked at the crossing of the horizontal zero point (Figure 22d). Figure 22a, b, c, and d: Continuous tracking task (CTT). a. During the baseline and retention, the repeated and random waveforms segments were performed in separate blocks (4 blocks [15 trials of repeated then 15 trials random sequences]). b./c. Pictorial of the CTT apparatus used to perform the task during baseline and retention testing, and practice. Participants operated a joystick to move a closed red dot inside an open black circle on a computer screen. d. During practice days (2-6), individuals tracked continuous 20-s waveform segments (multiple overlapping lines represent different trials); repeating sequences were flanked with random sequences in a single block (5 blocks [10 trials of repeated and random sequences]), unbeknownst to participants. Baseline(Day 1)Practice Days (2-6)Retention Test (Day 7)RandomRepeatedRepeatedRandomRepeatedRandom15 trials15 trials15 trials15 trialsBlock 1 (x 4)Block 1 (x 4)1 trial x 10 Block 1 (x 5)Recognition Test (Day 7)AB C D 117 Recognition test Following the retention test on day 7, participants were shown 10 blocks of continuous target movement and asked to decide if they recognized any as the repeated pattern that they practiced. Three of the 10 were “true” repeating sequences (i.e., the same as the repeated practice pattern); seven were foils that had not been previously viewed by the study participants. Individuals who identified the repeated sequence at a better than chance rate (i.e., two of three repeated sequences identified correctly as being recognized were considered to have gained explicit awareness of the repeating sequence (Figure 22a) (Boyd & Winstein, 2004b; Meehan, Randhawa, et al., 2011; Wadden et al., 2015). Motor performance Motor performance was evaluated using root mean squared error (RMSE), which is the average difference between the target pattern and participant movements, and reflects overall tracking errors in the kinematic pattern. RMSE was calculated separately for random and repeating segments on every trial. Mean RMSE was calculated for every block across pre-test (15 trials; 4 blocks), practice days (10 trials; 5 blocks), and retention (15 trials; 4 blocks). Practice exponential curve Individual changes in implicit sequence-specific performance and motor control, as measured by the RMSE for repeated and random sequences, respectively, were fit to an exponential function with a least squares regression analysis. The RMSE value for each trials across the baseline (day 1), and five days of task performance (days 2 to 6), for each participant, were parameterized using the following equation (Brown & Heathcote, 2003) (equation 1): 1. 𝑬(𝑹𝑴𝑺𝑬N) = 𝑨 + 𝑩𝒆-α*N 118 E(RMSEN) is the expected value of RMSE on practice trial N; A is the expected value of RMSE after practice has been completed (asymptote parameter); B is the change in the expected value of RMSE from the beginning of practice to the end of practice (change score parameter); Alpha (α) is the exponential learning rate parameter (see Figure 23 for HC single-subject example) (Heathcote et al., 2000). A custom MATLAB (Mathworks, Natick, MA, USA) script was used for analyses. Figure 23: Skill acquisition follows an exponential decay as performance improves. Fitting root mean squared error (RMSE) data extracted from a motor skill practice to the function, E(RMSEN) = A + B-αN, produced practice parameters based on the nonlinear decay of performance during skill acquisition (Heathcote et al., 2000). Data shown here were derived from a sample healthy control (HC) participant. Curves were divided by a normalization factor of A + B; A = asymptote value, B = change score. Retention test score A retention test score (RTS) was calculated to determine the change in motor behaviour associated with learning. By accounting for early practice performance and the change score 00.20.40.60.811.21 51 101 151 201 251Normalized RMSETrialBAE(RMSEN) = A +Be-αN 119 parameter, B, as predicted from the exponential function, a direct comparison between practice and retention was possible. The mean RMSE for block 1 for the pre-test (day 1) and retention test were calculated (RMSEPT and RMSERT, respectively) (Krakauer, Pine, Ghilardi, & Ghez, 2000; Modabber, Neva, Gill, Budge, & Henriques, 2008). The RTS was determined as follows (equation 2): 2. RTS = (RMSEPT – RMSERT) / B In this equation, a higher score indicates better retention (i.e., reduced loss of motor performance following practice to retention) of the motor skill. The RTS is interpreted as the increase or decrease in motor learning-related change following the retention interval (i.e., period between practice and retention) (Haibach et al., 2011). Predicting dose of practice to asymptote To evaluate the predictive capabilities of the exponential function to determine the optimal dose of practice for motor learning, the exact trial was calculated that indicated individual performance was within 1% tolerance of the A value. The number of trials until asymptote was determined by the following equation (equation 3): 3. TrialA_n= -ln (0.01* A/B)/α Trialn is the expected trial N until asymptote; A is the expected value of RMSE after practice has been completed (asymptote parameter); B is the change in the expected value of RMSE from the beginning of practice to the end of practice (change score parameter); Alpha (α) is the exponential learning rate parameter (Figure 23) (Heathcote et al., 2000). Statistical analysis Primary analyses. To compare performance curve profiles during the acquisition of repeated and random sequences between ST and HC groups, separate (repeated and random) 120 between-GROUP (ST, HC) multivariate analyses of variance (MANOVAs) were used to evaluate differences in motor performance on the exponential practice parameter dependent measures (A, B, α), derived from the exponential function. To control for the effect of initial performance on motor performance-related change, the mean RMSE for the pre-test (day 1) were entered in the analyses. Based on objective evaluation of the normality of data distributions, the Shapiro-Wilk test with a significance level set at p < 0.001 (Gamst et al., 2008), indicated that B values for repeated and random sequences were non-normal. Consequently, for all statistical analyses, square-root transformations were applied to all variables in the MANOVAs. To assess differences in motor learning-related change, a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA was performed, with the RTS ((RMSEPT – RMSERT) / B) as the dependent measure. To evaluate the relationship between exponential practice parameters and RTS in the HC and ST groups, simple linear regression analyses were conducted (R2), in which the rate of skill acquisition parameter, α, was regressed on the RTS separately for repeated and random sequences2. Because stroke-related movement deficits may interfere with objective measurements of motor behaviour, we conducted hierarchical regression analyses (R2) in the ST group, designed to examine the relationship between α and RTS while accounting for individuals’ level of motor function. These hierarchical regression analyses were conducted separately for the repeated and random sequences; paretic WMFT rate was entered as a predictor in the first block, and the rate of skill acquisition parameter, α, was entered in the second block; and these variables regressed 121 on the RTS2. The variance inflation factor (VIF) and tolerance statistics indicated minimal collinearity within the data as VIF value was under 2.0 (Field, 2009). To investigate differences in the predicted numbers of trials until asymptote, a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA, with predicted trialA_n as the dependent variable. The Shapiro-Wilk test of normality indicated the predicted trialA_n measures for repeated or random sequences were not normally distributed (p < 0.001). Due to violations in normality following square-root transformations, predicted trialA_n was log-transformed to achieve normality for statistical analyses (p > 0.001). Secondary analyses. Conventional motor performance and learning analyses were performed to compare with our performance curve findings in the primary analyses. Baseline (day 1) motor performance on the random and repeated sequences was evaluated with a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA, with RMSE score as the dependent variable. Performance of the repeated and random sequences during practice sessions (days 2 to 6) was examined using a three-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) × SESSION (days 2 to 6; within-subjects factor) mixed-model ANOVA, with RMSE score as the dependent variable. To compare findings to primary analyses (practice performance curves), and to control for the effect of initial performance on motor performance-related change, the mean RMSE for the baseline (day 1) for the average repeated and random sequences was entered in the analysis. 2 There was no statistically significant relationship between the exponential function parameters alpha (α) and B within the ST and HC groups (p = 0.425). Therefore, the parameter B was used to calculate the retention score and did not confound the correlational analysis between practice parameter alpha (α) and the retention score. 122 To assess motor sequence learning at retention (day 7), a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA was performed, with RMSE score as the dependent variable. Similar to practice, the mean RMSE for the baseline (day 1) for the average repeated and random sequences was entered as a covariate. To evaluate the relationship between exponential practice parameters and absolute retention in the HC and ST groups, simple linear regression analyses (R2) were conducted, in which the rate of skill acquisition parameter, α, was regressed on mean RMSE at retention separately for repeated and random sequences. All data were visually inspected for skewness and kurtosis and objectively tested for normality with the Shapiro-Wilk test with a significance level set at p < 0.001 (Gamst et al., 2008). Homogeneity of variance assumptions were assessed with Levene's tests. Post hoc pairwise comparisons were performed via univariate ANOVAs, in the event of a significant main statistical test. A Bonferroni correction was used on post hoc analyses to correct for multiple comparisons. Effect sizes were reported as partial eta-squared (ηρ²) where 0.01 is considered a relatively small effect, 0.06 moderate and more than 0.14, a large effect (Gray & Kinnear, 2012). Data are presented in the text as mean (M) plus or minus standard deviation (SD) or standard error (SE). For all statistical tests, significance level was set to p ≤ 0.05. SPSS 22.0 (SPSS Inc., Chicago, IL, USA) statistical software was used for analyses. 123 Results Curve Fitting: goodness-of-fit Following exponential curve fitting, one individual in the HC group was excluded from subsequent analyses for being an extreme outlier; demonstrating a negative B value and an alpha value (rate of skill acquisition) of a high growth factor (α = 0.85). On a group-level, the goodness-of-fit for the exponential curve fitting method (Lang & Bastian, 2001) was calculated. Average R2 for the repeated sequence for the HC and ST groups were 0.37 (SD = 0.20) and 0.26 (SD = 0.17), respectively. For the random sequence, for the HC and ST groups, average R2 was 0.30 (SD = 0.15) and 0.27 (SD = 0.17), respectively. To visually compare conventional methods (average performance for each block across practice days) and parametrizing data, all individual raw performance data and normalized curves have been plotted for both HC and ST groups (Figure 24a, b and 25a, b). If 10% of the variance was explained by the fit, this was deemed an adequate representation of the data (Lang & Bastian, 2001). Here, we were able to account for 30% or more of the variance for the HC group and 26% for the ST group. Therefore, based on this criterion, our group data adequately represented by our curve fitting approach. For repeated sequence performance, normalized practice curves across all trials and mean RMSE data for each block of practice are shown in Figures 24 and 25 for the HC and ST groups, respectively. For each participant, curves were divided by a normalization factor of A + B, where A = asymptote value and B = change score. 124 Figure 24a and b: Healthy control (HC) repeated sequence performance data. a. Normalized practice curves across all trials for the HC group. b. Mean root mean squared error (RMSE) data for each block of practice across pre- and practice days (1 to 6) for the HC group. For each participant, curves were divided by a normalization factor of A(i) + B(i); A = asymptote value; B = change score; i = participant. Figure 25a and b: Individuals with stroke repeated sequence performance data. a. Normalized practice curves across all trials for the ST group. b. Mean root mean squared error (RMSE) data for each block of practice across pre- and practice days (1 to 6) for the ST group. For each participant, curves were divided by a normalization factor of A(i) + B(i); A = asymptote value; B = change score; i = participant. a. b.0510152025303540Day1_1Day1_2Day1_3Day1_4Day2_1Day2_2Day2_3Day2_4Day2_5Day3_1Day3_2Day3_3Day3_4Day3_5Day4_1Day4_2Day4_3Day4_4Day4_5Day5_1Day5_2Day5_3Day5_4Day5_5Day6_1Day6_2Day6_3Day6_4Day6_5Mean RMSEBlocksHC1 HC2 HC3 HC4 HC5HC6 HC7 HC8 HC9 HC1000.20.40.60.811.21 11 21 31 41 51 61 71 81 91 101111121131141151161171181191201211221231241251261271281291301Normalized RMSETrialsHC1 HC2 HC3 HC4 HC5 HC6HC7 HC8 HC9 HC10 HCMean00.20.40.60.811.21 11 21 31 41 51 61 71 81 91 101111121131141151161171181191201211221231241251261271281291301Normalized RMSETrialsST1 ST2 ST3 ST4 ST5ST6 ST7 ST8 ST9 ST10ST11 ST12 ST13 ST14 STMean0510152025303540Day1_1Day1_2Day1_3Day1_4Day2_1Day2_2Day2_3Day2_4Day2_5Day3_1Day3_2Day3_3Day3_4Day3_5Day4_1Day4_2Day4_3Day4_4Day4_5Day5_1Day5_2Day5_3Day5_4Day5_5Day6_1Day6_2Day6_3Day6_4Day6_5Mean RMSETrialsST1 ST2 ST3 ST4 ST5 ST6 ST7ST8 ST9 ST10 ST11 ST12 ST13 ST14a. b. 125 Differences in performance curve metrics between ST and HC groups: primary analyses Using our exponential function to characterize change in motor performance across practice of the repeated sequence, when controlling for baseline performance, there was a significant difference between the groups (ST, HC) for the exponential practice parameters derived from the repeated sequences (A, B, α), Wilks’ λ = 0.62, F(3,19) = 3.85, p = 0.026, ηρ² = 0.38, large effect size. Post hoc univariate ANOVAs showed that a significant difference for the exponential practice parameters was observed for the extracted α value, F(1,21) = 5.12, p = 0.034, ηρ² = 0.20, large effect size (Figure 26); individuals in the HC group (M = 0.026, SD = 0.0221) had a faster rate of skill acquisition compared to the ST group (M = 0.011 , SD = 0.00960) for repeated sequence trials. There were no significant differences between groups for the exponential practice parameters A or B for repeated sequence practice, F(1,21) ≤ 0.3.60, ps ≥ 0.071. There was no significant main effect of GROUP (ST, HC) on the practice parameters for random sequence detected by the MANOVA (A, B, α), Wilks’ λ = 0.80, F(3,19) = 1.57, p = 0.23, ηρ² = 0.20, large effect size. 126 Figure 26: Repeated sequence practice curves. Curves show a significant difference between the healthy control (HC) and stroke (ST) groups for exponential practice parameters for the extracted Alpha (α) value, F(1,21) = 5.12, p = 0.034, ηρ² = 0.20, large effect size. * indicates a statistically significant at p ≤ 0.05. Differences in motor learning curve metrics in retention: primary analyses For RTS (used to characterize performance lost or gained following the retention interval), there was no main effect of SEQUENCE, (F(1, 22) = 0.40, p = 0.53 , ηρ² = 0.018, small effect size), or GROUP, (F(1, 22) = 0.20, p = 0.66, ηρ² = 0.009, small effect size). However, the ST group demonstrated observably more loss in performance from practice to retention for repeated sequence performance than the HC group. The RTS for the ST group’s repeated sequence was 0.55 (SD = 0.385), and for the HC group RTS was 0.63 (SD = 0.248); this translates to a 45% and 37% loss in performance over the retention interval for the ST and HC groups, respectively. Conversely for the random sequence trials, the ST group demonstrated observably reduced loss in random trial performance over the retention interval than the HC group. The ST group had a 00.20.40.60.811.21 21 41 61 81 101 121 141 161 181 201 221 241 261 281 301 321Normalized RMSETrialsSTHCα = 0.026 α = 0.011 *p = 0.034 127 RTS of 0.62 (SD = 0.258) and for the HC group, RTS = 0.46 (SD = 0.321); this translates to a 38% and 54% loss in performance over the retention interval for the ST and HC groups, respectively. Nevertheless, there was no significant GROUP × SEQUENCE interaction (p = 0.13, ηρ² = 0.10, moderate effect size). Relationship between α and retention score: primary analyses Simple linear regressions were used to evaluate the relationships between α and the RTS ((RMSEPT – RMSERT) / B) for both repeated and random sequence performance for the HC group. The practice parameter, α, accounted for 46.5% of the variance in repeated sequence RTSs (R2 = 0.46, F(1,8) = 6.95, p = 0.030) (Figure 27a), indicating a significant relationship between rate of skill acquisition in practice and motor learning-related change at the retention test. This relationship was not significant for random sequence performance (R2 =0.09, F(1,8) < 1, p > 0.05). A similar analysis to that above was also performed on the ST group data. The practice parameter, α, again accounted for significant variance in repeated sequence RTSs (R2 = 0.32, F(1,12) = 5.57, p = 0.036), but no relation was shown for the random sequence (R2 = 0.07, F(1,12) < 1, p > 0.05, Figure 27b). In the hierarchical regression models for the ST group, paretic WMFT rate accounted for significant variance in RTS for both repeated and random sequences (R2 = 0.42, F(1,12) = 8.83, p = 0.012; R2 =0.41, F(1,12) = 8.34, p = 0.014, respectively). The addition of α in the second block improved the repeated sequence model (ΔR2 = 0.17, ΔF(1,11) = 4.43, Δp = 0.059), but this was not significant. The addition of α did not significantly improve the random sequence model (ΔR2 = 0.06, ΔF(1,11) =1.23, Δp = 0.29). The VIF was 1.07. 128 Figure 27a and b: Repeated sequence performance. a. Relationships of retention test score (RTS) and Alpha (α) during repeated sequence performance in healthy individuals (HC group) (p ≤ 0.05). b. Relationships of RTS and Alpha (α) during repeated sequence performance in individuals with stroke (ST group) (p ≤ 0.05). Predicted dose of practice between ST and HC groups: primary analyses Based on our calculation for the predicted number of trials until asymptote, there was a significant main effect of SEQUENCE, F(1, 22) = 8.84, p = 0.007, ηρ² = 0.29, large effect size, whereby there were more predicted trials until asymptote in performance for repeated sequence compared to random trial performance. However, there was no main effect of GROUP, F(1, 22) = 0.99, p = 0.33, ηρ² = 0.043, small effect size, for predicted trialA_n. For the repeated sequences; ST trials = 1755.18 (SD = 3504.68), HC trials = 414.34 (SD = 436.662), and for the random sequences; ST trials = 328.27 (SD = 399.151); HC trials = 520.41 (SD = 802.027). Discrete motor performance and learning measure: secondary analyses At baseline (day 1), there was no main effect of SEQUENCE on mean RMSE, F(1, 22) = 1.20, p = 0.28, ηρ² = 0.052, small effect size, and no GROUP × SEQUENCE interaction, F(1, 22) = 1.07, p = 0.31, ηρ² = 0.046., small effect size Observably, for repeated sequence performance, participants in the ST group had higher RMSE scores (M = 15.59, SD = 4.872) than the HC group (M = 13.50, SD = 8.739). Similarly, for random trial performance, participants in the ST 00.20.40.60.811.21.40 0.01 0.02 0.03 0.04 0.05 0.06 0.07Retention Test ScoreAlpha (α )Retention test ScoreAlpha (α)a. b.0.00.20.40.60.81.01.21.40.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07R2 = 0.46 R2 = 0.32 129 group’s mean RMSE (M = 15.56, SD = 5.030) was higher than the HC group (M = 12.17, SD = 5.519). Across the five days of practice there was a significant main effect of GROUP, with the ST group (M = 11.95, SE = 0.926) demonstrating a higher RMSE than the HC group (M = 8.68, SE = 1.096), F(1, 21) = 5.91, p = 0.024, ηρ² = 0.22, large effect size. When controlling for baseline performance (day 1), there was no main effect of SESSION (days 2-5), F(2.7, 57.0) = 0.76, p = 0.56, ηρ² = 0.035, small effect size; RMSE did not show a significant change across practice days for both the ST and the HC groups (p > 0.05). There was no main effect of SEQUENCE, as both groups demonstrated comparable performance on random trials and the repeated sequence, F(1, 21) = 2.08, p = 0.16, ηρ² = 0.09, moderate effect size. Additionally, there was no significant SESSION × SEQUENCE interaction effect, F(2.5, 52.2) = 0.88, p = 0.48, ηρ² = 0.040, small effect size, demonstrating equivalent RMSE for the repeating and random sequences across practice days (2-5). At retention (day 7), when controlling for baseline performance (day 1), there was no significant main effect of SEQUENCE on mean RMSE; both the ST and HC groups performed the repeated sequence (ST: M = 11.85, SD = 4.689; HC: M = 8.19, SD = 1.80) and random trials (ST: M =11.23, SD = 4.364; HC: M = 7.75, SD = 1.536) with comparable accuracy, F(1, 21) = 0.041, p = 0.84, ηρ² = 0.002, small effect size. As observed by the significant main of effect of GROUP on mean RMSE, F(1, 21) = 4.49, p = 0.046, ηρ² = 0.18, large effect size, the HC group (M = 8.19, SE = 1.79) performed the tracking task with less error than the ST group (M = 11.70, SE = 0.99). There was no significant GROUP × SEQUENCE interaction, F(1, 21) = 0.023, p = 0.88, ηρ² = 0.001, small effect size). 130 Relationship between α and absolute retention: secondary analyses Simple linear regressions were used to evaluate the relationships between α and the mean RMSE at retention for both repeated and random sequence performance for the HC group. Based on the significant linear regression model, (R2 = 0.59, F(1,8) = 11.45, p = 0.010), a positive relationship between rate of skill acquisition in practice and absolute mean RMSE at the retention test was observed. This relationship was not significant for random sequence performance (R2 = 0.21, F(1,8) = 2.09, p = 0.19). For the ST group, the practice parameter, α, did not account for significant variance in the repeated sequence’s (R2 = 0.014, F(1,12) = 0.17, p = 0.69), or random sequences’ (R2 = 0.43, F(1,12) = 0.53, p = 0.48) mean RMSE at the retention test. Testing of explicit knowledge Participants could only recognize sequences at a chance level and failed to gain explicit knowledge of the repeated sequence. The ST group correctly identified 53.4% of sequences; the HC group accurately recognized 53.3%. The difference between recognition rates was not statistically significant (p > 0.05). Discussion Here we showed that individual motor performance of a CTT can be adequately modelled by an exponential function (based on the goodness-of-fit data) and that this function is significantly related to the change in performance following the retention interval of a newly practiced motor skill. During motor sequence acquisition, as observed in our primary analyses, individuals with chronic stroke demonstrated significantly slower improvements in motor performance of a repeated sequence embedded in the CTT, compared to a HC group. Inferior motor sequence performance and retention by individuals with stroke as compared to matched 131 controls has been previously reported (Meehan, Randhawa, et al., 2011; Wadden et al., 2015). In the present study, as observed in our secondary analyses, there was a significant difference between ST and HC groups across practice days and at the retention test for performance of the CTT. However, for both groups, there was no change in tracking performance for repeated and random sequences across practice days. Furthermore, at the retention test, based on the motor learning-related change measures of performance, there was no significant difference between tracking accuracy of the repeated sequence and random trials, and thus groups did not experience implicit motor sequence learning. Our primary analysis of performance curves predicted a greater number of trials until asymptote for the repeated sequence compared to random trials performance for both ST and HC groups. Therefore, this lack of implicit-sequence specific learning may reflect the need for greater dose of practice of repeated sequences until motor learning-related change is achieved. However, further research is needed to investigate the presence of implicit-sequence specific learning during the performance of continuous tracking task in older adults and individuals with stroke. Furthermore, it is possible that decomposing tracking error in components of spatial accuracy and temporal precision may have helped differentiate the evolution of disparate aspects of skill acquisition that were not detected by examining RMSE (Mang et al, 2014). The current study expands on previous stroke-related motor learning research (Deuschl et al., 1996; Ioffe et al., 2006; Lang & Bastian, 1999, 2001) showing here that the rate of motor skill acquisition differs between individuals with stroke and healthy controls, and that the rate of acquisition is predictive of some measures of retention. The ST and HC groups showed a positive relationship between a faster rate of change in motor performance during practice and change in performance following the retention interval for the repeated sequence. Interestingly, a 132 near significant relationship was shown for individuals with stroke after we accounted for overall motor function by including WMFT rate in our model. Although, there was not a significant change in variance accounted for when this variable was added into a predictive model, the final model (WMFT rate and rate of motor skill acquisition) showed a strong trend (p = 0.059) with a large effect size (r = 0.77) and explained 58.9% of the variance in the change in performance following the retention interval. Further investigation is needed to determine if this relationship is robust and to add power to the current analyses (an often-encountered issue when working with stroke populations). A pre- to post-testing approach ignores large amounts of information that characterizes individual capacity for change, rates of improvement, and time to asymptote. Within-individual variability has been reported to be between 37% to 53% of between-individual variability, which challenges the idea that motor performance on one day represents a person’s characteristic performance (Nesselroade & Salthouse, 2004). Further, when motor performance scores, such as clinical outcome measures, are used to prognosticate and determine whether therapeutic resources should be deployed, a measurement at a single point in time may not adequately reflect the full potential of the individual. Here we provide evidence based on quantifying the trend in performance over multiple days of practice, that for healthy individuals and individuals with stroke, evaluating the rate of motor skill acquisition during implicit sequence-specific learning may enhance our ability to predict capacity for retention of a new motor skill following practice. This finding extends upon previous motor learning literature (Lee, 2004) that demonstrates performance across multiple practice sessions may be useful in determining the retention of motor skills under certain conditions, as opposed to examining performance within a single discrete episode. Specifically, practicing multiple tasks under conditions with high contextual 133 interference (i.e., randomly switching between tasks) has been shown to reduce performance in practice, but result in enhanced long-term retention performance (Guadagnoli et al., 1999; Lee & Magill, 1983; Magill & Hall, 1990; Rey et al., 1987; Shea & Morgan, 1979). Our findings extend previous results, and present the rate of motor skill acquisition as a practice outcome measure that goes beyond describing the performance-learning relationship to predicting motor learning-related change based on performance curves. We used a relative retention measure to quantify the amount of performance that was lost, or gained, following the retention interval. In clinical settings, relative retention is an important method to measure the effectiveness of interventions and compare differences in motor learning-related change in a population with a heterogeneous range of clinical impairments (Aman, Elangovan, Yeh, & Konczak, 2014). RTS was calculated from the ratio of the exponential derived parameter B (similar to the change score from the beginning to the end of practice) as well as the change in performance from the beginning to retention. This is a unique approach because a change in practice performance (y values) across the entire practice period (x values) is based on the expected asymptotic value A. Parameterizing practice and learning to an individual’s predicted performance values may offer more insight into her or his potential capacity for change beyond constraints associated with a limited number of trials. The rate of motor skill acquisition, which for an exponential function reflects a fixed amount of what remains to be acquired (i.e., attain asymptotic value), positively related to reduced loss in performance over the retention interval for both ST and HC groups during the performance of the repeated sequence. Individuals that quickly achieved their asymptotic value in tracking accuracy for the repeated sequence showed reduced loss in performance, or a relative gain in performance, following the retention interval. Interestingly, for the HC group, higher rates of 134 motor skill acquisition in practice were related to less absolute tracking accuracy in the retention test. Therefore, the absolute and relative retention measures provided different information about performance at retention, and motor performance-related change following the retention interval, respectively, as well as the relationship with rate of motor skill acquisition. A slower rate of skill acquisition during repeated sequence practice may reflect continuous trial-and-error processes as HC individuals gradually learn the sequence regularities, potentially with less consistent movement patterns but a lowered overall mean tracking accuracy at retention. The asymptotic value we derived in the present work describes the estimated value of an individual’s performance when an apparent plateau is achieved. Some individuals may achieve their asymptote relatively quickly, while others may take substantially more trials to do so, which will significantly impact the α value. We did not observe a significant difference in the amount of predicted trials to asymptote between individuals with stroke and healthy controls. This may result from our testing of individuals with relatively mild levels of stroke-related motor impairment (average upper extremity Fugl-Meyer assessment motor score was 52 of 66). However, calculating predicted time to asymptote (A) immediately following practice has the potential to guide decisions regarding the dose of practice required for optimal improvement in motor behaviour and functional outcomes. Optimal dose of practice is an attractive concept in the field of rehabilitation following stroke because the dose required for neuroplastic change to occur is extremely high (Lohse, Lang, et al., 2014). The optimal dose of practice is characterized by an individualized number of repetitions that a patient would need to maximize retention of the desired task. Due to the exponential nature of learning, if dose prescription could be calculated based on performance data from individual performance curves, and therapists could quantify amount of practice necessary for the retention of motor skills, then the somewhat abstract 135 concept of dose could become a tangible measure. However, further investigation is needed to determine if this concept can be applied to all motor skills, beyond motor skills examined using laboratory tasks. In addition, this predictive methodological approach may not only specify whether the individual may require more practice trials, but may also provide an indication of the challenge level of the practice session. If the practice or rehabilitation session is too easy, then the predicted rate of skill acquisition during practice will be high and the number of trials to asymptote will be low; this may translate into lower long-term retention of the motor skill performance (Guadagnoli & Lee, 2004; Wright et al., 2016). One of the emphasized downfalls of the use of performance curves is that they are often erroneously interpreted as learning curves (Schmidt & Lee, 2013). Based on findings from the current and previous studies (Lakhani et al., 2016; Park & Schweighofer, 2017; Schaefer & Duff, 2014), the parameters of the curve (rate of skill acquisition) have unique relationships with processes of motor learning. In a recent study from our laboratory, in young healthy adults, the rate of motor improvement in practice for a semi-immersed, virtual reality motor task was inversely related to rate of motor performance in retention (Lakhani, et al., 2016; see also Chapter 2). Establishing the relationships between performance curves and long-term motor learning is a useful tool in rehabilitation; however, it is important to establish the generalized versus task-specific relationships between parameters estimated from the exponential function. In the present study, we were the first to determine a positive relationship between rate of implicit motor sequence acquisition and change in performance following the retention interval for healthy individuals and individuals with stroke. Thus, the present study contains a roadmap for future rehabilitation research that employs predictive models of motor learning based on individual performance curves and clinical characteristics of motor function. 