Investigating the impact of diffuse axonal injury on executive ... - TSpace

INVESTIGATING THE IMPACT OF DIFFUSE AXONAL INJURY ON WORKING MEMORY PERFORMANCE FOLLOWING TRAUMATIC BRAIN INJURY USING FUNCTIONAL AND DIFFUSION NEUROIMAGING METHODS.

by

Gary R. Turner

A thesis submitted in conformity with the requirements for the degree of Doctorate of Philosophy, Graduate Department of Psychology in the University of Toronto

© Copyright by Gary R. Turner, 2008

ii Investigating the impact of diffuse axonal injury on working memory performance following traumatic brain injury using functional and diffusion neuroimaging methods. Gary R. Turner, Ph.D., 2008 Department of Psychology, University of Toronto

Abstract

Traumatic brain injury (TBI) is a leading cause of disability globally. Cognitive deficits represent the primary source of on-going disability in this population, yet the mechanisms of these deficits remain poorly understood. Here functional and diffusion-weighted imaging techniques were employed to characterize the mechanisms of neurofunctional change following TBI and their relationship to cognitive function. TBI subjects who had sustained moderate to severe brain injury, demonstrated good functional and neuropsychological recovery, and screened positive for diffuse axonal injury but negative for focal brain lesions were recruited for the project. TBI subjects and matched controls underwent structural, diffusion-weighted and functional MRI. The functional scanning paradigm consisted of a complex working memory task with both load and executive control manipulations. Study one demonstrated augmented functional engagement for TBI subjects relative to healthy controls associated with executive control processing but not maintenance operations within working memory. In study two, multivariate neuroimaging analyses demonstrated that activity within a network of bilateral prefrontal cortex (PFC) and posterior parietal regions was compensatory for task performance in the TBI sample. Functional connectivity analyses revealed that a common network of bilateral PFC regions was active in both groups during working memory performance, although this activity was behaviourally relevant at lower

iii levels of task demand in TBI subjects relative to healthy controls. In study three, diffusionimaging was used to characterize the impact of diffuse white matter pathology on these neurofunctional changes. Unexpectedly, decreased white matter integrity was not correlated with working memory performance following TBI. However, markers of white matter pathology did inversely correlate with the compensatory functional changes observed previously. These results implicate diffuse white matter pathology as a primary mechanism of functional brain change following TBI. Moreover, reactive neurofunctional changes appear to mediate the impact of diffuse injury following brain trauma, suggesting new avenues for neurorehabilitation in this population.

iv Acknowledgements This work would not have been possible without the generous support and guidance of my supervisor, Brian Levine. His intellectual mentorship and unwavering support over the course of my doctoral studies has been invaluable. I consider our collaborations on this project merely the prelude to what I hope will be an enduring friendship and research partnership. I also thank my committee members, Morris Moscovitch and Randall McIntosh for their guidance and support throughout the project. I owe an enormous debt of gratitude as well to Robin Green, who initially set me on course for this wondrous adventure and who has become a mentor, life coach and close personal friend over these last few years. I am deeply grateful for the enormous support of the Levine lab research team who provided invaluable assistance with this project and others throughout my tenure in the lab. Specifically, for their help on this dissertation project, I offer my heartfelt thanks to Adriana Restagno and Marina Mandic. Also for their friendship and support I especially thank Charlene O’Connor as well as Nathan Spreng, Eva Svoboda, Nadine Richard, and, for always keeping me grounded, Asaf Gilboa. My family has continued to be an unending source of strength, support, encouragement and love through all of my endeavours and I owe them everything. All I have been able to accomplish over the tenure of my dissertation studies I owe first and foremost to my life partner Marcelo Martins. His support, patience and generosity has provided me with the strength to weather the hard spots and the confidence to celebrate the successes. I could not imagine this journey without him.

v Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .iv Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .viii General Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Chapter 1: Augmented neural activity during working memory following diffuse axonal injury ………. . . . . . . . . . . . . . . . . . . . . . . . . . .13 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Chapter 2: Compensatory neural recruitment during verbal working memory performance after TBI: evidence for an altered functional engagement hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .32 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .40 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Chapter 3: Diffuse axonal injury as a mechanism for functional brain changes following traumatic brain injury: an integrated diffusion-weighted and functional imaging study. . . . . . . . . . . . . . . . . . .65

vi

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .65 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .67 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .78 Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 General Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .98

vii List of Tables Chapter 1 Table 1.1: TBI patient demographics, acute injury characteristics, structural neuroimaging data, and neuropsychological test data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .16 Table 1.2: Activation cluster maxima corresponding to maximal BOLD signal changes during the Alphabetize vs. Maintain conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Chapter 2 Table 2.1: Cluster maxima from the behaviour PLS (ST-bPLS) analysis for the control group (See figure 2.1a). . . . . . . . . . . . . . . . . . . . . . . . . . 45 Table 2.2: Cluster maxima from the behaviour PLS (ST-bPLS) analysis for the TBI group (See figure 2.1b). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .46 Table 2.3: Cluster maxima from the combined behaviour and seed (ST-bPLS & ST-sPLS) analysis for the left inferior frontal gyrus seed (see figure 2.3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Table 2.4: Cluster maxima from the combined behaviour and seed (ST-bPLS & ST-sPLS) analysis for the right posterior middle frontal gyrus (BA 46 / 44) seed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Chapter 3 Table 3.1: Voxel cluster coordinate and maxima for the control versus TBI whole brain FA comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 Table 3.2: Suprathreshold activation cluster maxima from LV 1 (p < .001) in the Seed and Genu PLS analysis (see Figure 3.4). . . . . . . . . . . ….. .86 Table 3.3: Suprathreshold activation cluster maxima from LV 1 (p < .001) in the Behaviour, Seed and Genu PLS analysis (TBI group only, see Figure 3.5). Seed was placed in right lateral middle frontal gyrus (BA 46/44). All abbreviations as in Table 3.2.. . . . . . . . . . . .……88 Table 3.4: Regions where white matter FA is predicted by BOLD response in right GFm during Alphabetize 5 task (TBI participants only). Letters correspond to Figure 3.6 . . . . . . . …. .90

viii List of Figures Chapter 1 Figure 1.1: Schematic of fMRI behavioural paradigm. . . . . . . . . . . . ……………18 Figure 1.2: Behavioural data for Alphaspan task during fMRI scanning. . . . . . . …23 Figure 1.3: Areas of maximal BOLD signal change in the Executive Demand contrast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Figure 1.4: Region of interest (ROI) analysis indicating mean % change difference between Alphabetize and Maintain conditions for the control group and individual TBI participants . . . . . . . . . . . . . . . . ..28 Chapter 2 Figure 2.1.a: Brain regions demonstrating significant correlations between brain response and task accuracy in control participants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Figure 2.1.b: Brain regions demonstrating significant correlations between brain response and task accuracy in TBI participants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Figure 2.2: Correlations between brain activity in anterior middle frontal gyrus seed regions (highlighted in Figure 2.1). . . . . . . . . . …47 Figure 2.3: Brain regions demonstrating reliable and positive correlations with left inferior frontal gyrus (BA 44/6) activity and task accuracy during Alphabetize 5 trials in the control group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Figure 2.4: Brain regions demonstrating reliable and positive correlations with left inferior frontal gyrus (BA 44/6) activity and task accuracy during Alphabetize 5 trials in the control group. . . . . . . . . . . 54 Figure 2.5: Conceptual representation of combined behavioural and seed partial least squares analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . … .57 Chapter 3 Figure 3.1: Group differences in voxel-wise distribution of white matter FA values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .79 Figure 3.2: Regions demonstrating significant differences in white matter FA between control and TBI participants. . . . . . . . . . . . . . . . . ..80

ix

Figure 3.3: Correlations between callosal FA and Accuracy on the Alphabetize 5 letter working memory task . . . . . . . . . . . . . . . . . . . . . . 82 Figure 3.4: Brain regions demonstrating reliable and positive correlations with right GFm seed (A5, F3, F5) and negative correlations with Genu FA (all tasks) in TBI subjects . . . . . . . . . . . . . . . . . . . . . . …85 Figure 3.5: Brain regions demonstrating reliable and positive correlations with right GFm seed (A5, F3, F5) and negative correlations with Genu FA (all tasks) in TBI subjects. . . . . . . . . . . . . . . . . . . . . . . . .87 Figure 3.6: Regions where activity (i.e. BOLD response) in right middle frontal gyrus during Alphabetize 5 condition predicted lower white matter FA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 89

INTRODUCTION

Dissertation objectives and key research questions Traumatic brain injury (TBI) represents a significant and growing public health concern. World-wide incidence rates for TBI range from 180 to 500 cases per 100,000 population per year (CDC, 2001), with trauma-related brain injury poised to become the third leading cause of death and disability globally by 2020 (Povlishock & Katz, 2005). TBI survivors often experience chronic disability resulting in significant personal and societal costs and cognitive deficits have been identified as the primary source of lingering disability in this population (CDC, 2001). While the neural mechanisms underlying these deficits are not well characterized, diffuse pathology, such as diffuse axonal injury (DAI), is suspected to play a significant role and has been shown to be the primary determinant of overall recovery in more severe cases (Ross, Temkin, Newell, & Dikmen, 1994). Diffuse neuropathology is ubiquitous in TBI. Incidence has been estimated at 40-50% of hospitalizations and rises to 100% in motor vehicle accidents where there was loss of consciousness (Meythaler, Peduzzi, Eleftheriou, & Novack, 2001). Our increasing reliance upon motorized transport and advances in acute care treatment of brain trauma suggests that the number of people living with TBI and, more specifically, DAI will continue to grow over the coming decades. Thus, investigating the relationship between DAI and cognition in this population is emerging as a priority in health care research. There have been numerous attempts to correlate measures of diffuse neuropathology and cognitive performance. Early investigations relied on gross measures of diffuse injury,

1

2 such as ventricular volume (Meyers, Levin, Eisenberg, & Guinto, 1983; Wilson, 1990; Vilkki, Holst, Ohman, Servo, & Heiskanen, 1992). More recently, emerging techniques, such as diffusion-weighted imaging and magnetic resonance spectroscopy, have facilitated greater precision in the measurement of diffuse neuropathology in vivo, yet the correlation between measures of white matter volume and/or structural integrity and cognitive performance is equivocal. In an earlier report, Tomaiuolo and colleagues (2004) identified a negative relationship between corpus callosum and fornix volume and neuropsychological test performance, however many of the observed correlations were found to be insignificant in a follow-up report (Tomaiuolo et al., 2005). Salmond et al. (2006) reported a significant relationship between diffusivity (a measure of white matter integrity) and a measure of learning and memory. Yet they failed to identify similar correlations with indices of motor or attentional functioning, nor did they observe significant correlations between a second scalar measure of diffusion (fractional anisotropy – FA) and any of their cognitive measures. Kraus et al. (2007) recently reported correlations between white matter FA and indices of executive functioning, attention and memory across a range of TBI severity. However, they did not characterize the incidence of focal lesions in their sample, making it difficult to attribute their findings to diffuse injury. Nakayama et al. (2006) failed to identify a relationship between memory performance and white matter integrity in the corpus callosum in a sample of DAI participants without focal lesions. As evident from this brief review, efforts to characterize the relationship between diffuse axonal injury and cognitive functioning following TBI are complicated by myriad factors which likely contribute to the equivocality of the findings. These include task-related differences, overlapping contributions of focal and diffuse injury and methodological differences (e.g. region of interest versus whole-brain analysis, or the choice of a diffusion

3 metric). Another potential source of variability, which heretofore has not been considered in this population, is the role of neural or cognitive reorganization following TBI in mediating the relationship between DAI and cognitive performance. There is a growing body of literature involving the utilization of fMRI methods to examine functional brain changes during cognitive task performance following TBI (see Levine et al., 2006 for a review). While similar interpretive challenges appear in this literature (i.e. task-, injury- and analysisrelated variability), there is converging evidence that the functional neural correlates of cognition are altered following TBI. Moreover, advances in our understanding of cognition as an emergent property of activity in widely distributed neural networks (McIntosh, 1999; Postle, 2006) raise important questions with respect to the potential interactions amongst diffuse axonal injury, functional reorganization and cognitive performance in this population. The goal of my dissertation research is to investigate the interaction of traumaticallyinduced DAI, functional brain activity and cognition using functional and structural neuroimaging methods. Specifically, I will address three research questions: (i)

Does TBI alter the functional neuroanatomy of cognitive task performance?

(ii)

Does cognitive performance depend on functional reorganization following TBI?

(iii)

How do structural and functional brain changes interact following TBI? As the overarching theme of my dissertation project involves the application of

neuroimaging methods to investigate the impact of diffuse axonal injury on the neural and behavioural correlates of cognitive functioning, the following section provides a brief review of the pathobiology of TBI, with a particular emphasis on the mechanisms of DAI. This is followed by a short primer on the benefits of adopting neuroimaging methods in neuropsychological investigations. In the final section, I review general methods for the

4 project and briefly review the rationale, research hypotheses and methods for each of the three accompanying papers.

Pathobiology of traumatic brain injury Early models of TBI identified oedema as the primary, if not the sole, source of neuropathology following brain trauma. These ideas have since been supplanted with a more sophisticated model of neuropathological change associated with TBI (see Povlishock & Christman, 1995; Povlishock & Katz, 2005 for comprehensive reviews of TBI-related neuropathology). Pathological reports (e.g. Adams, Gennarelli, & Graham, 1982; Strich, 1956), animal models (e.g. Gennarelli et al., 1982) and, increasingly, neuroimaging methods (Arfanakis et al., 2002; Huisman et al., 2004; Kraus et al., 2007) have elucidated the cellular mechanisms and structural consequences associated with a complex series of neuropathological cascades involving molecular, metabolic and ischemic changes that are initiated when the brain is rapidly displaced within the cranium. Traumatically-induced neuropathology may be characterized along two principal dimensions: focal versus diffuse injury and primary versus secondary processes (Povlishock & Katz, 2005). Focal injury, or contusions, result from haemorrhagic lesions occuring at grey-white matter junctions or along the cortical surface which may be abraded by contact with bony ridges of the cranium. Contusions of the latter type typically occur in regions of the orbital frontal cortex and anterior temporal lobes adjacent to the bony ridges of the skull surrounding the orbits. Haemorrhagic lesions may also occur at (or opposite to) the location of impact, a process known as coup-contra-coup injury. While cortical contusions represent the primary form of focal brain injury, secondary focal damage may occur as a result of microvascular changes

5 attributable to physical shear and tensile forces or due to ischemic injury secondary to metabolic disturbances. Diffuse injury, the focus of the current research, was initially presumed to occur immediately upon impact as a result of shear and strain forces. These mechanical forces were assumed to produce physical disconnection of proximal and distal segments of the axon, a model of DAI which was derived from early pathologist reports of ‘retraction balls’ appearing at the ends of the disconnected axonal segments. Subsequent research revealed a more complex process involving both primary and secondary neuropathological cascades. Initially, as the axolemma is displaced by the physical force of the trauma, its flow disrupts intra-axonal neurofillaments which, in turn, impedes afferent and efferent transit along the length of the axon. Accumulating organelles at the site of the disruption produce axonal swelling and eventual degradation of fibre integrity leading, over the course of several days, weeks and perhaps months, to axotomy or disconnection of the distal segment from the neuronal soma. While axonal disconnection due to mechanical shear forces is still presumed to occur in cases of severe injury, the discovery of these secondary neuropathological processes suggests that DAI may be considerably more prevalent, and at lower severity levels, than earlier pathological reports suggested. This possibility is further supported by emerging data suggesting that even small focal alterations in membrane permeability, secondary to mechanical strain and shear forces, may trigger auto-destructive cascades which also lead to eventual axotomy. Moreover, a recent report by Smith and Meany (2000) identified these auto-destructive processes in both myelinated and unmyelinated neurons where the axonal membrane remained intact. In these cases, physical strain forces on unmyelinated axons appeared to disrupt the sodium calcium pumps, disrupting ionic homeostasis within the axon fibre and thereby initiating the autodestructive cascades that

6 break down neurofillaments. Povlishock and Katz (2005), in their recent review, suggested that similar pathological mechanisms may not be limited to the axolemma but may also occur at the site of the neuronal soma. In sum, diffuse axonal injury is a primary consequence of TBI. While early reports of DAI implicated mechanical shearing and tensile forces as the sole mechanism of injury, more recent data has demonstrated that the effect of these physical forces is mediated through a complex series of secondary intracellular cascades. Physical forces disrupt ionic homeostasis and result in delayed axotomy that occurs over days, weeks or months following the initial injury. Moreover, these autodestructive secondary cascades may or may not involve axonal swelling (often used as an acute marker of injury severity). Axonal damage may also occur as the result of focal alterations in membrane permeability or altered functioning of ionic pumps which regulate homeostasis within the axolemma and may impact either myelinated or unmyelinated fibres or both. These data suggest that DAI following TBI may more extensive than previously understood at all levels of TBI severity, implicating this neuropathological process as a potential mechanism for the patterns of cognitive dysfunction observed following brain trauma.

