UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Functional brain networks underlying working memory performance in schizophrenia : a multi-experiment… Sanford, Nicole A. 2019

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2020_may_sanford_nicole.pdf [ 12.46MB ]
Metadata
JSON: 24-1.0387449.json
JSON-LD: 24-1.0387449-ld.json
RDF/XML (Pretty): 24-1.0387449-rdf.xml
RDF/JSON: 24-1.0387449-rdf.json
Turtle: 24-1.0387449-turtle.txt
N-Triples: 24-1.0387449-rdf-ntriples.txt
Original Record: 24-1.0387449-source.json
Full Text
24-1.0387449-fulltext.txt
Citation
24-1.0387449.ris

Full Text

    FUNCTIONAL BRAIN NETWORKS UNDERLYING WORKING MEMORY PERFORMANCE IN SCHIZOPHRENIA: A MULTI-EXPERIMENT APPROACH  by  NICOLE A. SANFORD B.A., McGill University, 2010 M.Sc., University of British Columbia, 2014  A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Neuroscience)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  December 2019      © Nicole A. Sanford, 2019  ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the dissertation entitled:  Functional Brain Networks Underlying Working Memory Performance in Schizophrenia: A Multi-Experiment Approach  submitted by Nicole Sanford  in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Neuroscience   Examining Committee: Todd Woodward, Psychiatry Supervisor  Christine Tipper, Psychiatry Supervisory Committee Member  Alan Kingstone, Psychology University Examiner Anthony Herdman, Audiology and Speech Sciences University Examiner Martin Lepage, Psychiatry (McGill University) External Examiner   Additional Supervisory Committee Members: Rebecca Todd, Psychology Supervisory Committee Member Lawrence Ward, Psychology Supervisory Committee Member   iii Abstract Working memory (WM), defined as actively holding and/or manipulating information in mind, is central in guiding behaviour. WM is also a core domain of impairment in schizophrenia which substantially impacts functional outcome. Although the dorsolateral prefrontal cortex (DLPFC) has been implicated as a source of WM deficits in schizophrenia, these deficits may be better characterized within the framework of functional connectivity of distributed brain regions. However, in task-state functional magnetic resonance imaging (fMRI) research, the sluggish he-modynamic response hinders the separation of cognitive sub-processes in WM tasks. Moreover, findings from an individual fMRI task may not be broadly clinically meaningful even if reliable effects are detected. The present research used whole-brain, multi-experiment, functional con-nectivity analyses to obtain more refined characterizations of WM networks and their activity in schizophrenia across a variety of cognitive tasks. Study 1 demonstrated a novel method of com-bining a verbal WM task with a thought generation task, which produced a finer delineation of networks than when the WM task was analysed alone. Study 2 reported individual analyses of four tasks (i.e., verbal WM, thought generation, visuospatial WM, and set-switching Stroop tasks), providing basic characterizations of their dominant networks in healthy individuals. Final-ly, study 3 consolidated all four datasets into a unified analysis to examine differences between healthy controls and schizophrenia patients in the resulting networks, as well as correlations be-tween these networks and task performance. A visual attention network – engaged during encod-ing of memory sets, and diminished in patients – was associated with accuracy in the verbal and visuospatial WM tasks, and with WM capacity measured in separate out-of-scanner testing ses-sions. A frontoparietal network including the DLPFC – possibly underlying internally-oriented attention – exhibited hypoactivity in patients as expected, but was not correlated with behaviour- iv al WM measures. These findings suggest that dysfunction in a given network cannot be assumed to underlie poor task performance, as this may depend on the cognitive sub-process it supports. This work also demonstrates that a network may be concealed in an individual task when it does not account for a distinct portion of variance, yet may exhibit reliable activity when examined across multiple tasks.  v Lay Summary Working memory is a core aspect of cognitive impairment in schizophrenia which sub-stantially impacts daily living. This research examined brain networks engaged during working memory and other cognitive tasks in individuals with a diagnosis of schizophrenia and in healthy controls. A multi-experiment analysis was used which allowed for the examination of activity in a given network across different types of cognitive demands. This resulted in a finer separation of brain networks which had been obscured when analyzed using more conventional methods. While schizophrenia patients exhibited diminished activity in a number of networks, one net-work was particularly notable because its activity was correlated with accuracy in verbal and visuospatial working memory tasks, and with working memory span. As this network was en-gaged during the initial encoding of items to be remembered, the results suggest that working memory deficits in schizophrenia may be due to problems during early memory encoding.   vi Preface This dissertation is the original intellectual product of the author, Nicole Sanford. The analyses reported in this dissertation comprise secondary use of data from previously completed studies as well as new data collected for this work. All data collection carried out at the Universi-ty of British Columbia (UBC) was approved by the UBC Clinical Research Ethics Board (Certif-icate numbers: H07-02786 and H14-02687).  The Spatial Capacity (SCAP) Task data, analysed in Chapters 4 and 5, were obtained from the OpenfMRI database (https://openfmri.org/dataset/ds000030/; accession number is ds000030), and originally collected as part of the UCLA Consortium for Neuropsychiatric Phe-nomics LA5c Study (Poldrack et al., 2016). Data for the Thought Generation Task (TGT), ana-lyzed in Chapters 3-5, were obtained from a previously completed study (certificate H07-02786) carried out in the Cognitive Neuroscience of Schizophrenia Lab at UBC, under the supervision of Dr. Todd Woodward. Primary data collection of the verbal Working Memory (WM) task and the Task-Switch Inertia (TSI) task was also carried out under the supervision of Dr. Woodward (certificate H14-02687); I was involved in development of the research design, testing protocols, hypotheses, and assisted with data collection. Dr. Jennifer Whitman programmed the first ver-sion of the verbal WM task implemented in this work, and I added minor revisions to facilitate task instruction. I programmed the TSI task and carried out initial pilot testing. The work reported in Chapter 3, “Finer Separation of Working Memory Networks Using Multi-Experiment fMRI-CPCA”, has been accepted for publication in the journal Cortex as “Sanford, N., Whitman, J. C., & Woodward, T. S. W. (In Press). Task-merging for finer separa-tion of functional brain networks in working memory. Cortex.” Some revisions have been made for coherence with the rest of this dissertation. As first author, I designed and carried out the  vii analysis, and wrote the original manuscript. Dr. Whitman programmed the original version of the WM task as stated above, and was involved in discussions of the results and in suggesting revi-sions to an earlier draft. Dr. Woodward was the principal investigator on this research and con-tributed throughout the project with respect to design, analysis, and manuscript revisions.  viii Table of Contents Abstract ......................................................................................................................................... iii Lay Summary ................................................................................................................................ v Preface ........................................................................................................................................... vi Table of Contents ....................................................................................................................... viii List of Tables ............................................................................................................................... xv List of Figures .............................................................................................................................. xx List of Abbreviations ............................................................................................................. xxviii Acknowledgements ................................................................................................................... xxx Chapter 1: Introduction ............................................................................................................... 1 1.1. Neurocognitive Impairment in Schizophrenia ......................................................... 1 1.1.1. Background ................................................................................................... 1 1.1.2. Neurological underpinnings of working memory deficits in schizophrenia . 3 1.2. Functional Brain Networks and Working Memory Capacity .................................. 4 1.2.1. Functional brain networks underlying working memory .............................. 4 1.2.2. Functional connectivity during working memory in schizophrenia ............. 5 1.3. Measuring Task-State Functional Connectivity ...................................................... 7 1.3.1. Measuring connectivity with functional magnetic resonance imaging......... 7 1.3.2. Limitations of task-state connectivity research............................................. 8 1.3.3. Multi-experiment comparisons ................................................................... 10 1.4. Dissertation Overview ........................................................................................... 11 1.4.1. Aims ............................................................................................................ 11 1.4.2. Outline......................................................................................................... 12  ix Chapter 2: Methods .................................................................................................................... 13 2.1. Participants ............................................................................................................. 13 2.2. fMRI Tasks ............................................................................................................ 14 2.2.1. Working Memory Task (WM) .................................................................... 14 2.2.2. Spatial Capacity Task (SCAP) .................................................................... 15 2.2.3. Task-Switch Inertia Task (TSI) .................................................................. 15 2.2.4. Thought Generation Task (TGT) ................................................................ 18 2.3. fMRI Data Acquisition and Preprocessing ............................................................ 20 2.3.1. Data acquisition .......................................................................................... 20 2.3.2. Preprocessing .............................................................................................. 20 2.4. Constrained Principal Component Analysis for fMRI .......................................... 21 2.4.1. General framework ..................................................................................... 21 2.4.2. Matrix equations ......................................................................................... 22 2.4.3. Multi-experiment fMRI-CPCA ................................................................... 26 2.5. Chapter 2 Tables .................................................................................................... 29 2.6. Chapter 2 Figures ................................................................................................... 30 Chapter 3: Finer Separation of Working Memory Networks Using Multi-Experiment fMRI-CPCA................................................................................................................................. 34 3.1. Background ............................................................................................................ 34 3.2. Aims and Hypotheses ............................................................................................ 35 3.3. Methods ................................................................................................................. 37 3.3.1. Participants .................................................................................................. 37 3.3.2. Tasks ........................................................................................................... 38  x 3.3.3. Functional connectivity analysis ................................................................. 39 3.4. Results .................................................................................................................... 40 3.4.1. WM task fMRI-CPCA results ..................................................................... 40 3.4.2. Multi-experiment fMRI-CPCA results ....................................................... 42 3.5. Discussion of Multi-Experiment fMRI-CPCA ...................................................... 47 3.6. Chapter 3 Tables .................................................................................................... 52 3.7. Chapter 3 Figures ................................................................................................... 69 Chapter 4: Characterizing Dominant Networks Underlying Neurocognitive Tasks Using Single-Experiment fMRI-CPCA................................................................................................ 84 4.1. Aims and Hypotheses ............................................................................................ 84 4.2. Analysis 1: Working Memory Task ....................................................................... 86 4.2.1. Methods....................................................................................................... 86 4.2.2. WM task performance results ..................................................................... 88 4.2.3. WM functional connectivity results ............................................................ 88 4.2.4. Summary of WM task results ..................................................................... 91 4.3. Analysis 2: Spatial Capacity (SCAP) Task ............................................................ 92 4.3.1. Methods....................................................................................................... 92 4.3.2. SCAP task performance results .................................................................. 93 4.3.3. SCAP functional connectivity results ......................................................... 94 4.3.4. Summary of SCAP task results ................................................................... 97 4.4. Analysis 3: Task-Switch Inertia (TSI) Task .......................................................... 98 4.4.1. Methods....................................................................................................... 98 4.4.2. TSI task performance results .................................................................... 101  xi 4.4.3. TSI functional connectivity results ........................................................... 102 4.4.4. Summary of TSI task results ..................................................................... 104 4.5. Analysis 4: Thought Generation Task (TGT) ...................................................... 105 4.5.1. Methods..................................................................................................... 105 4.5.2. TGT functional connectivity results ......................................................... 107 4.5.3. Summary of TGT task results ................................................................... 109 4.6. Discussion of Single Experiment Analyses ......................................................... 110 4.7. Chapter 4 Tables .................................................................................................. 113 4.8. Chapter 4 Figures ................................................................................................. 140 Chapter 5: Identifying Networks Underlying Working Memory Deficits in Schizophrenia Using Multi-Experiment fMRI-CPCA .................................................................................... 161 5.1. Aims and Hypotheses .......................................................................................... 161 5.2. Methods ............................................................................................................... 162 5.2.1. Datasets ..................................................................................................... 162 5.2.2. Analysis..................................................................................................... 166 5.3. Task Performance Results ................................................................................... 172 5.3.1. WM task performance............................................................................... 172 5.3.2. SCAP task performance ............................................................................ 172 5.3.3. TSI task performance ................................................................................ 173 5.4. Overview of fMRI-CPCA Results ....................................................................... 175 5.5. Default Mode Network (Component 1) ............................................................... 176 5.5.1. Anatomical characteristics ........................................................................ 176 5.5.2. DMN: WM task results ............................................................................. 176  xii 5.5.3. DMN: SCAP task results .......................................................................... 177 5.5.4. DMN: TSI task results .............................................................................. 178 5.5.5. DMN: TGT task results ............................................................................ 179 5.5.6. Summary of DMN results ......................................................................... 179 5.6. Internal Attention Network (Component 2) ......................................................... 180 5.6.1. Anatomical characteristics ........................................................................ 180 5.6.2. Internal attention network: WM task results ............................................. 180 5.6.3. Internal attention network: SCAP task results .......................................... 181 5.6.4. Internal attention network: TSI task results .............................................. 183 5.6.5. Internal attention network: TGT task results ............................................ 184 5.6.6. Summary of internal attention network results ......................................... 184 5.7. Sensorimotor Network (Component 3)................................................................ 185 5.7.1. Anatomical characteristics ........................................................................ 185 5.7.2. Sensorimotor network: WM task results................................................... 185 5.7.3. Sensorimotor network: SCAP task results ................................................ 186 5.7.4. Sensorimotor network: TSI task results .................................................... 187 5.7.5. Sensorimotor network: TGT task results .................................................. 188 5.7.6. Summary of sensorimotor network results ............................................... 189 5.8. Motor Response Network (Component 4) ........................................................... 190 5.8.1. Anatomical characteristics ........................................................................ 190 5.8.2. Motor response network: WM task results ............................................... 190 5.8.3. Motor response network: SCAP task results ............................................ 191 5.8.4. Motor response network: TSI task results ................................................ 192  xiii 5.8.5. Motor response network: TGT task results ............................................... 193 5.8.6. Summary of motor response network results............................................ 193 5.9. Visual Attention Network (Component 5) ........................................................... 194 5.9.1. Anatomical characteristics ........................................................................ 194 5.9.2. Visual attention network: WM task results ............................................... 194 5.9.3. Visual attention network: SCAP task results ............................................ 195 5.9.4. Visual attention network: TSI task results ................................................ 197 5.9.5. Visual attention network: TGT task results .............................................. 198 5.9.6. Summary of visual attention network results............................................ 198 5.10. Occipital Network (Component 7) .................................................................... 199 5.10.1. Anatomical characteristics ...................................................................... 199 5.10.2. Occipital network: WM task results ....................................................... 199 5.10.3. Occipital network: SCAP task results ..................................................... 200 5.10.4. Occipital network: TSI task results ......................................................... 201 5.10.5. Occipital network: TGT task results ....................................................... 202 5.10.6. Summary of occipital network results .................................................... 203 5.11. Correlations Between Task Performance and HDR Increases/Decreases ......... 203 5.11.1. WM task performance and HDRs ........................................................... 203 5.11.2. SCAP task performance and HDRs ........................................................ 205 5.11.3. TSI task performance and HDRs ............................................................ 208 5.12. Discussion of 4-Task fMRI-CPCA Results ....................................................... 209 5.12.1. WM task .................................................................................................. 209 5.12.2. SCAP task ............................................................................................... 210  xiv 5.12.3. TSI task ................................................................................................... 211 5.12.4. TGT task ................................................................................................. 213 5.12.5. Task-related network functions............................................................... 214 5.12.6. Group differences in network activity .................................................... 218 5.13. Chapter 5 Tables ................................................................................................ 222 5.14. Chapter 5 Figures ............................................................................................... 242 Chapter 6: Conclusion .............................................................................................................. 297 6.1. Summary .............................................................................................................. 297 6.2. Comparison of Findings Across Chapters ........................................................... 298 6.2.1. Two- versus four-task CPCA results ........................................................ 298 6.2.2. Single- versus four-task CPCA results ..................................................... 299 6.3. Implications ......................................................................................................... 303 6.4. Limitations ........................................................................................................... 306 6.5. Future Directions ................................................................................................. 308 6.6. Conclusions .......................................................................................................... 309 Bibliography .............................................................................................................................. 311  xv List of Tables Table 2.1. fMRI acquisition parameters for all datasets included in present research. ................ 29 Table 3.1. Demographic information for the WM task and the TGT task datasets (from WM-TGT multi-experiment analysis with healthy controls only). ........................................... 52 Table 3.2. WM task fMRI-CPCA, response/attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. 53 Table 3.3. WM task fMRI-CPCA, default mode network (DMN, component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. 55 Table 3.4. WM task fMRI-CPCA, visual attention network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. ...... 58 Table 3.5. WM-TGT multi-experiment fMRI-CPCA, components 6 and 7 (occipital and auditory regions, respectively): Results of repeated measures analyses of variance (ANOVAs). . 60 Table 3.6. WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .................................................................................................................................. 61 Table 3.7. WM-TGT multi-experiment fMRI-CPCA, visual attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. ........................................................................................................ 63 Table 3.8. WM-TGT multi-experiment fMRI-CPCA, internal attention network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. ........................................................................................................ 65  xvi Table 3.9. WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each peak. .................................................................................................................... 67 Table 4.1. Demographic information for the WM task, SCAP task, TSI task, and TGT task datasets (from single-task analyses with healthy controls only). .................................... 113 Table 4.2. Mean WM, SCAP, and TSI task performance results (percent correct and mean reaction time for each condition; standard deviations in parentheses). .......................... 114 Table 4.3. WM task fMRI-CPCA, response/attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak.......................................................................................................................................... 115 Table 4.4. WM task fMRI-CPCA, visual attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 117 Table 4.5. WM task fMRI-CPCA, default mode network (DMN, component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak.......................................................................................................................................... 119 Table 4.6. SCAP task fMRI-CPCA, external attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak.......................................................................................................................................... 121 Table 4.7. SCAP task fMRI-CPCA, default mode network (DMN, component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak.......................................................................................................................................... 123  xvii Table 4.8. TSI task fMRI-CPCA, DMN/occipital network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 125 Table 4.9. TSI task fMRI-CPCA, response network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 127 Table 4.10. TSI task fMRI-CPCA, evaluation network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 129 Table 4.11. TGT task fMRI-CPCA, language network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 131 Table 4.12. TGT task fMRI-CPCA, posterior-medial DMN (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 132 Table 4.13. TGT task fMRI-CPCA, anterior/posterior-lateral DMN (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. ................................................................................................................................ 135 Table 4.14. TGT task fMRI-CPCA, auditory network (component 4): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 138 Table 5.1. Demographic information for healthy controls and schizophrenia patients in the WM and TSI tasks dataset (4-task fMRI-CPCA study). ......................................................... 222 Table 5.2. Demographic information for healthy controls and schizophrenia patients in the SCAP task dataset (4-task fMRI-CPCA study). ........................................................................ 223 Table 5.3. Demographic and clinical information for healthy controls and schizophrenia patients in the TGT dataset (4-task fMRI-CPCA study). ............................................................. 224  xviii Table 5.4. Mean Working Memory (WM) task performance for each participant group and full sample (percent correct and mean reaction time for each task condition; standard deviations in parentheses). .............................................................................................. 225 Table 5.5. Mean Spatial Capacity (SCAP) task performance for each participant group and full sample (percent correct and mean reaction time for each task condition; standard deviations in parentheses). .............................................................................................. 226 Table 5.6. Mean TSI task performance for each participant group and full sample (percent correct and mean reaction time for each condition; standard deviations in parentheses).......................................................................................................................................... 227 Table 5.7. 4-task fMRI-CPCA, default mode network (DMN, component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak.......................................................................................................................................... 228 Table 5.8. 4-task fMRI-CPCA, internal attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 230 Table 5.9. 4-task fMRI-CPCA, sensorimotor network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 232 Table 5.10. 4-task fMRI-CPCA, motor response network (component 4): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 233 Table 5.11. 4-task fMRI-CPCA, visual attention network (component 5): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .... 234 Table 5.12. 4-task fMRI-CPCA, occipital network (component 7): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. .................... 235  xix Table 5.13. 4-task fMRI-CPCA, default mode network (DMN, component 1): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. ......................................... 236 Table 5.14. 4-task fMRI-CPCA, internal attention network (component 2): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. ......................................... 237 Table 5.15. 4-task fMRI-CPCA, sensorimotor network (component 3): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. .......................................................... 238 Table 5.16. 4-task fMRI-CPCA, motor response network (component 4): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. .......................................................... 239 Table 5.17. 4-task fMRI-CPCA, visual attention network (component 5): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. .......................................................... 240 Table 5.18. 4-task fMRI-CPCA, occipital network (component 7): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. ................................................................ 241   xx List of Figures Figure 2.1. Working Memory (WM) task design; example of a trial presented in the 4-letter load condition with a 4-second delay. ...................................................................................... 30 Figure 2.2. Spatial Capacity (SCAP) task design; stimuli downloaded from software developer’s publicly available code library (https://poldracklab.stanford.edu/softwaredata). ............. 30 Figure 2.3. Task-Switch Inertia (TSI) task design. ....................................................................... 31 Figure 2.4. Thought Generation Task (TGT) design; examples of trials presented in a thought-generation block. ............................................................................................................... 32 Figure 2.5. Schematic overview of fMRI-constrained principal component analysis (fMRI-CPCA). .............................................................................................................................. 33 Figure 3.1. WM task fMRI-CPCA, response/attention network (component 1): Anatomical and temporal characteristics. ................................................................................................... 69 Figure 3.2. WM task fMRI-CPCA, default mode network (DMN, component 2): Anatomical and temporal characteristics. ................................................................................................... 70 Figure 3.3. WM task fMRI-CPCA, visual attention network (component 3): Anatomical and temporal characteristics. ................................................................................................... 71 Figure 3.4. Surface representations and WM HDR shapes in the single-experiment vs. the multi-experiment fMRI-CPCA. .................................................................................................. 72 Figure 3.5. WM-TGT multi-experiment fMRI-CPCA, blood flow artifact (component 5): Anatomical and temporal characteristics. ......................................................................... 73 Figure 3.6. WM-TGT multi-experiment fMRI-CPCA, occipital (de)activation (component 6): Anatomical and temporal characteristics. ......................................................................... 74 Figure 3.7. WM-TGT multi-experiment fMRI-CPCA, auditory network (component 7): Anatomical and temporal characteristics. ......................................................................... 75 Figure 3.8. WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Anatomical and temporal characteristics. ......................................................................... 76  xxi Figure 3.9. WM task from the WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Estimated HDRs illustrating delay × time interaction. ........................... 77 Figure 3.10. WM-TGT multi-experiment fMRI-CPCA, visual attention network (component 2): Anatomical and temporal characteristics. ......................................................................... 78 Figure 3.11. WM task from the WM-TGT multi-experiment fMRI-CPCA, visual attention network (component 2): Estimated HDRs illustrating delay × time interaction. ............. 79 Figure 3.12. WM-TGT multi-experiment fMRI-CPCA, internal attention network (component 3): Anatomical and temporal characteristics. .................................................................... 80 Figure 3.13. WM task from the WM-TGT multi-experiment fMRI-CPCA, internal attention network (component 3): Graphs illustrating effects of cognitive load and delay length. . 81 Figure 3.14. WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Anatomical and temporal characteristics. .................................................................... 82 Figure 3.15. WM task from the WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Graphs illustrating effects of cognitive load and delay length. .... 83 Figure 4.1. WM task fMRI-CPCA, response/attention network (component 1): Anatomical and temporal characteristics. ................................................................................................. 140 Figure 4.2. WM task fMRI-CPCA, response/attention network (component 1): Estimated HDRs illustrating delay × time interaction. ............................................................................... 141 Figure 4.3. WM task fMRI-CPCA, visual attention network (component 2): Anatomical and temporal characteristics. ................................................................................................. 142 Figure 4.4. WM task fMRI-CPCA, visual attention network (component 2): Estimated HDRs illustrating delay × time interaction. ............................................................................... 143 Figure 4.5. WM task fMRI-CPCA, default mode network (DMN, component 3): Anatomical and temporal characteristics. ................................................................................................. 144 Figure 4.6. WM task fMRI-CPCA, default mode network (DMN, component 3): Graphs illustrating effects of cognitive load and delay length. ................................................... 145  xxii Figure 4.7. SCAP task performance results, graphs illustrating effects of cognitive load and delay length............................................................................................................................... 146 Figure 4.8. SCAP task fMRI-CPCA, external attention network (component 1): Anatomical and temporal characteristics. ................................................................................................. 147 Figure 4.9. SCAP task fMRI-CPCA, external attention network (component 1): Graphs illustrating effects of cognitive load and delay length. ................................................... 148 Figure 4.10. SCAP task fMRI-CPCA, DMN (component 2): Anatomical and temporal characteristics. ................................................................................................................. 149 Figure 4.11. SCAP task fMRI-CPCA, default mode network (DMN, component 2): Graphs illustrating effects of cognitive load and delay length. ................................................... 150 Figure 4.12. TSI task fMRI-CPCA, DMN/occipital network (component 1): Anatomical and temporal characteristics. ................................................................................................. 151 Figure 4.13. TSI task fMRI-CPCA, DMN/occipital network (component 1): Graphs illustrating effects of stimulus congruency and task-switch condition. ............................................ 152 Figure 4.14. TSI task fMRI-CPCA, response network (component 2): Anatomical and temporal characteristics. ................................................................................................................. 153 Figure 4.15. TSI task fMRI-CPCA, response network (component 2): Estimated HDRs illustrating congruency × time interaction. ..................................................................... 154 Figure 4.16. TSI task fMRI-CPCA, evaluation network (component 3): Anatomical and temporal characteristics. ................................................................................................................. 155 Figure 4.17. TSI task fMRI-CPCA, evaluation network (component 3): Estimated HDRs illustrating congruency × time interaction. ..................................................................... 156 Figure 4.18. TGT task fMRI-CPCA, language network (component 1): Anatomical and temporal characteristics. ................................................................................................................. 157 Figure 4.19. TGT task fMRI-CPCA, posterior-medial DMN (component 2): Anatomical and temporal characteristics. ................................................................................................. 158  xxiii Figure 4.20. TGT task fMRI-CPCA, anterior/posterior-lateral DMN (component 3): Anatomical and temporal characteristics. ........................................................................................... 159 Figure 4.21. TGT task fMRI-CPCA, auditory network (component 4): Anatomical and temporal characteristics. ................................................................................................................. 160 Figure 5.1. WM task performance: Percentage of correct responses for each load condition, averaged over delay to illustrate load × group interaction. ............................................. 242 Figure 5.2. SCAP task performance: Percentage of correct responses for each load condition, averaged over delay to illustrate main effect of load. ..................................................... 242 Figure 5.3. TSI task performance: Percentage of correct responses for each task condition, illustrating significant congruency × task-switch interaction. ........................................ 243 Figure 5.4. 4-task fMRI-CPCA: Summary of components 1-5 and 7. ....................................... 244 Figure 5.5. 4-task fMRI-CPCA, blood flow artifact (component 6): Anatomical and temporal characteristics. ................................................................................................................. 245 Figure 5.6. 4-task fMRI-CPCA, artifact (component 8): Anatomical and temporal characteristics.......................................................................................................................................... 246 Figure 5.7. 4-task fMRI-CPCA, default mode network (DMN, component 1): Dominant 10% of component loadings ........................................................................................................ 247 Figure 5.8. WM task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots.......................................................................................................................................... 248 Figure 5.9. WM task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating effects of delay and load. ................................................................................................ 249 Figure 5.10. SCAP task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for all task conditions ............................................................................................. 250 Figure 5.11. SCAP task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating effects of cognitive load and delay length. ..................................................................... 251 Figure 5.12. TSI task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for all word-reading conditions. ...................................................................................... 252  xxiv Figure 5.13. TSI task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating effects of stimulus congruency and task-switch condition. ............................................ 253 Figure 5.14. TGT task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for both task conditions. .................................................................................................. 254 Figure 5.15. 4-task fMRI-CPCA, internal attention network (component 2): Dominant 10% of component loadings ........................................................................................................ 255 Figure 5.16. WM task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all task conditions................................................................... 255 Figure 5.17. WM task from the 4-task fMRI-CPCA, internal attention network (component 2): Graphs illustrating effects of cognitive load and delay length. ...................................... 256 Figure 5.18. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all task conditions................................................................... 257 Figure 5.19. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Graphs illustrating effects of cognitive load and delay length. ...................................... 258 Figure 5.20. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDRs illustrating group differences. ............................................................. 259 Figure 5.21. TSI task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all word-reading conditions. .................................................. 260 Figure 5.22. TSI task from 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDRs illustrating stimulus congruency and task-switch effects. .................. 261 Figure 5.23. TGT task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for both task conditions. .............................................................. 262 Figure 5.24. 4-task fMRI-CPCA, sensorimotor network (component 3): Dominant 10% of component loadings ........................................................................................................ 263 Figure 5.25. WM task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDR plots for all task conditions................................................................... 263  xxv Figure 5.26. WM task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDRs illustrating delay × time interaction .................................................... 264 Figure 5.27. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDR plots ...................................................................................................... 265 Figure 5.28. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Graphs illustrating effects of cognitive load and delay length. ...................................... 266 Figure 5.29. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Mean predictor weights illustrating load × group interaction, explained by significant difference in quadratic contrast. ...................................................................................... 267 Figure 5.30. TSI task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDR plots for all word-reading conditions. .................................................. 267 Figure 5.31. TSI task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDRs illustrating stimulus congruency and task-switch effects. .................. 268 Figure 5.32. TGT task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Estimated HDR plots for both task conditions. .............................................................. 269 Figure 5.33. 4-task fMRI-CPCA, motor response network (component 4): Dominant 10% of component loadings ........................................................................................................ 270 Figure 5.34. WM task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots for all task conditions................................................................... 270 Figure 5.35. WM task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots illustrating delay × time interaction ............................................. 271 Figure 5.36. WM task from the 4-task fMRI-CPCA, motor response network (component 4): mean predictor weights illustrating load × group interaction. ........................................ 271 Figure 5.37. SCAP task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots for all task conditions................................................................... 272 Figure 5.38. SCAP task from the 4-task fMRI-CPCA, motor response network (component 4): Graphs illustrating effects of delay length. ..................................................................... 273  xxvi Figure 5.39. SCAP task from the 4-task fMRI-CPCA, motor response network (component 4): Graphs illustrating group differences. ............................................................................. 274 Figure 5.40. TSI task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots for all word-reading conditions. .................................................. 275 Figure 5.41. TSI task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots illustrating stimulus congruency and task-switch effects. ........... 276 Figure 5.42. TSI task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDRs illustrating group × time interaction. .................................................. 277 Figure 5.43. TGT task from the 4-task fMRI-CPCA, motor response network (component 4): Estimated HDR plots for both task conditions. .............................................................. 278 Figure 5.44. 4-task fMRI-CPCA, visual attention network (component 5): Dominant 10% of component loadings ........................................................................................................ 279 Figure 5.45. WM task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDR plots for all task conditions................................................................... 279 Figure 5.46. WM task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDRs illustrating delay × time interaction. ................................................... 280 Figure 5.47. WM task from the 4-task fMRI-CPCA, visual attention network (component 5): Plots illustrating the load × time × group interaction. .................................................... 281 Figure 5.48. SCAP task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDR plots for all task conditions................................................................... 282 Figure 5.49. SCAP task from the 4-task fMRI-CPCA, visual attention network (component 5): Graphs illustrating effects of cognitive load and delay length. ...................................... 283 Figure 5.50. SCAP task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDRs illustrating group differences. ............................................................. 284 Figure 5.51. TSI task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDR plots for all word-reading conditions. .................................................. 285  xxvii Figure 5.52. TSI task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDRs illustrating stimulus congruency and task-switch effects. .................. 286 Figure 5.53. TGT task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDR plots for both task conditions. .............................................................. 287 Figure 5.54. TGT task from the 4-task fMRI-CPCA, visual attention network (component 5): Estimated HDRs illustrating group × time interaction. .................................................. 288 Figure 5.55. 4-task fMRI-CPCA, occipital network (component 7): Dominant 10% of component loadings ........................................................................................................................... 289 Figure 5.56. WM task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDR plots for all task conditions.................................................................................... 289 Figure 5.57. WM task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDRs illustrating delay × time interaction ..................................................................... 290 Figure 5.58. SCAP task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDR plots for all task conditions................................................................... 291 Figure 5.59. SCAP task from the 4-task fMRI-CPCA, occipital network (component 7): Graphs illustrating effects of cognitive load and delay length. ................................................... 292 Figure 5.60. TSI task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDRs for all word-reading conditions. ........................................................................... 293 Figure 5.61. TSI task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDRs illustrating effects of stimulus congruency and task-switch condition. ............... 294 Figure 5.62. TGT task from the 4-task fMRI-CPCA, occipital network (component 7): Estimated HDR plots for both task conditions. ............................................................................... 295 Figure 5.63. TGT task from the 4-task fMRI-CPCA, occipital network (component 7): Graphs illustrating group differences. ......................................................................................... 