136 Limitations and future directions There are several methodological considerations in the present study, which may be considered limitations. Primarily, we studied a relatively small (n = 14) group of mildly impaired individuals with stroke. This limits our ability to form robust associations between practice parameters, clinical measures, and motor learning, due to relatively low statistical power and an under-representative sample of participants compared to the general outpatient stroke population. To increase the ecological validity between clinical research and rehabilitation settings, where tests are administered by researchers and treatments are delivered by clinicians, respectively, a deeper understanding of the performance and learning relationship must be achieved in larger and more diverse patient populations, across a variety of motor tasks. In addition, our regression model demonstrated a moderate effect for our ST group, predicted explained 58.9% of variance in implicit-sequence learning. Therefore, further investigation is need to determine if this effect size, and thus strength of prediction, can be furthered increased if, for example, the optimal dose or intensity level is individualized prescribed within the ST group. We do not claim that our curve fitting approach is better than other methods; instead we present it as one possibility for capturing different data that is useful in understanding patterns of change and doses of practice associated with motor learning. We present the idea that information about patterns of change may be more helpful in the rehabilitation setting than is pre- to post-test characterizations of behavioural change. Better characterization of skill acquisition measures could enable data-driven manipulations of motor practice (to ensure that the adequate dose of practice is delivered) and the task (to optimize rates of change). Together, these shifts can enhance the impact of practice or rehabilitation sessions to optimize motor learning. 137 Conclusions While rates of skill acquisition have been used to quantify performance following stroke, no other study has investigated their association with retention performance to date. Further, the use of exponential functions to estimate the number of trials until plateau in motor behaviour has not yet been considered. Therefore, based on the present work we propose an innovative use for these curvilinear measures. Individualized dose of practice is an important step in the field of motor rehabilitation following stroke, not only to attain economic efficiency, but also to work towards optimizing personalized treatment plans. Currently, the most documented use of performance curves in a practical setting is the evaluation of skill acquisition for healthcare professionals (Eversbusch & Grantcharov, 2004; Flamme, Stukenborg-Colsman, & Wirth, 2005; Hernandez et al., 2004). In a rehabilitation setting, where individuals with stroke are receiving interventions to relearn motor skills, performance curves could be used to determine an individualized set number of trials that predict when performance will reach a plateau. In addition, the rate of skill acquisition between different interventions could be compared to determine practice conditions (i.e., feedback, contextual interference) that yield a more productive method of learning. For example, if the physiotherapist understands the normal or excepted rate of skill acquisition, then she or he can determine suitable practice conditions to positively affect the length of time and difficulty level required for achieving the intended outcome. If the rate of skill acquisition is abnormally slow, the physiotherapist could decrease the level of difficulty (e.g., increase the target size, shorten the distance of a reaching task, and so on), until the patient can switch to more difficult task conditions. If a final level of behaviour only assesses performance, large amounts of valuable information, which could have been used to update how interventions were being administered, are lost. Prescribing rehabilitative 138 exercises based on predictive values could help to construct an idealized practice paradigm on a subject-wise basis. 139 Chapter 4: Compensatory motor network connectivity: motor sequence learning after subcortical stroke Introduction The ability to acquire a movement skill without conscious awareness of improvements in performance is a fundamental aspect of motor learning. This process, known as implicit motor learning, is preserved in individuals with stroke and encompasses a large portion of motor skill rehabilitation (Boyd & Winstein, 2006; Boyd, Quaney, Pohl, & Winstein, 2007; Boyd & Winstein, 2001; Meehan, Randhawa, et al., 2011). However, it is unclear how the brain reorganizes to support the behavioural improvements observed after motor skill practice in individuals with stroke (Meehan, Randhawa, et al., 2011). In part, our failure to fully understand how the stroke-damaged brain reorganizes to support motor learning stems from methodological approaches in previous work that have mainly focused on functionally segregated brain structures (i.e., region of interest [ROI]–based and voxel-wise analyses) based on the locus of the lesion (Meehan, Randhawa, et al., 2011). Recent advances in neuroimaging analysis techniques promote a shift away from mapping brain function to individual areas, and place a greater emphasis on the assessment of integrated and reorganized functional brain networks, comprised of spatially discrete nodes bearing anatomical and functional connections (Westlake & Nagarajan, 2011). For the present study, we applied a novel multivariate approach to an fMRI analysis and identified functional brain networks underlying sequence-specific implicit motor learning. In previous work, researchers illustrated differences in regional brain activation between healthy and stroke individuals during implicit motor learning (Meehan, Randhawa, et al., 2011). 140 A comparison of differences in regional brain activation between these groups provides an indication of maladaptive recruitment, or lack of recruitment, of discrete brain regions following motor sequence learning in individuals with chronic stroke (Meehan, Randhawa, et al., 2011). Meehan et al. (2011) performed separate ANOVAs for baseline and retention on voxel-wise activation maps, comparing groups (healthy controls, stroke participants) and sequences of movement (repeated, random). Based on this approach, researchers observed significantly increased activation in the dorsal premotor cortex (PMd) and decreased activation in the dorsolateral prefrontal cortex (DLPFC) at retention for the healthy controls compared to individuals with stroke. This activation pattern suggests a transition from feedback mechanisms to feed-forward memory-based control during motor learning in healthy adults, as learning progresses (Abe et al., 2007; Debaere, Wenderoth, Sunaert, Van Hecke, & Swinnen, 2004). Individuals with stroke did not show a concomitant decrease in activation in the DLPFC with motor learning. This finding may indicate continued reliance on the prefrontal-based attentional network and compensatory regions that included the primary somatosensory cortex (S1), ipsilesional insula, bilateral superior frontal, and middle temporal gyri (Meehan, Randhawa, et al., 2011). This past work suggested that variances exist within the DLPFC-premotor network between healthy individuals and individuals with chronic stroke, yet it did not allow us to consider changes in network connectivity supporting implicit learning. This is important because post-stroke behavioural impairments are associated with disruptions in communication within distributed brain networks that are specific to behavioural domains (Carter, Shulman, & Corbetta, 2012). A different approach would be to investigate functional connectivity commonalities within a motor network across groups and then identify the connectivity differences between these groups. Thus, in our current study a multivariate approach was applied 141 to a previously published fMRI data set (Meehan, Randhawa, et al., 2011) to allow for the direct comparison to the univariate approach, as well as the characterization of functional network connectivity across groups. Subsequently, this network was used to identify differences in functional connectivity between groups that may explain disparities in motor skill learning between health controls and individuals with stroke. Functional connectivity analyses of fMRI data can be used to investigate temporal correlations between the hemodynamic responses (HDR) of spatially distant brain areas. Past work in this field has used methods such as independent component analyses (ICA) (Coynel et al., 2010), principal component analyses (PCA) (Ward, Brown, Thompson, & Frackowiak, 2003), and functional connectivity matrices and graph theory methods (Wang et al., 2010) to study activation patterns based on models of motor skill learning (Hikosaka et al., 2002). Limitations of standard ICA and PCA approaches involve difficulties with: (1) relating derived brain networks to behavioural tasks carried out while participants are in the magnetic resonance (MR) scanner; and (2) simultaneously comparing activity between two groups on one task-related network (Metzak et al., 2012). These limitations can be addressed using constrained principal component analysis (CPCA), a statistical technique that combines multivariate multiple regression and principal component analysis in a unified framework (Takane & Hunter, 2001; Takane & Shibayama, 1991; Woodward et al., 2006). When applied to fMRI (fMRI-CPCA; www.nitrc.org/ projects/fmricpca) (Woodward et al., 2006; Woodward et al., 2013), this technique allows isolation of task-relevant blood-oxygen-level-dependent (BOLD) signal fluctuations. The analyzed BOLD signal is constrained to the aspect of variance in BOLD signal that is predictable from the experimental design (i.e., presentation of stimuli), producing the derivation of images based on the degree to which one or more task-related functional networks 142 are involved in each experimental condition for each participant. The computations of functional networks are based on the analysis of interrelationships among cortical structures involved in the experimental task (Woodward et al., 2013). For the present study, we used whole-brain fMRI-CPCA to evaluate shared functional networks that may activate (or deactivate) during sequence-specific implicit motor learning for healthy controls and individuals with stroke. Individuals in our stroke group had right hemispheric subcortical lesions, which allowed us to use a data-driven approach to evaluate the connectivity within whole-brain networks rather than restricting our analysis to predefined ROIs (model-driven). This factor was key in our ability to run an unrestricted, task-based, whole-brain connectivity analysis (fMRI-CPCA). In previous research, at rest, connectivity between regions within the motor network are altered in the lesioned brain, and the change in functional connectivity between specific regions is predictive of stroke recovery (Wang et al., 2010). However, little is known about how implicit motor learning task-based networks are altered in the injured brain. fMRI-CPCA has the advantage of identifying shared functional networks underlying the performance of repeated versus random tracking movements, and thereby the degree of functional connectivity associated with learned sequences of movement, as compared to changes in generalized motor control. Using between-group comparisons, this method provides a statistical test of the degree to which brain reorganization after stroke is affecting shared functional networks, and the relationship between changes in functional connectivity and motor sequence learning. Given the results from previous multivariate analyses (Coynel et al., 2010; Wang et al., 2010), we hypothesized that in our primary whole-brain analysis, greater connectivity within a motor network would be observed during repeated compared to random performance at a 143 retention test. Second, based on our previous fMRI univariate study (Meehan, Randhawa, et al., 2011), we hypothesized that there would be a significant difference between network activation during repeated sequence performance for healthy controls compared to individuals with stroke. Third, we predicted that a secondary connectivity analysis, which exploits functional connectivity within regions restricted to the motor network from our primary analysis, would reveal a specific compensatory stroke-affected motor network. Fourth, we hypothesized that the level of connectivity within this stroke-affected motor network would be related to the change in motor behaviour for the implicit motor learning shown during the retention test following five days of motor skill practice. To test these hypotheses, we reanalyzed a previously published fMRI data set for which a univariate analysis was performed to identified different brain regions associated with sequence-specific implicit motor learning between healthy individuals and individuals with stroke (Meehan, Randhawa, et al., 2011). Methods Participants Nine first-time, right-hemisphere ischemic stroke (ST) participants with chronic (> 6 months) subcortical lesions (six men, mean [M] age = 63.8, SD = 6.2 years; M Fugl Meyer motor upper extremity [UE-FM] score = 54.3, SD = 13.0) and nine age-matched healthy controls (HC; four men, M age = 63.1, SD = 7.0 years) were recruited for the present study (Table 5; Figure 28). Exclusion criteria included: (1) a score below the 25th percentile on the Mini-Mental Status Exam (MMSE), using age adjusted norms (Crum, Anthony, Bassett, & Folstein, 1993); (2) left handedness; (3) neurological impairment or disease of individuals in the HC group (Lundy-Ekman, 1998); (4) inability to perform the task due to any orthopedic condition or color-blindness; or (5) contraindications or ineligibility to undergo magnetic resonance imaging (MRI). 144 Participants were recruited from the University of British Columbia (UBC), the local community, and the Brain Behaviour Lab database. Each participant’s consent was obtained according to the Declaration of Helsinki; the Research Ethics Boards at UBC approved all aspects of this work. Table 5: Participant characteristics Sex Age Mean (SD) Post-Strokea Mean (SD) UE-FMb Mean (SD) ST group 6 M; 3 F 63.88 (6.214) 53.22 (49.789) 54.33 (12.952) HC group 4 M; 5 F 63.11 (7.001) SD, standard deviation; ST, stroke; HC, healthy control; M, males; F, females. aPost-stroke duration is in months. bUE-FM = upper extremity portion of the Fugl-Meyer assessment (range for UE-FM is 0-66; lower scores denote less hemiparetic arm function). 145 Figure 28: T1 weighted images with lesion location. Participants’ anatomical images, including highlighted lesions in red for the stroke (ST) group. Behavioural task Participants tracked a moving, white, circular target on a computer monitor that followed a sine-cosine waveform. The ST group operated a joystick with their hemiparetic left arm; the HC group used their non-dominant left arm. Movements were represented as a red dot on a 19” computer monitor placed directly in front of participants. Unbeknownst to participants, one segment of each tracking trial of the moving target followed a predefined pattern constructed from a modified version of the method introduced by Wulf and Schmidt (1997). This pattern was repeated throughout practice (days 2 to 6) and during fMRI acquisition (baseline and retention, days 1 and 7). The random segments of the tracking task did not contain a pattern and a different 146 random sequence was used for every trial; however, the same set of random movements was used across participants (see Figure 29 for practice apparatus). During each day of practice, participants executed a total of 250 repetitions of the random and repeated sequences partitioned over five blocks each containing 10 sequence repetitions. The task was performed in a block design during fMRI acquisition (i.e., day 1 and day 7 only). Each block contained either random or repeated sequences (40 s rest and 150 s stimulus presentation; 40 s rest and 150 s stimulus presentation; 40 s rest) presented in a counterbalanced order within a run, and a total of four runs were performed. Following the fMRI retention test on day 7, a test was performed to assess whether participants had gained explicit awareness of the repeating sequence. Participants were asked if they recognized any of the ten, 10-s blocks of the sequence presented to them as they repeated the pattern they had practiced. Three of the 10 blocks presented during the test were the “true” repeating sequences and seven were randomly generated sequences (“foils”). If a participant correctly identified the repeated sequence at a better than chance level (i.e., correctly identified two of three repeated sequences as being repeated, and correctly identified four of the seven novel sequences as having never been seen before), then the participant was considered to have gained explicit awareness of the repeating sequence and was excluded from the final sample (Boyd et al., 2009; Vidoni & Boyd, 2008). In the current study, all individuals in both groups failed to gain explicit awareness; thus, none of the participants were excluded from the analyses based on failing to achieve implicit motor learning. To evaluate motor performance, our behavioural outcome measure for this continuous tracking task (CTT) was root mean squared error (RMSE). RMSE is representative of the overall 147 tracking error integrating both temporal and spatial measurements of time lag and distance from the moving target, respectively (Meehan, Randhawa, et al., 2011). Figure 29: Continuous tracking task (CTT). Pictorial of the CTT apparatus used to perform the task during the five days of practice. Participants operated a joystick to move a closed red dot inside an open black circle to track 20-s waveform segments (multiple overlapping lines represent different trials) separated by random and repeating sequences on a computer screen. The functional magnetic resonance imaging (fMRI) design performed during baseline (day 1) and retention (day 7) used random and repeated sequences that were counterbalanced across scans and performed over 150-s blocks. Imaging protocol MR imaging was performed at the UBC MRI Research Centre on a Philips Achieva 3.0 T whole-body MRI scanner (Philips Healthcare, Andover, MD, USA) using a sensitivity encoding head coil (SENSE). Blood-oxygen-level-dependent (BOLD) images were acquired axially using echo-planar images (EPI) with a single-shot readout (TR = 2,000 ms, TE = 30 ms, flip angle θ = 90°, FOV = 240 mm, 36 slices, 3 mm thickness with a 1 mm gap). A high-resolution anatomical scan was collected (TR = 12.4 ms, TE = 5.4 ms, flip angle θ = 8°, FOV = 256 mm, 170 slices, 1 mm thickness) for later co-registration with the functional maps. 148 fMRI data analysis Preprocessing Brain images were preprocessed using statistical parametric mapping (SPM 8) software (Wellcome Department of Cognitive Neurology, London, UK) in the MATLAB environment (MATLAB version 7.6, The MathWorks Inc., Natick, MA, USA). Images were corrected for slice timing and realigned for motion correction. The functional and anatomical images were co-registered, normalized into MNI standard space (Montreal Neurological Institute, Quebec, Canada), and spatially smoothed using a Gaussian kernel of 8 mm full width at half maximum (FWHM). Using icb2tal transform software (Research Imaging Institute of the University of Texas Health Science Center, San Antonio, TX, USA), the fMRI-CPCA results were converted to Talairach space to report results in Talairach standard space (Talairach & Tournoux, 1988). fMRI-CPCA fMRI-CPCA MATLAB-based software was applied to extract functionally-connected brain networks associated with motor sequence learning at the baseline (day 1) and retention (day 7) (MATLAB version 13.0, The MathWorks Inc., Natick, MA, USA). The theory and proofs of CPCA are detailed in previously published work (Takane & Hunter, 2001; Takane & Shibayama, 1991); for applications to fMRI data, refer to (Metzak et al., 2011; Rapin et al., 2012; Woodward et al., 2006; Woodward et al., 2013). In short, when applied to fMRI data, three main steps are carried out (see Figure 30). In the first step, referred to as the external analysis, a multivariate least-square multiple regression is carried out to isolate the BOLD signal variability that is predictable from the stimulus timing model; in this case, the stimulus timing model is a hemodynamic response (HDR) shape derived from the timing of the blocks in the tracking task convolved with a hemodynamic response function obtained from SPM 8. In the 149 second step, the predicted scores from the multivariate multiple regression are submitted to a PCA using singular value decomposition. In a third step, weights are computed which, when they are applied to the stimulus timing model, produce component scores. These weights, termed “predictor weights,” reflect the intensity of the functional network for each subject and condition (i.e., repeated versus random). Further statistical analyses can then be performed on the predictor weights to examine group– and condition-level connectivity differences on the functional network(s) derived from the PCA. Secondary region of interest fMRI-CPCA The whole-brain functional network that explained the most variance from each of the baseline and retention fMRI CPCA analyses was used to perform ROI-based analyses exclusively within individuals with stroke. This secondary analysis assessed stroke-specific spatial distribution of activation within the task-dependent networks. Binary masks (ROIs) were created for each day (day1, day 7), which included the most extreme 30% of voxels (highest component loadings). New masked baseline and retention fMRI-CPCA analyses were performed for the ST group data. 150 Figure 30: Functional magnetic resonance imaging and constrained principal component analysis (fMRI-CPCA) methodology. The fMRI-CPCA methods are delineated for whole-brain and masked analyses outlining the chronological steps (input, analytic processing, output) used to derive the functional connectivity networks (see Methods for details). Statistical inference procedure Functional connectivity network: Inspection of the scree plot (Cattel, 1966; Cattel & Vogelmann, 1977) of singular values suggested that a single functional network should be extracted for further analysis for both baseline and retention whole-brain analyses. Similarly, a single functional network was extracted for the secondary ROI-based fMRI-CPCA analyses at baseline and retention as demonstrated by the scree plots. To test for differences in the estimated BOLD response associated with motor performance, separate GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model analyses of variance (ANOVAs) for each day (baseline, retention) were performed on the predictor weights of the functional network resulting from the PCA. To test our a priori hypotheses, first, planned matched paired t-tests were performed for the BOLD response associated with repeated and random sequence performance 151 within HC and ST groups. Second, a planned independent t-test was performed for estimated BOLD response associated with repeated sequence performance between HC and ST groups. In the secondary connectivity analysis, to test for differences in the estimated BOLD response within the motor network for the stroke group, separate SEQUENCE (repeated, random) univariate ANOVAs for each day (baseline, retention) were performed on the predictor weights of the functional network. Post hoc linear contrasts were used to further examine significant interactions from these ANOVAs. Pre-training performance: Day 1 (baseline) motor performance on the random and repeated sequences was evaluated with a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA with mean RMSE as the dependent variable. Practice performance: Performance of the repeated and random sequences during practice sessions (days 2 to 6) was examined using a three-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) × SESSION (days 2 to 6; within-subjects factor) mixed-model ANOVA with mean RMSE as the dependent variable. Retention performance: To assess motor sequence learning at retention (day 7), a two-factor GROUP (HC, ST; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA was performed with mean RMSE as the dependent variable. Post hoc linear contrasts were used to further examine significant interactions from practice and retention performance ANOVAs. Behavioural data are presented from previously published work (Meehan, Randhawa, et al., 2011). Functional connectivity network and motor performance: To assess the relationship between level of functional connectivity within the motor network and implicit motor learning 152 for the stroke group, a Pearson’s correlational analysis (r) was conducted on predictor weights extracted from the secondary analysis and mean RMSE on retention (absolute retention value) for both repeated and random conditions. In addition, age, post-stroke duration, and upper extremity Fugl-Meyer (UE-FM) were correlated with predictor weights, as these measures have been previously shown to have a strong relationship with neurophysiological measures such as WM integrity (fractional anisotropy [FA]) (Borich, Mang, et al., 2012). All data were visually inspected for skewness and kurtosis and objectively tested for normality with the Shapiro-Wilk test with a significance level set at p < 0.001 (Gamst et al., 2008). In the event of a violation of sphericity (significant Mauchly’s test, p < 0.05), the Greenhouse-Geisser correction was applied to adjust degrees of freedom. Effect sizes were reported as partial eta-squared (ηρ²) where applicable, where 0.01 is considered a relatively small effect, 0.06 moderate and more than 0.14, a large effect (Gray & Kinnear, 2012). In addition to reporting ηρ², Pearson’s correlation coefficients (r) were used to improve the interpretability; where r squared is the portion of the total variability explained by the effect in question. Values of r between 0.10 and 0.30 are considered a small effect, between 0.30 and 0.50 a moderate effect and more than 0.50 are considered a large effect (Field, 2009). The 95% confidence intervals (CIs) of the MDs were used to describe the effect of the motor sequence practice on the dependent measures. Data are presented in the text as mean (M) and standard deviations (SD), and mean difference (MD) and standard error (SE). Significance level for all statistical tests was set at p ≤ 0.05. All statistical procedures were conducted using SPSS software (SPSS 19.0, SPSS Inc., Chicago, IL, USA). 