Functional neuroimaging methods in TBI Price and Friston (2002) have argued that functional neuroimaging may complement structural neuroimaging and behavioural assessments of patients in a number of ways, two of which are relevant here. First, the authors argue that functional imaging may reveal cognitive responses not evident through behavioural measures alone. As discussed above, the variability reported in investigations of DAI and cognition may be attributable to the instantiation of novel or supplementary cognitive processes following TBI, thereby

7 compensating for diffuse structural damage. Price and Friston (1999) have labelled this ‘cognitive reorganization’. Second, they argue that functional imaging may reveal neural mechanisms of recovery following injury through processes such as reverse diaschisis, perilesional recruitment, ‘unmasking’ of normally inhibited systems or learning-dependent plasticity. These processes fall into the category of ‘neural reorganization’ under the authors’ schema of reorganization following brain damage. With respect to the present research objective, evidence for neural or cognitive reorganization following DAI would suggest that cognitive deficits attributable to DAI may be the result of absent or functionally inadequate reorganization, opening potential avenues for rehabilitation interventions. This possibility of using functional neuroimaging methods to observe such functional recruitment has been identified as an important tool in the armamentarium of rehabilitationists in a recent review (Chen, Abrams, & D'Esposito, 2006). A final point raised by Price and Friston (2002) is worth reviewing briefly here. While counter-intuitive with respect to standard neuropsychological investigations, determination of cognitive or neural reorganization in a patient population using fMRI methods is critically dependent upon their capacity to perform the task at an equivalent level to controls. Performance differences on the behavioural measure of interest raise the possibility that brain differences may simply be reflecting these performance differences and not ‘reorganization’ per se. I return to this issue in the ‘General Methods’ section below.

So far, in this general introduction, I have presented data that diffuse injury following TBI may be more extensive than previously understood. In this context, the correlations between DAI and cognitive functioning that have been reported to date in the literature are surprisingly tenuous. Heterogeneity in methods, cognitive measures, and patient sample

8 characteristics undoubtedly contribute to this paradox. Here I propose that variability in cognitive and/or neural reorganization is a significant contributor to these divergent findings, and if so, may offer a window of intervention for remediation or compensation of cognitive deficits following TBI. In the final section, I propose a plan of research using functional and diffusion neuroimaging techniques to interrogate this proposition.

General Methods Participants One of the primary goals of the current research was to reduce heterogeneity within the TBI patient cohort recruited into this study, thus allowing us to attribute functional brain changes to specific neural mechanisms following TBI. There has been considerable heterogeneity in participant injury characteristics in previous reports investigating the role of DAI in cognitive functioning, thus limiting the study of brain structure, function and behavioural interactions. I address this issue directly in a number of ways. First, the TBI patients recruited for this research (characterized in detail in paper 1) are free of significant premorbid or secondary neurological illness or psychiatric diagnosis. Furthermore, selection criteria were devised to maximize homogeneity and minimize variability in this notoriously heterogeneous patient population. Specifically, we selected participants who had sustained TBI as a result of a motor-vehicle accident where there was documented loss of consciousness and indication of significant post-traumatic amnesia (i.e. moderate to severe TBI). Incidence of DAI has been estimated near 100% when these criteria are met (Meythaler et al., 2001). Participants were also excluded if there was evidence of macroscopic focal lesions or if there was no indication of DAI-lesions on neuroradiological report. Finally from this pool of ‘pure’ DAI subjects, we selected only those who were in

9 the chronic recovery period and who had demonstrated good functional and neuropsychological recovery. While these criteria may limit the generalizability of these data in a clinical sense, they are critical to parsing the impact of DAI on the neural basis of cognitive recovery following TBI.

Functional Imaging Paradigm Diffuse neuropathology such as DAI is pervasive following brain trauma, thus one might predict that TBI would be associated with relative deficits in more integrative cognitive domains, where processing demands are distributed across the cerebrum. This is indeed the case. Deficits in executive functions, which are mediated by prefrontal cortical regions and their connections to posterior and subcortical brain structures, are a hallmark of TBI (see Turner & Levine, 2004 for a review). Working memory, or the capacity to hold information ‘in mind’ when it is no longer present in the environment, and to manipulate, monitor and update it in the service of behaviour (Goldman-Rakic, 1987), is considered to rely upon these prefrontally-mediated neural networks. Working memory is perhaps the most clearly characterized domain of higher cognition both in terms of its behavioural and neural correlates (Postle, 2006). Recently, Vallat-Azouvi, Weber, Legrand and Azouvi (2007) convincingly demonstrated that working memory is disrupted following TBI and that deficits are specific to executive control processes as opposed to simple maintenance operations (see D'Esposito, Cooney, Gazzaley, Gibbs, & Postle, 2006 for further evidence). This is consistent with a theoretical dissociation first proposed by Baddeley (1986) and allows for clear predictions with respect to the preferential impact of DAI on executive control processes in working memory following TBI. Here I have selected a working memory paradigm where both executive and storage demands could be explicitly manipulated in an

10 effort to optimally challenge mechanisms of neural or cognitive reorganization in our TBI sample. Moreover, to maximize sensitivity to potential brain changes following TBI, I selected a paradigm for which the functional neuroanatomy has been well characterized in healthy control participants (Alphaspan, Craik, 1990; Postle, Berger, & D'Esposito, 1999 and see Chapter 1 for a complete task description).

In the final section of this general introduction I provide a brief synopsis of the three research papers that comprise my dissertation thesis project. I present the general research question being addressed, the methodological approach and a brief review of the methods and the rationale for each study vis-a-vis the overall objective of the dissertation research.

Research study overview Chapter 1. Does TBI alter the functional neuroanatomy of cognitive task performance? This question has been examined previously (e.g. McAllister et al., 1999; McAllister et al., 2001; Perlstein et al., 2004). In this study, I expand upon these earlier reports in three ways: (i) by isolating the contribution of DAI (see ‘Participants’ above) (ii) matching performance between TBI and healthy controls (see ‘Functional Imaging’ above) and (iii) by parsing the impacts of TBI on executive control and storage processes in working memory. These have been confounded in previous reports, potentially masking neurofunctional changes attributable to either. The principal objective of this first study is to demonstrate that altered neural recruitment could be observed during executive control processing in a sample of ‘pure’ DAI subjects who were multiple years post-injury, who had demonstrated good functional and neuropsychological recovery, and whose behavioural performance was equivalent to a

11 non-injured comparison group. Study one is foundational in that it replicates and advances the findings of these previous reports and sets the stage for examining whether the presence of functional brain changes correspond to behaviour and thus mediate the impact of DAI following brain trauma.

Chapter 2. Does cognitive performance depend on functional reorganization following TBI? Having investigated the functional correlates of higher cognitive functioning (i.e. executive processing in working memory) in Chapter 1, in this series of analyses I explore whether evidence of altered functional recruitment, secondary to DAI, may be related to behavioural performance. Given the distributed nature of DAI, I employed multivariate analysis methods to identify behaviourally relevant functional brain networks engaged at differing levels of working memory task demand following DAI. The principal goal of the second study was to explicitly correlate measures of functional reorganization and cognitive task performance following DAI.

Chapter 3. How do structural and functional brain changes interact following TBI? The goal of the first two studies was to investigate whether DAI was specifically associated with altered functional brain activity during performance of a working memory task and whether this pattern of neural reorganization was behaviourally relevant. Interpretation of these reports depends on the assumption that DAI is present. While this assumption is likely valid given the fact that all patients had a significant TBI, there was no quantification of DAI. In this final report I asked whether quantitative measures of DAI would directly predict performance on a working memory task and to what extent this relationship was dependent upon the neural reorganization investigated in studies 1 and 2. Here I integrated functional

12 neuroimaging, diffusion-weighted imaging and working memory behavioural data for the first time in this population. The principal objective of this study was to document the extent to which altered functional brain activity mediated the impact of diffuse injury following TBI.

In summary, my dissertation research combines functional and diffusion-weighted neuroimaging methods to investigate the impact of DAI on the functional neuroanatomy of higher cognitive functioning, specifically executive control within working memory. To assess this relationship, I ask three questions: Does DAI specifically result in altered functional recruitment following TBI? Are these changes behaviourally relevant? How does DAI interact with neurofunctional changes to influence behavioural and functional outcomes following TBI? Increasing evidence points to DAI as a primary source of cognitive deficit following TBI. In the three Chapters that follow 1 , I have employed functional and structural neuroimaging methods to investigate why this may be the case and what the implications may be for functional recovery.

1

The three chapters that follow this General Introduction are presented here in manuscript format. At the time of writing, report 1 is in submission at Neurology, reports 2 and 3 are in preparation for submission. The three reports have been reformatted into standardized APA format for consistency within this thesis document.

13 CHAPTER 1 Augmented Neural Activity During Working Memory Following Diffuse Axonal Injury

Abstract Deficits in working memory are commonly observed following traumatic brain injury (TBI), with executive control processes preferentially impacted relative to storage and rehearsal. Previous activation functional neuroimaging investigations of working memory in patients with TBI have reported altered functional recruitment, but methodological issues including sample heterogeneity (e.g. variability in injury mechanism, severity, neuropathology or chronicity), under-specified definitions of ‘working memory’, and behavioural differences between TBI and control groups have hindered interpretation of these changes. Here, executive control processing in working memory was explicitly engaged during fMRI in a sample of carefully selected chronic stage, moderate to severe TBI patients with diffuse axonal injury (DAI), but without focal lesions. Despite equivalent task performance, we observed a pattern of greater recruitment of inter- and intrahemispheric regions of prefrontal cortex (PFC) and posterior cortices in our DAI sample. Enhanced activations were recorded in left dorsolateral PFC (middle frontal gyrus), right ventrolateral PFC (inferior frontal gyrus), as well as bilateral posterior parietal cortices and left temporal-occipital junction. Region of interest analyses confirmed that these effects were robust across individual patients and could not be attributed to load factors or slowed speed of processing. These data suggest that augmented functional recruitment in the context of normal behavioural performance may be a neural marker of capacity or efficiency limits that can affect functional outcome following TBI with diffuse injury.

14 Introduction Deficits in working memory are commonly observed in patients with traumatic brain injury (TBI, McAllister, Flashman, Sparling, & Saykin, 2004). Working memory, however, is not a unitary process; it can be separated into storage and executive control components (Baddeley, 1986). There is substantial evidence that these components can be dissociated at the neural as well as at the behavioural level (D'Esposito, Cooney, Gazzaley, Gibbs, & Postle, 2006; Goldman-Rakic, 1987), with the executive control component strongly associated with dorsolateral prefrontal cortex (dlPFC, Curtis & D'Esposito, 2003). This functional localization creates a puzzle when considering patients with TBI, where ventral, not dorsal frontal regions, are vulnerable to focal lesions (Courville, 1937), and where focal lesions may be absent even in the presence of marked cognitive deficits (Scheid, Walther, Guthke, Preul, & von Cramon, 2006). This raises the possible influence of diffuse axonal injury (DAI). DAI is a ubiquitous consequence of rapid deceleration of the head, followed by disrupted ionic homeostasis, cytoskeleton compromise, and ultimately, disconnection of the distal axonal segment from the soma (Povlishock & Katz, 2005). dlPFC functions may be preferentially impacted by DAI as a result of its extensive reciprocal connections with almost all cortical and subcortical structures in the brain (Petrides & Pandya, 1999). While DAI is likely to substantially contribute to the executive control deficits in TBI, extant evidence to support this hypothesis is weak. Investigations of the functional neuroanatomy of working memory following significant TBI have suggested increased recruitment of frontal regions (Christodoulou et al., 2001; McAllister et al., 1999; Perlstein et al., 2004), but these are complicated by methodological factors such as task complexity (in particular, confounding working memory load effects with the critical executive control

15 processes), performance impairments in patients, and the inclusion of patients with focal and diffuse injuries (for a review see Levine et al., 2006). More evidence is found in mild TBI (Chen et al., 2004; McAllister et al., 1999), but these studies were conducted at the acute phase, with little expected long-term functional consequences relative to patients with more significant TBI. In this study, we assessed patients with severe TBI and documented DAI but without large focal lesions. We have adopted a working memory paradigm with wellestablished functional neuroanatomy that is capable of separating executive control and load factors, with good convergent validity with other measures of executive control in working memory (Kane et al., 2004). Finally, we have used a standardized pre-scan training regimen to equate task performance between TBI and control subjects during fMRI scanning.

Methods Participants Eight patients (6 males) with moderate to severe TBI were recruited along with twelve (12) healthy control subjects (8 males). TBI patients and control subjects were matched on age [t(18) = .785 (p > .05; NS)] and education [t(18) = -1.99 (p > .05; NS)]. All subjects were right-handed, native English speakers and were screened for previous neurological injury, major medical conditions affecting cognition, history of psychiatric illness, and the use of medications affecting cognition (Levine et al., in press). Two other sets of control subjects were used for the purposes of characterizing TBI patients’ neuropsychological status and structural neuroimaging data (see Table 1.1). Table 1.1: TBI patient demographics, acute injury characteristics, structural neuroimaging data, and neuropsychological test data.

16 Subject #

Demographics Age (years) Education (years) Injury Characteristics Glascow Coma Score Loss of Consciousness (hrs) Post Traumatic Amnesia (hrs) Severity Classification 3

Quantitative Structural Imaging (μl x 10 Gray Matter White Matter Cerebral Spinal Fluid Select Neuropsychological Tests Shipley Institute of Living Scale (verbal) Verbal Fluency (# words generated) Symbol-Digit, written (# correct) Symbol-Digit, oral (# correct) Trail-making test (B-A) Wisconsin Card Test (perseverations) Self-Ordered Pointing (total errors)

1133

2930

3639

3649

1054

3645

3653

3646

Mean (SD)

Control

33 14

26 14

32 15

30 17

27 17

43 15

36 14

28 16

32 (6) 15 (1)

28 (8) 15 (2)

4.5 336 1008 Sev.

15 0.25 336 Mod.

9 5 504 Mod.

10.5 168 n/a Mod.

9 26 240 Sev.

9 < 0.25 n/a Mod.

9 6 4 Mod.

6 96 1080 Sev.

9 (8.3-9.4) 26 (5.5-132) 420 (264-882)

61.8 41.5 20.7

62 41.5 20.6

61.4 43.3 19.2

55.9 44.4 23.9

60 42.4 21.2

60.5 (2.3)* 42.2 (1.6)* 21.5 (1.7)*

63.1 (1.6) 46.8 (1.8) 16.3 (1.9)

34 35 51 64 12 13 6

29 45 46 67 33 17 13

28 49 69 88 25 11 4

28 38 55 65 35 16 10

36 33 45 49 25 14 8

31 (4) 41 (6) 52 (8) 65 (12) 27 (8) 13 (3) 8 (3)

30 (4) 40 (10) 59 (10) 72 (12) 31 (15) 20 (9) 6 (4)

) 61.5 39.8 23.1 34 48 54 64 24 11 4

26 44 47 58 33 10 10

Notes. Three separate control groups were used. Demographics are compared to the functional imaging control subjects (N=12); structural neuroimaging data are compared to age-matched healthy adults (N=12); neuropsychological data are compared to age-, education-, and socioeconomic-matched healthy adults (N=27). Data are means and standard deviations, except for acute injury characteristics, which are medians and intraquartile ranges. * Indicate significant differences from controls. Bold text in Select Neuropsychological Tests signify impaired performance.

TBI patients were recruited from consecutive admissions as part of the Toronto TBI study (Fujiwara, Schwartz, Gao, Black, & Levine, in press; Levine et al., in press). All patients had sustained a TBI as a result of a motor vehicle accident and were in the chronic stage of recovery at the time of study participation. Despite their significant TBIs, all patients demonstrated good functional recovery as evidenced by a return to pre-injury employment or academic status. Injury severity was determined by GCS as documented by the trauma team leader’s score at the time of discharge from the Trauma Unit, corresponding to the recommended 6-hour GCS score (Teasdale & Jennett, 1974). Severity in two cases (1054, 2930) was upgraded from that indicated by the GCS due to extended loss of consciousness, extended post-traumatic amnesia, or both. Seven of eight patients underwent a separate structural MRI TBI protocol (for details, see Levine et al., in press). Radiological interpretation indicated evidence of DAI-related neuropathology (hemosiderin deposits)

17 localized to the frontal lobes in all seven patients, with additional pathology visualized in the parietal lobe (2 patients), the temporal lobe (1 patient), the corpus callosum (1 patient), the basal ganglia (2 patients), and the thalamus (1 patient). No patient had lesions greater than 3 mm in diameter. Whole-brain volumetric measures of gray matter, white matter, and cerebrospinal fluid (CSF) following our published image analysis methods (Levine et al., in press) were available for six patients, all of whom showed evidence of atrophy relative to age- and education-matched control subjects. Taken together, the radiological interpretation and significant volume loss in the TBI patients are consistent with DAI. Neuropsychological test data were compared with a local normative sample of age-, education-, and socioeconomically-matched control subjects. TBI patients demonstrated average to high average performance on the verbal subtask of the Shipley Institute of Living Scale (Zachary, 1986). This premorbid estimate, combined with matched education levels between our TBI and fMRI control subjects, minimize group differences in native verbal capacity limitations, an important criterion for a study of verbal working memory. TBI patients also performed normally on other neuropsychological tests of attention and executive functioning, including (with one exception) a task explicitly measuring executive control within working memory (Petrides & Milner, 1982).

Behavioural Task We employed a modified version of the Alphaspan task (Craik, 1986), previously used in neuroimaging studies to separate executive control processes from storage and rehearsal in verbal working memory (D'Esposito, Postle, Ballard, & Lease, 1999; Postle et al., 1999). On each trial, subjects studied a letter set consisting of 3 or 5 consonant letter strings (set size or ‘load’ manipulation) and were asked to ‘MAINTAIN’ the letter set over a brief

18 delay or ‘ALPHABETIZE’ the letters into the correct alphabetical position during the delay (executive demand manipulation). At the end of the delay, a probe was presented consisting of a letter and an ordinal position (e.g. L-4? – ‘Was ‘L’ the 4th letter in the set?’). On Maintain trials, the probe referred to the ordinal position in the original letter set while on Alphabetize trials the probe referred to the letter position following alphabetization of the list. Probability of a correct probe was set at 0.5 for all trials in all conditions (see Figure 1.1 for a schematic of the task design). Executive control was operationalized as the difference between Alphabetize and Maintain conditions (Postle et al., 1999). Figure 1.1: Schematic of fMRI behavioural paradigm. Warning cue (“Alphabetize” or “Maintain”) was presented at the beginning of each trial block. ITI = intertribal interval. 3 / 5 letters (1.5 / 2.5 s)

Delay ( 7 sec )

+

DR H C M

Time(s)

0

1

2

3

Probe (2s on + 2s off)

4

5

6

7

ITI

avg of 5 sec M – 4?