296  xxviii List of Abbreviations AHPQ Annett Hand Preference Questionnaire ANOVA Analysis of variance BA Brodmann area BOLD Blood oxygen level-dependent ci Task-switch from incongruent colour-naming block cn Task-switch from neutral colour-naming block (C)PCA (Constrained) principal component analysis CRT Cognitive remediation therapy DLPFC Dorsolateral prefrontal cortex DMN Default mode network DSM Diagnostic and Statistical Manual of Mental Disorders FIR Finite impulse response (f)MRI (Functional) magnetic resonance imaging HDR Hemodynamic response HRF Hemodynamic response function ICA Independent component analysis IQ Intelligence quotient ITI Inter-trial interval ITP Increase-to-peak MINI Mini International Neuropsychiatric Interview MNI Montreal Neurological Institute PLS Partial least squares  xxix ROI Region of interest RT Reaction time RTB Return-to-baseline SANS Scale for the Assessment of Negative Symptoms SAPS Scale for the Assessment of Positive Symptoms SCAP Spatial Capacity SD Standard deviation SEM Standard error of the mean SES Socioeconomic status SIRP Sternberg Item Recognition Paradigm SMA Supplementary motor area SSPI Signs and Symptoms of Psychotic Illness TGT Thought Generation Task TR Time to repetition TSI Task-Switch Inertia UBC University of British Columbia WAIS Wechsler Adult Intelligence Scale WI Incongruent word-reading sitmulus WM Working memory WMS Wechsler Memory Scale WN Neutral word-reading stimulus   xxx Acknowledgements I would like to express my sincere gratitude to my supervisor, Dr. Todd Woodward, for his mentorship, support, and opportunities provided over the years. I would also like to thank current and past members of the Cognitive Neuroscience of Schizophrenia (CNOS) Lab who have made this research possible, including but not limited to Sarah Flann, Jennifer Riley, Kel-sey Block, Meighen Roes, Julia O’Loughlin, Jessica Khangura, Marina Ren, and Judy So. I owe an enormous thank you to John Paiement for his development of the fMRI-CPCA software, as well as a thank you to Ryan Lim and Hafsa Zahid for updating its features in recent years. I am also grateful to Dr. Jennifer Whitman, Dr. Paul Metzak, and Dr. Katie Lavigne for their early guidance when I first joined the CNOS lab. I would also like to thank the members of my supervisory committee, Dr. Christine Tip-per, Dr. Rebecca Todd, and Dr. Lawrence Ward, for their helpful feedback throughout my doc-toral studies. I am incredibly grateful to the University of British Columbia, Cordula and Gunter Paetzold, and the BC Children’s Hospital Research Institute for their financial support. Many thanks to the team of technologists at the University of British Columbia 3T MRI Research Cen-tre for their vital role in data collection.  Finally, I would like to thank my family for their encouragement, and most especially my husband, Eric Nanka, whose abundant support and enthusiasm has carried me through many challenges.   1 Chapter 1: Introduction 1.1. Neurocognitive Impairment in Schizophrenia 1.1.1. Background Schizophrenia is a severe and chronic neuropsychiatric disorder that typically manifests in late adolescence or early adulthood (Karlsgodt et al., 2008). Schizophrenia is clinically de-fined by three clusters of symptoms: (1) positive symptoms, such as reality distortion symptoms (i.e., delusions and hallucinations) and disorganization (e.g., disorganized speech, inappropriate affect), (2) negative symptoms, which are the absence of normally present behaviours and mood (e.g., anhedonia, flattened affect, and avolition), and (3) neurocognitive impairment, including deficits in working memory, processing speed, verbal memory, and a variety of other domains (Diagnositic and Statistical Manual of Mental Disorders, DSM-V; American Psychiatric Association, 2013). While positive symptoms are the most visible symptom cluster, neurocogni-tive impairment is more predictive of functional outcome, such as maintaining employment and self-sufficiency (Lepage, Bodnar, & Bowie, 2014; McGurk, Mueser, Harvey, LaPuglia, & Marder, 2003). Neurocognitive impairment is persistent and is independent of positive symptom severity (Keefe et al., 2006), which typically cycles through periods of acute psychosis and re-mission over the course of weeks or months. Further, neurocognitive impairment is present be-fore the onset of a first episode of psychosis, and may comprise vulnerability markers for transition to psychosis in high-risk individuals (Carrión et al., 2018; Lutgens, Lepage, Iyer, & Malla, 2014). These deficits are present in drug-naïve schizophrenia patients (Fatouros-Bergman, Cervenka, Flyckt, Edman, & Farde, 2014), and are not explained by comorbid substance abuse (Pencer & Addington, 2003; Wobrock et al., 2013). Therefore, neurocognitive impairment is  2 considered to be a core aspect of schizophrenia rather than a secondary outcome of the illness progression, medication effects, or comorbidities. Working memory (WM), defined as actively holding and/or manipulating information in temporary store (Baddeley, 2012), is fundamental to executive processes such as cognitive con-trol and is a prevailing domain of interest in cognitive neuroscience (D'Esposito & Postle, 2015). WM is also considered to be a core domain of neurocognitive impairment in schizophrenia, as identified by the Measurement and Treatment Research to Improve Cognition in Schizophrenia initiative (MATRICS; Green et al., 2004) and the Cognitive Neuroscience Treatment Research to Improve Cognition in Schizophrenia initiative (CNTRICS; Carter et al., 2008). While there have been several conceptualizations of WM, the most influential is the multicomponent model first proposed by Baddeley and Hitch (1974). The original model describes distinct processes that produce a unified manifestation of WM capacity. These include two possible “buffer” systems underlying temporary memory storage, which are referred to as the phonological loop (e.g., ver-bal rehearsal of a set of digits) and the visuospatial sketchpad (e.g., internally visualizing the lo-cations of stimuli). In addition, a central executive system is described as a more complex component that controls attention. By contrast, state-based models conceptualize WM as either attention to activated long-term memory (for maintaining symbolic stimuli in mind, such as let-ters, digits, etc.) or sensorimotor recruitment (for maintaining basic perceptual features such as location, colour, etc.). Activated long-term memory models posit that when symbolic stimuli are presented, their long-term memory representations are accessed and subsequently maintained in an elevated state of attention until this information is no longer needed. Sensorimotor recruitment models posit that the same systems that are engaged in the perception of information also support the short-term retention of that information. The commonality among all WM models is that it is  3 conceptualized as a process involving engagement of attention to internal representations, but does not refer to executive control per se (which may be a domain-general skill). The ability to actively hold a number of items in mind may be a crucial and fundamental building block for guiding behaviour, and as such is tied to daily living. This is exemplified in its relationship to functional outcome in schizophrenia, as WM capacity has been found to predict interpersonal behaviour, community activities, and work skills in schizophrenia patients either directly or indi-rectly through mediating effects of social and/or functional competence and living skills (Bowie et al., 2008). WM function also predicts short-term clinical outcome following a first episode of psychosis (Bodnar, Malla, Joober, & Lepage, 2008), and as such, its underlying neurological mechanisms may have direct clinical implications. 1.1.2. Neurological underpinnings of working memory deficits in schizophrenia WM deficits in schizophrenia have generally been thought to arise from dysfunction in the dorsolateral prefrontal cortex (DLPFC, encompassing the middle and superior frontal gyri, including Brodmann areas 9 and 46; Glausier & Lewis, 2018; Rajkowska & Goldman-Rakic, 1995). The most widely accepted neurobiological mechanism of this is insufficient dopamine D1 transmission in the prefrontal cortex (Arnsten, Girgis, Gray, & Mailman, 2017; O'Tuathaigh, Moran, Zhen, & Waddington, 2017; Slifstein et al., 2015). The relation of WM demand to DLPFC activity reflects an inverted-U shaped function, with activation increasing with increas-ing demand to the point that capacity is reached, at which point activation declines (Van Snellenberg et al., 2016). In schizophrenia, this hypothetical curve is thought to be shifted to the left, such that WM capacity is reached at a lower level of demand (Callicott et al., 2003; Karlsgodt et al., 2007; Manoach, 2003).  4 Schizophrenia has been theorized to be a disorder of brain connectivity (Canuet, Aoki, Ishii, & Maestú, 2016; Karlsgodt et al., 2008), a concept referred to as the “disconnection hy-pothesis” (Friston, 1999). Moreover, although the key role of the prefrontal cortex in WM and other executive functions is well-known (Duncan & Owen, 2000; Stuss, 2011; Yuan & Raz, 2014), it has become apparent with the growth of functional connectivity-based approaches that WM is supported by coordinated activation of regions distributed across remote brain regions rather than within the prefrontal cortex alone (e.g., Cohen & D'Esposito, 2016). Resting-state research also suggests that the brain is intrinsically organized as a collection of networks charac-terized by coordinated fluctuations of activity across distributed brain regions (Buckner, Krienen, Castellanos, Diaz, & Yeo, 2011; Choi, Yeo, & Buckner, 2012; Yeo et al., 2011). Therefore, the relationship between DLPFC function and cognitive deficits in schizophrenia may be better un-derstood within the framework of its functional integration with the rest of the brain. 1.2. Functional Brain Networks and Working Memory Capacity 1.2.1. Functional brain networks underlying working memory It has been proposed that there is a “core” WM network comprising DLPFC, premotor cortex, anterior insula, supplementary motor area (SMA), and intraparietal sulcus (Rottschy et al., 2012). However, it is also known that such patterns of frontoparietal activity are ubiquitous across tasks involving goal-directed cognitive demand (Duncan, 2010; Duncan & Owen, 2000; Fedorenko, Duncan, & Kanwisher, 2013). The question of whether WM is subserved by a “mul-tiple demands” network, or whether there is a core WM network that appears to be ubiquitous because WM itself is fundamental to many cognitive tasks, is as yet unresolved. What is evident, however, is that WM is not a unitary cognitive construct; distinct processes are expected to be involved in a WM task, including – at minimum – encoding external stimuli, maintaining inter- 5 nal representations, and selecting/executing a response. This notion is supported by findings from data-driven approaches such as component-based analyses, which allow for multiple net-works to emerge that exhibit coordinated activation and/or deactivation in a cognitive task. The emergence of multiple frontoparietal networks subserving WM is exemplified in studies using methods including but not limited to independent component analysis (ICA; e.g., Meda, Stevens, Folley, Calhoun, & Pearlson, 2009; Steffener, Habeck, & Stern, 2012; Wong & Stevens, 2012), constrained principal component analysis (CPCA; e.g., Braunlich, Gomez-Lavin, & Seger, 2015; Metzak et al., 2011; Metzak et al., 2012; Woodward et al., 2006; Woodward, Feredoes, Metzak, Takane, & Manoach, 2013), and partial least squares (PLS; e.g., Kim et al., 2010). 1.2.2. Functional connectivity during working memory in schizophrenia In discussing findings related to DLPFC dysfunction in schizophrenia in the context of brain connectivity, it is important to distinguish between results from studies that used a region of interest (ROI) selection approach (e.g., seed-based connectivity) to examine connectivity with DLPFC versus studies that used a whole-brain approach. While ROI-driven approaches are im-portant for answering specific questions about a previously identified effect, an overly-specific focus on one region may result in overlooking other anatomical configurations that contribute to WM deficits in schizophrenia. To that end, a number of connectivity studies using whole-brain, data-driven approaches have indeed found that networks comprising DLPFC exhibit patterns suggesting an attenuation and/or inefficiency of activity in schizophrenia patients during WM tasks. For example, using ICA, Kim et al. (2009) found evidence of inefficient processing in a frontoparietal network comprising left-lateralized DLPFC, inferior parietal lobule, and cingulate gyrus, that was engaged during a Sternberg Item Recognition Paradigm (SIRP) WM task. This network was more engaged in individuals with schizophrenia compared with healthy controls at  6 a medium level of cognitive load, but was less engaged in schizophrenia patients at lower and higher levels of cognitive load. However, it is not clear whether this network subserved a specif-ic cognitive sub-process such as encoding or maintenance. In a similar ICA study, Meda et al. (2009) also found a frontoparietal network including left-lateralized DLPFC that differed be-tween schizophrenia patients and healthy controls, which was specifically correlated with the encoding phase of the task (although it is important to note that only the encoding and response phases were modelled in this analysis). In this study, patients exhibited less engagement in this network compared with controls at all levels of cognitive load that were implemented. Using CPCA, Metzak et al. (2012) found that individuals with schizophrenia exhibited greater engage-ment than healthy controls of a network including DLPFC at a medium level of cognitive load, but the groups exhibited similar degrees of engagement at lower and higher levels of cognitive load. This network appeared to underlie the encoding and maintenance phases of the task. Using a PLS approach, Kim et al. (2010) found that individuals with schizophrenia exhibited an exag-gerated inverted-U relationship between activity and cognitive load in a frontoparietal network including DLPFC, despite comparable levels of task performance between schizophrenia patients and healthy controls.  Together, some important observations should be noted from these studies. First, the na-ture of group differences clearly varies with task design and cognitive load, which is an issue that has been previously noted in research focused specifically on the DLPFC (Karlsgodt et al., 2009). Second, in all of these studies, group differences were not limited to a DLPFC-related network; rather, other independently-activated task-related networks exhibited differences be-tween patients and controls as well, including aberrant patterns of default mode network (DMN) deactivation (Kim et al., 2009; Meda et al., 2009; Metzak et al., 2012). This highlights the im- 7 portance of comprehensive investigations with respect to identifying brain activity underlying WM deficits in schizophrenia; ideally, studies should minimize a priori assumptions about the location or the timing of meaningful brain activity, and should allow for the evaluation of several networks engaged over the course of the post-stimulus time series. Most importantly, the variety of findings also raises the question of whether clinically meaningful conclusions can be drawn from the results of any individual WM study in isolation. 1.3. Measuring Task-State Functional Connectivity 1.3.1. Measuring connectivity with functional magnetic resonance imaging Functional magnetic resonance imaging (fMRI) measures brain activity by detecting changes associated with blood flow using the blood oxygen level-dependent (BOLD) contrast. The BOLD contrast reflects neural activity-dependent changes in the relative concentration of oxygenated and deoxygenated blood. This is possible due to the different magnetic properties of deoxyhemoglobin (dHb) and oxygenated hemoglobin (Hb). The BOLD contrast increases upon neural activation due to the increased cerebral blood flow, which overcompensates for the de-crease in oxygen and delivers an oversupply of oxygenated blood. Although the BOLD signal is an indirect measure of neural activity, there is a strong correlation between BOLD responses and local field potentials, the electrophysiological signals generated by the electric currents flowing from neurons within a small region of neural tissue (Logothetis & Wandell, 2004). fMRI pro-vides a time series of BOLD signal from hundreds of thousands of brain points (voxels), with each point being just a few millimeters across. As the BOLD response primarily reflects the in-put and local processing of neuronal information (Logothetis & Wandell, 2004), this allows for inferences of localized neural activity with high spatial resolution. However, the biological  8 mechanism underlying the BOLD signal – namely, the oversupply of oxygenated blood – takes several seconds to peak, and as such, neural events cannot be measured in real time. Some approaches being increasingly used in fMRI include multivariate modeling of vari-ance sources (e.g., PLS, CPCA, and ICA; McIntosh & Lobaugh, 2004; Metzak et al., 2011; Stone, 2004), multivoxel pattern analysis and machine learning (Mahmoudi, Takerkart, Regragui, Boussaoud, & Brovelli, 2012), and topological network analysis using graph theory (Rubinov & Sporns, 2010). Each of these approaches has certain advantages and disadvantages for modeling the functional brain networks underlying cognitive processes, but fundamentally, all connectivity methods involve computing the intercorrelations in BOLD signal among all voxels (or select network nodes) in the brain over time, as opposed to treating each voxel as an independent entity. Methods employing multivariate modeling of variance sources utilize dimen-sion reduction to identify components that reflect highly intercorrelated voxels. Notably, these approaches allow for the whole brain to be analyzed, and can produce several network configura-tions which capture distinct sources of variance in BOLD signal. These approaches do not re-quire the selection of ROIs, and in many cases, they allow the post-stimulus time courses associated with each network to be evaluated. 1.3.2. Limitations of task-state connectivity research In task-state research, the timing of evoked hemodynamic responses (HDRs), and how they differ between task conditions, provides a basis for interpreting the function of a brain net-work in relation to the task design. However, the nature of the BOLD signal presents significant challenges in delineating task-related networks and their evoked HDRs. One limitation in task-state fMRI research is that task-related brain activity accounts for only a fraction of BOLD signal changes over and above ongoing fluctuations in brain activity and other physiological processes.  9 A conventional solution to this problem entails modelling the recorded BOLD signal by convolv-ing the post-stimulus time series with either canonical hemodynamic response functions (HRFs) or empirically-derived HRFs from simple motor responses (e.g., Aguirre, Zarahn, & D’Esposito, 1998). For example, a typical use in ICA is to obtain the spatial map of each component by ap-plying ICA to the preprocessed BOLD data, and then determine which components underlie task-related processes by computing the fit between HRF-convolved event timing and the con-structed time courses of each component (e.g., Meda et al., 2009). However, relying exclusively on models that assume specific HDR shapes introduces the risk of missing valuable information provided by HDR shapes that do not match the models, but are nevertheless elicited by cognitive processes engaged during the task. Another limitation of BOLD data is that neural activity can-not be measured in real time because the biological process on which functional images rely (i.e., the oversupply of oxygenated blood to the area(s) of increased activity) is maximized several seconds after the actual change in neural activity, and the resultant HDR can take 20 seconds or longer to exhibit a full response and return to baseline (Logothetis & Wandell, 2004). Although jittered and partial trial designs help ameliorate confounds arising from HDRs overlapping across task trials (Burock, Buckner, Woldorff, Rosen, & Dale, 1998; Dale, 1999; Serences, 2004), this does not improve the temporal resolution of within-trial activations. A major implication of these limitations is that coinciding cognitive processes and their underlying brain activity may blur to-gether, such that a given network and its HDR may actually reflect a combination of multiple cognitive processes, despite being theoretically separable onto different cognitive operations and different networks.   10 1.3.3. Multi-experiment comparisons Toward this goal of characterizing task-related brain activity, it is beneficial to compare different experiments reported in existing literature with theoretically meaningful differences and similarities in elicited cognitive operations. In theory, this should help form a basis for determin-ing what function(s) a network serves by identifying conditions under which the network is en-gaged versus conditions when it is not. This is the basic idea behind coordinate-based meta-analyses such as activation likelihood estimation (ALE), which allows investigators to create brain activation maps from published experimental contrasts (Turkeltaub, Eden, Jones, & Zeffiro, 2002). Although coordinate-based meta-analyses are valuable for creating brain activa-tion maps that reflect consensus across vast collections of studies, they do not provide the corre-sponding timing information that would allow for a more refined evaluation of network function based on the HDRs elicited by a range of tasks. Moreover, comparisons between studies are in-herently confounded by differences in study design, analysis methods, and other decisions made by investigators (Carp, 2012). A framework that allows an evaluation of HDR shapes for a given brain network across different datasets, without imposing restrictive assumptions about the shape of the HDR or the spatial configuration of the brain networks, could be more informative. A mul-tivariate approach that allows such a comparison between tasks/datasets has been previously put forward (Lavigne, Metzak, & Woodward, 2015; Lavigne & Woodward, 2018; Ribary et al., 2017), which is referred to as multi-experiment constrained principal component analysis for fMRI (fMRI-CPCA, described in detail in the next chapter). This approach was first used to compare HDR shapes from different versions of the same cognitive task (examining the construct of evidence integration; Lavigne, Metzak, et al., 2015), and was more recently used with task merging to clarify findings regarding speech versus non-speech auditory perception in  11 schizophrenia patients (Lavigne & Woodward, 2018). In the present study, task merging with fMRI-CPCA was used to characterize functional brain networks underlying temporally correlat-ed WM processes, by allowing a finer separation of networks than could be observed from the analysis of a single task in isolation. This also allowed for a more refined evaluation of the re-sulting networks by examining activity across different tasks and cognitive demands, and in turn, a more refined understanding of the networks that exhibited some irregularity in schizophrenia patients. 1.4. Dissertation Overview 1.4.1. Aims The general aims of this research were to characterize the brain networks underlying WM and related cognitive functions, and to identify networks that may underlie WM deficits in indi-viduals with schizophrenia. More specifically, this worked aimed to: 1. Obtain a more refined characterization of functional brain networks subserving dis-tinct WM processes using multi-experiment functional connectivity analyses of task-state fMRI data. 2. Identify networks underlying task performance and WM capacity. 3. Identify networks which exhibit aberrant patterns of activity in schizophrenia patients across different cognitive tasks, with an emphasis on networks underlying WM pro-cesses. The present research used whole-brain, data-driven functional connectivity approaches to achieve these goals. It was expected that networks reflecting multiple temporally correlated pro-cesses within a given task would separate into constituent networks when multiple tasks with rel-evant overlapping and non-overlapping cognitive demands were simultaneously analysed using a  12 unified computational framework. Although a multi-experiment approach was used to character-ize networks engaged across a variety of cognitive constructs, the overarching purpose of this work was centered around identifying aberrant patterns of connectivity in schizophrenia with re-spect to WM capacity in particular. 1.4.2. Outline Chapter 2 provides an overview of the methods that were used in this work, and the find-ings are reported in Chapters 3-5. Chapter 3 describes a multi-experiment fMRI connectivity study combining a verbal WM task with a thought generation task. This study illustrates the ad-vantages of a multi-experiment design while keeping the collection of results relatively limited, and reflects a preliminary step that was carried out in addressing the first aim. Chapter 4 reports four single-experiment fMRI connectivity analyses – including the verbal WM task and thought generation task examined in Chapter 3, as well as a set-switching Stroop task and a visuospatial WM task – and provides a general characterization of the networks that dominate the task-related BOLD signal in healthy individuals. Chapter 5 reports a multi-experiment analysis combining all four of the tasks analyzed in Chapter 4. In line with the overarching goal of this work, this four-task analysis includes examining differences between healthy controls and schizophrenia pa-tients, as well as correlations between network (de)activation and task performance/WM capaci-ty. Finally, the findings of this research are summarized and discussed in Chapter 6.   13 Chapter 2: Methods The present research employed task-state fMRI and multivariate functional connectivity analyses to better understand the neural underpinnings of WM deficits in schizophrenia. Of par-ticular interest was delineating activity in networks that replicate across different cognitive tasks, as widely observed networks may be more clinically meaningful with respect to neurocognitive deficits in schizophrenia than task-specific patterns. As the data and tasks employed overlap across analysis chapters, information regarding participants, tasks, data processing, and statistical computations are detailed in the following sections. 2.1. Participants Study-specific sample characteristics and testing procedures are described in the appro-priate chapters. Across all studies, participants were eligible if they were between 19-60 years of age, fluent in English, had an estimated IQ > 70, had intact or corrected visual acuity of at least 20/50, and intact colour perception. Exclusion criteria included a history of neurological illness or infection (e.g., stroke, epilepsy, encephalitis, etc.), history of head trauma resulting in loss of consciousness for > 10 minutes, and current or recent alcohol/substance dependence. Further ex-clusion criteria for healthy controls included a recurrent history of psychiatric illness or a family history of psychosis. Patients were required to have a primary diagnosis of schizophrenia or re-lated disorder (e.g., schizoaffective disorder). Exclusion criteria related to MRI contraindications included being pregnant, wearing a pacemaker, having had aneurysm surgery, having metal im-plants, having metallic fragments in or near eyes, and the participant feeling that they would be uncomfortable in the confined space of the MRI scanner.  14 2.2. fMRI Tasks 2.2.1. Working Memory Task (WM)  The WM task design is presented in Figure 2.1. A modified Sternberg Item Recognition Paradigm (SIRP; Sternberg, 1966) was administered in which a string of 4 or 6 upper case con-sonants was displayed for 4 seconds, and then a single probe letter was displayed for 2 seconds. Participants were asked to respond “yes” or “no” with a button-press as to whether this probe letter was part of the first string of letters, using an MRI-compatible response device. Both the cognitive load (4 or 6 letters in the item set) and delay period (0 or 4 seconds between the letter string and probe) were manipulated so as to facilitate identification of the functional brain net-works distinctly involved in encoding, maintenance, and responding. The task was programed using Neurobehavioral Systems Presentation software (https://www.neurobs.com/). The 4-letter strings were flanked by a pound character (“#”) on each end to match stimulus size between the two cognitive load conditions, and the text was presented in white, 40-point Arial font against a black background. A fixation marker was displayed throughout inter-trial intervals (ITIs) lasting 2, 4, 6, or 12 seconds. Stimuli were projected onto a screen attached to the bore of the MRI scan-ner, and participants viewed the reflection of this image in a mirror positioned in their visual field. The task took approximately 25 minutes to complete over two 12.5-minute runs, with a to-tal of 112 trials (14 trials per condition per run, which were randomly generated within each task run). A 30-second rest break was included halfway through each task run, during which the text “Short break. Please remain still” was displayed. Prior to the scanning session, participants com-pleted a practice run with a computerized version of the task to familiarize themselves with the experimental procedure. Once in the scanner, participants were reminded of the instructions and were asked to provide “yes” and “no” button-press responses to demonstrate that they under- 15 stood which buttons to press before starting the task. The task was automatically initiated by a pulse from the MRI scanner, following an initial four dummy scans that were carried out to al-low the magnetization to stabilize to a steady state. A 5-second countdown was displayed prior to the first trial of the task run and at the end of the 30-second rest break. 2.2.2. Spatial Capacity Task (SCAP) The data for the SCAP fMRI task were obtained from the OpenfMRI database (https://openfmri.org/dataset/ds000030/; accession number is ds000030). The SCAP task, shown in Figure 2.2, is also an item-recall task, but it is a visuospatial memory task with no verbal con-tent. First, a target array of either 1, 3, 5, or 7 yellow dots (positioned pseudo-randomly around a central fixation) is presented during a 2-second encoding period. After a 1.5-second (1.5s), 3-second (3.0s), or 4.5-second (4.5s) delay, a single green dot (i.e., the probe) is displayed for 3 seconds. Participants are asked to respond with a button-press as to whether the probe dot was in the same position as one of the target dots. 4 load levels × 3 delay lengths resulted in 12 task conditions, and 4 trials per condition are presented in the task (total = 48 trials). A central fixa-tion is visible throughout the trials. Jittered ITIs ranged from 1-4 seconds (mean = 2 seconds), during which the central fixation is visible. The task was programmed in MATLAB (Mathworks) using the Psychtoolbox program (http://psychtoolbox.org/). Participants viewed the task stimuli through MRI-compatible goggles and responded with their right hand on an MR-compatible but-ton box. Before starting the in-scanner task, subjects underwent a training session outside of the scanner, and once in the scanner were reminded of the instructions. 2.2.3. Task-Switch Inertia Task (TSI) The TSI task design is presented in Figure 2.3. This task is a set-switching Stroop task which involves responding to Stroop stimuli in alternating blocks of colour-naming (i.e., naming  16 the font colour of text displayed) and word-reading (i.e., reading the word displayed) of neutral and incongruent stimuli. A neutral word-reading stimulus is one in which a colour word (“GREEN”, “RED”, “YELLOW”, or “BLUE”) is displayed in white font against a black back-ground, and a neutral colour-naming stimulus is one in which a string of Xs is displayed in green, red, yellow, or blue font. Incongruent stimuli are colour words (“GREEN”, “RED”, “YELLOW”, or “BLUE”) displayed in incongruent green, red, yellow, or blue font. Congruent Stroop stimuli, in which the font colour matches the colour word displayed, were not used in this study because the cognitive and neurological processes affected by response inhibition demands – not facilitation – were of primary interest. Moreover, the facilitation effect may be absent un-der task-switching conditions anyway (Rogers & Monsell, 1995).  The TSI task was programmed using Neurobehavioral Systems Presentation software (https://www.neurobs.com/). This task consisted of 12 blocks of 10 trials and lasted approxi-mately 11 minutes. Each time the task was switched, an instruction screen was displayed for 9 seconds with reminders for which buttons to use for responding. Stimuli were presented in 40-point Helvetica font against a black background. Two commercially-available MRI-compatible fiber optic response devices with two buttons each were used for the participants’ responses. For each trial, the stimulus was displayed for 1900ms, followed by a blank screen for 100ms. The response was recorded within the total 2000ms trial, which was followed by a jittered ITI of 900ms–6000ms during which a fixation cross was displayed. In total, 30 incongruent and 30 neu-tral colour-naming trials and 30 incongruent and 30 neutral word-reading trials are presented. The task was automatically initiated by a pulse from the MRI scanner, following an initial four dummy scans that were carried out to allow the magnetization to stabilize to a steady state. A 20-second rest break was included halfway through the task run, during which the text “Short break.  17 Please remain still” was displayed. In a previous study published by our lab, incongruent and neutral stimuli were interspersed within each block of colour-naming and word-reading (Woodward, Leong, Sanford, Tipper, & Lavigne, 2016). However, a consideration in this new version was to examine whether the slowing of responses due to task-switching is dependent on the level of prepotent response inhibition required in the previous task set, which would suggest that shifting ability is impacted by response inhibition demands and not merely the change of task rules per se. Therefore, in this revised version of the task, the colour-naming blocks alternat-ed between exclusively neutral stimuli and exclusively incongruent stimuli, whereas the word-reading blocks always consisted of a mix of neutral and incongruent stimuli.  Immediately prior to the scanning session, participants completed a practice run outside of the scanner to familiarize themselves with the testing procedure and key-response mapping (left middle finger = “green”, left index finger = “red”, right index finger = “yellow”, and right middle finger = “blue”). Beginning with neutral word stimuli, participants practiced responding, and received on-screen feedback after errors (“Incorrect”) or responses exceeding 1900ms (“Too slow”). After the participant responded with at least 90% accuracy over 10 trials, the presentation of neutral word stimuli stopped and switched to a series of neutral colour-naming stimuli, which was followed by a series of both neutral and incongruent word-reading stimuli, and finally a se-ries of incongruent colour-naming stimuli. Once in the scanner, participants were reminded of the instructions, and asked to provide “green”, “red”, “yellow” and “blue” responses to demon-strate that they understood which buttons to press, and then finally completed one more practice run in the scanner to ensure that they had memorized which buttons to press.  18 2.2.4. Thought Generation Task (TGT) The TGT task design is presented in Figure 2.4. This task was originally designed to in-vestigate the neurological basis of internal generation versus external processing of speech in schizophrenia, and specifically for examining differences between hallucinating and non-hallucinating patients (Lavigne, Rapin, et al., 2015; Rapin et al., 2012). Participants were pre-sented with an object noun and its corresponding image (e.g., pillow) for five seconds and in-structed to either mentally generate or listen to a simple definition of the word (e.g., “Something you rest your head on when sleeping”). The two experimental conditions (i.e., generating and hearing) were presented in alternating blocks of 15 trials (30 trials total for each condition across two runs), with a 60-second rest break in between the two conditions. Trials were cued with the words “something you…” or “listen…” presented under the images in the generating and hearing conditions, respectively. Stimuli were randomly assigned to each condition for each participant. Participants were administered a post-scan questionnaire where they were asked, for each trial, whether they generated a definition and, if so, what that definition was. Hearing tests were car-ried out using an audiometer (AMBCO 650AB, http://www.ambco.com/) to ensure absence of hearing impairment prior to testing. The task was automatically initiated by a pulse from the MRI scanner, following an initial four dummy scans that were carried out to allow the magneti-zation to stabilize to a steady state. Three 30-word lists of nouns were created using the MRC psycholinguistic database (Coltheart, 1981). Only nouns with scores >500 on familiarity, concreteness, and imageability criteria ratings were chosen (maximum value = 700). The three word lists (i.e., list A, B and C) were matched by mean values on these parameters. All nouns were objects of neutral affective content within the categories of food, houseware, furniture, clothing, and transportation devices.  19 Each audio stimulus was recorded by a female native English speaker in a quiet room, and lasted on average 2.22 seconds (SD = 0.62). Two out of the three word lists were randomly assigned to the two conditions for each participant, resulting in six potential word set combinations (i.e., AB, BA, AC, CA, CB, and CB), which were counterbalanced across participants. Prior to fMRI scanning, participants were familiarized with the experimental procedure in a computerized practice run, using 10 words that were different from those presented in the scanner. In addition, to facilitate the generation of definitions while in the scanner, participants practiced audibly generating definitions for the 30 words in the thought generation condition. No training was carried out for the hearing condition material to minimize the likelihood of partici-pants self-generating definitions during hearing trials. The task was programmed using Neurobe-havioral Systems Presentation software (https://www.neurobs.com/). Stimuli were projected onto a screen placed at the bore of the MRI scanner, and participants viewed a reflection of the image in a mirror positioned in their visual field. In the hearing condition, the audio file containing the definition was presented 700ms after the onset of the word and illustration. The recorded defini-tions always began with the words “Something you”, and participants were instructed that men-tally generated definitions should also start with “Something you”, to ensure that at least some words were mentally generated on every trial, and to minimize confounds between conditions. To prevent participants from internally reviewing the most recently generated or heard defini-tion, a display of generic circles moving in an orbiting motion was presented during the ITIs of either 2, 4, 6, 8, 16, or 20 seconds (mean = 4.46 seconds). The order of presentation of the ITIs, conditions, and words within each block were randomized.  20 2.3. fMRI Data Acquisition and Preprocessing 2.3.1. Data acquisition Task-specific scanning parameters and protocols are presented in Table 2.1. In all studies, functional image volumes were collected using a T2*-weighted sequence covering the whole brain. A T1-weighted structural image was also acquired from each participant for co-registration of functional images with structural templates used in the fMRI normalization proce-dure. Data collection for the WM task, TSI task, and TGT was carried out at the UBC MRI Re-search Centre on a Philips Achieva 3.0 Tesla scanner with quasar dual gradients (maximum gradient amplitude, 80 mT/m; maximum slew rate, 200 mT/m/s). The WM data and TSI data were collected in the same study in largely overlapping samples (participants completed the TSI task and then the WM task in the same scanning session), with the exception of a subset of the WM data which had been previously collected in an earlier study overlapping with the TGT da-taset. The data for the SCAP task were obtained from the OpenfMRI database (https://openfmri.org/dataset/ds000030/; accession number is ds000030). The MRI data were originally acquired on one of two 3.0 Tesla Siemens Trio scanners at either the Ahmanson-Lovelace Brain Mapping Center (Siemens version syngo MRI B15) or the Staglin Center for Cognitive Neuroscience (version syngo MR B17). Complete details for this project are provided in Poldrack et al. (2016).  2.3.2. Preprocessing Functional images were reconstructed offline and preprocessed using Statistical Paramet-ric Mapping 12 (SPM12; Wellcome Trust Centre for Neuroimaging, UK). First, each partici-pant’s functional and T1 scans were manually inspected and reoriented to the anterior commis-sure. Next, each functional run was slice timing-corrected, realigned to the first image of the run,  21 co-registered to the structural (T1) image, normalized to the Montreal Neurological Institute (MNI) T1 brain template (voxel size = 3 × 3 × 3 mm) using the segmentation and normalization algorithms applied in SPM12, and spatially smoothed with an 8 × 8 × 8 mm full width at half maximum Gaussian filter. Runs for which realignment exceeded 4 mm or 4° on any scan were excluded from all analyses. The six realignment parameters (i.e., translation in x, y, and z dimen-sions, and rotation in pitch, roll, and yaw directions) were regressed out of the BOLD time series prior to statistical analyses to minimize effects of head movement. Linear and quadratic trends, which may arise as an artifact of scanner drift, were also regressed out of the BOLD time series. 2.4. Constrained Principal Component Analysis for fMRI 2.4.1. General framework Functional connectivity analyses were carried out using constrained principal component analysis for fMRI (fMRI-CPCA; Metzak et al., 2011). CPCA is a general method for structural analysis of multivariate data, which combines multivariate multiple regression with principal component analysis (PCA) into a unified framework (Hunter & Takane, 2002; Takane & Hunter, 2001). fMRI-CPCA applies this framework to reveal temporally orthogonal sources of task-related BOLD signal by employing PCA on the portion of variance in activity that is predictable from the task timing. In the present research, data-driven estimates of the post-stimulus time courses of hemodynamic responses (HDRs) were obtained by using a finite impulse response (FIR) model in the regression stage of the CPCA. A FIR model makes no assumptions about the shape of the HDR; instead, it produces separate estimates for each post-stimulus scan for each condition and each participant. Performing PCA after regression with a FIR model allows inves-tigators to (1) identify the spatial configurations of multiple functional brain networks simultane-ously involved in performing a cognitive task, (2) estimate the post-stimulus HDR of each  22 functional network, and (3) statistically test the effect of experimental manipulations and group differences in HDR shapes in each functional brain network. The estimated HDR shapes aid in-terpretation of the cognitive processes engaged from examining the relative timing and duration of the different peaks, which in turn helps inform our understanding of the role of the functional brain systems involved. A general overview of the steps involved in a typical fMRI-CPCA is presented in Figure 2.5. The mathematical equations as applied to fMRI data are detailed below. 2.4.2. Matrix equations 2.4.2.1. Multivariate multiple regression Two matrices are constructed for the fMRI-CPCA analysis. The first matrix, Z, contains the intensity values for the preprocessed BOLD time series of each voxel, with one column per voxel and one row per scan. The second matrix, G, consists of a FIR basis set, with one column per estimated post-stimulus scan per condition, and one row per acquired scan. The value 1 is placed in rows of G for which BOLD signal amplitude is to be estimated, and the value 0 in all other rows (i.e., “mini boxcar” functions). The G matrix estimates subject- and condition-specific effects by including a separate FIR basis set for each condition and for each subject. Each col-umn of Z and G is standardized for each subject separately. The first step in CPCA involves partitioning the total variability in Z (BOLD signal) into variance that is predictable by G (design matrix) and variance that is not predictable by G, pro-ducing a matrix of predicted scores (the GC matrix) and a matrix of residuals (the E matrix). To achieve this, multivariate least-squares linear multiple regression is carried out, whereby the BOLD time series (Z) is regressed onto the design matrix (G):  23   𝑠Z𝑏 =  𝑠G𝑒C𝑏 +  𝑠E𝑏 (1) where Z = the dependent variable, G = the independent variable, C = the weights applied to G to produce a matrix of predicted scores (GC), E is a matrix of residuals, s = acquired scans, b = voxels, and e = FIR estimates. The C matrix represents condition-specific regression weights, which are akin to the beta images produced by conventional univariate fMRI analyses, and are obtained by:    𝑒C𝑏 = ( 𝑒G𝑠′G𝑒 ) −1 𝑒G𝑠′Z𝑏  (2) Therefore, GC represents the variability in Z that is predictable from the design matrix G (i.e., the task-related variability in Z).  2.4.2.2. Principal component analysis (PCA) The next step in CPCA is to apply PCA, a special case of singular value decomposition, to the GC matrix so that the resulting components can be examined. PCA uses orthogonal trans-formation to convert a set of observed variables into a set of linearly uncorrelated variables called principal components, such that the first principal component accounts for the largest pos-sible amount of variance in the data, and each succeeding component has the highest possible variance under the constraint that it is orthogonal to the preceding components. PCA is typically used for dimension reduction, as it allows a large set of variables to be reduced to a few principal components that reflect common variance among the variables that are most highly related to each component. In the case of fMRI-CPCA, it is performed on the vertically concatenated GC matrices of the entire sample of participants. As such, each component is constructed from the temporal intercorrelations of estimated BOLD activity among all voxels in the brain – that is, their functional connectivity. This is an important distinction from the patterns of co-activation that arise from traditional mass-univariate approaches to task-state fMRI research.  24 The singular value decomposition of GC (comprising all subjects and scans) results in:  [ 𝑠U𝑚D𝑚V𝑏 ′] ≈  𝑠G𝑒C𝑏 (3) where U = matrix of left singular vectors, D = diagonal matrix of singular values, and V = matrix of right singular vectors after reduction of dimensionality to m components. Each column of (VD √b − 1⁄ ), where b is the number of columns (i.e., voxels) in Z, is overlaid on a structural brain image to allow spatial visualization of the brain regions dominating each functional net-work. The rescaled VD is referred to as a loading matrix, and the values are correlations between the component scores (U) and the columns (i.e., voxels) in GC. The number of components to retain is typically chosen based on the amount of variance each one accounts for, indicated by their respective eigenvalues (i.e., the squared singular values obtained from singular value decomposition of the mean centered and normalized data matrix). In the present research, the plotted eigenvalues (i.e., the scree plot) were evaluated to identify the components contributing a meaningful and unique portion of variance (Cattell, 1966; Cattell & Vogelmann, 1977). If more than one component is extracted, the interpretability of the compo-nents can be increased by redistributing the variance via rotating the components. The rotation method used in the present research, varimax rotation (Kaiser, 1958), redistributes the variance captured by the original eigenvectors, stopping optimization when the matrix of component load-ings optimally matches a "simple structure", which is characterized by the following conditions: (1) any given variable (i.e., voxel) has a high loading on a single component (i.e., brain network), and near-zero loadings on the others, and (2) any given component consists of only a small sub-set of variables with high loadings, with the remaining variables all having near-zero loadings.  25 2.4.2.3. Predictor weights Post-stimulus HDRs are estimated by regressing the component scores of each compo-nent extracted from GC onto the independent variables in G (i.e., FIR estimates), producing a matrix of predictor weights, P, that estimate the degree to which each variable in G is related to the corresponding components extracted from GC. That is, predictor weights in matrix P are the weights that, when applied to each column of the matrix of predictor variables (G), create U (i.e., component scores) such that:   𝑠U𝑚 =  𝑠G𝑒P𝑚  (4) Each subject- and condition-specific set of predictor weights is expected to take the shape of an HDR when plotted over post-stimulus time, with the highest values corresponding to the HDR peaks. Note that positive predictor weights do not necessarily reflect activation, as this de-pends on whether the component loadings are positive or negative. To avoid confusion as to whether a given HDR shape reflects activation or deactivation, all plots in the present research are oriented such that values above the X axis reflect activation, and values below the X axis re-flect deactivation (which means that the Y axis may be reversed in some instances). As fMRI-CPCA predictor weights are produced for each component for each combina-tion of post-stimulus time bin, task condition, and participant, activity in each of the extracted networks is statistically examined by applying repeated measures analysis of variance (ANOVA) to the predictor weights to examine effects of post-stimulus time and experimental conditions in each task. As a further step in fMRI, to verify that the resulting components primarily capture biologically plausible BOLD signal rather than noise, a visual examination of their HDR shapes and anatomical organization is performed, and repeated measures ANOVA is carried out on the predictor weights to confirm the presence of significant effects of post-stimulus time. In the pre- 26 sent research, Mauchley’s test of sphericity was carried out for each analysis, and the Green-house-Geisser correction was applied in cases where the assumption of sphericity was violated; corrected degrees of freedom are reported where the correction changed statistical results. In-spection of the HDR shapes while considering the experimental designs facilitated assignment of each network to one or more cognitive processes. 2.4.3. Multi-experiment fMRI-CPCA As the regression step in CPCA (equation 1) is performed separately for each subject, the G matrices need not be comprised of the same number of conditions or columns across all sub-jects; this allows for a simple extension of the CPCA framework to multi-experiment analyses. In simultaneous analyses of multiple datasets, identifying networks which are spatially identical across tasks, an advantage of fMRI-CPCA is that it can describe how the time-courses of post-stimulus activity in those networks vary (or do not vary) across tasks. Some examples of other multi-experiment approaches include joint sparse representation analysis (jSRA; Ramezani, Marble, Trang, Johnsrude, & Abolmaesumi, 2015), joint independent component analysis (jICA; Calhoun et al., 2006), multimodal canonical correlation analysis (mCCA; Sui et al., 2010), com-bined CCA + ICA (Sui et al., 2010), and multi-task PLS (Grady, Springer, Hongwanishkul, McIntosh, & Gordon Winocur, 2006). jSRA, jICA, mCCA, and CCA + ICA are all second-level analyses that involve identifying common sources from experimental contrast images extracted from different tasks using statistical matches (e.g., beta weights) to modelled HDR shapes, and as such, these methods do not provide the corresponding temporal information that would allow for a more refined evaluation of network function based on the HDRs elicited by a range of tasks. PLS is somewhat similar to CPCA in that it identifies patterns of brain activity that covary with features of the experimental design, but due to differences in the nature of the matrices that  27 are covaried, does not provide an estimated condition-specific HDR shape as an integral part of the output (see Metzak et al., 2011, p. 869, for a detailed explanation of how the general framework of CPCA differs mathematically and conceptually from PLS and other multivariate methods). Moreover, multi-task PLS is used to identify what is common between tasks without allowing for activations that may be exclusive to a particular task.  