153 Results Whole-brain functional connectivity during motor sequence learning At baseline, the functional network included right precentral gyrus, left superior parietal lobule, and the cerebellum bilaterally (Table 6 and Figure 31). This component accounted for 32.91% of the variance predictable from the HDR model of the block design on baseline. The functional network distinguished at retention (day 7) included right postcentral gyrus, superior parietal lobule, occipital lobe, and the cerebellum bilaterally (Table 7 and Figure 31). This component accounted for 23.8% of the variance predictable from the HDR model of the block design at retention. Separate ANOVAs were conducted on component one-predictor weights from baseline and retention. There was a significant main effect of SEQUENCE at retention, but not at baseline (non-significant baseline main effect; p = 0.084). At retention, analysis of the predictor weights showed a significant main effect of SEQUENCE, F(1, 16) = 14.86, p = 0.001, r = 0.69, large effect size, a significant main effect of GROUP, F(1, 16) = 5.94, p = 0.027, r = 0.52, large effect size, and a near significant GROUP × SEQUENCE interaction effect, F(1, 16) = 4.36, p = 0.053, r = 0.46, moderate effect size . While the interaction did not meet conventional significance (p ≤ 0.05), based on interpretation of the r, the interaction effect accounted for 21% of the total variance. Preplanned comparisons showed a significant increase in network activity for healthy participants during repeated, relative to random sequence performance (MD = 0.21, SE = 0.052), t(8) = 4.13, p = 0.003, r = 0.82, 95% CI [0.096, 0.34], large effect size. No such differences were observed for individuals with stroke (p = 0.24) (Figure 31). Our second planned comparisons showed a significant increase in network activity for healthy participants compared to individuals with stroke (MD = 0.98, SE = 0.38), t(16) = 2.57, p = 0.020, r = 0.54, 95% CI [0.17, 1.80], large 154 effect size, during the repeated sequence. Based on interpretation of the r, the main effect of sequence accounted for 67% of the total variance of functional connectivity for the healthy participants, while the main effect of GROUP (HC, ST) for repeated sequence functional connectivity accounted for 29% of the total variance. Therefore, there was a greater difference in functional connectivity between sequences (repeated, random) within the HC group than between the HC and ST groups for functional connectivity during the repeated sequence. ROI functional connectivity for the ST group At baseline, the functional network included left lingual gyrus and right parietal lobule, middle temporal gyrus, and superior and inferior frontal gyri (Table 6 and Figure 31). This component accounted for 36.79% of the variance predictable from the HDR model of the block design at baseline. The functional network noted at retention (day 7) included left medial frontal gyrus and right insular cortex, lateral occipital cortex, inferior temporal gyrus, and bilateral cerebellum (anterior lobe and culmen; Table 7 and Figure 31). This component accounted for 30.36% of the variance predictable from the HDR model of the block design on retention. Separate ANOVAs were performed on the predictor weights for baseline and retention. Neither of these analyses revealed a main effect of sequence (p = 0.12, p = 0.83, respectively) for the ST group. 155 Table 6: Cluster volumes, baseline (day 1) Showing data for whole-brain and masked component one at baseline (day 1) for the healthy control (HC) and stroke (ST) groups, respectively, with anatomical descriptions and Talairach coordinates for peaks within each cluster. 156 Table 7: Cluster volumes, retention (day 7) Cluster volumes for whole-brain and masked component one at retention (day 7) for the healthy control (HC) and stroke (ST) groups, respectively, with anatomical descriptions and Talairach coordinates for peaks within each cluster. 157 Figure 31a and b: Plots of whole-brain and masked predictor weights on retention, and images for component one networks revealed by functional magnetic resonance imaging and constrained principal component analysis (fMRI-CPCA) at baseline and retention. a. The graph displays the mean predictor weights plotted as a function of condition (repeated, random) and group (stroke, healthy control) and analysis (whole brain and region of interest [ROI]-based fMRI-CPCA) (error bars are standard errors) for baseline (day 1) and retention (day 7). b. The top panel displays the positive loadings of component one for baseline (day 1). The whole-brain network is displayed in red for the dominant 5% of loadings (red, min. = 0.21; white, max. = 0.40) and the masked network in blue for the dominant 5% of loadings (blue, min. = 0.12; turquoise, max. = 0.15) of the motor network overlaid on a structural image. The bottom panel displays the positive loadings of component one for retention (day 7). The whole-brain network is displayed in red for the dominant 5% (red, min. = 0.17; white, max. = 0.30) of the motor network and the masked network in blue for the dominant 5% (blue, min. = 0.11; turquoise, max. = 0.14) of the motor network overlaid on a structural image. 158 Behavioural task performance At baseline (day 1) participants in the ST group had higher RMSE scores than the HC group (MD = 3.28, SE = 1.52), F(1,16) = 4.70, p = 0.046, ηρ² = 0.23, large effect size. However, as hypothesized, at baseline there was no effect of SEQUENCE on RMSE scores (F(1,16) < 1, p > 0.05). Across practice there was a significant main effect of GROUP, with the ST group demonstrating a higher RMSE than the HC group (MD = 3.48, SE = 1.23), F(1,16) = 8.07, p = 0.012, ηρ² = 0.33, large effect size. Yet, as demonstrated by the main effect of SESSION (days 2-5), during practice both the ST and the HC groups reduced RMSE, F(2.15, 34.48) = 9.13, p = 0.001, ηρ² = 0.36, large effect size. Additionally, there was a SEQUENCE × SESSION interaction, as both groups demonstrated greater change in repeated sequence performance compared to random sequences, F(2.06, 32.90) = 3.58, p = 0.038, ηρ² = 0.18, large effect size. Based on this significant interaction, two separate post hoc repeated measures ANOVAs for SEQUENCE (repeated, random) were performed. There was a significant main effect of SESSION, F(1.97, 31.58) = 15.42, p < 0.001, ηρ² = 0.49, large effect size, for the repeated sequence across groups, but no main effect of SESSION, F(2.31, 37.05) = 1.85, p = 0.17, ηρ² = 0.10, moderate effect size, for random sequences across groups. At retention, as observed by the main effect of SEQUENCE on RMSE score, both the ST and HC groups were more accurate during the repeated sequence than during the random sequence (MD = 0.97, SE = 0.26) (F(1,16) = 14.34, p = 0.002, ηρ² = 0.47, large effect size). The HC group performed the tracking task with less error than the ST group as observed by a significant main of effect of GROUP, F(1,16) = 6.70, p = 0.020, ηρ² = 0.29, large effect size. 159 However, there was no significant GROUP × SEQUENCE interaction, F(1,16) = 1.83, p = 0.19, ηρ² = 0.10, moderate effect size) (Figure 32). Figure 32: Tracking performance. Tracking performance for baseline (day 1), and five days of practice and retention (day 7) for the healthy control (HC) and stroke (ST) groups. The average root mean squared error (RMSE) is displayed across days for the continuous tracking task (CTT) (bars represent standard error). Relationship between functional connectivity and motor learning In the ST group, there was a significant relationship between the ROI-based fMRI-CPCA predictor weights and RMSE during implicit motor learning at the retention test. Specifically, individuals who demonstrated less tracking error during implicit motor learning showed greater motor network connectivity at retention (r = –0.73, p = 0.025; Figure 33). Age, post-stroke 160 duration and Fugl-Myer were not significantly correlated with implicit motor learning (all rs < + 0.43; ps > 0.05). Figure 33: Relationship between functional connectivity in the masked motor network and repeated tracking performance for the stroke (ST) group at retention (day 7). This graph depicts the significant negative correlation between the predictor weights for the masked motor network and tracking performance (average root mean squared error [RMSE]) at retention (day 7) for the ST group. Discussion In the current study, we used fMRI-CPCA to evaluate the neural networks involved in sequence-specific implicit motor learning in healthy individuals and individuals with chronic stroke. The whole-brain functional connectivity analysis demonstrated a well-defined and characteristic large-scale motor network during the performance of the CTT. Separate analyses at baseline and retention testing following five days of motor skill practice resulted in activation of a motor network for both repeated and random conditions. At baseline, repeated and random conditions shared similar levels of connectivity within this motor network. At retention, for the performance of both the repeated and random sequences, the HC group showed significantly greater motor network activation compared to the ST group. For the sequences, there was greater 161 motor network activation for the repeated sequence compared random sequences across groups. The GROUP × SEQUENCE interaction effect trended towards significance (p = 0.053); there was an observably larger difference between motor network activation for repeated and random sequences in the HC group compared to the ST group. Our primary hypothesis was that there would be between-group differences in motor network activity during repeated versus random sequence performance at a retention test. Therefore, we performed planned paired t-tests on motor activity patterns from repeated versus random sequences for each group separately. The HC group showed activation in the motor network during the performance of both repeated and random sequences, with significantly greater functional connectivity observed during the repeated sequence. The motor network in healthy individuals included a large cluster of bihemispheric activation encompassing the sensory and motor cortices, and parietal lobule. Bilateral cerebellar activation and occipital lobule activation was also found in this network. Conversely, for the ST group, there was no significant difference between repeated and random sequence motor network activation at retention. Further highlighting functional connectivity variances, our second planned comparisons showed a significant difference between the HC and ST groups for motor network activation during repeated sequence performance. Specifically, the HC group exhibited significantly greater activation during repeated sequence performance compared to the ST group. This finding is likely related to whole-brain fMRI-CPCA results that suggest high inter-subject variability in the spatial distribution of neural activity during performance of the repeated sequence after stroke. Our secondary analysis constrained connectivity within the now-defined motor network to assess the level of functional connectivity specific to the ST group. At retention, the results of 162 the ROI-based fMRI-CPCA analysis demonstrated similar activation patterns within a subset of regions in the motor network, during repeated and random conditions. This subset predominately included larger clusters of functional integration within the left medial frontal gyrus and right cerebellar lobule, as well as the right insular cortex, occipital lobule, middle temporal gyrus, and left cerebellar culmen. The level of inter-regional functional connectivity was significantly correlated with performance during the continuous tracking task; individuals with higher network connectivity performed the task with less tracking error following five days of motor skill practice. Motor network This is the first study to use fMRI-CPCA to assess the functional networks associated with sequence-specific implicit motor learning after stroke. The motor network recruited during the performance of the repeated sequence in healthy individuals is consistent with previous work using other functional connectivity analysis methods (Coynel et al., 2010; Sun et al., 2007). The present results are unique because fMRI-CPCA allows for whole-brain analysis of functionally connected areas that are specifically related to the experimental task employed; and in the current study, this approach revealed a task-related motor network previously characterized as the M1-premotor-parietal-cerebellar circuit (Tamas Kincses et al., 2008). This network has been shown to demonstrate substantial inter– and intra–hemispheric functional connectivity, as spatial and motor representations are established during non-dominant hand motor sequence learning (Dayan & Cohen, 2011; Sun et al., 2007). The absolute activation level of brain regions within this network has been shown to decrease with increasing practice, which is believed to reflect improved efficiency in visuospatial processing, spatio-motor integration, and motor execution for the repeated sequence (Dayan & Cohen, 2011; Gobel, Parrish, & Reber, 2011; Tamas Kincses et 163 al., 2008). In the present study, a similar motor network was observed in a separate analysis during baseline performance. However, no significant difference between repeated and random conditions was observed in this network. Our findings from the retention test analysis revealed an increase in coordinated neural activity in healthy controls, but not stroke participants. Unlike traditional functional connectivity analysis methods, fMRI-CPCA extracts networks exclusively from variance in BOLD signal predictable from the hemodynamic response modelled from the task timing (Woodward et al., 2009). In the present study, there was an improvement in neural efficiency demonstrated by the expression of a motor network. This reflects greater synchronization within the M1-premotor-parietal-cerebellar circuit that was associated with more accurate tracking for the repeated condition at retention. Lesion disruption of neural networks Unlike healthy controls, individuals with stroke did not demonstrate a pattern of coordinated activity within the whole-brain motor network. This finding supports other work suggesting that a high degree of inter-participant variability exists during motor learning in the stroke-affected brain (Price & Friston, 2002; Westlake & Nagarajan, 2011). Different reorganization patterns may emerge within a group of individuals with stroke, potentially due to the diverse repertoire of disrupted connections remote of the lesion site (Wang et al., 2010), as well as a large range of discrepancies between participants’ motor abilities and task dose and difficulty. Resting-state functional connectivity within the motor network appears to become less optimized (i.e., greater randomization of connectivity), with greater time following the recovery from stroke (one week to one year post-stroke). Greater network randomization occurs from non-optimized outgrowth of new and potentially aberrant connections, which has been positively related to motor recovery after stroke (Wang et al., 2010). Following stroke, individual brains 164 may undergo varied reorganization strategies to create operable, yet random, functional networks to perform the CTT and implicitly acquire a motor sequence, by compensatory mechanisms. Furthermore, in the current study, individuals in the stroke group were in the chronic phase of stroke recovery, and thus it is possible that functional brain networks have shifted towards random topological patterns of brain activity (Wang et al., 2010). Subsequently, although spatial variability of functional networks increases after brain injury, our masked secondary analysis may have isolated regions of preserved structural integrity, resulting in commonalities in connectivity functionally important in implicit motor learning. We tested the relationship between the variability of network connectivity and motor learning in our secondary analysis by constraining (i.e., masking) the brain to assess the intraregional connectivity within the motor network in individuals with stroke. The spatial distribution of this analysis was limited to voxels within the motor network, thus reducing irrelevant activation and maximizing the variance of the BOLD signal fluctuations within the motor network. This method elucidated a more restricted functional network with a high level of intrinsic functional connectivity as compared to the whole-brain analysis in the stroke group. From our previously published fMRI data set (Meehan, Randhawa, et al., 2011), there were overlaps between the regions that comprised the univariate and multivariate compensatory motor networks in the current study, respectively, which included ipsilesional insular cortex and contralesional temporal gyrus activation. In our secondary analysis, the largest cluster of activation was within the left medial frontal gyrus (Brodmann’s area 6 [BA 6]). Meehan et al. (2011) found activation in the medial frontal gyrus after stroke significantly correlated with tracking error (i.e., RMSE) at retention (Meehan, Randhawa, et al., 2011). However, the specific region of activity within the medial frontal gyrus in the present study and Meehan et al. (2011) 165 studies did not overlap. This difference likely stems from methodological differences between the two analytical approaches. The secondary analysis from the Meehan et al. (2011) study relied on coordinates from peak activation derived from the healthy control group, whole-brain univariate analysis to determine and illustrate the relationship between level of activation of the medial frontal gyrus and behavioural performance at retention (Meehan, Randhawa, et al., 2011). Conversely, in the present study, our secondary analysis relied on multiple regions derived from the healthy control group, as well as whole-brain multivariate analysis to determine and illustrate the relationship between level of intraregional connectivity and behavioural performance. Findings from both univariate and multivariate analyses demonstrate the importance of the medial frontal gyrus during motor learning following stroke; however, sub-regional discrepancies exist between the coordinates associated with the level of peak activation and those associated with coordinated activity with other regions. Applying a multivariate analysis to the present data set allowed the BOLD signal in regions that followed similar temporal patterns of activation to emerge as a part of the task-dependent network. Multivariate analysis increases the sensitivity of signal detection, as regions with weaker signals that still contribute to a task-dependent network might be negated in whole-brain analysis due to voxel-wise statistical inference (Norman, Polyn, Detre, & Haxby, 2006). At retention, the stroke group demonstrated task-dependent variance of the BOLD response within a larger cluster of activation in the left medial frontal gyrus as well as the right cerebellar lobule, which was not observed in the whole-brain univariate analysis (Meehan, Randhawa, et al., 2011). Compared to connectivity within the motor network demonstrated by the HC group, a different pattern of lateralization of brain activity was observed in the ST group; contralesional cortical involvement of the motor cortex and subcortical involvement of the ipsilesional cerebellum were 166 noted during tracking performance at retention. The hemispheric laterality of connectivity observed within the motor network in the secondary analysis in the ST group, in comparison with the whole-brain analysis in the HC group, may be the result of disruption within the ipsilesional dentatothalamocortical and corticopontine tracts, which are known to connect the ipsilesional motor cortex with the contralesional cerebellar hemisphere (Middleton & Strick, 1994, 1997). People in our lab have previously studied the relationship between corticospinal tract (CST) integrity, as measured by WM tractography in the posterior limb of the internal capsule (PLIC), and motor skill learning in a cohort of individuals with chronic stroke affecting the middle cerebral artery using diffusion tensor imaging (DTI) (Borich et al., 2014). They showed that the degree of motor learning–related change in the performance of a complex visuomotor task is associated with post-training ipsilesional WM integrity, as measured by FA in the PLIC. In the present study, the contralesional cortical to ipsilesional cerebellum functional connectivity pattern observed in the compensatory motor network may indicate underlying structural damage of the ipsilesional CST tractography, because the internal capsule encompasses fibres of the dentatothalamocortical and corticopontine tracts (Axer & Keyserlingk, 2000). To determine if WM integrity influences GM functional connectivity and motor recovery, future studies should assess the underlying tractography of the motor network (dentatothalamocortical and corticopontine tracts), as well as the relationship between the underlying WM integrity and motor skill learning in individuals with chronic stroke. Individuals with stroke who demonstrated greater connectivity within the ROI-based fMRI-CPCA network (secondary analysis) tracked with less error during the repeated sequence condition at retention. This finding differs from previous literature, where it has been reported that greater activation of the ipsilesional, and reduced activation of the contralesional, 167 hemispheres predicts motor recovery following stroke (Carey, Abbott, Egan, Bernhardt, & Donnan, 2005). However, the functional network observed is highly dependent on the nature of the task, as challenging motor-related tasks have been shown to recruit different brain regions (Schaechter & Perdue, 2008). The CTT, employed in the present study, is a complex visuomotor task that involves a high level of motor dexterity and control, with which individuals manipulate a joystick using a specified range of movements and velocity. Our findings are in accordance with Schaechter and Perdue (2008), who concluded that increased activation in the contralesional cortical network was dependent on motor skill challenge in chronic stroke patients who were deemed to have good motor recovery based on their level of upper limb function. In the present study, contralesional motor and ipsilesional cerebellar connectivity may reflect the demands of the task, as well as alterations in the structural WM of the underlying ipsilesional PLIC. A limitation of the present study is the small sample size. The GROUP × SEQUENCE interaction effect for motor network activation at the retention test trended towards significance, with a moderate effect size; the interaction accounted for 21% variance in motor network activity. Therefore, the findings of this study must be interpreted with caution, and future research is needed to determine whether these motor network differences can be generalized to all individuals with subcortical stroke. The GROUP × SEQUENCE interaction effect with motor network activity may be practically meaningful, indicating a true disruption of activity in the lesion-affected hemisphere during implicit motor learning; however, difficulties attend in the interpretation of effect sizes in neuroimaging data (Fox et al., 2016), especially where these may be the result of variability in active brain regions across individuals. With an observed moderate effect size, however, we are assured that there are differences of motor network activation between our groups, yet, within the sample of healthy individuals and individuals with stroke, 168 certain individuals did notably differ from the group mean. With a greater sample size, those individuals in the stroke group may have been identified as characteristically different (i.e., low versus high UE-FM scores), further reducing the variance of the group. However, the focus of the present study was testing the fMRI-CPCA method, and comparing identified networks to prior studies to build upon past studies. An additional caution relates to the homogeneity of lesion location within the stroke group. While the similarity of stroke lesion locations within the right basal ganglia enables a group analysis, it does limit the generalizability of these data. Conclusion Findings from the present study support the existence of a functional motor network for healthy individuals and a variable compensatory motor network for individuals with chronic stroke that is active during delayed retention of a complex visuomotor task. Recruitment of this motor network was observed during a retention test for healthy individuals, but after stroke, we noted a high degree of inter-participant variability in network activity within the stroke-affected brain. An increase in coordinated neural activity within the primary motor (M1)-premotor-parietal-cerebellar circuit, observed for the healthy control group, may have been limited in the stroke group due to lesion-related changes in functional connectivity. Our secondary analysis of the motor network demonstrated greater connectivity of a compensatory network following stroke. Individuals with stroke demonstrated the capacity to implicitly learn a motor sequence; and functional connectivity within the compensatory motor network may be a neurophysiologic biomarker of implicit motor learning. Individuals with greater functional connectivity within this network had superior motor performance following five days of motor practice. The findings from the present study provide a baseline for comparison of an optimal efficiency motor network, setting the stage for further fMRI-CPCA motor learning studies. fMRI-CPCA is a task- 169 based multivariate approach that provides statistical information about the importance of functional brain networks to each condition and subject, allowing direct-group comparisons. Thus, future studies can evaluate individualized predictor weight values of the motor network in larger stroke groups to determine if subgroups have different levels of connectivity. To enhance the ecological validity of fMRI studies in rehabilitation settings, we recommend longer practice paradigms in general, with difficulty matched to individuals’ own motor abilities, to determine if a common network can be identified with an increased number of repetitions in individuals with chronic stroke. Thus, the present study provides insights into novel, task-based fMRI methods as a way to describe and evaluate the (re)organization of functional networks following stroke. 