8

9

10 11 12 13 14 15 16 17 18

Prior to scanning, each participant was provided with step by step instructions as to how to perform the tasks, followed by 20 trials (5 trials in each of the 4 conditions). During scanning, subjects completed 28 trials of each of the four task conditions, Alphabetize 3 and 5 letter sets (A3, A5) and Maintain 3 and 5 letter sets (M3, M5) in a single scan session. Within each session four 12-min scans were acquired. Trials were grouped by executive demand with 2 blocks of seven trials at each level of executive demand presented during a single scan acquisition. Set size was randomized within each block. At the beginning of each trial block, subjects were presented with a warning screen to alert them as to the

19 upcoming condition (i.e. either “Maintain” or “Alphabetize”). To alert the participant to the upcoming letter set, a grey rectangle was presented at the center of the screen for 500 ms. Letter strings were presented at a standard timing rate of 500 ms per letter (1500 ms for three item set sizes, 2500 ms for five items). The letter set was then replaced by a fixation cross presented at the center of the screen for 7000 ms. At the end of this delay period, the probe was presented for 2 seconds, with an additional 2 second period of blank screen during which responses were allowed. Trials were separated by a variable inter trial interval averaging 5 seconds across all trials in a single scan session. Total stimulus onset asynchrony was 18000 ms (3 letter trials) or 19000 ms (5 letter trials).

fMRI scanning and analyses Scanning was performed on a whole-body 3.0 T MRI system (Signa 3T94 hardware, VH3M3 software; General Electric Healthcare, Waukesha, WI) with a standard quadrature bird-cage head coil. Subjects were placed in the scanner in supine position, with their head firmly placed in a vacuum pillow to minimize head movement. Earplugs were provided to reduce the noise from the scanner, and sensors were placed on subjects’ right index finger and around the chest to monitor heart rate and respiration. A volumetric anatomical MRI was performed before functional scanning, using standard high-resolution 3D T1-weighted fast spoiled gradient echo (FSPGR) images (TR/TE=7.2/3.1 ms, inversion-recovery prepared T1=300 ms, flip angle 15°, 256×192 acquisition matrix, 124 axial slices 1.4 mm thick, voxel size=0.86×0.86 cm, FOV=22×16.5 cm). Functional imaging was performed to measure the blood oxygenation level dependent (BOLD) effect (Ogawa, Lee, Kay, & Tank, 1990). Scans were obtained using a single-shot T2*-weighted pulse with spiral in-out, achieving 26 slices, each 5 mm thick (TR/TE=2000/30 ms, flip angle 70°, 64×64 acquisition

20 matrix, voxel size=3.125×3.125, slice spacing=0, FOV=20×20 cm). Data processing and analyses were performed using Analysis of Functional NeuroImages software (Cox & Hyde, 1997). The initial 10 time points from each functional image volume were excluded from the analyses to allow for stabilization of the magnetic field gradients. Time-series data were spatially co-registered (aligned volumetrically to a reference image within the run, using the 3dvolreg program in AFNI) to correct for small head motion using a 3-D Fourier transform interpolation, and the linear trends were removed. Uncorrected head motion (spikes) was identified through visual inspection and reduced through averaging the two surrounding time points. Physiological motion (respiration and heart beat) was also removed through linear filtering. Finally, slice timing correction was carried out to account for the time dependent discrepancy between the initial and final slice acquisitions. Following preprocessing, the data were submitted to a deconvolution analysis using the AFNI plugin 3dDeconvolve. The functional data were modeled using a general linear model to derive parameter estimates and corresponding t-statistics for the impulse response functions corresponding to the four primary working memory conditions (Alphabetize 3/5, Maintain 3/5) as well as the two types of probe trials (probe 3-letter, probe 5-letter). We constrained the shape of the haemodynamic response function by convolving the stimulus time-series with a gamma function to obtain ideal waveform time-series. The resulting whole brain, voxel-based maps of the parameter estimates and their statistical (T-statistic) assessments (i.e. the within subject, within condition activation maps) were transformed into stereotaxic space (Cox & Hyde, 1997; Talairach & Tournoux, 1988) by aligning the anterior–posteriorcommissure, inferior–superior and left-right maxima, as well as mid-saggital reference points of the structural scans and scaling them to fit the Talairach–Tournoux Atlas brain. Activation maps were spatially smoothed with a Gaussian filter with a full width at half

21 maximum value of 8.0 mm to minimize individual variation of the anatomical landmarks. These steps were performed to facilitate the subsequent group analysis, which consisted of voxel-wise, mixed effects (groups fixed, conditions fixed, subjects random), three-factor ANOVA for unbalanced designs. This program is included as a MATLAB script with the AFNI software package. The three fixed factors each had two levels: Group (TBI, Control subjects), Executive Demand (Alphabetize, Maintain) and Set Size (3 letters, 5 letters). The fourth random factor, subject, was nested within Group. Task effects within groups were thresholded at an individual voxel probability of p150 μl (3 acquisition voxels), and having a connectivity radius of 2 mm (i.e. two clusters need to be separated by at least 2 mm to be considered different). For the interaction of Group x Condition (GxED), we report clusters of activation thresholded at p < .001, cluster size > 150 μl. Region-of-interest (ROI) analyses were drawn at these cluster maxima to assess the impact of inter-individual variability and potential speed of processing effects. Parameter estimates from each voxel for each condition were extracted from these ROIs and outliers defined as having scores more than 1.5 times the inter-quartile range of their respective groups for each condition were trimmed using a Winsorizing procedure to 2 SD above or below the true mean (i.e. exclusive of the outlier). This procedure allows the maintenance of extreme observations without unduly influencing statistical estimates.

Results Behaviour One TBI patient was identified as an outlier for accuracy (TBI # 3646). Trial-by-trial analysis for this patient revealed a consistent pattern of timing out on Alphabetize trials, greater for 5 than for 3 letter sets sizes. This pattern is indicative of a task-specific

22 decrement at response as opposed to disengagement or poor arousal. On those trials where responses were recorded, performance was within the average range of the TBI group for both 3- and 5-letter sets (78% and 67% respectively). As we were interested in neural response during encoding and delay trial epochs, and not at probe, we did not exclude this patient’s data. For the purposes of statistically assessing group effects on task performance, Alphabetize accuracy values for this subject were trimmed to 2 SD below the true mean using a Winsorizing procedure. A two group (control vs. TBI) x 2 (executive demand) x 2 (set size) repeated measures ANOVA revealed a significant main effect of executive demand (ED) and set size (Set) [F(1) = 18.51 (ED); F(1) = 11.42 (Set), p < .01 for both comparisons] with poorer performance observed during the Alphabetize and set size 5 conditions. The interaction between ED and Set was non-significant [F(1) = 0.02, NS]. Importantly, there was no main effect of group [F(1) = 2.54; NS] and no group x ED [F(1) = 0.094, NS] or group x Set [F(1) = 0.182, NS] interactions, indicating that the main effects of condition were equivalent across both the control and TBI groups (Figure 1.2, top panel). Post-hoc analyses revealed that there were no group differences on any of the tasks (p > .05, all comparisons). Analyses including the uncorrected outlier data did not alter these results.

23 Figure 1.2: Behavioural data for Alphaspan task during fMRI scanning. Panel A: task accuracy (% correct) for TBI and control subjects. Panel B: reaction time (msec). Error bars represent 95% confidence intervals. 1.00

A. Accuracy

0.90 0.80

Control

0.70

TBI

0.60 0.50 Alpha 3

Alpha 5

Maintain 3 Maintain 5 Task

B. RT (msecs)

2500 2000 1500

Control

1000

TBI

500 0 Alpha 3

Alpha 5

Maintain 3 Maintain 5 Task

On measures of reaction time, there was a main effect of condition [F(3) = 43.81, p < .001], and group [F(1) = 6.41; p < .05] but no group by task interaction [F(3) = 1.54; NS]. Post-hoc testing revealed that control subjects exhibited faster responding at probe for all tasks with the exception of Maintain 3 (p < .05, all comparisons). Task main effects were driven by longer reaction times at probe on 5-letter versus 3-letter set size trials, irrespective of executive demand (p < .001 for all comparisons; See Figure 1.2).

24 Neuroimaging Analyses Executive control was associated with increased activity in left lateral frontal, posterior parietal and insular cortices, replicating prior work with this paradigm (Postle et al., 1999; Figure 1.3, panel A; Table 1.2). The same contrast in TBI patients revealed a more extensive pattern of suprathreshold activation (Figure 1.3, panel B; Table 1.2). Brain voxels demonstrating greater Alphabetization versus Maintain activity (t > 4.96, p < 10-4 ) in this contrast represented .004% of total brain volume in the TBI sample as compared to .0008% in control subjects. Specific areas of activation included the lateral PFC bilaterally, the insular and superior parietal cortices bilaterally, and right temporal-occipital junction (TOJ). Group (TBI vs. Control) x Executive Demand (Alphabetize vs. Maintain) interactions were observed in left middle frontal gyrus (GFm), right inferior frontal gyrus (GFi), and left posterior regions including inferior and superior parietal cortices. Activity in these areas was significantly greater during Alphabetize than Maintain trials and this difference was significantly greater in TBI patients relative to control subjects (Figure 1.3, panel C; Table 1.2). An area of left angular gyrus, Brodmann Area (BA) 39 exhibited a significant Group X Executive Demand interaction associated with greater activation during Maintain trials in the control subjects. We used ROI analyses to confirm that these interactions were reliable (see Methods). Across ROIs an average of 77% of the TBI sample showed an increase in BOLD signal during Alphabetize trials that was greater than the 95th percentile for controls (Figure 1.4). A small sub-cluster threshold activation in the right medial frontal polar cortex (BA 10) demonstrated a significant Group x Executive Demand x Set size interaction. Post-hoc analyses revealed that this interaction was driven by increased activity during Alphabetize 5 relative to Maintain 5 trials in TBI patients over and above the control sample; a similar, albeit attenuated, pattern was observed in the 3-letter set

25 conditions. No similar interactions were observed within lateral PFC, suggesting that group differences in executive control processes mediated by this region were not impacted by load as defined in this study. Figure 1.3: Areas of maximal BOLD signal change in the Executive Demand contrast (i.e. Alphabetize > Maintain conditions), collapsed across set size. t=7

A

D.

Right IFG (BA 44)

t=-7

B

t=7

Alpha 5 - Maintain 5 (% chg)

TBI Control

0.2 0.15 0.1 0.05 0 -0.05 -0.1 0-2

2-4

4-6

6-8

8-10

10-12

12-14

14-16

Tim e (seconds)

Left MFG (BA 46)

C

Z=6

1

Z=16

2

Z=26

t=-7 F=16

F=0

Alpha 5 - Maintain 5 (% chg)

TBI 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 -0.05 -0.1

Control

0-2

2-4

4-6

6-8

8-10

10-12

12-14

14-16

Tim e (seconds)

Notes: Panel A = Control subjects; Panel B = TBI patients. Panel C. Areas demonstrating a significant Group by Executive Demand interaction. All suprathreshold cluster maxima for each group and the interaction are reported in Table 1.2.. Panel D.1 & D.2. Temporal differences (% change in BOLD signal) between Alphabetize and Maintain set size 5 conditions for control subjects and TBI patients, cross-referenced to numbers as specified in panel C. Green and red bars below the x-axis represent the onset of stimulus presentation and onset of probe respectively.

26

Table 1.2. Activation cluster maxima corresponding to maximal BOLD signal changes during the Alphabetize vs. Maintain conditions (i.e. main effect of Executive Demand). Region Lat. BA

x

Control y z

x

TBI y z

Group x ED x y z

Frontal Cortex Frontopolar Medial Frontal Gyrus

R

10

{1 55 11}

L L L

46 46 8

-40 28 20

R R L

44 46 44

R L L

6 6 6

R L

13 13

31 22 -31 22

L L R

40 7 7

-16 -64 53

L

39

{-52 -64 35}

Occipital Cortex Superior Occ. Gyrus R

39

Dorsolateral Middle Frontal Gyrus

Ventrolateral Inferior Frontal Gyrus

Premotor Superior Frontal Gyrus Middle Frontal Gyrus Insula

-46 43 14 -40 26 20 -49 28 44

-43 46 11 -49 25 47 55 13 14

46 28 11 -37

4 29 4 16 53 -19 4 62 -40 10 62 2 5

31 19 -34 19

-19 4 62 -34 10 56

2 5

Posterior/Subcortical Parietal Cortex Inferior lobule Superior lobule

Angular Gyrus

Thalamus Anterior Nucleus

R

-19 -70 56 34 -73 50 22 -67 56

-55 -40 50 -22 -70 56

-46 -70 38 31 -73 29

7 -13 17

Notes: Maxima are presented for each group and for the Group by Executive Demand interaction. Coordinates are reported in standard space according to the stereotaxic map of Talairach and Tournoux (1988). Individual voxel probability was set at p < 10-4 for each group and 10-3 for the interaction, cluster size > 150ul. Coordinates in parentheses indicate negative activations.

Although TBI subjects did not show significant evidence of slowing on neuropsychological tasks (see Table 1.1), the reaction time differences between our groups on 3 of the 4 experimental tasks suggested that slowed processing speed during alphabetization may have contributed to a cumulative enhancement of the haemodynamic

27 response. To test this hypothesis, we extracted ROI time series data for the 5 letter set sizes for the Alphabetize and Maintain conditions (where we expected speed of processing differences to be most prominent) in two anterior lateral PFC ROIs demonstrating reliable Group by ED interactions. In right GFi, differences in BOLD signal between the two conditions peaked between 8-10 seconds after stimulus onset in both groups (Figure 1.3, panel D.1), whereas time to peak occurred earlier in our TBI sample in the left GFm ROI (6-8 seconds after stimulus onset; Figure 1.3, panel D.2). Neither of these anterior lateral PFC regions demonstrated a significant interaction of group (TBI, Control) X trial time (TR1- TR8) [RGFi: F(1,7) = 1.83, p < .05; LGFm: F(1,7) = 1.67, p < .05]. Thus, we could find no evidence that ‘time on task’ significantly altered the haemodynamic response in our TBI patients relative to the control group. In sum, our group-level and single-subject ROI analyses confirm that the neural signature of executive control in working memory is altered following TBI and that these brain changes cannot be accounted for by differences in behavioural performance, variability within our patient sample or time on task differences.

28 Figure 1.4: Region of interest (ROI) analysis indicating mean % change difference between Alphabetize and Maintain conditions for the control group and individual TBI participants. Right IFG (BA 44)

Left MFG (BA 46)

2.5

3

2

2.5

Left MFG (BA 8) 2

1.5 2 1.5 1 1.5 1 1

0.5

0.5

Alphabetize vs. Maintain (% chg)

0.5 0 0 0 - 0.5

-0.5

-1

- 0.5

Set 3

Set 5

-1

Left MFG (BA 6) 3

Set 3

Set 5

-1

Left iPL (BA 40) 2

Set 3

Set 5

Left AG (BA 39) 2.5

2.5

2 1.5

2

1.5 1

1.5

1

1

0.5

0.5

0.5

0 0

0

-0.5 - 0.5

-0.5

-1

-1

Set 3

Set 5

-1

Set 3

Set 5

-1.5

Set 3

Set 5

Notes: Scatter plots for each ROI include (i) group mean (filled squares) and 95% confidence interval of the mean (error bars) for the control sample and (ii) individual scores for TBI participants (empty squares). Data is presented for three and five letter sets separately.

Discussion Our results demonstrate that DAI following TBI is associated with altered functional brain response during executive control processing in working memory. Previous studies investigating the functional neuroanatomy of working memory in TBI have been confounded by heterogeneity in patient selection or task characteristics (e.g. Christodoulou et al., 2001; McAllister et al., 1999). Our patients had evidence of DAI by both clinical radiological and volumetric criteria. None had neuroradiological evidence of significant

29 focal lesions. We were thus able to specifically examine the impact of DAI on executive control processes to a greater degree than heretofore possible. While the patients had sustained moderate to severe brain injury, all had returned to pre-injury employment or academic status and were several years post-injury, thus avoiding confounds related to recovery phase or performance differences. Finally, the functional neuroanatomy of our task is well characterized, has good convergent validity with other executive measures, and the design allowed us to independently manipulate load and executive control demands. By eliminating these sources of patient- and task-related variance, we showed that the functional neuroanatomy of executive control processing is specifically altered by DAI following moderate to severe brain injury. Follow-up analyses indicated that our results were reliable across the TBI sample and could not be attributed to slowing. These data are consistent with the characterization of working memory as an emergent property of coordinated activity within a distributed network of brain regions (Postle, 2006; Wager & Smith, 2003). We have demonstrated that diffuse injury, in the absence of focal brain pathology, is sufficient to alter brain networks subserving working memory processing. Our findings replicate earlier work demonstrating the involvement of PFC and left perisylvian regions during performance of healthy subjects on this task (Postle et al., 1999). However, we observed a significant group by executive demand interaction within lateral PFC regions bilaterally, as well as in left posterior parietal cortices. These data demonstrate that DAI is associated with augmented functional recruitment within this network of brain regions, particularly during tasks requiring high executive control. Moreover, equivalent group performance on our tasks suggests that executive control is facilitated by supplemental recruitment within these regions following DAI.