While all of the above approaches involve identifying sets of brain regions that overlap across different tasks, multi-experiment fMRI-CPCA expands on this to allow statistical evalua-tion of network-specific HDR time courses, which may or may not be engaged in all tasks/conditions. Importantly, multi-experiment fMRI-CPCA produces estimated HDRs and connectivity-based spatial maps in a data-driven manner (i.e., no reliance on statistical matches to assumed HRFs, or a set of a-priori assumed ROIs) while keeping the spatial pattern of each network constant across the tasks to be compared, regardless of whether it is a within- or be-tween-subjects design. The degree to which each functional brain network replicates its in-volvement across tasks is determined by comparing the magnitude and temporal pattern of the HDR shapes associated with each network (Lavigne, Metzak, et al., 2015; Lavigne & Woodward, 2018). This provides an opportunity to use differences between tasks, and the cogni-tive operations they putatively elicit, to help determine the cognitive function served by a given network (Ribary et al., 2017). For example, if a particular cognitive process is engaged in two tasks, but its timing differs due to different task designs, HDRs for the underlying network should be elicited in both tasks but display different shapes. However, if a cognitive process is elicited by one task but not the other, the task not eliciting this process should show a flat HDR shape for that network. This raises the possibility of a finer delineation of networks to emerge: when two (or more) distinct processes consolidate into one network in a single task due to the  28 low temporal resolution of fMRI, it is possible that these processes will separate onto different networks when one process is engaged in both tasks (resulting in an HDR for both tasks), but the other is engaged in only one task (resulting in an HDR for only one task). The next chapter pre-sents an example of these concepts being applied in a preliminary multi-experiment analysis that was designed to obtain finer separation of task-related networks – and the cognitive processes they support – engaged in a verbal WM task.    29 2.5. Chapter 2 Tables Table 2.1. fMRI acquisition parameters for all datasets included in present research.  WM SCAP TSI TGT Functional images     Sequence T2*-weighted gradi-ent-echo spin pulse T2*-weighted echo-planar imaging (EPI) T2*-weighted gradi-ent-echo spin pulse T2*-weighted gradi-ent-echo spin pulse Slices 35 34 35 36 Slice thickness/gap 3/1 mm 4/0 mm 3/1 mm 3/1 mm Acquisition matrix 96 × 96 64 × 64 96 × 96 80 × 80 Repetition time (TR) 2000 ms 2000 ms 2000 ms 2500 ms Echo time (TE) 30 ms 30 ms 30 ms 30 ms Flip angle (FA) 90° 90° 90° 90° Field of view (FOV) 288 × 288 × 139 mm 192 × 192 × 136 mm 288 × 288 × 139 mm 240 × 240 × 143 mm Volumes per run 374 291 330 176 Runs 2 1 1 2  WM SCAP TSI TGT Structural image     Sequence T1-weighted Fast Field Echo (FFE)  T1-weighted magneti-zation-prepared rapid gradient-echo (MPRAGE) T1-weighted Fast Field Echo (FFE)  T1-weighted Fast Field Echo (FFE)  Slices 190 sagittal 176 sagittal 190 sagittal 182 coronal Slice thickness/gap 1/0 mm 1/0 mm 1/0 mm 1/0 mm Acquisition matrix 256 × 256 256 × 256 256 × 256 256 × 256 Repetition time (TR) 8.1 ms 1.9 s 8.1 ms 8.1 ms Echo time (TE) 3.7 ms 2.26 ms 3.7 ms 3.7 ms Flip angle (FA) 8° 7° 8° 8° Field of view (FOV) 190 × 256 × 256 mm 176 × 256 × 256 mm 190 × 256 × 256 mm 256 × 182 × 256 mm Voxel dimensions 1 × 1 × 1 mm 1 × 1 × 1 mm 1 × 1 × 1 mm 1 × 1 × 1 mm Total scanning time 7 min 23 sec 6 min 3 sec 7 min 23 sec 6 min 22 sec     30 2.6. Chapter 2 Figures Figure 2.1. Working Memory (WM) task design; example of a trial presented in the 4-letter load condition with a 4-second delay. Participants were instructed to try to remember the string of let-ters displayed at the start of each trial. As the four task conditions were randomly intermixed within each fMRI run, it was explained that there would be either four or six letters to remember on any given trial, and that on some trials there would be a four-second delay before the probe letter was displayed, while on other trials the probe letter would be displayed immediately after the initial string of letters disappeared from the screen. Participants were instructed to use their right index finger to respond “yes” if the probe letter had been included in the letter string dis-played at the start of the trial, and use their right middle finger to respond “no” if the probe letter had not been included in the letter string.   Figure 2.2. Spatial Capacity (SCAP) task design; stimuli downloaded from software developer’s publicly available code library (https://poldracklab.stanford.edu/softwaredata). Participants are provided with the instructions: “You will see either 1, 3, 5, or 7 yellow dots on the screen. After these dots disappear, a green dot will appear. Press the FIRST button if the green dot is in the same place as one of the yellow dots. Press the SECOND button if the green dot is in a different place than one of the yellows.”    31 Figure 2.3. Task-Switch Inertia (TSI) task design. Task blocks began with the instructions “NAME THE COLOUR” or “NAME THE WORD”, with reminders for which buttons to press at the bottom of the screen. Each block consisted of 10 trials, and a total of 12 blocks were com-pleted (3 neutral colour-naming, 3 incongruent colour-naming, and 6 word-reading). Word-reading blocks consisted of 5 neutral and 5 incongruent stimuli pseudo-randomly distributed throughout the block so that no more than 3 of one type of stimulus appeared in the first 5 trials. cn = neutral colour-naming; ci = incongruent colour-naming; W = word-reading; ITI = inter-trial interval.    32 Figure 2.4. Thought Generation Task (TGT) design; examples of trials presented in a thought-generation block. Participants were instructed to either mentally generate (generating condition) or to listen to (hearing condition) a simple definition of a word (e.g., “Something you rest your head on when sleeping” for the word “pillow”). The conditions were cued with the words “some-thing you…” or “listen…” presented under the images for the generating and hearing conditions, respectively. ITI = inter-trial interval.   33 Figure 2.5. Schematic overview of fMRI-constrained principal component analysis (fMRI-CPCA). Step 1: Multivariate least-squares regression of blood oxygen level-dependent (BOLD) intensity values, Z, onto finite impulse response (FIR) basis set, G (performed for each partici-pant). The C matrix comprises condition-specific regression weights. GC represents the variabil-ity in Z that is predictable from G (i.e., the task-related variability in BOLD signal). Individual participant GC matrices are concatenated to create the resulting GC matrix in the following step. Step 2: Singular value decomposition is used to extract components in GC that represent tempo-rally orthogonal functional brain networks in which BOLD activity is directly related to the task timing. This produces a (rescaled) matrix, VD, containing the component loadings which are thresholded and overlaid on a brain template, and matrix U, containing the component scores for each participant. By convention, positive loadings are depicted in hot colours, and negative load-ings are depicted in cool colours. Step 3: For each participant, component scores (U) are re-gressed onto G to produce predictor weights (P) estimating the intensity of the component at each time bin coded in G. Predictor weights should take the shape of a hemodynamic response (HDR) when plotted over post-stimulus time, and may be input to conventional statistical anal-yses such as a repeated measures analysis of variance (ANOVA).     34 Chapter 3: Finer Separation of Working Memory Networks Using Multi-Experiment fMRI-CPCA 3.1. Background Working memory (WM), defined as the ability to actively hold information in mind in order to guide behavior (Baddeley & Hitch, 1974), is a prevailing construct of interest in cogni-tive neuroscience (D'Esposito & Postle, 2015) and its neural circuitry may have implications in cognitive remediation for schizophrenia (Li et al., 2015). It has been proposed that there is a “core” network underlying WM processes, comprising dorsolateral prefrontal cortex (DLPFC), premotor cortex, anterior insula, supplementary motor area (SMA), and intraparietal sulcus (Rottschy et al., 2012). However, WM is not a unitary cognitive construct; distinct processes are expected to be involved in a WM task, including – at minimum – encoding external stimuli, maintaining internal representations, and selecting/executing a response. Although it is plausible for a given brain region to support a variety of cognitive operations (e.g., Cohen, Sreenivasan, & D'Esposito, 2014), a more refined characterization of functional brain networks underlying WM should be possible. However, given the limited temporal resolution of fMRI data, network de-compositions derived from functionally correlated voxels could be confounded by coincid-ing/overlapping task demands such as internal maintenance of items in memory and preparation/execution of motor responses. In a Sternberg item recognition paradigm (SIRP), one way to address the problem of low temporal resolution in fMRI data is to independently vary the degree of cognitive load (i.e., number of items to remember) and the duration of the delay period. With these manipulations, functional brain networks underlying encoding and maintenance should exhibit activity that in-creases with greater cognitive load, but those underlying motor responses should not. In addition,  35 functional brain networks underlying maintenance and retrieval/response should exhibit activity in which the timing depends on delay period length, but those underlying encoding should not. However, these experimental manipulations could affect multiple processes (e.g., increasing the number of items to remember may result in both an increase in encoding demands and an in-crease in perceptual processing), and the inability to randomize the order of task phases is a pre-vailing limitation, as the order must be: 1. encoding, 2. delay, and then 3. response. In addition to utilizing strategic task designs, another way to delineate HDRs of networks underlying WM sub-processes is by treating the different task phases as distinct epochs (encod-ing, delay, and probe), and deriving spatial maps of connectivity by, for example, applying a seed-based approach (Gazzaley, Rissman, & D’Esposito, 2004; Rissman, Gazzaley, & D'Esposito, 2004) or independent component analysis (Meda et al., 2009; Steffener et al., 2012; Wong & Stevens, 2012). However, WM sub-processes may not fit neatly into one task phase or be elicited by the onset of a particular stimulus. For example, there could be preparatory activity that initiates within the maintenance phase but would be more accurately characterized as a re-sponse/retrieval process, or there could be encoding processes that extend into the delay period. It is also important to note that the actual cognitive processes engaged during a task may underlie a variety of different types of task demands, and demonstrating that a network is correlated with one particular task phase does not necessarily reveal what function that network serves per se. 3.2. Aims and Hypotheses Two approaches were implemented to optimally separate and assign plausible cognitive processes to networks underlying a SIRP task involving memorizing letter strings: (1) independ-ent manipulation of cognitive load and delay period length, and (2) multi-experiment fMRI-CPCA combining WM data with a different cognitive task, the Thought Generation Task (TGT;  36 Lavigne, Rapin, et al., 2015). A number of considerations were made in the decision to use the TGT as a comparison task. Firstly, the TGT task does not involve an overt response (button-press or otherwise), which offers a clear distinction to help separate motor response processes from cognitive processes involved in WM; that is, as motor response processes will be engaged in the WM task only, a response-based HDR should be elicited for the WM task but not the TGT task. Although a basic perceptual or motor task would provide a similar manipulation (e.g., visu-al fixation task without any motor response, or a motor response task without WM cognitive de-mands), the TGT task also entails some manipulation of higher-level cognitive demand which could be relevant to WM. For example, the TGT task’s two conditions (silent thought generation and speech perception) provide a contrast between an internally-oriented cognitive process and an externally-oriented stimulus perception process, and this comparison could be analogous to the contrast between the WM process of internally rehearsing a sequence of letters in the absence of their external representation and encoding the letters presented on-screen. In summary, the purpose of utilizing the TGT data in the present study was to advance our understanding of net-works engaged during the WM task, as separation of networks in the multi-experiment analysis was expected to be observed that were not separable in the WM dataset on its own; studies utiliz-ing the TGT task as a construct of interest on its own have been previously published (Lavigne, Rapin, et al., 2015; Rapin et al., 2012), and it is analyzed in detail in Chapter 4, and so the TGT single-experiment fMRI-CPCA results are not discussed in the present chapter.  Previously published fMRI-CPCA results on WM and TGT tasks allowed for some gen-eral predictions to be made. First, in the WM task, frontoparietal connectivity coordinated with primary and association visual cortices was expected to emerge, with the magnitude of coordi-nated activation being dependent on cognitive load, and the activation initiating early in the post- 37 stimulus time series (Braunlich et al., 2015; Metzak et al., 2011; Metzak et al., 2012; Woodward et al., 2013). It was also expected that deactivation of the default mode network (DMN; Buckner, Andrews-Hanna, & Schacter, 2008) would be sustained throughout the WM trial phases, with the magnitude of deactivation also being dependent on cognitive load (Braunlich et al., 2015; Metzak et al., 2011; Metzak et al., 2012; Woodward et al., 2013). The respective activation and deactivation of these networks were expected to be similarly engaged in the TGT task (Lavigne, Rapin, et al., 2015). Finally, it was predicted that a third network would exhibit activation in re-gions involved in volitional control over motor responses, including the SMA, primary and pre-motor cortices contralateral to the response hand, basal ganglia, and cerebellar activations ipsilateral to the response hand. This “response” network was predicted to exhibit late, staggered onsets, dependent on the delay duration in the WM task (Braunlich et al., 2015; Woodward et al., 2013), but to produce a flat HDR in the TGT task since this task did not involve a motor re-sponse. Internally oriented WM processes related to maintenance and/or recall may temporally coincide with response anticipation and execution; therefore, in the current study, because inter-nally oriented cognitive processes – but not response processes – are expected to overlap be-tween the two tasks (i.e., in the thought generation condition for the TGT task), some separation of these processes onto distinct networks in the multi-experiment analysis was expected. 3.3. Methods 3.3.1. Participants 69 participants from two datasets were included in the present analysis, all of whom met the general inclusion criteria for healthy controls as detailed in Chapter 2 (Section 2.1). 37 partic-ipants completed the WM task and 32 non-overlapping participants completed the TGT task. Task performance on the fMRI WM task was examined to confirm participant engagement dur- 38 ing the task, and any runs in which a participant achieved < 60% correct responses were exclud-ed from the analysis (note that participants had a 50% chance of guessing correctly on any given trial). A brief demographic questionnaire was administered to all participants to obtain their age, highest level of education, and any history of head injury, neurological conditions, drug use, and medication. In addition, visual acuity and color vision were assessed to ensure that participants were able to view the tasks on an fMRI presentation screen. Handedness was measured with the Annett Hand Preference Questionnaire (AHPQ; Annett, 1970), and IQ was estimated with the Quick IQ test (Mortimer & Bowen, 1999). Participants were recruited via posters on the Univer-sity of British Columbia (UBC) campus, community bulletin boards, and on electronic bulletin boards such as Craigslist. Participants provided informed consent at the start of their testing ses-sions, and were compensated $10 per hour for time spent participating in the study plus $10 for travel expenses. Participants were also given a copy of their T1 MRI scan on a disc. Demograph-ic information for each group of participants is provided in Table 3.1. Details regarding fMRI data acquisition and preprocessing are provided in Chapter 2 (Section 2.3 and Table 2.1) 3.3.2. Tasks 3.3.2.1. WM task design  Details regarding WM task design and administration are presented in Chapter 2 (Sec-tion 2.2.1 and Figure 2.1). In summary, a string of 4 or 6 upper case consonants was displayed for 4 seconds, and then a single probe letter was displayed for 2 seconds. Participants were asked to respond “yes” or “no” with a button-press as to whether this probe letter was part of the first string of letters. Both the cognitive load (4 or 6 letters in the item set) and delay period (0 or 4 seconds between the letter string and probe) were manipulated so as to facilitate identification of the functional brain networks distinctly involved in encoding, maintenance, and responding.  39 3.3.2.2. TGT task design Details regarding TGT task design and administration are presented in Chapter 2 (Section 2.2.4 and Figure 2.4). Briefly, participants were presented with an object noun and its corre-sponding image (e.g., pillow) for five seconds and instructed to either mentally generate or listen to a simple definition of the word (e.g., “Something you rest your head on when sleeping”). The two experimental conditions (i.e., generating and hearing) were presented in alternating blocks of 15 trials. 3.3.3. Functional connectivity analysis Multi-experiment fMRI-CPCA was carried out as described in detail in Chapter 2 (Sec-tion 2.4 and Figure 2.5). fMRI-CPCA combines multivariate multiple regression with PCA to reveal temporally orthogonal sources of post-stimulus BOLD activity, by employing PCA on the portion of variance in BOLD signal that is predictable from the task timing (as specified by a FIR model). In the present analysis, the time bins for which a FIR basis function was specified were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time for the WM task with TR = 2,000ms, and 25 seconds of post-stimulus time for the TGT task with TR = 2,500ms). For the WM task, each level of cognitive load and delay duration was modelled, resulting in G matrices with 374 rows (scans) and 40 columns (4 conditions × 10 post-stimulus time bins) per task run. For the TGT task, both the generating and the hearing condition were modelled, result-ing in G matrices with 176 rows (scans) and 20 columns (2 conditions × 10 post-stimulus time bins) per task run. fMRI-CPCA produces predictor weights for each component (i.e., network) for each combination of post-stimulus time bin, task condition, and participant. Therefore, the estimated HDRs in each of the extracted networks was statistically examined by applying repeated  40 measures ANOVAs to the predictor weights to examine effects of post-stimulus time and exper-imental conditions in each task. For the WM task, a 2 (cognitive load; 4 vs. 6 letters) × 2 (delay duration; 0 vs. 4 seconds) × 10 (post-stimulus time bins) ANOVA was performed for each net-work. For the TGT task, a 2 (condition; hearing vs. generating) × 10 (post-stimulus time bins) ANOVA was performed for each network. Post hoc analyses of significant main effects and in-teractions were performed for networks extracted from the multi-experiment fMRI-CPCA only. For the WM task, post-hoc analyses were performed to examine effects of cognitive load, delay duration, load × delay, and delay × time. Main effects of load and delay were examined using polynomial contrasts. For the TGT task, post-hoc analyses were performed to examine effects of condition and condition × time. Effects of condition were examined using polynomial contrasts. In both tasks, interactions involving time were examined using repeated measures contrasts be-tween adjacent time bins. 3.4. Results 3.4.1. WM task fMRI-CPCA results A more extensive analysis of the WM task, with a larger sample size, is reported in Chap-ter 4. A brief summary is provided here for comparison with the subsequent multi-experiment results. Three components were extracted from the WM data, accounting for 16.83%, 9.70%, and 9.51% of the variance in task-related BOLD signal, respectively (after varimax rotation). These networks are referred to as (1) response/attention network, (2) default mode network (DMN), and (3) visual attention network. The response/attention network (component 1) consisted of activation primarily in bilat-eral SMA, paracingulate gyrus, dorsal anterior cingulate, and insula, left somatomotor areas, and right cerebellum. The HDR onsets were staggered according to the delay length (Figure 3.1; ana- 41 tomical coordinates in Table 3.2). The 2 (load) × 2 (delay) × 10 (time) repeated measures ANOVA revealed significant main effects of cognitive load, F(1, 36) = 21.053, p < .001, ηp2 = .369; delay, F(1, 36) = 32.473, p < .001, ηp2 = .474; and time, F(9, 324) = 25.336, p < .001, ηp2 = .413. Further, significant interactions emerged for load × time, F(9, 324) = 8.400, p < .001, ηp2 = .189; and delay × time, F(9, 324) = 78.078, p < .001, ηp2 = .684; but not for load × delay or load × delay × time (ps > .15). Together, these results indicate involvement of this network in retriev-al/response, with the staggered somatomotor activity suggesting motor-response processes (i.e., the “yes”/”no” button-press); however, the main effect of cognitive load, due to greater engage-ment of this network in the 6-letter condition than in the 4-letter condition (mean predictor weights = 0.07 and 0.03, respectively), suggests a role in cognition/attention as well. The DMN (component 2) consisted primarily of bilateral deactivation in known DMN regions, including precuneus, posterior cingulate gyrus, anterior paracingulate gyrus, superior lateral occipital cortex, and middle temporal gyrus (Figure 3.2; anatomical coordinates in Table 3.3). Component 2 is referred to as the DMN for consistency with fMRI literature convention (Buckner et al., 2008; Raichle, 2015; Yeo et al., 2011). Significant main effects emerged for load, F(1, 36) = 16.320, p < .001, ηp2 = .312; delay, F(1, 36) = 21.157, p < .001, ηp2 = .370; and time, F(9, 324) = 36.854, p < .001, ηp2 = .506. Significant interactions emerged for load × delay, F(1, 36) = 9.007, p = .005, ηp2 = .200; load × time, F(9, 324) = 6.016, p < .001, ηp2 = .143; and delay × time, F(9, 324) = 15.512, p < .001, ηp2 = .301; but not load × delay × time (p > .08). The load-dependence and extension of the HDR into late post-stimulus time points when a delay pe-riod was present suggests a cognitive process engaged throughout the trial.  42 The visual attention network (component 3) consisted primarily of bilateral activation in occipital cortex (extending dorsally into the superior parietal lobule), SMA, precentral gyrus, and thalamus, which peaked relatively early following trial onsets (Figure 3.3; anatomical coordi-nates in Table 3.4). Significant main effects emerged for load, F(1, 36) = 22.020, p < .001, ηp2 = .380; delay, F(1, 36) = 13.918, p = .001, ηp2 = .279; and time, F(9, 324) = 129.140, p < .001, ηp2 = .782. Significant interactions emerged for load × time, F(9, 324) = 13.983, p < .001, ηp2 = .280; and delay × time, F(9, 324) = 20.250, p < .001, ηp2 = .360; but not for load × delay or load × de-lay × time (ps > .25). The activation of frontoparietal and primary/association visual cortices, as well as the dependence of HDR magnitude on cognitive load, suggest involvement of this net-work in visual attention; the absence of a second HDR peak in the 4-second delay conditions suggests that this activity underlies some cognitive process over and above basic visual percep-tion. 3.4.2. Multi-experiment fMRI-CPCA results 3.4.2.1. Descriptive summary  Seven components were extracted from the combined WM and TGT data, accounting for 9.79%, 7.80%, 7.26%, 6.98%, 4.15%, 3.30%, and 2.54% of the variance in task-related BOLD signal, respectively (after varimax rotation). All components were interpretable and accorded with the event timing of both tasks; however, components 5-7 appeared to underlie basic sensory or vascular processes that were not of interest, and so are not discussed in detail in this chapter, but are presented in Figures 3.5 to 3.7, with statistical results listed in Table 3.5. Components 1-3 all exhibited some configuration of frontoparietal connectivity including activation in dorsome-dial prefrontal cortex, and therefore merited further examination. Component 4 appeared to com-prise DMN deactivation, which has been proposed to be important for WM and other cognitive  43 tasks (e.g., Santangelo & Bordier, 2019; Vatansever, Menon, Manktelow, Sahakian, & Stamatakis, 2015; Woodward et al., 2016). Components 1-4, which accounted for 34.53% of var-iance in task-related BOLD signal, are described below for clarity, followed by statistical results. These networks are referred to as (1) response network, (2) visual attention network, (3) internal attention network, and (4) DMN. Figure 3.4 presents a side-by-side comparison between these four components and the three components extracted in the single-experiment WM analysis, in-cluding surface representations and HDR plots for the WM task. The response network (component 1), which overlapped with component 1 in the single-experiment WM analysis, consisted of activation primarily in the SMA, dorsal anterior cingulate, left somatomotor areas, cerebellum, and insula (Figure 3.8; anatomical coordinates in Table 3.6). Activation of this network was suppressed in the TGT task, but exhibited a pattern of activation in the WM task consistent with the timing of response processes (i.e., late initiation of the HDR, with the onsets in the 4-second delay conditions occurring approximately 4 seconds after the on-sets in the 0-second delay conditions). Although this network was similar to component 1 in the single-experiment analysis, the prefrontal cortex and insula activation was slightly posterior to that of the single-experiment results (Figure 3.1), even in regions such as the medial prefrontal cortex that were common to the two analyses. Further, the HDR was less load-dependent than that of the single-experiment WM analysis, suggesting isolation of response processes. These observations suggest that this network underlies motor responses with the right hand. The visual attention network (component 2), which was quite similar to the visual atten-tion network in the single-experiment WM analysis (component 3), consisted of activation in visual cortex (extending dorsally into parietal regions), SMA, precentral gyrus, and thalamus (Figure 3.10; anatomical coordinates in Table 3.7). Unlike component 1, component 2 was active  44 in both the TGT and WM task. As in component 3 from the single-experiment WM analysis, ac-tivation was load-dependent and peaked early in the WM task, but did not exhibit a second peak when the probe stimulus was displayed after a 4-second delay. This suggests that the visual ac-tivity was related to some attentional process over and above primary visual perception, particu-larly during the encoding phase of the WM task. Unlike the visual attention network (component 2), the internal attention network (com-ponent 3) was not identified in the single-experiment analysis, but emerged by way of combining the WM data with the TGT data. The internal attention network consisted of activation of paracingulate and superior frontal gyrus, DLPFC, frontal poles, anterior insula, and supra-marginal gyrus (Figure 3.12; anatomical coordinates in Table 3.8). In the WM task, peak activa-tion was strong for the maintenance phase of trials in which there was a 4-second delay before the probe was displayed, but modest for trials in which there was no delay, especially for the 4-letter condition. In the TGT task, activity was much greater in the generating condition than in the hearing condition. A possible explanation for this is that this network subserves volitional attention to internal stimuli representations. The DMN, (component 4), similar to the DMN in the single-experiment analysis (com-ponent 3), primarily consisted of bilateral deactivation of the precuneus, posterior cingulate gy-rus, anterior paracingulate gyrus, superior lateral occipital cortex, and middle temporal gyrus (Figure 3.14; anatomical coordinates in Table 3.9). As in the single-experiment analysis, this de-activation was load-dependent and sustained throughout the trials. Component 4 is referred to as the DMN for consistency with fMRI literature convention (Buckner et al., 2008; Raichle, 2015; Yeo et al., 2011).  45 3.4.2.2. Statistical results (WM task) For the response network in the WM task, the 2 (load) × 2 (delay) × 10 (time) repeated measures ANOVA revealed a significant main effect of time, F(2.77, 99.83) = 4.347, p = .008, ηp2 = .108; as well as significant interactions of load × time, F(3.37, 121.33) = 3.899, p = .008, ηp2 = .098; and delay × time, F(9, 324) = 63.435, p < .001, ηp2 = .638. No other main effects or interactions emerged for the response network (all ps > .08 for load, delay, load × delay, and load × delay × time). The delay × time interaction appeared to be driven by staggered onsets consistent with the probe timing (Figure 3.9). For the visual attention network, significant main effects emerged for load, F(1, 36) = 6.032, p = .019, ηp2 = .144; delay, F(1, 36) = 35.197, p < .001, ηp2 = .494; and time, F(9, 324) = 162.133, p < .001, ηp2 = .818. Significant interactions emerged for load × time, F(9, 324) = 17.142, p < .001, ηp2 = .323; and delay × time, F(9, 324) = 22.107, p < .001, ηp2 = .380; but not for load × delay or load × delay × time (both ps > .07). The main effect of load was due to greater mean activation in the 6-letter condition than in the 4-letter condition (mean predictor weights = 0.049 and 0.064 for the 4-letter and 6-letter conditions, respectively). The main effect of delay was due to greater mean activation in the 0s delay condition than in the 4s delay condition (mean predictor weights = 0.08 and 0.04 for 0s and 4s delay, respectively). Delay effects appeared to be driven by a more sustained HDR in the 0s delay condition, perhaps due to the appearance of the probe stimulus immediately after the encoding phase (Figure 3.11).   For the internal attention network, significant main effects emerged for cognitive load, F(1, 36) = 69.448, p < .001, ηp2 = .659; delay, F(1, 36) = 43.505, p < .001, ηp2 = .547; and time, F(9, 324) = 22.954, p < .001, ηp2 = .389. Significant interactions emerged for load × delay,  46 F(1, 36) = 4.625, p = .038, ηp2 = .114; load × time, F(9, 324) = 15.451, p < .001, ηp2 = .300; de-lay × time, F(9, 324) = 31.150, p < .001, ηp2 = .464; and load × delay × time, F(4.34, 156.29) = 3.356, p = .009, ηp2 = .085. The main effect of load was due to greater mean activation in the 6-letter condition compared with the 4-letter condition (mean predictor weights = 0.023 and 0.106 for 4 letters and 6 letters, respectively). The main effect of delay was due to greater mean activa-tion in the 4s delay condition than in the 0s delay condition (mean predictor weights = 0.03 and 0.10 for 0s and 4s delay, respectively). The effect of delay was greater in the 6-letter condition than in the 4-letter condition (Figure 3.13A). Delay effects were driven by not only a more sus-tained HDR in the 4s delay condition, but also a greater HDR peak compared with that of the 0s delay condition (Figure 3.13B). Finally, for DMN deactivation, significant main effects emerged for cognitive load, F(1, 36) = 22.033, p < .001, ηp2 = .380; delay, F(1, 36) = 26.635, p < .001, ηp2 = .425; and time, F(9, 324) = 42.880, p < .001, ηp2 = .544. Significant interactions emerged for load × delay, F(1, 36) = 6.511, p = .015, ηp2 = .153; load × time, F(9, 324) = 6.994, p < .001, ηp2 = .163; and delay × time, F(9, 324) = 18.695, p < .001, ηp2 = .342. The main effect of load was driven by greater mean DMN deactivation in the 6-letter condition compared with the 4-letter condition (mean predictor weights = 0.140 and 0.190 for 4 letters and 6 letters, respectively). The main ef-fect of delay was due to greater mean activation in the 4s delay condition than in the 0s delay condition (mean predictor weights = 0.14 and 0.19 for 0s and 4s delay, respectively). The effect of delay was greater in the 6-letter condition than in the 4-letter condition (Figure 3.15A). Effects related to delay appeared to be driven by a more sustained HDR in the 4s delay condition than in  47 the 0s delay condition, but not a greater magnitude of deactivation in the 4s delay condition per se (Figure 3.15B). 3.4.2.3. Statistical results (TGT task) For the response network in the TGT task, the 2 (condition) × 10 (time) ANOVA re-vealed a main effect of time, F(9, 279) = 32.778, p < .001, ηp2 = .514. Neither the condition nor the condition × time interaction were significant (both ps > .25). For the visual attention network a significant main effect emerged for time, F(9, 279) = 97.039, p < .001, ηp2 = .758, but not for condition or a condition × time interaction (both ps > .06).  For the internal attention network, significant main effects emerged for condition, F(1, 31) = 11.157, p = .002, ηp2 = .265; and time, F(9, 279) = 14.945, p < .001, ηp2 = .325. The condition × time interaction was also significant, F(9, 279) = 9.851, p < .001, ηp2 = .241. Effects related to condition were due to greater mean activation in the generating condition compared with the hearing condition as a whole (mean predictor weights = 0.045 and 0.018 for generating and hearing, respectively), due to there being minimal response in the hearing condition (Figure 3.12B).  Finally, analysis of the DMN deactivation revealed a significant main effect of time, F(9, 279) = 18.292, p < .001, ηp2 = .371, but no effect of condition or condition × time interaction (ps > .45). 3.5. Discussion of Multi-Experiment fMRI-CPCA In the present analysis, a multi-experiment approach was demonstrated that helped delin-eate networks underlying a WM task. In addition to separation of task-based networks, the multi-experiment results extended previous analyses of brain networks involved in WM by allowing  48 the examination of temporal characteristics of these networks across different types of task de-mands. A particularly notable finding from the multi-experiment analysis was the emergence of a network exhibiting a plausible mechanism for attention to internal stimuli representations in both tasks, which is referred to as the “internal attention” network. The results of the single-experiment WM analysis replicated previous findings using fMRI-CPCA and other methods, with somatomotor activity seemingly underlying the retriev-al/response phase (based on its late, staggered peaks; component 1, Figure 3.1), visual, frontal, and thalamic activity initiating during encoding (having an early onset and return to baseline; component 3, Figure 3.3), and the DMN (component 2, Figure 3.2) exhibiting sustained, load-dependent deactivation (Braunlich et al., 2015; Metzak et al., 2011; Metzak et al., 2012; Woodward et al., 2006; Woodward et al., 2013). In previous studies, it was proposed that deacti-vation of the DMN supports WM maintenance, suggested from the observation that this load-dependent deactivation tends to peak intermediately between networks seeming to underlie en-coding and retrieval/response. However, the current set of multi-experiment results suggests that the internal attention network is a better candidate for maintenance processes. First, the onset of DMN deactivation occurred quite early in the post-stimulus time series. Second, neither the pre-sent results nor a comparable previous study (Woodward et al., 2013) showed a meaningful de-crease in magnitude of DMN deactivation when there was no delay before the probe stimulus. A functional brain network supporting maintenance-related processes would be expected to initiate activation several seconds into the post-stimulus time series – following networks activated from the trial onset – and show a strong increase in peak magnitude when the maintenance period is longer; both of these characteristics were observed in the internal attention network, but not in the DMN.  49 The response network that emerged from the multi-experiment analysis (component 1; Figure 3.8) was similar to the response/attention network in the single-experiment WM analysis (component 1; Figure 3.1) with respect to its spatial configuration and staggered activation pat-terns. However, in the single-experiment analysis, a main effect of cognitive load emerged for the response/attention network, suggesting that the network also supported cognitive/attentional processes; by contrast, the response network in the multi-experiment analysis appeared to be more specifically related to the motor response, supported in part by the observation that this network was suppressed in the TGT task (which did not require a response), and that it did not exhibit a main effect of cognitive load. Interestingly, insula and prefrontal activations in the re-sponse/attention network were situated more anteriorly in the single-experiment analysis than in the response network extracted from the multi-experiment analysis (compare Figures 3.1A and 3.8A, respectively), which may reflect meaningful anatomical differences with respect to higher-level cognition versus motor planning (Fuster, 2004). These findings suggest that the anatomical and functional characteristics of the response/attention network from the single-experiment anal-ysis may comprise aspects of both the response network and internal attention network, as will be discussed below, while the multi-experiment version was a clearer delineation of the response network. This illustrates an advantage of simultaneously analyzing multiple tasks; that is, the combining of cognitive/behavioral processes into a single network was reduced when a task with relevant overlapping and non-overlapping cognitive demands was simultaneously analyzed. The most striking result in the multi-experiment analysis was the emergence of a novel network (component 3, internal attention) that was considerably more engaged in the WM 4s de-lay condition than in the 0s delay condition (Figure 3.13B), and the onset of which followed that of the visual attention network, but preceded the response network (Figure 3.4D). Although pre- 50 vious research has identified networks underlying WM which spatially resemble this internal at-tention network (Emch, von Bastian, & Koch, 2019; Meda et al., 2009; Rottschy et al., 2012; Vasic, Walter, Sambataro, & Wolf, 2009; Wolf et al., 2008), the present study expands on this in providing the task-related time series, which strongly suggest engagement of this network not only during WM, but also during internal thought that is not required to be held in memory (in the TGT task). Moreover, the present results help clarify which aspect of a WM task this network underlies. Comparing the series of WM task HDRs across networks (as displayed in Figure 3.4D), it is possible that activation of the visual attention network supports visual attention to the to-be-encoded letter string at the trial onset, and its diminishing intensity reflects a coordinated shift from external to internal attention as the internal attention network becomes engaged; this may be required in order to maintain a mental representation of the stimuli. As to why this net-work is somewhat active in the 0s delay conditions despite the absence of the requirement to maintain items in memory, it could be the case that internal attention either (1) initiates during the late encoding phase of the task in anticipation of the possible maintenance phase, (2) overlaps with response anticipation and execution, or (3) some combination of both, as participants do not know until completion of the encoding phase whether or not there will be a delay period. There-fore, it is plausible that both internal (maintenance) processes and response anticipation process-es become engaged before the end of the encoding phase and exhibit similar HDR shapes (when no delay actually occurs), particularly if sustaining internal attention into the response phase is required for accurate recall. This similarity in timing could explain why component 1 in the sin-gle-experiment analysis appeared to be comprised of both cognitive and motor response process-es. The presence of an overlapping cognitive process between WM and TGT (i.e., internal representations of stimuli) and a non-overlapping process (i.e., motor response) may have al- 51 lowed these temporally correlated processes to separate onto different networks when the two datasets were analyzed simultaneously using multi-experiment fMRI-CPCA.  Although descriptive labels were assigned for the purposes of discussion (“response”, “visual attention”, etc.), the postulated function assigned to each network may not be the only function it may serve across the vast range of cognitive constructs examined in the wider cogni-tive neuroscience literature – the complexities of structure-function mapping are well-summarized by Poldrack (2010). For example, it cannot be determined from the present findings whether the internal attention network has a broad role of generating internal representations of task-relevant stimuli, or whether it underlies verbal tasks only. A larger set of tasks that, togeth-er, independently manipulate internal representations of various modalities, inner speech, cogni-tive control, memory demand, and a range of other constructs would ideally be combined in order to better infer the functional specificity of the internal attention network and other patterns of connectivity observed in the present study. However, the limited experimental design was ap-propriate for this relatively narrow, proof-of-concept study which aimed to demonstrate how the addition of even one carefully chosen task can provide valuable information that could be missed in the analysis of a single cognitive construct.    52 3.6. Chapter 3 Tables Table 3.1. Demographic information for the WM task and the TGT task datasets (from WM-TGT multi-experiment analysis with healthy controls only). Standard deviations in parentheses.  WM (n = 37) TGT (n = 32) Age 28.16 (8.70) 28.75 (8.58) Years of education 15.86 (2.16) 15.58 (1.81) Quick estimated IQ 99.59 (11.89) 97.09 (11.21) Gender distribution (female/male) 27/10 13/19 Handedness (L/M/R) 3/2/32 3/1/28 Socioeconomic status factor score 58.24 (15.68) 65.75 (14.97) Note. IQ = intelligence quotient; L = left-handed; M = mixed handedness; R = right-handed; WM = Working Memory; TGT = Thought Generation Task.   53 Table 3.2. WM task fMRI-CPCA, response/attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task response network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 3,993 107,811     Precentral gyrus   4 -36 -22 59 Postcentral gyrus   3 -45 -31 53 Postcentral gyrus   3 -51 -25 44 Central opercular cortex   48 -54 -19 20 Precentral gyrus   6 -33 -7 62 Supplementary motor area   6 -3 -4 53 Superior lateral occipital cortex   7 -12 -70 53 Parietal operculum cortex   48 -57 -40 23 Cluster 1: right hemisphere       Paracingulate gyrus   32 3 17 47        Cluster 2: left hemisphere 1,456 39,312     Cerebellum VI   n/a -30 -55 -31 Lingual gyrus   18 -6 -73 -13 Cluster 2: bilateral       Lingual gyrus   17 0 -79 2 Cluster 2: right hemisphere       Cerebellum V   n/a 18 -52 -22 Cerebellum VIIIa   n/a 18 -64 -49 Cerebellum V   n/a 6 -61 -16 Cerebellum crus II   n/a 6 -73 -37 Cerebellum V   n/a 12 -58 -28        Cluster 3: left hemisphere 930 25,110     Insular cortex   48 -30 20 5 Precentral gyrus   44 -54 8 29 Precentral gyrus   6 -54 8 23 Insular cortex   48 -39 -1 8 Middle frontal gyrus   46 -36 35 26 Frontal pole   46 -30 47 20 Temporal pole   48 -51 11 -4 (Table 3.2 continued on next page)    (Table 3.2, continued from previous page) 54 WM task response network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Putamen   48 -30 -16 2        Cluster 4: right hemisphere 549 14,823     Postcentral gyrus   48 57 -16 26 Superior parietal lobule   40 39 -46 47 Postcentral gyrus   3 54 -19 38 Posterior supramarginal gyrus   2 45 -37 50        Cluster 5: right hemisphere 438 11,826     Insular cortex   47 33 23 2 Inferior frontal gyrus, pars opercularis   6 57 11 20        Cluster 6: right hemisphere 227 6,129     Frontal pole   46 36 41 26        Cluster 7: right hemisphere 177 4,779     Middle frontal gyrus   6 33 -1 59        Cluster 8: left hemisphere 63 1,701     Thalamus   n/a -12 -19 8        Cluster 9: left hemisphere 54 1,458     Inferior lateral occipital cortex   37 -48 -64 5 Middle temporal gyrus, temporooccipital part   21 -48 -52 8        Cluster 10: right hemisphere 39 1,053     Thalamus   n/a 12 -16 8 Caudate   n/a 15 -1 14        Cluster 11: left hemisphere 16 432     Cerebellum VIIb   n/a -33 -58 -49        Cluster 12: right hemisphere 7 189     Precuneus cortex   7 9 -70 50        Cluster 13: left hemisphere 2 54     Intracalcarine cortex   18 -15 -70 2  55 Table 3.3. WM task fMRI-CPCA, default mode network (DMN, component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task default mode network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 27 729     Supplementary motor area   6 -3 8 56        Cluster 2: left hemisphere 9 243     Precentral gyrus   6 -54 2 44        Negative loadings              Cluster 1: left hemisphere 2,843 76,761     Superior frontal gyrus   9 -21 35 44 Cluster 1: right hemisphere       Frontal pole   10 3 56 11 Superior frontal gyrus   9 24 29 44 Frontal pole   9 15 47 38 Frontal pole   9 18 44 41        Cluster 2: left hemisphere 2,476 66,852     Posterior cingulate gyrus   23 -3 -46 35 Cluster 2: bilateral       Precuneus cortex   31 0 -70 29 Posterior cingulate gyrus   23 0 -16 38        Cluster 3: left hemisphere 853 23,031     Superior lateral occipital cortex   39 -48 -70 29        Cluster 4: right hemisphere 848 22,896     Superior lateral occipital cortex   39 48 -64 26 Middle temporal gyrus, temporooccipital part   21 60 -40 -4        Cluster 5: right hemisphere 269 7,263     Anterior middle temporal gyrus   21 57 -4 -16 Temporal pole   38 42 20 -28 (Table 3.3 continued on next page)    (Table 3.3, continued from previous page) 56 WM task default mode network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Temporal pole   21 48 8 -31        Cluster 6: left hemisphere 237 6,399     Anterior middle temporal gyrus   21 -54 -4 -19 Temporal pole   21 -48 11 -28 Frontal orbital cortex   38 -42 26 -16 Anterior inferior temporal gyrus   20 -51 -4 -37        Cluster 7: left hemisphere 106 2,862     Cerebellum crus II   n/a -21 -79 -37        Cluster 8: right hemisphere 92 2,484     Cerebellum crus II   n/a 27 -76 -37        Cluster 9: right hemisphere 91 2,457     Frontal orbital cortex   47 42 32 -10 Inferior frontal gyrus, pars triangularis   45 54 26 11 Inferior frontal gyrus, pars triangularis   45 51 29 8 Inferior frontal gyrus, pars triangularis   45 51 29 -1        Cluster 10: right hemisphere 27 729     Planum temporale   42 60 -28 17        Cluster 11: right hemisphere 23 621     Lingual gyrus   37 27 -40 -10        Cluster 12: left hemisphere 19 513     Lingual gyrus   37 -27 -43 -10        Cluster 13: right hemisphere 12 324     Heschl's gyrus   48 57 -10 5        Cluster 14: left hemisphere 9 243     Middle temporal gyrus, temporooccipital part   37 -63 -55 2        (Table 3.3 continued on next page)           (Table 3.3, continued from previous page) 57 WM task default mode network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 15: right hemisphere 5 135     Amygdala   n/a 21 -4 -19 Amygdala   n/a 30 2 -19        Cluster 16: right hemisphere 4 108     Cerebellum IX   n/a 9 -52 -46        Cluster 17: left hemisphere 4 108     Temporal pole   48 -33 2 -19   58 Table 3.4. WM task fMRI-CPCA, visual attention network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task visual attention network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 3,364 90,828     Occipital pole   18 -18 -91 -7 Superior lateral occipital cortex   19 -24 -70 32 Superior lateral occipital cortex   7 -21 -61 50 Temporal occipital fusiform cortex   37 -36 -49 -19        Cluster 2: bilateral 3,164 85,428     Vermis crus II   n/a 0 -73 -28 Cluster 2: right hemisphere       Occipital fusiform gyrus   18 21 -88 -4 Superior lateral occipital cortex   19 27 -67 35 Superior lateral occipital cortex   7 24 -61 56 Temporal occipital fusiform cortex   37 36 -46 -19 Cerebellum crus II   n/a 9 -76 -40        Cluster 3: left hemisphere 774 20,898     Precentral gyrus   6 -51 -1 47 Precentral gyrus   44 -42 5 29 Precentral gyrus   6 -45 2 32 Precentral gyrus   6 -57 2 26 Middle frontal gyrus   6 -27 -1 50        Cluster 4: left hemisphere 224 6,048     Supplementary motor area   6 -3 5 62        Cluster 5: right hemisphere 140 3,780     Precentral gyrus   6 54 -4 47        Cluster 6: right hemisphere 57 1,539     Precentral gyrus   6 33 -19 71 Precentral gyrus   4 42 -16 65        (Table 3.4 continued on next page)    (Table 3.4, continued from previous page) 59 WM task visual attention network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 7: right hemisphere 43 1,161     Precentral gyrus   44 42 5 26        Cluster 8: left hemisphere 40 1,080     Thalamus   n/a -21 -28 -4        Cluster 9: right hemisphere 39 1,053     Precentral gyrus   6 30 -1 47        Cluster 10: right hemisphere 35 945     Thalamus   n/a 21 -28 -1        Cluster 11: bilateral 26 702     Vermis IX   n/a 0 -58 -37        Cluster 12: left hemisphere 13 351     Anterior supramarginal gyrus   40 -42 -37 41        Cluster 13: right hemisphere 9 243     Cerebellum VIIb   n/a 24 -70 -49        Cluster 14: right hemisphere 5 135     Anterior supramarginal gyrus   2 45 -34 44        Negative loadings              Cluster 1: left hemisphere 19 513     Lingual gyrus   17 -9 -76 -7        Cluster 2: right hemisphere 2 54     Occipital pole   18 12 -88 20   60 Table 3.5. WM-TGT multi-experiment fMRI-CPCA, components 6 and 7 (occipital and auditory regions, respectively): Results of repeated measures analyses of variance (ANOVAs). Significant results are presented in bold font. Component 5 reflected artifact signal and was not analyzed. Component 6 (occipital (de)activation) WM task DF DFerror F p ηp2 Load 1 36 4.84 .034* .119 Delay 1 36 6.717 .014* .157 Time 9 324 6.691 < .001*** .157 Load × delay 1 36 3.044 .090 .078 Load × time 4.17 149.96 3.748 .006** .094 Delay × time 9 324 12.397 < .001*** .256 Load × delay × time 5.30 190.86 2.073 .066 .054 TGT task DF DFerror F p ηp2 Condition 1 31 1.468 .235 .045 Time 9 279 48.944 < .001*** .612 Condition × time 3.22 99.95 0.758 .529 .024 Component 7 (auditory network) WM taska DF DFerror F p ηp2 - - - - - - TGT task DF DFerror F p ηp2 Condition 1 31 39.018 < .001*** .557 Time 9 279 92.004 < .001*** .748 Condition × time 9 279 51.879 < .001*** .626 aAnalysis of the auditory network (component 7) was not performed for WM task due to absence of meaningful HDR shape (see Figure 3.7, WM conditions). DF = degrees of freedom; * = p < .05; ** = p < .01; *** = p < .001.  