170 Chapter 5: White matter biomarkers associated with motor change in individuals with stroke: A continuous theta burst stimulation study Introduction Recovery of movement after stroke is often incomplete (Stewart & Cramer, 2013). Incomplete recovery from stroke has led to interest in adjunct interventions that may be paired with standard therapy to magnify the effects of rehabilitation. One such intervention is non-invasive brain stimulation (Brodie, Meehan, et al., 2014; Carey et al., 2014; Meehan, Dao, et al., 2011). Repetitive transcranial magnetic stimulation (rTMS) is a form of non-invasive brain stimulation that, depending on stimulation parameters, enhances or suppresses local cortical or corticospinal excitability (Pascual-Leone et al., 1998). The effects of rTMS on cortical excitability tend to outlast the stimulation period for approximately 60 minutes or more (Huang & Rothwell, 2004; Huang et al., 2009). During this period of modulated activity, delivering motor skill practice theoretically may enhance motor learning, in part due to a less inhibited brain state that is more labile to neuroplastic change (Brodie, Meehan, et al., 2014; Cassidy et al., 2015; Meehan, Dao, et al., 2011). Theta burst stimulation (TBS) is a patterned form of rTMS. When TBS is presented continuously for 30 seconds or more (continuous TBS [cTBS]), it tends to suppresses excitability of the primary motor cortex (M1) (Huang et al., 2005). In healthy individuals, cTBS, when paired with behavioural training has been used to alter normal functioning of a specific area of the brain to determine the area’s role in the behaviour of interest (Clerget, Badets, Duque, & Olivier, 2011; 171 Iezzi et al., 2010; Li Voti et al., 2014). Specifically, when paired with behavioural training, cTBS applied over the: (1) ipsilateral M1 impaired practice and retention performance of a simple finger abduction motor task (Iezzi et al., 2010); (2) Broca’s area (Brodmann’s area 44 [BA44]) reduced performance of key-press motor sequence task at a retention test (Clerget et al., 2011); and (3) lateral cerebellum impaired the performance of index finger abductions, reaching-to-grasp and reaching-to-point movements at a retention test (Li Voti et al., 2014). Theoretically, in individuals with stroke cTBS over the contralesional hemisphere influences the excitability of the ipsilesional hemisphere, thus decreasing interhemispheric inhibition (Meehan, Dao, et al., 2011; Schambra, Sawaki, & Cohen, 2003). This is particularly relevant post-stroke as there is strong evidence for an imbalance in interhemispheric signals such that the contralesional cortex may exert increased inhibition on the ipsilesional cortex (Murase, Duque, Mazzocchio, & Cohen, 2004). When paired with motor skill practice of a tracking task, cTBS applied over the contralesional hemisphere can facilitate motor learning in individuals with chronic stroke, perhaps due to a release of this interhemispheric inhibition (Meehan, Dao, et al., 2011). Yet there is a large degree of individual variability in response to repetitive brain stimulation (Brodie, Borich, et al., 2014; Carey et al., 2014). As a result, recent work has focused on the investigation of biomarkers of inter-individual variability in response to non-invasive brain stimulation, directed towards identifying characteristics of “responders” versus “non-responders” that may be targeted to enhance the efficacy of future interventions (Brodie, Borich, et al., 2014; Carey et al., 2014). The primary motor and somatosensory cortices (M1 and S1, respectively) are two brain areas that support motor recovery (Carey et al., 2006). While hyperexcitability from both the contralesional M1 (M1c) and S1 (S1c) correlates with reduced post-stroke motor function 172 (Calautti et al., 2007), few studies have extended non-invasive brain stimulation sites beyond contralesional M1 (Brodie, Meehan, et al., 2014; Meehan, Dao, et al., 2011). Meehan et al. (2011) compared the effects of cTBS over M1c versus S1c followed by motor practice, and found comparable improvements in movement time during motor skill practice, regardless of stimulation site; both M1c and S1c stimulation groups showed larger amounts of motor learning-related change compared to sham stimulation paired with motor skill practice (Meehan, Dao, et al., 2011). In the present study, we first considered the effects of pairing cTBS over M1c or S1c with motor skill practice. We discovered a high degree of variability in individual response to this intervention. Therefore we sought to define what characterized a “responder” to cTBS over the M1c or S1c. After stroke, knowledge of the residual functional architecture of the brain may allow more accurate identification of biomarkers of an individual’s capacity to change motor behaviour (Ward, 2017). Differences in residual brain structure after stroke could be a source of variability in individual responses to repetitive brain stimulation when it is paired with motor skill practice (Brodie, Borich, et al., 2014; Carey et al., 2014). Immediately following stroke, there is a disruption to the underlying white matter (WM) structures that lead to neural pathway adaptation (Wang et al., 2006). Assessment of WM structures may be helpful to identify individuals’ capacity for motor skill learning post-stroke (and hence, motor recovery) (Ward, 2015a). To date, most research has focused on regional WM evaluation of individual tracts (Brodie, Borich, et al., 2014; Carey et al., 2014); however, recovery from stroke involves a network of bihemispheric pathways that extend between the contralesional and ipsilesional M1, secondary motor areas, and ipsilesional cerebellum (Ward, 2015a). In the current work, we hypothesized that characterizing a specialized WM motor network associated with motor learning would explain response to cTBS 173 paired with motor skill practice in individuals with chronic stroke. We identified this WM network using a functional connectivity analysis to quantify activation associated with motor learning in healthy individuals (Chapter 4; Wadden et al., 2015). We employed an fMRI-guided tractography method to constrain WM connections associated with our previously identified gray matter (GM) motor learning network (Chapter 4; Wadden et al., 2015). We named the resultant network the “constrained motor connectome” (CMC) and hypothesized that individual capacity for motor learning–related change following cTBS paired with motor skill practice would relate to its residual integrity (Chapter 4; Wadden et al., 2015). In the field of stroke rehabilitation research, the identification of the factors that characterize responders versus non-responders to an intervention is of growing interest (Auriat et al., 2015; Carey et al., 2014). Typically, the basis of assessing the response to rTMS is characterized in terms of M1 excitability. In the present study, we were interested in assessing behavioural response following cTBS paired with motor skill practice. In supplement to neurophysiological outcomes (e.g., motor evoked potentials [MEPs]), behavioural response provides meaningful data about the effectiveness of this intervention on motor learning-related change (Rothwell, 2016), and can be a more pragmatic anchor point to assess the success of an adjunct intervention from a clinical perspective. In behavioural intervention studies, responders are identified as individuals who demonstrate an improvement in function and/or reduction in impairment (Lang et al., 2015). Following an intervention, some individuals may reach a plateau in performance improvement, while others demonstrate additional improvement beyond this plateau, when given the opportunity for extended practice (Hardwick, Rajan, Bastian, Krakauer, & Celnik, 2016; Wadden et al., 2017). In the present study, we characterized individuals as 174 responders or non-responders based on positive or negative changes in motor performance derived from individual performance curves, respectively. Thus, the overarching objective of the current study was to establish if a new brain-based biomarker, termed the CMC, could be used to determine the capacity to respond to cTBS over M1c or S1c paired with motor skill practice. For the current study, we were interested in the effects of cTBS over M1c and S1c on a composite score of serial targeting task (STT) performance, known as the response time total (RTT; time to initiate movement + movement time). First, we evaluated the effects of stimulation group (M1c, S1c, and sham) on RTT at baseline, across 5 days of practice and at a delayed retention test. Next, we were specifically interested in the effect of stimulation within our motor practice responders group. After the identification of motor practice responders and non-responders for repeated sequence performance, we hypothesized improvements in motor performance would be similar for the M1c and S1c stimulation within the motor practice responder group. Therefore, for our next set of analyses, M1c and S1c groups were combined into a contralesional sensorimotor (SMc; M1c and S1c) group, comprised of individuals who received active cTBS over M1c or S1c (SMc-cTBS group). We hypothesized that active cTBS over SMc paired with motor skill practice would facilitate greater improvements in motor sequence learning compared to sham stimulation (Sham group) in the motor practice responder group. For final analyses, we hypothesized that the integrity of a WM network of sensorimotor regions defined by our past work (Wadden et al., 2015), the CMC, would explain the variability of response in individuals receiving cTBS over the SMc paired with motor skill practice (SMc-cTBS group). 175 Methods Participants Twenty-eight individuals (mean [M] age = 63.0; standard deviation [SD] = 12.88 years; 7 females) who demonstrated chronic stroke-related unilateral upper limb deficits were recruited. All experimental sessions were completed at the University of British Columbia (UBC). Ethical approval was granted from Clinical Research Ethics Board of UBC. All participants provided written informed consent in accordance with the Declaration of Helsinki. This study was a subcomponent of a planned randomized controlled trial (RCT) registered with clinicaltrials.gov (NCT01371409). For this subcomponent study, individuals from the larger study RCT (n = 35) were selected based on the collection of a diffusion-weighted imaging (DWI) magnetic resonance (MR) scan (n = 28). Inclusion criteria were: 1) chronic cortical or subcortical stroke (≥ 6 months ago), 2) upper extremity Fugl-Meyer assessment (UE-FM) motor impairment score greater than or equal to 15, and 3) a Montreal Cognitive Assessment (MoCA) score greater than or equal to 20. Exclusion criteria were: 1) history of seizure/epilepsy, head trauma, a major psychiatric diagnosis, neurodegenerative disorder, or substance abuse, 2) taking any gamma-aminobutyric acid (GABA)-ergic, N-methyl-D-aspartate (NMDA)-receptor antagonist, or other drug known influence known neural receptors that facilitate neuroplasticity, or 3) contraindication to transcranial magnetic stimulation (TMS) or magnetic resonance imaging (MRI). Experimental design Participants were initially assigned to one of three stimulation groups: 1) M1c TBS (contralesional primary motor cortex), 2) S1c TBS (contralesional primary somatosensory cortex), or 3) Sham cTBS over contralesional M1 (Figure 34a and b). Each individual was 176 pseudo-randomly assigned to a stimulation group based on UE-FM score to ensure even distribution of impairments across groups (M1c, n = 9; S1c, n = 11; Sham, n = 8). The experimental protocol consisted of seven sessions over 14 days (Figure 34). In session 1, participants underwent MR scanning, clinical assessments of motor function and impairments (Wolf Motor Function Test [WMFT], UE-FM, respectively), and baseline performance on the experimental motor learning task (STT). During sessions 2 to 6, participants received cTBS over the contralesional hemisphere according to stimulation group (M1c, S1c, Sham), immediately before STT practice. In session 7, an STT retention test was employed to assess motor learning, with no cTBS. Figure 34: Experiment design and apparatus. a. Experimental design. b. Serial targeting task (STT) apparatus and target locations. MR, magnetic resonance; cTBS, continuous theta burst stimulation. In session 1, to assess upper extremity motor function, the WMFT was performed by a licensed physical therapist. Mean performance time to complete 15 items of the WMFT with the paretic and non-paretic arms was determined. Participants’ WMFT rate was calculated to Session 1• MR Imaging• Clinical Assessment • STT Initial TestPractice Sessions 2-6• cTBS• STT PracticeSession 7• STT Retention Testa.b. 177 determine how many times an individual could complete the task continuously for 60 seconds (60 seconds divided by the mean performance time); if an individual could not perform the task in 120 seconds a score of 0 was given (Hodics et al., 2012). In addition, individuals’ physical impairment level was assessed via the UE-FM (range for UE-FM motor test is 0-66; lower scores denote less hemiparetic arm function). Participant characteristics are shown in Table 8. 178 Table 8: Participant characteristics Participant Stimulation group (M1c = 1; S1c = 2; Sham = 3) Lesion location (C = cortical; SC = subcortical) MoCA UE-FMa PSDb Age ST1 1 C 28.0 55.0 270.0 59.0 ST2 1 SC 26.0 63.0 37.0 50.0 ST3 1 SC 30.0 62.0 67.0 65.0 ST4 1 SC 26.0 56.0 94.0 64.0 ST5 1 SC 26.0 59.0 12.0 82.0 ST6 1 SC 23.0 41.0 196.0 46.0 ST7 1 SC 26.0 62.0 20.0 62.0 ST8 1 SC 26.0 16.0 22.0 57.0 ST9 1 SC 25.0 30.0 160.0 57.0 ST10 2 SC 25.0 59.0 82.0 67.0 ST11 2 C 27.0 60.0 142.0 73.0 ST12 2 SC 27.0 56.0 83.0 71.0 ST13 2 C 25.0 60.0 35.0 85.0 ST14 2 SC 29.0 62.0 81.0 76.0 ST15 2 SC 29.0 54.0 23.0 60.0 ST16 2 SC 24.0 7.0 94.0 57.0 ST17 2 C 21.0 62.0 24.0 55.0 ST18 2 SC 23.0 11.0 36.0 93.0 ST19 2 C 29.0 18.0 33.0 33.0 ST20 2 C 27.0 62.0 31.0 69.0 179 C = cortical; M1c = contralesional primary motor cortex; PSD = post-stroke duration; S1c = contralesional primary somatosensory cortex; SC = subcortical; UE-FM = upper extremity Fugl-Meyer assessment. Serial targeting task Participants performed the STT seated at a computer desk. The paretic hand (pronated) was used to grasp the computer mouse (housed in a custom frame) to control the movements of an on-screen cursor (Wheel Mouse Optical, Microsoft Corporation, Redmond, WA, USA). Individuals were instructed to move the cursor as quickly and accurately as possible to a series of sequentially appearing targets. The location of the participant’s hand in space was occluded by an opaque surface affixed above the hand. Embedded within the series of targets was a repeated six-element sequence that was flanked by a six-element random sequence. The participant performed four blocks of the STT during each practice session (2 to 6). Each block was comprised of nine random sequences that each contained six movements each, and eight repeated sequences that also contained six movements each. Participants performed one block of the STT in sessions 1 (baseline) and 7 (retention) to index motor learning (Meehan, Dao, et al., 2011). Participant Stimulation group (M1c = 1; S1c = 2; Sham = 3) Lesion location (C = cortical; SC = subcortical) MOCA UE-FMa PSDb Age ST21 3 SC 28.0 23.0 41.0 63.0 ST22 3 SC 30.0 35.0 27.0 56.0 ST23 3 SC 28.0 58.0 20.0 71.0 ST24 3 SC 26.0 49.0 155.0 76.0 ST25 3 C 28.0 57.0 15.0 69.0 ST26 3 SC 24.0 61.0 18.0 79.0 ST27 3 SC 26.0 29.0 47.0 51.0 ST28 3 SC 21.0 57.0 27.0 83.0 180 Exponential curve fitting The primary dependent measure was response time total (RTT). Participants’ RTT was calculated as the time to initiate movement plus movement time. The sum RTT for all six movements within the repeated and random sequences was calculated separately. The RTT for repeated and random sequences in each block across the seven sessions (baseline, Session 1; practice, Sessions 2 to 6; and retention, Session 7) of task performance for each participant were subsequently fit to separate exponential functions using the following equation (Brown & Heathcote, 2003; Wadden et al., 2017) (equation 1): E(RTTN) = A +Be-αN E(RTTN) is the expected value of RTT on practice trial N. A is the expected values of RTT after practice has been completed (asymptote parameter). B is the expected change in RTT from sessions 1 to 7 (change score parameter). Alpha (α) is the exponential motor skill acquisition rate parameter (Heathcote et al., 2000). Our primary outcome measure was B, which reflects an individual’s capacity for motor change. A custom MATLAB (Version R2013b, The Mathworks Inc., Natick, MA, USA) script was used for all analyses. Motor practice responder versus non-responder: The B score was used to differentiate between responders and non-responders. A positive B score reflects an individual’s capacity for motor learning based on the performance plateau prediction, while a negative B score indicates an absence of motor learning related change (Yamashita, Kawato, & Imamizu, 2015). 181 Transcranial magnetic stimulation procedures All participants were screened for contraindications to rTMS (Rossi, Hallett, Rossini, & Pascual-Leone, 2011). All TMS was performed using MagStim Rapid2 and Plus1 magnetic stimulators and a 70-mm diameter air-cooled figure-of-eight coil (Magstim Co. Ltd., Whitland, Carmarthenshire, UK) on sessions 2 to 6. During all TMS procedures, participants were seated in a reclining chair with their hands placed in a relaxed position (elbow at 180 degrees flexion, forearm pronated). Coil positioning was continuously monitored using a BrainsightTM neuronavigation system, which displayed each individual’s T1-weighted MRI. The participants’ motor hotspot (M1c) and resting motor threshold (RMT) were determined as the site that evoked a measurable MEP greater than or equal to 50µV peak-to-peak for 5 out of ten trials in the extensor carpi radialis muscle (ECR), at the lowest stimulus intensity at rest. After identifying the ECR motor hotspot and RMT, active motor threshold (AMT) was determined as the lowest intensity to evoke a 200 µV MEP in at least five out of 10 consecutive TMS stimuli (Rossini et al., 1994), while participants maintained a 20% maximal isometric voluntary hand grip contraction. Hand grip contraction was monitored online using a handgrip dynamometer that translated force in N to a voltage signal (National Instruments Corporation, Austin, Texas, USA), which was displayed on a computer monitor. Thereafter, cTBS was then applied with the participant at rest in the theta burst pattern of stimulation: three stimuli delivered at 50 Hz, grouped and delivered at 5 Hz, in continuous blocks for a total of 600 stimuli over 40 seconds (Huang et al., 2005), at an intensity of 80% AMT. Sham stimulation was performed with a dedicated coil that looked and sounded like active stimulation but did not induce any current. cTBS was delivered over the determined M1c “hotspot” or two cm posterior to this location for S1c, consistent with previous work (Brodie, Meehan, et al., 2014; Meehan, Dao, et al., 2011), 182 while sham was delivered over M1c hotspot. Both M1c (including Sham) and S1c were located and real-time position monitored using the BrainsightTM neuronavigation system. Directly following cTBS completion on sessions 2 to 6, participants completed motor skill practice of the STT. Magnetic resonance imaging protocol MR acquisition was conducted at the UBC MRI Research Centre on a Philips Achieva 3.0 T whole-body MRI scanner (Phillips Healthcare, Andover, MA, USA) using an eight-channel sensitivity encoding head coil (SENSE factor = 2.4) and parallel imaging. Anatomical scan: a high-resolution T1-weighted anatomical scan (TR = 7.47 ms, TE = 3.65 ms, flip angle Ɵ = 6̊, FOV = 256 × 256 mm, 160 slices, 1 mm3 isotropic voxel) was collected. Diffusion-weighted magnetic resonance imaging (DW-MRI): one high-angular resolution diffusion imaging (HARDI) scan was performed using a single-shot echo-planar imaging (EPI) sequence (TR = 7096 ms, TE = 60 ms, FOV = 224 × 224 mm, 70 slices, voxel dimensions = 2.2 × 2.2 × 2.2 mm3). Diffusion weighting was applied across 60 independent non-collinear orientations (b = 700 s/mm2), along with five un-weighted images (b = 0 s/mm2). DWIs were corrected for motion and distortion using the software package ExploreDTI v4.2.2 (www.exploredti.com; Leemans and Jones, 2009). During motion and distortion correction, signal intensity was modulated and the b-matrix was rotated (Leemans & Jones, 2009). Depending on the WM pathways of interest, two commonly used tractography approaches were employed based on their advantages and disadvantages: constrained spherical deconvolution (CSD) and diffusion tensor imaging (DTI) (Nimsky, Bauer, & Carl, 2016). A recent review by Nimsky et al. (2016) recommended to select the tractography method, or 183 combination of methods, most appropriate for the specific objective. When the objective of the tractography is to reduce the likelihood of missing prominent pathways, as was the case with our study, CSD was chosen as the best method, due to its relatively high-sensitivity to fibre orientations. CSD can also identify a greater number of connections in an area with multiple tract orientations; however, this can also increase the number of false positives (Thomas et al., 2014). Therefore, we used CSD to analyze known pathways of the corticospinal tract (CST), to ensure the inclusion of all tracts where we could manually control their inclusion and exclusion (see Methods, below; use of “SEED” and “AND” Boolean commands), to reduce false positives. DTI is recommended if the objective is to reduce the possibility of identifying spurious pathways where long-range axonal connectivity is of interest (Nimsky et al., 2016). Therefore, for the constrained motor connectome (CMC) analysis, which is a broader network than the CST, comprised of long-range pathways, we used the more conservative DTI tractography because we had no a priori guidelines to identify tracts within this motor network. CSD and DTI tractography were performed on DW images using the ExploreDTI software package. Corticospinal tract (CST): DW images remained in native space. At each voxel of the CST, CSD-based deterministic whole-brain fibre tractography was initiated using the following parameters: seedpoint resolution of 2 mm3, 0.2 mm step size, maximum turning angle greater than 40°, and fibre length range of 50 to 500 mm (Reijmer et al., 2012). The first cross-sectional region of interest (ROI) for the CST was delineated bilaterally (non-lesioned [NL]-CST and lesioned [L]-CST) in the axial plane, as in previous work (Hong, Son, & Jang, 2010). First, a “SEED” ROI was constructed around the posterior limb of the internal capsule (PLIC) at the level of the anterior commissure (Borich, Wadden, & Boyd, 2012). Second, a logical ‘‘AND’’ ROI was constructed around the CST at the level of the mid-pons (Kwon et al., 2011). The 184 “AND” function constrained the reconstruction to fibres passing through both the “SEED” and “AND” ROI (Figure 35a). Constrained Motor Connectome (CMC): Using the CMC, the motor network mask was in Montreal Neurological Institute (MNI) standard space; all imaging data were converted first to MNI space. Standard (MNI) and native space transformations have been compared in a prior study and no differences were observed between fractional anisotropy (FA) values of the CST in individuals with stroke (Park, Kou, Boudrias, Playford, & Ward, 2013). At each voxel of the CMC, DTI-based deterministic whole-brain fibre tractography was initiated using the following parameters: seedpoint resolution of 2 mm3, FA threshold 0.2, 0.2 mm step size, maximum turning angle greater than 40 ̊, and fibre length range of 50 to 500 mm. FA values, the most commonly reported measure of white matter microstructural integrity after stroke, were extracted from reconstructed tracts and used for statistical analyses (Jang, 2010). FA is a quantitative, unit-less measure of diffusion behaviour of water in the brain; a value of zero indicates diffusion of water is isotropic and a value of one specifies a preferred direction of diffusion along one axis (Mukherjee, Berman, Chung, Hess, & Henry, 2008). We used a previously defined functional motor network associated with motor learning in healthy individuals. Selected binary masks of cortical GM clusters of activation were used as ROIs for WM tractography. Each of the four clusters encompassed multiple brain regions. Cluster one included the S1, M1, precentral gyrus, bilaterally, and the right intraparietal, superior parietal, and inferior parietal cortices. Cluster two included lobule V, VI, VIIIa and VIIIb of the cerebellum, bilaterally. Cluster three included right lobule VI and VIIa Crus of the cerebellum. Cluster four included left intraparietal and superior parietal cortices (Table 9 and Figure 36). The GM cortical clusters from the functional motor network were derived from a whole-brain 185 connectivity analysis in MNI space, allowing for the clusters of activation from the fMRI connectivity analysis to be overlaid on the DW images converted to MNI space. The functional motor network ROIs were used to isolate the underlying WM fibre tracts of the CMC (Figure 35b). In our previous fMRI study (Wadden et al., 2015), all individuals used the left hand during performance of an upper extremity motor task during fMRI. Thus, the functional motor network comprising the binary mask used to extract the CMC data was specific to left-hand performance, while the motor task in the current study was performed outside the scanner with the paretic-hand. Therefore, to ensure the functional motor activity typically associated with unilateral task performance was similar across individuals, all brains with a left lesion were flipped during preprocessing (Schaapsmeerders et al., 2016; Takenobu et al., 2014). Figure 35a and b: Diffusion tensor imaging (DTI). Fractional anisotropy (FA) maps were created for individuals, followed by whole-brain tractography. This is an example of a single subject with a left hemispheric lesion. a. Regions of interest (ROIs) for the non-lesioned (NL) and lesioned (L) corticospinal tracts (CST) were manually drawn on the native diffusion-weighted (DW) image, followed by tractography. FA values were derived from the constrained motor connectome (CMC), as well as the NL- and L-CST. b. The functional motor network mask (gray ROIs) was extracted and overlaid on the DW Montreal Neurological Institute standard space-alined (MNI) image, followed by tractography of the CMC (posterior and anterior views). Posterior View Anterior ViewCMCCSTL -CST NL -CST b.a. 186 Table 9: Constrained motor connectome (CMC) CMC MNI coordinate (X Y Z) mm2 Right postcentral gyrus 36 –28 70 4,171 Left cerebellum (V) –16 –52 –22 704 Left superior parietal lobule –32 –56 62 124 Right cerebellum (VI) 26 –60 –26 114 The areas of the motor learning network were used as regions of interest, overlaid on diffusion-weighted images prior to tractography. MNI, Montreal Neurological Institute standard space. Figure 36: Constrained motor connectome (CMC). The motor network mask (represented in red) was overlaid on the diffusion-weighted image to create the CMC. Statistical analyses Four main investigative steps were performed to determine WM biomarkers of cTBS over the contralesional hemisphere paired with motor skill practice in individuals with chronic stroke. These steps included: (1) determining the effect of stimulation group (M1c, S1c, Sham) on motor sequence learning (baseline, Session 1; practice, Sessions 2 to 6; and retention, Session 7); (2) separating individuals into motor practice responders and non-responders. For motor skill RLRL 187 responders, first, determining the effect of M1c and S1c stimulation on improvements in motor performance. Second, determining the effect of active stimulation (SMc-cTBS group; M1c and S1c groups combined) compared to sham stimulation (Sham) on improvements in motor performance on motor practice responders; (3) assessing differences in demographic and clinical measures between motor practice responders and non-responders for SMc-cTBS group; and (4) testing for differences in WM biomarkers between motor practice responders and non-responders for the SMc-cTBS group. Differences in motor sequence learning between cTBS groups in individuals with stroke during practice and retention Baseline performance: Session 1 (baseline) motor performance on the random and repeated sequences was evaluated with a two-factor GROUP (M1c, S1c, Sham; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model analysis of variance (ANOVA) with mean RTT as the dependent variable. Practice performance: Performance of the repeated and random sequences during cTBS paired with motor skill practice (Sessions 2 to 6) was examined using a three-factor GROUP (M1c, S1c, Sham; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) × SESSION (Sessions 2 to 6; within-subjects factor) mixed-model ANOVA with mean RTT as the dependent variable. There were two missing data points for practice days for two participants in the Sham group (n = 6). Retention performance: To assess motor sequence learning at retention (session 7), a two-factor GROUP (M1c, S1c, Sham; between-subjects factor) × SEQUENCE (repeated, random; within-subjects factor) mixed-model ANOVA was performed with mean RTT as the dependent variable. 188 Effect of cTBS on motor sequence learning for motor practice responders Motor practice responders were identified as individuals who demonstrated a positive B score (B > 0) for repeated sequences, and therefore an improvement in performance. Motor practice non-responders were identified as individuals who demonstrated a negative B score (B < 0), and therefore an absence of motor learning–related change (see Figure 37 for subject-specific examples). We were specifically interested in testing the effect of stimulation (M1c, S1c, Sham) on improvements in performance for the motor practice responder group. From a previous study conducted in our laboratory, Meehan et al. (2011) showed similar effects of cTBS over M1c and to S1c paired with motor skill practice of the serial targeting task (STT) on time to initiate movement and movement time. Based on the findings of Meehan et al. (2011) findings and our hypothesis, that in the present study there would be similarities between the effect of cTBS over M1c and S1c on RTT, we performed a planned independent t-test on the B score between M1c and S1c groups for motor practice responders. Next, based on previous findings, we had an a priori hypothesis that groups receiving active inhibitory stimulation over the contralesional hemisphere (M1c, S1c groups) would demonstrate greater improvements in motor performance compared to individuals receiving sham stimulation (Sham group) (Boggio et al., 2006; Fregni et al., 2006; Grefkes et al., 2010; Meehan, Dao, et al., 2011; Takeuchi et al., 2005; Tretriluxana, Kantak, Tretriluxana, Wu, & Fisher, 2013). To evaluate the effects of receiving contralesional stimulation, M1c and S1c groups were combined into a contralesional sensorimotor (SMc-cTBS group; M1c and S1c) group. Combining stimulation groups was supported by previous findings that demonstrated strong connections between M1-S1 cortices during non-invasive brain stimulation (Enomoto et al., 2001). To evaluate the effects of receiving cTBS over SMc compared to sham stimulation, 189 based on our hypothesis, we performed a planned one-tailed independent t-test on B score between SMc-cTBS and Sham groups for motor practice responders. Figure 37a, b, c, and d: Active continuous theta burst stimulation (cTBS) motor practice responders and non-responders. The top panel (a, b) represents motor practice responders (positive B score) following contralesional cTBS, delivered over contralesional primary somatosensory cortex (S1c; B score = 14.27; Figure 37a) or primary motor cortex (M1c; B score = 12.30; Figure 37b), paired with motor skill practice. The bottom panel (c, d) represents motor practice non-responders (negative B score) following contralesional cTBS, delivered over contralesional primary somatosensory cortex (S1c; B Score = –4.35; Figure 37c) or primary motor cortex (M1c; B score = –1.76; Figure 37d) paired with motor skill practice. 