30 Altered functional recruitment following brain injury has been characterized as a decline in the efficiency of neural processing operations (Chen et al., 2006). TBI patients often report subjective changes associated with cognitively demanding tasks in spite of normal task performance, a dissociation that presents significant challenges with respect to diagnostics and treatment planning in the absence of measurable signs of cognitive dysfunction in patients with TBI, as well as patients with other neuropathologies causing diffuse or multifocal damage (e.g., dementia, multiple sclerosis). Reduced cognitive efficiency, as indexed by altered functional recruitment during neuropsychologically normal task performance, may serve as a novel metric, reflecting the costs of normal behavioural performance at the neural level. Indeed, these measures may correlate well with subjective complaints of patients with diffuse damage, but without demonstrable neuropsychological deficits. Two further clinical implications emerge from these findings. First, although augmented recruitment was observed here in a sample of well recovered patients without focal injury, we predict that patients with more severe behavioural or functional deficits would similarly demonstrate this pattern of enhanced neural recruitment during cognitively demanding tasks, albeit with diminishing behavioural gains. Such a pattern has recently been reported by Newsome and colleagues (2007) and suggests that these data may have clinical relevance across the spectrum of TBI severity. Second, TBI is commonly associated with deficits in arousal, chronic pain, anxiety and depressive symptomology, each of which has been associated with similar patterns of augmented functional neural recruitment to those reported here (see Hillary, Genova, Chiaravalloti, Rypma, & DeLuca, 2006 for a review). While the cumulative impact of such co-morbidity on higher cognitive functions following TBI has yet to be fully investigated, it is plausible that any reduction in neural processing

31 efficiency resulting from brain injury may be exacerbated by secondary impacts of co-morbid physical or psychological impairments. The impact of such interactions may impact the pace or extent of recovery in these patients and presents a challenge both for clinicians and researchers in assessing the implications of such co-morbidity. In sum, DAI is associated with augmented functional recruitment during high-level executive control tasks. Efforts to diagnose and remediate executive control deficits following TBI have been hampered by a dearth of assessment tools with sufficient sensitivity to detect executive deficits in a structured laboratory setting (Levine, Katz, Black, & Dade, 2002). We propose that functional brain data provides a potential neural biomarker of altered executive control functions that may be discernable even in patients without frank neuropsychological or functional deficits (see Chen et al., 2006 for a further discussion). Such neural biomarkers may be particularly beneficial in the armamentarium of clinicians working with TBI, where executive control deficits are often reported as a primary complaint but which remain frustratingly impervious to standard neuropsychological or structural neuroimaging investigations.

32 CHAPTER 2 Compensatory neural recruitment during verbal working memory performance after TBI: evidence for an altered functional engagement hypothesis.

Abstract There is increasingly convergent evidence that traumatic brain injury is associated with altered patterns of neural recruitment during performance of working memory tasks. Typically, this has manifest as increased recruitment of homologous regions of prefrontal cortex (e.g. right ventrolateral PFC during performance of a verbal working memory task). However the behavioural correlates of these functional changes are poorly understood. We used fMRI to examine neural activity in a sample of TBI patients and matched healthy controls performing working memory tasks with varying memory load and executive demand. In part one of this report, we correlated brain responses and task performance using multi-variate techniques. We identified networks within left and right PFC that uniquely and positively correlated with performance in our control and TBI samples respectively, demonstrating, for the first time in a TBI sample, evidence of compensatory functional recruitment. In part 2 of the report we investigated whether these compensatory brain changes were facilitated by functional reorganization (i.e. recruitment of novel brain regions not engaged by our control sample on any of our tasks) or altered functional engagement (i.e. differential recruitment of similar brain regions between our groups). This combined functional connectivity and brain-behaviour analysis provide strong support for the altered functional engagement hypothesis. Implications for our understanding of neuroplastic change and the design of rehabilitation interventions following brain injury are considered.

33 Introduction

There is increasingly convergent evidence that traumatic brain injury (TBI) is associated with altered patterns of neural recruitment during performance of working memory tasks (McAllister et al., 1999; Perlstein et al., 2004), typically manifested as increased activity in homologous regions of prefrontal cortex (e.g. right ventrolateral PFC during performance of a verbal working memory task: Christodoulou et al., 2001; McAllister et al., 1999) and more dispersed activity adjacent to areas implicated in task performance in noninjured controls (e.g. more extensive recruitment within left inferior and middle frontal gyri: Christodoulou et al., 2001). However, the functional implications of this altered recruitment are less well understood. Only two previous studies have directly investigated brain and behaviour correlations during working memory performance in TBI with conflicting results. McAllister et al. (1999) reported a positive correlation between activity in left frontal PFC regions and accuracy on a verbal working memory task (N-back) across all participants in their study (they did not report group specific correlations). In contrast, Newsome et al. (2007), using a similar paradigm, observed brain-behaviour correlations only in posterior cortices and only during the lowest working memory load. Several other reports investigating the neural correlates of working memory following TBI did not directly correlate brain and behavioural measures, although a review of the brain and behaviour data from these reports suggests similarly equivocal findings. Christodoulou and colleagues (2001) observed decreased lateralization (i.e. increased right PFC activity) and increased dispersion (i.e. increased activity in areas adjacent to those observed in healthy controls) during performance on a modified N-back task in a sample of moderate to severe TBI patients. However performance in the TBI group was significantly lower than that of

34 healthy matched controls, suggesting that this increased brain activation was not sufficient to support normal task performance in the TBI group. A similar result was observed by Perlstein et al. (2004), with increased right PFC activity insufficient to elicit comparable Nback performance between the control and TBI samples. In contrast, McAllister et al. (1999, 2001), observed increased right PFC activity with increasing working memory load in the context of equivalent group performance, suggesting that this additional recruitment may have been compensatory. Although there have been several consistently reported findings across the few studies that have examined functional brain activity in TBI (e.g. reduced hemispheric lateralization, see Hillary et al., 2006 for a review), the heterogeneity in range of TBI severity and chronicity likely affects interpretation of brain-behaviour patterns in these studies, irrespective of whether these were directly assessed. Moreover, performance variability is now considered a hallmark of neurocognitive functioning in TBI patients (Levine & Downey-Lamb, 2002; Stuss & Levine, 2002). Such variability, whether in brain or behavioural measures or both, serves to significantly dampen signal to noise ratios in correlational analyses and presents a significant barrier to the detection of consistent brainbehavior patterns in this population. In the current report, we investigate these correlations in a carefully screened sample of moderate to severe, chronic-phase TBI survivors, each of whom demonstrated good neuropsychological and functional recovery after their injury. All had evidence of diffuse axonal injury on neuroradiological report but were screened for focal lesions greater than 3mm. The patients were scanned with fMRI using a working memory paradigm that varied load and executive demands independently, permitting the examination of brain-behaviour correlations across multiple levels of working memory task demand. In Research Study 1

35 above, we replicated others’ findings of more widespread activation, including increased right frontal activity, associated with increased executive demands in the TBI patients. In this study, we adopted multivariate network analyses methods to identify networks of behaviourally relevant brain regions in our TBI group, as these methods are well-suited to the detection of functional neuroanatomical changes in distributed systems. Here we report the results of two multivariate analyses. The first directly examined the functional correlations between brain activity and performance on a verbal working memory task using spatial-temporal behavioural partial least squares (ST-bPLS, McIntosh, Chau, & Protzner, 2004). This technique has demonstrated enhanced statistical power relative to univariate approaches and is robust to outlier influences, thus allowing us to extract correlational data that might otherwise be obscured by sample heterogeneity using more traditional univariate analysis methods (McIntosh et al., 2004). The second analysis examined the functional connectivity of those brain regions demonstrating robust brain and behaviour correlations in study one using seed-based PLS analysis techniques. Our interest in this interaction of functional connectivity and brain-behavioural correlations emerged from observing the striking overlap between patterns of altered functional recruitment subserving working memory functioning after brain injury and in healthy subjects under conditions of high cognitive demand (Curtis & D'Esposito, 2003; Rypma, Prabhakaran, Desmond, Glover, & Gabrieli, 1999). These data suggest that TBI and cerebral challenge in the healthy brain may result in the engagement of similar brain regions to support task performance. To our knowledge, there exists no direct empirical data to support this (but see Hillary et al., 2006 for a conceptual review).

36 General Method

Participants Eight participants (6 males) with moderate to severe traumatic brain injury were recruited for this study. All subjects were part of the Toronto TBI study and were recruited based upon consecutive admissions to a level 1 trauma center. Further details with respect to the patient demographics, injury characteristics and recruitment inclusion and exclusion criteria of the larger sample have been reported elsewhere (Levine et al., in press). Demographic and injury characteristics for the subset of patients included in the present report have been described in Research Study 1 and are only briefly reviewed here. TBI participants had sustained closed head injury as a result of a motor vehicle accident and were in the chronic stage of recovery at the time of study participation. The injury severity profile of the group included 5 moderate and 3 severe TBI participants as determined by trauma GCS, loss of consciousness and extent of post-traumatic amnesia. Exclusion criteria included previous head injury, significant psychiatric history or evidence of current or recent alcohol and drug abuse. TBI patients were also excluded from participation if they had evidence of focal lesions greater than 3mm in diameter based on high resolution structural MRI. All TBI participants in the current study had evidence of DAI-related neuropathology (hemosiderin deposits) on neuroradiological report and each had demonstrated good functional recovery (return to work or school). Seven of the eight TBI participants underwent extensive behavioural testing as part of the larger Toronto TBI Study and data are reported in Chapter 1. Twelve (12) neurologically normal participants (8 males) were recruited for the current study. TBI and control participants were matched on age [t(18) = .785, p > .05; NS] and education [t(18) = -1.99, p > .05; NS]. All control participants were

37 right-handed and were native English speakers and were screened for previous neurological injury, history of psychiatric illness, or drug use.

Behavioural Task In the present study we employed a modified version of the Alphaspan protocol described by Postle and colleagues (1999) and based on earlier work by Craik (1986). For each task trial, participants were required to study a letter set consisting of either 3 or 5 consonant letter strings (set size or ‘load’ manipulation) and asked to either ‘MAINTAIN’ the letter set over a brief delay or ‘ALPHABETIZE’ the letters into their correct alphabetical position during the delay period (executive demand manipulation). At the end of the delay, a probe was presented consisting of a letter and an ordinal position (e.g. L-4? – ‘Was ‘L’ the 4th letter in the set?’). On Maintain trials, the probe referred to the ordinal position in the original letter set while on Alphabetize trials the probe referred to the letter position following alphabetization of the list. Probability of a correct probe was set at 0.5 for all trials in all conditions. Prior to scanning, each participant completed a training session consisting of step by step instructions for each task condition. Once there were no further questions for the administrator, all subjects completed 20 further trials (5 trials in each of the 4 conditions) immediately prior to entering the MR scanner. During scanning, participants completed 28 trials of each of the four task conditions (Alphabetize 3 & 5 letter sets, Maintain 3 & 5 letter sets) during a single scanning session. Within each session a total of four individual scans were acquired. Trials were grouped by executive demand with 2 blocks of seven trials at each level of executive demand presented during a single scan acquisition. Total stimulus onset asynchrony was 18000 ms (3 letter trials) or 19000 ms (5

38 letter trials). Each individual scan acquisition was 12 minutes in duration (see Chapter 1, Figure 1.1 for a full trial schematic).

fMRI scanning and analyses Scanning was performed at Sunnybrook Health Sciences Centre on a researchdedicated whole-body 3.0 T MRI system (Signa 3T94 hardware, VH3M3 software; General Electric Healthcare, Waukesha, WI) with a standard quadrature bird-cage head coil. Participants were placed in the scanner in supine position, with their head firmly placed in a vacuum pillow to minimize head movement. Earplugs were provided to reduce the noise from the scanner, and sensors were placed on participants’ right index finger and around the chest, to monitor heart rate and respiration. A volumetric anatomical MRI was performed before functional scanning, using standard high-resolution 3D T1-weighted fast spoiled gradient echo (FSPGR) images (TR/TE=7.2/3.1 ms, inversion-recovery prepared T1=300 ms, flip angle 15°, 256×192 acquisition matrix, 124 axial slices 1.4 mm thick, voxel size=0.86×0.86 cm, FOV=22×16.5 cm). Functional imaging was performed to measure the blood oxygenation level dependent (BOLD) effect (Ogawa et al., 1990). Scans were obtained using a single-shot T2*-weighted pulse with spiral in-out, achieving 26 slices, each 5 mm thick (TR/TE=2000/30 ms, flip angle 70°, 64×64 acquisition matrix, 26 axial slices 5 mm thick, voxel size=3.125×3.125, slice spacing=0, FOV=20×20 cm). Data processing was performed using Analysis of Functional NeuroImaging (AFNI) software (http://afni.nimh.nih.gov/, Cox, 1996). Time series data were spatially coregistered to correct for head motion by using a 3D Fourier transform interpolation. Motion-corrected images were then spatially transformed to an fMRI spiral scan template generated from 30 subjects scanned locally. This template was registered to the MNI305 template. The

39 transformation of each subject to the spiral template was achieved using a 12-parameter affine transform with sinc interpolation as implemented in SPM99 (http://www.fil.ion.ucl.ac.uk/spm/, Friston et al., 1995). Images were smoothed with an 8mm isotropic Gaussian filter before analysis. For each subject, “brain” voxels in a specific image were defined as voxels with an intensity greater than 15% of the maximum value in that image. The union of masks was used for group analyses as described below.

Partial Least Squares (PLS) Analyses The primary image analysis was done with spatio-temporal partial least squares (STPLS) (McIntosh, Bookstein, Haxby, & Grady, 1996). ST-PLS operates on the entire data structure at once, which requires that the data be in matrix form. One data matrix is made for each group. Within group, the rows of the data matrix are arranged as follows: condition blocks are stacked, and each subject has a row of data within each condition block. With n subjects and k conditions, there are n_k rows in the matrix. The columns of the data matrix contain the signal intensity measure at each voxel at each time point. The first column has intensity for the first voxel at the first time point, the second column has the intensity for the first voxel at the second time point. With m voxels and t time points, there are m _ t columns in the matrix. The haemodynamic response function (HRF) for any given condition normally lasts for several scans; thus a “lag window” is defined for a short signal segment within a trial that represents the response of each voxel. In the current experiment, the lag window is 6 (TR = 2; 12 s), beginning at trial onset minus 1 TR. The HRF for each trial is expressed as the intensity difference from trial onset. Two ST-PLS analyses were conducted. The first, behavioural ST-PLS (ST-bPLS), was the primary analysis to test the hypothesis of an interaction between task conditions and group in terms of brain–behaviour correlations.

40 This addressed the fundamental question as to whether those regions demonstrating functional relevance differed between our two groups. From this analysis we identified seed voxels to be used in the combined behaviour and functional connectivity analysis described in part two of this report. This second analysis used a behaviour-seed PLS (ST-sPLS) technique to address two questions: (i) does the behavioural relevance of brain regions demonstrating reliable brain-behaviour correlations depend upon their functional connectivity both inter- and intra-hemispherically and (ii) are these relationships altered by TBI?