61 Table 3.6. WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM-TGT response network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 4,880 131,760     Postcentral gyrus   3 -42 -22 56 Postcentral gyrus   3 -54 -22 44 Central opercular cortex   48 -51 -22 20 Supplementary motor area   6 -3 -7 53 Insular cortex   48 -39 -1 8 Precentral gyrus   6 -57 5 32 Cluster 1: right hemisphere       Superior frontal gyrus   6 27 -4 62 Posterior cingulate gyrus   23 12 -22 41        Cluster 2: right hemisphere 1,982 53,514     Postcentral gyrus   48 57 -19 35 Superior parietal lobule   2 36 -43 59 Precentral gyrus   6 57 8 23 Precentral gyrus   6 57 8 17 Insular cortex   48 39 -1 11 Precuneus cortex   5 12 -58 59 Precuneus cortex   31 12 -43 53        Cluster 3: left hemisphere 417 11,259     Cerebellum VI   n/a -30 -49 -31 Cluster 3: right hemisphere       Cerebellum V   n/a 18 -52 -22        Cluster 4: right hemisphere 169 4,563     Inferior lateral occipital cortex   37 45 -61 2        Cluster 5: right hemisphere 154 4,158     Cerebellum VIIIb   n/a 18 -61 -52 Vermis VIIIa   n/a 3 -70 -37        (Table 3.6 continued on next page)    (Table 3.6, continued from previous page) 62 WM-TGT response network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 6: left hemisphere 153 4,131     Inferior lateral occipital cortex   37 -48 -67 5        Cluster 7: left hemisphere 36 972     Thalamus   n/a -12 -19 5        Cluster 8: left hemisphere 31 837     Frontal pole   46 -30 38 26        Cluster 9: right hemisphere 27 729     Frontal pole   46 30 41 26        Cluster 10: left hemisphere 22 594     Cerebellum VIIIa   n/a -30 -55 -49        Cluster 11: right hemisphere 15 405     Thalamus   n/a 12 -16 5        Cluster 12: left hemisphere 12 324     Cerebellum VIIb   n/a -15 -70 -49 Cerebellum VIIIa   n/a -21 -64 -52   63 Table 3.7. WM-TGT multi-experiment fMRI-CPCA, visual attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM-TGT visual attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 6,388 172,476     Occipital pole   18 -18 -91 -7 Temporal occipital fusiform cortex   37 -36 -49 -19 Superior lateral occipital cortex   19 -24 -70 32 Superior lateral occipital cortex   7 -21 -64 50 Cluster 1: bilateral       Vermis crus II   n/a 0 -73 -28 Cluster 1: right hemisphere       Occipital fusiform gyrus   18 21 -88 -4 Superior lateral occipital cortex   19 27 -70 32 Temporal occipital fusiform cortex   37 36 -46 -19 Superior lateral occipital cortex   7 24 -61 53 Cerebellum VI   n/a 9 -73 -22        Cluster 2: left hemisphere 755 20,385     Precentral gyrus   6 -51 -1 47 Precentral gyrus   44 -42 5 29 Precentral gyrus   6 -57 2 23 Middle frontal gyrus   6 -27 -4 47        Cluster 3: right hemisphere 392 10,584     Precentral gyrus   6 57 -1 47 Middle frontal gyrus   6 33 -1 47 Precentral gyrus   6 36 -19 71 Precentral gyrus   4 42 -16 65        Cluster 4: left hemisphere 218 5,886     Supplementary motor area   6 -3 5 62        Cluster 5: right hemisphere 48 1,296     Precentral gyrus   44 42 8 26        (Table 3.7 continued on next page)    (Table 3.7, continued from previous page) 64 WM-TGT visual attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 6: left hemisphere 43 1,161     Thalamus   n/a -21 -28 -1        Cluster 7: right hemisphere 32 864     Thalamus   n/a 21 -28 -1        Cluster 8: bilateral 15 405     Vermis IX   n/a 0 -58 -37        Negative loadings              Cluster 1: left hemisphere 4 108     Lingual gyrus   18 -9 -73 -7   65 Table 3.8. WM-TGT multi-experiment fMRI-CPCA, internal attention network (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM-TGT internal attention network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: bilateral 4,030 108,810     Superior frontal gyrus   6 0 17 53 Cluster 1: left hemisphere       Insular cortex   47 -30 23 2 Inferior frontal gyrus, pars opercularis   48 -48 14 29 Middle frontal gyrus   6 -36 -1 59 Middle frontal gyrus   45 -42 32 29 Frontal pole   46 -33 50 17 Superior frontal gyrus   6 -18 11 65 Caudate   n/a -15 8 5 Thalamus   n/a -12 -16 8        Cluster 2: right hemisphere 1,656 44,712     Insular cortex   47 33 23 2 Middle frontal gyrus   45 42 35 29 Middle frontal gyrus   6 39 5 56 Inferior frontal gyrus, pars opercularis   44 54 14 17        Cluster 3: left hemisphere 932 25,164     Superior parietal lobule   40 -33 -52 44 Superior lateral occipital cortex   7 -12 -73 53        Cluster 4: left hemisphere 548 14,796     Occipital pole   17 -6 -97 5 Cluster 4: bilateral       Lingual gyrus   17 0 -82 -4 Cluster 4: right hemisphere       Cerebellum VI   n/a 9 -73 -25 Occipital pole   17 12 -94 11 Cerebellum VI   n/a 27 -61 -28        (Table 3.8 continued on next page)           (Table 3.8, continued from previous page) 66 WM-TGT internal attention network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 5: right hemisphere 304 8,208     Superior parietal lobule   40 33 -52 41        Cluster 6: left hemisphere 132 3,564     Cerebellum VI   n/a -30 -58 -31        Cluster 7: right hemisphere 116 3,132     Cerebellum VIIb   n/a 27 -70 -52        Cluster 8: left hemisphere 83 2,241     Inferior temporal gyrus, temporooccipital part   37 -48 -55 -16        Cluster 9: right hemisphere 45 1,215     Caudate   n/a 12 11 2 Caudate   n/a 15 5 11        Cluster 10: left hemisphere 28 756     Posterior supramarginal gyrus   42 -60 -43 20        Cluster 11: left hemisphere 13 351     Cerebellum crus I   n/a -9 -76 -25        Cluster 12: left hemisphere 7 189     Posterior middle temporal gyrus   21 -57 -31 -4        Cluster 13: left hemisphere 3 81     Precentral gyrus   4 -33 -25 53   67 Table 3.9. WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each peak. WM-TGT default mode network (component 4) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 4 108     Supplementary motor area   6 -3 8 56        Negative loadings              Cluster 1: left hemisphere 3,729 100,683     Superior frontal gyrus   9 -21 35 44 Cluster 1: right hemisphere       Frontal pole   10 3 56 11 Superior frontal gyrus   9 24 29 44        Cluster 2: bilateral 1,749 47,223     Posterior cingulate gyrus   23 0 -49 35 Cluster 2: right hemisphere       Superior lateral occipital cortex   19 15 -85 41        Cluster 3: left hemisphere 932 25,164     Superior lateral occipital cortex   39 -51 -70 29        Cluster 4: right hemisphere 840 22,680     Superior lateral occipital cortex   39 51 -64 26        Cluster 5: right hemisphere 261 7,047     Anterior middle temporal gyrus   21 57 -4 -19 Temporal pole   38 42 17 -28        Cluster 6: left hemisphere 229 6,183     Anterior middle temporal gyrus   21 -54 -4 -19 Temporal pole   21 -48 8 -28        Cluster 7: right hemisphere 93 2,511     Frontal orbital cortex   47 42 32 -10 (Table 3.9 continued on next page)    (Table 3.9, continued from previous page) 68 WM-TGT default mode network (component 4) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Inferior frontal gyrus, pars triangularis   45 51 26 14 Frontal orbital cortex   47 30 32 -13        Cluster 8: left hemisphere 33 891     Cerebellum crus I   n/a -27 -76 -37        Cluster 9: right hemisphere 22 594     Cerebellum crus II   n/a 27 -76 -37        Cluster 10: left hemisphere 4 108     Frontal orbital cortex   38 -42 26 -13        Cluster 11: right hemisphere 2 54     Frontal orbital cortex   48 27 20 -16   69 3.7. Chapter 3 Figures Figure 3.1. WM task fMRI-CPCA, response/attention network (component 1): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.21, max = 0.37; no negative loadings above threshold). Images are displayed in neuro-logical orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. 4L = 4 letters; 6L = 6 letters.   70 Figure 3.2. WM task fMRI-CPCA, default mode network (DMN, component 2): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.16, max = 0.19; blue/green = negative loadings, min = -0.27, max = -0.16). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. Y axis is reversed (negative down, positive up) to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis reflect deactivation). 4L = 4 letters; 6L = 6 letters.   71 Figure 3.3. WM task fMRI-CPCA, visual attention network (component 3): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.21, max = 0.37). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condi-tion. 4L = 4 letters; 6L = 6 letters.    72 Figure 3.4. Surface representations and WM HDR shapes in the single-experiment vs. the multi-experiment fMRI-CPCA. A (top left): surface representations of components 1-3 from the single-experiment WM analysis. B (bottom left): predictor weights plotted over post-stimulus time for components 1-3 from the WM single-experiment analysis, averaged over load; component 2 predictor weights have been multiplied by -1 so that values above X axis reflect activation for all components. C (top right): surface representations of components 1-4 from the multi-experiment analysis (response, visual attention, internal attention, and default mode networks). D (bot-tom right): WM task predictor weights plotted over post-stimulus time for components 1-4 from the multi-experiment analysis (re-sponse, visual attention, internal attention, and default mode networks), averaged over load; DMN predictor weights have been multiplied by -1 so that values above X axis reflect activation for all components. Comp = component; DMN = default mode network.    73 Figure 3.5. WM-TGT multi-experiment fMRI-CPCA, blood flow artifact (component 5): Ana-tomical and temporal characteristics. Many peak areas were outside of the brain, likely reflecting blood drainage after a neural response. A (top): dominant 10% of loadings (red/yellow = posi-tive loadings, min = 0.10, max = 0.21; no negative loadings above threshold). Images are dis-played in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. 4L = 4 letters; 6L = 6 letters.   74 Figure 3.6. WM-TGT multi-experiment fMRI-CPCA, occipital (de)activation (component 6): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = posi-tive loadings, min = 0.08, max = 0.29; blue/green = negative loadings, min = -0.17, max = -0.08). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. 4L = 4 letters; 6L = 6 letters.   75 Figure 3.7. WM-TGT multi-experiment fMRI-CPCA, auditory network (component 7): Anatom-ical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive load-ings, min = 0.07, max = 0.24; blue/green = negative loadings, min = -0.11, max = -0.07). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. 4L = 4 let-ters; 6L = 6 letters.   76 Figure 3.8. WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Ana-tomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.15, max = 0.30; no negative loadings above threshold). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. 4L = 4 letters; 6L = 6 letters.    77 Figure 3.9. WM task from the WM-TGT multi-experiment fMRI-CPCA, response network (component 1): Estimated HDRs illustrating delay × time interaction. Asterisks indicate signifi-cant delay × time contrast between adjacent time bins. *** = p < .001.  -.30-.20-.10.00.10.20.30.402 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)Delay  Time0s delay4s delay****** 78 Figure 3.10. WM-TGT multi-experiment fMRI-CPCA, visual attention network (component 2): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = posi-tive loadings, min = 0.11, max = 0.54; blue/green = negative loadings, min = -0.12, max = -0.11). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. 4L = 4 letters; 6L = 6 letters.    79 Figure 3.11. WM task from the WM-TGT multi-experiment fMRI-CPCA, visual attention net-work (component 2): Estimated HDRs illustrating delay × time interaction. Asterisks indicate significant delay × time contrast between adjacent time bins. * = p < .05; ** = p < .01; *** = p < .001.  -.20-.10.00.10.20.30.40.502 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)Delay  Time0s delay4s delay************* 80 Figure 3.12. WM-TGT multi-experiment fMRI-CPCA, internal attention network (component 3): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.14, max = 0.30; no negative loadings above threshold). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. Asterisks indicate significant condition × time contrasts between adjacent time bins in the TGT task. 4L = 4 letters; 6L = 6 letters; * = p < .05; ** = p < .01; *** = p < .001.    81 Figure 3.13. WM task from the WM-TGT multi-experiment fMRI-CPCA, internal attention net-work (component 3): Graphs illustrating effects of cognitive load and delay length. A (top): mean predictor weights illustrating load × delay interaction. B (bottom): estimated HDRs illus-trating delay × time interaction (asterisks indicate significant delay × time contrast between adja-cent time bins). * = p < .05; ** = p < .01; *** = p < .001.    82 Figure 3.14. WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.13, max = 0.14; blue/green = negative loadings, min = -0.24, max = -0.13). Images are displayed in neurological orientation (left is left) with MNI z-axis coordinates. B (bottom): mean predictor weights plotted over post-stimulus time for each task and condition. Y axis has been reversed to facilitate interpretation (i.e., values above X axis indicate activation of blue/green voxels, and values below X axis indicate deactivation of blue/green voxels). 4L = 4 letters; 6L = 6 letters.    83 Figure 3.15. WM task from the WM-TGT multi-experiment fMRI-CPCA, default mode network (DMN, component 4): Graphs illustrating effects of cognitive load and delay length. Y axes are reversed (negative up, positive down) to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis reflect deactivation). A (top): mean predictor weights illus-trating load × delay interaction. B (bottom): predictor weights plotted over post-stimulus time to illustrate delay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins). * = p < .05; ** = p < .01; *** = p < .001.    84 Chapter 4: Characterizing Dominant Networks Underlying Neurocognitive Tasks Using Single-Experiment fMRI-CPCA 4.1. Aims and Hypotheses The previous chapter presented a preliminary multi-experiment fMRI-CPCA that was de-signed to obtain a more refined characterization of the functional brain networks underlying a Sternberg verbal WM task. Although the findings demonstrated that the addition of one task with relevant overlapping and non-overlapping cognitive processes can provide meaningful infor-mation that expands on a single-task analysis, a number of limitations should be noted. For ex-ample, the WM task and TGT are both highly verbal tasks; therefore, the addition of a visuospatial WM task, with a similar paradigm but non-verbal task content, would be valuable for determining the degree to which the observed networks are specific to tasks involving inner speech. Further, a more powerful manipulation may be required for dissociating internal versus external attention, as neither the WM task nor the TGT involve an externally-oriented task condi-tion that is both challenging and not concurrent with memory encoding. The addition of a task that engages other executive function domains, such as inhibition and task-switching, could help determine whether the network tentatively described as “volitional internal attention” is specifi-cally internally-oriented. Therefore, a 4-task fMRI-CPCA was designed, including the WM task and TGT task analysed in the previous chapter, with the addition of the Spatial Capacity (SCAP) task and the Task-Switch Inertia (TSI) task detailed in Chapter 2 (Sections 2.2.2 and 2.2.3, re-spectively). The main purpose of the present chapter was to describe the fundamental character-istics of the dominant networks underlying these tasks, as observed in healthy control participants. This includes a more detailed examination of the WM task with a larger sample size than the previous chapter (including participants who had also completed the TGT task, and so  85 had been excluded from the WM dataset so as not to have a mix of overlapping and non-overlapping participants in the TGT-WM analysis), although the same three networks were ex-pected to emerge (i.e., frontoparietal network underlying the response phase of the task, a visual attention network underlying the encoding phase, and sustained DMN deactivation). It was pre-dicted that these three networks would also emerge in the SCAP task, which encompasses the same WM task stages of encoding, maintenance, and recall/response. For the TSI task, it was expected that at least one bilateral, frontoparietal task-positive network and DMN deactivation would emerge, consistent with previous research of Stroop task-switching (Woodward et al., 2016). For the TGT data, it was predicted that, along with DMN deactivation, a left-dominant network comprising known language areas (pars opercularis of the inferior frontal gyrus and pos-terior superior temporal gyrus), SMA, and lateral occipital cortex would emerge and be more en-gaged in the thought generation condition than in the hearing condition, consistent with a previous analysis of this task (Lavigne, Rapin, et al., 2015). A number of factors were considered in the decision to carry out a new analysis of the TGT task rather than referring to published findings for comparison. First and foremost, the published analysis was carried out on a sample comprising healthy controls, schizophrenia patients, and bipolar disorder patients, whereas the intent of this chapter was to present “control” analyses derived from non-clinical participants. In addition, this work utilized more up-to-date preprocessing procedures (i.e., use of the SPM12 software package rather than SPM8), and although this may not necessarily result in meaningful differences between otherwise identical analyses, it would nevertheless add a confound to the present work.  86 4.2. Analysis 1: Working Memory Task 4.2.1. Methods 4.2.1.1. Participants 54 healthy participants who met all general inclusion criteria detailed in Chapter 2 (Sec-tion 2.1) were included in the present analysis. Task performance on the fMRI WM task was ex-amined to confirm participant engagement during the task, and any runs in which a participant achieved < 60% correct responses were excluded from the analysis (note that participants had a 50% chance of guessing correctly on any given trial). Demographic information for this sample is provided in Table 4.1. A brief questionnaire was administered to all participants to obtain their age, highest level of education, and any history of head injury, neurological conditions, drug use, and medication. In addition, visual acuity and color vision were assessed to ensure that partici-pants were able to view the task on an fMRI presentation screen. Handedness was measured with the AHPQ (Annett, 1970), and IQ was estimated with the Quick IQ test (Mortimer & Bowen, 1999). Participants were recruited via posters on the UBC campus, community bulletin boards, and on electronic bulletins such as Craigslist. Participants provided informed consent at the start of their testing sessions, and were compensated $10 per hour for time spent participating in the study plus $10 for travel expenses. Participants were also given a copy of their T1 MRI scan on a disc. 4.2.1.2. WM task design Details regarding the WM task design and procedures are presented in Chapter 2 (task protocol described in Section 2.2.1 and Figure 2.1; fMRI data acquisition and preprocessing de-scribed in Section 2.3 and Table 2.1). In summary, a string of 4 or 6 upper case consonants was displayed for 4 seconds, and then a single probe letter was displayed for 2 seconds. Participants  87 responded “yes” or “no” with a right-handed button press as to whether this probe letter was part of the first string of letters. Both the cognitive load (4 or 6 letters in the item set) and delay peri-od (0 or 4 seconds between the letter string and probe) were manipulated so as to facilitate iden-tification of the functional brain networks distinctly involved in encoding, maintenance, and response/recall. 4.2.1.3. WM data analyses WM task performance was measured as percentage of correct responses for each task condition. To examine effects of cognitive load and delay duration on task performance, these measures were entered into a 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) repeated measures ANOVA, and significant effects were further examined with polynomial contrasts. For the WM task functional connectivity analysis, a single-experiment fMRI-CPCA was carried out as described in Chapter 2 (Section 2.4). Each level of cognitive load and delay dura-tion was specified in the design matrix (i.e., G matrix), and the time bins for which a FIR basis function was specified were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,000ms), resulting in G matrices with 374 rows (scans) and 40 columns (4 con-ditions × 10 post-stimulus time bins) per task run. After the regression and PCA steps, within-subject factors of cognitive load (4 vs. 6 letters), delay length (0s vs. 4s), and time (10 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 2 (load) × 2 (delay) × 10 (time) ANOVA for each component. Significant effects of load, delay, load × de-lay, and delay × time were further examined. Post hoc analyses of load and delay were carried out using polynomial contrasts, and interactions involving time were examined using repeated measures contrasts between adjacent time bins.  88 4.2.2. WM task performance results WM task performance for the sample is listed in Table 4.2. The 2 (load) × 2 (delay) ANOVA revealed a significant main effect of load, F(1, 53) = 51.851, p < .001, ηp2 = .495; and delay, F(1, 53) = 10.319, p = .002, ηp2 = .163. The load × delay interaction was not significant (p > .10). The main effect of load was due to lower accuracy in the 6-letter condition compared with the 4-letter condition (mean correct = 93.68% and 87.00% for 4 letters and 6 letters, respec-tively). The main effect of delay was due to lower accuracy in the 4s delay condition compared with the 0s delay condition (mean correct = 91.80% and 88.89% for 0s delay and 4s delay, re-spectively). 4.2.3. WM functional connectivity results Three components were extracted from the WM task-related variance in BOLD signal. After varimax rotation, the components accounted for 16.28%, 9.56%, and 9.14% of the vari-ance, respectively. For the purpose of discussion, these components are referred to as (1) re-sponse/attention network, (2) visual attention network, and (3) DMN. 4.2.3.1. WM task, response/attention network (component 1) Response/attention network (component 1) estimated HDR plots are presented in Figure 4.1B. This network consisted of activation in bilateral SMA, dorsal anterior cingu-late/paracingulate gyrus, and insula, as well as left pre/postcentral gyri and slightly right-dominant cerebellum (see Figure 4.1A for anatomical visualization, and Table 4.3 for locations of cluster peaks). The 2 (load) × 2 (delay) × 10 (time) repeated measures ANOVA revealed sig-nificant main effects of cognitive load, F(1, 53) = 24.062, p < .001, ηp2 = .312; delay, F(1, 53) = 45.662, p < .001, ηp2 = .463; and time, F(9, 477) = 40.996, p < .001, ηp2 = .436. Significant inter- 89 actions emerged for load × time, F(9, 477) = 13.603, p < .001, ηp2 = .204; and delay × time, F(9, 477) = 121.475, p < .001, ηp2 = .696. The main effect of cognitive load was due to greater mean activation in the 6-letter condi-tion than in the 4-letter condition (mean predictor weights = 0.036 and 0.074 for 4 letters and 6 letters, respectively). The main effect of delay was due to greater mean activation in the 4s delay condition than in the 0s delay condition (mean predictor weights = 0.030 and 0.080 for 0s delay and 4s delay, respectively). The delay × time interaction was due to the staggered onsets and peaks according to delay length, which occurred late in the post-stimulus time series (Figure 4.2). 4.2.3.2. WM task, visual attention network (component 2) Visual attention network (component 2) estimated HDR plots are presented in Figure 4.3B. This network consisted of bilateral activation in the occipital poles (extending dorsally into superior lateral occipital cortex as well as ventrally into temporal-occipital fusiform cortex), SMA, precentral gyri, and thalamus (see Figure 4.3A for anatomical visualization, and Table 4.4 for locations of cluster peaks). The repeated measures ANOVA revealed significant main effects of load, F(1, 53) = 19.054, p < .001, ηp2 = .264; delay, F(1, 53) = 31.072, p < .001, ηp2 = .370; and time, F(9, 477) = 222.024, p < .001, ηp2 = .807. Significant interactions emerged for load × time, F(9, 477) = 23.689, p < .001, ηp2 = .309; and delay × time, F(9, 477) = 33.278, p < .001, ηp2 = .386.  The main effect of load was due to greater mean activation in the 6-letter condition than in the 4-letter condition (mean predictor weights = 0.066 and 0.091 for 4 letters and 6 letters, re-spectively). The main effect of delay was due to greater mean activation in the 0s delay condition than in the 4s delay condition (mean predictor weights = 0.092 and 0.064 for 0s delay and 4s de- 90 lay, respectively). The delay × time interaction was due to a slightly more sustained HDR in the 0s delay condition than in the 4s delay condition, although the timing of onsets and peaks did not differ between conditions (Figure 4.4). 4.2.3.3. WM task, DMN (component 3) DMN (component 3) estimated HDR plots are presented in Figure 4.5B. This network consisted primarily of deactivation in bilateral precuneus, posterior cingulate gyri, superior frontal gyri, frontal poles, superior lateral occipital cortex, middle temporal gyri, and cerebellar crus II, as well as small clusters of activation in the left precentral gyrus/SMA (see Figure 4.5A for anatomical visualization, and Table 4.5 for locations of cluster peaks). The repeated measures ANOVA revealed significant main effects of load, F(1, 53) = 31.711, p < .001, ηp2 = .374; delay, F(1, 53) = 21.821, p < .001, ηp2 = .292; and time, F(9, 477) = 48.699, p < .001, ηp2 = .479. Signifi-cant interactions emerged for load × delay, F(1, 53) = 5.509, p = .023, ηp2 = .094; load × time, F(9, 477) = 7.627, p < .001, ηp2 = .126; delay × time, F(9, 477) = 22.062, p < .001, ηp2 = .294; and load × delay × time, F(4.43, 234.83) = 3.162, p = .012, ηp2 = .056.  The main effect of load was due to greater mean deactivation in the 6-letter condition than in the 4-letter condition (mean predictor weights = 0.111 and 0.164 for 4 letters and 6 let-ters, respectively). The main effect of delay was due to greater mean deactivation in the 4s delay condition than in the 0s delay condition (mean predictor weights = 0.120 and 0.155 for 0s delay and 4s delay, respectively). The load × delay interaction was due to this delay effect being more pronounced in the 6-letter condition than in the 4-letter condition (Figure 4.6A). The delay × time interaction was due to the staggered peaks corresponding to delay length, although the early onsets and initial HDR decreases were similar across delay conditions (Figure 4.6B).  91 4.2.4. Summary of WM task results Unsurprisingly, WM task accuracy was higher in the 4-letter condition than in the 6-letter condition, and when no delay occurred between the encoding and recall phases; however, no load × delay interaction emerged. As in Chapter 3, three components emerged from the fMRI data. The response/attention network (component 1; Figure 4.1), replicating the re-sponse/attention network from the WM task analysis in Chapter 3 (see component 1, Figure 3.1), mainly comprised left-lateralized somatomotor regions as well as widespread activation in dorsal medial prefrontal cortex. Although the late, staggered onsets and the left-lateralized somatomotor activity suggest that this network underlies right-handed button-press responses, it is also im-portant to note the substantial effect of cognitive load, with greater activity in the 6-letter condi-tion than in the 4-letter condition. As discussed in Chapter 3, it is possible that this network reflects a blurring of maintenance and response processes. The visual attention network (compo-nent 2; Figure 4.3) comprised widespread activation in visual cortex as well as SMA and precen-tral gyri, which peaked early and was slightly more sustained in the 0s delay condition (likely due to the immediate appearance of the probe after the encoding phase); this replicates the visual attention network that emerged in the previous WM task analysis (see component 3, Figure 3.3) and in the multi-experiment WM-TGT analysis (see component 2, Figure 3.10) reported in Chapter 3. Finally, the DMN (component 3; Figure 4.5) exhibited early onsets but staggered peaks according to delay length; this replicates the DMN from the previous WM analysis (see component 2, Figure 3.2) and the multi-experiment WM-TGT analysis (see component 4, Figure 3.14). The main effect of delay appeared to be due to a more sustained HDR in the 4s delay con-dition rather than a greater magnitude of deactivation per se.  92 4.3. Analysis 2: Spatial Capacity (SCAP) Task 4.3.1. Methods 4.3.1.1. Participants The data for the SCAP fMRI task were obtained from the OpenfMRI database (https://openfmri.org/dataset/ds000030/; accession number is ds000030). Complete details re-garding this project are provided in Poldrack et al. (2016). Participants included in the present analysis consisted of 119 healthy adults who met all general inclusion criteria detailed in Section 2.1. Task performance on the fMRI SCAP task was examined to confirm participant engagement during the task, and any participants who achieved < 60% correct responses were excluded from the analysis (note that participants had a 50% chance of guessing correctly on any given trial). Demographic information is presented in Table 4.1, including age, gender distribution, handed-ness as measured with the Edinburgh Handedness Inventory (Oldfield, 1971), and educational achievement. 4.3.1.2. SCAP task design Details regarding SCAP task design and procedures are provided in Chapter 2 (task pro-tocol described in Section 2.2.2 and Figure 2.2; fMRI data acquisition and preprocessing de-scribed in Section 2.3 and Table 2.1). In summary, a target array of either 1, 3, 5, or 7 yellow dots, positioned pseudo-randomly around a central fixation, was presented during a 2-second en-coding period. After a 1.5s, 3.0s, or 4.5s delay, a single green dot (probe) was displayed for 3 seconds. Participants were asked to respond with a button-press as to whether the probe dot was in the same position as one of the target dots. 4 load levels × 3 delay durations resulted in 12 task conditions.  93 4.3.1.3. SCAP data analyses SCAP task performance was measured as percentage of correct responses for each condi-tion. To examine effects of cognitive load and delay duration on task performance, these accura-cy measures were entered into a 4 (cognitive load; 1, 3, 5, or 7 dots) × 3 (delay duration; 1.5s, 3.0s, or 4.5s) repeated measures ANOVA. As the load and delay factors comprised >2 levels, post-hoc polynomial analyses of significant main effects involved linear, quadratic, and (for load only) cubic contrasts, which allowed for the observation of ceiling and/or floor effects in manip-ulations of task difficulty. For the SCAP task functional connectivity analysis, a single-experiment fMRI-CPCA was carried out as described in Chapter 2 (Section 2.4). Each level of cognitive load and delay duration was specified in the design, and the time bins for which a FIR basis function was speci-fied were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,000ms), resulting in G matrices with 291 rows (scans) and 120 columns (12 conditions × 10 post-stimulus time bins) per participant. After the regression and PCA steps, within-subject factors of cognitive load (1, 3, 5, or 7 dots), delay length (1.5s, 3.0s, or 4.5s), and time (10 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 4 (load) × 3 (delay) × 10 (time) ANOVA for each component. Significant effects of load, delay, and de-lay × time were further examined. Post hoc analyses of load and delay effects were carried out using polynomial contrasts (as in the analysis of task performance), and interactions involving time were examined using repeated measures contrasts between adjacent time bins. 4.3.2. SCAP task performance results SCAP task performance for the sample is listed in Table 4.2. The 4 (load) × 3 (delay) ANOVA revealed significant main effects of load, F(3, 354) = 52.178, p < .001, ηp2 = .307; and  94 delay, F(2, 236) = 7.324, p = .001, ηp2 = .058. The load × delay interaction was also significant, F(5.15, 607.33) = 3.889, p = .002, ηp2 = .032. The main effect of load was due to significant linear and cubic contrasts, with the linear contrast having the greater effect size (linear: F(1, 118) = 145.055, p < .001, ηp2 = .551; cubic: F(1, 118) = 7.893, p = .006, ηp2 = .063; quadratic p > .10). This was due to a general tendency for lower accuracy with higher cognitive load (mean correct = 92.65%, 84.03%, 82.63%, and 76.89% for 1, 3, 5, and 7 dots, respectively; see Figure 4.7A). Paired t-tests comparing adjacent load conditions revealed significant differences between 1 versus 3 dots (mean difference = 8.613, SEM = 1.181, t(118) = 7.295, p < .001) and 5 versus 7 dots (mean difference = 5.742, SEM = 1.296, t(118) = 4.432, p < .001), but not 3 versus 5 dots (p > .25). The main effect of de-lay was due to a significant linear contrast (linear: F(1, 118) = 12.019, p = .001, ηp2 = .092; quad-ratic: p > .30). This was due to a general trend for decreasing accuracy with longer delay duration (mean correct = 86.34%, 83.51%, and 82.30% for 1.5s, 3.0s, and 4.5s delay, respective-ly; see Figure 4.7B). Paired t-tests comparing adjacent delay conditions revealed a significant difference between 1.5s versus 3.0s delay (mean difference = 2.836, SEM = 1.118, t(118) = 2.536, p = .013), but no difference between 3.0s versus 4.5s delay (p > .20). 4.3.3. SCAP functional connectivity results Two components were extracted from the SCAP task-related variance in BOLD signal. After varimax rotation, the components accounted for 14.64% and 11.41% of the variance, re-spectively. For the purpose of discussion, these components are referred to as (1) external atten-tion network and (2) DMN.  95 4.3.3.1. SCAP task, external attention network (component 1) External attention network (component 1) estimated HDR plots are presented in Figure 4.8B. This network was primarily comprised of bilateral activation in the SMA/paracingulate gyri, precentral gyri, superior parietal lobules, superior and inferior lateral occipital cortex, and temporal-occipital fusiform cortex, as well as small clusters in anterior insula and thalamus (see Figure 4.8A for anatomical visualization and Table 4.6 for locations of cluster peaks). The 4 (load) × 3 (delay) × 10 (time) repeated measures ANOVA revealed main effects of load, F(3, 354) = 24.016, p < .001, ηp2 = .169; delay, F(2, 236) = 19.956, p < .001, ηp2 = .145; and time, F(9, 1062) = 203.944, p < .001, ηp2 = .633. Significant interactions included load × delay, F(6, 708) = 5.254, p < .001, ηp2 = .043; load × time, F(27, 3186) = 11.973, p < .001, ηp2 = .092; delay × time, F(18, 2124) = 54.033, p < .001, ηp2 = .314; and load × delay × time, F(54, 6372) = 6.326, p < .001, ηp2 = .051.  Post-hoc analyses of the main effect of load revealed that both the linear and quadratic contrasts were significant, with the linear contrast having the greater effect size (linear: F(1, 118) = 43.570, p < .001, ηp2 = .270 ; quadratic: F(1, 118) = 3.938, p = .050, ηp2 = .032; cubic: p > .45). This was due to a tendency for greater mean activation with greater cognitive load (mean predictor weights = 0.034, 0.047, 0.060, and 0.064 for 1, 3, 5, and 7 dots, respectively; see Figure 4.9A). Paired t-tests between adjacent load conditions revealed that only the contrasts be-tween 1 versus 3 dots and between 3 versus 5 dots were significant (1 vs. 3 dots: mean differ-ence = 0.013, SEM = 0.004, t(118) = 3.608, p < .001; 3 vs. 5 dots: mean difference = 0.012, SEM = 0.004, t(118) = 3.425, p = .001; other p > .15). Post-hoc analyses of the main effect of delay revealed that only the quadratic contrast was significant (F(1, 118) = 37.304, p < .001, ηp2 =  96 .240; linear p > .10). This was due to a tendency for greater mean activation with longer delay durations (mean predictor weights = 0.044, 0.049, and 0.062 for 1.5s, 3.0s, and 4.5s delay, re-spectively; see Figure 4.9B). Paired t-tests between adjacent delay conditions revealed that only the contrast between 3.0s delay versus 4.5s delay was significant (mean difference = 0.013, SEM = 0.003, t(118) = 4.083, p < .001; 1.5s vs. 3.0s p > .05). The delay × time interaction was due in part to HDR peaks occurring at later time bins with longer delay durations (Figure 4.9C). Further, the 3.0s and 4.5s delay conditions exhibited multiple peaks; the first peak occurred at 8 seconds for both delay conditions, and was followed by a later peak at 12 seconds in the 3.0s de-lay condition and at 14 seconds in the 4.5s delay condition. The 1.5s delay condition did not ex-hibit this pattern, possibly because the 1.5s delay was too brief for the HDR to exhibit a detectable decrease before the probe was displayed (see Figure 4.9C). 4.3.3.2. SCAP task, DMN (component 2) DMN (component 2) estimated HDR plots are presented in Figure 4.10B. This network was primarily comprised of bilateral deactivation in regions of the DMN, including frontal poles/paracingulate gyri, superior frontal gyri, inferior frontal gyri (pars triangularis), frontal or-bital cortex, posterior cingulate gyri, precuneus, superior lateral occipital cortex, angular gyri, planum temporale, middle temporal gyri, parahippocampal gyri, and cerebellar crus I (see Figure 4.10A for anatomical visualization and Table 4.7 for locations of cluster peaks). The repeated measures ANOVA revealed significant main effects of load, F(3, 354) = 22.016, p < .001, ηp2 = .157; delay, F(2, 236) = 15.859, p < .001, ηp2 = .118; and time, F(9, 1062) = 128.499, p < .001, ηp2 = .521. Significant interactions included load × delay, F(5.44, 641.42) = 2.812, p = .013, ηp2 = .023; load × time, F(27, 3186) = 12.578, p < .001, ηp2 = .096; delay × time, F(18, 2124) = 15.272, p < .001, ηp2 = .115; and load × delay × time, F(54, 6372) = 3.529, p < .001, ηp2 = .029.  97 Post-hoc analyses of the main effect of load revealed that the linear and quadratic con-trasts were both significant, with the linear contrast having the greater effect size (linear: F(1, 118) = 49.137, p < .001, ηp2 = .294; quadratic: F(1,118) = 10.331, p = .002, ηp2 = .081; cubic: p > .35 ). This was due to a general tendency for greater mean deactivation with greater cogni-tive load (mean predictor weights = 0.030, 0.050, 0.056, and 0.061 for 1, 3, 5, and 7 dots, respec-tively; see Figure 4.11A). Paired t-tests between adjacent load conditions revealed that only the contrast between 1 versus 3 dots was significant (mean difference = 0.020, SEM = 0.004, t(118) = 5.168, p < .001; other ps > .15). Post-hoc analyses of the main effect of delay revealed that both the linear and quadratic contrasts were significant, with the linear contrast having the greater effect size (linear: F(1, 118) = 26.901, p < .001, ηp2 = .186; quadratic: F(1, 118) = 4.395, p = .038, ηp2 = .036). This was due to a general tendency for greater mean deactivation with long-er delay durations (mean predictor weights = 0.040, 0.053, and 0.055 for 1.5s, 3.0s, and 4.5s de-lay conditions, respectively; see Figure 4.11B). The delay × time interaction was due to more sustained HDRs with longer delay durations, and a later peak in the 4.5s delay condition com-pared with the 1.5s and 3.0s delay conditions (see Figure 4.11C). 4.3.4. Summary of SCAP task results As in the verbal WM task, response accuracy was higher in low load conditions and in shorter delay conditions. Two components emerged from the fMRI data, which resembled pat-terns that would be expected for an external attention network (component 1; Figure 4.8) and de-activation of the DMN (component 2; Figure 4.10). Both of these networks were load-dependent (i.e., showing a greater degree of (de)activation with greater cognitive load), and both showed longer HDRs with longer trial durations. However, the external attention network appeared to respond separately to the target array and to the probe, as suggested by the presence of two peaks  98 (except in the 1.5s delay condition, which may be too brief for the HDR to exhibit a detectable decline). The decreases that emerged in the external attention network midway through the post-stimulus time series suggest that this network is unlikely to support WM maintenance. Although the DMN onsets and initial HDR decreases were similar across delay conditions, this deactiva-tion did not exhibit multiple peaks within the post-stimulus time series despite being more sus-tained with longer trial durations. Neither network showed a greater magnitude of peak (de)activation with longer delay durations, suggesting that processes occurring after the encoding phase (e.g., maintenance and recognition) do not rely on further engagement of these networks over and above their initial responses.  4.4. Analysis 3: Task-Switch Inertia (TSI) Task 4.4.1. Methods 4.4.1.1. Participants 27 healthy participants who met all general inclusion criteria detailed in Chapter 2 (Sec-tion 2.1) were included in the present analysis. Task performance on the TSI task was examined to confirm participant engagement during the task, and any participants who achieved < 30% correct responses were excluded from the analysis (note that participants had a 25% chance of guessing correctly on any given trial). Demographic information for this sample is provided in Table 4.1. A brief questionnaire was administered to all participants to obtain their age, highest level of education, and any history of head injury, neurological conditions, drug use, and medica-tion. In addition, visual acuity and color vision were assessed to ensure that participants were able to view the task on an fMRI presentation screen. Handedness was measured with the AHPQ (Annett, 1970), and IQ was estimated with the Quick IQ test (Mortimer & Bowen, 1999). Partic-ipants were recruited via posters on the UBC campus, community bulletin boards, and on elec- 99 tronic bulletins such as Craigslist. Participants provided informed consent at the start of their testing sessions, and were compensated $10 per hour for time spent participating in the study plus $10 for travel expenses. Participants were also given a copy of their T1 MRI scan on a disc.  4.4.1.2. TSI task design Details regarding TSI task design and procedures are provided in Chapter 2 (task protocol described in Section 2.2.3 and Figure 2.3; fMRI data acquisition and preprocessing described in Section 2.3 and Table 2.1). In summary, the TSI task is a set-switching Stroop task that involves responding to Stroop stimuli in alternating blocks of colour-naming and word-reading of neutral and incongruent stimuli. A neutral word-reading stimulus is one in which a colour word (“GREEN”, “RED”, “YELLOW”, or “BLUE”) is displayed in white font against a black back-ground, and a neutral colour-naming stimulus is one in which a string of Xs is displayed in green, red, yellow, or blue font. Incongruent stimuli are colour words (“GREEN”, “RED”, “YELLOW”, or “BLUE”) displayed in incongruent green, red, yellow, or blue font. In this ver-sion of the TSI task, each block consisted of 10 trials. Word-reading blocks consisted of an equal mixture of neutral and incongruent stimuli, whereas colour-naming blocks consisted of either 10 neutral or 10 incongruent stimuli. Therefore, word-reading trials may be further categorized as “colour-neutral task-switch” (where the preceding block of trials consisted of neutral colour-naming stimuli) and “colour-incongruent task-switch” (where the preceding block of trials con-sisted of incongruent colour-naming stimuli). Therefore, 4 word-reading conditions, each con-sisting of 15 trials, were examined in the present study: (1) neutral stimulus following a neutral colour-naming block (cn-WN), (2) neutral stimulus following an incongruent colour-naming block (ci-WN), (3) incongruent stimulus following a neutral colour-naming block (cn-WI), and  100 (4) incongruent stimulus following an incongruent colour-naming block (ci-WI). Colour-naming trials were not analysed in the present study. 4.4.1.3. TSI data analyses TSI task performance was measured as either percentage of correct responses or mean re-action time (RT) for each condition. To examine effects of task condition, separate analyses were carried out on percent correct and RT measures, both of which were examined in a 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; preceding block = neutral vs. incongruent colour-naming) repeated measures ANOVA on word-reading trials only. Post-hoc analyses were carried out for significant effects of congruency, task-switch, and congruency × task-switch using polynomial contrasts. For the TSI task functional connectivity analysis, a single-experiment fMRI-CPCA was carried out as described in Chapter 2 (Section 2.4). Each level of stimulus congruency and task-switch condition was specified in the design (but only for word-reading trials), and the time bins for which a FIR basis function was specified were scans 1-8 following trial onset (i.e., 16 sec-onds of post-stimulus time with TR = 2,000ms), resulting in G matrices with 330 rows (scans) and 32 columns (4 conditions × 8 post-stimulus time bins) per participant. After the regression and PCA steps, within-subject factors of stimulus congruency (neutral vs. incongruent stimulus), task-switch condition (preceding block = neutral vs. incongruent colour-naming), and time (8 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 2 (congruency) × 2 (task-switch) × 8 (time) repeated measures ANOVA for each component. Post-hoc analyses of significant effects of congruency, task-switch, congruency × task-switch, congruency × time, and task-switch × time were carried out, with effects of congruency and task- 101 switch examined using polynomial contrasts, and interactions involving time examined using repeated measures contrasts between adjacent time bins. 4.4.2. TSI task performance results TSI task performance (i.e., percent correct and mean RT for each task condition) for the sample is listed in Table 4.2. Accuracy and RTs were analysed separately as reported below. 4.4.2.1. Response accuracy Examining percentage of correct responses, the 2 (stimulus congruency) × 2 (task-switch condition) ANOVA revealed a significant main effect of congruency, F(1, 26) = 48.098, p < .001, ηp2 = .649. Neither the main effect of task-switch nor the congruency × task-switch interac-tion were significant (both ps > .25). The main effect of congruency was due to greater mean ac-curacy in the neutral stimulus condition than in the incongruent condition (mean correct = 97.04% and 86.42% for neutral and incongruent stimuli, respectively). 4.4.2.2. Reaction time (RT) Examining mean RTs, the 2 (congruency) × 2 (task-switch) ANOVA revealed significant main effects of congruency, F(1, 26) = 154.452, p < .001, ηp2 = .856; and task-switch, F(1, 26) = 11.649, p = .002, ηp2 = .309. The congruency × task-switch interaction was not significant (p > .10). The main effect of congruency was due to longer RTs in the incongruent stimulus condi-tion than in the neutral condition (mean RTs = 913.41ms and 1165.20ms for neutral and incon-gruent stimuli, respectively). The main effect of task-switch was due to longer RTs following an incongruent colour-naming block than following a neutral colour-naming block (mean RTs = 1014.57ms and 1064.04ms following neutral and incongruent colour-naming blocks, respective- 102 ly). The absence of a significant congruency × task-switch interaction suggests that the task-switch effect on RTs was the same for neutral as for incongruent word-reading trials. 4.4.3. TSI functional connectivity results Three components were extracted from the TSI task-related variance in BOLD signal. After varimax rotation, the components accounted for 12.44%, 12.05%, and 7.92% of the vari-ance, respectively. For the purpose of discussion, these components are referred to as (1) DMN/occipital network, (2) response network, and (3) evaluation network. 4.4.3.1. TSI task, DMN/occipital network (component 1) DMN/occipital network (component 1) estimated HDR plots are presented in Figure 4.12B. This network was comprised of bilateral deactivation in regions of the DMN (including frontal poles, superior frontal gyri, precuneus, and posterior cingulate cortex) as well as broad occipital deactivation originating in the medial occipital lobes (including lingual gyri, intracal-carine cortex, occipital fusiform cortex, occipital poles, and superior lateral occipital cortex; see Figure 4.12A for anatomical visualization and Table 4.8 for locations of cluster peaks). The 2 (congruency) × 2 (task-switch) × 8 (time) repeated measures ANOVA revealed main effects of stimulus congruency, F(1, 26) = 66.643, p < .001, ηp2 = .719; and time, F(7, 182) = 39.834, p < .001, ηp2 = .605. Significant interactions emerged for congruency × task-switch, F(1, 26) = 23.297, p < .001, ηp2 = .473; congruency × time, F(7, 182) = 43.209, p < .001, ηp2 = .624; task-switch × time, F(3.09, 80.42) = 6.219, p = .001, ηp2 = .193; and congruency × task-switch × time, F(7, 182) = 6.936, p < .001, ηp2 = .211. The main effect of stimulus congruency was due to greater mean deactivation in the in-congruent condition compared with the neutral condition (mean predictor weights = -0.026 and 0.132 for neutral and incongruent stimuli, respectively). This was driven by the combination of  103 more extreme peak deactivation in the incongruent condition and a much greater post-HDR rise above baseline in the neutral condition (Figure 4.13A). Peak deactivation was also later in the incongruent colour-naming than the neutral colour-naming task-switch condition (Figure 4.13B). The congruency × task-switch interaction appeared to be due to a more pronounced effect of stimulus congruency following neutral colour-naming (Figure 4.13C). 4.4.3.2. TSI task, response network (component 2) Response network (component 2) estimated HDR plots are presented in Figure 4.14B. This network was primarily comprised of bilateral activation in somatomotor regions including SMA, pre/postcentral gyri, and cerebellar regions, as well as bilateral superior parietal lobules, superior and inferior lateral occipital cortex, central opercular cortex, thalamus, and small clus-ters in the insula (see Figure 4.14A for anatomical visualization and Table 4.9 for locations of cluster peaks). The repeated measures ANOVA revealed significant main effects of stimulus congruency, F(1, 26) = 11.399, p = .002, ηp2 = .305; and time, F(7, 182) = 92.386, p < .001, ηp2 = .780. Significant interactions emerged for congruency × time, F(7, 182) = 6.525, p < .001, ηp2 = .201; and congruency × task-switch × time, F(7, 182) = 6.460, p < .001, ηp2 = .199. The main effect of stimulus congruency was due to greater mean activation in the neutral stimulus condition than in the incongruent stimulus condition (mean predictor weights = 0.024 and -0.047 for neutral and incongruent stimuli, respectively). In addition, there was greater post-peak suppression in the incongruent stimulus condition than in the neutral stimulus condition (Figure 4.15). 4.4.3.3. TSI task, evaluation network (component 3) Evaluation network (component 3) estimated HDR plots are presented in Figure 4.16B. This network was comprised primarily of bilateral activation in superior and middle frontal gyri,  104 frontal poles, frontal orbital cortex, inferior frontal gyri (pars opercularis), anterior insula, supe-rior lateral occipital cortex, angular gyri, temporooccipital/posterior middle and inferior temporal gyri, and cerebellar crus I and II (see Figure 4.16A for anatomical visualization and Table 4.10 for locations of cluster peaks). The repeated measures ANOVA revealed significant main effects of congruency, F(1, 26) = 8.773, p = .006, ηp2 = .252; and time, F(2.88, 74.98) = 2.939, p = .041, ηp2 = .102. A significant interaction emerged for congruency × time, F(7, 182) = 7.314, p < .001, ηp2 = .220. The main effect of congruency was due to greater mean activation in the incongruent stimulus condition than in the neutral stimulus condition (mean predictor weights = -0.046 and 0.049 for neutral and incongruent stimuli, respectively). This was due to the HDR for the neutral stimulus condition being suppressed throughout the post-stimulus time series (Figure 4.17). 4.4.4. Summary of TSI task results While both response accuracy and RT were affected by stimulus congruency (i.e., higher accuracy and shorter RTs for neutral stimuli than for incongruent stimuli), response accuracy was not affected by task-switch condition. However, RTs were slowed following a task-switch from an incongruent colour-naming block, and this effect was the same for neutral as for incon-gruent stimuli. Three components were extracted from the TSI fMRI data. The DMN/occipital network (component 1; Figure 4.12) comprised a combination of DMN regions and widespread occipital deactivation, which was dependent on both stimulus congruency and task-switch condition. This network exhibited greater deactivation for incongruent stimuli than for neutral stimuli, and this effect was magnified following a task-switch from an incongruent colour-naming block. The re-sponse network (component 2; Figure 4.14) was dominated by bilateral activation of somatomo- 105 tor regions, such as the SMA, pre/postcentral gyri, and cerebellum. The bilateral distribution of somatomotor activity is consistent with the task design, which required both hands for button-press responses. This network peaked earlier than the other two networks, and exhibited greater activation for neutral stimuli than for incongruent stimuli. The evaluation network (component 3; Figure 4.16) was comprised of frontoparietal connectivity, with peaks in more anterior prefrontal cortex and more posterior parietal regions as compared to the response network. This evaluation network was only engaged in the incongruent stimulus condition, which could suggest that it un-derlies an externally-oriented salience or stimulus conflict detection process. However, this seems unlikely given its later onsets and peaks compared with those of the response network, as a bottom-up attention network would be expected to initiate at the trial onset. Therefore, a more plausible explanation for the effect of stimulus congruency is that this network subserves a pro-cess akin to reflecting on whether the response matched the current task rule (i.e., focus on the word rather than the font colour), which is ambiguous for incongruent but not neutral stimuli. Although no significant task-switch effect emerged in the evaluation network, it is possible that this may be due to insufficient statistical power (suggested by the large standard errors around the predictor weights), and it should be noted that the HDR shapes suggest a trend for greater engagement following an incongruent colour-naming block (Figure 4.16B). 