05101520253035401 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151RTT (s)TrialsS1c – Subject S305101520253035401 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151RTT(s)S1c – Subject S10 05101520253035401 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171M1c – Subject M305101520253035401 11 21 31 41 51 61 71 81 91 101111121131141151161171TrialsM1c – Subject M5a.c. d.b. 190 Clinical measures for SMc-cTBS motor practice responders and non-responders: To investigate the motor practice responder profile of cTBS to the sensorimotor (SMc-cTBS group; M1c and S1c) region, we conducted independent group t-tests to assess differences in demographic and clinical characteristics between motor practice responders and non-responders identified from the SMc-cTBS groups only. Demographic and clinical dependent variables included age, PSD, UE-FM score, and paretic WMFT rate. Fisher’s exact test was used to assess differences between motor practice responders and non-responders in stroke location (C; SC) for these individuals (Table 10; see Figure 38 for stroke locations). Table 10: Motor practice responder descriptive profile Table shows participant characteristics for motor practice responders and non-responders in the SMc-cTBS group (M1c and S1c combined). C = cortical; M1c = contralesional primary motor cortex; PSD = post-stroke duration; S1c = contralesional primary somatosenory cortex; SC = subcortical; UE-FM = upper-extremity Fugl Meyer assessment; WMFT = Wolf Motor Function Test. Group Stim Group Age (yrs)Stroke Location(C, SC) PSD(months)UEFMPareticWMFT RateMean SD Mean SD Mean SD Mean SDResponders n = 12M1c = 5 S1c = 763.8 16.06 C = 5SC = 786.0 78.86 46.8 20.08 38.7 17.91 Nonrespondersn = 8M1c = 4 S1c = 464.4 10.88 C = 2SC = 663.6 50.44 49.1 20.11 42.5 21.14 191 Figure 38: Lesion figure for M1c and S1c (contralesional sensorimotor [SMc])-cTBS motor practice responders versus non-responders. WM tractography for SMc-cTBS motor practice responders and non-responders: A multivariate analysis of variance (MANOVA) was used to assess differences in WM-FA from the CMC, and NL– and L-CST between motor practice responders and non-responders within the SMc-cTBS group (S1c, M1c). The dependent variables for the MANOVA were FA values for each ROI (CMC, and NL, and L-CST). Post hoc univariate ANOVAs were performed on significant (p ≤ 0.05) MANOVAs. In the event of a violation of sphericity (significant Mauchly’s test, p ≤ 0.05), the Greenhouse-Geisser correction was applied. Levene’s test for equality of variances was used to SMc-cTBS active “Nonresponders” SMc-cTBS active “Responders” 192 test for homogeneity of variance, and degrees of freedom were adjusted when the test was significant (p ≤ 0.05). The 95% confidence intervals (CIs) of the mean difference (MD) were used to describe the effect of stimulation on improvements in motor performance (B score). Effect sizes were reported as partial eta-squared (ηρ²) where 0.01 is considered a relatively small effect, 0.06 moderate and more than 0.14, a large effect (Gray & Kinnear, 2012). Significance level for all statistical tests was set at p ≤ 0.05, and post hoc tests, Bonferroni corrected for multiple comparisons, were conducted when appropriate. Data are presented in the text as M plus or minus SD or standard error (SE). All statistical procedures were conducted using SPSS software (version 21.0, SPSS Inc., Chicago, IL, USA). Results Repeated versus random sequence performance and learning between cTBS stimulation groups Baseline performance: During initial STT performance (Session 1), the random and repeated sequences trended towards, but were not, statistically different (F(1,25) = 4.09, p = 0.054, ηρ² = 0.141, large effect size; Figure 39a). There was no baseline difference in performance level between groups (M1c, S1c, Sham) as shown by the lack of significant main effect of GROUP (M1c, S1c, Sham) (F(2,25) = 0.48, p = 0.63, ηρ² = 0.037, small effect size). Additionally, there was no significant GROUP × SEQUENCE interaction in Session 1 (F(2,25) = 0.084, p = 0.92, ηρ² = 0.007, small effect size). Practice performance: All groups (M1c, S1c, Sham) demonstrated improved motor performance on the STT, evidenced by an observed decrease in RTT across sessions 2 to 6 and a significant main effect of SESSION (F(2.5,58.7) = 4.51, p = 0.009, ηρ² = 0.16, large effect size; Figure 39b). Mauchly’s test indicated that the assumption of sphericity had been violated (χ2(9) = 193 27.63, p = 0.001); therefore, degrees of freedom were corrected using Greenhouse-Geisser estimate of sphericity (ε = 0.64) for the main effect of SESSION. In addition, individuals showed superior repeated (M = 13.88, SE = 0.90) compared to random sequence (M = 15.59, SE = 0.99) performance across practice sessions, as revealed by the significant main effect of SEQUENCE (F(1,23) = 19.63, p = 0.0019, ηρ² = 0.46, large effect size). However, there was no main effect of GROUP (M1c, S1c, Sham) (F(2,23) = 0.066, p = 0.94, ηρ² = 0.006, small effect size), no significant interaction for SEQUENCE × SESSION (F(4,92) = 0.37, p = 0.83, ηρ² = 0.016, small effect size), or for GROUP × SEQUENCE (F(2,23) = 0.29, p = 0.75, ηρ² = 0.025, small effect size). Retention test performance: Motor learning–related change was shown by a main effect of SEQUENCE that confirmed all groups were faster for repeated (M = 12.34, SE = 0.77) compared to random sequence (M = 14.20, SE = 0.86) during retention performance (F(1,25) = 29.94, p < 0.001, ηρ² = 0.55, large effect size; Figure 39c). However, the main effect of GROUP (M1c, S1c, Sham) (F(2,25) = 0.38, p = 0.68, ηρ² = 0.030, small effect size), and the GROUP × SEQUENCE interaction (F(2,25) = 0.64, p = 0.53, ηρ² = 0.049, small effect size) were not significant. 194 Figure 39a, b, and c: Mean response time total (RTT) for repeated and random sequences. Stimulation groups (contralesional primary motor cortex [M1c], contralesional primary somatosensory cortex [S1c], and Sham) demonstrated similar performances for repeated and random sequences on Session 1 (baseline), 2 to 6, and 7 (retention). a. All groups demonstrated initial faster RTTs for repeated compared to random sequence performance. b. Collapsed across groups, all individuals demonstrated faster performance for repeated compared to random sequences across the five days of practice: F(1,25) = 29.94, p < 0.001. c. All groups demonstrated faster RTTs for repeated compared to random sequence during retention performance: F(1,23) = 19.63, p < 0.001. Error bars are the standard deviation (SD) of the mean. Motor practice responders: Overall, there were 17 motor practice responders and 11 non-responders, as indicated by a positive B score for responders and a negative B score for non-responders (M1c: 5 responders, 4 non-responders; S1c: 7 responders, 4 non-responders; sham: 5 responders, 3 non-responders) (see Figure 40 for normalized B score; normalization factor of A + B; A = asymptote value; B = change score). For motor practice responders, the first planned 051015202530M1c S1c ShamResponse Total Time (s)GroupsBaselineRandomRepeated0510152025302 3 4 5 6Response Total Time (s)SessionsPractice SessionsRandomRepeated051015202530M1c S1c ShamResponse Total Time (s)GroupsRetentionRandomRepeateda. b.c. 195 comparison demonstrated no significant difference for B score between M1c (M = 5.87, SD = 5.154) and S1c (M = 5.66, SD = 4.922) groups for the performance of the repeated sequence (t(10) = 0.074, p = 0.94, 95% CI [ -6.32, 6.76]). Following the amalgamation of M1c and S1c groups into the SMc-cTBS group, the second planned comparison demonstrated a significantly larger improvement in motor performance (B score) for the SMc-cTBS group (M = 5.74, SD = 4.784) compared to the Sham group (M = 3.06, SD = 1.146), t(13.54) = 1.82, p = 0.045, 95% CI [-0.48, 5.86]). 196 Figure 40a, b, and c: Motor skill acquisition curves for contralesional primary somatosensory cortex (S1c), contralesional primary motor cortex (M1c), and sham stimulation groups. Single-subject normalized motor skill exponential curves for active continuous theta burst (cTBS) stimulation groups (S1c, M1c) (top panel; a, b) delivered over S1c (Figure 40a) and M1c (Figure 40b), and for the sham stimulation group (bottom panel; Figure 40c). Dotted black line indicates individuals with negative B scores and identified as motor skill non-responders. Solid gray line indicates individuals with positive B scores and identified as motor practice responders. Dashed red line represents the mean motor skill acquisition curve for each stimulation group. 00.20.40.60.822.214.171.124 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171Normalized Response Time (s)TrialsM1c GroupM1 M2 M3 M4M5 M6 M7 M8M9 M1c Mean00.511.522.51 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171Normalized Response Time (s)TrialslSham GroupSham1 Sham2 Sham3 Sham4Sham5 Sham6 Sham7 Sham8Sham9 Sham Mean1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 161 171TrialsS1c GroupS1 S2 S3 S4S5 S6 S7 S8S9 S10 S11 S1c Meana. b.c. 197 Clinical baseline measures for SMc-cTBS group: In the combined SMc-cTBS group, 12 of 20 participants responded positively to cTBS paired with motor skill practice, as evidenced by a positive B score. When considering the individuals in the SMc-cTBS group, independent group t-tests and a Fisher’s exact test (binary data for stroke location [C: 1; SC: 0]) demonstrated no significant differences in demographic (age: t(18) = 0.08; p = 0.93) or clinical characteristics (stroke location: p = 0.64; PSD: t(18) = 0.70, p = 0.49; UE-FM: t(18) = 0.25, p = 0.81; paretic WMFT rate: t(18) = 0.44, p = 0.67) between responders (n = 12) and non-responders (n = 8). Thus, participants were well-matched for clinical and demographic data, indicating that these factors did not significantly contribute to individuals’ capacity for improved motor learning following cTBS and motor skill practice. White matter tractography for SMc-cTBS group: In the combined SMc-cTBS group, following the GROUP (responder, non-responder) × WM-FA (NL-CST, L-CST, CMC) MANOVA, there was a significant main effect of GROUP (responder, non-responder) for WM-FA in the NL– and L-CST, and CMC (Wilks’ λ= 0.62, F(3, 16) = 3.24, p = 0.05, ηρ² = 0.38, large effect size). Post hoc univariate tests revealed that WM-FA from tracts within the CMC (F(1, 18) = 7.69, p = 0.013; see Table 11, Figure 41) were significantly higher (greater FA) in responders (CMC-FA: M = 0.48, SD = 0.0149) compared to non-responders (CMC-FA: M = 0.46, SD = 0.0113). However, FA from the NL– and L-CST did not significantly differ between responders and non-responders (F(1, 18) ≤ 3.34, p ≥ 0.084; see Table 11). Therefore, group differences in the microstructural integrity of the CMC network had better predictive value than CST tracts. 198 Table 11: Comparison (mean and SD) of responder versus non-responder DWI characteristics in SMc-cTBS group. DWI Responders Non-responders F-test df (1,18) p-value Mean SD Mean SD NL-CST 0.50 0.02 0.48 0.04 3.34 0.084 L-CST 0.42 0.08 0.46 0.05 0.94 0.345 CMC 0.48 0.01 0.46 0.01 7.69 0.013* Significant effect of diffusion-weighted fractional anisotropy (DW-FA) between groups. Wilks’ Lambda = 0.62, F(3,16) = 3.24, p = 0.050. Post hoc univariate tests revealed FA from tracts of the CMC (F(1, 18) = 7.69, p = 0.013) was significantly higher in responders versus non-responders. 199 Figure 41: Comparison of responder versus non-responder of CMC in the SMc-cTBS group. Error bars represent SD. Figure 42: Subject-specific examples of white matter tractography. Examples of DW-WM tracts from the a) non-lesioned corticospinal tract (NL-CST), b) lesioned corticospinal tract (NL-CST), and (c) constrained motor connectome (CMC) for a motor practice responder in the SMc-cTBS group. 0.440.450.460.470.480.490.50Nonresponder ResponderFractional Anisotropy (FA)p = 0.013*NL-CST L-CST CMCa. b. c. 200 Figure 43: Statistical study design and results flow chart. Four ordered investigative steps were performed to determine the motor practice responder profile for continuous theta burst stimulation (cTBS) paired with motor skill practice. These steps included the following. First, we determined the effect of stimulation group (contralesional primary motor cortex [M1c]; contralesional primary somatosensory cortex [S1c]; sham stimulation) on motor sequence learning. At retention, all groups demonstrated superior performance for repeated compared to random sequence (p < 0.05), but there was no significant effect of the stimulation group on motor sequence learning (p > 0.05). Second, to further investigate the effect of stimulation group on motor sequence learning, for B score of the repeated sequence (representing change in response time total [RTT] across practice when fitted to an exponential function), we performed planned comparisons between stimulation groups among the motor practice responders only. We compared the difference between the M1c and S1c groups for B scores for the motor practice responders only. Then we combined M1c and S1c groups into a sensorimotor group (SMc-cTBS) to compare the effect of active stimulation to sham stimulation on B score for the motor practice responders only. There was no significant difference between improvement in motor performance for the M1c and S1c groups (p > 0.05); however, there was a difference for the SMc-cTBS group compared to sham stimulation (p < 0.05). Third, to understand the responder profile of SMc-cTBS paired with motor skill practice, we assessed the differences in demographic and clinical measures between motor practice responders and non-responders. There were no significant differences in any measures (p > 0.05). Fourth, to investigate potential biomarkers of response to SMc-cTBS paired with motor skill practice, we assessed differences in white matter (WM) biomarkers between motor practice responders and non-responders for the SMc-cTBS group and showed a significant difference for fracional anisotropy in the constrained motor connectome (FA-CMC) between responder and non-responder groups (p < 0.05). 1. Was there learning on the repeated sequence? 2. Did groups (M1c, S1c, Sham) demonstrate greater improvements in repeated compared to random sequence performance? For the motor practice responder group, was there greater improvements in motor performance for cTBS stimulation (M1c + S1c) versus sham stimulation? M1c and S1c were combined to investigate the effect of contralesional sensorimotor (SMc) cTBSstimulation compared to sham stimulation on B score. For the SMc-cTBS stimulation group, were there differences in clinical measures between motor practice responder and non-responder groups? For the SMc-cTBS stimulation group, was there a difference in WM FA for the NL- and L- CST and CMC between motor practice responder and non-responder groups?SMc-cTBS motor practice responders demonstrate differences in WM FA for the CMC compared to non-responders. Study Questions126.96.36.199.Impact on analysisResult: 1. At retention, main effect of sequence, and 2. no interaction of group x sequence. Impact: Repeated sequence B score was used to determine motor practice responder versus non-responder. Result: SMc group demonstrated a significantly greater improvement in motor performance than the sham stimulation group. Impact: For the next set of analyses, we looked at the combined SMc group to investigate responder profile of cTBS.Result: No differences in clinical measures between motor practice responders and non-responders for SMc-cTBS stimulation group. Impact: Investigate white matter (WM) difference for non-lesioned and lesioned corticospinal tracts (CST), and constrained motor connectome (CMC) for responder groups.YESYESNOYES 201 Discussion We demonstrated that the residual white matter integrity of the CMC was significantly different between motor practice responders and non-responders to contralesional cTBS paired with motor skill practice. Following our intervention, we showed that, independent of receiving cTBS over the (1) contralesional primary motor cortex (M1c) or (2) contralesional primary somatosensory cortex (S1), or (3) sham stimulation, as a group, individuals with chronic stroke demonstrated the ability to learn a motor sequence. This was supported by significant differences across the repeated and random sequences at retention across all groups, with lower RTTs for the repeated versus random sequence. The lack of behavioural differences across stimulation groups is consistent with variable inter-individual responses to non-invasive brain stimulation observed in previous studies (Brodie, Borich, et al., 2014; Carey et al., 2014). This finding motivated our investigation into a sub-group of “responders”. In the motor practice responders (i.e., those individuals who showed a decrease in response times across practice of the repeated sequence based on curve fitting procedures), there were differences in the means for the M1c and S1c cTBS groups in comparison to sham stimulation, demonstrated by a significant one-tailed t-test. M1c and S1c showed larger B values, indexing greater improvements in response times compared to the sham group. However, similar to past work (Meehan et al. 2011), there was no motor learning-related difference in our two stimulation groups, M1c and S1c (see also Enomoto et al., 2001; Meehan et al., 2011; Mochizuki et al., 2007). Thus, we combined the groups to assess the responder profile of the SMc-cTBS group. Response to cTBS paired with motor skill practice was determined by diffusivity properties (FA) within a network that has been previously identified as important for 202 motor learning in healthy older adults, that is, the white matter (WM) integrity of a motor learning network (i.e., the constrained motor connectome, CMC). Our finding that a complex WM motor network known to support motor learning (the CMC), is related to the responsiveness of individuals to cTBS paired with motor practice, extends previous findings showing that greater FA, which is considered to reflect higher white matter microstructural integrity, of the corticospinal tract and corpus callosum are important in stroke recovery (Brodie, Meehan, et al., 2014; Carey et al., 2014; Lindenberg, Renga, Zhu, Nair, & Schlaug, 2010; Mang et al., 2015; Zhu, Lindenberg, Alexander, & Schlaug, 2010). Similar to previous studies, responsiveness to non-invasive brain stimulation was not explained by standard demographics, such as age, or stroke severity as measured by UE-FM or WMFT (Brodie, Borich, et al., 2014; Carey et al., 2014). Contrary to prior literature, post-stroke duration (Lee et al., 2015; Rose, Patten, McGuirk, Lu, & Triggs, 2014), stroke location (Ameli et al., 2009; Emara, El Nahas, Elkader, Ashour, & El Etrebi, 2009), and corticospinal tract microstructural integrity (Brodie, Borich, et al., 2014; Carey et al., 2014) did not characterize responsiveness to cTBS paired with motor skill practice. Inconsistency in measures that explain variability in response to non-invasive brain stimulation may reflect the lack of generalization between stimulation protocols (i.e., continuous versus intermittent TBS; brain region stimulated [M1 versus S1]; contralesional versus ipsilesional hemisphere) (Auriat et al., 2015). To further the field of rTMS and post-stroke recovery, future work is needed to determine the factors that elucidate response variability under specific non-invasive brain stimulation protocols. Following stroke, spared bihemispheric neuronal connections between direct pathways of the M1 and the CST, as well as indirect pathways such as the reticulospinal and/or rubrospinal, may contribute to positive capacity for motor change (Rüber, Schlaug, & Lindenberg, 2012; 203 Schulz et al., 2017; Wadden et al., 2015; Ward, 2015a). Given the bihemispheric representation within the CMC, our findings may reflect the overlap between pathways in the CMC and those involved in interhemispheric signaling during cTBS. We hypothesize, based on previous work (Carey et al., 2014), that the delivery of cTBS over a disordered structural network (i.e., low FA of connecting pathways) may attenuate cTBS-induced modulation of activity in targeted areas. In the present study, disruptions in the CMC could attenuate the neuroplastic effects of brain stimulation paired with motor skill practice (Carey et al., 2014; Hummel & Cohen, 2006). This conjecture is in accordance with opinions based on a current review in the field of brain stimulation, suggesting that individuals should be stratified based on specific, pre-existing neurobiological integrity associated with the clinical intervention of interest (Ward, 2015a). Our methodological approach for evaluating motor performance and stratifying responders and non-responders was closely based on previous segregation procedures (Carey et al., 2014). We employed a curve fitting technique to categorize motor learning-related change from individual data across the entire practice period (Wadden et al., 2017; Yamashita et al., 2015). Assessment of each individual’s capacity for motor learning–related change in this manner is not constrained to a predetermined set number of trials, but is based on the curvilinear pattern of performance change throughout the entire period of skill evolution. As the field of stroke rehabilitation works to identify biomarkers for stroke recovery, this presents a refined and more sensitive method for capturing behavioural states of recovery that could be applied to other interventions (Burke & Cramer, 2013). In the motor practice responder group (positive B score), there was a significant difference when comparing SMc-cTBS stimulation with sham stimulation on improvements in response time. There was no significant difference between cTBS groups (M1c, S1c). These data align with our previous research, where no difference in improvements in 204 movement time was observed between cTBS stimulation groups (S1c, M1c), only between cTBS stimulation groups (SMc-cTBS) and sham stimulation (Meehan, Dao, et al., 2011). Improvements in the performance of complex motor skills involve broad networks and strengthened connections between the sensory and motor cortices (Meehan, Dao, et al., 2011). Improvements in response time may reflect enhancments in the encoding processes for force, target direction, and egocentric coordinate transformations that occur between motor and sensory cortices (Meehan, Dao, et al., 2011). The interaction between M1 and S1 during skilled learning is critical, and the similar behavioural findings from the present study may reflect the reciprocal strengthening of connections in individuals with undisrupted WM linkage between regions; or our findings may alternatively indicate that individuals with a more intact motor network at baseline have a greater capacity for motor recovery following this intervention. Furthermore, rather than an isolated effect of cTBS on M1 versus S1, these regions in individuals with preserved WM integrity of the CMC work in concert to promote neuroplastic change and motor learning in individuals with chronic stoke. Limitations The identification of specific biomarkers that elucidate the responder versus non-responder distinction is an important first step in understanding the mechanisms of non-invasive brain stimulation paired with motor skill practice (Bernhardt et al., 2016). A limitation of this study is the relatively small sample size (n = 28; M1c = 9, S1c = 11, Sham = 8). A larger sample may help to verify the CMC as a biomarker of cTBS response. Beyond our planned comparisons (two-tailed and one-tailed independent t-tests), we observed a lack of behavioural effects, and interactions between groups and sequences (random, repeated) using inferential statistics for performance of the serial targeting task. Furthermore, it is important to consider the B value, 205 which measures the expected change score and was used to differentiate between motor practice responders and non-responders, may be indicative of poor and good early performance, respectively. The motor skill non-responders performed the task faster earlier in practice, and therefore demonstrated a ceiling effect and less improvements in performance over the 5 subsequent days of practice. This observation indicates that individuals might benefit from practicing a motor skill that is tailored to individual motor abilities, allowing them to enhance skill acquisition over the entire course of the training period. While additional independent t-tests demonstrated there was no statistical difference between the SMc-cTBS motor practice responders and non-responders for repeated sequence mean RTT on the baseline (Session 1, t(18) = 1.14, p = 0.27, the group means showed motor practice responders had worse baseline performance (M = 16.07, SD = 6.25) compared to the motor practice non-responders (M = 12.70, SD = 6.81). In addition, the predicted asymptote value, A, which reflects estimated plateau in performance, was not statistically different between motor practice responders and non-responders, t(18) = 0.44, p = 0.67, demonstrating similar practice-end motor performance levels. However, future research should investigate the response to cTBS over the contralesional hemisphere paired with motor skill practice in a more homogeneously impaired group of individuals with stroke. Secondly, the CMC is a group-level approach; the diffusion-weighted images are normalized to MNI space to overlay a common motor network mask. Ideally, individualized masks created from an fMRI motor learning experiment prior to receiving a non-invasive brain stimulation intervention, in combination with the CMC, may help predict individualized responses to cTBS paired with motor skill practice. Our findings will support future work to 206 investigate the possible usefulness of using fMRI-guided DWI as a methodological approach to identify biomarkers of recovery. Future studies Individuals with stroke develop compensatory patterns of activation that promote rapid changes in motor function (Wang et al., 2010). However, these compensatory patterns of activity may have long-term detrimental effects (Levin et al., 2009), that are independent of improvements in motor impairment. Shifting individuals into a more normal pattern of activation early post-stroke may be an important mechanism for long-term motor recovery. As such, the capacity to determine responder and non-responder biomarker profiles for specific types of brain stimulation protocols is an important field of inquiry. Future studies need to determine individual functional and structural connectivity patterns associated with changes in motor function that evolve naturally (sham stimulation) compared to bio-electrically-induced changes via non-invasive brain stimulation. Serial imaging of fMRI and DTI connectivity has been suggested as means for determining the relationship between behavioural and brain changes; however, many studies only examine changes pre- and post–intervention (Burke & Cramer, 2013) and consider individuals in the chronic phase of stroke recovery. Formulating experimental designs to investigate individual differences throughout interventions is essential to the understanding of variations in outcome measures and, furthermore, is central to deriving maximal individualized treatment effects. Indeed, not all individuals demonstrate improvements over the same planned trajectory or number of practice sessions practice (Hardwick et al., 2016; Wadden et al., 2017); thus, individualized interventions are needed, based on persons’ own potential for improvement. 207 Conclusions The residual WM structure of a novel motor network in the brain, in the chronic phase of stroke, has emerged as a potential biomarker of motor recovery. The underlying neurophysiological mechanisms that yield the relationship between WM pathways and response to repetitive non-invasive brain stimulation needs further investigation. As a recent review outlines, findings from repetitive non-invasive brain stimulation studies, used to enhance the potential effects of activity-dependent plasticity, have resulted in positive outcomes in stroke populations (Hsu et al., 2012; Ward, 2015b). However, the effects of non-invasive brain stimulation are known to be variable, which suggests there are specific underlying mechanisms that drive activity-dependent plasticity following non-invasive brain stimulation paired with motor practice (Ward, 2015b). The findings from the present study demonstrate the potential importance of evaluating widespread, functionally relevant WM networks to characterize the response profile of the primary dependent outcome measure. 