Attention and Inference in Melancholic Depression

Matthew Paul Hyett BSc (Psych); PGDipPsych

A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy in Psychiatry at the University of New South Wales

SCHOOL OF PSYCHIATRY

FACULTY OF MEDICINE

January 2015

PLEASE TYPE THE UNIVERSITY OF NEW SOUTH WALES Thesis/Dissertation Sheet

Surname or Family name: Hyett

First name: Matthew Other name/s: Paul

Abbreviation for degree as given in the University calendar: PhD

School: Psychiatry Faculty: Medicine

Title: Attention and Inference in Melancholic Depression

Abstract 350 words maximum: (PLEASE TYPE) Melancholia has long been positioned as a quintessentially biological depressive condition. Impairments in attention are prominent, particularly in shifting attention away from internal states, but a detailed neurocognitive understanding of these deficits is lacking. This thesis hence sought to clarify the cognitive and neurobiological mechanisms of attention deficits in melancholia. Analytic methods spanning cognitive and brain network modelling were employed to explore the biases and inflexibility of attention in melancholia, compared to non-melancholic depressed and healthy individuals. The first study (Chapter 2) investigates disrupted attentional inference to emotional stimuli across sub-types of depression and healthy participants. I hypothesised that both depressed groups would show impaired discriminability of emotional signals, and that melancholia would be characterised by decreased sensitivity to emotional stimuli. Signal detection data .from an attentional control task were modelled using hierarchical (Bayesian) statistics. Melancholia was associated with disrupted sensitivity of emotional signals, and poorer discriminability of neutral signals, hence likely reflecting distorted attentional inference. The second study (Chapter 3) explores resting state functional brain network effective connectivity across melancholic, non-melancholic and control groups. Interactions between cortical systems corresponding to attention, executive control and interoception - derived from independent component analysis (ICA) - were modelled using dynamic causal modelling (DCM). Analyses supported the hypothesis that relationships amongst networks subserving attention and interoception would be disrupted in melancholia. This study revealed a specific 'dysconnectivity' between brain regions underpinning attention and interoception in melancholia. In the third study (Chapter 4), I advanced an in-scanner naturalistic film viewing paradigm to quantify brain networks underling the shifting of attention from rest to dynamic processing of exogenous emotional stimuli, employing the same groups as Chapter 3. I hypothesised that cortical systems would remain in an "at-rest" state in melancholia, reflecting impaired attentional shifting to exogenous stimuli. Surprisingly, neuronal activity in systems supporting attention and interoception were increased in melancholia compared to controls during negative film viewing. I speculate that these findings reflect ineffective neuronal adaptation during attentional resource allocation to emotional material in melancholia. Preliminary analyses (presented in Chapter 5) highlight that impaired attentional set-shifting performance is associated with disruptions to these neuronal systems, hence pointing to a disorder-specific behavioural analogue of the neurobiological findings. The studies comprising this thesis offer a unique cognitive and neurobiological explanation for attentional deficits in melancholia, and act to explain aspects of its clinical presentation in terms of impaired redirection of attention away from persistent and dysphoric internal states.

Declaration relating to disposition of project thesis/dissertation

I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral theses only). dt,wUJlt. ... . ~ ...... :l!:.t: .. Y. ..'z:g~~-:- .. ~~ ...... <,~~) ...... :.: ...... Witness

The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.

FOR OFFICE USE ONLY Date of completion of requirements for Award:

THIS SHEET IS TO BE GLUED TO THE INSIDE FRONT COVER OF THE THESIS Originality Statement

I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project’s design and conception or in style, presentation and linguistic expression is acknowledged.

Matthew Paul Hyett

5 January 2015

Page | II

Copyright Statement

I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or hereafter known, subject to the provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the abstract of my thesis in Dissertations Abstract International (this is applicable to doctoral theses only).

I have either used no substantial portions of copyright material in my thesis or I have obtained permission to use copyright material where permission has not been granted.

I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.

Note: Copyright has been obtained from each journal for inclusion of the published material in this thesis.