4.5. Analysis 4: Thought Generation Task (TGT) 4.5.1. Methods 4.5.1.1. Participants 32 healthy participants who met all general inclusion criteria detailed in Chapter 2 (Sec-tion 2.1) were included in the present analysis. Demographic information for this sample is pro-vided in Table 4.1. A brief questionnaire was administered to all participants to obtain their age,  106 highest level of education, and any history of head injury, neurological conditions, drug use, and medication. In addition, visual acuity and color vision were assessed to ensure that participants were able to view the task on an fMRI presentation screen. Handedness was measured with the AHPQ (Annett, 1970), and IQ was estimated with the Quick IQ test (Mortimer & Bowen, 1999). Participants were recruited via posters on the UBC campus, community bulletin boards, and on electronic bulletins such as Craigslist. Participants provided informed consent at the start of their testing sessions, and were compensated $10 per hour for time spent participating in the study plus $10 for travel expenses. Participants were also given a copy of their T1 MRI scan on a disc. 4.5.1.2. TGT task design Details regarding the TGT task design and procedures are provided in Chapter 2 (task protocol described in Section 2.2.4 and Figure 2.4; fMRI data acquisition and preprocessing de-scribed in Section 2.3 and Table 2.1). In summary, participants were presented with an object noun and its corresponding image (e.g., pillow) for five seconds and instructed to either mentally generate or listen to a function of the noun (e.g., “Something you rest your head on when sleep-ing”). The two experimental conditions (i.e., generating and hearing) were presented in alternat-ing blocks of 15 trials (30 trials total for each condition across two runs). 4.5.1.3. TGT data analysis No overt behavioural responses are recorded in the TGT task, and so only functional connectivity was examined. A single-experiment fMRI-CPCA was carried out for the TGT task as described in Chapter 2 (Section 2.4). The generating and hearing conditions were both speci-fied in the design, and the time bins for which a FIR basis function was specified were scans 1-8 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,500ms). This resulted in G matrices with 176 rows (scans) and 16 columns (2 conditions × 8 post-stimulus time bins) per  107 task run. After the regression and PCA steps, within-subject factors of condition (generating vs. hearing) and time (8 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 2 (condition) × 8 (time) repeated measures ANOVA for each component. Post-hoc analyses of significant effects of condition and condition × time were carried out, with the interaction involving time examined using repeated measures contrasts between adjacent time bins. 4.5.2. TGT functional connectivity results Four components were extracted from the TGT task-related variance in BOLD signal. After varimax rotation, the components accounted for 14.52%, 9.49%, 5.80%, and 5.33% of the variance, respectively. For the purpose of discussion, these components are referred to as (1) language network, (2) posterior-medial DMN, (3) anterior/posterior-lateral DMN, and (4) auditory network. 4.5.2.1. TGT task, language network (component 1) Language network (component 1) estimated HDR plots are presented in Figure 4.18B. This network consisted of left-lateralized activation in posterior middle frontal gyrus, inferior frontal gyrus (pars opercularis and pars triangularis), frontal orbital cortex, and superior lateral occipital cortex, as well as bilateral (but slightly left-dominant) posterior superior temporal gyri, bilateral SMA/superior frontal gyrus, inferior lateral occipital cortex, occipital poles, and occipi-tal fusiform cortex (see Figure 4.18A for anatomical visualization and Table 4.11 for locations of cluster peaks). The 2 (condition) × 8 (time) repeated measures ANOVA revealed significant main effects of condition, F(1, 31) = 9.590, p = .004, ηp2 = .236; and time, F(7, 217) = 85.500, p < .001, ηp2 = .734. A significant interaction emerged for condition × time, F(2.44, 75.51) = 6.474, p = .001, ηp2 = .173.  108 The main effect of condition was due to greater mean activation in the generating condi-tion than in the hearing condition (mean predictor weights = 0.129 and 0.100 for generating and hearing, respectively). The HDR shape for the generating condition also exhibited a steeper in-crease to peak and steeper post-peak decline as compared to the shape of the hearing condition HDR (Figure 4.18B). 4.5.2.2. TGT task, posterior-medial DMN (component 2) Posterior-medial DMN (component 2) estimated HDR plots are presented in Figure 4.19B. This network primarily consisted of deactivation of posterior medial regions of the DMN including posterior cingulate gyrus, precuneus cortex, and cuneal cortex, as well as anterior cin-gulate cortex, SMA, pre/postcentral gyri, superior frontal gyri, superior and inferior lateral occip-ital cortex, anterior supramarginal gyri, and lingual gyri (see Figure 4.19A for anatomical visualization and Table 4.12 for locations of cluster peaks). The repeated measures ANOVA re-vealed a significant main effect of time, F(7, 217) = 9.831, p < .001, ηp2 = .241. Neither the main effect of task condition nor the condition × time interaction were statistically significant (both ps > .20). 4.5.2.3. TGT task, anterior/posterior-lateral network (component 3) Anterior/posterior-lateral DMN (component 3) estimated HDR plots are presented in Figure 4.20B. This network primarily consisted of bilateral deactivation in anterior and lateral posterior regions of the DMN including frontal poles, superior and middle frontal gyri, paracingulate gyri, superior lateral occipital cortex, and angular gyri, as well as middle temporal gyri and bilateral regions of the cerebellum including crus I and II. Small clusters of activation emerged as well in occipital fusiform cortex (see Figure 4.20A for anatomical visualizations and Table 4.13 for locations of cluster peaks). The repeated measures ANOVA revealed a significant  109 main effect of time, F(7, 217) = 19.615, p < .001, ηp2 = .388. Neither the main effect of task con-dition nor the condition × time interaction were statistically significant (both ps > .60). 4.5.2.4. TGT task, auditory network (component 4) Auditory network (component 4) estimated HDR plots are presented in Figure 4.21B. This network consisted of bilateral activation in primary auditory cortex and surrounding re-gions, including anterior superior temporal gyrus, temporal fusiform cortex, temporal-occipital fusiform cortex, inferior frontal gyrus (pars triangularis and pars opercularis), precentral gyrus, and thalamus, as well as small clusters of deactivation in inferior lateral occipital cortex (see Figure 4.21A for anatomical visualization and Table 4.14 for locations of cluster peaks). The re-peated measures ANOVA revealed significant main effects of condition, F(1, 31) = 107.394, p < .001, ηp2 = .776; and time, F(7, 217) = 23.187, p < .001, ηp2 = .428. A significant interaction emerged for condition × time, F(7, 217) = 83.205, p < .001, ηp2 = .729.  The main effect of condition was due to greater mean activation in the hearing condition than in the generating condition (mean predictor weights = -0.041 and 0.050 for generating and hearing conditions, respectively). This network was only engaged during the hearing condition, with onsets initiating at the 5-second post-stimulus time bin, and peaking at around 10 seconds (see Figure 4.21B). 4.5.3. Summary of TGT task results Four components were extracted from the TGT data. The language network (component 1; Figure 4.18) comprised frontotemporal connectivity and widespread activity in occipital cor-tex. Several regions of this network were left-lateralized, including inferior frontal gyrus (pars triangularis), middle frontal gyrus, and superior lateral occipital cortex/superior parietal lobule, which, along with the activation in posterior superior temporal gyri, is consistent with known  110 language-related patterns of activity (i.e., Broca’s area in the inferior frontal gyrus and Wernicke’s area in posterior superior temporal gyrus; Flinker et al., 2015). This network was the only one that was more engaged in the generating condition than in the hearing condition. The posterior-medial DMN (component 2; Figure 4.19) consisted primarily of deactivation in poste-rior medial regions of the DMN – especially the precuneus – as well as in pre/postcentral gyri, and dorsal medial prefrontal regions including SMA and dorsal anterior cingulate. By contrast, the anterior/posterior-lateral DMN (component 3; Figure 4.20) consisted of deactivation of ante-rior portions of the DMN (frontal poles, superior frontal gyri) and the lateral posterior areas (su-perior lateral occipital cortex, temporooccipital part of the middle temporal gyri), and it peaked earlier than the posterior-medial DMN. Neither of these task-negative networks exhibited differ-ences between task conditions. Finally, the auditory network (component 4; Figure 4.21) reflect-ed activation of primary auditory cortex in response to the stimulus presented in the hearing condition.  4.6. Discussion of Single Experiment Analyses Four single-experiment fMRI-CPCAs were carried out in the present study, consisting of verbal WM (WM task), visuospatial WM (SCAP task), Stroop task-switching (TSI task), and internal thought generation versus speech perception (TGT). Two to four components were ex-tracted from each dataset. As expected, a commonality among all tasks was the involvement of at least one task-positive network including activation in dorsal medial prefrontal cortex and which was dependent on cognitive demand, and at least one task-negative network consisting of deacti-vation of the DMN (or particular sub-regions).  In both the WM task and the TSI task, two task-positive networks emerged that com-prised dorsal medial prefrontal cortex and parietal regions, with one of the two appearing to un- 111 derlie response processes. Therefore, it is surprising that only one task-positive network emerged in the SCAP task, given its similarity to the verbal WM task with respect to the putative process-es engaged (i.e., encoding, maintenance, and recall/response). The different component struc-tures between these WM tasks could be due to the nature of the task design and/or cognitive demand – namely, verbal versus non-verbal memory store – but it is possible that additional net-works emerge in the SCAP task when combined with other tasks, as observed in the verbal WM task in Chapter 3. The TGT functional connectivity results were more distinct in that two task-negative networks emerged, each of which was dominated by different subdivisions of the DMN (posteri-or-medial and anterior/posterior-lateral; Figures 4.19 and 4.2, respectively). It has been suggest-ed that language-related tasks are more likely to produce a decomposition of the DMN into sub-regions due to its partial overlap with language systems (Seghier, Fagan, & Price, 2010), and the impact of semantic processing may play contrasting roles on different sub-regions of the DMN (Seghier & Price, 2012). Although neither of the DMN sub-networks in the TGT task exhibited differences between task conditions, the networks did show differences in HDR shapes. Notably, the anterior/posterior-lateral DMN (component 3) was the first to peak of all four networks that emerged and exhibited a relatively brief response, while the posterior-medial DMN (component 2) peaked later and was slightly more sustained (compare Figures 4.19B and 4.20B). Also unique to the TGT task was the presence of a left-lateralized task-positive network (i.e., the language network, component 1) that was more engaged in the thought generation con-dition than in the hearing condition (Figure 4.18), which is likely due to the language production and processing demands of the task. This raises the question as to why a similar network did not emerge in the TSI task, which is also intrinsically tied to language (i.e., making a semantic  112 judgement of a word presented on-screen). A possible explanation is that when a cognitive task – such as the TSI task – engages a more prepotent response to a highly familiar word, it does not elicit language-specific activity to the same degree as higher-level sentence produc-tion/comprehension. Although meaningful insights may be gained from the individual analyses of these tasks, it is important to note the uncertainty of the degree to which any given network reflects task-specific activity rather than a more replicable pattern. Although this is not necessarily a limita-tion with respect to basic cognitive neuroscience research, in the context of schizophrenia re-search it may be important to identify networks that replicate across several cognitive tasks when the aim is to identify clinically meaningful patterns of brain activity. Therefore, the final analysis chapter involves combining these four tasks into a multi-experiment fMRI-CPCA with the over-arching goal of characterizing networks that may underlie WM deficits in schizophrenia.   113 4.7. Chapter 4 Tables Table 4.1. Demographic information for the WM task, SCAP task, TSI task, and TGT task da-tasets (from single-task analyses with healthy controls only). Standard deviations are in paren-theses.  WM (n = 54) SCAP (n = 119) TSI (n = 27) TGT (n = 32) Age 29.65 (9.41) 31.22 (8.80) 29.37 (8.63) 28.75 (8.58) Years of education 15.79 (2.02) 15.03 (1.71) 16.31 (1.86) 15.58 (1.81) Quick estimated IQ 100.06 (11.99) - 101.22 (12.34) 97.09 (11.21) Gender distribution (female/male) 36/18 64/55 20/7 13/19 Handedness (L/M/R) 5/2/47 0/0/119 1/1/25 3/1/28 Socioeconomic status factor score 59.63 (14.27) - 54.89 (12.62) 65.75 (14.97) Note. Years of education missing for 1 WM participant. IQ = intelligence quotient; L = left-handed; M = mixed handedness; R = right-handed; WM = Working Memory Task; SCAP = Spa-tial Capacity Task; TSI = Task-Switch Inertia Task; TGT = Thought Generation Task.     114 Table 4.2. Mean WM, SCAP, and TSI task performance results (percent correct and mean reac-tion time for each condition; standard deviations in parentheses). Task/Condition Percent correct RT (milliseconds)a WM task   4 letters   0s delay 94.64 (7.20) 1,035.83 (170.42) 4s delay 92.72 (7.18) 1,021.14 (157.92) 6 letters   0s delay 88.96 (9.37) 1,108.63 (153.17) 4s delay 85.05 (12.00) 1,089.91 (180.02) SCAP task   1 dot   1.5s delay 96.01 (11.27) 938.85 (226.01) 3.0s delay 92.86 (14.61) 902.74 (220.40) 4.5s delay 89.08 (19.43) 971.08 (271.09) 3 dots   1.5s delay 84.87 (23.52) 1,097.02 (271.97) 3.0s delay 80.04 (24.05) 1,143.21 (292.22) 4.5s delay 87.18 (21.06) 1,041.19 (273.98) 5 dots   1.5s delay 84.45 (19.80) 1,215.67 (297.27) 3.0s delay 84.66 (18.44) 1,129.39 (264.22) 4.5s delay 78.78 (25.34) 1,191.11 (315.12) 7 dots   1.5s delay 80.04 (20.99) 1,281.85 (337.10) 3.0s delay 76.47 (19.34) 1,212.79 (283.01) 4.5s delay 74.16 (20.82) 1,201.38 (293.21) TSI task   Colour-namingb   Neutral 95.56 (4.43) 884.21 (95.31) Incongruent 87.65 (9.99) 1,120.85 (108.54) Word-reading   Neutral (all) 97.04 (3.38) 913.48 (104.37) cn-WN 97.04 (3.85) 899.77 (103.00) ci-WN 97.04 (5.01) 927.05 (120.95) Incongruent (all) 86.42 (8.52) 1,164.92 (139.98) cn-WI 87.90 (9.79) 1,129.36 (144.58) ci-WI 84.94 (11.78) 1,201.03 (165.16) aRT was not analyzed for the WM task or SCAP task, but is presented here for interest. bColour-naming trials were not analyzed, but task performance is presented here for interest. RT = reaction time; cn = task-switch from neutral colour-naming block; ci = task-switch from incongru-ent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.  115 Table 4.3. WM task fMRI-CPCA, response/attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task response/attention network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 4,036 108,972     Precentral gyrus   4 -36 -22 59 Postcentral gyrus   3 -45 -28 53 Precentral gyrus   6 -33 -7 62 Postcentral gyrus   3 -51 -25 44 Central opercular cortex   48 -54 -19 20 Supplementary motor area   6 -3 -4 53 Superior lateral occipital cortex   7 -12 -70 53 Parietal operculum cortex   48 -57 -40 23 Cluster 1: right hemisphere       Paracingulate gyrus   32 3 17 47        Cluster 2: left hemisphere 1,341 36,207     Cerebellum VI   n/a -33 -55 -31 Cluster 2: bilateral       Lingual gyrus   17 0 -79 2 Cluster 2: right hemisphere       Cerebellum VI   n/a 21 -52 -25 Cerebellum VIIIa   n/a 18 -64 -49 Cerebellum V   n/a 6 -64 -16 Cerebellum crus II   n/a 6 -73 -34 Cerebellum V   n/a 12 -58 -28        Cluster 3: left hemisphere 971 26,217     Insular cortex   48 -30 20 5 Precentral gyrus   44 -54 8 29 Precentral gyrus   48 -51 5 11 Insular cortex   48 -39 -1 8 Frontal pole   46 -30 47 20 Middle frontal gyrus   46 -36 35 29 Inferior frontal gyrus, pars opercularis   45 -51 11 -1        (Table 4.3 continued on next page)    (Table 4.3, continued from previous page) 116 WM task response/attention network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 4: right hemisphere 529 14,283     Postcentral gyrus   48 57 -16 26 Postcentral gyrus   3 54 -19 38 Posterior supramarginal gyrus   40 39 -46 44 Superior parietal lobule   2 42 -40 53 Anterior supramarginal gyrus   2 48 -34 50        Cluster 5: right hemisphere 471 12,717     Insular cortex   47 33 23 2 Inferior frontal gyrus, pars opercularis   6 57 11 20 Inferior frontal gyrus, pars opercularis   48 54 11 8        Cluster 6: right hemisphere 227 6,129     Frontal pole   46 36 44 26        Cluster 7: right hemisphere 186 5,022     Middle frontal gyrus   6 33 -1 59        Cluster 8: left hemisphere 67 1,809     Thalamus   n/a -12 -19 8        Cluster 9: left hemisphere 58 1,566     Inferior lateral occipital cortex   37 -51 -67 2 Middle temporal gyrus, temporooccipital part   21 -48 -52 8        Cluster 10: right hemisphere 33 891     Thalamus   n/a 12 -16 8 Caudate   n/a 15 -1 14        Cluster 11: left hemisphere 16 432     Cerebellum VIIb   n/a -33 -58 -49        Cluster 12: right hemisphere 11 297     Precuneus cortex   7 9 -70 50        Cluster 13: left hemisphere 7 189     Cerebellum VIIb   n/a -15 -70 -49  117 Table 4.4. WM task fMRI-CPCA, visual attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task visual attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 6,617 178,659     Occipital pole   18 -18 -91 -7 Superior lateral occipital cortex   7 -21 -64 50 Superior lateral occipital cortex   19 -24 -67 32 Cluster 1: bilateral       Vermis crus II   n/a 0 -73 -28 Cluster 1: right hemisphere       Occipital fusiform gyrus   18 21 -88 -4 Superior lateral occipital cortex   19 27 -67 35 Superior lateral occipital cortex   7 27 -61 53 Temporal occipital fusiform cortex   37 36 -46 -19 Cerebellum crus II   n/a 9 -76 -40        Cluster 2: left hemisphere 719 19,413     Precentral gyrus   6 -51 -1 47 Precentral gyrus   6 -57 2 23 Middle frontal gyrus   6 -27 -1 50        Cluster 3: left hemisphere 220 5,940     Supplementary motor area   6 -3 5 62        Cluster 4: right hemisphere 175 4,725     Precentral gyrus   6 54 -1 44        Cluster 5: right hemisphere 39 1,053     Precentral gyrus   44 42 8 26        Cluster 6: right hemisphere 36 972     Precentral gyrus   6 30 -1 47        Cluster 7: right hemisphere 34 918     Thalamus   n/a 21 -28 -1        Cluster 8: left hemisphere 32 864     Thalamus   n/a -21 -28 -4        (Table 4.4 continued on next page)    (Table 4.4, continued from previous page) 118 WM task visual attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 9: bilateral 23 621     Vermis IX   n/a 0 -58 -37        Cluster 10: right hemisphere 14 378     Precentral gyrus   6 33 -19 71        Cluster 11: right hemisphere 11 297     Cerebellum VIIb   n/a 24 -70 -52        Cluster 12: right hemisphere 4 108     Anterior supramarginal gyrus   2 42 -34 44        Cluster 13: left hemisphere 3 81     Cerebellum X   n/a -21 -40 -43        Cluster 14: left hemisphere 2 54     Superior parietal lobule   40 -39 -40 44        Negative loadings              Cluster 15: left hemisphere 17 459     Lingual gyrus   18 -9 -73 -7        Cluster 16: left hemisphere 6 162     Occipital pole   18 -9 -94 20        Cluster 17: right hemisphere 2 54     Occipital pole   18 12 -88 23  119 Table 4.5. WM task fMRI-CPCA, default mode network (DMN, component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. WM task default mode network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 19 513     Supplementary motor area   6 -3 8 56        Cluster 2: left hemisphere 5 135     Precentral gyrus   6 -54 2 44        Negative loadings              Cluster 3: left hemisphere 2,791 75,357     Superior frontal gyrus   10 -6 56 29 Superior frontal gyrus   9 -21 35 44 Cluster 3: right hemisphere       Frontal pole   10 3 56 14 Frontal pole   10 3 59 5 Superior frontal gyrus   9 24 32 47 Frontal pole   9 18 44 41 Middle frontal gyrus   46 39 20 44        Cluster 4: left hemisphere 2,565 69,255     Posterior cingulate gyrus   23 -3 -46 35 Cluster 4: bilateral       Precuneus cortex   31 0 -70 29 Posterior cingulate gyrus   23 0 -16 38        Cluster 5: left hemisphere 816 22,032     Superior lateral occipital cortex   39 -45 -73 29        Cluster 6: right hemisphere 774 20,898     Superior lateral occipital cortex   39 51 -64 23        Cluster 7: right hemisphere 422 11,394     Anterior middle temporal gyrus   21 57 -4 -16 Temporal pole   38 42 20 -28 Inferior frontal gyrus, pars triangularis   45 54 26 11 Inferior frontal gyrus, pars triangularis   45 51 29 8 Temporal pole   21 48 8 -31 (Table 4.5 continued on next page)    (Table 4.5, continued from previous page) 120 WM task default mode network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Frontal orbital cortex   47 42 32 -10 Planum temporale   48 57 -10 5        Cluster 8: left hemisphere 224 6,048     Anterior middle temporal gyrus   20 -54 -7 -16 Temporal pole   21 -48 11 -28 Frontal orbital cortex   38 -42 26 -16        Cluster 9: left hemisphere 150 4,050     Cerebellum crus II   n/a -21 -79 -37        Cluster 10: right hemisphere 115 3,105     Cerebellum crus II   n/a 27 -76 -37        Cluster 11: left hemisphere 31 837     Lingual gyrus   37 -27 -43 -10        Cluster 12: right hemisphere 18 486     Lingual gyrus   37 27 -40 -10        Cluster 13: right hemisphere 11 297     Temporal pole   38 30 5 -19        Cluster 14: right hemisphere 5 135     Cerebellum IX   n/a 9 -52 -46        Cluster 15: left hemisphere 5 135     Middle temporal gyrus, temporooccipital part   37 -63 -55 -1  121 Table 4.6. SCAP task fMRI-CPCA, external attention network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. SCAP task external attention network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 7,388 199,476     Superior lateral occipital cortex   7 -18 -64 53 Superior parietal lobule   40 -33 -46 50 Superior lateral occipital cortex   19 -27 -73 29 Middle frontal gyrus   6 -27 -4 56 Inferior lateral occipital cortex   37 -48 -67 -7 Supplementary motor area   32 -3 8 50 Precentral gyrus   44 -51 5 32 Occipital fusiform gyrus   18 -9 -73 -19 Cluster 1: right hemisphere       Superior lateral occipital cortex   7 24 -61 53 Superior lateral occipital cortex   7 21 -64 50 Superior parietal lobule   7 30 -52 53 Superior lateral occipital cortex   7 27 -70 38 Inferior temporal gyrus, temporooccipital part   37 48 -61 -10 Occipital pole   17 12 -91 11 Lingual gyrus   17 9 -88 -4        Cluster 2: right hemisphere 560 15,120     Middle frontal gyrus   6 30 -1 59 Precentral gyrus   44 51 8 29        Cluster 3: left hemisphere 24 648     Intracalcarine cortex   17 -12 -73 11        Cluster 4: right hemisphere 23 621     Insular cortex   48 33 20 5        Cluster 5: left hemisphere 12 324     Thalamus   n/a -12 -19 8  (Table 4.6 continued on next page)     (Table 4.6, continued from previous page) 122 SCAP task external attention network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 6: right hemisphere 11 297     Intracalcarine cortex   17 12 -67 11        Cluster 7: left hemisphere 5 135     Insular cortex   48 -30 20 5        Cluster 8: right hemisphere 3 81     Thalamus   n/a 12 -19 8   123 Table 4.7. SCAP task fMRI-CPCA, default mode network (DMN, component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. SCAP task default mode network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Negative loadings              Cluster 1: left hemisphere 3,431 92,637     Frontal pole   10 -18 59 23 Superior frontal gyrus   8 -3 35 56 Middle frontal gyrus   9 -36 20 50 Cluster 1: bilateral       Paracingulate gyrus   32 0 50 14 Frontal pole   9 0 62 17 Cluster 1: right hemisphere       Frontal pole   9 18 47 41 Frontal pole   9 15 53 38 Superior frontal gyrus   8 6 41 53 Frontal pole   10 18 59 26 Middle frontal gyrus   9 42 20 47        Cluster 2: left hemisphere 1,255 33,885     Precuneus cortex   30 -6 -55 11 Cluster 2: bilateral       Posterior cingulate gyrus   31 0 -46 35 Posterior cingulate gyrus   23 0 -22 41 Precuneus cortex   31 0 -67 32        Cluster 3: right hemisphere 876 23,652     Angular gyrus   39 51 -58 32 Posterior middle temporal gyrus   21 63 -22 -10 Posterior middle temporal gyrus   21 66 -28 -4        Cluster 4: left hemisphere 519 14,013     Superior lateral occipital cortex   39 -48 -64 32        Cluster 5: left hemisphere 504 13,608     Posterior middle temporal gyrus   21 -60 -22 -13 Posterior middle temporal gyrus   21 -63 -43 -1 (Table 4.7 continued on next page)    (Table 4.7, continued from previous page) 124 SCAP task default mode network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Planum temporale   22 -57 -25 11 Planum temporale   42 -60 -28 14 Central opercular cortex   48 -57 -7 5        Cluster 6: right hemisphere 339 9,153     Cerebellum crus I   n/a 27 -82 -31        Cluster 7: left hemisphere 305 8,235     Frontal orbital cortex   47 -48 26 -7        Cluster 8: left hemisphere 300 8,100     Cerebellum crus I   n/a -24 -82 -34        Cluster 9: right hemisphere 209 5,643     Frontal orbital cortex   38 51 26 -7        Cluster 10: right hemisphere 205 5,535     Planum temporale   48 54 -22 11 Central opercular cortex   48 57 -7 8 Central opercular cortex   48 60 -4 5 Insular cortex   48 39 -16 -4        Cluster 11: bilateral 26 702     Thalamus   n/a 0 -13 8        Cluster 12: left hemisphere 25 675     Posterior parahippocampal gyrus   20 -24 -22 -13        Cluster 13: left hemisphere 18 486     Anterior parahippocampal gyrus   34 -24 -1 -16        Cluster 14: right hemisphere 11 297     Hippocampus   n/a 24 -19 -13        Cluster 15: left hemisphere 3 81     Planum polare   48 -42 -19 -1   125 Table 4.8. TSI task fMRI-CPCA, DMN/occipital network (component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TSI task DMN/occipital network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Negative loadings              Cluster 1: left hemisphere 6,522 176,094     Occipital pole   18 -9 -91 20 Lingual gyrus   18 -9 -79 -7 Precuneus cortex   17 -3 -61 11 Posterior cingulate gyrus   30 -3 -52 20 Superior lateral occipital cortex   39 -48 -67 23 Posterior cingulate gyrus   31 -3 -37 44 Posterior parahippocampal gyrus   37 -27 -34 -16 Posterior parahippocampal gyrus   30 -24 -40 -13 Anterior cingulate gyrus   23 -3 -13 41 Lingual gyrus   37 -21 -46 -10 Posterior parahippocampal gyrus   30 -24 -22 -16 Cluster 1: bilateral       Lingual gyrus   17 0 -88 -1 Precuneus cortex   31 0 -73 29 Posterior cingulate gyrus   31 0 -49 32 Cluster 1: right hemisphere       Superior lateral occipital cortex   19 21 -85 26 Lingual gyrus   18 15 -76 -7 Intracalcarine cortex   17 9 -85 2 Precuneus cortex   30 9 -52 5 Precuneus cortex   17 6 -58 11 Posterior cingulate gyrus   30 3 -52 20 Lingual gyrus   37 24 -52 -10 Precuneus cortex   5 3 -46 59        Cluster 2: left hemisphere 1,038 28,026     Superior frontal gyrus   9 -24 35 44 Cluster 2: bilateral       Frontal pole   10 0 59 2 Cluster 2: right hemisphere       Frontal pole   9 24 38 44 (Table 4.8 continued on next page)    (Table 4.8, continued from previous page) 126 TSI task DMN/occipital network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Frontal pole   8 27 35 47 Frontal pole   9 15 53 41        Cluster 3: right hemisphere 289 7,803     Anterior superior temporal gyrus   22 57 -4 -13 Temporal pole   38 42 20 -28        Cluster 4: left hemisphere 214 5,778     Temporal pole   21 -51 8 -28 Anterior middle temporal gyrus   22 -57 -4 -13        Cluster 5: right hemisphere 14 378     Amygdala   n/a 24 -4 -19        Cluster 6: left hemisphere 8 216     Anterior parahippocampal gyrus   28 -21 -4 -19 Temporal pole   34 -30 2 -19  127 Table 4.9. TSI task fMRI-CPCA, response network (component 2): Clusters for the most ex-treme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TSI task response network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 5,979 161,433     Postcentral gyrus   3 -51 -22 38 Postcentral gyrus   40 -39 -37 50 Superior parietal lobule   7 -30 -52 59 Middle frontal gyrus   6 -27 -4 59 Precentral gyrus   6 -54 5 32 Precentral gyrus   6 -36 -7 62 Superior lateral occipital cortex   7 -18 -61 56 Superior lateral occipital cortex   19 -24 -70 29 Central opercular cortex   48 -51 5 2 Insular cortex   48 -39 -1 8 Cluster 1: bilateral       Supplementary motor area   6 0 -1 56 Cluster 1: right hemisphere       Postcentral gyrus   3 45 -28 50 Superior frontal gyrus   6 30 -4 62 Superior parietal lobule   2 33 -46 59 Postcentral gyrus   4 51 -19 47 Precentral gyrus   6 39 -10 62 Postcentral gyrus   3 36 -37 59 Central opercular cortex   48 57 -16 20 Superior lateral occipital cortex   7 15 -67 56 Superior lateral occipital cortex   19 27 -67 35        Cluster 2: left hemisphere 1,785 48,195     Cerebellum VI   n/a -27 -55 -25 Inferior lateral occipital cortex   19 -45 -67 -4 Inferior temporal gyrus, temporooccipital part   37 -45 -58 -13 Cluster 2: bilateral       Vermis VI   n/a 0 -67 -16 Vermis crus II   n/a 0 -73 -34 (Table 4.9 continued on next page)           (Table 4.9, continued from previous page) 128 TSI task response network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 2: right hemisphere       Cerebellum VI   n/a 24 -55 -25 Cerebellum VI   n/a 30 -52 -28 Cerebellum VIIIa   n/a 27 -58 -52 Cerebellum VIIb   n/a 15 -70 -49 Inferior lateral occipital cortex   37 48 -67 -1 Inferior temporal gyrus, temporooccipital part   37 45 -58 -13 Inferior lateral occipital cortex   19 39 -79 -4        Cluster 3: right hemisphere 119 3,213     Precentral gyrus   6 54 8 35        Cluster 4: left hemisphere 96 2,592     Cerebellum VIIIa   n/a -27 -61 -52        Cluster 5: left hemisphere 47 1,269     Thalamus   n/a -12 -19 8        Cluster 6: right hemisphere 35 945     Thalamus   n/a 12 -16 8        Cluster 7: left hemisphere 13 351     Insular cortex   48 -30 17 5        Cluster 8: right hemisphere 6 162     Insular cortex   48 33 17 5        Cluster 9: right hemisphere 4 108     Occipital fusiform gyrus   18 27 -85 2  129 Table 4.10. TSI task fMRI-CPCA, evaluation network (component 3): Clusters for the most ex-treme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TSI task evaluation network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 4,742 128,034     Frontal pole   45 -42 47 5 Middle frontal gyrus   9 -45 17 44 Middle frontal gyrus   8 -30 17 56 Frontal orbital cortex   38 -48 26 -10 Inferior frontal gyrus, pars opercularis   38 -54 20 -1 Insular cortex   47 -36 20 -4 Frontal pole   46 -24 53 23 Cluster 1: bilateral       Superior frontal gyrus   9 0 35 41 Superior frontal gyrus   8 0 26 50 Cluster 1: right hemisphere       Middle frontal gyrus   44 45 20 41 Frontal pole   47 36 53 -1 Frontal pole   46 39 53 8 Middle frontal gyrus   8 33 14 56 Frontal pole   47 51 38 -7 Frontal orbital cortex   38 48 23 -7 Frontal orbital cortex   47 36 23 -7        Cluster 2: left hemisphere 1,171 31,617     Cerebellum crus I   n/a -42 -70 -34 Cerebellum crus II   n/a -36 -70 -46 Cerebellum crus I   n/a -30 -67 -31 Cerebellum crus I   n/a -27 -70 -28 Cerebellum crus I   n/a -9 -79 -28 Cluster 2: right hemisphere       Cerebellum crus I   n/a 12 -79 -28 Cerebellum crus II   n/a 12 -82 -43 Cerebellum crus II   n/a 33 -73 -49 Cerebellum crus I   n/a 36 -67 -31        (Table 4.10 continued on next page)    (Table 4.10, continued from previous page) 130 TSI task evaluation network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 3: left hemisphere 942 25,434     Angular gyrus   39 -42 -58 50 Angular gyrus   40 -42 -49 38        Cluster 4: right hemisphere 863 23,301     Angular gyrus   40 42 -55 50 Superior lateral occipital cortex   7 36 -61 56        Cluster 5: left hemisphere 212 5,724     Inferior temporal gyrus, temporooccipital part   37 -57 -55 -13 Posterior inferior temporal gyrus   20 -63 -40 -16 Posterior middle temporal gyrus   21 -63 -34 -7 Posterior middle temporal gyrus   21 -63 -28 -7        Cluster 6: right hemisphere 127 3,429     Posterior middle temporal gyrus   21 63 -25 -10 Middle temporal gyrus, temporooccipital part   21 63 -46 -4 Inferior temporal gyrus, temporooccipital part   37 60 -55 -13        Cluster 7: left hemisphere 18 486     Superior lateral occipital cortex   7 -9 -70 62 Precuneus cortex   7 -3 -67 50        Cluster 8: right hemisphere 10 270     Frontal pole   9 24 44 38   131 Table 4.11. TGT task fMRI-CPCA, language network (component 1): Clusters for the most ex-treme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TGT language network (component 1) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 6,280 169,560     Temporal occipital fusiform cortex   19 -30 -61 -16 Inferior lateral occipital cortex   18 -27 -88 5 Posterior superior temporal gyrus   22 -57 -37 2 Superior lateral occipital cortex   7 -27 -67 44 Cluster 1: right hemisphere       Temporal occipital fusiform cortex   37 33 -52 -19 Occipital pole   18 24 -91 8        Cluster 2: left hemisphere 1,015 27,405     Middle frontal gyrus   6 -48 5 50 Inferior frontal gyrus, pars opercularis   48 -45 17 23 Inferior frontal gyrus, pars triangularis   45 -48 29 20 Inferior frontal gyrus, pars triangularis   47 -48 26 -1 Frontal orbital cortex   47 -36 29 -1        Cluster 3: right hemisphere 333 8,991     Posterior superior temporal gyrus   22 54 -28 2 Planum temporale   22 57 -19 2 Posterior superior temporal gyrus   22 60 -16 -1        Cluster 4: left hemisphere 299 8,073     Superior frontal gyrus   6 -3 11 59        Cluster 5: right hemisphere 47 1,269     Cerebellum VIIb   n/a 24 -70 -52        Cluster 6: right hemisphere 38 1,026     Middle frontal gyrus   6 51 5 53        Cluster 7: left hemisphere 19 513     Thalamus   n/a -24 -28 -4   132 Table 4.12. TGT task fMRI-CPCA, posterior-medial DMN (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TGT posterior-medial DMN (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Negative loadings              Cluster 1: left hemisphere 7,287 196,749     Precentral gyrus   4 -12 -37 47 Postcentral gyrus   2 -24 -40 62 Cuneal cortex   18 -12 -73 20 Cuneal cortex   18 -6 -79 29 Posterior cingulate gyrus   31 -12 -22 38 Anterior cingulate gyrus   24 -6 5 41 Precentral gyrus   4 -42 -16 50 Anterior supramarginal gyrus   2 -60 -28 32 Precentral gyrus   6 -24 -7 50 Superior lateral occipital cortex   7 -12 -73 47 Lingual gyrus   18 -15 -64 -4 Precentral gyrus   6 -24 -16 68 Superior frontal gyrus   6 -24 -4 62 Superior frontal gyrus   6 -24 -4 56 Anterior supramarginal gyrus   2 -48 -34 35 Postcentral gyrus   3 -57 -10 38 Cluster 1: bilateral       Supplementary motor area   6 0 -10 59 Cluster 1: right hemisphere       Precuneus cortex   5 6 -49 56 Precuneus cortex   7 9 -64 53 Superior parietal lobule   5 18 -46 68 Cuneal cortex   18 18 -67 20 Precuneus cortex   7 9 -76 35 Anterior cingulate gyrus   24 3 5 44 Middle frontal gyrus   9 27 32 41 Cuneal cortex   19 15 -82 23 Superior frontal gyrus   8 24 11 59 Precentral gyrus   6 6 -16 47 Superior parietal lobule   2 42 -37 56 Anterior supramarginal gyrus   2 57 -28 32 (Table 4.12 continued on next page)    (Table 4.12, continued from previous page) 133 TGT posterior-medial DMN (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Postcentral gyrus   4 54 -16 44 Lingual gyrus   18 15 -67 -4 Anterior supramarginal gyrus   3 57 -22 41 Precentral gyrus   3 39 -13 41 Precentral gyrus   6 45 -10 44 Superior frontal gyrus   6 12 -10 71 Precentral gyrus   4 12 -22 74 Postcentral gyrus   48 60 -16 23 Postcentral gyrus   48 54 -13 26        Cluster 2: right hemisphere 420 11,340     Inferior lateral occipital cortex   37 45 -64 14 Superior lateral occipital cortex   19 39 -70 32 Angular gyrus   22 48 -49 23        Cluster 3: left hemisphere 108 2,916     Anterior cingulate gyrus   24 -6 38 11 Anterior cingulate gyrus   32 -12 44 8 Cluster 3: right hemisphere       Frontal pole   10 9 59 14 Paracingulate gyrus   10 3 53 5 Anterior cingulate gyrus   32 6 41 8 Paracingulate gyrus   32 6 50 20        Cluster 4: left hemisphere 107 2,889     Inferior lateral occipital cortex   37 -45 -67 11        Cluster 5: left hemisphere 34 918     Frontal pole   46 -27 38 29        Cluster 6: left hemisphere 26 702     Cerebellum VI   n/a -30 -43 -40        Cluster 7: left hemisphere 23 621     Cerebellum VIIIb   n/a -15 -58 -55        (Table 4.12 continued on next page)    (Table 4.12, continued from previous page) 134 TGT posterior-medial DMN (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 8: right hemisphere 13 351     Cerebellum VIIIb   n/a 24 -52 -52        Cluster 9: right hemisphere 5 135     Insular cortex   48 36 -7 -1        Cluster 10: right hemisphere 4 108     Posterior supramarginal gyrus   40 39 -43 35        Cluster 11: left hemisphere 3 81     Precentral gyrus   48 -60 -4 17   135 Table 4.13. TGT task fMRI-CPCA, anterior/posterior-lateral DMN (component 3): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TGT anterior/posterior-lateral DMN (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: right hemisphere 40 1,080     Occipital fusiform gyrus   19 30 -67 -10 Temporal occipital fusiform cortex   37 33 -52 -13        Cluster 2: left hemisphere 19 513     Occipital fusiform gyrus   19 -30 -67 -10        Cluster 3: left hemisphere 4 108     Temporal occipital fusiform cortex   37 -36 -46 -16        Negative loadings              Cluster 1: left hemisphere 4,227 114,129     Frontal pole   46 -21 56 23 Frontal pole   10 -3 56 -1 Superior frontal gyrus   9 -24 35 44 Middle frontal gyrus   46 -39 23 41 Frontal pole   11 -21 56 -1 Paracingulate gyrus   32 -3 35 35 Frontal pole   47 -33 50 -7 Frontal pole   47 -42 41 -7 Frontal pole   47 -45 38 -10 Anterior cingulate gyrus   24 -3 26 23 Frontal pole   48 -30 35 14 Cluster 1: bilateral       Frontal pole   10 0 59 -4 Superior frontal gyrus   8 0 32 56 Cluster 1: right hemisphere       Middle frontal gyrus   9 30 29 47 Superior frontal gyrus   8 27 26 56 Frontal pole   10 6 62 5 Paracingulate gyrus   32 3 47 23 Frontal pole   10 6 56 11 Frontal pole   9 15 53 29 (Table 4.13 continued on next page)    (Table 4.13, continued from previous page) 136 TGT anterior/posterior-lateral DMN (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Middle frontal gyrus   9 42 20 47 Frontal pole   9 18 50 32 Frontal pole   47 39 47 5 Frontal pole   47 33 56 -1 Frontal pole   11 18 59 -10 Frontal pole   45 48 38 -1 Frontal pole   47 42 50 -7 Frontal pole   47 45 44 -7 Frontal pole   47 48 38 -13        Cluster 2: left hemisphere 1,307 35,289     Cerebellum crus I   n/a -33 -76 -37 Cluster 2: right hemisphere       Cerebellum crus II   n/a 24 -82 -37 Cerebellum crus I   n/a 48 -70 -37        Cluster 3: left hemisphere 1,191 32,157     Angular gyrus   39 -54 -55 44 Superior lateral occipital cortex   19 -42 -79 35 Superior lateral occipital cortex   19 -39 -79 41 Superior lateral occipital cortex   39 -45 -67 50 Superior lateral occipital cortex   39 -57 -67 26 Superior lateral occipital cortex   39 -51 -73 29 Middle temporal gyrus, temporooccipital part   37 -66 -46 -7 Middle temporal gyrus, temporooccipital part   37 -63 -55 -7 Posterior middle temporal gyrus   21 -63 -22 -13 Middle temporal gyrus, temporooccipital part   37 -60 -61 -10 Inferior lateral occipital cortex   37 -60 -64 -1 Posterior middle temporal gyrus   20 -54 -16 -16        Cluster 4: right hemisphere 969 26,163     Superior lateral occipital cortex   39 51 -73 35 Superior lateral occipital cortex   39 54 -61 41 Angular gyrus   39 60 -55 35 Superior lateral occipital cortex   39 48 -67 47 Superior lateral occipital cortex   39 60 -61 20 Inferior lateral occipital cortex   37 60 -67 8 Middle temporal gyrus, temporooccipital part   21 66 -55 5 (Table 4.13 continued on next page)    (Table 4.13, continued from previous page) 137 TGT anterior/posterior-lateral DMN (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 5: right hemisphere 101 2,727     Posterior middle temporal gyrus   21 66 -16 -16 Anterior middle temporal gyrus   21 54 -1 -28        Cluster 6: left hemisphere 74 1,998     Superior lateral occipital cortex   7 -3 -70 68 Superior lateral occipital cortex   7 -6 -79 62        Cluster 7: left hemisphere 45 1,215     Precuneus cortex   31 -3 -55 32 Cluster 7: bilateral       Precuneus cortex   7 0 -67 47        Cluster 8: right hemisphere 23 621     Frontal orbital cortex   47 33 20 -10        Cluster 9: left hemisphere 15 405     Insular cortex   48 -33 17 -10        Cluster 10: right hemisphere 9 243     Posterior middle temporal gyrus   21 72 -34 -7        Cluster 11: bilateral 5 135     Posterior cingulate gyrus   23 0 -25 41        Cluster 12: right hemisphere 2 54     Inferior temporal gyrus, temporooccipital part   20 66 -52 -13   138 Table 4.14. TGT task fMRI-CPCA, auditory network (component 4): Clusters for the most ex-treme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. TGT auditory network (component 4) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 3,130 84,510     Heschl's gyrus   48 -54 -13 2        Cluster 2: right hemisphere 3,126 84,402     Anterior superior temporal gyrus   48 60 -7 -1 Anterior temporal fusiform cortex   36 36 2 -37 Anterior temporal fusiform cortex   20 39 -7 -40        Cluster 3: left hemisphere 1,401 37,827     Temporal occipital fusiform cortex   37 -30 -55 -13 Posterior temporal fusiform cortex   37 -30 -37 -19 Posterior temporal fusiform cortex   37 -27 -43 -16 Thalamus   n/a -15 -28 -4 Cluster 3: right hemisphere       Temporal occipital fusiform cortex   37 30 -52 -13 Posterior temporal fusiform cortex   37 30 -37 -19 Thalamus   n/a 15 -25 -4        Cluster 4: left hemisphere 192 5,184     Inferior frontal gyrus, pars triangularis   45 -48 23 14 Inferior frontal gyrus, pars opercularis   48 -45 14 23        Cluster 5: right hemisphere 80 2,160     Inferior frontal gyrus, pars opercularis   48 45 17 23 Inferior frontal gyrus, pars triangularis   45 48 26 17        Cluster 6: right hemisphere 44 1,188     Precentral gyrus   6 54 2 47        Cluster 7: left hemisphere 28 756     Precentral gyrus   6 -54 -7 50        (Table 4.14 continued on next page)    (Table 4.14, continued from previous page) 139 TGT auditory network (component 4) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 8: left hemisphere 16 432     Frontal orbital cortex   47 -36 32 -10        Negative loadings              Cluster 1: left hemisphere 8 216     Inferior lateral occipital cortex   37 -45 -70 5        Cluster 2: right hemisphere 6 162     Inferior lateral occipital cortex   37 42 -64 2    140 4.8. Chapter 4 Figures Figure 4.1. WM task fMRI-CPCA, response/attention network (component 1): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.20, max = 0.37; no negative loadings above threshold). Images are displayed in neuro-logical orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. 4L = 4 letters; 6L = 6 letters.   141 Figure 4.2. WM task fMRI-CPCA, response/attention network (component 1): Estimated HDRs illustrating delay × time interaction. Asterisks indicate significant delay × time contrast between adjacent time bins. ** = p < .01; *** = p < .001.   -.20-.10.00.10.20.30.40.502 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)Delay  Time0s delay4s delay******** 142 Figure 4.3. WM task fMRI-CPCA, visual attention network (component 2): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.12, max = 0.62; blue/green = negative loadings, min = -0.16, max = -0.12). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. 4L = 4 letters; 6L = 6 letters.   143 Figure 4.4. WM task fMRI-CPCA, visual attention network (component 2): Estimated HDRs illustrating delay × time interaction. Asterisks indicate significant delay × time contrast between adjacent time bins. * = p < .05; *** = p < .001.   -.20-.10.00.10.20.30.40.502 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)Delay  Time0s delay4s delay***** 144 Figure 4.5. WM task fMRI-CPCA, default mode network (DMN, component 3): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.15, max = 0.18; blue/green = negative loadings, min = -0.25, max = -0.15). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. Y axis has been reversed (nega-tive up, positive down) to facilitate interpretation (i.e., values above X axis reflect activation of blue/green voxels, and values below X axis reflect deactivation of blue/green voxels). 4L = 4 let-ters; 6L = 6 letters.   145 Figure 4.6. WM task fMRI-CPCA, default mode network (DMN, component 3): Graphs illustrat-ing effects of cognitive load and delay length. A (top): mean predictor weights illustrating signif-icant load × delay interaction. B (bottom): predictor weights averaged over load to illustrate delay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins). Y axes are reversed (negative up, positive down) to facilitate interpretation (i.e., val-ues above X axis reflect activation, and values below X axis reflect deactivation). * = p < .05; ** = p < .01; *** = p < .001.    146 Figure 4.7. SCAP task performance results, graphs illustrating effects of cognitive load and delay length. A (top): mean percentage of correct responses for each load level. B (bottom): mean percentage of correct responses for each delay length. Asterisks indicate significant paired t-tests between adjacent conditions. * = p < .05; *** = p < .001.     147 Figure 4.8. SCAP task fMRI-CPCA, external attention network (component 1): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.39, max = 0.59; no negative loadings above threshold). Images are displayed in neuro-logical orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time (each load level displayed on a separate graph).    148 Figure 4.9. SCAP task fMRI-CPCA, external attention network (component 1): Graphs illustrat-ing effects of cognitive load and delay length. A (top left): mean predictor weights illustrating main effect of load (asterisks indicate significant paired t-tests between adjacent load condi-tions). B (top right): mean predictor weights illustrating main effect of delay (asterisks indicate significant paired t-tests between adjacent delay conditions). C (bottom): predictor weights av-eraged over load to illustrate delay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins; contrast with greatest effect size is flagged). a = linear ef-fect of delay; b = quadratic effect of delay; * = p < .05; ** = p < .01; *** = p < .001.    149 Figure 4.10. SCAP task fMRI-CPCA, DMN (component 2): Anatomical and temporal character-istics. A (top): dominant 10% of loadings (blue/green = negative loadings, min = -0.50, max = -0.33; no positive loadings above threshold). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time (each load level displayed on a separate graph). Y axes are reversed to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis reflect deactivation).   150 Figure 4.11. SCAP task fMRI-CPCA, default mode network (DMN, component 2): Graphs illus-trating effects of cognitive load and delay length. Y axes are reversed (negative up, positive down) to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis reflect deactivation). A (top left): mean predictor weights illustrating significant main effect of load (asterisks indicate significant paired t-tests between adjacent load conditions). B (top right): mean predictor weights illustrating significant main effect of delay (asterisks indicate significant paired t-tests between adjacent delay conditions). C (bottom): predictor weights av-eraged over load to illustrate delay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins; contrast with greatest effect size is flagged). a = linear ef-fect of delay; b = quadratic effect of delay; *** = p < .001.   151 Figure 4.12. TSI task fMRI-CPCA, DMN/occipital network (component 1): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (blue/green = negative loadings, min = -0.34, max = -0.20; no positive loadings above threshold). Images are displayed in neuro-logical orientation (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each word-reading condition. Y axis is reversed to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis reflect deactivation). cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.   152 Figure 4.13. TSI task fMRI-CPCA, DMN/occipital network (component 1): Graphs illustrating effects of stimulus congruency and task-switch condition. Y axes are reversed (negative up, posi-tive down) to facilitate interpretation (i.e., values above X axis reflect activation, and values be-low X axis reflect deactivation). A (top left): predictor weights averaged over task-switch to illustrate congruency × time interaction (asterisks indicate significant congruency × time con-trasts between adjacent time bins). B (top right): predictor weights averaged over congruency to illustrate task-switch × time interaction (asterisks indicate significant task-switch × time con-trasts between adjacent time bins). C (bottom): mean predictor weights illustrating significant congruency × task-switch interaction. WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus; cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; * = p < .05; ** = p < .01; *** = p < .001.   153 Figure 4.14. TSI task fMRI-CPCA, response network (component 2): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.19, max = 0.32; no negative loadings above threshold). Images are displayed in neurological orienta-tion (left is left) with MNI coordinates. B (bottom): predictor weights for each word-reading condition plotted over post-stimulus time. cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.    154 Figure 4.15. TSI task fMRI-CPCA, response network (component 2): Estimated HDRs illustrat-ing congruency × time interaction. Asterisks indicate significant congruency × time contrasts be-tween adjacent time bins. WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus; *** = p < .001.  -.40-.30-.20-.10.00.10.20.30.40.502 4 6 8 10 12 14 16Predictor WeightsPost-Stimulus Time (Seconds)Congruency  TimeWNWI*** 155 Figure 4.16. TSI task fMRI-CPCA, evaluation network (component 3): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.16, max = 0.26; no negative loadings above threshold). Images are displayed in neurological orienta-tion (left is left) with MNI coordinates. B (bottom): predictor weights for each word-reading condition plotted over post-stimulus time. cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.   156 Figure 4.17. TSI task fMRI-CPCA, evaluation network (component 3): Estimated HDRs illus-trating congruency × time interaction. Asterisks indicate significant congruency × time contrasts between adjacent time bins. WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus; ** = p < .01; *** = p < .001.  -.30-.20-.10.00.10.20.30.402 4 6 8 10 12 14 16Predictor WeightsPost-Stimulus Time (Seconds)Congruency  TimeWNWI******* 157 Figure 4.18. TGT task fMRI-CPCA, language network (component 1): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.15, max = 0.35; no negative loadings above threshold). Images are displayed in neurological orienta-tion (left is left) with MNI coordinates. B (bottom): predictor weights plotted over post-stimulus time for each task condition. Asterisks indicate significant condition × time contrasts between adjacent time bins. * = p < .05; ** = p < .01; *** = p < .001.   158 Figure 4.19. TGT task fMRI-CPCA, posterior-medial DMN (component 2): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (blue/green = negative loadings, min = -0.20, max = -0.12; no positive loadings above threshold). Images are displayed in neuro-logical orientation (left is left) with MNI coordinates. B (bottom): predictor weights for each task condition plotted over post-stimulus time. Y axis is reversed (negative up, positive down) to facilitate interpretation (i.e., values above X axis reflect activation, and values below X axis re-flect deactivation).   159 Figure 4.20. TGT task fMRI-CPCA, anterior/posterior-lateral DMN (component 3): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.10, max = 0.12; blue/green = negative loadings, min = -0.15, max = -0.10). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights for each task condition plotted over post-stimulus time. Y axis is reversed (negative up, positive down) to facilitate interpretation (i.e., values above X axis reflect activation in blue/green voxels, and values below X axis reflect deactivation in blue/green voxels).   160 Figure 4.21. TGT task fMRI-CPCA, auditory network (component 4): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.06, max = 0.39; blue/green = negative loadings, min = -0.07, max = -0.06). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): predictor weights for each task condition plotted over post-stimulus time. Asterisks indicate significant condition × time contrasts between adjacent time bins. *** = p < .001.     161 Chapter 5: Identifying Networks Underlying Working Memory Deficits in Schizophrenia Using Multi-Experiment fMRI-CPCA 5.1. Aims and Hypotheses The overarching goal of this work was to characterize task-induced functional brain net-works that may underlie WM deficits in schizophrenia. As demonstrated in Chapter 3, multi-experiment fMRI-CPCA may produce a more refined understanding of networks engaged during a cognitive task, and allows for the direct evaluation of the extent to which a given network is engaged across different tasks. However, limitations in interpreting the functions of networks observed in the verbal WM-TGT design were noted, which could be addressed in part with the addition of a visuospatial WM task and an externally-oriented, set-switching Stroop task. Each of these four tasks were analysed in the previous chapter. The aim of the following study was to characterize the functions of task-related networks simultaneously derived from all four tasks of interest, their relationships with task performance, and the nature of potentially suboptimal activ-ity in schizophrenia patients. Based on the results of the single-task fMRI-CPCAs (Chapter 4) and the preliminary 2-task fMRI-CPCA with the WM and TGT tasks (Chapter 3), it was hypothesized that networks underlying motor responses (dominated by activity in somatomotor cortex), internally-oriented attention (including DLPFC, dorsal anterior/paracingulate gyrus, and intraparietal sulcus), and visual attention (including SMA, lateral occipital activation extending dorsally into posterior pa-rietal cortex, and thalamus) as well as DMN deactivation would emerge to varying degrees across all tasks. More basic visual and auditory sensory networks were expected to separate out as well, which was observed in the preliminary multi-experiment CPCA in Chapter 3, but the networks were not discussed in detail. The DMN sub-region networks observed in the TGT task  162 in the previous chapter (components 2 and 3) were not expected to emerge, as they did not emerge when the TGT data was combined only with the WM task, and neither the SCAP nor the TSI task exhibited decomposition of DMN regions either. Based on previous research suggesting a link between the DLPFC and cognitive deficits in schizophrenia (Barch & Ceaser, 2012; Glausier & Lewis, 2018; Potkin et al., 2009), suboptimal patterns of activity were expected to emerge in a network resembling the internal attention network (component 3 from the WM-TGT analysis in Chapter 3), reflecting cognitive inefficiency (i.e., hyperactivity at low levels of cogni-tive load) and/or hypoactivity as a whole compared with healthy controls. 5.2. Methods 5.2.1. Datasets The WM task, SCAP task, TSI task, and the TGT task analysed in the previous chapter were entered into a 4-task fMRI-CPCA. The task paradigms and information regarding data ac-quisition and preprocessing for each dataset are detailed in Chapter 2 (Sections 2.2 and 2.3 for tasks and fMRI data information, respectively). Each dataset comprised a healthy control group and a schizophrenia patient group. Although the healthy control participants largely overlapped with those included in the respective single-task CPCAs reported in Chapter 4, in some cases they only comprised sub-samples as described below. 5.2.1.1. WM and TSI participants Participants for the WM and TSI tasks were drawn from the same dataset, and so only participants who completed both the WM and TSI tasks were included in this analysis. This re-sulted in a sample size of 54 for both tasks (26 healthy controls and 28 schizophrenia patients). Demographic and neuropsychological information for these participants is presented in Table 5.1, including age, Quick estimated IQ (Mortimer & Bowen, 1999), scaled digit span score from  163 the Wechsler Adult Intelligence Scale subtest (WAIS; Wechsler, 2008), gender distribution, handedness as measured with the AHPQ (Annett, 1970), socioeconomic status (SES), education-al achievement, and clinical information (patients only). The participant groups differed in age (mean age greater in patients; mean difference = 11.624 years, SEM = 2.56, t(50.50) = 4.539, p < .001), scaled digit span score (digit span higher in controls; mean difference = 1.319 digits, SEM = 0.607, t(52) = 2.174, p = .034), SES factor scores (mean factor score lower in controls; mean difference = 19.462, SEM = 4.674, t(46) = 4.232, p < .001), and years of education (mean years of education greater in controls; mean difference = 1.658 years, SEM = 0.603, t(52) = 2.750, p = .008).  The patient group consisted of outpatients with a diagnosis of schizophrenia or related disorder (i.e., schizoaffective or psychotic disorder not otherwise specified) being followed by a mental health care provider at the time of participation. Diagnoses were confirmed for each par-ticipant using the Mini International Neuropsychiatric Interview (MINI; Sheehan et al., 1998), which was also used to screen healthy controls for psychiatric illness. Symptom severity in pa-tients was assessed with the Signs and Symptoms of Psychotic Illness semi-structured interview (SSPI; Liddle, Ngan, Duffield, & Warren, 2002). All patients reported taking antipsychotic med-ication at the time of testing, but one patient did not have up-to-date medication information. Of the 27 patients with complete medication information, 25 patients reported taking atypical anti-psychotics (e.g., clozapine, olanzapine, risperidone) and 2 patients reported taking typical anti-psychotic medication (e.g., loxapine, fluphenazine) as their primary medication. 9 patients reported taking an additional antipsychotic, 7 of which were atypical, and 2 of which were typi-cal antipsychotics. 4 patients reported taking antianxiety medication, 9 reported taking an antide-pressant, 3 reported taking lithium, 2 reported taking an anticonvulsant, and 5 reported taking an  164 anticholinergic as secondary medication. Further information regarding illness duration and se-verity is provided in Table 5.1. 5.2.1.2. SCAP participants The original study from which these data were downloaded comprised a considerably greater number of healthy controls than schizophrenia patients (119 controls and 44 patients with useable fMRI data), and this dataset as a whole was larger than the other sample sizes for the tasks included in the present study (163 vs. 54-60). Therefore, to avoid over-representing the SCAP healthy controls in the component solution extracted from the combined datasets, as well as to reduce computing time, 44 controls were randomly selected from the larger sample includ-ed in Chapter 4, which had similar demographic characteristics to the full sample from which they were drawn (compare Tables 4.1 and 5.2). This resulted in the SCAP dataset comprising a sample size of n = 88 (44 healthy controls and 44 schizophrenia patients). Demographic and neu-ropsychological information for the controls and patients included in this chapter is presented in Table 5.2, including age, raw digit span score as measured with the Wechsler Memory Scale (WMS-IV; Wechsler, 2009), gender distribution, handedness as measured with the Edinburgh Handedness Inventory (Oldfield, 1971), educational achievement, and clinical information (pa-tients only). The participant groups differed in age (mean age greater in patients; mean differ-ence = 5.477 years, SEM = 1.912, t(86) = 2.864, p = .005), gender distribution (greater proportion of males in the patient group; 𝜒2(1) = 8.017, p = .005), and years of education (mean years of education greater in controls; mean difference = 2.341 years, SEM = 0.383, t(86) = 6.108, p < .001), which was also the case when patients were compared with the full sample of healthy controls (all ps < .05). Participants also differed in WMS total raw digit span score  165 (higher digit span in controls; mean difference = 7.636 digits, SEM = 1.131, t(86) = 6.753, p < .001); however, standard scaled scores were not available for this dataset. The patient group consisted of outpatients with a primary diagnosis of schizophrenia or schizoaffective disorder, based on the Structured Clinical Interview for DSM-IV Axis I Disor-ders (SCID-I; First, Spitzer, Gibbon, & Williams, 2002). Symptom severity was assessed with the Scale for the Assessment of Negative Symptoms (SANS; Andreasen, 1984a) and the Scale for the Assessment of Positive Symptoms (SAPS; Andreasen, 1984b). Medication information was not available for 4 patients. Of the 40 patients with available medication history, 39 reported taking at least one atypical antipsychotic. 4 patients reported taking both an atypical and a typical antipsychotic; no patients reported taking only a typical antipsychotic. 1 patient did not report taking any antipsychotic medication. 18 patients reported taking an antidepressant, 11 reported taking an anticonvulsant, 5 reported taking an anticholinergic, and 3 reported taking lithium. Fur-ther information regarding illness duration and severity is provided in Table 5.2. 5.2.1.3. TGT participants The healthy controls included in this chapter were the same participants as those included in the previous TGT analyses in chapters 3 and 4, resulting in a total sample size of n = 60 (32 healthy controls and 28 schizophrenia patients). Demographic and neuropsychological infor-mation for the two groups is provided in Table 5.3, including age, Quick IQ (Mortimer & Bowen, 1999), WAIS scaled digit span (Wechsler, 2008), gender distribution, AHPQ handedness (Annett, 1970), SES, educational achievement, and clinical information (patients only). The groups differed in age (mean age greater in patients; mean difference = 4.786 years, SEM = 2.376, t(58) = 2.014, p = .049), years of education (mean years of education greater in controls; mean difference = 1.257 years, SEM = 0.566, t(58) = 2.220, p = .030), and scaled digit span  166 scores (higher digit span in controls; mean difference = 1.887 digits, SEM = 0.726, t(57) = 2.598, p = .012). The patient group consisted of outpatients with a diagnosis of schizophrenia or related disorder (i.e., schizoaffective or psychotic disorder not otherwise specified) being followed by a mental health care provider at the time of participation. Diagnoses were confirmed for each par-ticipant using the MINI (Sheehan et al., 1998), which was also used to screen healthy controls for psychiatric illness. Symptom severity in patients was assessed using the SSPI semi-structured interview (Liddle et al., 2002). All patients but three reported taking antipsychotic medication at the time of testing, and medication details were missing for one participant. Of the 24 patients with current medication information, 22 patients reported taking atypical antipsychotics and 2 patients reported taking typical antipsychotic medication as their primary medication. 11 patients reported taking an additional antipsychotic, 8 of which were atypical, and 3 of which were typi-cal antipsychotics. 8 patients reported taking antianxiety medication, 10 reported taking an anti-depressant, 2 reported taking lithium, 3 reported taking an anticonvulsant, and 2 reported taking an anticholinergic as secondary medication. Further information regarding illness duration and severity is provided in Table 5.3. 5.2.2. Analysis 5.2.2.1. Task performance Task performance was examined in the WM task, SCAP task, and TSI task only, as no overt behavioural response was recorded in the TGT task. For the WM task, task performance was calculated as percentage of correct responses for each condition. To examine effects of task condition and differences between healthy controls and schizophrenia patients, these accuracy measures were entered into a 2 (cognitive load; 4 vs. 6 letters) × 2 (delay duration; 0s vs. 4s) ×  167 2 (group) mixed-model ANOVA. As age was a potential confound in this sample, group effects were further examined in a follow-up ANOVA with the same factors, but performed on the re-siduals obtained from regressing out variance in accuracy predicted by age. Digit span and years of education were not statistically controlled for because these variables are likely to be intrinsi-cally tied to differences between healthy individuals and people with a diagnosis of schizophre-nia in the general population. For the SCAP task, task performance was also calculated as percentage of correct re-sponses for each condition. To examine effects of task condition and differences between healthy controls and schizophrenia patients, these accuracy measures were entered into a 4 (cognitive load; 1, 3, 5, or 7 dots) × 3 (delay duration; 1.5s, 3.0s, or 4.5s) × 2 (group) ANOVA. As age and gender were potential confounds in this sample, group effects were further examined in a follow-up ANOVA performed on the residuals obtained from regressing out variance predicted by age and gender. Educational achievement and digit span were not statistically controlled. For the TSI task, task performance was measured as either percentage of correct respons-es or mean reaction time (RT) for each condition. To examine effects of task condition and dif-ferences between controls and patients, separate analyses were carried out on percent correct and RT measures, both of which were examined in a 2 (stimulus congruency; neutral vs. incongru-ent) × 2 (task-switch condition; preceding block = neutral vs. incongruent colour-naming) × 2 (group) ANOVA. As in the WM task (which comprised the same sample of participants), group effects were further examined in follow-up ANOVAs performed on the residuals obtained from regressing out variance predicted by age.  168 5.2.2.2. Functional connectivity analyses Multi-experiment fMRI-CPCA was carried out as described in Chapter 2 (Section 2.4 and Figure 2.5). To examine effects of task condition and group differences on HDRs, a mixed-model ANOVA was performed for each component extracted and for each task as described be-low. For the WM task, each level of cognitive load and delay duration was specified in the de-sign matrix (i.e., G matrix), and the time bins for which a FIR basis function was specified were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,000ms), re-sulting in G matrices with 374 rows (scans) and 40 columns (4 conditions × 10 post-stimulus time bins) per task run. After the regression and PCA steps, within-subject factors of cognitive load (4 vs. 6 letters), delay length (0s vs. 4s), and time (10 post-stimulus time bins) were exam-ined with the resulting predictor weights, resulting in a 2 (load) × 2 (delay) × 10 (time) × 2 (group) ANOVA for each component. Significant effects of load, delay, load × delay, delay × time, and any effects/interactions involving group differences were further examined. Post hoc analyses of load and delay were carried out using polynomial contrasts, and interactions involv-ing time were examined using repeated measures contrasts between adjacent time bins. To verify whether group effects remained after accounting for age, a follow-up ANOVA was carried out for each component with the same factors entered into the model, but performed on the residuals obtained from regressing out variance predicted by age. For the SCAP task, each level of cognitive load and delay duration was specified in the design, and the time bins for which a FIR basis function was specified were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,000ms), resulting in G matrices with 291 rows (scans) and 120 columns (12 conditions × 10 post-stimulus time bins) per partici- 169 pant. After the regression and PCA steps, within-subject factors of cognitive load (1, 3, 5, or 7 dots), delay length (1.5s, 3.0s, or 4.5s), and time (10 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 4 (load) × 3 (delay) × 10 (time) × 2 (group) ANOVA for each component. Significant effects of load, delay, delay × time, and any ef-fects/interactions involving group differences were further examined. Post hoc analyses of load and delay effects were carried out using polynomial contrasts, and interactions involving time were examined using repeated measures contrasts between adjacent time bins. To verify whether group effects remained after accounting for differences in age and gender distribution, a follow-up ANOVA was performed with the same factors entered into the model, but computed on the residuals obtained from regressing out variance predicted by age and gender. For the TSI task, each level of stimulus congruency and task-switch condition was speci-fied in the design (but only for word-reading trials), and the time bins for which a FIR basis function was specified were scans 1-10 following trial onset (i.e., 20 seconds of post-stimulus time with TR = 2,000ms), resulting in G matrices with 330 rows (scans) and 40 columns (4 con-ditions × 10 post-stimulus time bins) per participant. After the regression and PCA steps, within-subject factors of stimulus congruency (neutral vs. incongruent stimulus), task-switch condition (preceding block = neutral vs. incongruent colour-naming), and time (10 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 2 (congruency) × 2 (task-switch) × 10 (time) × 2 (group) ANOVA for each component. Post-hoc analyses of significant effects of congruency, task-switch, congruency × task-switch, congruency × time, task-switch × time, and any effects/interactions involving group differences were carried out, with effects of congruency and task-switch examined using polynomial contrasts, and interactions involving time examined using repeated measures contrasts between adjacent time bins. To verify whether  170 group effects remained after accounting for differences in age, a follow-up ANOVA was per-formed with the same factors entered into the model, but computed on the residuals obtained from regressing out variance predicted by age. Finally, for the TGT task, the generating and hearing conditions were both specified in the design, and the time bins for which a FIR basis function was specified were scans 1-10 fol-lowing trial onset (i.e., 25 seconds of post-stimulus time with TR = 2,500ms). This resulted in G matrices with 176 rows (scans) and 20 columns (2 conditions × 10 post-stimulus time bins) per task run. After the regression and PCA steps, within-subject factors of condition (generating vs. hearing) and time (10 post-stimulus time bins) were examined with the resulting predictor weights, resulting in a 2 (condition) × 10 (time) × 2 (group) ANOVA for each component. Post-hoc analyses of significant effects of condition, condition × time, and any effects/interactions involving group differences were carried out, with interactions involving time examined using repeated measures contrasts between adjacent time bins. To verify whether group effects re-mained after accounting for differences in age, a follow-up ANOVA was performed with the same factors entered into the model, but performed on the residuals obtained from regressing out variance predicted by age. 5.2.2.3. Correlations between task performance and brain activity Correlations between task performance and HDR shapes were examined for the WM, SCAP, and TSI tasks. Past work has shown that the HDR increase and the subsequent return to baseline can reflect separable cognitive/behavioural processes (Lavigne, Menon, & Woodward, 2016); therefore, in all three of these tasks, HDR increase-to-peak (ITP) and return-to-baseline (RTB) measures were computed separately by averaging across time bins that were selected by examining HDR shapes. For ITP measures, time bins included the HDR peak and the time bins  171 preceding it up to (but not including) the baseline point. For RTB measures, time bins included the time bin immediately following the peak and the subsequent time bins down to the lowest point (or highest, in cases where the HDR reflected deactivation rather than activation). In cases where the onset and/or termination of the HDR was ambiguous, contrasts were performed be-tween adjacent time bins to determine whether there was a significant increase/decrease from one time bin to the next; for example, if there was no significant difference between pre-peak time points 1 and 2, then the second time point would be marked as the baseline point. As the purpose of this analysis was to gain a general sense of patterns predicting task performance, pre-dictor weights were averaged across task conditions where appropriate. Due to the large number of statistical tests being carried out for this section, a threshold of p < .01 was set as the signifi-cance threshold, and results with a p value higher than this but < .05 are noted as trend correla-tions. To examine relationships between task performance and brain activity in the WM task, two-tailed Pearson correlations were performed between accuracy (averaged over load condi-tions) and the corresponding ITP/RTB for each delay condition. The delay conditions were sepa-rated out because the difference in timing of the task phases results in correspondingly different timing of HDR phases. Participants’ WAIS digit span scores were examined as an additional be-havioural measure of WM capacity. WAIS digit span scores are scaled according to an individu-al’s age, and so it is a practical measure of WM capacity that is not confounded by age (and was verified to not be directly correlated with age in this sample, p > .18). For the SCAP task, two-tailed Pearson correlations were performed between accuracy (averaged over load conditions) and the corresponding ITP/RTB for each delay condition, as in the WM task. As standard WMS scores were not available for this dataset, but digit span was  172 nevertheless a neuropsychological measure of interest, partial correlations were performed be-tween raw digit span and HDR ITPs/RTBs while controlling for age. For the TSI task, two-tailed Pearson correlations were performed between task perfor-mance (percent correct or mean RTs) and HDR ITP/RTB for each component. Task performance and HDRs were averaged across all task conditions for this analysis. Although WAIS digit span was not expected to be correlated with the TSI task, it was included as a behavioural measure of interest for comparison with results from the WM task (which was completed by the same partic-ipants). 5.3. Task Performance Results 5.3.1. WM task performance WM task performance (i.e., percent correct in each task condition) for each participant group and the total sample is listed in Table 5.4. The 2 (load) × 2 (delay) × 2 (group) ANOVA revealed a significant main effect of load, F(1, 52) = 31.166, p < .001, ηp2 = .375, due to greater accuracy in the 4-letter condition than in the 6-letter condition (means = 92.66% correct and 87.37% correct for 4 letters and 6 letters, respectively); all other ps > .06. After removing vari-ance predicted by age, a significant load × group interaction emerged, F(1, 52) = 4.417, p = .040, ηp2 = .078. This interaction appeared to be due to healthy controls showing greater improvement in accuracy at the lower cognitive load (see Figure 5.1). 5.3.2. SCAP task performance SCAP task performance (i.e., percent correct in each task condition) for each participant group and the total sample is listed in Table 5.5. The 4 (load) × 3 (delay) × 2 (group) ANOVA revealed significant main effects of load, F(3, 258) = 44.075, p < .001, ηp2 = .339; and group, F(1, 86) = 31.397, p < .001, ηp2 = .267, which remained significant after removing variance pre- 173 dicted by age and gender (p < .001). Significant interactions emerged for load × delay, F(6, 516)= 3.401, p = .003, ηp2 = .038; and, initially, load × group, F(3, 258) = 2.742, p = .044, ηp2 = .031. However, the load × group interaction did not remain significant after removing vari-ance predicted by age and gender (p > .25). The main effect of load was due to both linear and quadratic effects, with the linear effect having the greater effect size (linear: F(1, 86) = 132.270, p < .001,ηp2 = .606; quadratic: F(1, 86) = 7.103, p = .009, ηp2 = .076; cubic: p > .40). This was due to a general tendency for lower accuracy with higher cognitive load (mean correct = 89.96%, 81.82%, 74.62%, and 72.82% for 1, 3, 5, and 7 dots, respectively; see Figure 5.2). Paired t-tests comparing adjacent load conditions revealed significant differences between 1 versus 3 dots (mean difference = 8.144, SEM = 1.444, t(87) = 5.639, p < .001) and 3 versus 5 dots (mean difference = 7.197, SEM = 1.689, t(87) = 4.262, p < .001), but not 5 versus 7 dots (p > .35). The main effect of group was due to controls having greater mean accuracy than patients (mean correct = 86.27% and 73.34% for controls and patients, respectively). 5.3.3. TSI task performance TSI task performance (i.e., percent correct and mean RT for each task condition) for each participant group and the total sample is listed in Table 5.6. Accuracy and RTs were analysed separately as reported below. 5.3.3.1. Response accuracy Examining percent correct in each task condition, the 2 (stimulus congruency) × 2 (task-switch condition) × 2 (group) ANOVA revealed significant main effects of congruency, F(1, 52) = 89.808, p < .001, ηp2 = .633; task-switch, F(1, 52) = 10.105, p = .002, ηp2 = .163; and  174 group, F(1, 52) = 14.074, p < .001, ηp2 = .213. Significant interactions emerged for congruency × task-switch, F(1, 52) = 15.190, p < .001, ηp2 = .226; task-switch × group, F(1, 52) = 5.021, p = .029, ηp2 = .088; and congruency × task-switch × group, F(1, 52) = 7.736, p = .008, ηp2 = .130. However, neither the task-switch × group interaction nor the congruency × task-switch × group interaction remained significant after removing variance predicted by age (ps > .06). The main effect of congruency was due to greater mean accuracy in the neutral stimulus condition than in the incongruent condition (mean correct = 93.02% and 80.74% for neutral and incongruent stimuli, respectively). The main effect of task-switch was due to greater mean accu-racy following a neutral colour-naming block than following an incongruent colour-naming block (mean correct = 88.89% and 84.88% following neutral and incongruent colour-naming, respectively). The congruency × task-switch interaction was due to a more pronounced congru-ency effect following an incongruent colour-naming block; that is, while accuracy on neutral stimulus trials was similar regardless of whether the preceding colour-naming block consisted of neutral or incongruent stimuli, accuracy on incongruent stimulus trials was much lower when preceded by an incongruent rather than neutral colour-naming block (Figure 5.3). The main ef-fect of group was due to controls having greater mean accuracy compared with patients (mean correct = 91.99% and 82.14% for controls and patients, respectively). 5.3.3.2. Reaction time (RT) Examining mean RTs, the 2 (congruency) × 2 (task-switch) × 2 (group) ANOVA re-vealed significant main effects of congruency, F(1, 52) = 245.043, p < .001, ηp2 = .825; task-switch, F(1, 52) = 18.067, p < .001, ηp2 = .258; and group, F(1, 52) = 5.367, p = .024, ηp2 = .094. A significant congruency × group interaction initially emerged, F(1, 52) = 7.308, p = .009, ηp2 =  175 .123. However, neither the main effect of group nor the congruency × group interaction remained significant after removing variance predicted by age (both ps > .08). The main effect of congruency was due to longer RTs in the incongruent stimulus condi-tion than in the neutral condition (mean RTs = 970.10ms and 1180.25ms for neutral and incon-gruent stimuli, respectively). The main effect of task-switch was due to longer RTs following an incongruent colour-naming block than following a neutral colour-naming block (mean RTs = 1048.42ms and 1102.67ms following neutral and incongruent colour-naming blocks, respective-ly). 5.4. Overview of fMRI-CPCA Results Although the scree plot suggested that either a 4-component or 8-component solution may be appropriate, the less conservative option was selected to allow for the examination of small components that may only be engaged in one task. Therefore, 8 components were extract-ed, which after varimax rotation accounted for 5.83%, 5.80%, 5.55%, 4.83%, 4.41%, 4.21%, 3.08%, and 2.76% of the task-related variance in BOLD signal, respectively. Components 1, 2, 3, 4, 5, and 7 appeared to originate primarily in grey matter and exhibited plausible HDR shapes in all tasks, and therefore were selected for further examination. Figure 5.4 presents a visual sum-mary of components 1-5 and 7 for all tasks, including surface representations and HDR shapes. Components 6 and 8 appeared to reflect artifacts originating from non-cortical sources, and so are not discussed further but are presented in Figures 5.5 and 5.6, respectively. The following sections report the anatomical characteristics and the mixed-model ANOVA results for each of these six components. For the purpose of discussion, each component was assigned a descriptive label following extensive evaluation of its anatomical and functional characteristics; therefore, the following labels are used hereafter to refer to each component: (1) default mode network  176 (DMN), (2) internal attention network, (3) sensorimotor network, (4) motor response network, (5) visual attention network, and (7) occipital network. 5.5. Default Mode Network (Component 1) 5.5.1. Anatomical characteristics The DMN (component 1) was primarily characterized by bilateral deactivation in the me-dial frontal cortex (frontal poles, superior frontal gyri, and anterior paracingulate gyri), middle frontal gyri, medial parietal cortex (posterior cingulate gyri and precuneus), superior lateral oc-cipital cortex, middle temporal gyri, and temporal poles. See Figure 5.7 for anatomical visualiza-tion and Table 5.7 for locations of cluster peaks. 5.5.2. DMN: WM task results DMN (component 1) estimated HDR plots for the WM task are presented in Figure 5.8. The 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group) ANOVA revealed sig-nificant main effects of load, F(1, 52) = 39.858, p < .001, ηp2 = .434; delay, F(1, 52) = 15.108, p < .001, ηp2 = .225; and time, F(9, 468) = 61.987, p < .001, ηp2 = .544. Significant interactions emerged for load × delay, F(1, 52) = 6.654, p = .013, ηp2 = .113; load × time, F(9, 468) = 17.185, p < .001, ηp2 = .248; and delay × time, F(9, 468) = 39.697, p < .001, ηp2 = .433. No significant ef-fects related to group differences emerged (all ps > .05). Although a delay × time × group trend interaction initially appeared to be worth examining (p = .053), this interaction was concluded to be non-significant after removing variance predicted by age (p > .20). Post-hoc analyses of the main effects of load and delay revealed that there was a greater degree of mean deactivation in the 6-letter load condition compared with the 4-letter load condi-tion (mean predictor weights = 0.135 and 0.090, respectively), and a greater degree of mean de-activation in the 4s delay condition compared with the 0s delay condition (mean predictor  177 weights = 0.128 and 0.097, respectively). Delay effects appeared to be driven by the presence of a more sustained HDR in the 4s delay conditions rather than the magnitude of suppression being greater per se (Figure 5.9A). The load × delay interaction was due to the effect of delay length being greater within the 6-letter condition than within the 4-letter condition (Figure 5.9B).  5.5.3. DMN: SCAP task results DMN (component 1) estimated HDR plots for the SCAP task are presented in Figure 5.10. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(3, 258) = 13.372, p < .001, ηp2 = .135; delay, F(2, 172) = 5.770, p = .004, ηp2 = .063; and time, F(9, 774) = 77.063, p < .001, ηp2 = .473. Significant interactions emerged for load × time, F(27, 2322) = 6.174, p < .001, ηp2 = .067; delay × time, F(18, 1548) = 10.187, p < .001, ηp2 = .106; and load × delay × time, F(19.66, 1690.61) = 2.207, p = .002, ηp2 = .025. No significant effects emerged involving group differences (all ps > .09). Post-hoc analyses of the main effect of load revealed that the linear and quadratic con-trasts were both significant, with the linear contrast having the greater effect size (linear: F(1, 86) = 27.824, p < .001, ηp2 = .244; quadratic: F(1, 86) = 14.666, p < .001, ηp2 = .146; cubic: p > .90). This was due to a general tendency for increasing deactivation with increasing cogni-tive load, except for the change from 5 dots to 7 dots (mean predictor weights = 0.041, 0.067, 0.080, and 0.078 for 1, 3, 5, and 7 dots, respectively; see Figure 5.11A). Further examination of differences between adjacent conditions showed that only the contrast between 1 and 3 dots was significant (mean difference = 0.026, SEM = 0.007, t(87) = 3.764, p < .001; other ps > .10). Ex-amination of the main effect of delay revealed that the linear and quadratic contrasts were both  178 significant, with the linear contrast having the greater effect size (linear: F(1, 86) = 6.686, p = .011, ηp2 = .072; quadratic: F(1, 86) = 4.808, p = .031, ηp2 = .053). This was due to a general ten-dency for increasing deactivation with increasing delay duration, except for the change from 3.0s delay to 4.5s delay conditions (mean predictor weights = 0.057, 0.073, and 0.070 for 1.5s, 3.0s, and 4.5s delay, respectively; see Figure 5.11B). Further examination of differences between ad-jacent conditions showed that only the difference between 1.5s delay and 3.0s delay was signifi-cant (mean difference = 0.016, SEM = 0.005, t(87) = 3.304, p = .001; other p > .50). Similar to the pattern observed in the verbal WM task, delay effects appeared to be due to DMN deactiva-tion being more sustained in trials with longer delay periods rather than exhibiting a greater magnitude of deactivation per se (see Figure 5.11C). 5.5.4. DMN: TSI task results DMN (component 1) estimated HDR plots for the TSI task are presented in Figure 5.12. The 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; following neu-tral vs. incongruent colour-naming) × 10 (time) × 2 (group) ANOVA revealed significant main effects of congruency, F(1, 52) = 7.163, p = .010, ηp2 = .121; and time, F(9, 468) = 50.062, p < .001, ηp2 = .491. Significant interactions emerged for congruency × task-switch, F(1, 52) = 16.987, p < .001, ηp2 = .246; congruency × time, F(4.18, 217.18) = 3.746, p = .005, ηp2 = .067; task-switch × time, F(3.88, 201.89) = 2.959, p = .022, ηp2 = .054; and congruency × task-switch × time, F(4.63, 241.01) = 3.722, p = .004, ηp2 = .067. No significant effects emerged involving group differences before or after removing variance predicted by age (all ps > .30). The main effect of congruency was due to there being a greater degree of mean deactiva-tion in the incongruent stimulus condition compared with the neutral stimulus condition (mean  179 predictor weights = -0.010 and 0.030 for neutral and incongruent stimuli, respectively), which emerged during the HDR increase (Figure 5.13A). The congruency × task-switch interaction was due to there being a significant effect of stimulus congruency following an incongruent colour-naming block (mean difference between neutral and incongruent stimuli = 0.090, SEM = 0.021, t(53) = 4.215, p < .001) but not following a neutral colour-naming block (p > .50; see Figure 5.13B). The task-switch × time interaction appeared to be due to there being a less extreme re-sponse in the incongruent task-switch condition as compared to the neutral task-switch condition after averaging across stimulus congruency (i.e., less peak deactivation and less post-peak acti-vation); however, this was a result of the congruency × task-switch × time interaction apparent in Figure 5.12, whereby the HDR peak deactivation is dampened in the neutral condition but not the incongruent condition, and the post-peak rise above baseline is dampened in the incongruent stimulus condition but not the neutral stimulus condition. 5.5.5. DMN: TGT task results DMN (component 1) estimated HDR plots for the TGT task are presented in Figure 5.14. The 2 (condition; generating vs. hearing) × 10 (time) × 2 (group) ANOVA revealed a significant main effect of time, F(9, 522) = 35.652, p < .001, ηp2 = .381. A significant interaction emerged for condition × time, F(3.17, 184.03) = 2.600, p = .050, ηp2 = .043. No significant effects involv-ing group differences emerged before or after removing variance predicted by age (all ps > .25). The condition × time interaction was due to the later HDR peak observed in the thought genera-tion condition than in the hearing condition (Figure 5.14). 5.5.6. Summary of DMN results The DMN (component 1; Figures 5.7 to 5.14) exhibited deactivation that was dependent on cognitive demand in all tasks. In the WM and SCAP tasks, deactivation was greater in condi- 180 tions of higher load. In the TSI task, the DMN exhibited a greater degree of deactivation for in-congruent stimuli than for congruent stimuli, and this effect was magnified when the preceding colour-naming block consisted of incongruent stimuli (Figure 5.13B). In the WM and SCAP tasks, HDRs initiated early in the post-stimulus time series, and peaks were staggered according to delay duration (Figures 5.9A and 5.11C for WM and SCAP, respectively). However, the mag-nitude of the peaks did not continually increase with longer delays, suggesting that the cognitive processes that occur later in the trial (e.g., maintenance and recall) do not rely on further deacti-vation in this network over and above its initial decline from the trial onset. In the TGT task, de-activation peaked later and to a slightly greater degree in the generating condition than in the hearing condition (Figure 5.14), which may reflect the greater cognitive demand required for vo-litional internal thought generation than for passive speech perception. No group differences emerged for this network in any of the tasks examined. 5.6. Internal Attention Network (Component 2) 5.6.1. Anatomical characteristics The internal attention network (component 2) was primarily characterized by bilateral ac-tivation in prefrontal cortex (middle frontal gyri, frontal poles, dorsal paracingulate gyrus), ante-rior insula, parietal regions peaking in posterior supramarginal gyri and right superior parietal lobule, cerebellum (especially right lobe VI), and subcortical regions including bilateral caudate and thalamus. See Figure 5.15 for anatomical visualization and Table 5.8 for coordinates of peak locations. 5.6.2. Internal attention network: WM task results Internal attention network (component 2) estimated HDR plots for the WM task are pre-sented in Figure 5.16. The 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group)  181 ANOVA revealed significant main effects of load, F(1, 52) = 129.511, p < .001, ηp2 = .714; de-lay, F(1, 52) = 59.859, p < .001, ηp2 = .535; and time; F(9, 468) = 52.562, p < .001, ηp2 = .503. Significant interactions emerged for load × delay, F(1, 52) = 9.358, p = .004, ηp2 = .153; load × time, F(9, 468) = 24.996, p < .001, ηp2 = .325; delay × time, F(9, 468) = 84.202, p < .001, ηp2 = .618; delay × time × group, F(3.37, 175.18) = 3.321, p = .017, ηp2 = .060; and load × delay × time, F(4.78, 248.49) = 2.513, p = .033, ηp2 = .046. However, the delay × time × group interaction did not remain significant after removing variance predicted by age (p > .10).  The main effect of load was due to greater engagement of this network in 6-letter condi-tion compared with 4-letter condition (mean predictor weights = 0.043 and 0.119 for 4 letters and 6 letters, respectively). The main effect of delay was due to greater engagement in 4s delay con-dition compared with 0s delay condition (mean predictor weights = 0.053 and 0.109 for 0s and 4s delays, respectively). The load × delay interaction was due to the effect of delay length being more pronounced in the 6-letter condition than in the 4-letter condition (Figure 5.17B). Notably, the delay × time interaction was not only due to the staggered peaks corresponding to delay length, but also there appeared to be a continual increase in activation throughout the 4-second delay (Figure 5.17A). 5.6.3. Internal attention network: SCAP task results Internal attention network (component 2) estimated HDR plots for the SCAP task are presented in Figure 5.18. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(2.66, 229.10) = 6.449, p = .001, ηp2 = .070; delay, F(2, 172) = 16.816, p < .001, ηp2 = .164; and time, F(9, 774) = 65.446, p < .001, ηp2 = .432. Significant interactions emerged for time × group, F(2.97, 255.41) =  182 3.404, p = .019, ηp2 = .038; load × delay, F(6, 516) = 2.184, p = .043, ηp2 = .025; load × time, F(27, 2322) = 3.498, p < .001, ηp2 = .039; load × time × group, F(14.21, 1221.97) = 1.709, p = .047, ηp2 = .019; delay × time, F(18, 1548) = 20.137, p < .001, ηp2 = .190; delay × time × group, F(18, 1548) = 3.230, p < .001, ηp2 = .036; and load × delay × time, F(54, 4644) = 4.131, p < .001, ηp2 = .046. Follow-up analyses of the main effect of load revealed that the linear and quadratic con-trasts were both significant, with the linear contrast having the greater effect size (linear: F(1, 86) = 9.819, p = .002, ηp2 = .102; quadratic: F(1, 86) = 9.422, p = .003, ηp2 = .099; cubic: p > .80). This was due to a general trend for greater activation with greater cognitive load, except for the change from 1 dot to 3 dots (mean predictor weights = 0.048, 0.046, 0.052, and 0.071 for 1, 3, 5, and 7 dots, respectively; see Figure 5.19A). Paired t-tests between adjacent load conditions revealed a significant difference only between the 5-dots and 7-dots conditions (mean differ-ence = 0.020, SEM = 0.006, t(87) = 3.231, p = .002). Examination of the main effect of delay revealed that the linear and quadratic contrasts were both significant, with the linear contrast having the greater effect size (linear: F(1, 86) = 25.341, p < .001, ηp2 = .228; quadratic: F(1, 86) = 4.388, p = .039, ηp2 = .049). This was due to a tendency for increasing activation with increasing delay duration (mean predictor weights = 0.038, 0.059, and 0.065 for 1.5s, 3.0s, and 4.5s delay, respectively; see Figure 5.19B), and further examination of differences between adjacent condi-tions showed that only the difference between 1.5s delay and 3.0s delay was significant (mean difference = 0.022, SEM = 0.004, t(87) = 4.986, p < .001). The delay × time interaction was due to the progressively later and higher peaks with longer delay durations (Figure 5.19C).  183 The time × group and delay × time × group interactions remained significant after remov-ing variance predicted by age and gender (both ps < .05), but the load × time × group interaction did not (p > .10). Post-hoc analysis of the time × group interaction revealed that group differ-ences were largely due to a greater increase and steeper decrease of HDRs in healthy controls as compared to schizophrenia patients (Figure 5.20A). Further, differences between HDR peaks across delay conditions appeared to be more pronounced in controls as compared to patients, as evident from an apparently greater peak in the 4.5s delay condition than in the other conditions being exhibited in healthy controls but not in schizophrenia patients (Figure 5.20B, see HDR shapes from 12-18 seconds in particular). 5.6.4. Internal attention network: TSI task results Internal attention network (component 2) estimated HDR plots for the TSI task are pre-sented in Figure 5.21. The 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; following neutral vs. incongruent colour-naming) × 10 (time) × 2 (group) ANOVA revealed significant main effects of congruency, F(1, 52) = 24.651, p < .001, ηp2 = .322; task-switch, F(1, 52) = 5.209, p = .027, ηp2 = .091; and time, F(9, 468) = 16.938, p < .001, ηp2 = .246. Significant interactions emerged for congruency × time, F(9, 468) = 13.732, p < .001, ηp2 = .209; congruency × time × group, F(3.85, 200.20) = 2.498, p = .046, ηp2 = .046; task-switch × time, F(9, 468) = 6.476, p < .001, ηp2 = .111; and congruency × task-switch × time, F(9, 468) = 6.423, p < .001, ηp2 = .110. The congruency × time × group interaction did not remain significant after removing variance predicted by age (p > .50). The main effect of congruency was due to there being greater mean activity in the incon-gruent condition than in the neutral condition (mean predictor weights = -0.030 and 0.041 for  184 neutral and incongruent stimuli, respectively), due to the minimal response elicited in the neutral stimulus condition (Figure 5.22A). The main effect of task-switch condition was due to there be-ing greater mean activity following an incongruent colour-naming block than following a neutral colour-naming block (mean predictor weights = 0.001 and 0.010 for neutral and incongruent task-switch conditions, respectively). Further, activity following an incongruent colour-naming block was more sustained (Figure 5.22B), particularly during incongruent stimulus trials (Figure 5.21).  5.6.5. Internal attention network: TGT task results Internal attention network (component 2) estimated HDR plots for the TGT task are pre-sented in Figure 5.23. The 2 (condition; generating vs. hearing) × 10 (time) × 2 (group) ANOVA revealed significant main effects of condition, F(1, 58) = 39.113, p < .001, ηp2 = .403; and time, F(9, 522) = 10.800, p < .001, ηp2 = .157. A significant interaction emerged for condition × time, F(9, 522) = 21.131, p < .001, ηp2 = .267. No significant effects involving group differences emerged before or after removing variance predicted by age (all ps > .40). The condition effects were evidently due to the engagement of this network being exclusive to the generating condition (no main effect of time emerged when the hearing condition was examined in isolation; p > .20). 5.6.6. Summary of internal attention network results The internal attention network (component 2; Figures 5.15 to 5.23) was comprised of frontoparietal activity including dorsal paracingulate gyrus, DLPFC, and intraparietal sulcus as well as anterior insula. Activity in this network was dependent on cognitive load and exhibited staggered peaks in the WM and SCAP tasks, and showed greater magnitude of activation with longer delay durations (Figures 5.17A and 5.19C for WM and SCAP, respectively). The internal attention network responded minimally to neutral stimuli in the TSI task (Figure 5.22A), and ex- 185 hibited a more sustained HDR following an incongruent colour-naming block (Figure 5.22B). In the TGT task, this network was engaged during thought generation but not the hearing condition (Figure 5.23). Group differences emerged in the SCAP task which suggested that activity in this network is relatively attenuated in schizophrenia patients (Figure 5.20A), and does not exhibit as pronounced increases with longer delay durations as in healthy controls (Figure 5.20B). Group effects also initially emerged in the WM task and TSI task, but these differences did not remain significant after accounting for age differences. 5.7. Sensorimotor Network (Component 3) 5.7.1. Anatomical characteristics The sensorimotor network (component 3) was primarily characterized by (de)activation in auditory cortex and surrounding areas, including bilateral central opercular cortex, posterior insula, postcentral gyri, SMA/dorsal anterior cingulate, and lingual gyri, as well as left precentral gyrus, left temporal occipital fusiform cortex, and right cerebellum VI. See Figure 5.24 for ana-tomical visualization and Table 5.9 for coordinates of peak locations. 5.7.2. Sensorimotor network: WM task results Sensorimotor network (component 3) estimated HDR plots for the WM task are present-ed in Figure 5.25. The 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(1, 52) = 4.011, p = .050, ηp2 = .072; delay, F(1, 52) = 8.836, p = .004, ηp2 = .145; and time, F(9, 468) = 42.390, p < .001, ηp2 = .449. Signifi-cant interactions emerged for delay × time, F(9, 468) = 66.551, p < .001, ηp2 = .561; and delay × time × group, F(4.13, 214.96) = 2.993, p = .018, ηp2 = .054. However, the delay × time × group interaction was no longer significant after removing variance predicted by age (p > .40).  186 The main effect of load was due to greater mean activation in the 4-letter condition than in the 6-letter condition (mean predictor weights = 0.034 and 0.022 for 4 letters and 6 letters, re-spectively). The main effect of delay was due to greater mean activation in the 4s delay condition than in the 0s delay condition (mean predictor weights = 0.020 and 0.036 for 0s and 4s delay, respectively), which could be due to the 4s delay condition exhibiting a later HDR as well as a slightly higher peak than the 0s delay condition (Figure 5.26B). The delay × time interaction was due to the staggered time courses consistent with delay length, with both the onsets and the peaks occurring later in the 4s delay condition (Figure 5.26A). 5.7.3. Sensorimotor network: SCAP task results Sensorimotor network (component 3) estimated HDR plots for the SCAP task are pre-sented in Figure 5.27. Unlike in the verbal WM task, HDRs in the SCAP task reflected deactiva-tion rather than activation. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(3, 258) = 12.159, p < .001, ηp2 = .124; delay, F(2, 172) = 8.203, p < .001, ηp2 = .087; and time, F(9, 774) = 27.779, p < .001, ηp2 = .244. Significant interactions emerged for load × delay, F(6, 516) = 3.496, p = .002, ηp2 = .039; load × time, F(27, 2322) = 5.167, p < .001, ηp2 = .057; delay × time, F(18, 1548) = 12.217, p < .001, ηp2 = .124; and load × delay × time, F(54, 4644) = 2.408, p < .001, ηp2 = .027. Initially, no significant effects involving group differences emerged (all ps > .05); however, the load × group interaction became significant after removing variance predicted by age and gender, F(2.51, 215.64) = 3.018, p = .039, ηp2 = .034. Post-hoc analyses of the main effect of load revealed that both the linear and quadratic contrasts were significant, with the linear contrast having the greater effect size (linear:  187 F(1, 86) = 20.392, p < .001, ηp2 = .192; quadratic: F(1, 86) = 10.268, p = .002, ηp2 = .107; cubic: p > .07). This was due to a tendency for more extreme negative predictor weights with increas-ing load, except for the change from 3 dots to 5 dots (mean predictor weights = -0.015, -0.047, -0.046, and -0.053 for 1, 3, 5, and 7 dots, respectively; Figure 5.28A). Paired t-tests between ad-jacent conditions revealed that this difference was significant only for the contrast between the 1-dot and 3-dots conditions (mean difference = 0.032, SEM = 0.007, t(87) = 4.560, p < .001; other ps > .20). Post-hoc analyses of the main effect of delay revealed that both the linear and quadrat-ic contrasts were significant, with the quadratic contrast having the greater effect size (linear: F(1, 86) = 6.856, p = .010, ηp2 = .074; quadratic: F(1, 86) = 9.551, p = .003, ηp2 = .100). This was due to a tendency for more extreme negative predictor weights with increasing delay length, ex-cept for the change from 3.0s to 4.5s delay conditions (mean predictor weights = -0.029, -0.049, and -0.042 for 1.5s, 3.0s, and 4.5s delay, respectively; Figure 5.28B). Paired t-tests between ad-jacent conditions revealed that only the change from 1.5s delay to 3.0s delay was significant (mean difference = 0.020, SEM = 0.005, t(87) = 4.001, p < .001; other p > .10). The delay × time interaction was due to the difference in timing of peak deactivation across delay conditions, with largely similar initial HDR decreases but an increasingly delayed return to baseline with increas-ing delay length (Figure 5.28C). The load × group interaction appeared to be primarily due to patients exhibiting greater suppression in the 1-dot condition compared with controls (Figure 5.29), although independent t-tests for each load level were non-significant (all ps > .08). 5.7.4. Sensorimotor network: TSI task results Sensorimotor network (component 3) estimated HDR plots for the TSI task are presented in Figure 5.30. The 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; neutral vs. incongruent colour-naming) × 10 (time) × 2 (group) ANOVA revealed significant  188 main effects of congruency, F(1, 52) = 15.862, p < .001, ηp2 = .234; and time, F(9, 468) = 5.548, p < .001, ηp2 = .096. Significant interactions emerged for time × group, F(4.29, 222.87) = 2.807, p = .024, ηp2 = .051; congruency × task-switch, F(1, 52) = 10.427, p = .002, ηp2 = .167; congruen-cy × time, F(9, 468) = 12.586, p < .001, ηp2 = .195; task-switch × time, F(4.10, 213.27) = 3.683, p = .006, ηp2 = .066; and congruency × task-switch × time, F(9, 468) = 9.354, p < .001, ηp2 = .152. The time × group interaction did not remain significant after removing variance predicted by age (p > .25). The main effect of congruency was due to greater mean activation in trials with neutral stimuli (mean predictor weights = 0.024 and -0.027 for neutral and incongruent stimulus condi-tions, respectively), an effect which emerged early in the HDR but then diminished towards the end of the post-stimulus time series (Figure 5.31A). This congruency effect was much more pro-nounced following incongruent colour-naming blocks than following neutral colour-naming blocks (Figure 5.31B). The task-switch × time interaction was due to a small difference between task-switch conditions in the contrast between the first two post-stimulus time bins (p = .032; change from the first to second time bin = 0.030 and -0.014 following neutral and incongruent colour-naming blocks, respectively). 