General Results Behavioural Results We chose to examine accuracy as our behavioural measure for these analyses as we were principally interested in examining the functional brain correlates of executive processes (i.e. manipulation) within working memory. We maintain that accuracy provides a purer measure of the outcome of these processes and is less susceptible than reaction time measures to extraneous processes known to be active at probe (e.g. scanning, retrieval, motor speed, but see Hillary et al., 2006 for a discussion of this issue). Full behavioural analyses are presented in Chapter 1 and only the between group results relative to the current study are described here. These behavioural data were used for both study phases described below. One TBI patient was identified as an outlier for accuracy (TBI # 3646). We used a winsorizing procedure by which the outlier values for this subject’s accuracy on Alphabetize 3/5 trials were trimmed to 2 SD below the true mean of the sample (i.e. exclusive of the outlier). To confirm that our results were not unduly influenced by this outlier correction we

41 ran all analyses with this subject’s original data. There was a marginal but insignificant impact of the correction on the robustness of the behavioural PLS output (see ST-bPLS Results below). There was no impact on the behavioural results. Statistical analysis was carried out using a two group (control vs. TBI) x 2 (executive demand) x 2 (set size) repeated measures ANOVA. There was no main effect of group [F(1) = 2.54; NS] and no group x ED [F(1)=0.094, NS] or group x Set [F(1)=0.182, NS] interactions, indicating that the main effects of condition were stable across both the control and TBI groups. Post-hoc analyses revealed that there were no group differences on any of the tasks (p > .05, all comparisons).

fMRI analyses I: brain and behaviour correlations (behavioural PLS) In Chapter 1, we documented evidence of increased neural response during executive control processing in working memory in patients with ‘pure’ DAI (Chapter 1, Figure 1.3, Table 1.2). This was the case even though performance was matched between our two groups. However, few previous reports have directly investigated the link between these altered patterns of brain response and working memory behaviour in a TBI sample, leaving open the question as to whether these brain changes are compensatory, deleterious or incidental to task performance following TBI. The presence of brain changes in the context of preserved behavioural performance in our earlier report suggested that such changes are unlikely to be deleterious to task performance. However, it remains possible that different patterns of neural recruitment, for example within versus between cerebral hemispheres, may not be equivalently related to behaviour. Moreover, given that DAI has also been associated with widespread disruptions in cerebral microvasculature and metabolic changes (Povlishock & Katz, 2005), one might predict that the altered pattern of functional recruitment observed in our previous report may reflect physiological or systemic brain

42 changes that, while co-occuring with working memory processing, are nonetheless incidental to behavioural performance. To directly test the hypothesis that the altered functional recruitment observed during executive control processing in working memory following DAI is compensatory, we used ST-bPLS, a multivariate technique to assess brain-behaviour correlations simultaneously across the whole-brain. By adopting this technique, we were able to identify those brain regions that were correlated with behaviour in healthy control subjects and assess whether these brain-behaviour correlations are altered by DAI. Spatial-temporal behavioural partial least squares is a variant of ST-PLS that identifies latent variables (LVs) that capture task- and group-dependent patterns of brain– behaviour correlations (McIntosh et al., 2004). The correlation of behaviour measures and the fMRI signal is computed across subjects within each task, producing within-task brain– behaviour correlations for each of the four task conditions and the two groups. Singular value decomposition of the brain–behaviour correlation matrix produces three new matrices: voxel saliences, singular values, and task saliences. The variation across the task saliences indicates whether a given LV represents a similarity or difference in the brain–behaviour correlation across tasks and between our two groups. This can also be shown by calculation of correlation between the brain scores (dot product of the voxel salience and fMRI data) and behaviour data for each task. The voxel saliences give the corresponding spatiotemporal activity pattern. They are displayed as a singular image, which shows voxels that are weighted in proportion to the strength and direction (positive or negative) of their brain– behaviour correlation. Statistical assessment for ST-bPLS is done using permutation tests for the LVs and bootstrap estimation of standard errors for the voxel saliences. The permutation test assesses whether the effect represented in a given LV is sufficiently strong, in a statistical sense, to be different from random noise. The SE estimates of the saliences

43 from the bootstrap tests are used to assess the reliability of the nonzero saliences on significant LVs.

fMRI results I A two-group ST-bPLS with accuracy as the behavioural measure identified one significant LV (LV 1, p < .01) and a second that fell just below the standard threshold for statistical significance (p < .064). When we ran these analyses using the uncorrected data for TBI patient # 8, LV 2 fell just within the significance threshold (p < .05). However, we believe that the outlier corrected data represents a truer sampling of the subject’s task performance. For this reason, and to protect against a spurious inflation of the brainbehaviour correlations due to an extreme score for this subject, we report the more conservative, outlier corrected data throughout the remainder of this report. LV 1 identified regions of reliable and positive correlations between brain response and task accuracy across all tasks in the TBI sample. In our control sample, only performance on theA5 and M3 conditions was reliably correlated with activity in these regions. While the A5 correlation was robust and reliable (r=0.85), the M3 correlation was considerably smaller (r=0.48) and only marginally reliable based on bootstrap estimates of standard error, potentially reflecting near ceiling performance for controls on this task [X=0.96(0.03)] and will not be addressed further. Positive brain weights for LV 1 were observed in bilateral dlPFC, middle frontal gyrus (GFm), BA 46; right ventrolateral prefrontal cortex (vlPFC), inferior frontal gyrus (GFi), BA 44; anterior and posterior cingulate gyrus (GC), BA 32 & 30 respectively; left superior parietal cortex and bilateral regions of the TOJ, BA 18/19 near the fusiform (GF) and lingual gyri (GL). The second LV identified a main effect of group for the A3, A5 and M5 tasks. Dominant positive weights

44 (positive correlations with behaviour in the control group) were observed in left vlPFC, inferior frontal gyrus, BA 44 as well as in the supra-marginal gyrus (Gsm) of the left inferior parietal lobule (LPi), BA 40. In contrast, dominant negative weights (positive correlations with behaviour in the TBI group) were observed in right GFm, BA 46; right Gsm (BA 40) as well as right frontal polar cortex, BA 10, left insula, left precuneus (BA 7) and posterior cingulate gyrus. Single group analyses yielded similar results confirming these findings of a hemispheric dissociation in brain behaviour correlates between our two groups (Figure 2.1, Tables 2.1 & 2.2). Figure 2.1a. Brain regions demonstrating significant correlations between brain response and task accuracy in control participants (yellow clusters). (BSR = 4, p < 10-4; cluster = 15). Figure 2.1b. Brain regions demonstrating significant correlations between brain response and task accuracy in the TBI group. (BSR = 4, p < 10-4; cluster = 10). 8 4-6s

6-8s

8-10s -8 10

12

14

16

18

20

22

24

26

28

10

12

14

16

18

20

22

24

26

28

Notes: Arrows indicate cluster maxima that differentiated our two groups with respect to brain behaviour correlations during the Alphabetize tasks. These include left inferior and middle frontal gyri (controls) and two regions of right anterior middle frontal gyrus (TBI). These voxels were used as seeds to explore changes in functional connectivity in the subsequent series of analyses. Colour bar indicates bootstrap ratio range (BSR), a measure conceptually similar to a Z-score, that is used to determine reliability of the brain /behaviour correlation pattern.

45 Table 2.1: Cluster maxima from the behaviour PLS (ST-bPLS) analysis for the control group (See figure 2.1a). Lag 0 0

x -36 24

y -78 -67

z -10 -13

Anat GL GF

BA 18 19

BSR 7.8919 7.1583

Clust. 58 54

1 1 1 1 1 1 1 1

-67 -40 -63 -51 -24 44 -24 28

9 48 -46 -30 -55 -28 -78 -78

29 23 10 31 58 53 -13 -13

GFi GFm GTs Gsm PCu LPi GF GF

44 46/9 22 40 7 40 18 18

8.0633 6.5961 6.3414 6.6677 5.4227 5.273 7.4925 6.4417

20 22 18 44 27 30 108 75

2 2 2 2 2 2

-44 -24 -51 28 -44 -48

26 -59 -29 -90 -84 -63

-18 62 35 -16 23 -10

GFi PCu LPi GF GO GOm

47 7 40 18 19 19

6.7831 8.1431 5.9971 6.8112 6.4364 6.1715

21 17 24 91 26 17

3

48

-64

-27

Cereb

7.4412

58

Lag 0 0 0

x -8 4 55

y 70 70 -61

z -7 -7 33

Anat GFs GFs Gsm

BA 10 10 40

BSR -6.9307 -5.4135 -6.1099

Clust. 40 18 16

3 3

-20 -4

-36 -84

50 37

PCu CU

7 19

-8.9998 -7.0375

30 16

Notes: Cluster threshold was set by a bootsrap ratio > 4, p < .0001, and a minimum cluster size of 15 voxels. Lag refers to the temporal window with lag 0 corresponding to 2-4 seconds after stimulus onset. Each lag represents a 2 second time window. Voxel coordinates are reported in accordance with the stereotaxic atlas of Talairach and Tournoux (1988). Anat = anatomical region, abbreviations are consistent with those found in the atlas. BA=Brodmann area, BSR=bootstrap ratio, Clust.= cluster size (in voxels). Top section of the table indicates those brain regions that demonstrated a positive correlation with task accuracy. Bottom portion indicates those regions that were inversely correlated with accuracy in the control sample.

46 Table 2.2. Cluster maxima from the behaviour PLS (ST-bPLS) analysis for the TBI group (see figure 2.1b). Lag 0

x 24

y -72

z 29

Anat GO

BA 19

BSR 6.9401

Clust. 11

1

24

-73

26

GO

19

10.2363

19

2 2

24 40

-73 12

26 3

GO Insula

19

9.1954 7.3828

12 23

4 4

-59 -67

-9 -42

-20 9

GTi GTs

20 22

6.3222 6.1611

19 15

Lag 0 0 0 0

x -20 8 -20 0

y -82 -82 -25 -35

z -13 -13 -29 2

Anat GFm GL BS B. Stem

BA 18 18

BSR -6.2519 -5.7971 -6.978 -5.4666

Clust. 12 18 15 24

1 1 1

20 0 -20

-86 -25 -86

-13 -32 -19

GFm B. Stem Cereb

18

-7.1859 -10.132 -6.4683

22 16 19

2 2 2 2 2 2

40 36 -28 24 -20 12

47 21 -57 -86 -29 -29

16 25 25 -13 -36 -39

GFm GFm PCu GFm BS BS

46 46 18 18

-11.9796 -7.8931 -7.6894 -6.4147 -7.5555 -6.2537

28 12 14 13 16 19

3 3 3 3

44 36 -28 -24

47 21 -50 -56

16 25 17 54

GFm GFm GTm LPs

46 46 39 7

-9.9943 -6.0106 -5.0793 -7.1324

16 13 14 18

4

-16

-83

-26

Cereb

-5.2311

14

Notes: Cluster threshold was set by a bootstrap ratio > 4, p < .0001, and a minimum cluster size of 15 voxels. Lag refers to the temporal window with lag 0 corresponding to 2-4 seconds after stimulus onset. Each lag represents a 2 second time window. Voxel coordinates are reported in accordance with the stereotaxic atlas of Talairach and Tournoux (1988). Anat = anatomical region, abbreviations are consistent with those found in the atlas. BA=Brodmann area, BSR=bootstrap ratio, Clust.= cluster size (in voxels). Top section of the table indicates those brain regions that demonstrated a negative correlation with task accuracy. The bottom portion indicates those regions that were positively correlated with accuracy in the TBI sample.

To better represent this group X lateralization interaction within PFC, we extracted the brain-behaviour correlations for each task from those PFC regions demonstrating

47 maximal positive correlations in the ST-bPLS analysis for each group. This resulted in 4 regions of interest (ROI). For the TBI group, maximal saliency was observed in right PFC in an area bordering right GFm and GFi (46/45) and a more anterior GFm region bordering BA 46/10. For the control group, maximal saliency was observed in the left lateral PFC in a region bordering the inferior and pre-central frontal gyri (BA 44/6) and a more anterior segment of middle frontal gyrus (BA 9/46), which appeared homologous to the right anterior PFC region observed for the TBI sample. Figure 2.2 presents the correlational data for these bilateral anterior PFC regions for each task and group.

Brain / Behaviour (accuracy) correlations (r-values)

Figure 2.2. Correlations between brain activity in anterior middle frontal gyrus seed regions (highlighted in Figure 2.1). Talairach coordinate of left middle frontal gyrus (GFm): -40, 48, 23 ; right GFm: 40, 47, 16. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Left GFm Right GFm

A3

A5

M3

Controls

M5

A3

A5

M3

M5

TBI

Interim discussion I: evidence for compensatory functional recruitment following TBI We used ST-bPLS to investigate whether the altered functional recruitment of brain regions observed during working memory performance was related to task accuracy. We hypothesized, based on previous reports from the functional neuroimaging literature (e.g.

48 McAllister et al., 1999; Perlstein et al., 2004), that the recruitment of areas within right PFC, which we had reported previously in this TBI sample, would be positively correlated with task accuracy (i.e. compensatory). These data are consistent with our hypothesis. Activity in a right lateralized network, including two distinct regions of right lateral PFC in the vicinity of the anterior middle frontal gyrus on the border of BA 10/46 and a more dorsal region of GFm on the border of BA 44/45/46 was highly correlated with working memory task accuracy in our TBI group. Similar regions of right lateral PFC were observed by Perlstein et al. (2004) to track positively with working memory load in their TBI sample. This region has also been implicated in supraspan working memory performance in non-injured participants (e.g. Rypma et al., 1999) and during working memory performance in healthy aging (Erickson et al., 2007). However, these data represent the first time that activity in this region has been directly and positively correlated with performance on a working memory task in a TBI sample. This finding contrasts with the recent report of Newsome et al. (2007), who failed to find brain behaviour correlations in any PFC region during a demanding 2-back working memory task in a severe TBI sample. This is an unexpected finding given that PFC activity has often been associated with N-back performance, particularly during the 2-back condition (McAllister et al., 1999; McAllister et al., 2001). While the exact cause of this discrepancy between our two reports is unclear, it is notable that their TBI sample included only severe TBI participants who had extensive focal and diffuse brain damage and who were scanned in the acute injury phase (12-18 weeks postinjury). Both the severity and recency of the injuries in their sample would be expected to produce significant variability in both brain and behavioural responses, which may have attenuated correlations in their analyses. Moreover, the authors did report brain-behaviour correlations in posterior brain regions during the less demanding 1-back condition, where

49 overall variability would be expected to be lower, thus providing some support for this interpretation. In contrast to the right lateralized brain and behaviour correlations observed in our TBI sample, task accuracy in our control sample during the Alphabetize tasks and Maintain 5 task was predominantly associated with activity in a network of left lateralized brain regions, including dorsal and ventral aspects of PFC (BA 46/44), although this network expanded to include right lateral PFC during Alphabetize 5 trials. This pattern replicates many previous reports implicating vlPFC and dlPFC in working memory tasks involving a significant executive control component (Curtis & D'Esposito, 2003 and see Cabeza and Nyberg, 2000 for a review}. Figure 2.2 illustrates the striking group by hemisphere interaction evident in our part one results. These data are consistent with previous reports investigating working memory performance after TBI and extend those findings by demonstrating the functional relevance of this altered functional recruitment. fMRI analyses II: brain, behaviour and functional connectivity (behaviour & seed PLS). The results of study one supported our prediction that right hemisphere recruitment during working memory performance was compensatory for TBI participants and, consistent with previous report, this compensatory recruitment was also evident in healthy controls under conditions of high working memory demand (i.e. Alphabetize 5). This overlapping pattern of recruitment between our two groups suggested the possibility that the compensatory functional recruitment identified within PFC following TBI (Part I) does not represent the instantiation of novel PFC networks to support working memory performance after injury but rather reflects the unmasking of extant but behaviourally latent functional connectivity within PFC. To our knowledge, the behavioural correlates of altered functional connectivity following TBI have not been investigated previously.

50 To examine this question, we entered the 4 PFC ROIs identified in our first series of analyses as seeds into a combined behaviour and functional connectivity analysis. Each of these regions (left inferior frontal gyrus, left middle frontal gyrus, right anterior middle frontal gyrus, right posterior middle frontal gyrus) demonstrated highly reliable and significant correlations with behaviour that differentiated our two groups. Given the common overlap of right PFC regions associated with brain damage (or the highest level of working memory demand in our control sample), we predicted that the supplementary recruitment of right PFC that has been commonly reported following TBI represents the activation of existing, but behaviourally latent, bilateral PFC networks (i.e. functional engagement) as opposed to the instantiation of novel neural networks (i.e. functional reorganization). To test this prediction, we used combined behaviour and seed analysis (Grady et al., 2003 and see below) to characterize brain areas that are concurrently behaviourally relevant and functionally connected to each of these seed regions, thereby distinguishing behaviourally latent from behaviourally relevant brain networks in our two groups across a range of working memory task demands. We conducted spatial-temporal seed PLS (ST-sPLS, McIntosh et al., 2004) analyses to identify latent variables (LV) that capture task- and group- dependent changes in functional connectivity between the seed ROI and the rest of the brain (i.e. brain-seed correlations). The correlation of the fMRI signal for the seed and for the rest of the brain is computed across subjects within groups within each task, resulting in a matrix of within group and within-task brain-seed correlation maps. Singular value decomposition of the brain-seed correlation matrix produces three new matrices: the singular image of voxel saliences, singular values, and task saliences. The variation across the task saliences indicates whether a given LV represents a similarity or difference in the brain-seed correlation across

51 tasks. This can also be shown by calculation of correlation between the brain scores (dot product of the voxel salience and fMRI data) and seed fMRI signal for each task. The voxel saliences give the corresponding spatiotemporal activity pattern. Statistical assessment is similar to that used for ST-bPLS. We combined this seed analysis with a behavioural PLS (ST-bPLS, described above) in a single analysis in order to simultaneously identify patterns of functional connectivity and behavioural correlations that were both common to and divergent between our two groups.

fMRI results II We entered the 4 behaviourally relevant seeds within PFC into the combined seed and behavioural analysis. Specifically, we were interested in identifying those patterns of brain activity demonstrating a reliable main effect of group or group by task interaction for both behaviour and connectivity with the seed region. We present the results from each of the four seeds individually below followed by a summary of the findings for these analyses. Left inferior frontal gyrus (GFi, BA 44/6). LV 1 (p < .001) reflected behaviour and seed correlations for the Alphabetize and Maintain 3 tasks in the control sample (Figure 2.3, Table 2.3). For the TBI group, no task demonstrated both behavioural and seed correlations; however seed correlations alone were stable for all tasks. Positive correlations with both task accuracy and left GFi activity for controls but only GFi activity in TBIs were observed in left GFm and right inferior parietal lobule as well in left occipital cortex and bilateral cerebellar cortices (Figure 2.3, Table 2.3).

52

Figure 2.3. Brain regions demonstrating reliable and positive correlations with left inferior frontal gyrus (BA 44/6) activity and task accuracy during Alphabetize and Maintain 3 trials in the control group but only with seed activity in the TBI group (LV 1).

2-4

4-6

Brain Scores

Controls

TBI

50

50

0

0

-50

-50

-100

-100

r =-0.86

0.7

0.8

0.9

r =-0.19

1.0

0.6

0.7

0.8

0.9

Brain Scores

Accuracy (Alphabetize 5) 50

50

0

0

-50

-50

-100

r =-0.85

-100 0

0.5

1.0

r =-0.64

0

0.5

1.0

L.GFi activity (Alphabetize 5) Notes: Only the task and seed activity correlation was reliable in TBI participants. (BSR = 6, p < 10-4; cluster = 5). Image volumes range from Z=10, Z=28 (step=2).