208 Chapter 6: General discussion Overview To understand the impact of rehabilitation interventions, it is critical to employ methods that characterize how individuals, including people with stroke, change their behaviour and brain networks as they learn new motor skills. Conventionally, change in motor performance is quantified with discrete measures (i.e., repeated measures analysis of variance [ANOVA] for blocks or sessions of mean response time) of behaviour taken pre– and post-practice or across practice blocks (Borich et al., 2014; Boyd et al., 2009; Boyd et al., 2007; Boyd et al., 2008). However, motor skill acquisition follows a characteristic pattern of change; gradual improvements in motor performance occur with increasing practice time (Karni et al., 1998). The first objective of this thesis was to move beyond pre– and post-testing of motor skills, and through individualized curve fitting techniques, offer added insight into the relationship between performance and learning during motor skill acquisition (Chapters 2 and 3). Typically, patterns of brain activity are altered in concert with changes in behaviour during motor skill acquisition (Grafton et al., 2002; Karni et al., 1995; Karni et al., 1998). However, after stroke it is common to find patterns of brain activity that differ from that noted in healthy individuals (Meehan, Randhawa, et al., 2011). However, the functional significance of these differences is poorly understood. The second objective of this thesis was to investigate functionally connected gray matter (GM) networks associated with motor learning (Chapter 4), as well as the relationship between the underlying white matter (WM) of motor networks and response to rTMS post-stroke (Chapter 5). The major findings of each research chapter are summarized below. 209 Summary of findings by chapter Chapter 2: The primary aim of this chapter was to evaluate learner adaptive practice conditions in a young, healthy population to inform future work on the challenge point framework and curve fitting. While the individualized “challenge point” remains an elusive yet important notion, the findings from Chapter 2 reinforce the significance of increasing cognitive effort during practice to support enhanced motor learning. During the performance of the discrete pairing task (DPT), a slower rate of improvement during practice (Study 1, baseline), and high contextual interference (CI; Study 2, learner adaptive) was associated with superior performance on a retention test. This finding motivated our investigation of the relationship between practice curve parameters and motor learning–related change across different tasks (i.e., motor adaptation tasks) and other populations (i.e., older adults, individuals with stroke). Furthermore, our findings highlight the merit of individualizing practice schedules to enhance skill acquisition and retention. Chapter 3: In Chapter 3, across five days of practice, rate of improvement was faster in older adults than in individuals with stroke, during continuous tracking task (CTT) performance of a repeated sequence. However, this was a single-task learning design (not multi-task, as with the studies detailed in Chapter 2), and at a comparably fixed difficulty level throughout the period of skill evolution. With populations such as older adults or individuals with neurological disease, in whom there may be cognitive and/or motor impairments, a faster rate of motor skill acquisition until an asymptote in performance occurred was an important factor in the maintenance, or improvement, of performance following the retention interval (end of practice to the 24-hr delayed retention test). Wolf Motor Function Test (WMFT) at baseline was predictive of motor performance-related change in the stroke group. Findings from Chapter 3 highlighted, 210 the potential use of rate of motor skill acquisition from curve fitting to prescribe and manipulate individualized interventions during motor sequence learning in future studies. The behavioural differences observed between healthy individuals and individuals with stroke motivated the investigation of the neurological differences associated with reorganization of brain networks during motor learning. Chapter 4: From the larger sample of individuals recruited for Chapter 3, Chapter 4 involved the selection of a smaller group of individuals with right subcortical hemisphere lesion for a question-focused fMRI study. Employing the same task and curve fitting methods as Chapter 3, following five days of motor skill practice, individuals with stroke did not show the same level of functional network integration as healthy individuals in a whole-brain functional connectivity analysis. This was presumably due to the heterogeneity of functional reorganization following stroke. This finding was the motive for a secondary analysis that was performed using a binary mask of the functional network activated from the primary whole-brain analyses to assess within-network connectivity of a constrained network. Within the constrained network of brain regions, for the stroke group, connectivity within a smaller motor network correlated with motor sequence performance on the retention test. The findings from Chapter 4 underscore the significance of a functionally integrated motor network as an important neurophysiological predictor of motor learning–related change in individuals with stroke, as opposed to focusing on isolated brain regions. Chapter 5: Building on the behavioural models of motor learning in Chapters 2 and 3, and the functional motor network extracted in Chapter 4, the final study described in Chapter 5, aimed to characterize the response profile of individuals with chronic stroke who received non-invasive brain stimulation (cTBS over the contralesional hemisphere) paired with motor skill 211 practice. Individuals that demonstrated a positive response to cTBS over the contralesional sensorimotor (SMc) paired with motor skill practice of the serial targeting task (STT), as measured by an improvement in their capacity for motor learning–related change, had greater fractional anisotropy (FA) values in a white matter (WM) motor network (“constrained motor connectome” [CMC]) compared to individuals who did not improve on the task. Findings from Chapter 5 support the idea that the underlying WM pathways of a motor learning–related network are a putative biomarker with which to characterize the response profile to cTBS over the contralesional hemisphere. Overall, the research outlined in this thesis covers a diverse but related set of research hypotheses and interpretations, surrounding the deeper investigation into novel behavioural and neurological measures that may influence motor learning rehabilitation interventions, as well as the identification of novel motor learning biomarkers in individuals with chronic stroke. The findings from this thesis suggest that future research investigations that connect behavioural and neurological measures to individual patterns of motor learning–related change may lead to better designed rehabilitation interventions post-stroke. Moreover, our observations underscore that assessments of motor performance and learning and related biomarkers must be multi-faceted: encompassing individually tailored learning strategies to enhance skill acquisition and retention; more sensitive measures of change in motor skill, based on a greater number of data points; and network-based evaluations of neural structural and functional integrity. Challenge in practice relates to enhanced motor learning for young, healthy adults Chapter 2 presented the results from a within-subjects study of young, healthy individuals designed to investigate the challenge point framework (CPF) that used a computer-controlled, learner-adapted algorithm during practice of a motor sequence task (DPT). The experimental 212 studies in Chapter 2 were conducted for two reasons: (1) to investigate if rate of skill acquisition is related to retention (on a related task, i.e., Study 1); and (2) to determine whether knowledge of individuals’ rate of motor skill acquisition and baseline performance (from Study 1) could be used to manipulate contextual interference (CI) and enhance motor learning in subsequent practice (Study 2). The studies from this chapter were largely motivated by findings from Choi et al. (2008), which showed enhanced learning of a motor adaptation task using an individualized, computer-controlled algorithm to manipulate the level of nominal difficulty of practice based on current performance level. To assess if the benefits of learner-adapted practice conditions for learning could transfer to a motor sequence task, and to improve upon the algorithm by incorporating individual skill levels, in the second part of this study individual practice schedules were manipulated based on individual rates of skill acquisition and response times from three practice phases identified in the first half of the study (i.e., Study 1). Participants practiced three different motor sequence tasks at three levels of individualized CI across separate days. In Study 1 of Chapter 2, a slower rate of motor skill acquisition in practice, along with more time spent in phase one (cognitive phase) of practice, were related to faster response times in retention. This finding is in accordance with interpretations from the CPF and CI literature; practicing under conditions with greater challenge/difficulty, which disrupt initial practice performance, enhances long-term motor learning (Guadagnoli & Lee, 2004; Lin et al., 2012a; Wright et al., 2016). This finding is in line with another recent study (Lakhani et al., 2016), where motor performance data was fitted to an exponential function in order to investigate the relationship between changes in motor behaviour and activity-dependent changes in myelin in the brain of healthy individuals over 10 days of motor skill practice (Lakhani et al., 2016). As with my study, the practice paradigm was performance-dependent, such that when participants’ 213 performance met a predetermined criterion, the difficulty of the task was progressively increased. Individuals with slower rates of skill acquisition across the first nine training sessions demonstrated a faster rate of improvement on the last day of skilled practice (Lakhani et al., 2016). This finding is neurologically supported by the negative relationship between slower rates of motor skill acquisition and the greatest increases (days 1 to 10) in myelin water fraction in the contralateral intraparietal sulcus (IPS), an area associated with the demands (i.e., visuomotor attention) of the task (Lakhani et al., 2016). While the motor tasks (semi-immersive virtual reality-based intercept and release and DPT) and duration of practice (nine days and one day) differed, findings from Lakhani et al. (2016) and Study 1 of Chapter 2, respectively, demonstrate the potential merit of a slower rate of skill acquisition to promote gradual behavioural adaptations and enhance consolidation of motor skills in young, healthy adults. Thus, this thesis provides a novel contribution concerning the relationship between the rate of motor skill acquisition and learning-related change, which has been confirmed in a separate data set and found to be related to myelin changes (Lakhani et al., 2016). These findings should motivate further research to investigate this relationship in other laboratory tasks, applied situations and clinical populations. Overall, findings from Study 2 in Chapter 2 demonstrated that practicing at an individualized “challenge point” that generated a condition of high CI, and potentially high cognitive effort, enhanced motor learning occurred during the DPT in young, healthy individuals. Higher individualized CI condition during practice led to the fastest responses times in the retention test. This finding is supported by previous CI literature, where it has been shown that practice conditions with more rather than less interference diminished performance during practice trials actually enhanced performance on a delayed retention test (Brady, 1998, 2004; 214 Magill & Hall, 1990; Shea & Morgan, 1979; Wright, Magnuson, & Black, 2005). Response times from Study 2 in Chapter 2 showed a linear relationship with decreasing practice condition difficulty (i.e., high, moderate to low levels of CI) and performance on a delayed retention test, rather than the inverted U-shape outlined in the CPF. The inverted U-shape is presented to show the existence of an optimal challenge point, which is at the highest point of the U; above and below this point practice conditions yield too high or low levels of challenge, respectively. While findings from Study 2 of Chapter 2 demonstrated the efficiency of a learner-adapted algorithm (without human intervention) to produce high levels of challenge based on individualized performance during practice, we did not generate a condition that was above the optimal challenge point, and thus presented level of challenge that was too high. In future work, to determine the effectiveness of computer-controlled, learner-adapted algorithms for motor sequence learning, the high learner adaptive difficulty condition must be compared to a randomized CI schedule (independent of performance). This would provide evidence for the CPF (i.e., optimal challenge point) and support for performance-dependent practice based on learner adaptive algorithms. Determining the challenge point parameters for learner-adapted algorithms remains difficult (Choi et al., 2011; Guadagnoli & Lee, 2004). The continued investigation into the learner-adapted algorithm parameters for specific tasks and populations is warranted. For example, in the Choi et al. (2008, 2011) studies, the performance rate parameter and performance reference value were set empirically based on pilot data from healthy individuals. When the learner-adapted algorithm was used in a stroke population (Choi et al., 2011), compared to healthy control performance, there was an increased deviation in current performance from the performance reference value, derived from healthy control data. This finding indicates that the 215 rate of the performance parameter in the algorithm was too high or low for the stroke population. Furthermore, because rehabilitation interventions occur for a longer period, it is unknown whether the parameters of the learner-adapted algorithm, which were based on one baseline practice session, are relevant, or if they need to be updated using a new “baseline” session. Rate of motor skill acquisition is associated with motor performance–related change in individuals with stroke Nonlinear, exponential fits have been used on motor performance data following stroke (Lang & Bastian, 1999, 2001). However, the approach presented in this thesis advances past efforts by employing the rate of skill acquisition extracted during practice to inform the retention of a motor skill. Although it is important to understand how performance evolves with practice, it is equally important to understand the relationship between practice parameters and motor learning outcomes. The findings from Chapter 3 showed that individuals with chronic stroke (ST group) had significantly slower rates of improvement in implicit, sequence-specific motor performance, compared to a healthy control (HC) group, when RMSE performance data were fitted to an exponential function. Based on mean difference in RMSE at retention, there was no evidence of implicit, sequence-specific motor learning for either HC or ST groups. Although this was not expected, other groups have also shown altered implicit learning of continuous, compared to discrete, tracking tasks (Chambaron, Ginhac, Ferrel-Chapus, & Perruchet, 2006; Van Ooteghem et al., 2008). It may be that the embedded sequence (continuous waveform) required more practice to become learned, particularly for these older adults. While through conventional analyses, I did not detect group-level differences between random and repeated sequences at retention, I did observe significant relationships between rate of improvement in practice and motor learning-related change at retention for repeated sequence performance only. 216 In Chapter 3, individuals practiced a continuous tracking task (CTT) over 5 days of practice with no explicit knowledge of the presence of a repeating sequence. The HC and ST groups showed a positive relationship between rate of change in implicit, sequence-specific motor performance during practice and motor learning-related change at the delayed retention test (i.e., reduced loss of motor performance over the retention interval). Individuals that showed a faster rate of improvement during one-day of practice in Chapter 2, were more likely to have spent a greater amount of time in the late phases of practice (Phase II/III), near a plateau in performance, than individuals with a slower rate of improvement. Practicing in Phase III, and achieving a plateau in performance, has been suggested by researchers to reflect expertise at a skill, and a ceiling effect, as performance is near perfect (Doyon et al., 2003). Individuals with faster rate of improvement, may have achieved a consistent movement pattern with minimal variation (Shumway-Cook & Woollacott, 2007). However, based on RMSE scores in Chapter 3, a ceiling effect was not achieved. It is possible, due to the nature of the task (i.e., minimal instructions), that a “false plateau” in performance was obtained; with different training methods (i.e., feedback on performance, knowledge of results [KR]) further improvements possibly could have been observed (Ericsson, Krampe, & Tesch-Römer, 1993). I hypothesize that a faster rate of improvement and practice in Phase III was associated with the achievement of a consistent movement pattern (not necessarily optimal) that was less susceptible to a loss of performance over the retention interval during implicit repeating sequence tracking. In Chapter 3, for absolute retention measures, the HC group (older adults) showed a positive relationship between slower rate of change in implicit sequence-specific motor performance and less error in motor performance at the retention test. The slower rate of improvement associated with better performance at retention (i.e., learning) may reflect 217 continuous trial-and-error processes as individuals slowly learn the repeated sequence regularities and track with greater accuracy. Less time spent in the later phases of practice, and greater variations of movement patterns make individuals more susceptible to relative loss of performance in the retention interval. However, overall, HC individuals with slower rates of improvement tracked with greater absolute accuracy at retention. The relationship between rate of improvement and absolute RMSE at retention for repeated sequence performance was not observed in the ST group. The heterogeneity of cognitive and motor impairments may have weakened this association within the stroke group. The differential findings for the relationship between rate of skill acquisition and motor learning–related change in young versus older adults and individuals with stroke may reflect varying levels of motor ability. Based on the level of impairment in different populations, processes of learning may be uniquely influenced. For example, under heightened cognitive demands (i.e., high CI), older adults have been shown to activate different brain networks, reflective of distinct memory processes, compared to younger adults when learning a motor sequence task (Lin et al., 2012a, 2012b). Depending on the demands of the task (i.e., greater or less information processing; explicit versus implicit), determining appropriate level of challenge of the task is fundamental to skill acquisition and learning of new skills. These studies are the building blocks for the systematic development of a framework based on curve parameters to aid in the creation of individualized stroke rehabilitation interventions that incorporate CPF to enhance long-term motor learning. Performance curve to predict individualized dose in stroke rehabilitation Currently, the field of stroke rehabilitation accepts that “more is better” with regards to prescribing a dose of treatment; or the dose is indiscriminately assigned to all individuals with 218 stroke, regardless of their level of motor impairment and function (Bernhardt et al., 2016). This approach may impede the progress of the field, and is a less than sophisticated method with which to prescribe the treatment needed to drive neuroplastic change. The methods from Chapter 3 offer a first insight for future studies to use of curve fitting to enable the generation of practice that is not based on a pre-determined number of trials, but instead on each individual’s pattern of change. In Chapter 3, we introduced an equation, based on the exponential function, to determine the trial number at which an individual reaches their predicted asymptotic performance measure. There were significantly more predicted trials until asymptote in performance for both healthy individuals and individuals with stroke for the repeated sequence compared to random sequences across practice. Interestingly, compared to Chapter 4, implicit sequence-specific learning was not shown in Chapter 3 as demonstrated by the lack of significant difference between random and repeated performance error at a retention test. While there was an overlap between individuals in Chapter 3 and 4, there was a 40% difference of participants between testing samples. Reasons for this difference may be due to severity of stroke; upper extremity Fugl-Meyer (UE-FM) was marginally higher (showing less motor impairment) for the ST group in Chapter 4 (M = 54.3, SD = 12.95) compared to Chapter 3 (M = 52.7, SD = 13.0), suggesting that those with lower baseline motor impairment might be more susceptible to motor learning-related change in performance at retention, as opposed to simple improvements in general motor skill. The lack of replication of the implicit motor learning effect speaks to the high degree of inter-individual variability in learning in people with stroke and older adults. Stroke severity has been shown to influence the magnitude of implicit learning across acquisition and retention of motor sequence tasks (Boyd et al., 2007). Additionally, findings from a meta-analysis of implicit learning of the hemiparetic 219 upper extremity in individuals with stroke showed no significant implicit learning for discrete and continuous tracking tasks (Kal et al., 2016). In apparently healthy older adult populations varied findings for the degree of implicit motor learning have also been observed (Boyd et al., 2008; Chambaron et al., 2009; Chambaron et al., 2006). These varied findings not only speak to the influence of age, and stroke severity but potentially day-to-day factors such as motivation, fatigue, and attention that attribute to the inter-individual variability that impacts implicit motor sequence learning (Lewthwaite & Wulf, 2017). There is still much to investigate on the interaction between individual abilities (i.e., cognitive and motor abilities, motor impairment level), and internal (i.e., attention, fatigue, motivation) and external (i.e., nominal task difficulty) factors and implicit motor sequence learning. Moreover, these observations highlight the need to evaluate learning throughout the entire practice process, rather than at discrete pre- and post-training time-points. Based on the findings from Chapter 3, manipulating the rate of improvement during practice (i.e., challenge of practice) and practicing until a predicted asymptotic performance trial may influence the effects of implicit sequence-specific learning outcomes in older adults and individuals with stroke. Before disparities in findings are exclusively attributed to age-related and/or motor impairments post-stroke, individualized approaches that involve consideration of the level of task difficulty and dose of practice are warranted. However, the conflicting findings between Chapter 3 and 4 highlight the need for future research to determine if implicit sequence-specific learning of a continuous tracking task can be generalized to a larger group of older adults and individuals with stroke. Future research is needed to test the efficiency of such an approach in laboratory and clinical settings, and across varied motor skills. Given the large amount of heterogeneity in the 220 presentation of motor deficits and the amount of change associated with motor learning (or rehabilitation) in older adults and those with stroke, using performance curves to characterize responses may offer a more precise approach with which to capture changes in motor behaviour after stroke. Functionally connected motor learning–related network Univariate fMRI analysis techniques have advanced our theoretical understanding of the neurophysiological processes involved in motor learning; however, our knowledge of how activity in brain networks shift to support learning remains incomplete. Using multivariate statistical MRI analysis techniques allows for a more thorough comprehension of the communicative processes between brain regions. In Chapter 4, I employed a multivariate analysis to extract a motor-related functional brain network during motor sequence learning through a technique known as “constrained principal component analysis” (CPCA) (Woodward et al., 2006). Most network analyses in stroke populations employ resting state fMRI; however, here I used task-based fMRI, which plays to the strengths of the CPCA approach as it computes functional brain networks for which variance has been constrained to the task. This allows determination of functionally-relevant motor learning–related networks during task performance. Analyses were not constrained to ROIs selected a priori; instead the goal was to compare BOLD signals within the motor learning network between healthy individuals and individuals with stroke. While the ability to learn and relearn motor skills is preserved after stroke, the findings from Chapter 4 demonstrated reduced functional connectivity in the motor learning network for individuals with stroke compared with healthy individuals. Based on findings from Chapter 3, this neurological finding raised more questions about individualized practice paradigms and the potential relationships that may exist with patterns of brain activation. Specifically, lowered 221 functional connectivity may reflect behavioural processes that indicate individuals with neural damage need different, and potentially individualized, doses of practice before the same level of functional network integration can be observed. Accordingly, the continuous tracking task (CTT)’s level of difficulty may have initially been too high, with relatively high amounts of cognitive load for some individuals, and resulted in differential recruitment patterns of a task-relevant motor learning network. Reduced functional network integration has been observed during practice of a sequence task under high CI (increased practice difficulty), whereby brain regions that are typically recruited in the early phase of practice remained activated in the later phases. Wymbs & Grafton (2009) observed activation of the cortico-cerebellar circuits, which are predominantly active in the early phase during motor sequence learning (Doyon et al., 2003), and into the later phase of learning during practice of a motor sequence task under high CI conditions. Depending on the level of neuronal damage following stroke, the additional and sustained recruitment of multiple networks (i.e., cortico-cerebellar and striatal circuits) may necessitate a reduction in the difficulty of the task or an increase in the dose of practice. Future research should focus on developing individualized practice paradigms to determine if individuals with stroke can obtain similar levels of functional network integration, and thus task-relevant reorganization of motor networks to relearn and retain motor skills. Methodological approaches must become more sophisticated, integrative, and personalized to broaden theoretical frameworks and questions about the interactions between behavioural and neurophysiological networks. A relationship between individual differences in motor sequence learning-related network activation and retention performance is demonstrated in this thesis. This begins a line of inquiry surrounding the key idea that if practice conditions are adapted to the skill level of the learner, network connectivity could be optimized. Through 222 individualized practice, neural resources devoted to motor skill acquisition can be harnessed, rather than activation of unassociated brain regions relating to external factors, such as fatigue, lapse in attention, and motivation, that are incurred as a result of the functional task difficulty being too high or low, respectively. Differences in performance exist as a result of the skill level of the individual and practicing at an optimal challenge point may result in similar levels of activation and strength of connectivity within an optimized motor sequence-learning network. To date, the influence of adaptive practice schedules on the motor sequence learning-related network during practice and retention are unknown. Biomarkers for motor learning in chronic stroke Integrative methodological approaches advance our understanding of dynamic networks that support mechanisms of motor learning. Along with previous research (Grefkes & Ward, 2013; Wang et al., 2010), the findings from Chapter 4 provide evidence for disruptions within the functional motor network that lead to clinical implications. These disruptions, which are a result of tissue loss from insufficient blood flow to part of the brain (i.e., ischemic stroke), create functional imbalances in the motor network and are hypothesized to be rooted in damage to white axonal tracts connecting brain motor areas (Granziera et al., 2012; Turken et al., 2008). Thus, the motivation for Chapter 5 was to assess the underlying white matter (WM) of a motor learning-related functional brain network in individuals with stroke, based on data from Chapter 4. In the majority of past studies, only isolated WM pathways have been assessed to predict motor recovery in stroke (Borich et al., 2014; Borich, Mang, et al., 2012; Mang et al., 2015). In Chapter 5, I investigated a broad WM