5.7.5. Sensorimotor network: TGT task results Sensorimotor network (component 3) estimated HDR plots for the TGT task are present-ed in Figure 5.32. The 2 (condition; generating vs. hearing) × 10 (time) × 2 (group) ANOVA re-vealed significant main effects of condition, F(1, 58) = 45.638, p < .001, ηp2 = .440; and time, F(9, 522) = 58.533, p < .001, ηp2 = .502. A significant interaction emerged for condition × time,  189 F(9, 522) = 34.763, p < .001, ηp2 = .375. No significant effects involving group emerged before or after removing variance predicted by age (all ps > .20). The main effect of condition was due to greater engagement of this network in the hear-ing condition than in the generating condition (mean predictor weights = 0.015 and 0.043 for generating and hearing, respectively). Along with peak activation being considerably greater in the hearing condition, the HDR also exhibited greater post-peak suppression than that of the gen-erating condition (Figure 5.32). 5.7.6. Summary of sensorimotor network results The sensorimotor network (component 3; Figures 5.24 to 5.32) largely comprised audito-ry cortex, left-lateralized somatomotor areas, and insula and dorsal anterior cingulate just poste-rior to the insula and dorsal anterior cingulate peaks located in the internal attention network (component 2; Figure 5.15). In the TGT task, this network was engaged during the hearing con-dition but showed little activation in the generating condition (Figure 5.32), which is in line with the activation of auditory cortex. In the WM, SCAP, and TSI tasks, there was a greater degree of activation (or lower degree of suppression) in conditions of lower task difficulty, and in the TSI task, this effect was magnified following an incongruent colour-naming block (Figure 5.31B). A load × group interaction emerged in the SCAP task which appeared to be due to a trend of pa-tients exhibiting greater suppression of this network compared with controls in the 1-dot load condition but not at the higher load conditions (Figure 5.29). Group differences in the WM task and TSI task did not remain significant after accounting for age differences.  190 5.8. Motor Response Network (Component 4) 5.8.1. Anatomical characteristics The motor response network (component 4) was characterized by bilateral (de)activation in somatomotor regions (pre/postcentral gyri and SMA), superior frontal gyri, superior parietal lobules, and lateral occipital cortex. See Figure 5.33 for anatomical visualization and Table 5.10 for coordinates of peak locations. 5.8.2. Motor response network: WM task results Motor response network (component 4) estimated HDR plots for the WM task are pre-sented in Figure 5.34. The 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(1, 52) = 12.471, p = .001, ηp2 = .193; and time, F(9, 468) = 8.360, p < .001, ηp2 = .138. Significant interactions emerged for load × group, F(1, 52) = 4.933, p = .031, ηp2 = .087; time × group, F(2.47, 128.63) = 6.074, p = .001, ηp2 = .105; load × delay × group, F(1, 52) = 5.359, p = .025, ηp2 = .093; load × time, F(9, 468) = 8.047, p < .001, ηp2 = .134; delay × time, F(9, 468) = 67.358, p < .001, ηp2 = .564; delay × time × group, F(3.92, 203.95) = 3.256, p = .013, ηp2 = .059; and a trend-level interaction of load × delay × time, F(5.53, 287.68) = 2.161, p = .052, ηp2 = .040.  Effects of load were driven in part by greater mean activation in the 4-letter condition than in the 6-letter condition (mean predictor weights = 0.041 and 0.021, respectively). Load ef-fects were particularly pronounced at or following HDR peaks, and there appeared to be greater post-response suppression in the 6-letter condition than in the 4-letter condition (Figure 5.34). The delay × time interaction was due to the staggered time courses according to delay length, with both the onsets and the peaks occurring later in the 4s delay condition (Figure 5.35).  191 The load × group interaction was due to the load effect (4 letters > 6 letters) being more pronounced in healthy controls than in schizophrenia patients (mean difference = 0.034 and 0.008 in controls and patients, respectively; see Figure 5.36). This interaction remained after re-moving variance predicted by age, F(1, 52) = 5.919, p = .018, ηp2 = .102. Neither the time × group, load × delay × group, nor the delay × time × group interactions remained significant after removing variance predicted by age (all ps > .08). 5.8.3. Motor response network: SCAP task results Motor response network (component 4) estimated HDR plots for the SCAP task are pre-sented in Figure 5.37. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of delay, F(2, 172) = 14.063, p < .001, ηp2 = .141; and time, F(9, 774) = 49.754, p < .001, ηp2 = .366. Significant interactions emerged for de-lay × group, F(2, 172) = 3.791, p = .024, ηp2 = .042; time × group, F(3.03, 260.45) = 3.206, p = .023, ηp2 = .036; load × time, F(27, 2322) = 2.870, p < .001, ηp2 = .032; delay × time, F(18, 1548) = 15.352, p < .001, ηp2 = .151; and load × delay × time, F(24.99, 2149.38) = 1.913, p = .004, ηp2 = .022. Post-hoc analyses of the main effect of delay revealed that only the linear contrast was significant, F(1, 86) = 29.953, p < .001, ηp2 = .258 (quadratic: p > .90). This effect was due to consistently greater mean activation with longer delay duration (mean predictor weights = 0.044, 0.057, and 0.070 for 1.5s, 3.0s, and 4.5s delay conditions, respectively; Figure 5.38B). Paired t-tests between adjacent delay conditions were both significant (1.5s vs. 3.0s: mean difference = 0.013, SEM = 0.005, t(87) = 2.923, p = .004; 3.0s vs. 4.5s: mean difference = 0.013, SEM = 0.006, t(87) = 2.288, p = .025). However, this effect appeared to be due to the pattern of more  192 sustained HDRs with longer delays rather than greater magnitude of activation per se (Figure 5.38B). Both the delay × group and time × group interactions remained significant after removing variance predicted by age and gender (both ps < .05). Post-hoc analyses of the delay × group in-teraction showed that the groups exhibited different degrees of change from one condition to the next; while patients showed a steeper increase from 1.5s delay to 3.0s delay and little additional increase for 4.5s delay, controls showed a steep increase from 3.0s delay to 4.5s delay (Figure 5.39A); however, independent t-tests for each delay condition were non-significant (all ps > .10). Averaging across task conditions, controls showed a greater HDR increase compared with pa-tients, but a similar post-peak return to baseline (Figure 5.39B). 5.8.4. Motor response network: TSI task results Motor response network (component 4) estimated HDR plots for the TSI task are pre-sented in Figure 5.40. The 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; neutral vs. incongruent colour-naming block) × 10 (time) × 2 (group) ANOVA re-vealed significant main effects of congruency, F(1, 52) = 12.851, p = .001, ηp2 = .198; time, F(9, 468) = 53.319, p < .001, ηp2 = .506; and group, F(1, 52) = 5.040, p = .029, ηp2 = .088. Signifi-cant interactions emerged for time × group, F(9, 468) = 7.271, p < .001, ηp2 = .123; congruency × time, F(9, 468) = 20.166, p < .001, ηp2 = .279; task-switch × time, F(4.64, 241.32) = 2.613, p = .029, ηp2 = .048; and congruency × task-switch × time, F(5.07, 263.64) = 3.950, p = .002, ηp2 = .071. The main effect of congruency was due to greater mean activation in the neutral condi-tion than in the incongruent condition (mean predictor weights = 0.024 and -0.031 for neutral and incongruent stimuli, respectively), which emerged early in the HDR (Figure 5.41A). Alt- 193 hough the initial HDR increase and peak was similar between task-switch conditions, the task-switch × time interaction appeared to be driven by a steeper post-peak return-to-baseline and a greater degree of subsequent suppression in the neutral colour-naming task-switch condition compared with the incongruent colour-naming task-switch condition (Figure 5.41B). The main effect of group did not remain significant after removing variance predicted by age (p > .09). However, the time × group interaction remained significant (p < .05) and was due to patients exhibiting a more gradual return to baseline rather than a greater magnitude of activa-tion per se, as the initial HDR increase and peak was similar between groups (Figure 5.42). 5.8.5. Motor response network: TGT task results Motor response network (component 4) estimated HDR plots for the TGT task are pre-sented in Figure 5.43. Unlike all other tasks in this analysis, this network exhibited deactivation rather than activation in the TGT task. The 2 (condition; generating vs. hearing) × 10 (time) × 2 (group) ANOVA revealed only a significant main effect of time, F(9, 522) = 101.098, p < .001, ηp2 = .635. 5.8.6. Summary of motor response network results The motor response network (component 4; Figures 5.33 to 5.43) primarily comprised bi-lateral somatomotor regions, evidently underlying motor responses (or suppression thereof, in the TGT task). In both the WM and the SCAP task, onsets were staggered – consistent with the timing of button presses – and activity diminished or plateaued midway through the post-stimulus time series before increasing again in conditions with a long delay (Figures 5.35 and 5.38B for WM and SCAP, respectively). In the TSI task, activation was greater for neutral stimu-li than for incongruent stimuli (Figure 5.41A), and post-peak activation was slightly more sus-tained following an incongruent colour-naming block (Figure 5.41B). Group differences  194 emerged in all tasks except TGT. In the WM task, a load × group interaction emerged that was due to controls exhibiting lower mean activation in the 6-letter condition than in the 4-letter con-dition, while patients showed little of this effect and thus exhibited greater activation than con-trols in the 6-letter condition (Figure 5.36). In the TSI task, patients sustained activity in this network considerably longer than did controls (Figure 5.42). In the SCAP task, patients showed less activation during the increase-to-peak phase of the HDR (Figure 5.39B).  5.9. Visual Attention Network (Component 5) 5.9.1. Anatomical characteristics The visual attention network (component 5) was primarily characterized by bilateral (de)activation in occipital cortex (occipital poles, lateral occipital cortex, and occipital fusiform gyri), precentral gyri, middle frontal gyri, SMA/superior frontal gyrus, and thalamus. See Figure 5.44 for anatomical visualization and Table 5.11 for coordinates of peak locations. 5.9.2. Visual attention network: WM task results Visual attention network (component 5) estimated HDR plots for the WM task are pre-sented in Figure 5.45. The 2 (load; 4 letters vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(1, 52) = 13.512, p = .001, ηp2 = .206; delay, F(1, 52) = 68.578, p < .001, ηp2 = .569; and time, F(9, 468) = 361.523, p < .001, ηp2 = .874. Significant interactions emerged for load × time, F(9, 468) = 29.900, p < .001, ηp2 = .365; and delay × time, F(9, 468) = 38.746, p < .001, ηp2 = .427. Initially, no significant effects emerged involving group differences (all ps > .07). Post-hoc analyses revealed that there was greater mean activation in the 6-letter load condition than in the 4-letter load condition (mean predictor weights = 0.076 and 0.090 for 4 let-ters and 6 letters, respectively), and greater mean activation in the 0s delay condition than in the  195 4s delay condition (mean predictor weights = 0.103 and 0.064, respectively). Although the in-crease-to-peak phase of the HDR was the same across delay conditions, the 0s delay condition exhibited a more gradual return to baseline, most likely due to the immediate presentation of the probe stimulus following the encoding period (Figure 5.46). In the 4s delay condition, however, there was no presence of a second peak that would correspond with the presentation of the probe after a delay, suggesting that this activation was more important during the encoding phase of the task. Contrary to the initial ANOVA, the load × time × group interaction became significant only after removing variance predicted by age, F(4.79, 249.08) = 2.656, p = .025, ηp2 = .049. This interaction was due to patients exhibiting less activation in this network and a slightly more gradual return to baseline that was statistically significant from 10 to 12 seconds (see Figure 5.47). 5.9.3. Visual attention network: SCAP task results Visual attention network (component 5) estimated HDR plots for the SCAP task are pre-sented in Figure 5.48. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(3, 258) = 48.089, p < .001, ηp2 = .359; delay, F(2, 172) = 15.610, p < .001, ηp2 = .154; and time, F(9, 774) = 124.160, p < .001, ηp2 = .591. Significant interactions emerged for time × group, F(2.95, 253.39) = 4.311, p = .006, ηp2 = .048; load × delay, F(6, 516) = 3.785, p = .001, ηp2 = .042; load × time, F(27, 2322) = 39.679, p < .001, ηp2  = .316; load × time × group, F(13.20, 1134.91) = 1.854, p = .031, ηp2 = .021; delay × time, F(18, 1548) = 10.548, p < .001, ηp2 = .109; and load × delay × time, F(54, 4644) = 9.550, p < .001, ηp2  = .100.  196 Post-hoc analyses of the main effect of load revealed that the linear and quadratic con-trasts were significant, with the linear contrast having the greater effect size (linear: F(1, 86) = 88.975, p < .001, ηp2 = .509; quadratic: F(1, 86) = 41.258, p < .001, ηp2 = .324; cubic p > .90). This was due to a tendency for greater activation with increasing load (mean predictor weights = -0.020, 0.008, 0.023, and 0.023 for 1, 3, 5, and 7 dots, respectively; Figure 5.49A). Paired t-tests between adjacent load conditions were significant for 1 versus 3 dots (mean difference = 0.028, SEM = 0.004, t(87) = 7.218, p < .001) and 3 versus 5 dots (mean difference = 0.015, SEM = 0.004, t(87) = 3.359, p = .001), but not 5 versus 7 dots (p > .90). Post-hoc analyses of the main effect of delay revealed a significant quadratic contrast, F(1, 86) = 30.480, p < .001, ηp2 = .262 (linear p > .50). This was due to mean activation in the 3.0s delay condition being lower than both the 1.5s delay condition (mean difference = 0.012, SEM = 0.003, t(87) = 4.444, p < .001) and the 4.5s delay condition (mean difference = 0.013, SEM = 0.003, t(87) = 5.074, p < .001; Figure 5.49B). Although the initial HDR increase was the same across delay conditions, the de-lay × time effect appeared to be due to a more gradual return to baseline with shorter delay peri-ods, as observed in the WM task (Figure 5.49C). Both the time × group and load × time × group interactions remained significant after re-moving variance predicted by age and gender (both ps < .05). Across task conditions, the time × group interaction appeared to be due to an attenuated response in schizophrenia patients (i.e., less increase-to-peak and less extreme post-peak suppression; Figure 5.50A). The load × time × group interaction appeared to be due to differences in the post-peak return to baseline phase of the HDR, which may amount to controls exhibiting a steeper decrease below baseline particular-ly for conditions of higher cognitive load (Figure 5.50B).  197 5.9.4. Visual attention network: TSI task results Visual attention network (component 5) estimated HDR plots for the TSI task are pre-sented in Figure 5.51. The 2 (stimulus congruency: neutral vs. incongruent) × 2 (task-switch condition: following neutral vs. incongruent colour-naming block) × 10 (time) × 2 (group) ANOVA revealed significant main effects of congruency, F(1, 52) = 96.571, p < .001, ηp2 = .650; task-switch, F(1, 52) = 4.218, p = .045, ηp2 = .075; and time, F(9, 468) = 39.661, p < .001, ηp2 = .433. Significant interactions emerged for congruency × time, F(9, 468) = 82.180, p < .001, ηp2 = .612; task-switch × time, F(9, 468) = 5.007, p < .001, ηp2 = .088; and congruency × task-switch × time, F(9, 468) = 32.732, p < .001, ηp2 = .386. No significant effects emerged involving group differences before or after removing variance predicted by age (all ps > .10). The main effect of congruency was due to greater mean activation in trials with neutral stimuli than with incongruent stimuli (mean predictor weights = 0.032 and -0.055 for neutral and incongruent stimuli, respectively). Further, activation in the neutral stimulus condition was somewhat sustained throughout the post-stimulus time series, whereas activity in the incongruent condition was brief and was followed by a substantial suppression below baseline (Figure 5.52A). The main effect of task-switch was due to greater mean activation following incongruent colour-naming blocks than following neutral colour-naming blocks (mean predictor weights = -0.012 and -0.009 for neutral and incongruent colour-naming, respectively), and the peak activa-tion was followed by greater post-peak suppression in the incongruent task-switch condition (Figure 5.52B). Differences in HDR shapes between stimulus congruency conditions were par-ticularly pronounced following an incongruent colour-naming block (Figure 5.51).  198 5.9.5. Visual attention network: TGT task results Visual attention network (component 5) estimated HDR plots for the TGT task are pre-sented in Figure 5.53. The 2 (condition; generating vs. hearing) × 10 (time) × 2 (group) ANOVA revealed significant main effects of time, F(9, 522) = 302.867, p < .001, ηp2 = .839; and group, F(1, 58) = 4.926, p = .030, ηp2 = .078. Significant interactions emerged for time × group, F(2.62, 151.71) = 6.259, p = .001, ηp2 = .097; and condition × time, F(3.50, 202.92) = 2.995, p = .025, ηp2 = .049. The condition × time interaction was due to the steeper HDR increase and earlier peak in the generating condition than in the hearing condition (Figure 5.53). Both the main effect of group and the time × group interaction remained significant after removing variance predicted by age (both ps < .05). The main effect of group was due to greater mean activation in schizophre-nia patients than in healthy controls (mean predictor weights = 0.057 and 0.081 for controls and patients, respectively). Patients also exhibited a steeper HDR increase and decrease compared with controls (Figure 5.54). 5.9.6. Summary of visual attention network results The visual attention network (component 5; Figures 5.44 to 5.54) comprised activation in visual cortex, SMA, precentral gyri, and thalamus, but it did not appear to underlie primary visu-al perception per se. A substantial effect of cognitive load emerged in the WM task, despite the fact that the number of items on-screen was the same regardless of whether 4 or 6 letters were displayed, as the 4-letter string was flanked by pound (‘#’) signs. Moreover, this network did not exhibit a second peak corresponding to the presentation of the probe stimulus in the 4s delay condition. An effect of task condition emerged in the TGT task (i.e., greater peak in the generat-ing condition; Figure 5.53) despite the fact that stimulus size and duration was identical between  199 conditions, and stimuli were randomly assigned to task conditions across participants. Similarly, the effect of task-switching in the TSI task, in which a greater response and greater post-peak suppression was exhibited following an incongruent colour-naming block (Figure 5.52B), cannot be explained by basic visual perception demands, as the proportion of neutral versus incongruent stimuli was the same regardless of task-switch condition. Group differences in the WM and SCAP tasks amounted to attenuated and/or more gradual responses in patients which, given the early onset of this network, may reflect impaired encoding into memory (Figures 5.47 and 5.50 for WM and SCAP, respectively). However, patients exhibited greater activation than controls in the TGT task (Figure 5.54). 5.10. Occipital Network (Component 7) 5.10.1. Anatomical characteristics The occipital network (component 7) was primarily characterized by widespread bilateral (de)activation in occipital cortex (especially medial areas), including lingual gyrus, intracalcarine cortex, cuneal cortex, and superior lateral occipital cortex, as well as precuneus and right angular gyrus. See Figure 5.55 for anatomical visualization and Table 5.12 for coordinates of peak loca-tions. 5.10.2. Occipital network: WM task results Occipital network (component 7) estimated HDR plots for the WM task are presented in Figure 5.56, reflecting deactivation in this network. The 2 (load; 4 vs. 6 letters) × 2 (delay; 0s vs. 4s) × 10 (time) × 2 (group) ANOVA revealed a significant main effect of time, F(9, 468) = 32.195, p < .001, ηp2 = .382; and a significant interaction emerged for delay × time, F(9, 468) = 14.881, p < .001, ηp2 = .223. No significant effects involving group differences emerged before or after removing variance predicted by age (all ps > .07).  200 While the decrease-to-peak phases of the HDRs were the same between delay conditions, the 0s delay condition showed a more gradual return to baseline (Figure 5.57), mirroring the pat-tern observed in the visual attention network (component 5; Figure 5.46). However, the 0s delay condition also showed a continual increase following peak deactivation, whereas the post-peak increase in the 4s delay condition plateaued from the 12-second post-stimulus time bin onwards (Figure 5.57). 5.10.3. Occipital network: SCAP task results Occipital network (component 7) estimated HDR plots for the SCAP task are presented in Figure 5.58, reflecting activation rather than deactivation in this network. The 4 (load; 1, 3, 5, or 7 dots) × 3 (delay; 1.5s, 3.0s, or 4.5s) × 10 (time) × 2 (group) ANOVA revealed significant main effects of load, F(3, 258) = 17.176, p < .001, ηp2 = .166; delay, F(2, 172) = 6.536, p = .002, ηp2 = .071; and time, F(9, 774) = 22.094, p < .001, ηp2 = .204. Significant interactions emerged for load × delay, F(6, 516) = 5.586, p < .001, ηp2 = .061; load × time, F(27, 2322) = 4.895, p < .001, ηp2 = .054; delay × time, F(18, 1548) = 12.186, p < .001, ηp2 = .124; and load × delay × time, F(54, 4644) = 5.950, p < .001, ηp2 = .065. No significant effects involving group differences emerged before or after removing variance predicted by age and gender (all ps > .10). Post-hoc analyses of the main effect of load revealed significant linear and cubic con-trasts, with the linear contrast having the greater effect size (linear: F(1, 86) = 33.879, p < .001, ηp2 = .283; cubic:  F(1, 86) = 8.301, p = .005, ηp2 = .088; quadratic: p > .80). This was due to a general tendency for greater activation with greater cognitive load (mean predictor weights = 0.001, 0.001, -0.027, and -0.028 for 1, 3, 5, and 7 dots, respectively; see Figure 5.59A), and spe-cifically in the change from 3 to 5 dots (mean difference = 0.027, SEM = 0.006, t(87) = 4.378,  201 p < .001; ps for 1 vs. 3 dots and 5 vs. 7 dots contrasts > .70). Post-hoc analyses of the main effect of delay revealed a significant linear contrast only (linear: F(1, 86) = 10.724, p = .002, ηp2 = .111; quadratic: p > .10). This was due to less activation with longer delay conditions (mean predictor weights = -0.022, -0.011, and -0.008 for 1.5s, 3.0s, and 4.5s delay, respectively; see Figure 5.59B). Paired t-tests between adjacent delay conditions revealed that the 1.5s versus 3.0s delay contrast was significant (mean difference = 0.011, SEM = 0.004, t(87) = 3.183, p = .002; other p > .60). The 1.5s delay condition exhibited continual activation to a later peak as compared with the 3.0s and 4.5s delay conditions, following a slight plateau occurring at around the same time as when peak activation occurs in the 3.0s and 4.5s delay conditions. A possible explanation for this is that the peak in the 1.5s delay condition is actually a secondary response – perhaps elicited by the probe stimulus – that ends up being higher than the first because the delay is too brief for a full return to baseline in the 1.5s delay condition. By contrast, the longer delay conditions ex-hibit a full decline and then a much smaller second peak later on in the post-stimulus time series (Figure 5.59C). 5.10.4. Occipital network: TSI task results Occipital network (component 7) estimated HDR plots are presented in Figure 5.60, re-flecting deactivation in this network. The 2 (stimulus congruency; neutral vs. incongruent) × 2 (task-switch condition; following neutral vs. incongruent colour-naming) × 10 (time) × 2 (group) ANOVA revealed significant main effects of congruency, F(1, 52) = 33.411, p < .001, ηp2 = .391; task-switch, F(1, 52) = 6.649, p = .013, ηp2 = .113; time, F(9, 468) = 71.731, p < .001, ηp2 = .580; and group, F(1, 52) = 4.822, p = .033, ηp2 = .085. Significant interactions emerged for congruency × task-switch, F(1, 52) = 52.272, p < .001, ηp2 = .501; congruency × time, F(9, 468) = 53.568, p < .001, ηp2 = .507; task-switch × time, F(9, 468) = 10.229, p < .001, ηp2 =  202 .164; and congruency × task-switch × time, F(9, 468) = 4.428, p < .001, ηp2 = .078. The main ef-fect of group did not remain significant after removing variance predicted by age (p > .10). The main effect of stimulus congruency was due to greater mean deactivation in the in-congruent condition compared with the neutral condition (mean predictor weights = -0.022 and 0.047 for neutral and incongruent stimuli, respectively). This was driven by the combination of more extreme peak deactivation in the incongruent condition and a much greater post-HDR rise above baseline in the neutral condition (Figure 5.61A). The main effect of task-switch was due to greater mean deactivation following incongruent colour-naming compared with neutral colour-naming blocks (mean predictor weights = 0.008 and 0.017 following neutral and incongruent colour-naming, respectively). Peak deactivation occurred later in the incongruent colour-naming than the neutral colour-naming task-switch condition (Figure 5.61B). Overall, the effect of stimulus congruency was only pronounced following an incongruent colour-naming block rather than a neutral colour-naming block (Figure 5.61C). 5.10.5. Occipital network: TGT task results Occipital network (component 7) estimated HDR plots for the TGT task are presented in Figure 5.62, reflecting activation in this network. The 2 (condition; generating vs. hearing) × 2 (time) × 2 (group) ANOVA revealed a significant main effect of time, F(9, 522) = 39.128, p < .001, ηp2 = .403. Significant interactions emerged for condition × group, F(1, 58) = 11.107, p = .002, ηp2 = .161; and time × group, F(3.89, 225.53) = 2.970, p = .021, ηp2 = .049. Both the condition × group and time × group interactions remained significant after re-moving variance predicted by age (both ps < .05). The condition × group interaction was due to the groups showing opposite effects of task condition (Figure 5.63). While healthy controls showed greater mean activation in the generating condition than in the hearing condition (mean  203 difference = 0.013, SEM = 0.005, t(31) = 2.479, p = .019), schizophrenia patients showed greater mean activation in the hearing condition than in the generating condition (mean difference = 0.012, SEM = 0.005, t(27) = 2.266, p = .032). Although initial activation and HDR peaks were similar between groups, a more prolonged post-peak response emerged in patients (Figure 5.63). 5.10.6. Summary of occipital network results The occipital network (component 7; Figures 5.55 to 5.63) comprised visual cortex (de)activation originating in the medial occipital lobes, which activated in the SCAP task and TGT (Figures 5.58 and 5.62, respectively), but deactivated in the WM task and TSI task (Figures 5.56 and 5.60, respectively). No main effect of cognitive load emerged in the WM task, and alt-hough there was greater activity with increasing load in the SCAP task, this is confounded by there being more complex visual stimuli at high load levels due to the greater number of dots displayed on-screen. A condition × group interaction emerged in the TGT task which reflected contrasting effects of condition between groups; that is, healthy controls exhibited lower mean activation of this network in the hearing condition than in the generating condition, whereas pa-tients exhibited lower mean activation of this network in the generating condition than in the hearing condition (Figure 5.63A). The basis of this effect was in the post-peak return to baseline and suppression below baseline, as patients exhibited more sustained activation following the peak as well as greater post-peak suppression towards the end of the time series (Figure 5.63B). 5.11. Correlations Between Task Performance and HDR Increases/Decreases 5.11.1. WM task performance and HDRs 5.11.1.1. Computations of WM HDR measures Due to the difference in trial lengths and event timing between 0s delay trials and 4s de-lay trials, predictor weights and fMRI task performance were averaged across load condition but  204 not delay condition, resulting in four HDR measures per component (increase-to-peak and re-turn-to-baseline for the 0s delay and 4s delay conditions). Correlations between HDR measures and WM task performance were only examined between the corresponding delay conditions; for example, a correlation between accuracy in the 0s delay condition and HDR measures in the 4s delay condition would not be performed. WAIS scaled digit span was included in the analysis as an additional behavioural WM measure.  For the DMN (component 1; Figure 5.9A), the 0s delay increase-to-peak (ITP) measure was computed by averaging across time bins from 6-10 seconds, and the return-to-baseline (RTB) measure was computed by averaging across time bins from 12-18 seconds. The DMN 4s delay ITP was computed by averaging time bins from 6-14 seconds, and RTB was computed by averaging time bins from 16-20 seconds. For the internal attention network (component 2; Figure 5.17A), the 0s delay ITP was computed by averaging time bins from 6-10 seconds, and RTB was computed by averaging time bins from 12-16 seconds. The internal attention network 4s delay ITP was computed by averaging time bins from 6-14 seconds, and RTB was computed by aver-aging time bins from 16-20 seconds. For the sensorimotor network (component 3; Figure 5.26), the 0s delay ITP was computed by averaging time bins from 10-12 seconds, and RTB was com-puted by averaging time bins from 14-18 seconds. The sensorimotor network 4s delay ITP was computed by averaging time bins from 14-16 seconds, and RTB was computed by averaging time bins from 18-20 seconds. For the motor response network (component 4; Figure 5.35), the 0s delay ITP was computed by averaging time bins from 8-10 seconds, and RTB was computed by averaging time bins from 12-16 seconds. The motor response network 4s delay ITP was com-puted by averaging time bins from 14-16 seconds, and RTB was computed by averaging time bins from 18-20 seconds. For the visual attention network (component 5; Figure 5.46), the 0s de- 205 lay ITP was computed by averaging time bins from 4-8 seconds, and RTB was computed by av-eraging time bins from 10-16 seconds. The visual attention network 4s delay ITP was computed by averaging time bins from 4-8 seconds, and RTB was computed by averaging time bins from 10-16 seconds. Finally, for the occipital network (component 7, Figure 5.57), the 0s delay ITP was computed by averaging time bins from 4-8 seconds, and RTB was computed by averaging time bins from 10-20 seconds. The occipital network 4s delay ITP was computed by averaging time bins from 4-8 seconds, and RTB was computed by averaging time bins from 10-12 seconds. 5.11.1.2. Results  Although no correlations reached a statistical threshold of p < .01, some trend correla-tions emerged in the visual attention network (component 5) with a more liberal threshold of p < .05. Of the fMRI WM task measures, only the 4s delay condition showed a small correlation be-tween the visual attention network ITP and response accuracy (r(52) = .270, p = .048). Interest-ingly, digit span also showed small correlations with component 5 ITP in both the 0s delay and 4s delay conditions (0s delay: r(52) = .307, p = .024; 4s delay: r(52) = .313, p = .021). Although these correlations should be interpreted with caution given the potentially inflated risk of Type I error, it is noteworthy that both the fMRI WM task performance and the out-of-scanner neuro-psychological measure of WM capacity showed the same relationship with the same network underlying this task.  5.11.2. SCAP task performance and HDRs 5.11.2.1. Computations of SCAP HDR measures A similar approach as taken in the WM task analysis above was applied to the SCAP task HDRs; that is, predictor weights and task performance were averaged across load conditions but not delay. This resulted in six HDR measures per component (ITP and RTB for 1.5s, 3.0s, and  206 4.5s delay conditions). WMS digit span was analysed as well, but as standard scores were not available for this dataset, partial correlations between HDR measures and digit span were per-formed while controlling for age. For the DMN (component 1; Figure 5.11C), the 1.5s delay ITP was computed by averag-ing time bins from 6-12 seconds, and the RTB was computed by averaging time bins from 14-20 seconds. The DMN 3.0s delay ITP was computed by averaging time bins from 6-12 seconds, and RTB was computed by averaging time bins from 14-20 seconds. The DMN 4.5s delay ITP was computed by averaging time bins from 6-14 seconds, and RTB was computed by averaging time bins from 16-20 seconds. For the internal attention network (component 2; Figure 5.19C), the 1.5s delay ITP was computed by averaging time bins from 6-10 seconds, and the RTB was com-puted by averaging time bins from 12-18 seconds. The internal attention network 3.0s delay ITP was computed by averaging time bins from 6-12 seconds, and RTB was computed by averaging time bins from 14-18 seconds. The internal attention 4.5s delay ITP was computed by averaging time bins from 6-14 seconds, and RTB was computed by averaging time bins from 16-20 sec-onds. For the sensorimotor network (component 3; Figure 5.28C), the 1.5s delay ITP was com-puted by averaging time bins from 6-8 seconds, and the RTB was computed by averaging time bins from 10-12 seconds. The sensorimotor network 3.0s delay ITP was computed by averaging time bins from 6-10 seconds, and RTB was computed by averaging time bins from 12-14 sec-onds. The sensorimotor network 4.5s delay ITP was computed by averaging time bins from 6-12 seconds, and RTB was computed by averaging time bins from 14-16 seconds. For the motor re-sponse network (component 4; Figure 5.38B), the 1.5s delay ITP was computed by averaging time bins from 6-10 seconds, and the RTB was computed by averaging time bins from 12-16 seconds. The motor response network 3.0s delay ITP was computed by averaging time bins from  207 6-12 seconds, and RTB was computed by averaging time bins from 14-16 seconds. The motor response network 4.5s delay ITP was computed by averaging time bins from 6-14 seconds, and RTB was computed by averaging time bins from 16-20 seconds. For the visual attention network (component 5; Figure 5.49C), the 1.5s delay ITP was computed by averaging time bins from 6-8 seconds, and the RTB was computed by averaging time bins from 10-14 seconds. The visual at-tention network 3.0s delay ITP was computed by averaging time bins from 6-8 seconds, and RTB was computed by averaging time bins from 10-16 seconds. The visual attention network 4.5s delay ITP was computed by averaging time bins from 6-8 seconds, and RTB was computed by averaging time bins from 10-12 seconds. The occipital network (component 7; Figure 5.59C) was not examined due to the ambiguity and inconsistency of HDR shapes across delay condi-tions. 5.11.2.2. Results The correlations between behavioural measures and HDR ITPs/RTBs in the SCAP task were similar to those observed in the WM task. Significant correlations emerged for the visual attention network (component 5), including a correlation between the 1.5s delay ITP and 1.5s delay accuracy (r(86) = .280, p = .008); and between the 3.0s delay ITP and 3.0s delay accuracy (r(86) = .284, p = .007). At a more liberal threshold of p < .05, trend correlations also emerged in the visual attention network between the 4.5s delay ITP and 4.5s delay accuracy (r(86) = .267, p = .012), and between 4.5s delay ITP and digit span while controlling for age (r(85) = .225, p = .036). Trend correlations also emerged for the motor response network (component 4) between the 4.5s delay ITP and 4.5s delay accuracy(r(86) = .260, p = .014); and between the 4.5s delay ITP and digit span while controlling for age (r(85) = .240, p = .025). No correlations emerged for RTBs with either SCAP task accuracy or digit span (all ps > .15).   208 5.11.3. TSI task performance and HDRs 5.11.3.1. Computations of TSI HDR measures TSI task performance measures (accuracy and RTs) and HDR ITP/RTB measures were averaged across all conditions for the following analysis (averaged time courses for all compo-nents are shown in Figure 5.4D). Digit span was also included as a behavioural measure of inter-est for comparison with the above results from the WM task. For the DMN (component 1), the ITP measure was computed by averaging across time bins from 4-8 seconds, and the RTB meas-ure was computed by averaging across time bins from 10-16 seconds. For the internal attention network (component 2), the ITP measure was computed by averaging across time bins from 4-6 seconds, and the RTB measure was computed by averaging across time bins from 8-16 seconds. For the sensorimotor network (component 3), the ITP measure was computed by averaging across time bins from 6-8 seconds, and the RTB measure was computed by averaging across time bins from 10-12 seconds. For the motor response network (component 4), the ITP measure was computed by averaging across time bins from 4-6 seconds, and the RTB measure was com-puted by averaging across time bins from 8-12 seconds. For the visual attention network (com-ponent 5), the ITP measure was computed by averaging across time bins from 4-6 seconds, and the RTB measure was computed by averaging across time bins from 8-12 seconds. Finally, for the occipital network (component 7), the ITP measure was computed by averaging across time bins from 4-8 seconds, and the RTB measure was computed by averaging across time bins from 10-18 seconds. 5.11.3.2. Results With a threshold of p < .01, a significant negative correlation emerged between the motor response network (component 4) RTB and TSI task accuracy (r(52) = -.375, p = .005). At a more  209 liberal threshold of p < .05, trend correlations emerged between task accuracy and the sensorimo-tor network (component 3) ITP (r(52) = .328, p = .015), and a negative trend correlation emerged between accuracy and the internal attention network (component 2) RTB (r(52) = -.290, p = .034). No correlations emerged with mean RT or digit span (all ps > .05). 5.12. Discussion of 4-Task fMRI-CPCA Results Eight components were extracted from the 4-task fMRI-CPCA. Components 1-5 and 7 (i.e., the DMN, internal attention, sensorimotor, motor response, visual attention, and occipital networks) reflected task-related functional brain networks engaged in the four tasks to varying degrees, while components 6 and 8 reflected artifacts arising from non-cortical sources. Results of these networks with respect to each task are discussed below, followed by a discussion of functions that may be attributed to the networks, and then the group differences that emerged. 5.12.1. WM task Most of the networks that emerged in this analysis were similar to those of the WM-TGT multi-experiment fMRI-CPCA in Chapter 3 – with the notable exception of the sensorimotor network, component 3 – and showed correspondingly similar HDR shapes consistent with earlier interpretations. Comparing the series of HDR onsets across networks (Figure 5.4B), activation of the visual attention network (component 5, replicating the “visual attention” network previously observed in Chapter 3) may support encoding of the letter string at the trial onset, and its dimin-ishing intensity could reflect a coordinated shift from external attention to internal attention as the internal attention network (component 2, replicating the “internal attention” network previ-ously observed) becomes engaged, which may be required so as to maintain a mental representa-tion of the stimuli. Components 1-4 (i.e., DMN, internal attention, sensorimotor, and motor response networks) all exhibited staggered peaks consistent with delay length, suggesting that  210 they underlie processes that initiate after or extend beyond the encoding phase of the task. How-ever, HDRs in components 1 and 2 (DMN and internal attention) also showed similar onsets across delay conditions (Figures 5.9A and 5.17A for the DMN and internal attention network, respectively), while HDRs in component 3 and 4 (sensorimotor and motor response) showed staggered onsets that initiated later in the post-stimulus time series (suggesting that they underlie processes that initiate after the probe appears; Figures 5.26 and 5.35 for the sensorimotor and motor response networks, respectively). The visual attention network (component 5) was the most highly correlated with task performance (and with digit span), but only in the 4s delay con-dition; that is, the condition with a greater requirement to encode items into memory store. The internal attention network (component 2) appeared to support and/or coincide with WM mainte-nance, but may not directly underlie the encoding of items into memory, as it was not correlated with task performance or digit span. The occipital network (component 7) displayed a similar time course as the visual attention network, but reflected deactivation rather than activation (Figure 5.4B), and was not dependent on cognitive load. 5.12.2. SCAP task For the most part, the networks exhibited similar patterns in the SCAP task as in the WM task, with the exceptions of the sensorimotor and occipital networks (components 3 and 7; com-pare Figure 5.4B and C). The sensorimotor network was much more suppressed during the SCAP task, with deactivation peaking just before the motor response (component 4) HDR peak. Nevertheless, its initial suppression followed by disinhibition later in the post-stimulus time se-ries is not entirely inconsistent with the pattern observed in the verbal WM task. The occipital network, however, was active in the SCAP task rather than being suppressed, and in the 3.0s and 4.5s delay conditions, exhibited multiple peaks that were consistent with the timing of visual  211 stimuli appearing on the screen (Figure 5.59C). Further, component 7 was more engaged in con-ditions of higher cognitive load in the SCAP task but not the WM task; this may be due to great-er visual processing demands (e.g., searching/saccades) with larger target arrays in the SCAP task. Otherwise, brain activity in the SCAP task exhibited similar patterns as the WM task in the DMN (component 1; early onsets and sustained, load-dependent deactivation), the internal atten-tion network (component 2; more gradual increases, with greater magnitude of load-dependent activation with longer delay duration), the motor response network (component 4; late, staggered onsets and peaks consistent with the timing of motor responses), and the visual attention network (component 5; early onsets and load-dependent peaks, and the most highly correlated with re-sponse accuracy and digit span). 5.12.3. TSI task Remarkably, the networks showed clearly separable HDR shapes in the TSI task despite trials being only 2 seconds in duration (see Figure 5.4D). The DMN (component 1) initiated ear-ly and was somewhat sustained as observed in other tasks, but not to the same degree as the oc-cipital network (component 7), which exhibited a substantial and sustained suppression for most of the post-stimulus time series before rising well above baseline (presumably during the ITI). The internal attention network (component 2) not only showed considerably greater engagement for incongruent stimuli than for neutral stimuli (Figure 5.22A), but it also exhibited a task-switch effect whereby the HDR was more sustained when the preceding colour-naming block consisted of incongruent stimuli than when it consisted of neutral stimuli (Figure 5.22B). In addition, a trend correlation emerged in which the degree of sustained activation in the RTB phase of the HDR was negatively correlated with response accuracy in the TSI task (i.e., higher accuracy cor-related with lower post-peak HDR values), despite activation in this network appearing to extend  212 beyond the button-press responses (based on its HDR timing relative to the motor response net-work, component 4; see Figure 5.4D). It could be the case that performance in the TSI task relies on the brain’s ability to rapidly switch between states from trial to trial, and that a rapid post-peak decline in this network reflects more efficient switching between networks, improving ac-curacy on the subsequent trial. Alternatively, it could be that the HDR for the internal attention network is more sustained following an incorrect response due to an error monitoring/evaluation process, and so individuals who produce more incorrect responses have on average higher RTB activity. Both explanations are possible, but the latter is further supported by the finding of a much more sustained HDR in trials with an incongruent word-reading stimulus following an in-congruent colour-naming block (i.e., the condition with the lowest accuracy; Figure 5.21), which may be explained by a post-response evaluation process such as reflecting on the current task rules. This is also consistent with existing literature linking post-response error processing (but not necessarily response conflict per se) to dorsal anterior cingulate–anterior insula connectivity and/or co-activation (e.g., Bastin et al., 2016; Edwards, Calhoun, & Kiehl, 2012; Hoffmann, Labrenz, Themann, Wascher, & Beste, 2014; Iannaccone et al., 2015). However, this explanation cannot be tested without a more direct comparison of brain activity during correct-response trials versus incorrect-response trials, which would not necessarily be appropriate with the present da-taset given the large disparity in the number of correct versus incorrect responses in these partic-ipants. The sensorimotor network (component 3) exhibited a similar pattern in the TSI task as in the WM task, having relatively late onsets and being more engaged in the (easier) neutral stimu-lus condition than in the incongruent stimulus condition (Figure 5.31A). Although a trend posi-tive correlation emerged between the sensorimotor network ITP and task accuracy, it is important to note that this phase of the HDR occurs late in the post-stimulus time series as a  213 whole as compared to the other networks (Figure 5.4D), and so activation is unlikely to produce correct responses. A possible explanation is that activation of this network is greater following correct responses. Unlike in the WM and SCAP tasks, the motor response network (component 4) initiated and peaked fairly early in the TSI task, which is consistent with participants having to make more rapid responses. The motor response network also exhibited a significant negative correlation between degree of sustained activation in the RTB phase of the HDR and response accuracy, which, as mentioned above with respect to the similar (trend) relationship that was ob-served between the internal attention network RTB and response accuracy, could indicate that rapid deactivation of this network is important for accuracy on the subsequent trial (e.g., reflect-ing a “reset” of motor response inhibition) or that this network deactivates quicker following a correct response, or some combination of both. The visual attention network (component 5) ex-hibited a similarly early and brief response as in the WM and SCAP tasks, although it was much less engaged in the TSI task – especially for incongruent stimuli – and was not correlated with task performance. 5.12.4. TGT task Although no overt behavioural response was recorded in the TGT task, variability in shapes and timing between networks emerged as observed in the other three tasks. The DMN (component 1) peaked slightly earlier in the hearing condition than in the generating condition (Figure 5.14), perhaps because deactivation of DMN is not as important for passive speech per-ception. The internal attention network (component 2) was engaged in the generating condition, but showed a completely flat HDR in the hearing condition (i.e., neither activated nor deactivat-ed in response to the trial onset; Figure 5.23). This was the only network that exhibited such a striking difference between conditions, demonstrating that it does not support any bottom-up,  214 externally-oriented attention processes that would be involved in the hearing condition. Con-versely, the sensorimotor network (component 3) was considerably more engaged in the hearing condition than in the generating condition – presumably underlying auditory perception – alth-ough it may play some role in attention as well, considering it was not completely suppressed in the generating condition (Figure 5.32). The motor response network (component 4) was sup-pressed for both conditions, consistent with the notion of it underlying motor responses (which are not involved in the TGT task), although it is noteworthy that it was actively suppressed be-low baseline rather than showing a flat HDR (Figure 5.43). This could be a result of participants being instructed to remain still during scanning, and this sample in particular may have been primed to respond, as these participants completed other tasks during the same scanning session. In summary, the DMN, internal attention, and visual attention networks showed greater engage-ment in the generating condition to some degree, although this was relatively slight in the DMN and visual attention network, and seemed to be more due to differences in timing of peaks rather than magnitude per se (with the generating condition peaking later in the DMN and earlier in the visual attention network). The sensorimotor network showed considerably more engagement in the hearing condition than in the generating condition, and the motor response network and oc-cipital network (component 7) showed no effects of condition. Like the SCAP task, the occipital network was activated rather than deactivated in the TGT, although its response was relatively brief compared with the other networks (Figure 5.4E). 5.12.5. Task-related network functions Deactivation in the DMN (component 1; Figures 5.7 to 5.14), comprising anterior medial frontal cortex, medial parietal cortex (posterior cingulate gyri and precuneus), superior lateral occipital cortex, middle temporal gyri, and temporal poles, is widely observed in task-state fMRI  215 literature (Raichle, 2015), and so this terminology is used for consistency with fMRI convention. As expected for the DMN, the degree of deactivation tended to be greater with more demanding conditions in all tasks, and HDRs initiated early and were usually more sustained than other net-works.  