53 Table 2.3. Cluster maxima from the combined behaviour and seed (ST-bPLS & ST-sPLS) analysis for the left inferior frontal gyrus seed (see figure 2.3) (LV 1). Lag 0 0 0

x -40 -63 44

y 44 9 -25

z 24 22 45

Anat GFm GFi LPi

BA 46/9 44 40

BSR -7.7347 -10.0308 -8.5795

Clust. 10 10 7

1 1 1 1 1

-36 -67 40 -24 16

48 9 -21 -48 -59

27 29 42 -18 -17

GFm GFi LPi Cereb Cereb

46/9 44 40

-8.7742 -21.0005 -7.3339 -7.4733 -8.1558

9 19 17 9 10

2 2 2

-67 -51 12

9 -63 -67

29 -10 -17

GFi GOm Cereb

44 19

-16.6289 -9.0696 -7.7589

12 8 11

4

-67

9

29

GFi

44

-9.3991

8

Notes: Cluster threshold was set by a bootstrap ratio of 6, p < 10-4; cluster = 5. Lag refers to the temporal window with lag 0 corresponding to 2-4 seconds after stimulus onset. Each lag represents a 2 second time window. Voxel coordinates are reported in accordance with the stereotaxic atlas of Talairach and Tournoux (1988). Anat. = anatomical region, abbreviations are consistent with those found in the atlas. BA=Brodmann area, BSR=bootstrap ratio, Clust.= cluster size (in voxels). Coordinates in bold type represent areas immediately adjacent to the seed voxel.

Left middle frontal gyrus (GFm, BA 46). LV 1 (p < .001) demonstrated reliable behaviour and seed correlations for both Alphabetize 3 and 5 tasks in the control group. As in the left GFi ROI, TBIs demonstrated reliable correlations only with seed activity for all tasks. Brain regions demonstrating positive correlations with both task accuracy and left GFm activity for controls but only left GFm activity in TBIs were observed in left inferior frontal gyrus, left inferior parietal lobule, bilateral superior parietal lobule and bilateral TOJ. Right posterior middle frontal gyrus (BA 46/44). LV 1 (p < .001) demonstrated a reliable pattern of behaviour and seed correlations for the Alphabetize and Maintain 5 conditions for the TBI group. Only the seed correlations were reliable for controls. Regions demonstrating positive correlations with both Alphabetize task accuracy and seed activity in TBI but only seed activity in controls were observed in bilateral anterior middle

54 frontal gyrus (BA 46/10), left posterior cingulate gyrus, left precuneus (BA 7) and left posterior middle temporal gyrus (BA 39; Figure 2.4, Table 2.4). Figure 2.4: Brain regions demonstrating reliable and positive correlations with right posterior middle frontal gyrus (BA 46/44) and task accuracy (LV 1).

2-4

4-6

Brain Scores

Controls

TBI

50

50

0

0

-50 -100

-50 r = -0.21 0.7

-100 0.8

0.9

1.0

r =-0.50 0.6

0.7

0.8

0.9

0.7

0.8

0.9

Brain Scores

Accuracy: Alphabetize 3 50

50

0

0

-50

-50

-100

-100

r = -0.59 0.7

0.8

0.9

1.0

r =-0.59 0.6

Accuracy: Alphabetize 5 Notes: Seed activity and accuracy correlations were reliable during Alphabetize and Maintain 5 tasks for TBI participants. For controls, only the seed correlations were reliable for the Alphabetize tasks. An inverse task/behaviour correlation was also observed for controls during Maintain 5 trials. (BSR = 6, p < 10-4; cluster = 5). Image volumes range from Z=10, Z=28 (step=2).

55 Table 2.4. Cluster maxima from the combined behaviour and seed (ST-bPLS & ST-sPLS) analysis for the right posterior middle frontal gyrus (BA 46/44) seed (LV 1). See figure 2.4.

Lag 0 0

x 32 -8

y 21 -33

z 25 -42

Anat GFm B. Stem

BA 46

BSR -9.3272 -7.2393

Clust. 13 17

1 1

36 -20

21 -57

25 21

GFm GC

46 30

-14.6958 -9.4378

19 18

2 2 2 2 2

40 -48 36 -24 -24

51 47 21 -48 -57

16 12 25 50 25

GFm GFi GFm PCu GC

46 46 46 7 31

-10.6319 -8.9892 -38.2769 -11.8536 -14.2909

11 13 33 20 37

3 3 3

36 -32 -16

20 -61 -60

21 21 47

GFi GTm PCu

45 39 7

-10.2296 -8.88 -10.3049

17 17 30

4

-20

-67

59

LPs

7

-7.7757

13

Notes: Cluster threshold was set by a bootstrap ratio of 6, p < 10-4; cluster = 5. Lag refers to the temporal window with lag 0 corresponding to 2-4 seconds after stimulus onset. Each lag represents a 2 second time window. Voxel coordinates are reported in accordance with the stereotaxic atlas of Talairach and Tournoux (1988). Anat. = anatomical region, abbreviations are consistent with those found in the atlas. BA=Brodmann area, BSR=bootstrap ratio, Clust.= cluster size (in voxels). Coordinates in bold type represent areas immediately adjacent to the seed voxel.

Right anterior middle frontal gyrus (BA 46/10). LV 1 (p < .001) expressed reliable behaviour and seed correlations for both groups. For the control group, correlations with behaviour and activity in the seed region were reliable only for the Alphabetize 5 task whereas this pattern was reliable for both Alphabetize 3 and 5 as well as the Maintain 5 tasks in the TBI group. Positive brain weights (i.e. regions demonstrating positive correlations with behaviour and seed) were observed in right posterior middle frontal gyrus, left anterior middle frontal gyrus, posterior cingulate gyrus, left superior parietal lobule and the inferior occipital cortex and TOJ bilaterally.

56 Interim discussion II. Evidence for altered functional engagement within right lateral PFC as a mechanism of compensatory reorganization. We used a combined behaviour and seed PLS analysis to test our hypothesis that the compensatory functional recruitment identified within PFC following TBI in study 1 did not represent true reorganization (i.e. the instantiation of novel PFC networks to support working memory performance) but rather reflected the engagement of extant but behaviourally latent functional connectivity within PFC. The results of these four analyses using task activity from highly behaviourally salient seed regions in left and right PFC are consistent with this functional engagement hypothesis. Moreover, these data suggest a differential pattern of step-wise functional recruitment of right PFC regions between our two groups as working memory task demands increased (see Figure 2.5 for a schematic representation of these results). Engagement of left lateral PFC regions (GFi and GFm) was associated with baseline working memory demands (i.e. Maintain 3 in our protocol) in both groups. However, when working memory demands were increased, either by increasing load (Maintain 5) or executive demand (Alphabetize 3), right anterior GFm became functionally connected to this baseline network in TBI and controls. However, the addition of this right hemisphere node was only behaviourally relevant in our control group. For the TBI sample, it is only when an additional region of right middle frontal gyrus at the border of BA 46/44 was incorporated into the network that activity in these regions was positively associated with performance on the higher demand working memory tasks. Importantly, this expanded network of right PFC regions is functionally connected in the control group during these tasks but is only behaviourally salient when both executive demand and load are increased from baseline levels (i.e. Alphabetize 5). These data suggest that bilateral functional connectivity within PFC is evident during working memory performance whenever load is

57 increased or when executive demands are present. However, the threshold at which this expanded right PFC network is functionally necessary for working memory performance, is reduced following TBI. Figure 2.5. Conceptual representation of combined behavioural and seed partial least squares analysis (study 2). b

Control

a

L

c d

b a

c d

b a

c d

a. L. GFi (-67, 9, 29) b. L GFm (-40, 48 23) c. R. GFm (40, 47, 16) d. R. GFm (36, 21, 25)

R

b a

c d

b a

c d

b a

c d

TBI Baseline (Maintain

+ Load or + Alpha

+ Load and + Alpha

Notes: Dashed lines connecting seed regions represent functional connectivity between these regions. Solid lines signify both functional connectivity and correlations between activity in the network and task accuracy. + Load or + Alpha corresponds to Maintain 5 letter and Alphabetize 3 letter conditions, respectively. + Load and + Alpha corresponds to Alphabetize 5 letter condition. a. Left inferior frontal gyrus (BA 44/6), b. left anterior middle frontal gyrus, c. right anterior middle frontal gyrus, d. right posterior middle frotnal gyrus. Coordinates in parentheses are in the space of Talairach and Tourneaux (1988).

General Discussion

There is increasingly convergent evidence that traumatic brain injury is associated with altered functional recruitment during the performance of cognitively demanding tasks such as those engaging executive control processes in working memory. However, the functional implications of this altered neural activity have been poorly characterized. To

58 address this issue directly, we conducted two analyses to investigate: (i) whether altered patterns of neural recruitment observed following TBI are functionally adaptive (i.e. compensatory) and if so, (ii) do these functional brain changes represent the instantiation of novel neural networks following brain trauma (i.e. true functional reorganization) or the activation of existing networks that are behaviourally inert in the undamaged or underchallenged brain (i.e. altered functional engagement). To address the first question we used multivariate network analysis (ST-bPLS) to identify patterns of brain activity that correlated with task accuracy. We identified two dominant patterns of brain activity that corresponded with task accuracy. The first described a network comprised of bilateral anterior GFm regions, an area of right GFm posterior and superior to the first, anterior cingulate and regions of the posterior, inferior temporal cortex bilaterally. Activity in this network of brain regions was highly positively correlated with accuracy on the Alphabetize 5-letter task for controls and all tasks in our TBI sample. A second significant pattern differentiated our two groups with respect to brainbehaviour correlations on the Alphabetize and Maintain 5-letter set size conditions. Areas of left inferior frontal gyrus and left posterior parietal cortex were positively correlated with activity in our control group. In contrast, performance on these same tasks were positively correlated with activity in right lateral PFC along the border of GFM and GFi, right posterior parietal and left lingual gyrus in our TBI group. These results provide strong support for our original hypothesis that recruitment of right lateral PFC during cognitively demanding tasks, as has now been reported in several studies, is compensatory. Few reports to date have directly correlated brain and behaviour measures in a TBI population. Our findings are consistent with those of McAllister et al. (1999) who observed positive correlations between activity in left GFi and accuracy on a

59 working memory task; a pattern we also observed in our healthy controls (they did not report group-specific correlations). However, our data contrast with those reported more recently by Newsome and colleagues (2007), who failed to observe significant correlations between brain response and performance during the 2-back condition of an N-back working memory task. This task is similar to that used by McAllister et al. (1999), where maximal brain response during the 2-back condition was observed in their sample of mild TBI patients. We suggest that patient heterogeneity in their sample may have limited the ability of Newsome et al. (2007) to detect significant correlations. While other reports examining functional brain changes during working memory performance following TBI did not report direct brain and behaviour correlations (e.g. Christodoulou et al., 2001; Perlstein et al., 2004), each of these studies reported recruitment of right lateral PFC regions during working memory task performance, consistent with our findings. Moreover, consistent with the strong group by laterality interaction we observed in our own data (Figure 2.2), Perlstein et al. (2004) also reported a group by laterality interaction, whereby increased working memory load was associated with increased activity in left lateral PFC for controls and right lateral PFC for TBI. To our knowledge, no previous report has investigated the impact of altered functional connectivity on working memory performance following TBI. We were particularly interested in these questions of functional connectivity, as our TBI sample was carefully screened for evidence of focal brain pathology, leaving us with a relatively ‘pure’ sample of DAI patients. Higher cognition is increasingly understood to be emergent from patterns of activity within large scale neural networks anchored by nodes within PFC (Grady et al., 2003; McIntosh, 1999). Given extensive reports of working memory deficits following TBI (e.g. D'Esposito, Cooney, Gazzaley, Gibbs, & Postle, 2006; and see Turner & Levine,

60 submitted), we surmised that such networks subserving working memory might be particularly sensitive to DAI. To test this hypothesis, we again employed multivariate methods (a combined behaviour and seed PLS) to examine the functional connectivity and behavioural relevance of networks anchored by four PFC ‘seed’ regions, identified in study 1 as being highly correlated with performance on our working memory tasks. Here our question was whether the compensatory recruitment of right PFC regions represented instantiation of novel neural networks or the engagement of existing functional connections that are behaviourally latent in the undamaged or under-challenged brain. The results from part two of this report are consistent with this latter hypothesis. We have demonstrated that lateral regions of PFC, including anterior middle frontal gyri bilaterally, left GFi and the border zone of right inferior and middle frontal gyri, are functionally connected in both groups, though their behavioural relevance is differentially altered both by task demands and by brain injury (see Figure 2.5 for an overview of these results). Specifically, left lateral PFC activity was common to both groups during performance of our low load, low executive demand working memory task (Maintain 3). Increases in either load (moving from 3 to 5 letter set size) or executive demand (alphabetizing the letter set), resulted in the engagement of right PFC in both groups to support task performance. However, this additional recruitment of right PFC was more extensive in the TBI group, and included both anterior and posterior regions of right GFm. This same network of regions was also functionally connected in the control group, but the network was only behaviourally relevant when both load and executive demand increased (i.e. Alphabetize 5 trials). Results from study 2 demonstrate that increasing either load or executive demand resulted in recruitment of right anterior PFC in addition to more commonly reported left

61 PFC regions. But this expanded network was sufficient to support behavioural performance only in our control group. TBI patients required additional recruitment of right posterior lateral PFC to support performance on these tasks. However, when both executive demand and load increased (i.e. during Alphabetize 5 trials), control performance was correlated with activity in this broader PFC network. Thus the stepwise pattern of behaviourally relevant functional recruitment observed in our controls as working memory demands increased was truncated for TBIs, where maximal right PFC recruitment was observed with any increase in task demand. A recent review by Hillary and colleagues (2006), addressed this question of altered functional recruitment during working memory performance under conditions of ‘cerebral challenge’ and their conclusions provide important context for our findings. Their review included studies examining both individual differences related to organic brain damage or psychological states such as fatigue or stress, and task-based differences such as increased working memory load or executive demands. As with the current report, the authors observed a consistently reported pattern of altered functional recruitment under all challenging conditions with increased recruitment of right PFC being the dominant feature. In contrast to the present findings, however, they concluded that this altered functional recruitment was not compensatory but rather was consistently associated with poorer performance on working memory tasks. They argue that this additional recruitment is therefore not directly compensatory but rather represents the engagement of cognitive control mechanisms to facilitate the establishment of new task subroutines. Our own data strongly suggest that the right hemisphere recruitment is directly compensatory, consistent with reports by McAllister et al. (1999, 2001). While it is outside the scope of this report to review data from other neurological populations, we would argue that, in the TBI literature,

62 increased right dlPFC activity represents functional recruitment that is fundamentally compensatory in nature. In two studies of TBI across the severity spectrum (McAllister et al., 1999; Current results, see Chapter 1), when task performance was equated, there remained clear evidence of increased right lateral PFC activity in the TBI groups. Moreover, the data reported here provide direct evidence that right lateral PFC activity is positively correlated with task performance (Figure 2.2). For those reports where performance was not equivalent, the indirect observation of an inverse relationship between performance and brain activity does not rule out a facilitative contribution to task performance from increased right PFC engagement. Indeed, Hillary and colleagues (2006) argue that the right PFC may compensate for altered task performance under conditions of cerebral challenge by implementing cognitive control processes to support novel task processing routines. Our findings are not inconsistent with this overall proposal. However, there remains a fundamental discrepancy between the two reports as to whether right PFC recruitment is associated with better or poorer working memory performance. While the publication of further data directly correlating brain and behaviour measures would be helpful in resolving this discrepancy, in the final section of the report we propose a tentative explanation that would serve to unify these accounts and provide some insight into the potential clinical ramifications of these data. As described above, our patient group represented a highly selective sample of TBI survivors. There was no evidence of macroscopic focal brain lesions, all participants were many years post-injury, and all had demonstrated good functional recovery. However, all of our TBI patients had evidence of diffuse axonal injury on neuroradiological report and we have previously reported quantitative structural MRI data demonstrating highly significant decreases in total volumes of white and gray matter in our TBI group compared to an age-

63 and education-matched sample (Chapter 1). This neuropathological profile is similar to that reported in studies of healthy aging, where diffuse white matter pathology and generalized reductions in gray and white matter have been observed (e.g. Raz et al., 2005; Tisserand et al., 2002). Interestingly, studies examining the functional neuroanatomy of working memory in healthy aging (e.g. Erickson et al., 2007; e.g. Rypma & D'Esposito, 2000) also report similar patterns of compensatory right prefrontal recruitment as we observed in our TBI sample. This convergence of neuropathology and functional recruitment patterns suggests that diffuse axonal injury may mimic healthy aging, at least with respect to the functional neuroanatomy of higher cognition. Hillary et al. (2006) propose that the compensatory nature of right prefrontal activity in healthy aging may reflect gradual alterations in neural networks more consistent with true functional reorganization over the life-span. This may also be the case in our TBI patients, each of whom was many years post-injury and, as is often the case in healthy aging, had few functional or neuropsychological impairments. Although we raise this possibility by way of resolving a conflict between our findings and the conclusions of Hillary et al. (2006), if this convergence truly reflects similar mechanisms of neuroplastic change in TBI and healthy aging, our results raise important questions with respect to the neurobehavioural sequelae and potential limitations of neuroplastic change as TBI and aging processes interact in long-term brain injury survivors.