brain network as a biomarker that supports motor learning post-stroke. Using a WM brain network, rather than a single WM pathway, was an important methodological approach to the identification of a novel motor learning biomarker. Following a 223 stroke, a focal area of the brain is damaged that then impacts the function of distal areas from the site of the lesion (e.g., via Wallerian degeneration), which may otherwise appear normal. However, to perform complex motor skills, the brain works as a network to accomplish the numerous requirements of the task (e.g., problem solving, movement); and to evaluate the functional integrity of the post-stroke brain, we must evaluate the entire network of associated regions, instead of only examining discrete brain areas or fibre bundles. Presently, there is a call to identify biological markers (“biomarkers”) which, in the field of stroke rehabilitation, are defined as “indicators of disease state that can be used clinically as a measure reflecting underlying molecular/cellular processes that may be difficult to measure directly in humans, and could be used to predict recovery/treatment response” (Bernhardt et al., 2016). The objective of Chapter 5 was therefore to identify a biomarker to determine the behavioural response to cTBS over the contralesional sensorimotor cortex (SMc) paired with motor skill practice. I employed fMRI-guided diffusion-weighted imaging (DWI) (Preti et al., 2014) to exploit the intimate relationship between brain structure and function, and to accurately identify meaningful pathways for motor learning post-stroke. Specifically, this methodological approach, fMRI-guided DWI, was used to assess the underlying WM microstructure of a previously known motor learning network, known as the constrained motor connectome (CMC), in Chapter 5. We categorized people with stroke as either responders or non-responders, based on their capacity for motor learning-related change in performance, which was determined from behavioural data fitted according to exponential curve fitting methods described in Chapters 2 and 3. The fMRI motor learning network used to constrain WM data was detailed in Chapter 4. Similar to previous studies evaluating anatomical differences associated with responders and non-responders to repetitive non-invasive brain stimulation (Brodie, Borich, et al., 2014; Carey 224 et al., 2014; Tamas Kincses et al., 2008), we found a significant difference in WM integrity of the CMC between responders and non-responders, which demonstrates the utility of this approach. The findings from Chapter 5 advance the field of motor rehabilitation based on rTMS interventions, in identifying potential biomarkers associated with individual patterns of behavioural change post-stroke. To further build upon this finding, the development of a multimodal prediction model for repetitive non-invasive brain stimulation paired with motor skill practice is warranted. For example, a recent systematic review of biomarkers (Kim & Winstein, 2017) and consensus panel paper (Boyd et al., 2017) recommended that for the development of a prediction model for motor recovery multiple neurologic and clinical biomarkers are required. The most accurate models to predict motor recovery included structural MRI using DWI-derived fractional anisotropy (FA) values and corticospinal tract (CST)-lesion overlap volume measures, in combination with other neuroimaging measures, predominantly TMS measures, and clinical measures such as Fugl-Meyer score, age, or chronicity (Kim & Winstein, 2017). The DWI-derived FA values of the CMC were significantly different between our responders and non-responders to cTBS over the contralesional sensorimotor cortex (SMc) paired with motor skill practice; but increasing the sample size (greater than 50 individuals, as suggested by Kim & Winstein 2017), and including derived FA values of the CMC in a regression model with TMS and clinical measures would be a preferable way to determine the level of variance accounted for by each measure to predict response to cTBS paired with motor skill practice. Due to the promising findings from Chapter 5, indicating the WM integrity of the CMC as a potential biomarker of cTBS paired with motor skill practice, the next important step will be to determine if it can predict response status when employed a priori. Future studies should look 225 at whether DWI-derived FA values of the CMC can accurately predict an individual’s capacity for change following training using cTBS paired with motor skill practice. Promisingly, Stinear et al. (2017) successfully implemented the Predict Recovery Potential (PREP) algorithm in a clinical rehabilitation setting, which uses TMS values, DTI-derived FA values from the CST, and upper extremity strength to determine individual’s potential for upper extremity recovery at 12-weeks post-stroke (Stinear et al., 2012). The PREP algorithm correctly predicted the primary clinical outcome, the Action Research Arm Test (ARAT) (Lyle, 1981), for 80% of patients. Also, prediction guided rehabilitation therapy reduced the inpatient length of stay by one week, and increased the confidence level and modification strategies of therapists compared to a control (conventional therapy) group (Stinear, Byblow, Ackerley, Barber, & Smith, 2017). This is a promising direction in the facilitation of personalized interventions based the identification of conceivable responders to specific forms of treatment. Limitations More quantitative studies on the challenge point framework (CPF), specifically the use of learner adaptive algorithms to determine individual level of difficulty, is needed to assess the effectiveness of such an approach for motor sequence learning. Choi et al. (2008) were the only past group to successfully implement a learner algorithm to manipulate difficulty of practice for a motor adaptation task. However, there were key differences in parameters of the algorithms (i.e., rate of skill acquisition, performance reference value) in Study 2 of Chapter 2 and the Choi et al. (2008) study. Based on the only study that addressed this question (using computer-controlled learner adaptive algorithm for a motor sequence task, Study 2 of Chapter 2), more research is needed to determine if an individualized, optimal challenge point can be extrapolated across different motor tasks. In the future, the nominal difficulty of the DPT task, or other motor 226 sequence tasks (i.e., SRT, CTT, or STT) should be increased (e.g., by increasing the length of a sequence, decreasing target size, or lengthening the distance of a reach task). This would help test the inverted U-relationship outlined in the CPF (between challenge/task difficulty and motor learning), and to determine whether an optimal challenge point — somewhere between low and high difficulty — can be identified on an individual basis to help structure practice and motor learning paradigms. The exponential function as a performance curve fitting method was selected due to its proven accuracy compared to the power function when investigating individual skill acquisition (Heathcote et al., 2000). Prior to the Heathcote et al. (2000) findings, practice data was typically modelled with a power function, which led to the idea of the “Power Law of Practice” (Newell & Rosenbloom, 1981). Heathcote et al. (2000) modelled data at the individual level, across a range of tasks, and concluded that an exponential function was a more accurate depiction of individual data and provided a superior fit to the power function. Despite the evidence supporting the goodness-of-fit of exponential functions, we only observed “adequate-good” fits (Chapters 2, 3 and 5). It is likely that the fits we observed in these studies is more a function of the limited number of practice trials rather than the exponential modelling of learning processes, although admittedly no comparisons were made with power function modeling. A limitation of the studies reported in Chapters 4 and 5, was the sample size. In Chapter 4, the number of individuals in this study (n = 9), precluded our ability to investigate functional connectivity based on clinical measures of motor impairments. There are known differences in the functional and structural residual integrity of the brain based on the severity of stroke (Hayward, 2016). Increasing the sample size would allow for a more comprehensive investigation into the relationship between levels of functional connectivity and functional motor 227 impairment (i.e., according to Fugl-Meyer scores). In Chapters 3, 4, and 5, all individuals were in the stage of chronicity (> 6 months post-stroke), which may limit interpretations of behavioural and neurological motor learning outcomes to the acute post-stroke phase, where the majority of rehabilitation therapy is typically delivered. Furthermore, in the studies reported in Chapter 4, we used a relatively homogeneous group of individuals with stroke, as the lesions were all constrained to subcortical regions of the right hemisphere, which is not representative of all post-stroke individuals, limiting broader application of the findings. However, this inclusion criterion (right subcortical lesions) increased our internal validity and power by delimiting our sample, at the expense of external validity that would require recruitment of individuals with a variety of lesion locations. On the retention test (day 7) for Chapter 4, the interaction of GROUP × SEQUENCE, as our measure of functional connectivity for the motor network, did not meet accepted standards of statistical significance (p = 0.053, r = 0.46). Therefore, the differences in functional connectivity between the groups (i.e., stroke and healthy controls) must be interpreted with caution, until more research has been conducted to verify these differences (and the relative strength of the differences). Additionally, in Chapter 5, a larger sample of individuals with stroke within each of the stimulation groups, and within a smaller range of time post-stroke, would have helped to increase the power of this study and prevent type-II errors (falsely accepting the null hypothesis). While lesion location was assessed, we included individuals with both cortical and subcortical lesions (increasing within-group variability). Increasing the sample size to allow for the assessment of individuals with a similar lesion location (i.e., basal ganglia versus thalamus) might help to further categorize responders and non-responders and explain the variability in the response to cTBS over the contralesional M1 and S1. 228 Finally, in Chapters 3, 4, 5, investigation into improvements of motor performance did not entirely account for changes in upper extremity movement as data were based on cursor position. While changes in the control of the cursor may reflect motor learning-related change, it may not directly reflect changes in upper extremity kinematics (Meehan, Dao, et al., 2011). Thus, we were not able to characterize upper extremity compensatory mechanisms employed by individuals post-stroke. Similar in Chapter 4 and Chapter 5, pre- and post fMRI measures, and pre-DTI measures, respectively, were used to quantify and predict changes. However, we cannot describe the neurological mechanisms individuals with stroke employed during skill acquisition, and thus the cause and effect between the intervention and outcome measures. Implications and future directions Further investigation of the performance curve as a way of individualizing practice is warranted in the field of motor skill rehabilitation in individuals with stroke. Based on findings from this thesis, rate of skill acquisition can help to yield methods of motor learning which are individualized to the skills of the learner; these may prove to be more effective than traditional methods of practice based on a one-size fits all model. For example, as discussed in Chapter 3, if the normal or expected rate of skill acquisition for a specific task is known for a clinical population (i.e., based on normative data for rate of skill acquisition), a clinician can determine a suitable practice environment to positively affect the length of time and difficulty needed for achieving good retention of a motor skill. During practice, if the rate of skill acquisition is abnormally slow, the level of difficulty could be decreased by, for example, increasing target size, shortening the distance of a reach task, and/or lowering the amount of CI between tasks. Future research is needed to determine task– and group-specific differences for rate of skill acquisition and motor learning, in order to prescribe appropriate practice paradigms. 229 In a rehabilitation setting, in which individuals with stroke receive interventions to learn new or relearn old motor skills, performance curves could be used to determine: (1) an individualized set number of trials that predict when performance will reach a plateau; and (2) individuals’ capacity for motor learning–related change (i.e., responder, non-responder). Clinicians could use curve fitting information to make informed decisions about subsequent practice sessions and help to encourage future success and learning, which may impact motivation to continue to practice. Based on findings from this thesis, researchers interested in the neurophysiology of motor rehabilitation should begin to investigate differences in biomarkers associated with individual performance. The optimization of practice paradigms can be driven by both behavioural and neurophysiological measures. For example, in addition to the rate of skill acquisition, the level of functional integration of the motor learning network could guide increases in task difficulty or cognitive load of the task. Researchers could integrate neurophysiology measures associated with practice conditions that yield enhanced retention to help individuals with stroke surpass the “false plateaus” in practice. Incorporating neurophysiology measures, such as level of functional connectivity, at the different phases of skill acquisition, could increase the sensitivity to refine the level of challenge in practice, while minimizing the risk of burnout and fatigue associated with tasks that are too difficult for an individual. With advancements in neurophysiological methods, the future of real-time monitoring of neural activation could also be solicited to help improve individualized rehabilitation and the optimization of practice paradigms that translate into long-term motor skill retention and neuroplasticity in individuals with stroke. 230 Conclusion The studies reported in this thesis provide an important contribution to the science of rehabilitation and implications for individualized prognostication or therapy. Each study brings forth a relatively novel behavioural and/or neurological methodological approach to the study of motor learning, particularly in people with stroke. The specific novel contribution of this thesis work include: (1) implementation of a computer-controlled, learner-adapted algorithm with which to manipulate individualized difficulty levels during motor sequence practice; (2) characterization of motor learning after stroke using an exponential curve fitting approach and showing that the rate of skill acquisition is related to performance at retention; (3) use of a constrained principal components analysis (CPCA) to extract a motor learning network in healthy individuals and inform a constrained motor learning network in individuals with stroke, as well as show a relationship between the degree of functional connectivity and motor learning; and (4) identification of a white matter network biomarker that can characterize individuals with chronic stroke who positively respond to cTBS delivered over the sensorimotor, contralesional hemisphere, when paired with motor skill practice. Therefore, the work in this thesis provides a roadmap for modelling behaviour to a performance curve and using MRI to assess a functional and structural motor network to enhance upper extremity motor recovery following stroke. The studies in this thesis further our understanding of behavioural and neurophysiological measures related to optimal motor learning paradigms as well as the delivery of effective treatments in stroke rehabilitation. The use of integrative methods that are driven by behavioural and neurophysiological mechanisms will improve motor recovery in the chronic phase of stroke. It is only through the connection between behaviour and neurophysiology, together with 231 integrated methods emphasizing intra-individual characteristics, that the most effective treatment plans for individuals with stroke can be developed and assessed. 232 Bibliography Abe, M., Hanakawa, T., Takayama, Y., Kuroki, C., Ogawa, S., & Fukuyama, H. (2007). Functional coupling of human prefrontal and premotor areas during cognitive manipulation. J Neurosci, 27(13), 3429-3438. Abrahamse, E. L., Ruitenberg, M. F., de Kleine, E., & Verwey, W. B. (2013). Control of automated behavior: insights from the discrete sequence production task. Front Hum Neurosci, 7, 82. Ackerman, P. L. (1986). Individual differences in information processing: An investigation of intellectual abilities and task performance during practice. Intelligence, 10(2), 101-139. Ackerman, P. L. (1987). Individual differences in skill learning: An integration of psychometric and information processing perspectives. Psychological Bulletin, 102(1), 3-27. Ackerman, P. L. (1988). Determinants of individual differences during skill acquisition: cognitive abilities and information processing. Journal of experimental psychology: General, 117(3), 288. Adams, J. A. (1984). Learning of movement sequences. Psychological Bulletin, 96(1), 3. Adams, J. A. (1987). Historical review and appraisal of research on the learning, retention, and transfer of human motor skills. Psychological Bulletin, 101(1), 41. Al-Ameer, H., & Toole, T. (1993). Combinations of blocked and random practice orders: Benefits to acquisition and retention. Journal of Human Movement Studies, 25(4), 177-192. Albaret, J.-M., & Thon, B. (1998). Differential effects of task complexity on contextual interference in a drawing task. Acta Psychologica, 100(1), 9-24. Alexander, A. L., Lee, J. E., Lazar, M., & Field, A. S. (2007). Diffusion tensor imaging of the brain. Neurotherapeutics, 4(3), 316-329. Aman, J. E., Elangovan, N., Yeh, I.-L., & Konczak, J. (2014). The effectiveness of proprioceptive training for improving motor function: a systematic review. Frontiers in human neuroscience, 8. Ameli, M., Grefkes, C., Kemper, F., Riegg, F. P., Rehme, A. K., Karbe, H., . . . Nowak, D. A. (2009). Differential effects of high-frequency repetitive transcranial magnetic stimulation over ipsilesional primary motor cortex in cortical and subcortical middle cerebral artery stroke. Ann Neurol, 66(3), 298-309. Anderson, J. R. (1982). Acquisition of cognitive skill. Psychological Review, 89(4), 369. Anderson, J. R. (1993). Problem solving and learning. American Psychologist, 48(1), 35. Auriat, A. M., Neva, J. L., Peters, S., Ferris, J. K., & Boyd, L. A. (2015). A Review of Transcranial Magnetic Stimulation and Multimodal Neuroimaging to Characterize Post-Stroke Neuroplasticity. Front Neurol, 6, 226. Axer, H., & Keyserlingk, D. G. (2000). Mapping of fiber orientation in human internal capsule by means of polarized light and confocal scanning laser microscopy. J Neurosci Methods, 94(2), 165-175. Bahrick, H. P., Fitts, P. M., & Briggs, G. E. (1957). Learning curves; facts or artifacts? Psychol Bull, 54(3), 255-268. Barker, A. T., Jalinous, R., & Freeston, I. L. (1985). Non-invasive magnetic stimulation of human motor cortex. The Lancet, 325(8437), 1106-1107. 233 Basser, P. J., & Jones, D. K. (2002). Diffusion-tensor MRI: theory, experimental design and data analysis–a technical review. NMR in Biomedicine, 15(7-8), 456-467. Battig, W., Thompson, R., & Voss, J. (1972). Intratask interference as a source of facilitation in transfer and retention. Topics in learning and performance, 131-159. Battig, W. F. (1979). The flexibility of human memory. Levels of processing and human memory, Lawrence Erlbaum Associates, Hillsdale, NJ, 23-44. Bernhardt, J., Borschmann, K., Boyd, L., Thomas Carmichael, S., Corbett, D., Cramer, S. C., . . . Ward, N. (2016). Moving rehabilitation research forward: Developing consensus statements for rehabilitation and recovery research. Int J Stroke, 11(4), 454-458. Bernhardt, J., Hayward, K. S., Kwakkel, G., Ward, N. S., Wolf, S. L., Borschmann, K., . . . Corbett, D. (2017). Agreed definitions and a shared vision for new standards in stroke recovery research: The Stroke Recovery and Rehabilitation Roundtable taskforce. International Journal of stroke, 12(5), 444-450. Bjork, R. A. (1994). Memory and metamemory considerations in the. Metacognition: Knowing about knowing, 185. Boggio, P. S., Alonso-Alonso, M., Mansur, C. G., Rigonatti, S. P., Schlaug, G., Pascual-Leone, A., & Fregni, F. (2006). Hand function improvement with low-frequency repetitive transcranial magnetic stimulation of the unaffected hemisphere in a severe case of stroke. American journal of physical medicine & rehabilitation, 85(11), 927-930. Book, W. F. (1908). The Psychology of Skill: With Special Reference to Its Acquisition in Typewriting: University of Montana. Borich, M. R., Brown, K. E., & Boyd, L. A. (2014). Motor skill learning is associated with diffusion characteristics of white matter in individuals with chronic stroke. J Neurol Phys Ther, 38(3), 151-160. Borich, M. R., Mang, C., & Boyd, L. A. (2012). Both projection and commissural pathways are disrupted in individuals with chronic stroke: investigating microstructural white matter correlates of motor recovery. BMC neuroscience, 13(1), 1. Borich, M. R., Wadden, K. P., & Boyd, L. A. (2012). Establishing the reproducibility of two approaches to quantify white matter tract integrity in stroke. Neuroimage, 59(3), 2393-2400. Boyd, L., & Winstein, C. (2006). Explicit information interferes with implicit motor learning of both continuous and discrete movement tasks after stroke. Journal of Neurologic Physical Therapy, 30(2), 46-57. Boyd, L. A., Edwards, J. D., Siengsukon, C. S., Vidoni, E. D., Wessel, B. D., & Linsdell, M. A. (2009). Motor sequence chunking is impaired by basal ganglia stroke. Neurobiol Learn Mem, 92(1), 35-44. Boyd, L. A., Hayward, K. S., Ward, N. S., Stinear, C. M., Rosso, C., Fisher, R. J., . . . Carey, L. M. (2017). Biomarkers of stroke recovery: Consensus-based core recommendations from the Stroke Recovery and Rehabilitation Roundtable. International Journal of stroke, 12(5), 480-493. Boyd, L. A., Quaney, B. M., Pohl, P. S., & Winstein, C. J. (2007). Learning implicitly: Effects of task and severity after stroke. Neurorehabilitation and neural repair, 21(5), 444-454. Boyd, L. A., Vidoni, E. D., & Siengsukon, C. F. (2008). Multidimensional motor sequence learning is impaired in older but not younger or middle-aged adults. Physical Therapy, 88(3), 351. 234 Boyd, L. A., Vidoni, E. D., & Wessel, B. D. (2010). Motor learning after stroke: is skill acquisition a prerequisite for contralesional neuroplastic change? Neuroscience letters, 482(1), 21-25. Boyd, L. A., & Winstein, C. J. (2001). Implicit motor-sequence learning in humans following unilateral stroke: the impact of practice and explicit knowledge. Neuroscience letters, 298(1), 65-69. Boyd, L. A., & Winstein, C. J. (2003). Impact of explicit information on implicit motor-sequence learning following middle cerebral artery stroke. Physical Therapy, 83(11), 976-989. Boyd, L. A., & Winstein, C. J. (2004a). Cerebellar stroke impairs temporal but not spatial accuracy during implicit motor learning. Neurorehabilitation and neural repair, 18(3), 134-143. Boyd, L. A., & Winstein, C. J. (2004b). Providing explicit information disrupts implicit motor learning after basal ganglia stroke. Learning & Memory, 11(4), 388-396. Brady, F. (1998). A theoretical and empirical review of the contextual interference effect and the learning of motor skills. Quest, 50(3), 266-293. Brady, F. (2004). Contextual interference: a meta-analytic study. Perceptual and Motor Skills, 99(1), 116-126. Brodie, S. M., Borich, M. R., & Boyd, L. A. (2014). Impact of 5-Hz rTMS over the primary sensory cortex is related to white matter volume in individuals with chronic stroke. Eur J Neurosci, 40(9), 3405-3412. Brodie, S. M., Meehan, S., Borich, M. R., & Boyd, L. A. (2014). 5 Hz repetitive transcranial magnetic stimulation over the ipsilesional sensory cortex enhances motor learning after stroke. Front Hum Neurosci, 8, 143. Brown, S., & Heathcote, A. (2003). Averaging learning curves across and within participants. Behavior Research Methods, Instruments, & Computers, 35(1), 11-21. Burke, E., & Cramer, S. C. (2013). Biomarkers and predictors of restorative therapy effects after stroke. Curr Neurol Neurosci Rep, 13(2), 329. Burke Quinlan, E., Dodakian, L., See, J., McKenzie, A., Le, V., Wojnowicz, M., . . . Cramer, S. C. (2015). Neural function, injury, and stroke subtype predict treatment gains after stroke. Annals of neurology, 77(1), 132-145. Calautti, C., & Baron, J.-C. (2003). Functional neuroimaging studies of motor recovery after stroke in adults a review. Stroke, 34(6), 1553-1566. Calautti, C., Naccarato, M., Jones, P. S., Sharma, N., Day, D. D., Carpenter, A. T., . . . Baron, J. C. (2007). The relationship between motor deficit and hemisphere activation balance after stroke: A 3T fMRI study. Neuroimage, 34(1), 322-331. Carey, J. R., Deng, H., Gillick, B. T., Cassidy, J. M., Anderson, D. C., Zhang, L., & Thomas, W. (2014). Serial treatments of primed low-frequency rTMS in stroke: Characteristics of responders vs. nonresponders. Restor Neurol Neurosci, 32(2), 323-335. Carey, L. M., Abbott, D. F., Egan, G. F., Bernhardt, J., & Donnan, G. A. (2005). Motor impairment and recovery in the upper limb after stroke: behavioral and neuroanatomical correlates. Stroke, 36(3), 625-629. Caria, A., Weber, C., Brötz, D., Ramos, A., Ticini, L. F., Gharabaghi, A., . . . Birbaumer, N. (2011). Chronic stroke recovery after combined BCI training and physiotherapy: a case report. Psychophysiology, 48(4), 578-582. 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. 235 Cassidy, J. M., Chu, H., Anderson, D. C., Krach, L. E., Snow, L., Kimberley, T. J., & Carey, J. R. (2015). A Comparison of Primed Low-frequency Repetitive Transcranial Magnetic Stimulation Treatments in Chronic Stroke. Brain Stimul, 8(6), 1074-1084. Cattel, R. B. (1966). The scree test for the number of factors. Multivariate Behavioral Research, 1(2), 245-276. Cattel, R. B., & Vogelmann, S. (1977). A comprehensive trial of the scree and kg criteria for determining the number of factors. Multivariate Behavioral Research, 12(3), 289-325. Chambaron, S., Berberian, B., Delbecque, L., Ginhac, D., & Cleeremans, A. (2009). Implicit motor learning in discrete and continuous tasks: Toward a possible account of discrepant results. Handbook of motor skills: Development, impairment, and therapy, 139-155. Chambaron, S., Ginhac, D., Ferrel-Chapus, C., & Perruchet, P. (2006). Implicit learning of a repeated segment in continuous tracking: A reappraisal. The Quarterly Journal of Experimental Psychology, 59(05), 845-854. Chen, R., Classen, J., Gerloff, C., Celnik, P., Wassermann, E., Hallett, M., & Cohen, L. (1997). Depression of motor cortex excitability by low-frequency transcranial magnetic stimulation. Neurology, 48(5), 1398-1403. Chen, R., Yung, D., & Li, J.-Y. (2003). Organization of ipsilateral excitatory and inhibitory pathways in the human motor cortex. J Neurophysiol, 89(3), 1256-1264. Choi, Y., Gordon, J., Park, H., & Schweighofer, N. (2011). Feasibility of the adaptive and automatic presentation of tasks (ADAPT) system for rehabilitation of upper extremity function post-stroke. Journal of neuroengineering and rehabilitation, 8(1), 1. Choi, Y., Qi, F., Gordon, J., & Schweighofer, N. (2008). Performance-based adaptive schedules enhance motor learning. Journal of motor behavior, 40(4), 273-280. Clerget, E., Badets, A., Duque, J., & Olivier, E. (2011). Role of Broca’s area in motor sequence programming: A cTBS study. Neuroreport, 22(18), 965-969. Cohen, R., & Sternad, D. (2009). Variability in motor learning: relocating, channeling and reducing noise. Experimental Brain Research, 193(1), 69-83. Cousineau, D., Helie, S., & Lefebvre, C. (2003). Testing curvatures of learning functions on individual trial and block average data. Behav Res Methods Instrum Comput, 35(4), 493-503. Cousineau, D., & Lacroix, G. L. (2006). Getting parameters from learning data. Tutorials in quantitative methods in psychology, 2, 77-83. Coynel, D., Marrelec, G., Perlbarg, V., Pelegrini-Issac, M., Van de Moortele, P. F., Ugurbil, K., . . . Lehericy, S. (2010). Dynamics of motor-related functional integration during motor sequence learning. Neuroimage, 49(1), 759-766. Cronbach, L. J. (1963). Educational psychology: Harcourt, Brace & World. Cross, E. S., Schmitt, P. J., & Grafton, S. T. (2007). Neural substrates of contextual interference during motor learning support a model of active preparation. Journal of Cognitive Neuroscience, 19(11), 1854-1871. Crum, R. M., Anthony, J. C., Bassett, S. S., & Folstein, M. F. (1993). Population-based norms for the Mini-Mental State Examination by age and educational level. Jama, 269(18), 2386-2391. Dayan, E., & Cohen, L. G. (2011). Neuroplasticity subserving motor skill learning. Neuron, 72(3), 443-454. 