The internal attention network (component 2; Figures 5.15 to 5.23) consisted of connec-tivity typically associated with WM, including the DLPFC, dorsal anterior cingulate cortex, ante-rior insula, and intraparietal sulcus (Rottschy et al., 2012). Findings from the WM and SCAP tasks are consistent with this characterization in that it was more engaged with longer mainte-nance periods, and was dependent on cognitive load. However, the TSI results are not necessari-ly consistent with such a characterization, as a WM network would not be expected to remain activated following the button-press response, which is what was exhibited in the TSI task (Figure 5.4D, compare components 2 and 4). This pattern is consistent with a more abstract pro-cess akin to error monitoring and/or keeping in mind the current task rules (i.e., focus on the word rather than the font colour), especially considering that activation was much greater for in-congruent stimuli than for neutral stimuli. Further, this network was engaged during the thought generation condition (but not hearing condition) in the TGT task, despite no requirement to maintain or manipulate information in mind. It seems likely that this network underlies internal-ly-oriented attention processes across all four tasks rather than WM per se, and as such is re-ferred to as the “internal attention” network. Although its role could be even more abstract, this may be an appropriate interpretation based on the findings in this particular set of tasks. The sensorimotor network (component 3; Figures 5.24 to 5.32) was primarily comprised of auditory perception regions, particularly responding to the hearing condition in the TGT; as such, it is surprising that this network displayed such marked responses in the WM and TSI tasks  216 (Figures 5.25 and 5.30, respectively). As the onsets were quite late, it is possible that the activa-tion that emerged in these tasks reflects disinhibition of attention to extraneous stimuli (e.g., awareness of scanner noise), and/or a salience-orienting process with the appearance of the ITI fixation image. Its connectivity between posterior insula and the sensorimotor regions observed here, as well as its functional differentiation from an anterior insula-DLPFC network (i.e., the internal attention network), replicates patterns of insula connectivity observed in resting-state research (Cauda et al., 2011). Considering the broad connectivity of auditory cortex with soma-tomotor and anterior cingulate cortex, this network is referred to as the “sensorimotor” network rather than strictly “auditory”. The motor response network (component 4; Figures 5.33 to 5.43) is referred to as such largely due to its somatomotor activity and the timing of HDR onsets, which were consistent with button-press responses in the WM, SCAP, and TSI tasks, and were suppressed during the TGT task (which does not involve a motor response). Moreover, this network did not show greater engagement in more demanding task conditions, and in fact tended to show the opposite effect where differences between conditions emerged (e.g., greater mean activation in the 4-letter condition than in the 6-letter condition in the WM task). In the visual attention network (component 5; Figures 5.44 to 5.54), which resembled the previously observed visual attention network in Chapter 3 (Figure 3.10), a particularly notable finding was that relationships between HDR ITP and task performance/digit span were consistent between the WM and SCAP tasks. That is, in both tasks, the ITP phase of the HDR was positive-ly correlated with response accuracy and with digit span measured in a separate testing session. Although some of these correlations were small (only trend relationships in the WM task, and for digit span in both tasks), it is quite compelling that these relationships emerged in non- 217 overlapping datasets and in tasks with different types of stimuli but similar WM paradigms, and yet activity in this network during the TSI task was not correlated with task performance or digit span (which was carried out by the same participants within the same scanning session as the verbal WM task). Although it is likely that this network supports encoding of the memory set in the WM and SCAP tasks, it is given the more generic description of “visual attention” due to its engagement in the other tasks as well (though to a lesser degree in the TSI task; Figure 5.51). In all four tasks, this network was among the earliest to initiate in the post-stimulus time series (Figure 5.4). Consistent with the early HDR, it has been suggested that activation in the superior region of dorsomedial prefrontal cortex (which emerged in this network) underlies energization of the required response set for a cognitive task (Stuss, 2006), and it could be the case that there is a less intense trial-by-trial energization in tasks in which trials are relatively rapid, such as the TSI task.  Finally, the occipital network (component 7; Figures 5.55 to 5.63) comprised visual cor-tex (de)activation originating in the medial occipital lobes, which activated in the SCAP task and TGT (Figure 5.4C and E), but deactivated in the WM task and TSI task (Figure 5.4B and D). Although the observed discrepancy between activation versus suppression across tasks is some-what difficult to explain, one possibility is that activity in this network is dependent on the de-gree to which bottom-up visual processing of stimulus features is tied to the task. In the WM and TSI tasks, in which the stimuli comprise letters or words to be rehearsed/read, the basic visual characteristics are either not relevant or must be actively ignored (in the TSI task). By contrast, in the SCAP task, encoding of basic visual features (namely, spatial location) is directly tied to the task demands, and the dependence on cognitive load may be explained by the greater com-plexity of the visual display and more saccades being required with more dots to encode. Alth- 218 ough visual processing is not a requirement of the TGT task, each trial began with a colourful image that was likely to elicit a visual response regardless of the task instructions, and the re-sponse was brief and peaked early. As activity primarily originated in the occipital lobes with little connectivity to remote brain regions, this network is simply referred to as the “occipital” network. 5.12.6. Group differences in network activity Summaries of the mixed model ANOVA results for group differences in each network are presented in Tables 5.13 (DMN), 5.14 (internal attention), 5.15 (sensorimotor network), 5.16 (motor response network), 5.17 (visual attention network), and 5.18 (occipital network). No sig-nificant group differences emerged in the DMN (component 1) for any of the tasks examined. This is somewhat surprising given previous research suggesting that the DMN may exhibit al-tered connectivity in schizophrenia (Hu et al., 2017). However, Hu and colleagues do note from their review of DMN research in schizophrenia (Hu et al., 2017) that findings of hyper- or hypo-activity in the DMN in schizophrenia vary considerably across task-state studies of medicated patients with chronic schizophrenia, and suggest that connectivity in the DMN may be modulat-ed by treatment and/or symptom severity. Therefore, a more comprehensive analysis of relation-ships with symptom severity, duration of illness, and other clinical factors may be more relevant for the DMN. In the internal attention network (component 2), significant group differences emerged in the SCAP task only, although the WM task participants showed a similar pattern prior to remov-ing variance predicted by age. In the SCAP task, patients exhibited a relatively attenuated re-sponse, which was especially pronounced in the 4.5s delay condition (Figure 5.20). This is consistent with the view that hypoactivity of DLPFC (or networks in which it is connected) con- 219 tributes to or underlies WM deficits in schizophrenia (Barch & Ceaser, 2012; Glausier & Lewis, 2018). However, this network was not correlated with task performance or digit span in either the WM or the SCAP task. Therefore, although there is evidence of impaired engagement of this network in schizophrenia patients as expected (at least in the SCAP task), it may not be the pri-mary basis of WM deficits in schizophrenia.  In the sensorimotor network (component 3), a significant load × group interaction emerged in the SCAP task which appeared to be due to a trend of schizophrenia patients exhibit-ing greater suppression of this network compared with controls in the 1-dot load condition but not at the higher load conditions (Figure 5.29). The greater suppression exhibited by schizophre-nia patients at low overt cognitive load could reflect a compensatory mechanism in line with the inefficiency hypothesis of cognitive deficits in schizophrenia (Liddle et al., 2013; Metzak et al., 2012). In the motor response network (component 4), group differences emerged in all tasks ex-cept TGT. In the WM task, healthy controls exhibited less activation in the 6-letter condition than in the 4-letter condition, while schizophrenia patients showed little of this effect and thus exhibited greater activation than controls in the 6-letter condition (Figure 5.36). In the TSI task, patients sustained activity in this network considerably longer than did controls (Figure 5.42). This effect cannot be explained by differences in reaction times, as mean RT did not differ be-tween groups in any TSI task condition, and RT was not correlated with either the ITP or RTB phase of the HDR. In both the WM and the TSI task, the group differences amounted to a pattern in which patients exhibited greater activation than healthy controls. Somewhat contradictory to this was that in the SCAP task, patients showed less activation during the initial HDR increase (Figure 5.39B). As there was a small (trend) positive correlation between the ITP part of this  220 network and SCAP task accuracy in the 4.5s delay condition, this could be due to controls pro-ducing more responses than patients. However, it may also be noted that the groups’ similarity in the post-peak return to baseline despite patients’ initially attenuated response may suggest that patients sustain activation of the network for longer than would be expected given their initial level of activation. As this response network, and in particular the RTB phase of the HDR, was the most highly correlated with task performance in the TSI task, it is possible that this lack of post-response suppression between task trials could explain performance deficits in schizophre-nia patients, as motor control and response inhibition is directly tied to the TSI task demands. A similar effect has been observed in schizophrenia patients during an auditory oddball task, which also requires increased control over motor responses (Lavigne et al., 2016). In the visual attention network (component 5), which was the most clearly related to WM capacity in both the WM task and SCAP task data, some load-related group effects emerged (see Figures 5.47 and 5.50 for WM and SCAP, respectively). Group differences in the WM and SCAP tasks amounted to attenuated and/or more gradual responses in schizophrenia patients – especially in conditions of high cognitive load – which may reflect impaired encoding into memory, given the early onset of this network. By contrast, patients exhibited greater HDRs than controls in the TGT task, regardless of task condition (Figure 5.54). However, it is unclear how this may relate to behaviour, as there is no overt measure of task performance in the TGT task. Finally, in the occipital network (component 7), a condition × group interaction emerged in the TGT task which reflected contrasting effects of condition between groups. That is, healthy controls exhibited lower mean activation of this network in the hearing condition than in the generating condition, whereas patients exhibited lower mean activation of this network in the generating condition than in the hearing condition (Figure 5.63A). The basis of this effect was in  221 the post-peak return to baseline and suppression below baseline, as patients exhibited more sus-tained activation following the peak as well as greater post-peak suppression towards the end of the time series (Figure 5.63B). Given that the initial HDR increase did not differ between groups, this may reflect differences in the degree to which visual fixation is maintained during the TGT task. 5.12.7. Clinical implications While schizophrenia patients exhibited differences from healthy controls in at least one task in almost all networks examined (except the DMN), the clearest link to WM deficits emerged in the visual attention network. As noted above, this is supported by the replication of its relationships with task performance and digit span in both the verbal WM and the SCAP da-tasets. Given the visual attention network’s early onset in all tasks, the reduced engagement in schizophrenia patients suggests that the source of WM deficits may lie in the energization of processes required for successful encoding of items into temporary store. As such, patients with WM deficits may benefit from treatments that focus on improving initial cognitive energization. In treatments that aim to modulate brain activity directly, the visual attention network (or specif-ic clusters within the network) could be a promising target. The next and final chapter expands on these conclusions and presents comparisons of findings across chapters.   222 5.13. Chapter 5 Tables Table 5.1. Demographic information for healthy controls and schizophrenia patients in the WM and TSI tasks dataset (4-task fMRI-CPCA study). Standard deviations are in parentheses.  Patients (n = 28) Controls (n = 26) All (n = 54) Test for  group  differences p-value for group differences Age 40.39 (10.54) 28.77 (8.20) 34.80 (11.08) t(50.5) = 4.539 < .001 Quick estimated IQ 96.54 (12.12) 101.12 (12.57) 98.74 (12.44) t(52) = 1.363 .179 WAIS scaled digit span 9.14 (2.16) 10.46 (2.30) 9.78 (2.30) t(52) = 2.174 .034 Gender (female/male) 15/13 19/7 34/20 𝜒2(1) = 2.000 .138 Handedness (L/M/R) 4/1/23 1/1/24 5/2/47 𝜒2(2) = 1.750 .417 Socioeconomic status factor score 74.50 (20.36) 55.04 (12.85) 65.13 (19.63) t(46) = 4.232 < .001 Years of education 14.55 (2.53) 16.21 (1.81) 15.35 (2.35) t(52) = 2.750 .008 Highest education completed      Doctorate degree 0 2 2 - - Master’s degree 3 2 5 - - Bachelor’s degree 7 15 22 - - Associate degree 1 1 2 - - Other post-secondary 1 1 2 - - High school diploma 15 5 20 - - High school not completed 1 0 1 - - Age at first diagnosis 25.11 (9.57) - - - - Illness duration in years 20.29 (12.32) - - - - SSPI factor scores      Anxiety (max = 4) 1.32 (1.25) - - - - Depression (max = 4) 1.04 (1.26) - - - - Psychomotor poverty (max = 16) 2.82 (3.01) - - - - Psychomotor excitation (max = 24) 1.96 (1.64) - - - - Disorganization (max = 16) 2.36 (1.45) - - - - Reality distortion (max = 8) 2.71 (2.27) - - - - Poor insight (max = 4) 0.86 (1.18) - - - - SSPI total score 13.14 (6.96) - - - - CPZ equivalent dose (mg) 1,317.85 (1,760.22) - - - - Note. Illness duration unknown for 4 participants. IQ = intelligence quotient; WAIS = Wechsler Adult Intelligence Scale; L = left; M = mixed; R = right; SSPI = Signs and Symptoms of Psychotic Illness; SSPI psychomotor pov-erty = sum of scores for anhedonia, underactivity, flattened affect, and poverty of speech; SSPI psychomotor excita-tion = sum of scores for elated mood, insomnia, overactivity, pressured speech, peculiar behaviour, and irritability/hostility; SSPI disorganization = sum of scores for attentional impairment, disorientation, inappropriate affect, and disordered form of thought; SSPI reality distortion = sum of global scores for delusions and hallucina-tions; CPZ = chlorpromazine (equivalent doses calculated for 26 patients, according to guidelines from the Clinical Handbook of Psychotropic Drugs, 22nd Edition; Procyshyn, Bezchlibnyk-Butler, & Jeffries, 2017).    223 Table 5.2. Demographic information for healthy controls and schizophrenia patients in the SCAP task dataset (4-task fMRI-CPCA study). Standard deviations are in parentheses.  Patients (n = 44) Controls (n = 44) All (n = 88) Test for group differences p-value for group differences Age 37.09 (9.09) 31.61 (8.84) 34.35 (9.33) t(86) = 2.864 .005 WMS raw digit span 22.89 (4.94) 30.52 (5.64) 26.70 (6.52) t(86) = 6.753 < .001 Gender (female/male) 11/33 24/20 35/53 𝜒2(1) = 8.017 .005 Handedness all right all right all right - - Years of education 12.66 (1.83) 15.00 (1.77) 13.83 (2.14) t(86) = 6.108 < .001 Highest education completed      Doctorate degree 0 0 0 - - Master’s degree 0 2 2 - - Bachelor’s degree 5 24 29 - - Associate degree 1 1 2 - - Some college 8 9 17 - - High school 19 6 25 - - High school not completed 8 1 9 - - Other 2 1 3 - - SAPS global scores      Bizarre behaviour 0.98 (1.39) - - - - Delusions 2.64 (1.43) - - - - Hallucinations 2.32 (1.78) - - - - Inappropriate affect 0.41 (0.92) - - - - Formal thought disorder 1.52 (1.53) - - - - SANS global scores      Alogia 0.95 (1.12) - - - - Anhedonia 2.34 (1.48) - - - - Attention 2.11 (1.37) - - - - Avolition 2.77 (1.55) - - - - Blunted affect 1.23 (1.27) - - - - CPZ equivalent dose (mg) 615.95 (827.90) - - - - Note. Education information missing for 1 schizophrenia patient. IQ = intelligence quotient; WMS = Wechsler Memory Scale; SAPS = Scale for the Assessment of Positive Symptoms; SANS = Scale for the Assessment of Negative Symptoms; CPZ = chlorpromazine (equivalent doses calculated for 35 patients, according to guidelines from the Clinical Handbook of Psychotropic Drugs, 22nd Edition; Procyshyn et al., 2017). Max score for all SAPS/SANS measures = 5.    224 Table 5.3. Demographic and clinical information for healthy controls and schizophrenia patients in the TGT dataset (4-task fMRI-CPCA study). Standard deviations are in parentheses.  Patients (n = 28) Controls (n = 32) All (n = 60) Test for group differences p-value for group differences Age 33.54 (9.83) 28.75 (8.58) 30.98 (9.42) t(58) = 2.014 .049 Quick estimated IQ 96.57 (11.28) 97.09 (11.21) 96.85 (11.15) t(58) = 0.180 .858 WAIS scaled digit span 10.50 (2.67) 12.39 (2.88) 11.49 (2.92) t(57) = 2.598 .012 Gender (female/male) 14/14 13/19 27/33 𝜒2(1) = 0.530 .466 Handedness (L/M/R) 1/1/26 3/1/28 4/2/54 𝜒2(2) = 0.811 .667 Socioeconomic status factor score 69.71 (28.42) 65.75 (14.97) 67.60 (22.17) t(40) = 0.662 0.512 Years of education 14.32 (2.55) 15.58 (1.81) 14.99 (2.26) t(58) = 2.220 .030 Highest education completed      Doctorate degree 0 0 0 - - Master’s degree 3 3 6 - - Bachelor’s degree 4 10 14 - - Some post-secondary 15 18 33 - - High school diploma 3 1 4 - - High school not completed 3 0 3 - - Age at first diagnosis 23.26 (6.24) - - - - Illness duration in years 16.65 (12.79) - - - - SSPI factor scores      Anxiety (max = 4) 1.26 (1.10) - - - - Depression (max = 4) 1.00 (1.24) - - - - Psychomotor poverty (max = 16) 4.11 (2.47) - - - - Psychomotor excitation (max = 24) 3.56 (3.53) - - - - Disorganization (max = 16) 2.19 (1.98) - - - - Reality distortion (max = 8) 3.70 (2.52) - - - - Poor insight (max = 4) 1.41 (1.34) - - - - SSPI total score 17.41 (7.96) - - - - CPZ equivalent dose (mg) 731.60 (891.93) - - - - Note. Digit span score is missing for 1 healthy control; age at first diagnosis is missing for 1 patient; illness duration in years is missing for 2 patients; all SSPI scores are missing for 1 patient. IQ = intelligence quotient; WAIS = Wechsler Adult Intelligence Scale; L = left; M = mixed; R = right; SSPI = Signs and Symptoms of Psychotic Illness; SSPI psychomotor poverty = sum of scores for anhedonia, underactivity, flattened affect, and poverty of speech; SSPI psychomotor excitation = sum of scores for elated mood, insomnia, overactivity, pressured speech, peculiar behaviour, and irritability/hostility; SSPI disorganization = sum of scores for attentional impairment, disorientation, inappropriate affect, and disordered form of thought; SSPI reality distortion = sum of global scores for delusions and hallucinations; CPZ = chlorpromazine (equivalent doses calculated for 23 patients, according to guidelines from the Clinical Handbook of Psychotropic Drugs, 22nd Edition; Procyshyn et al., 2017).    225 Table 5.4. Mean Working Memory (WM) task performance for each participant group and full sample (percent correct and mean reaction time for each task condition; standard deviations in parentheses). Reaction time (RT) values are in milliseconds.  Patients  Controls  All Condition % correct RTa  % correct RTa  % correct RTa 4 letters         0s delay 91.33 (8.66) 1,122.56 (149.98)  96.43 (5.71) 981.45 (132.29)  93.78 (7.76) 1,054.62 (157.42) 4s delay 88.90 (9.98) 1,046.31 (127.66)  94.37 (5.73) 965.01 (119.97)  91.53 (8.59) 1,007.16 (129.51) 6 letters         0s delay 86.61 (10.94) 1,177.05 (144.95)  89.29 (8.33) 1,065.36 (124.36)  87.90 (9.77) 1,123.27 (145.50) 4s delay 85.97 (11.04) 1,127.96 (132.87)  87.77 (10.84) 1,031.77 (132.82)  86.84 (10.88) 1,081.64 (140.25) aRT was not analyzed, but is presented here for interest.     226 Table 5.5. Mean Spatial Capacity (SCAP) task performance for each participant group and full sample (percent correct and mean reaction time for each task condition; standard deviations in parentheses). Reaction time (RT) values are in milliseconds.  Patients  Controls  All Condition % correct RTa  % correct RTa  % correct RTa 1 dot           1.5s delay 85.23 (20.40) 1,088.27 (250.33)  98.86 (5.27) 951.87 (194.63)  92.05 (16.32) 1,020.07 (233.24) 3.0s delay 86.36 (17.43) 1,062.80 (299.51)  95.45 (11.14) 921.58 (189.76)  90.91 (15.25) 992.19 (259.19) 4.5s delay 82.95 (21.41) 1,069.30 (330.80)  90.91 (15.34) 1,033.02 (262.89)  86.93 (18.94) 1,051.16 (297.62) 3 dots         1.5s delay 73.86 (25.26) 1,213.70 (310.31)  92.05 (15.97) 1,120.01 (214.87)  82.95 (22.91) 1,166.31 (268.95) 3.0s delay 73.86 (22.85) 1,300.35 (323.55)  81.25 (23.58) 1,195.70 (296.74)  77.56 (23.38) 1,248.02 (313.10) 4.5s delay 78.98 (22.84) 1,202.01 (287.29)  90.91 (15.34) 1,035.26 (250.30)  84.94 (20.25) 1,118.63 (280.69) 5 dots         1.5s delay 61.93 (25.57) 1,365.54 (360.86)  82.95 (20.01) 1,273.61 (296.15)  72.44 (25.15) 1,319.57 (331.43) 3.0s delay 71.02 (26.93) 1,275.69 (313.31)  85.23 (18.14) 1,130.00 (263.30)  78.13 (23.92) 1,202.85 (296.90) 4.5s delay 63.07 (22.55) 1,404.20 (353.65)  83.52 (20.13) 1,241.21 (328.00)  73.30 (23.61) 1,322.70 (348.87) 7 dots         1.5s delay 68.18 (27.68) 1,484.50 (306.39)  82.39 (18.35) 1,403.50 (362.73)  75.28 (24.42) 1,444.00 (336.28) 3.0s delay 62.50 (25.57) 1,412.03 (364.59)  76.70 (19.73) 1,272.11 (311.94)  69.60 (23.81) 1,342.07 (344.59) 4.5s delay 72.16 (21.71) 1,406.74 (304.13)  75.00 (19.44) 1,252.09 (313.23)  73.58 (20.54) 1,329.41 (316.63) aRT was not analyzed, but is presented here for interest.    227 Table 5.6. Mean TSI task performance for each participant group and full sample (percent cor-rect and mean reaction time for each condition; standard deviations in parentheses).  Patients  Controls  All Condition % correct RT  % correct RT  % correct RT Colour-naminga         Neutral 89.17 (11.39) 994.38 (137.18)  95.64 (4.50) 885.54 (96.94)  92.28 (9.29) 941.97 (130.51) Incongruent 65.83 (22.91) 1258.51 (111.25)  87.82 (10.15) 1113.82 (104.24)  76.42 (20.95) 1188.84 (129.44) Word-reading         Neutral (all) 89.17 (9.41) 1020.93 (106.01)  97.18 (3.36) 915.35 (105.97)  93.02 (8.17) 970.10 (117.72) cn-WN 89.05 (11.07) 990.23 (106.29)  97.18 (3.85) 901.07 (104.82)  92.96 (9.29) 947.30 (113.84) ci-WN 89.29 (10.79) 1053.45 (134.20)  97.18 (5.05) 929.49 (122.66)  93.09 (9.34) 993.77 (142.06) Incongruent (all) 75.12 (16.69) 1195.89 (125.76)  86.79 (8.46) 1163.40 (142.52)  80.74 (14.50) 1180.25 (133.82) cn-WI 81.90 (16.91) 1165.11 (141.47)  87.95 (9.98) 1132.77 (146.33)  84.81 (14.21) 1149.53 (143.39) ci-WI 68.33 (19.51) 1228.32 (149.41)  85.64 (11.42) 1193.55 (163.70)  76.67 (18.21) 1211.58 (155.95) aColour-naming trials were not analyzed, but task performance is presented here for interest. RT = reaction time (in milliseconds); cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimu-lus; WI = incongruent word-reading stimulus.     228 Table 5.7. 4-task fMRI-CPCA, default mode network (DMN, component 1): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. DMN (component 1) anatomical regions Cluster volumes BAs MNI coordinates  voxels mm3  x y z Negative loadings              Cluster 1: left hemisphere 4,161 112,347     Frontal pole   9 -18 41 44 Middle frontal gyrus   8 -33 23 50 Cluster 1: right hemisphere       Frontal pole   10 0 56 11 Frontal pole   9 18 41 47        Cluster 2: left hemisphere 1,124 30,348     Posterior cingulate gyrus   23 -3 -46 35        Cluster 3: left hemisphere 819 22,113     Superior lateral occipital cortex   39 -48 -67 35        Cluster 4: right hemisphere 637 17,199     Superior lateral occipital cortex   39 54 -61 29        Cluster 5: left hemisphere 390 10,530     Anterior middle temporal gyrus   21 -57 -7 -16 Posterior middle temporal gyrus   21 -60 -16 -13 Posterior middle temporal gyrus   21 -63 -43 -4        Cluster 6: right hemisphere 275 7,425     Posterior middle temporal gyrus   21 60 -10 -16 Posterior middle temporal gyrus   21 66 -34 -4 Temporal pole   21 51 5 -31        Cluster 7: left hemisphere 173 4,671     Frontal orbital cortex   47 -48 29 -10        Cluster 8: right hemisphere 101 2,727     Frontal orbital cortex   47 45 32 -10 Inferior frontal gyrus, pars triangularis   45 54 26 8 (Table 5.7 continued on next page)    (Table 5.7, continued from previous page) 229 DMN (component 1) anatomical regions Cluster volumes BAs MNI coordinates  voxels mm3  x y z Cluster 9: right hemisphere 76 2,052     Cerebellum crus I   n/a 30 -79 -37        Cluster 10: left hemisphere 60 1,620     Cerebellum crus I   n/a -30 -79 -37   230 Table 5.8. 4-task fMRI-CPCA, internal attention network (component 2): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. Internal attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 5,588 150,876     Insular cortex   47 -30 20 2 Middle frontal gyrus   44 -45 8 32 Middle frontal gyrus   48 -45 26 29 Middle frontal gyrus   6 -33 -1 59 Middle frontal gyrus   6 -30 2 56 Frontal pole   46 -33 50 14 Cluster 1: right hemisphere       Paracingulate gyrus   8 3 20 47 Insular cortex   47 33 23 -1 Middle frontal gyrus   45 45 32 26 Middle frontal gyrus   44 48 14 32 Middle frontal gyrus   8 33 5 56 Frontal pole   46 36 47 17 Inferior frontal gyrus, pars opercularis   48 51 14 11        Cluster 2: left hemisphere 1,008 27,216     Posterior supramarginal gyrus   40 -36 -49 44 Superior lateral occipital cortex   7 -12 -70 50        Cluster 3: right hemisphere 730 19,710     Superior parietal lobule   40 36 -49 44 Posterior supramarginal gyrus   40 45 -43 47 Precuneus cortex   7 12 -67 50        Cluster 4: right hemisphere 138 3,726     Cerebellum VI   n/a 30 -61 -28 Cerebellum VI   n/a 9 -73 -25        Cluster 5: right hemisphere 100 2,700     Caudate   n/a 15 8 8 Thalamus   n/a 12 -10 8 (Table 5.8 continued on next page)    (Table 5.8, continued from previous page) 231 Internal attention network (component 2) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Cluster 6: left hemisphere 96 2,592     Caudate   n/a -12 8 5        Cluster 7: left hemisphere 76 2,052     Cerebellum crus I   n/a -30 -61 -31        Cluster 8: left hemisphere 25 675     Inferior temporal gyrus, temporooccipital part   37 -48 -58 -10        Cluster 9: left hemisphere 24 648     Thalamus   n/a -12 -16 8        Cluster 10: left hemisphere 21 567     Cerebellum crus I   n/a -9 -76 -28        Cluster 11: right hemisphere 10 270     Cerebellum VIIb   n/a 27 -70 -49     232 Table 5.9. 4-task fMRI-CPCA, sensorimotor network (component 3): Clusters for the most ex-treme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. Sensorimotor network (component 3) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 3,501 94,527     Central opercular cortex   48 -54 -22 14 Central opercular cortex   48 -51 -7 5 Postcentral gyrus   4 -54 -19 47 Precentral gyrus   3 -45 -16 56 Postcentral gyrus   6 -33 -28 68        Cluster 2: right hemisphere 3,043 82,161     Central opercular cortex   42 60 -19 17 Central opercular cortex   48 54 -7 5 Insular cortex   48 39 5 5 Insular cortex   48 39 -16 2 Postcentral gyrus   3 57 -13 41        Cluster 3: bilateral 481 12,987     Supplementary motor area   6 0 -4 47        Cluster 4: right hemisphere 453 12,231     Cerebellum VI   n/a 21 -52 -19 Lingual gyrus   18 15 -58 2        Cluster 5: left hemisphere 292 7,884     Lingual gyrus   18 -15 -55 -4 Temporal occipital fusiform cortex   37 -21 -55 -19        Cluster 6: left hemisphere 38 1,026     Thalamus   n/a -12 -19 5        Cluster 7: right hemisphere 8 216     Thalamus   n/a 9 -16 5   233 Table 5.10. 4-task fMRI-CPCA, motor response network (component 4): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. Motor response network (component 4) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 7,593 205,011     Superior parietal lobule   40 -33 -40 56 Precentral gyrus   6 -27 -10 62 Precentral gyrus   6 -54 2 35 Superior lateral occipital cortex   19 -24 -73 35 Cluster 1: right hemisphere       Postcentral gyrus   40 36 -37 53 Postcentral gyrus   3 42 -31 50 Superior frontal gyrus   6 27 -7 59 Supplementary motor area   6 0 -7 56        Cluster 2: right hemisphere 94 2,538     Inferior lateral occipital cortex   37 51 -64 2        Cluster 3: left hemisphere 90 2,430     Inferior lateral occipital cortex   37 -48 -70 5        Cluster 4: right hemisphere 39 1,053     Precentral gyrus   6 54 5 32     234 Table 5.11. 4-task fMRI-CPCA, visual attention network (component 5): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neurological Institute (MNI) coordinates are listed for each cluster peak. Visual attention network (component 5) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Positive loadings              Cluster 1: left hemisphere 6,720 181,440     Occipital pole   18 -27 -91 8 Occipital pole   18 -18 -91 -4 Inferior lateral occipital cortex   19 -36 -82 -7 Superior lateral occipital cortex   7 -21 -61 50 Cluster 1: right hemisphere       Superior lateral occipital cortex   18 30 -88 8 Occipital fusiform gyrus   18 21 -88 -4 Inferior lateral occipital cortex   18 33 -85 5 Superior lateral occipital cortex   7 24 -58 53        Cluster 2: left hemisphere 585 15,795     Precentral gyrus   6 -51 -1 47 Precentral gyrus   44 -42 5 29 Precentral gyrus   6 -57 2 23 Middle frontal gyrus   6 -27 -4 50        Cluster 3: right hemisphere 240 6,480     Precentral gyrus   6 54 2 44        Cluster 4: left hemisphere 215 5,805     Supplementary motor area   6 -3 8 59        Cluster 5: left hemisphere 23 621     Thalamus   n/a -21 -31 -1        Cluster 6: right hemisphere 19 513     Thalamus   n/a 21 -28 -1        Cluster 7: right hemisphere 14 378     Middle frontal gyrus   6 30 -1 50  235 Table 5.12. 4-task fMRI-CPCA, occipital network (component 7): Clusters for the most extreme 10% of component loadings. Anatomical regions, Brodmann areas (BAs), and Montreal Neuro-logical Institute (MNI) coordinates are listed for each cluster peak. Occipital network (component 7) anatomical regions Cluster volumes BAs MNI coordinates voxels mm3  x y z Negative loadings              Cluster 1: left hemisphere 7,816 211,032     Lingual gyrus   17 -6 -79 -1 Intracalcarine cortex   17 -3 -82 2 Intracalcarine cortex   18 -6 -88 11 Superior lateral occipital cortex   19 -36 -79 20 Cluster 1: right hemisphere       Cuneal cortex   18 12 -85 20 Lingual gyrus   17 9 -76 -1 Superior lateral occipital cortex   39 42 -70 20 Precuneus cortex   7 6 -49 47 Angular gyrus   41 48 -46 20   236 Table 5.13. 4-task fMRI-CPCA, default mode network (DMN, component 1): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. No results were statistically significant. Default mode network (component 1) WM taska DF DFerror F p ηp2 Group 1 52 0.038 .845 .001 Load × group 1 52 0.020 .888 .000 Delay × group 1 52 1.660 .203 .031 Time × group 2.62 136.39 1.625 .192 .030 Load × delay × group 1 52 0.258 .613 .005 Load × time × group 3.96 205.81 1.477 .211 .028 Delay × time × group 3.34 173.51 1.384 .247 .026 Load × delay × time × group 4.85 252.01 0.577 .713 .011 SCAP taskb DF DFerror F p ηp2 Group 1 86 0.079 .779 .001 Load × group 2.62 225.46 0.128 .926 .001 Delay × group 2 172 1.081 .342 .012 Time × group 2.52 217.09 0.754 .500 .009 Load × delay × group 6 516 0.846 .534 .010 Load × time × group 12.25 1,053.10 1.117 .342 .013 Delay × time × group 8.33 716.79 1.789 .073 .020 Load × delay × time × group 19.54 1,680.14 1.191 .254 .014 TSI taska DF DFerror F p ηp2 Group 1 52 0.463 .499 .009 Congruency × group 1 52 0.002 .966 .000 Task-switch × group 1 52 0.055 .815 .001 Time × group 4.10 212.94 0.456 .773 .009 Congruency × task-switch × group 1 52 0.908 .345 .017 Congruency × time × group 4.20 218.49 0.336 .862 .006 Task-switch × time × group 3.81 198.00 0.919 .450 .017 Congruency × task-switch × time × group 4.55 236.37 0.185 .959 .004 TGT taska DF DFerror F p ηp2 Group 1 58 0.066 .798 .001 Condition × group 1 58 0.116 .734 .002 Time × group 3.57 206.95 1.351 .255 .023 Condition × time × group 3.17 183.68 0.609 .619 .010 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task.  237 Table 5.14. 4-task fMRI-CPCA, internal attention network (component 2): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. Significant results are presented in bold. Internal attention network (component 2) WM taska DF DFerror F p ηp2 Group 1 52 0.045 .832 .001 Load × group 1 52 0.409 .525 .008 Delay × group 1 52 0.422 .519 .008 Time × group 2.57 133.86 0.589 .598 .011 Load × delay × group 1 52 0.014 .905 .000 Load × time × group 3.90 202.83 0.812 .516 .015 Delay × time × group 3.33 173.16 2.058 .101 .038 Load × delay × time × group 4.81 249.92 0.545 .735 .010 SCAP taskb DF DFerror F p ηp2 Group 1 86 0.476 .492 .006 Load × group 2.66 229.04 2.015 .120 .023 Delay × group 2 172 0.609 .545 .007 Time × group 2.99 257.27 3.052 .029* .034 Load × delay × group 6 516 1.495 .178 .017 Load × time × group 14.18 1,219.05 1.435 .128 .016 Delay × time × group 10.05 864.71 2.272 .012* .026 Load × delay × time × group 22.11 1,901.06 0.923 .564 .011 TSI taska DF DFerror F p ηp2 Group 1 52 0.000 .994 .000 Congruency × group 1 52 0.000 .991 .000 Task-switch × group 1 52 0.388 .536 .007 Time × group 3.60 187.10 0.775 .530 .015 Congruency × task-switch × group 1 52 0.263 .611 .005 Congruency × time × group 3.84 199.87 0.834 .501 .016 Task-switch × time × group 4.25 221.08 1.001 .411 .019 Congruency × task-switch × time × group 3.50 181.88 0.283 .866 .005 TGT taska DF DFerror F p ηp2 Group 1 58 0.003 .959 .000 Condition × group 1 58 0.468 .496 .008 Time × group 3.34 193.71 0.748 .538 .013 Condition × time × group 3.65 211.69 0.406 .787 .007 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task; * = p < .05.  238 Table 5.15. 4-task fMRI-CPCA, sensorimotor network (component 3): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out vari-ance predicted by confounding variables. Significant results are presented in bold font. Sensorimotor network (component 3) WM taska DF DFerror F p ηp2 Group 1 52 1.272 .265 .024 Load × group 1 52 0.028 .868 .001 Delay × group 1 52 0.035 .853 .001 Time × group 3.21 166.70 0.183 .918 .003 Load × delay × group 1 52 0.062 .804 .001 Load × time × group 4.98 259.10 1.818 .110 .034 Delay × time × group 4.15 215.78 0.979 .422 .018 Load × delay × time × group 5.60 291.45 0.351 .899 .007 SCAP taskb DF DFerror F p ηp2 Group 1 86 0.736 .393 .008 Load × group 2.51 215.64 3.018 .039* .034 Delay × group 2 172 0.755 .471 .009 Time × group 2.71 233.16 0.453 .696 .005 Load × delay × group 6 516 1.931 .074 .022 Load × time × group 14.12 1,213.96 1.073 .378 .012 Delay × time × group 10.13 871.44 0.903 .531 .010 Load × delay × time × group 25.52 2,194.42 0.890 .623 .010 TSI taska DF DFerror F p ηp2 Group 1 52 0.816 .371 .015 Congruency × group 1 52 0.601 .442 .011 Task-switch × group 1 52 0.539 .466 .010 Time × group 4.36 226.73 1.247 .291 .023 Congruency × task-switch × group 1 52 0.802 .374 .015 Congruency × time × group 4.34 225.65 1.530 .190 .029 Task-switch × time × group 4.10 213.06 0.253 .911 .005 Congruency × task-switch × time × group 4.30 223.50 0.789 .541 .015 TGT taska DF DFerror F p ηp2 Group 1 58 0.972 .328 .016 Condition × group 1 58 0.238 .627 .004 Time × group 3.99 231.70 0.456 .768 .008 Condition × time × group 3.68 213.54 1.227 .301 .021 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task; * = p < .05.  239 Table 5.16. 4-task fMRI-CPCA, motor response network (component 4): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out vari-ance predicted by confounding variables. Significant results are presented in bold font. Motor network (component 4) WM taska DF DFerror F p ηp2 Group 1 52 0.417 .521 .008 Load × group 1 52 5.919 .018* .102 Delay × group 1 52 0.000 .984 .000 Time × group 2.43 126.20 2.433 .081 .045 Load × delay × group 1 52 2.435 .125 .045 Load × time × group 5.22 271.44 1.178 .320 .022 Delay × time × group 3.81 198.20 1.806 .132 .034 Load × delay × time × group 5.47 284.43 1.020 .409 .019 SCAP taskb DF DFerror F p ηp2 Group 1 86 0.865 .355 .010 Load × group 2.67 229.60 0.473 .679 .005 Delay × group 2 172 4.859 .009** .053 Time × group 2.99 257.41 2.651 .049* .030 Load × delay × group 4.88 419.40 0.367 .867 .004 Load × time × group 15.68 1,348.36 0.793 .693 .009 Delay × time × group 10.83 931.28 1.360 .188 .016 Load × delay × time × group 24.69 2,122.91 0.877 .638 .010 TSI taska DF DFerror F p ηp2 Group 1 52 2.862 .097 .052 Congruency × group 1 52 0.055 .815 .001 Task-switch × group 1 52 1.230 .272 .023 Time × group 4.23 219.89 3.203 .012* .058 Congruency × task-switch × group 1 52 0.168 .683 .003 Congruency × time × group 5.70 296.41 0.723 .624 .014 Task-switch × time × group 4.57 237.47 0.471 .781 .009 Congruency × task-switch × time × group 5.14 267.14 0.792 .559 .015 TGT taska DF DFerror F p ηp2 Group 1 58 0.955 .333 .016 Condition × group 1 58 0.427 .516 .007 Time × group 3.33 192.87 0.596 .636 .010 Condition × time × group 3.62 210.22 1.520 .202 .026 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task; * = p < .05; ** = p < .01.  240 Table 5.17. 4-task fMRI-CPCA, visual attention network (component 5): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out vari-ance predicted by confounding variables. Significant results are presented in bold font. Visual attention network (component 5) WM taska DF DFerror F p ηp2 Group 1 52 0.107 .744 .002 Load × group 1 52 0.622 .434 .012 Delay × group 1 52 0.238 .628 .005 Time × group 2.49 129.59 0.893 .431 .017 Load × delay × group 1 52 0.850 .361 .016 Load × time × group 4.79 249.08 2.656 .025* .049 Delay × time × group 3.95 205.52 0.558 .692 .011 Load × delay × time × group 5.26 273.57 0.759 .587 .014 SCAP taskb DF DFerror F p ηp2 Group 1 86 3.055 .084 .034 Load × group 2.60 223.55 0.106 .940 .001 Delay × group 2 172 2.242 .109 .025 Time × group 2.97 255.12 4.246 .006** .047 Load × delay × group 6 516 0.211 .973 .002 Load × time × group 13.11 1,127.70 1.890 .027* .022 Delay × time × group 8.55 734.89 0.851 .564 .010 Load × delay × time × group 21.60 1,857.23 1.349 .130 .015 TSI taska DF DFerror F p ηp2 Group 1 52 0.236 .629 .005 Congruency × group 1 52 0.072 .789 .001 Task-switch × group 1 52 0.612 .438 .012 Time × group 4.43 230.17 0.592 .686 .011 Congruency × task-switch × group 1 52 0.019 .891 .000 Congruency × time × group 4.56 237.08 0.391 .838 .007 Task-switch × time × group 5.16 268.08 0.574 .725 .011 Congruency × task-switch × time × group 3.84 199.69 0.825 .507 .016 TGT taska DF DFerror F p ηp2 Group 1 58 4.362 .041* .070 Condition × group 1 58 0.583 .448 .010 Time × group 9 522 7.025 <.001*** .108 Condition × time × group 3.49 202.36 1.017 .393 .017 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task; * = p < .05; ** = p < .01; *** = p < .001.  241 Table 5.18. 4-task fMRI-CPCA, occipital network (component 7): Mixed model ANOVA results for effects/interactions involving group differences for each task after regressing out variance predicted by confounding variables. Significant results are presented in bold font. Occipital network (component 7) WM taska DF DFerror F p ηp2 Group 1 52 0.235 .630 .004 Load × group 1 52 0.544 .464 .010 Delay × group 1 52 0.461 .500 .009 Time × group 2.83 147.00 0.753 .515 .014 Load × delay × group 1 52 1.623 .208 .030 Load × time × group 6.13 318.98 0.194 .980 .004 Delay × time × group 4.50 233.82 0.698 .610 .013 Load × delay × time × group 6.69 348.06 0.772 .606 .015 SCAP taskb DF DFerror F p ηp2 Group 1 86 0.003 .958 .000 Load × group 2.65 227.90 1.470 .227 .017 Delay × group 2 172 0.620 .539 .007 Time × group 3.23 278.19 0.245 .878 .003 Load × delay × group 5.09 437.62 0.859 .510 .010 Load × time × group 15.69 1,349.10 1.163 .293 .013 Delay × time × group 11.72 1,007.67 0.900 .544 .010 Load × delay × time × group 26.50 2,278.65 1.254 .174 .014 TSI taska DF DFerror F p ηp2 Group 1 52 2.491 .121 .046 Congruency × group 1 52 0.140 .710 .003 Task-switch × group 1 52 0.844 .363 .016 Time × group 4.40 228.71 1.198 .312 .023 Congruency × task-switch × group 1 52 0.427 .517 .008 Congruency × time × group 5.57 289.56 0.613 .708 .012 Task-switch × time × group 5.35 278.34 0.577 .729 .011 Congruency × task-switch × time × group 5.86 304.70 0.531 .780 .010 TGT taska DF DFerror F p ηp2 Group 1 58 0.263 .610 .005 Condition × group 1 58 9.122 .004** .136 Time × group 3.87 224.31 2.474 .047* .041 Condition × time × group 3.40 197.08 0.862 .473 .015 aStatistical results after removing variance predicted by age (WM, TSI, and TGT tasks). bStatistical results after removing variance predicted by age and gender (SCAP task). DF = degrees of freedom; WM = Working Memory; SCAP = Spatial Capacity; TSI = Task-Switch Inertia; TGT = Thought Generation Task; * = p < .05; ** = p < .01.  242 5.14. Chapter 5 Figures Figure 5.1. WM task performance: Percentage of correct responses for each load condition, aver-aged over delay to illustrate load × group interaction. Asterisk indicates significant interaction after removing variance predicted by age. * = p < .05.   Figure 5.2. SCAP task performance: Percentage of correct responses for each load condition, av-eraged over delay to illustrate main effect of load. Asterisks indicate significant paired t-tests be-tween adjacent load conditions. *** = p < .001.   75808590951004 letters 6 lettersPercent CorrectLoad  GroupControlsPatients*60657075808590951001 dot 3 dots 5 dots 7 dotsPercent Correct ****** 243 Figure 5.3. TSI task performance: Percentage of correct responses for each task condition, illus-trating significant congruency × task-switch interaction. WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus; cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; *** = p < .001.  707580859095100cn ciPercent CorrectCongruency  Task-SwitchWNWI*** 244 Figure 5.4. 4-task fMRI-CPCA: Summary of components 1-5 and 7. A (top): surface representa-tions of top 10% loadings for each component. B-E (bottom): predictor weights plotted over post-stimulus time for each task and component (components 1 and 7 weights have been multi-plied by -1 so that values below the X axis reflect deactivation and values above the X axis re-flect activation for all components). DMN = default mode network; C = component.   245 Figure 5.5. 4-task fMRI-CPCA, blood flow artifact (component 6): Anatomical and temporal characteristics. Many peak areas were outside of the brain, likely reflecting blood drainage after a neural response. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.14, max = 0.28; no negative loadings above threshold). Images displayed in neurological orien-tation (left is left) with MNI coordinates. B (bottom): HDR plots for each task. 4L = 4 letters; 6L = 6 letters; cn = task-switch from neutral colour-naming; ci = task-switch from incongruent colour-naming; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.    246 Figure 5.6. 4-task fMRI-CPCA, artifact (component 8): Anatomical and temporal characteristics. A (top): dominant 10% of loadings (red/yellow = positive loadings, min = 0.12, max = 0.23; no negative loadings above threshold). Images are displayed in neurological orientation (left is left) with MNI coordinates. B (bottom): HDR plots for each task. 4L = 4 letters; 6L = 6 letters; cn = task-switch from neutral colour-naming; ci = task-switch from incongruent colour-naming; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.   247 Figure 5.7. 4-task fMRI-CPCA, default mode network (DMN, component 1): Dominant 10% of component loadings (blue/green = negative loadings, min = -0.33, max = -0.18; no positive load-ings above threshold). Images are displayed in neurological orientation (left is left) with MNI coordinates.    248 Figure 5.8. WM task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots. Y axis is reversed (positive down, negative up) to facilitate interpretation (i.e., values below X axis reflect deactivation, and values above X axis reflect activation). 4L = 4-letter condition; 6L = 6-letter condition.  -.20-.10.00.10.20.30.40Predictor WeightsPost-Stimulus Time (Seconds)4L (0s delay)4L (4s delay)6L (0s delay)6L (4s delay)2        4         6         8        10       12       14       16        18      20 249 Figure 5.9. WM task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating ef-fects of delay and load. Y axis orientation is reversed to facilitate interpretation (i.e., values be-low X axis reflect deactivation, and values above X axis reflect activation). A (top): predictor weights plotted over post-stimulus time for each delay condition (asterisks indicate significant 0s vs. 4s delay × time contrasts between adjacent time bins). B (bottom): mean predictor weights illustrating significant load × delay interaction. * = p < .05; *** = p < .001.    250 Figure 5.10. SCAP task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for all task conditions (each load level displayed on a separate graph). Y axis is reversed (positive down, negative up) to facilitate interpretation of HDR shapes (values below X axis re-flect deactivation, and values above X axis reflect activation).    251 Figure 5.11. SCAP task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating effects of cognitive load and delay length. Y axes are reversed (negative up, positive down) to facilitate interpretation (i.e., values below X axis reflect deactivation and values above X axis reflect activation). A (top left): mean predictor weights illustrating main effect of load (asterisks indicate significant paired t-tests between adjacent load conditions). B (top right): mean predic-tor weights illustrating main effect of delay (asterisks indicate significant paired t-tests between adjacent delay conditions). C (bottom): predictor weights averaged over load to illustrate de-lay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins; contrast with the greatest effect size is flagged). a = linear effect of delay; b = quadratic ef-fect of delay; * = p < .05; ** = p < .01; *** = p < .001.     252 Figure 5.12. TSI task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for all word-reading conditions. Y axis is reversed to facilitate interpretation (values above X axis reflect activation, and values below X axis reflect deactivation). cn = task-switch from neu-tral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.   -.30-.20-.10.00.10.20.30Predictor WeightsPost-Stimulus Time (Seconds)cn-WNci-WNcn-WIci-WI2         4         6         8        10       12       14       16       18       20 253 Figure 5.13. TSI task from the 4-task fMRI-CPCA, DMN (component 1): Graphs illustrating ef-fects of stimulus congruency and task-switch condition. Y axis is reversed (negative up, positive down) to facilitate interpretation (values above X axis reflect activation, and values below X axis reflect deactivation). A (top): predictor weights plotted over post-stimulus time, illustrating con-gruency × time interaction (asterisks indicate significant congruency × time contrasts between adjacent time bins). B (bottom): mean predictor weights illustrating significant congruency × task-switch interaction. * = p < .05; *** = p < .001.    254 Figure 5.14. TGT task from the 4-task fMRI-CPCA, DMN (component 1): Estimated HDR plots for both task conditions. Y axis is reversed (negative up, positive down) to facilitate interpreta-tion (values above X axis reflect activation and values below X axis reflect deactivation). Aster-isks indicate significant condition × time contrasts between adjacent time bins. ** = p < .01; *** = p < .001.   -.30-.20-.10.00.10.20.30.40Predictor WeightsPost-Stimulus Time (Seconds)GeneratingHearing2.5        5       7.5       10      12.5      15     17.5      20      22.5      25***** 255 Figure 5.15. 4-task fMRI-CPCA, internal attention network (component 2): Dominant 10% of component loadings (red/yellow = positive loadings, min = 0.18, max = 0.36; no negative load-ings above threshold). Images are displayed in neurological orientation (left is left) with MNI coordinates.   Figure 5.16. WM task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all task conditions. 4L = 4 letters; 6L = 6 letters.  -.10.00.10.20.30.40.502 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)4L (0s delay)4L (4s delay)6L (0s delay)6L (4s delay) 256 Figure 5.17. WM task from the 4-task fMRI-CPCA, internal attention network (component 2): Graphs illustrating effects of cognitive load and delay length. A (top): predictor weights aver-aged over load to illustrate delay × time interaction (asterisks indicate significant delay × time contrasts between adjacent time bins). B (bottom): mean predictor weights illustrating signifi-cant load × delay interaction. ** = p < .01; *** = p < .001.    257 Figure 5.18. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all task conditions (each load level is presented on a separate graph).    258 Figure 5.19. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Graphs illustrating effects of cognitive load and delay length. A (top left): mean predictor weights illustrating main effect of load (asterisks indicate significant paired t-tests between adja-cent load conditions). B (top right): mean predictor weights illustrating main effect of delay (as-terisks indicate significant paired t-tests between adjacent delay conditions). C (bottom): predictor weights averaged over load to illustrate delay × time interaction (asterisks indicate sig-nificant delay × time contrasts between adjacent time bins; contrast with the greatest effect size is flagged). a = linear effect of delay; b = quadratic effect of delay; * = p < .05; ** = p < .01; *** = p < .001.   259 Figure 5.20. SCAP task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDRs illustrating group differences. A (top): predictor weights averaged over all task conditions to illustrate group × time interaction (asterisks indicate significant group × time con-trasts between adjacent time bins after removing variance predicted by age). B (bottom): predic-tor weights averaged over load to illustrate delay × time × group interaction (asterisks indicate significant contrasts between adjacent time bins underlying the delay × time × group interaction after removing variance predicted by age, and therefore indicate where the delay × time interac-tion differed between groups). a = significant group difference in the linear effect of delay; b = significant group difference in the quadratic effect of delay; * = p < .05; ** = p < .01.   260 Figure 5.21. TSI task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for all word-reading conditions. cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.  -.30-.20-.10.00.10.20.302 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)cn-WNci-WNcn-WIci-WI 261 Figure 5.22. TSI task from 4-task fMRI-CPCA, internal attention network (component 2): Esti-mated HDRs illustrating stimulus congruency and task-switch effects. Asterisks indicate signifi-cant condition × time contrasts between adjacent time bins. A (top): predictor weights averaged over task-switch to illustrate congruency × time interaction. B (bottom): predictor weights aver-aged over congruency to illustrate task-switch × time interaction. WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus; cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; ** = p < .01; *** = p < .001.   262 Figure 5.23. TGT task from the 4-task fMRI-CPCA, internal attention network (component 2): Estimated HDR plots for both task conditions. Asterisks indicate significant condition × time contrasts between adjacent time bins. ** = p < .01; *** = p < .001.   -.30-.20-.10.00.10.20.30.402.5 5 7.5 10 12.5 15 17.5 20 22.5 25Predictor WeightsPost-Stimulus Time (Seconds)GeneratingHearing************* 263 Figure 5.24. 4-task fMRI-CPCA, sensorimotor network (component 3): Dominant 10% of com-ponent loadings (red/yellow = positive loadings, min = 0.17, max = 0.33; no negative loadings above threshold). Images are displayed in neurological orientation (left is left) with MNI coordi-nates.   Figure 5.25. WM task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Esti-mated HDR plots for all task conditions. 4L = 4 letters; 6L = 6 letters.  -.20-.10.00.10.20.30.402 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)4L (0s delay)4L (4s delay)6L (0s delay)6L (4s delay) 264 Figure 5.26. WM task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Esti-mated HDRs illustrating delay × time interaction (asterisks indicate significant delay × time con-trasts between adjacent time bins). *** = p < .001.   -.20-.10.00.10.20.30.402 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)Delay  Time0s delay4s delay*** 265 Figure 5.27. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Es-timated HDR plots (each load level presented on a separate graph).    266 Figure 5.28. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Graphs illustrating effects of cognitive load and delay length. A (top left): mean predictor weights illustrating main effect of load (asterisks indicate significant paired t-tests between adja-cent load conditions). B (top right): mean predictor weights illustrating main effect of delay (as-terisks indicate significant paired t-tests between adjacent delay conditions). C (bottom): predictor weights averaged over load to illustrate delay × time interaction (asterisks indicate sig-nificant delay × time contrasts between adjacent time bins; contrast with the greatest effect size is flagged). a = linear effect of delay; b = quadratic effect of delay; * = p < .05; ** = p < .01; *** = p < .001.    267 Figure 5.29. SCAP task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Mean predictor weights illustrating load × group interaction, explained by significant difference in quadratic contrast.   Figure 5.30. TSI task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Esti-mated HDR plots for all word-reading conditions. cn = task-switch from neutral colour-naming block; ci = task-switch from incongruent colour-naming block; WN = neutral word-reading stimulus; WI = incongruent word-reading stimulus.  -.08-.07-.06-.05-.04-.03-.02-.01.00.01.021 dot 3 dots 5 dots 7 dotsMean Predictor WeightLoad  GroupControlsPatients-.30-.20-.10.00.10.20.302 4 6 8 10 12 14 16 18 20Predictor WeightsPost-Stimulus Time (Seconds)cn-WNci-WNcn-WIci-WI 268 Figure 5.31. TSI task from the 4-task fMRI-CPCA, sensorimotor network (component 3): Esti-mated HDRs illustrating stimulus congruency and task-switch effects. A (top): predictor weights averaged over task-switch to illustrate congruency × time interaction (asterisks indicate signif