Conclusion

In this report, we used multivariate methods to demonstrate that the altered functional recruitment we had reported in a previous investigation of executive control in working memory following TBI was compensatory. Increased activity in a network of PFC

64 regions, including right (TBI) and left (Controls) lateral PFC in the vicinity of the middle frontal gyrus, was positively correlated with accuracy on our working memory tasks, suggesting a significant group by task asymmetry in brain-behaviour correlations following TBI. Moreover, we were able to show, for the first time in this population, that this altered pattern of functional recruitment may not represent true cortical reorganization after brain injury but rather engagement of existing neural networks within PFC that are similarly engaged during increased working memory demands in healthy subjects (albeit at a lower threshold of task demand for the TBI participants). This characterization of altered functional recruitment following TBI as engagement rather than reorganization may have important therapeutic implications. Benefits of rehabilitation interventions based on a principle of effortful functional engagement have recently been demonstrated in healthy aging (Mahncke et al., 2006). Moreover, the correspondence of findings in our TBI sample with that reported in studies of normal aging raise intriguing questions with respect to the nature of, and capacity for, neuroplastic change in aging following diffuse axonal injury.

65

CHAPTER 3 Diffuse axonal injury as a mechanism for functional brain changes following traumatic brain injury: an integrated diffusion and functional imaging study.

Abstract Functional neuroimaging studies of patients with traumatic brain injury (TBI) have demonstrated compensatory functional recruitment during working memory performance, particularly on tasks requiring executive control processing. As diffuse axonal injury (DAI) is a primary pathology in TBI, we hypothesized that such alterations in functional recruitment occur in response to DAI. In this report, we use diffusion tensor imaging methods to quantify DAI and to in turn relate this to compensatory functional brain changes following brain trauma. Diffusion-weighted imaging data were acquired in a sample of moderate to severe TBI subjects with DAI and without large focal lesions and age- and education-matched, healthy controls. Fractional anisotropy (FA) values were derived from each subject’s diffusion-weighted images and compared between groups at the whole brain level and within regions of interest (ROI) placed in the genu and splenium of the corpus callosum for each subject. Interactions amongst FA, functional brain activity and working memory performance were investigated using both univariate and multivariate methods. Although FA was reduced in TBI patients in the genu and splenium of the corpus callosum and corresponding fibres in the forceps minor and major, these changes were uncorrelated with working memory performance. However, genu FA was negatively correlated with functional activation in a network of regions previously identified as compensatory, including right lateral prefrontal cortex. Activation in this region was also negatively correlated with FA in long-fibre tracts in the left hemisphere but positively correlated with FA in the right anterior limb

66 of the internal capsule. These results suggest that FA is reduced following TBI, even in the absence of focal brain pathology, and these changes in white matter integrity are related to altered functional brain response following TBI. Specifically, compensatory activity within right lateral PFC during working memory performance was associated with reduced FA in anterior callosal fibre tracts and increased FA in the right internal capsule, suggesting that functional engagement of this region may be facilitated through two complementary processes: (i) reduced trans-callosal inhibition and (ii) attentional control facilitated by relatively preserved thalamo-cortical fibre pathways.

67 Introduction

Although traumatic brain injury (TBI) is among the most common causes of neurologic disability, the precise relationships between brain injury characteristics and behavioural changes are unclear. One approach to this issue has been to challenge TBI patients with cognitive tasks while monitoring their brain activity with functional neuroimaging, such as fMRI. Recent studies of this sort have documented changes in activity associated with TBI, especially increased prefrontal activity during performance of cognitive tasks (for review see Levine et al., 2006 and see Chapters 1, 2). It is hypothesized that changes in functional brain activity following TBI are due to diffuse axonal injury (DAI), a primary neuropathology of TBI affecting brain interconnectivity. Accordingly, we observed increased working memory-related activation in a sample of TBI patients with significant DAI that were carefully screened for focal lesions that can confound such analyses (Chapters 1, 2). As the patients performed normally on the working memory task, we suggested that the altered functional brain activity may reflect a compensatory response to DAI. However, direct evidence linking DAI and functional brain changes in TBI is lacking. In this study, we use diffusion tensor imaging (DTI) methods, combined with functional neuroimaging methods, to probe the mechanisms underlying cognitive recovery in TBI patients with DAI. By integrating these imaging methods in a single set of analyses we aimed to characterize the contribution of DAI to both functional brain changes and behavioural performance following TBI. DTI is emerging as an important technique for investigating changes in white matter integrity following TBI (Arfanakis et al., 2002; Huisman et al., 2004; Xu, Rasmussen,

68 Lagopoulos, & Haberg, 2007). DTI uses magnetic resonance methods to create images based on the characteristics of water diffusion in the brain (Basser & Pierpaoli, 1996). The technique, based on the principles of Brownian motion of water through tissues, takes advantage of the fact that diffusion displacement distances are comparable with cell dimensions, allowing for the characterization of cell integrity or pathology. This is particularly true in myelinated axonal fibres where diffusion across the axon is smaller than along its length, meaning that water diffusion in white matter pathways of the brain is anisotropic (i.e. unequal in all dimensions). Thus DTI may be particularly suited to identifying and characterizing DAI resulting from TBI where rotational and acceleration/deceleration forces often lead to disruptions in axonal membrane integrity and permeability (Salmond et al., 2006). Diffusion-weighted imaging has provided important insights into the distribution and extent of DAI in vivo, often displaying better sensitivity to white matter abnormalities than traditional neuroimaging methods, particularly in the acute injury phase (Huisman et al., 2004). Diffusion indices also correlate well with measures of injury severity (Ptak et al., 2003) and outcome (Huisman et al., 2004; Wilde et al., 2006), suggesting that these methods may be important for characterizing neural and functional recovery both acutely and in chronic TBI. Similar correlations have been shown in other populations where white matter changes have been observed, such as multiple sclerosis (Audoin et al., 2007) and healthy aging (Madden et al., 2007). Fractional anisotropy (FA), a reliable, rotationally invariant scalar metric for measuring diffusion anisotropy (Alexander, Lee, Lazar, & Field, 2007; Basser & Pierpaoli, 1996, 1998), is the most commonly reported index of diffusion following brain injury and it has been shown to be the most discriminating scalar index of diffusion in TBI (Benson et al.,

69 2007). Reduced FA (i.e. poorer diffusion along white matter fibre tracts) has been most commonly reported in the corpus callosum, brain stem, fornix and sub-cortical white matter following TBI (Huisman et al., 2004; Salmond et al., 2006). This pattern of neuropathological change shows good concordance with data from experimental models of TBI and post-mortem investigations examining DAI–related injury throughout the brain (Povlishock & Jenkins, 1995). In the first section of this report we assess the sensitivity of DTI to white matter changes in a chronic TBI sample without focal lesions or gross neuropsychological or functional impairment. There are now several published reports investigating TBI-related neuropathology using diffusion-weighted imaging (Kraus et al., 2007; Nakayama et al., 2006; Xu et al., 2007). However, to our knowledge, no previous studies have explicitly investigated this question in a well recovered, chronic phase, ‘pure’ DAI sample. Next we investigate the relationship of white matter FA to performance on a working memory task that has served as a behavioural assay for investigating functional brain changes in our earlier reports (Chapter 1). FA has previously been shown to positively correlate with TBI patients’ overall mental status (e.g. Folstein Mental Status Exam, Nakayama et al., 2006) as well as with performance in specific domains of cognitive function, such as learning and memory (Salmond et al., 2006), inhibitory control and speed of processing (Wilde et al., 2006) and executive functioning and attention (Kraus et al., 2007). However, other reports have failed to find significant correlations between FA and standardized intelligence and memory batteries or with performance on a complex measure of working memory and attention (Nakayama et al., 2006). In this study, we used a verballybased paradigm known as Alphaspan (Craik, 1990) that enables independent manipulation of working memory load and executive control demands. Performance on this task has been

70 associated with functional brain response in a network of regions distributed both within and across hemispheres (Postle et al., 1999), thus allowing us to probe the contribution of FA in anterior-posterior long-fibre tracts and interhemispheric callosal fibres to working memory performance following TBI. In the final section of this investigation we examine the relationship of compensatory functional brain changes to white matter FA following TBI. Numerous studies have now reported functional brain changes during cognitive task performance following TBI (see Levine et al., 2006 for a review). The most commonly observed pattern involves reduced lateralization of functional brain response relative to healthy controls during performance of tasks typically associated with a highly lateralized brain response, particularly in regions of PFC (e.g. greater functional activity within right lateral PFC during an episodic memory task, Levine, Cabeza et al., 2002). In our prior study, the right lateral PFC was engaged by TBI patients with pure DAI at lower working memory task demands than controls; this regional activation was positively associated with working memory performance in the TBI patients (Chapter 2). As TBI patients’ task performance was equivalent to that of controls, we hypothesized that their activation patterns reflected the engagement of alternate or additional cortical resources in order to compensate for connectivity changes. We assessed this hypothesis by integrating FA into our brain-behavior analyses. While FA and fMRI datasets have been integrated with performance in some studies (Baird, Colvin, VanHorn, Inati, & Gazzaniga, 2005; Olesen, Nagy, Westerberg, & Klingberg, 2003), this has not been done in TBI (but for case study evidence, see Werring et al., 1998).

71 Method Participants Eight participants (6 males) with moderate to severe traumatic brain injury were recruited from consecutive admissions to a major urban trauma center as part of the Toronto TBI study. Further details with respect to the patient demographics, injury characteristics and recruitment inclusion and exclusion criteria of the larger sample have been reported elsewhere (Levine et al., in press). Demographic and injury characteristics for the subset of patients included in the present report have been described in Chapter 1 and are only briefly reviewed here. TBI participants had sustained closed head injury as a result of a motor vehicle accident and were in the chronic stage of recovery at the time of study participation. The injury severity profile of the group included 5 moderate and 3 severe TBI participants as determined by trauma GCS, loss of consciousness and extent of posttraumatic amnesia. Exclusion criteria included previous head injury, significant psychiatric history or evidence of current or recent alcohol and drug abuse. TBI patients were also excluded from participation if they had evidence of focal lesions greater than 3mm in diameter based on high resolution structural MRI. All TBI participants in the current study had evidence of DAI-related neuropathology (hemosiderin deposits) on neuroradiological report and each had demonstrated good functional recovery (return to work or school). Seven of the eight TBI participants underwent extensive behavioural testing as part of the larger Toronto TBI Study; these data are reported in the earlier paper (Chapter 1). Twelve (12) neurologically normal participants (8 males) were recruited as a comparison sample. Diffusion-weighted imaging data was unavailable for one of these participants. The absence of this participant did not alter our previously reported matching of age and education between our two groups [age: t(17) = .785, p > .05; NS; education: t(17) = -1.99, p > .05;

72 NS]. All comparison participants were right-handed, native English speakers and were screened for previous neurological injury, history of psychiatric illness, or drug use.

Behavioural Task We employed a modified version of the Alphaspan protocol described by Postle and colleagues (Postle et al., 1999) and based on earlier work by Craik (Craik, 1986). For each task trial, participants were required to study a letter set consisting of either 3 or 5 consonant letter strings (set size or ‘load’ manipulation) and asked to either ‘MAINTAIN’ the letter set over a brief delay or ‘ALPHABETIZE’ the letters into their correct alphabetical position during the delay period (executive demand manipulation). At the end of the delay, a probe was presented consisting of a letter and an ordinal position (e.g. L-4? – ‘Was ‘L’ the 4th letter in the set?’). On Maintain trials, the probe referred to the ordinal position in the original letter set while on Alphabetize trials the probe referred to the letter position following alphabetization of the list. Probability of a correct probe was set at 0.5 for all trials in all conditions. Prior to scanning, each participant completed a training session consisting of step-by-step instructions for each task condition. Once there were no further questions for the administrator, all subjects completed 20 further trials (5 trials in each of the 4 conditions) immediately prior to entering the MR scanner. During scanning, participants completed 28 trials of each of the four task conditions (Alphabetize 3 & 5 letter sets, Maintain 3 & 5 letter sets) during a single scanning session. Within each session a total of four individual scans were acquired. Trials were grouped by executive demand with 2 blocks of seven trials at each level of executive demand presented during a single scan acquisition. Total stimulus onset asynchrony was 18000 ms (3 letter trials) or 19000 ms (5 letter trials). Each individual scan acquisition was 12 minutes in duration.

73

Neuroimaging methods fMRI. Functional imaging scans were acquired at Sunnybrook Health Sciences Centre on a research-dedicated whole-body 3.0 T MRI system (Signa 3T94 hardware, VH3M3 software; General Electric Healthcare, Waukesha, WI) with a standard quadrature bird-cage head coil. Participants were placed in the scanner in supine position, with their head firmly placed in a vacuum pillow to minimize head movement. Earplugs were provided to reduce the noise from the scanner, and sensors were placed on participants’ right index finger and around the chest, to monitor heart rate and respiration. A volumetric anatomical MRI was performed before functional scanning, using standard high-resolution 3D T1-weighted fast spoiled gradient echo (FSPGR) images (TR/TE=7.2/3.1 ms, inversion-recovery prepared T1=300 ms, flip angle 15°, 256×192 acquisition matrix, 124 axial slices 1.4 mm thick, voxel size=0.86×0.86 cm, FOV=22×16.5 cm). Functional imaging was performed to measure the blood oxygenation level dependent (BOLD) effect (Ogawa et al., 1990). Scans were obtained using a single-shot T2*-weighted pulse with spiral in-out, achieving 26 slices, each 5 mm thick (TR/TE=2000/30 ms, flip angle 70°, 64×64 acquisition matrix, 26 axial slices 5 mm thick, voxel size=3.125×3.125, slice spacing=0, FOV=20×20 cm). Data processing was performed using Analysis of Functional NeuroImaging (AFNI) software (http://afni.nimh.nih.gov, Cox, 1996). Time series data were spatially coregistered to correct for head motion by using a 3D Fourier transform interpolation. Motion-corrected images were then spatially transformed to an fMRI spiral scan template generated from 30 subjects scanned locally. This template was registered to the MNI305 template. The transformation of each subject to the spiral template was achieved using a 12-parameter affine transform

74 with sinc interpolation as implemented in SPM99 (http://www.fil.ion.ucl.ac.uk/spm, Friston et al., 1995). Images were smoothed with an 8-mm isotropic Gaussian filter before analysis. For each subject, “brain” voxels in a specific image were defined as voxels with an intensity greater than 15% of the maximum value in that image. The union of masks was used for group analyses as described below.

Diffusion-Weighted Imaging. We used a single-shot spin echo planar sequence (TR/TE, 7400/87.1 ms; slice thickness, 4 mm; field of view, 25 cm2; number of experiments, 5; pixel matrix: 256x256) for diffusion tensor analysis. Diffusion gradients (b=1000 s/mm2) were always applied on two axes simultaneously around the 180 degree pulses. Diffusion properties were measured along six non-collinear directions. Diffusion-weighted magnetic resonance images were transferred to a workstation supplied by the manufacturer (Advantage Workstation, GE Medical Systems); structural distortion induced by large diffusion gradients was corrected on the basis of T2-weighted echo–planar images (b=0 s/mm2). The six elements of the diffusion tensor were estimated in each voxel, assuming a monoexponential relationship between signal intensity and the b matrix. The eigenvectors and eigenvalues of the diffusion tensor were determined by using multivariate analysis. Fractional anisotropy maps were generated on a voxel-by-voxel basis using DTI Studio software (Johns Hopkins Medical Imaging Center, Baltimore, Md.). For voxel based analyses (i.e. using whole-brain FA maps), spatial normalisation is an essential pre-processing step. As the contrast of the fractional anisotropy map is different from that of T1-weighted and T2-weighted and other template images in the statistical parametric mapping software package (SPM99; Wellcome, Department of Cognitive Neurology, London, UK), it is necessary to make a fractional anisotropy template

75 to normalise the individual fractional anisotropy maps correctly. We employed a method described by Nakayama et al. (2006) and Xu et al. (2007) to transform fractional anisotropy (FA) maps for all of our participants into a common template space, thus allowing us to examine voxel-wise group differences in FA. Importantly, we extended the work of Nakayama et al. (2006) by limiting our analysis to only those voxels that had been classified as white matter. By including only white matter voxels, we avoided confounding differences in white matter integrity (FA) with differences in white matter volume. A fractional anisotropy template was created based on FA maps from all of our participants (patients and controls). Individual fractional anisotropy maps were made by the Advantage Workstation (GE Medical Systems) and transferred to a Windows workstation with SPM99 running on MatlabV.6.5 (Mathworks, Natic, Massachusetts, USA). T2-weighted echo–planar images of all participants were transformed to the T2-weighted template. The parameter of the transformation was applied to the fractional anisotropy maps. Normalised individual fractional anisotropy maps were smoothed with an 8-mm full-width at half-maximum isotropic Gaussian kernel. Individual fractional anisotropy maps were normalised to the fractional anisotropy template. The normalised fractional anisotropy images were smoothed with a 4-mm full-width at half-maximum isotropic Gaussian kernel. Once the images had been spatially normalised and smoothed, group comparisons (two samples t test) were applied to calculate the statistical significance between the control and patient group on SPM99 (section 1). These whole-brain white matter FA maps were also used in our combined behavioural, DTI and fMRI analyses in sections 2 and 3. We also drew regions of interest (ROI) within the corpus callosum on each participant’s FA map using callosal boundaries defined by the schema of Witelson (1989). To mitigate against partial volume effects and avoid conflating callosal volume loss with

76 changes in white matter fibre tract integrity, ROIs were selectively drawn over the genu and splenium of the corpus callosum where white matter voxels are most easily and reliably identified. We adopted a standard ROI volume (54 voxels: 3 in y- and z-planes; 3 on either side of the midline in the x-plane), centered in these two regions of the callosum using neuroanatomical markers defined by Witelson (1989). We confirmed that all ROIs fit within the boundaries of these callosal structures by overlaying the ROI masks onto each participant’s T2-weighted image.