236 Debaere, F., Wenderoth, N., Sunaert, S., Van Hecke, P., & Swinnen, S. P. (2004). Changes in brain activation during the acquisition of a new bimanual coodination task. Neuropsychologia, 42(7), 855-867. Del Rey, P., Wughalter, E. H., & Whitehurst, M. (1982). The effects of contextual interference on females with varied experience in open sport skills. Research quarterly for exercise and sport, 53(2), 108-115. DeLeeuw, K. E., & Mayer, R. E. (2008). A comparison of three measures of cognitive load: Evidence for separable measures of intrinsic, extraneous, and germane load. Journal of Educational Psychology, 100(1), 223-234. Demirtas-Tatlidede, A., Alonso-Alonso, M., Shetty, R. P., Ronen, I., Pascual-Leone, A., & Fregni, F. (2015). Long-term effects of contralesional rTMS in severe stroke: safety, cortical excitability, and relationship with transcallosal motor fibers. NeuroRehabilitation, 36(1), 51-59. Deuschl, G., Toro, C., Zeffiro, T., Massaquoi, S., & Hallett, M. (1996). Adaptation motor learning of arm movements in patients with cerebellar disease. J Neurol Neurosurg Psychiatry, 60(5), 515-519. Di Pino, G., Pellegrino, G., Assenza, G., Capone, F., Ferreri, F., Formica, D., . . . Rothwell, J. C. (2014). Modulation of brain plasticity in stroke: a novel model for neurorehabilitation. Nature Reviews Neurology, 10(10), 597-608. Dite, W., Langford, Z. N., Cumming, T. B., Churilov, L., Blennerhassett, J. M., & Bernhardt, J. (2015). A Phase 1 exercise dose escalation study for stroke survivors with impaired walking. Int J Stroke. Dobkin, B. H. (2004). Strategies for stroke rehabilitation. The Lancet Neurology, 3(9), 528-536. Doyon, J., Bellec, P., Amsel, R., Penhune, V., Monchi, O., Carrier, J., . . . Benali, H. (2009). Contributions of the basal ganglia and functionally related brain structures to motor learning. Behav Brain Res, 199(1), 61-75. Doyon, J., & Benali, H. (2005). Reorganization and plasticity in the adult brain during learning of motor skills. Current opinion in neurobiology, 15(2), 161-167. Doyon, J., Penhune, V., & Ungerleider, L. G. (2003). Distinct contribution of the cortico-striatal and cortico-cerebellar systems to motor skill learning. Neuropsychologia, 41(3), 252-262. Doyon, J., Song, A. W., Karni, A., Lalonde, F., Adams, M. M., & Ungerleider, L. G. (2002). Experience-dependent changes in cerebellar contributions to motor sequence learning. Proceedings of the National Academy of Sciences, 99(2), 1017-1022. Doyon, J., Ungerleider, L., Squire, L., & Schacter, D. (2002). Functional anatomy of motor skill learning. Neuropsychology of memory, 3, 225-238. Emara, T., El Nahas, N., Elkader, H. A., Ashour, S., & El Etrebi, A. (2009). MRI can predict the response to therapeutic repetitive transcranial magnetic stimulation (rTMS) in stroke patients. Journal of vascular and interventional neurology, 2(2), 163. Enomoto, H., Ugawa, Y., Hanajima, R., Yuasa, K., Mochizuki, H., Terao, Y., . . . Kanazawa, I. (2001). Decreased sensory cortical excitability after 1 Hz rTMS over the ipsilateral primary motor cortex. Clinical Neurophysiology, 112(11), 2154-2158. Ericsson, K. A., Krampe, R. T., & Tesch-Römer, C. (1993). The role of deliberate practice in the acquisition of expert performance. Psychological Review, 100(3), 363. Eversbusch, A., & Grantcharov, T. (2004). Learning curves and impact of psychomotor training on performance in simulated colonoscopy: a randomized trial using a virtual reality 237 endoscopy trainer. Surgical Endoscopy And Other Interventional Techniques, 18(10), 1514-1518. Feldman, L. S., Cao, J., Andalib, A., Fraser, S., & Fried, G. M. (2009). A method to characterize the learning curve for performance of a fundamental laparoscopic simulator task: defining “learning plateau” and “learning rate”. Surgery, 146(2), 381-386. Field, A. (2009). Discovering statistics using SPSS: Sage publications. Fitts, P. M. (1966). Cognitive aspects of information processing: III. Set for speed versus accuracy. Journal of experimental psychology, 71(6), 849. Fitts, P. M., & Melton, A. W. (1964). Perceptual-motor skill learning. Categories of human learning, 47, 381-391. Fitts, P. M., & Posner, M. I. (1967). Human Performance: Brooks/Cole Publishing Company. Fitzgerald, P. B., Brown, T. L., Daskalakis, Z. J., Chen, R., & Kulkarni, J. (2002). Intensity-dependent effects of 1 Hz rTMS on human corticospinal excitability. Clinical Neurophysiology, 113(7), 1136-1141. Flamme, C., Stukenborg-Colsman, C., & Wirth, C. (2005). Evaluation of the learning curves associated with uncemented primary total hip arthroplasty depending on the experience of the surgeon. Hip international: the journal of clinical and experimental research on hip pathology and therapy, 16(3), 191-197. Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res, 12(3), 189-198. Foltys, H., Krings, T., Meister, I. G., Sparing, R., Boroojerdi, B., Thron, A., & Topper, R. (2003). Motor representation in patients rapidly recovering after stroke: a functional magnetic resonance imaging and transcranial magnetic stimulation study. Clin Neurophysiol, 114(12), 2404-2415. Fox, K. C., Dixon, M. L., Nijeboer, S., Girn, M., Floman, J. L., Lifshitz, M., . . . Christoff, K. (2016). Functional neuroanatomy of meditation: A review and meta-analysis of 78 functional neuroimaging investigations. Neuroscience & Biobehavioral Reviews, 65, 208-228. Fregni, F., Boggio, P. S., Valle, A. C., Rocha, R. R., Duarte, J., Ferreira, M. J., . . . Riberto, M. (2006). A sham-controlled trial of a 5-day course of repetitive transcranial magnetic stimulation of the unaffected hemisphere in stroke patients. Stroke, 37(8), 2115-2122. French, K. E., Rink, J. E., & Werner, P. H. (1990). Effects of contextual interference on retention of three volleyball skills. Perceptual and Motor Skills, 71(1), 179-186. Frensch, P. A., Lin, J., & Buchner, A. (1998). Learning versus behavioral expression of the learned: The effects of a secondary tone-counting task on implicit learning in the serial reaction task. Psychological Research, 61(2), 83-98. Friston, K. J. (1997). Eigenimages and multivariate analyses. Human Brain Function. Fugl-Meyer, A. R., Jaasko, L., Leyman, I., Olsson, S., & Steglind, S. (1975). The post-stroke hemiplegic patient. 1. a method for evaluation of physical performance. Scand J Rehabil Med, 7(1), 13-31. Gallahue, D., & Ozmun, J. (2012). Understanding motor development: Infants, children, adolescents, adults. In J. Goodway (Ed.): New York, NY: McGraw-Hill. Gamst, G., Meyers, L. S., & Guarino, A. (2008). Analysis of variance designs: A conceptual and computational approach with SPSS and SAS: Cambridge University Press. 238 Gillard, J., Papadakis, N., Martin, K., Price, C., Warburton, E., Antoun, N., . . . Pickard, J. (2001). MR diffusion tensor imaging of white matter tract disruption in stroke at 3 T. The British journal of radiology, 74(883), 642-647. Glautier, S. (2013). Revisiting the learning curve (once again). Frontiers in psychology, 4(982). Gobel, E. W., Parrish, T. B., & Reber, P. J. (2011). Neural correlates of skill acquisition: decreased cortical activity during a serial interception sequence learning task. Neuroimage, 58(4), 1150-1157. Grafton, S. T., Hazeltine, E., & Ivry, R. B. (2002). Motor sequence learning with the nondominant left hand. Experimental Brain Research, 146(3), 369-378. Granziera, C., Daducci, A., Meskaldji, D. E., Roche, A., Maeder, P., Michel, P., . . . Thiran, J.-P. (2012). A new early and automated MRI-based predictor of motor improvement after stroke. Neurology, 79(1), 39-46. Gray, C. D., & Kinnear, P. R. (2012). IBM SPSS statistics 19 made simple: Psychology Press. Green, T. D., & Flowers, J. H. (1991). Implicit versus explicit learning processes in a probabilistic, continuous fine-motor catching task. J Mot Behav, 23(4), 293-300. Grefkes, C., Nowak, D. A., Eickhoff, S. B., Dafotakis, M., Küst, J., Karbe, H., & Fink, G. R. (2008). Cortical connectivity after subcortical stroke assessed with functional magnetic resonance imaging. Annals of neurology, 63(2), 236-246. Grefkes, C., Nowak, D. A., Wang, L. E., Dafotakis, M., Eickhoff, S. B., & Fink, G. R. (2010). Modulating cortical connectivity in stroke patients by rTMS assessed with fMRI and dynamic causal modeling. Neuroimage, 50(1), 233-242. Grefkes, C., & Ward, N. S. (2013). Cortical Reorganization After Stroke How Much and How Functional? The Neuroscientist, 1073858413491147. Guadagnoli, M., Holcomb, W., & Weber, T. (1999). The relationship between contextual interference effects and performer expertise on the learning of a putting task. Journal of Human Movement Studies, 37(1), 19-36. Guadagnoli, M., & Lee, T. (2004). Challenge point: a framework for conceptualizing the effects of various practice conditions in motor learning. Journal of motor behavior, 36(2), 212-224. Gupta, R. K., Saksena, S., Hasan, K. M., Agarwal, A., Haris, M., Pandey, C. M., & Narayana, P. A. (2006). Focal Wallerian degeneration of the corpus callosum in large middle cerebral artery stroke: serial diffusion tensor imaging. Journal of Magnetic Resonance Imaging, 24(3), 549-555. Haibach, P., Reid, G., & Collier, D. (2011). Motor learning and development: Human Kinetics. Hanlon, R. E. (1996). Motor learning following unilateral stroke. Archives of physical medicine and rehabilitation, 77(8), 811-815. Harbourne, R. T., & Stergiou, N. (2009). Movement variability and the use of nonlinear tools: principles to guide physical therapist practice. Physical Therapy, 89(3), 267. Hardwick, R. M., Rajan, V. A., Bastian, A. J., Krakauer, J. W., & Celnik, P. A. (2016). Motor Learning in Stroke: Trained Patients Are Not Equal to Untrained Patients With Less Impairment. Neurorehabilitation and neural repair, 1545968316675432. Hayward, K., Schmidt, J., Lohse, K., Bernhart, J., Boyd, LA. (2016). What is the capacity of the damaged brain to recover in people with severe upper limb impairment. Journal of Neurology, Neurosurgery and Psychiatry, In review. Heathcote, A., & Brown, S. (2004). Beyond curve fitting? Comment on Liu, Mayer-Kress, and Newell (2003). Journal of motor behavior, 36(2), 225-232. 239 Heathcote, A., Brown, S., & Mewhort, D. (2000). The power law repealed: The case for an exponential law of practice. Psychonomic Bulletin & Review, 7(2), 185-207. Hebert, E. P., Landin, D., & Solmon, M. A. (1996). Practice schedule effects on the performance and learning of low-and high-skilled students: An applied study. Research quarterly for exercise and sport, 67(1), 52-58. Hernandez, J., Bann, S., Munz, Y., Moorthy, K., Datta, V., Martin, S., . . . Rockall, T. (2004). Qualitative and quantitative analysis of the learning curve of a simulated surgical task on the da Vinci system. Surgical Endoscopy And Other Interventional Techniques, 18(3), 372-378. Hikosaka, O., Nakamura, K., Sakai, K., & Nakahara, H. (2002). Central mechanisms of motor skill learning. Current opinion in neurobiology, 12(2), 217-222. Hirano, M., Kubota, S., Tanabe, S., Koizume, Y., & Funase, K. (2015). Interactions among learning stage, retention, and primary motor cortex excitability in motor skill learning. Brain Stimulation, 8(6), 1195-1204. Hodges, N. J., Lohse, K. R., Wilson, A., Lim, S. B., & Mulligan, D. (2014). Exploring the Dynamic Nature of Contextual Interference: Previous Experience Affects Current Practice But Not Learning. Journal of motor behavior, 46(6), 455-467. Hodics, T. M., Nakatsuka, K., Upreti, B., Alex, A., Smith, P. S., & Pezzullo, J. C. (2012). Wolf Motor Function Test for characterizing moderate to severe hemiparesis in stroke patients. Arch Phys Med Rehabil, 93(11), 1963-1967. Hong, J. H., Son, S. M., & Jang, S. H. (2010). Somatotopic location of corticospinal tract at pons in human brain: a diffusion tensor tractography study. Neuroimage, 51(3), 952-955. HSF. (2015). Heart and Stroke Foundation 2015 Stroke Report. Retrieved from Stroke Best Practices: Hsu, W.-Y., Cheng, C.-H., Liao, K.-K., Lee, I.-H., & Lin, Y.-Y. (2012). Effects of repetitive transcranial magnetic stimulation on motor functions in patients with stroke. Stroke, 43(7), 1849-1857. Huang, Y.-Z., Edwards, M. J., Rounis, E., Bhatia, K. P., & Rothwell, J. C. (2005). Theta burst stimulation of the human motor cortex. Neuron, 45(2), 201-206. Huang, Y.-Z., & Rothwell, J. C. (2004). The effect of short-duration bursts of high-frequency, low-intensity transcranial magnetic stimulation on the human motor cortex. Clinical Neurophysiology, 115(5), 1069-1075. Huang, Y.-Z., Rothwell, J. C., Lu, C.-S., Wang, J., Weng, Y.-H., Lai, S.-C., . . . Chen, R.-S. (2009). The effect of continuous theta burst stimulation over premotor cortex on circuits in primary motor cortex and spinal cord. Clinical Neurophysiology, 120(4), 796-801. Hummel, F. C., & Cohen, L. G. (2006). Non-invasive brain stimulation: a new strategy to improve neurorehabilitation after stroke? Lancet Neurol, 5(8), 708-712. Iezzi, E., Suppa, A., Conte, A., Agostino, R., Nardella, A., & Berardelli, A. (2010). Theta-burst stimulation over primary motor cortex degrades early motor learning. European Journal of Neuroscience, 31(3), 585-592. Ioffe, M. E., Ustinova, K. I., Chernikova, L. A., & Kulikov, M. A. (2006). Supervised learning of postural tasks in patients with poststroke hemiparesis, Parkinson's disease or cerebellar ataxia. Exp Brain Res, 168(3), 384-394. Isa, T., Ohki, Y., Alstermark, B., Pettersson, L.-G., & Sasaki, S. (2007). Direct and indirect cortico-motoneuronal pathways and control of hand/arm movements. Physiology, 22(2), 145-152. 240 Jang, S. H. (2009). The role of the corticospinal tract in motor recovery in patients with a stroke: a review. NeuroRehabilitation, 24(3), 285-290. Jang, S. H. (2010). Prediction of motor outcome for hemiparetic stroke patients using diffusion tensor imaging: a review. NeuroRehabilitation, 27(4), 367-372. Jang, S. H., Cho, S.-H., Kim, Y.-H., Han, B. S., Byun, W. M., Son, S.-M., . . . Lee, S. J. (2005). Diffusion anisotrophy in the early stages of stroke can predict motor outcome. Restorative neurology and neuroscience, 23(1), 11-17. Jang, S. H., You, S. H., Kwon, Y. H., Hallett, M., Lee, M. Y., & Ahn, S. H. (2005). Cortical reorganization associated lower extremity motor recovery as evidenced by functional MRI and diffusion tensor tractography in a stroke patient. Restor Neurol Neurosci, 23(5-6), 325-329. Johansen-Berg, H., Rushworth, M. F., Bogdanovic, M. D., Kischka, U., Wimalaratna, S., & Matthews, P. M. (2002). The role of ipsilateral premotor cortex in hand movement after stroke. Proceedings of the National Academy of Sciences, 99(22), 14518-14523. Johansen-Berg, H., Dawes, H., Guy, C., Smith, S. M., Wade, D. T., & Matthews, P. M. (2002). Correlation between motor improvements and altered fMRI activity after rehabilitative therapy. Brain, 125(12), 2731-2742. Johansson, B. B. (2000). Brain plasticity and stroke rehabilitation. Stroke, 31(1), 223-230. Jones, T. A. (2017). Motor compensation and its effects on neural reorganization after stroke. Nature Reviews Neuroscience, 18(5), 267-280. Kal, E., Winters, M., van der Kamp, J., Houdijk, H., Groet, E., van Bennekom, C., & Scherder, E. (2016). Is Implicit Motor Learning Preserved after Stroke? A Systematic Review with Meta-Analysis. PLoS One, 11(12), e0166376. Kaplan, P. E., Cailliet, R., & Kaplan, C. P. (2003). Rehabilitation of stroke: Butterworth-Heinemann. Karni, A., Meyer, G., Jezzard, P., & Adams, M. M. (1995). Functional MRI evidence for adult motor cortex plasticity during motor skill learning. Nature, 377(6545), 155. Karni, A., Meyer, G., Rey-Hipolito, C., Jezzard, P., Adams, M. M., Turner, R., & Ungerleider, L. G. (1998). The acquisition of skilled motor performance: fast and slow experience-driven changes in primary motor cortex. Proc Natl Acad Sci U S A, 95(3), 861-868. Keele, S. W., Ivry, R., Mayr, U., Hazeltine, E., & Heuer, H. (2003). The cognitive and neural architecture of sequence representation. Psychological Review, 110(2), 316. Keetch, K. M., & Lee, T. D. (2007). The effect of self-regulated and experimenter-imposed practice schedules on motor learning for tasks of varying difficulty. Research quarterly for exercise and sport, 78(5), 476-486. Kim, B., & Winstein, C. (2017). Can neurological biomarkers of brain impairment be used to predict poststroke motor recovery? A systematic review. Neurorehabilitation and neural repair, 31(1), 3-24. Kim, Y.-H., You, S. H., Ko, M.-H., Park, J.-W., Lee, K. H., Jang, S. H., . . . Hallett, M. (2006). Repetitive transcranial magnetic stimulation–induced corticomotor excitability and associated motor skill acquisition in chronic stroke. Stroke, 37(6), 1471-1476. Kirsch, W., & Hoffmann, J. (2012). Stimulus-dependent modulation of perceptual and motor learning in a serial reaction time task. Advances in cognitive psychology, 8(2), 155-164. Kitago, T., & Krakauer, J. W. (2013). Motor learning principles for neurorehabilitation. Handb Clin Neurol, 110, 93-103. 241 Kleim, J. A., Hogg, T. M., VandenBerg, P. M., Cooper, N. R., Bruneau, R., & Remple, M. (2004). Cortical synaptogenesis and motor map reorganization occur during late, but not early, phase of motor skill learning. J Neurosci, 24(3), 628-633. Krakauer, J. W. (2006). Motor learning: its relevance to stroke recovery and neurorehabilitation. Current Opinion in Neurology, 19(1), 84-90. Krakauer, J. W., Carmichael, S. T., Corbett, D., & Wittenberg, G. F. (2012). Getting neurorehabilitation right: what can be learned from animal models? Neurorehabilitation and neural repair, 26(8), 923-931. Krakauer, J. W., Pine, Z. M., Ghilardi, M. F., & Ghez, C. (2000). Learning of visuomotor transformations for vectorial planning of reaching trajectories. J Neurosci, 20(23), 8916-8924. Kumar, P., Kathuria, P., Nair, P., & Prasad, K. (2016). Prediction of upper limb motor recovery after subacute ischemic stroke using diffusion tensor imaging: a systematic review and meta-analysis. Journal of stroke, 18(1), 50. Kusano, Y., Seguchi, T., Horiuchi, T., Kakizawa, Y., Kobayashi, T., Tanaka, Y., . . . Hongo, K. (2009). Prediction of functional outcome in acute cerebral hemorrhage using diffusion tensor imaging at 3T: a prospective study. American Journal of Neuroradiology, 30(8), 1561-1565. Kwon, H. G., Lee, D. G., Son, S. M., Byun, W. M., Hong, C. P., Lee, D. H., . . . Jang, S. H. (2011). Identification of the anterior corticospinal tract in the human brain using diffusion tensor imaging. Neurosci Lett, 505(3), 238-241. Lai, S.-M., Studenski, S., Duncan, P. W., & Perera, S. (2002). Persisting consequences of stroke measured by the Stroke Impact Scale. Stroke, 33(7), 1840-1844. Lakhani, B., Borich, M. R., Jackson, J. N., Wadden, K. P., Peters, S., Villamayor, A., . . . Boyd, L. A. (2016). Motor skill acquisition promotes human brain myelin plasticity. Neural plasticity, 2016. Lane, N. E. (1987). Skill acquisition rates and patterns: Issues and training implications: Springer Science & Business Media. Lang, C. E., & Bastian, A. J. (1999). Cerebellar subjects show impaired adaptation of anticipatory EMG during catching. J Neurophysiol, 82(5), 2108-2119. Lang, C. E., & Bastian, A. J. (2001). Additional somatosensory information does not improve cerebellar adaptation during catching. Clin Neurophysiol, 112(5), 895-907. Lang, C. E., Lohse, K. R., & Birkenmeier, R. L. (2015). Dose and timing in neurorehabilitation: prescribing motor therapy after stroke. Curr Opin Neurol, 28(6), 549-555. Langhorne, P., Bernhardt, J., & Kwakkel, G. (2011). Stroke rehabilitation. Lancet, 377(9778), 1693-1702. Langhorne, P., Coupar, F., & Pollock, A. (2009). Motor recovery after stroke: a systematic review. The Lancet Neurology, 8(8), 741-754. Lavigne, K. M., Metzak, P. D., & Woodward, T. S. (2015). Functional brain networks underlying detection and integration of disconfirmatory evidence. Neuroimage, 112, 138-151. Lee, J. H., Kim, S. B., Lee, K. W., Kim, M. A., Lee, S. J., & Choi, S. J. (2015). Factors associated with upper extremity motor recovery after repetitive transcranial magnetic stimulation in stroke patients. Annals of rehabilitation medicine, 39(2), 268-276. Lee, T. D. (2012). Contextual interference. In N. Hodges & M. A. Williams (Eds.), Skill acquisition in sport: Research, theory and practice (pp. 79-93): Taylor & Francis. 242 Lee, T. D., & Magill, R. A. (1983). The locus of contextual interference in motor-skill acquisition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(4), 730-746. Lee, T. D., & Simon, D. A. (2004). Contextual interference. In N. Hodges & M. A. Williams (Eds.), Skill acquisition in sport: Research, theory and practice (pp. 29-44). Lee, T. D. S., D.A. (2004). Contextual interference. In M. A. Williams, & Hodges, N.J. (Ed.), SKill acquisition in sport: Research, theory and practice (pp. 29-44). London: Routledge. Leemans, A., & Jones, D. K. (2009). The B-matrix must be rotated when correcting for subject motion in DTI data. Magn Reson Med, 61(6), 1336-1349. Lefebvre, S., Dricot, L., Laloux, P., Gradkowski, W., Desfontaines, P., Evrard, F., . . . Vandermeeren, Y. (2015). Neural substrates underlying stimulation-enhanced motor skill learning after stroke. Brain, 138(1), 149-163. Levin, M. F., Kleim, J. A., & Wolf, S. L. (2009). What do motor "recovery" and "compensation" mean in patients following stroke? Neurorehabil Neural Repair, 23(4), 313-319. Lewthwaite, R., & Wulf, G. (2017). Optimizing motivation and attention for motor performance and learning. Current Opinion in Psychology. Li Voti, P., Conte, A., Rocchi, L., Bologna, M., Khan, N., Leodori, G., & Berardelli, A. (2014). Cerebellar continuous theta-burst stimulation affects motor learning of voluntary arm movements in humans. European Journal of Neuroscience, 39(1), 124-131. Li, Y., Wu, P., Liang, F., & Huang, W. (2015). The microstructural status of the corpus callosum is associated with the degree of motor function and neurological deficit in stroke patients. PLoS One, 10(4), e0122615. Lin, C.-H. J., Chiang, M.-C., Wu, A. D., Iacoboni, M., Udompholkul, P., Yazdanshenas, O., & Knowlton, B. J. (2012a). Age related differences in the neural substrates of motor sequence learning after interleaved and repetitive practice. Neuroimage, 62(3), 2007-2020. Lin, C.-H. J., Chiang, M.-C., Wu, A. D., Iacoboni, M., Udompholkul, P., Yazdanshenas, O., & Knowlton, B. J. (2012b). Enhanced motor learning in older adults is accompanied by increased bilateral frontal and fronto-parietal connectivity. Brain connectivity, 2(2), 56-68. Lindenberg, R., Renga, V., Zhu, L. L., Nair, D., & Schlaug, G. (2010). Bihemispheric brain stimulation facilitates motor recovery in chronic stroke patients. Neurology, 75(24), 2176-2184. Lindenberg, R., & Seitz, R. J. (2012). Impact of white matter damage after stroke Neuroimaging-Methods (pp. 233-244): InTech. Lloyd-Jones, D., Adams, R., Carnethon, M., De Simone, G., Ferguson, T. B., Flegal, K., . . . Greenlund, K. (2009). Heart disease and stroke statistics—2009 update. Circulation, 119(3), e21-e181. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95(4), 492. Logothetis, N. K. (2003). MR imaging in the non-human primate: studies of function and of dynamic connectivity. Current opinion in neurobiology, 13(5), 630-642. Lohse, K. R., Lang, C. E., & Boyd, L. A. (2014). Is more better? Using metadata to explore dose-response relationships in stroke rehabilitation. Stroke, 45(7), 2053-2058. 243 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, 130-141. Luft, A. R., McCombe-Waller, S., Whitall, J., Forrester, L. W., Macko, R., Sorkin, J. D., . . . Hanley, D. F. (2004). Repetitive bilateral arm training and motor cortex activation in chronic stroke: a randomized controlled trial. Jama, 292(15), 1853-1861. Lundy-Ekman, L. (1998). Neuroscience: fundamentals for rehabilitation. Philadelphia: WB Saunders. Lyle, R. C. (1981). A performance test for assessment of upper limb function in physical rehabilitation treatment and research. International Journal of Rehabilitation Research, 4(4), 483-492. Ma, L., Wang, B., Narayana, S., Hazeltine, E., Chen, X., Robin, D. A., . . . Xiong, J. (2010). Changes in regional activity are accompanied with changes in inter-regional connectivity during 4 weeks motor learning. Brain Research, 1318, 64-76. Maeda, F., Keenan, J. P., Tormos, J. M., Topka, H., & Pascual-Leone, A. (2000). Interindividual variability of the modulatory effects of repetitive transcranial magnetic stimulation on cortical excitability. Experimental Brain Research, 133(4). Maeda, T., Ishizaki, K.-i., & Yura, S. (2005). Can diffusion tensor imaging predict the functional outcome of supra-tentorial stroke? Brain and nerve, 57(1), 27-32. Magill, R. A., & Hall, K. G. (1990). A review of the contextual interference effect in motor skill acquisition. Human movement science, 9(3), 241-289. Mang, C. S., Borich, M. R., Brodie, S. M., Brown, K. E., Snow, N. J., Wadden, K. P., & Boyd, L. A. (2015). Diffusion imaging and transcranial magnetic stimulation assessment of transcallosal pathways in chronic stroke. Clin Neurophysiol, 126(10), 1959-1971. Mang, C. S., Snow, N. J., Campbell, K. L., Ross, C. J., & Boyd, L. A. (2014). A single bout of high-intensity aerobic exercise facilitates response to paired associative stimulation and promotes sequence-specific implicit motor learning. Journal of applied physiology, 117(11), 1325-1336. Mang, C. S., Snow, N. J., Wadden, K. P., Campbell, K. L., & Boyd, L. A. (2016). High-Intensity Aerobic Exercise Enhances Motor Memory Retrieval. Med Sci Sports Exerc, 48(12), 2477-2486. Martin, T., Keating, J., Goodkin, H., Bastian, A., & Thach, W. (1996). Throwing while looking through prisms. Brain, 119(4), 1183-1198. McLaughlin, A. C., Rogers, W. A., & Fisk, A. D. (2006). How effective feedback for training depends on learner resources and task demands. Paper presented at the Proceedings of the Human Factors and Ergonomics Society Annual Meeting. Meehan, S. K., Dao, E., Linsdell, M. A., & Boyd, L. A. (2011). Continuous theta burst stimulation over the contralesional sensory and motor cortex enhances motor learning post-stroke. Neurosci Lett, 500(1), 26-30. Meehan, S. K., Linsdell, M. A., Handy, T. C., & Boyd, L. A. (2011). Interhemispheric enhancement of somatosensory cortical excitability through contralateral repetitive transcranial magnetic stimulation. Clin Neurophysiol, 122(8), 1637-1644. Meehan, S. K., Randhawa, B., Wessel, B., & Boyd, L. A. (2011). Implicit sequence-specific motor learning after subcortical stroke is associated with increased prefrontal brain activations: an fMRI study. Hum Brain Mapp, 32(2), 290-303. Melton, A. W. (2014). Categories of Human Learning: Elsevier Science. 244 Mendis, S. (2013). Stroke disability and rehabilitation of stroke: World Health Organization perspective. International Journal of stroke, 8(1), 3-4. Metzak, P., Feredoes, E., Takane, Y., Wang, L., Weinstein, S., Cairo, T., . . . Woodward, T. S. (2011). Constrained principal component analysis reveals functionally connected load-dependent networks involved in multiple stages of working memory. Hum Brain Mapp, 32(6), 856-871. Metzak, P. D., Riley, J. D., Wang, L., Whitman, J. C., Ngan, E. T., & Woodward, T. S. (2012). Decreased efficiency of task-positive and task-negative networks during working memory in schizophrenia. Schizophr Bull, 38(4), 803-813. Middleton, F. A., & Strick, P. L. (1994). Anatomical evidence for cerebellar and basal ganglia involvement in higher cognitive function. Science, 266(5184), 458-461. Middleton, F. A., & Strick, P. L. (1997). Cerebellar output channels. Int Rev Neurobiol, 41, 61-82. Modabber, M., Neva, J., Gill, M., Budge, I., & Henriques, D. (2008). Learning and retaining visuomotor adaptation across time. Journal of Vision, 8(6), 610-610. Mukherjee, P., Berman, J., Chung, S., Hess, C., & Henry, R. (2008). Diffusion tensor MR imaging and fiber tractography: theoretic underpinnings. American Journal of Neuroradiology, 29(4), 632-641. Murase, N., Duque, J., Mazzocchio, R., & Cohen, L. G. (2004). Influence of interhemispheric interactions on motor function in chronic stroke. Annals of neurology, 55(3), 400-409. Muratori, L. M., Lamberg, E. M., Quinn, L., & Duff, S. V. (2013). Applying principles of motor learning and control to upper extremity rehabilitation. Journal of Hand Therapy, 26(2), 94-103. Nesselroade, J. R., & Salthouse, T. A. (2004). Methodological and theoretical implications of intraindividual variability in perceptual-motor performance. J Gerontol B Psychol Sci Soc Sci, 59(2), P49-55. Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of practice. Cognitive skills and their acquisition, 1, 1-55. Newell, K. M. (1991). Motor skill acquisition. Annu Rev Psychol, 42, 213-237. Newell, K. M., Liu, Y.-T., & Mayer-Kress, G. (2001). Time scales in motor learning and development. Psychological Review, 108(1), 57. Nichols-Larsen, D. S., Clark, P., Zeringue, A., Greenspan, A., & Blanton, S. (2005). Factors influencing stroke survivors’ quality of life during subacute recovery. Stroke, 36(7), 1480-1484. Nimsky, C., Bauer, M., & Carl, B. (2016). Merits and Limits of Tractography Techniques for the Uninitiated Advances and technical standards in neurosurgery (pp. 37-60): Springer. Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of learning: Evidence from performance measures. Cognitive psychology, 19(1), 1-32. Norman, K. A., Polyn, S. M., Detre, G. J., & Haxby, J. V. (2006). Beyond mind-reading: multi-voxel pattern analysis of fMRI data. Trends Cogn Sci, 10(9), 424-430. Nudo, R. J. (2013). Recovery after brain injury: mechanisms and principles. Frontiers in human neuroscience, 7, 887. Nudo, R. J., Wise, B. M., SiFuentes, F., & Milliken, G. W. (1996). Neural substrates for the effects of rehabilitative training on motor recovery after ischemic infarct. Science, 272(5269), 1791. 245 Page, S. J., Schmid, A., & Harris, J. E. (2012). Optimizi
UBC Theses and Dissertations
The behavioural and neural science of motor skill learning in healthy individuals and people with stroke Wadden, Katie P. 2017
Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
- 24-ubc_2017_november_wadden_katie.pdf [ 25.66MB ]
- JSON: 24-1.0357178.json
- JSON-LD: 24-1.0357178-ld.json
- RDF/XML (Pretty): 24-1.0357178-rdf.xml
- RDF/JSON: 24-1.0357178-rdf.json
- Turtle: 24-1.0357178-turtle.txt
- N-Triples: 24-1.0357178-rdf-ntriples.txt
- Original Record: 24-1.0357178-source.json
- Full Text