Analysis methods 1. Changes in white matter diffusivity following TBI. We examined this question in three ways. First, we plotted a histogram of average white matter voxel counts at each FA value (from 0 to 1, binned in units of .01) for both of our groups (see Benson et al., 2007). Voxel counts in each bin were normalized by total white matter voxels for each subject and averaged across subjects in each group. The between group contrast in the proportion of voxels occuring at each FA value is plotted in Figure 3.1. A higher proportion of voxels at lower FA values relative to controls may be considered a gross marker of white matter damage. Next, we contrasted average whole-brain, white matter masked, FA maps for our two groups using the contrast function included with the statistical parametric mapping analysis software (SPM 2, Friston et al., 1995). Finally, to compare FA between groups using ROI methods, the FA values for each ROI were extracted and averaged for each subject within our two groups. 2. FA and working memory performance. We directly correlated average ROI FA values for each subject with performance (% correct) on each of our working memory tasks for both TBI and comparison groups. We probed this brain and behaviour relationship further in our

77 TBI sample by entering task accuracy during high load working memory tasks (i.e. M5 and A5) into a simple regression analysis with the whole-brain, white matter masked FA maps using SPM 2. Here we were interested in identifying areas where task performance predicted white matter FA on a whole brain basis. 3.0 FA and fMRI We used two complementary approaches to relate the FA data to the fMRI data. The first examined regional FA in relation to whole-brain fMRI data using multivariate methods. The second examined regional fMRI data in relation to whole-brain FA data using univariate regression. 3.1 Multivariate integration of regional FA and whole-brain fMRI using multivariate methods. We used multivariate analysis techniques (partial least squares – PLS, see McIntosh et al., 2004) to integrate the results from our previous functional imaging investigation with the diffusionweighted data from the current study. PLS analysis methods have been fully described elsewhere (McIntosh et al., 1996; McIntosh et al., 2004) and are only briefly reviewed here. Using singular value decomposition, PLS generates latent variables (LVs) that express relationships between patterns of whole brain activation and one or more of the following: a) task manipulations, b) behavioural performance, or c) other individual difference measures such as seed voxel activity (to examine whole brain co-activation with activation in a specific region) or, in the case of this paper, FA. The statistical significance of each LV was assessed by permutation testing. The stability of each brain region’s contribution to the LV was determined through bootstrap resampling (subjects were resampled 500 times). Brain regions were considered reliable if they had a ratio of salience to standard error (hereafter referred to as the bootstrap ratio, interpreted similar to a Z-score (34)) greater than 4, corresponding to 99.9% confidence limits.

78 In our previous study with this sample, a combined seed voxel and behavioural PLS demonstrated compensatory recruitment of activity in a right prefrontal ROI in the vicinity of BA 46/44 (i.e. the seed voxel). In the present study, we asked whether this compensatory brain activity was related to callosal FA by including FA values from our manually-drawn ROIs in the multivariate model. 3.2. Relating fMRI to whole-brain FA. As in section 2, we were interested in expanding from our ROI analyses to explore the relationship between compensatory brain activity and FA at the whole-brain level. Following methods initially described by Baird et al. (2005), we extracted a measure of functional brain response from the right dlPFC seed region during performance of the most challenging condition (Alphabetize 5) for each TBI subject and entered it into a simple regression analysis with whole brain white matter FA maps using SPM 2. The goal of this analysis was to reveal regions where the level of functional brain response (that we had previously identified as compensatory during the A5 task), was correlated, either positively or negatively, with white matter FA across the whole brain.

Results

Based upon the analyses outlined above, we asked the following three questions to determine whether white matter changes were related to compensatory functional recruitment following TBI. 1. Is white matter FA reduced in well-recovered, chronic-phase TBI? We compared FA values between our two groups in three ways: (i) globally, by examining the relative distribution of white matter FA values; (ii) through whole-brain analysis, by examining regional differences in white matter using voxel-based methods and

79 (iii) using ROI methods to compare average FA between our two groups in ROIs drawn in the splenium and genu of the corpus callosum. In the first analysis, we observed global differences between the TBI and control groups in the whole brain distribution of voxel-wise FA values (Figure 3.1). Voxel counts peak at marginally significantly lower FA values in TBI patients as compared to healthy controls [TBI peak = 0.37; Control peak = .39. Difference = .02: t(16) = 1.404, p = 0.089, single tailed]. As seen in Figure 3.1, the proportion of white matter voxels is consistently higher for TBI than comparison subjects at lower FA values. However, this pattern is reversed as FA values increase. These results are consistent with findings recently reported by Benson et al. (2007) and demonstrates that global FA is reduced in our TBI sample, although detection of statistically reliable effects may have been hampered by low power. Figure 3.1: Group differences in voxel-wise distribution of white matter FA values.

FA values (Bin = 0-1.0; width =.01)

Notes: Voxel counts are normalized by total number of voxels for each subject after white matter masking. Bars represent differences between control and TBI groups with respect to the proportion of voxels occuring at each FA value (binned from .01 to 1.0).

Second, by directly comparing whole brain, white matter FA maps across groups, we identified significant differences at the juncture of callosal and anterior-posterior fibre pathways adjacent to the splenium (left) and genu (right). Significant differences were also

80 observed in the isthmus of the callosum, cingulum bundle, right cerebellum and in white matter underlying left posterior parietal cortex (t > 2.58, cluster size = 25 voxels; p < .05, cluster size corrected for multiple comparisons, Figure 3.2, Table 3.1). Importantly, the reverse contrast (i.e. Control FA > TBI FA) did not identify any suprathreshold voxels, suggesting that deceased FA in these regions is specific to the TBI participants. Figure 3. 2: Regions demonstrating significant differences in white matter FA between control and TBI participants. 10

c

d

b

b

a 0

No suprathreshold clusters were identified in the reverse contrast (i.e. Control > TBI). a=right cerebellum; b=right genu; c= right isthmus; d=left splenium. (t-contrast of parameter estimates, p < .01, 25 voxels; p < .05 corrected).

Table 3.1. Voxel cluster coordinate and maxima for the control versus TBI whole brain FA comparison (p < .01, 25 voxels; p < .01 corrected). No suprathreshold clusters were identified in the reverse contrast (i.e. Control > TBI). Cluster a. b. c. c. d. Not seen

X 14 14 -12 16 -22 -32

Y -74 38 -30 -4 -56 -20

Z -28 10 36 50 14 42

Size 35 104 67 27 25 80

Anatomical Location Cerebellum Genu (forceps minor) Isthmus / post.body Cingulum Splenium (forceps major) Sub-gyral (PPC)

Finally, we directly contrasted FA values extracted from the genu and splenium ROIs across groups. As expected, splenium FA was significantly lower in the TBI group relative to controls [t(17)= -2.895, p < .01] and genu FA was also reduced in TBI, although the

81 difference fell within the marginally significant range [t(17)= -1.958, p < .067]. To confirm that both the whole brain and ROI-based group results were convergent, we contrasted FA values from the whole brain analysis for voxels from the medial aspects of the genu and splenium corresponding to the ROI placements. Contrast t-values for these voxels in both genu and splenium regions were significant and negative [p < .05]. However, their spatial extent did not survive cluster thresholds for multiple comparison corrections and thus are not reported in Table 3.1.

2. Do changes in white matter FA predict working memory performance? To assess the relationship of FA to working memory performance, we examined correlations between FA values derived from the ROIs with working memory accuracy in the TBI participants and controls. No significant correlations with accuracy were observed in either anterior or posterior callosal ROIs for any of the working memory tasks (p > .05, all comparisons; see Figure 3.3 for ROI FA and Alphabetize 5 accuracy correlations).

82 Figure 3.3. Correlations between callosal FA and Accuracy on the Alphabetize 5 letter working memory task. No correlation (R-squared) was significant (p > 0.05). Genu FA

Splenium FA

TBI

R2 = -0.045

R2 = 0.055

R2 = -0.192

R2 = 0.027

Control

Accuracy (Alphabetize 5)

We further investigated whether whole brain FA was related to task performance by regressing task accuracy during Alphabetize and Maintain 5 tasks (A5, M5) onto whole brain FA. Only a single cluster of voxels demonstrated a significant and positive correlation between FA and accuracy and only for A5. This cluster was centered in an area of right medial posterior parietal cortex through which the posterior aspect of the cingulum bundle traverses (p < .01, cluster size = 25; p < .05, cluster size corrected for multiple comparisons).

83 3. Is reduced FA related to functional (compensatory) brain changes following TBI? In section 2 we established that there was no significant relationship between callosal or whole brain FA and working memory task performance. While one possible conclusion of this analysis is that DAI as indexed by FA does not relate to working memory performance, it is also possible that such a relationship is mediated by altered functional brain response. In other words, as a result of DAI, TBI patients engage alternate brain networks in order to maintain working memory task performance. We tested this hypothesis by correlating FA values in the anterior callosal ROI with working memory-related activation in a right posterior lateral prefrontal seed region (BA 44/46), previously shown to be correlated with working memory performance in our TBI sample (Chapter 2). As our previous functional connectivity and behavioural analysis emphasized patterns of altered functional recruitment within PFC, we focused the final series of analyses on the relationship between FA in the anterior corpus callosum and patterns of compensatory functional we had identified previously. Specifically, we conducted PLS analyses to examine the covariance structure between genu FA and seedvoxel network co-activation. To confirm that working memory performance did not significantly change this co-activation pattern (i.e. that behaviour and compensatory activation in right lateral PFC explained a common portion of the variance in brain response) we entered task accuracy into our model. This addition did not change the model results significantly. We therefore report results from the Genu—seed-voxel PLS that does not include the behavioural data. These analyses were extended in a second correlational analysis of whole brain FA and right prefrontal seed voxel activity. We predicted that activity in right lateral prefrontal cortex (and functionally connected regions in anterior middle frontal gyrus and posterior

84 parietal cortices) would be negatively correlated with callosal FA. If confirmed, this would provide evidence in support of a relationship between white matter DAI and altered functional response during working memory performance in TBI.

3.1 Partial least squares (PLS) analyses. Consistent with our prediction, the PLS analysis identified a network of brain regions that were functionally connected to our seed region in right lateral PFC and negatively correlated with FA in the callosal ROI. Importantly, in the TBI group, this pattern explained the largest portion of the variance (Genu-seed PLS: LV 1 = 32.21% of model variance, p < .001). Correlations with genu FA and seed activity were maximally differentiated during the high load working memory tasks (A5, M5 and marginally with M3) in the vicinity of the seed region as well as right anterior middle frontal gyrus, bilateral inferior parietal lobule and left temporal-occipital junction – a functional network that we had previously identified as compensatory for our TBI participants (Figure 3.4, Table 3.2).

85 Figure 3.4. Brain regions demonstrating reliable and positive correlations with right GFm seed (A5, M3, M5) and negative correlations with Genu FA (all tasks) in TBI subjects.

2-4

4-6

18

Brain Score

19

20

21

22

50

50

0

0

-50

-50

-100

r = - 0.56 -.70

-100 .74

Controls: Genu FA

.76

23

24

r = - 0.74 -.70

.74

.76

TBI: Genu FA

Notes: No reliable correlations were observed for any factor or task in controls. BSR = 4, p < .0001. (Min. cluster size = 15). Correlation plots present correlations between overall brain scores for each subject and group and genu FA during the Alphabetize 5 condition. Correlation was reliable for TBI but not controls.

86 Table 3.2. Suprathreshold activation cluster maxima from LV 1 (p < .001) in the Seed and Genu PLS analysis (see Figure 3.4). Gyrus GFd

BA 10

Lat. R

x 4

y 63

z 19

BSR 8.2685

Clust Size 64

Lag 2

10 10

R R

4 4

63 63

19 19

5.4935 5.321

21 17

3 4

10 10 9 46 / 44

R R R R

36 28 12 32

51 56 52 20

20 23 31 21

7.8056 5.5089 6.0768 9.3871

85 16 36 24

2 1 1 2

GFm

46

R

40

21

25

9.0666

18

1

GPrC GFs LPi

4 8 40

L L R

-16 -24 55

18 22 -22

54 50 23

5.2195 6.4578 5.8864

23 19 37

2 1 1

LPi LPi

40 40

R R

55 55

-18 -22

23 20

5.4297 6.0659

19 16

2 3

LPi GC GC GTs PCu PCu

40 24 31 22 7 7

L R R R R L

-32 4 20 28 12 -8

-36 -2 -33 -30 -25 -40

50 37 42 16 45 50

5.4958 6.1236 6.5077 5.8414 8.2193 9.2118

34 43 24 42 249 129

1 2 1 0 2 3

PCu PCu

7 7

L L

-12 -8

-40 -40

50 50

5.7872 5.9306

62 29

1 4

7 / 39 19 / 39 19 19 39 19 18 37

L L L L R L R R L L R

-20 -32 -28 -24 36 -44 8 40 -32 -20 32

-57 -61 -62 -73 -57 -85 -58 -51 -36 -51 -82

25 21 10 15 18 15 3 -14 -28 -14 -16

8.195 7.6538 7.4981 6.0536 5.646 6.9206 5.3186 6.7572 6.6985 5.6629 5.6185

70 74 33 72 18 47 16 68 34 31 16

2 1 3 0 2 3 0 3 3 1 2

GFd GFd

GFm GFs GFd GFm

GTm / PCu GTm GOm GOm GTm GTm GL Fusiform Cerebellum Cerebellum Cerebellum

Notes: Seed was placed in right lateral middle frontal gyrus (BA 46/44). Light grey font represents similar brain regions at different lag intervals. (GFd=medial frontal gyrus, GFm=middle frontal gyrus; GFs=superior frontal gyrus, GPrC=precentral gyrus; LPi=inferior parietal lobule; GC=cingulate gyrus; GTs=superior temporal gyrus; PCu=precuneus; GTm=middle temporal gyrus; GOm=middle occipital gyrus, GL=lingual gyrus).

When included in this model, working memory accuracy did not alter these results (Figure 3.5, Table 3.3), implying that neurofunctional changes were indeed mediating the impact of callosal FA loss on behaviour.

87 Figure 3.5. Brain regions demonstrating reliable and positive correlations with right GFm seed (A5, M3, M5) and negative correlations with Genu FA (all tasks) in TBI subjects.

2-4

4-6

18

Brain Score

19

20

21

22

23

50

50

50

0

0

0

-50

-50

-50

r = .42 0.6

0.7

0.8

Accuracy

0.9

r = 67 -.4 -.2 0

24

r = - .75 .2

.4

Seed activity

.6

0.6

0.7

0.75

0.8

Genu

Notes: Correlations with accuracy did not contribute to this pattern. BSR = 4, p < .0001. (Min. cluster size = 15). Correlation plots present correlations between overall brain scores and accuracy, seed activity and genu FA for each TBI subject. Only seed and genu FA correlations were significant. Accuracy did not contribute to this pattern.

88 Table 3.3. Suprathreshold activation cluster maxima from LV 1 (p < .001) in the Behaviour, Seed and Genu PLS analysis (TBI group only, see Figure 3.5). Seed was placed in right lateral middle frontal gyrus (BA 46/44). All abbreviations as in Table 3.2. Gyrus GFm GFm PCu GC PCu / CG LPs GTm GTm Cerebellum Cerebellum

BA 46 9 7 31 7 7 19 19 / 39

Lat. R R R L L L L L L R

x 36 32 9 -2 -13 -24 -24 -24 -30 21

y 45 14 -32 -35 -43 -59 -60 -67 -52 -82

z 27 31 45 33 40 49 20 16 -23 -25

BSR 6.8949 7.8619 6.1183 5.8776 4.8122 7.7053 6.599 7.8367 5.4126 5.8913

Clust Size 43 16 25 34 16 19 21 25 22 23

Lag 2 2 3 2 2 3 2 1 1 3

To summarize, activity in prefrontal and posterior areas functionally connected to a right lateral prefrontal seed region during working memory performance were associated with FA reductions in the anterior corpus callosum. These functional brain changes may therefore mediate the impact of structural brain changes on working memory performance following TBI, identifying a potential mechanism for behaviourally-relevant, altered functional recruitment in this population.

3.2. Correlating right lateral prefrontal activity with whole-brain, white matter FA The results reported in section 3.1 suggest that working memory task performance may be mediated by altered brain connectivity patterns in the presence of callosal DAI, as indexed by reduced FA. In this final set of analyses, we examined whether FA changes in other white matter pathways were associated with compensatory functional brain activity. Specifically, we correlated BOLD response with right lateral prefrontal seed voxel activity during the high load working memory tasks in our TBI sample with whole brain, white matter FA (Baird et al., 2005 and see Methods). The goal of this analysis was to (i) confirm our result from section 3.1 that anterior callosal FA was related to compensatory functional

89 recruitment in our TBI sample and (ii) explore whether FA in white matter outside of these callosal regions was related to this compensatory brain response. Consistent with our multivariate findings, seed voxel activity was significantly negatively correlated with white matter FA in the genu as well as in the splenium, intrahemispheric long-fibre tracts bilaterally, white matter underlying the left dorsolateral prefrontal cortex, right inferior parietal lobule, left temporal-occipital junction and the cerebellum. Positive correlations were observed in the anterior limb of the right internal capsule (A5; Figure 3.6, Table 3.4) as well as in external capsule bilaterally and in white matter underlying left superior parietal cortex (M5). Figure 3.6. Top panel . Regions where activity (i.e. BOLD response) in right middle frontal gyrus during Alphabetize 5 condition predicted lower white matter FA. Bottom panel. Activity in this same region predicted higher FA in ‘c’, anterior limb of internal capsule (t-contrasts of parameter estimates; p < .01, 25 voxels, corrected to p