Matthew Paul Hyett

5 January 2015

Page | III

Authenticity Statement

I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.

Matthew Paul Hyett

5 January 2015

Page | IV

Table of Contents

Chapter 1: General Introduction...... 1 1.1 Background ...... 1 1.2 Depression: Epidemiology and Classification ...... 4 1.2.1 The Burden of Depression ...... 4 1.2.2 Classifying Depression – From Phenomenology to the Laboratory ...... 5 1.3 Theories of Attention ...... 8 1.3.1 Parsing Attention into Distinct Behavioural Components ...... 9 1.3.2 Automatic and Controlled Attention...... 10 1.3.3 Neuropsychological Models of Attention ...... 10 1.3.4 Cognitive and Computational Modelling of Attentional Processes ...... 12 1.4 Functional Neuroimaging: Background and Analysis Methods ...... 14 1.4.1 Dynamics of Brain Organisation ...... 16 1.5 Neuroimaging Studies of Attention ...... 18 1.5.1 Segregation of Attentional Brain Networks ...... 19 1.6 Cognitive Deficits and Attentional Disturbances in Melancholia ...... 21 1.6.1 Relationships between Psychomotor Disturbance and Attention Deficits ...... 22 1.6.2 Selective Attention: Set-Shifting and Inhibition ...... 23 1.6.3 Controlled/Effortful and Automatic Attentional Processing ...... 24 1.6.4 Summary of the Neuropsychology of Attention in Depression ...... 25 1.7 Functional Imaging Studies of Depression ...... 25 1.7.1 Emotional Circuitry and Depressive Disorders ...... 26 1.7.2 Imaging Distinct Depressive Sub-types and Specific Clinical Features ...... 30 1.7.3 Task-Driven Attentional Disturbances in Depression ...... 31 1.7.4 Alternate Approaches to Studying Brain Function ...... 31 1.7.5 Emerging Network Approaches to Understanding the Neurobiology of Depression ...... 32 1.8 Computational Modelling of Attentional Disturbances in Melancholia ...... 33 1.9 An Introduction to the Studies ...... 33

Chapter 2: Bias and discriminability during emotional signal detection in melancholic depression ...... 35 2.1 Abstract ...... 35

Page | V

2.2 Introduction ...... 36 2.3 Methods ...... 39 2.3.1 Sample ...... 39 2.3.2 Psychiatric and Neurological Screening ...... 40 2.3.3 Neuropsychological Testing Procedures ...... 41 2.3.4 Hierarchical Bayesian Modelling of AGN Data...... 43 2.4 Results ...... 46 2.4.1 Sample Characteristics ...... 46 2.4.2 Group Effects of Mean Bias and Discriminability ...... 50 2.4.3 Comparing Standard Deviation Model Estimates ...... 53 2.4.4 Sensitivity and Robustness Analyses of Model Posteriors ...... 53 2.5 Discussion ...... 54

Chapter 3: The insula state of melancholia: Disconnection of interoceptive and attentional networks ...... 60 3.1 Abstract ...... 60 3.2 Introduction ...... 61 3.3 Methods ...... 63 3.3.1 Participants ...... 63 3.3.2 Depression Sub-typing Approach ...... 64 3.3.3 Q-sort Methodology to Derive Prototypic Melancholic Symptom Scores ...... 64 3.3.4 Imaging Data Acquisition and Analysis ...... 64 3.4 Results ...... 66 3.4.1 Clinical and Demographic Comparisons ...... 66 3.4.2 Overall Network (Node Degree) Effects...... 69 3.4.3 Specific Effective Connectivity (Group-Averaged Edge Degree) Effects ...... 71 3.4.4 Examining the Impact of Medications on Network Parameters ...... 73 3.5 Discussion ...... 73

Chapter 4: Scene unseen: Disrupted neuronal adaptation in melancholia during emotional film viewing ...... 79 4.1 Abstract ...... 79 4.2 Introduction ...... 80 4.3 Methods ...... 84

Page | VI

4.3.1 Participants ...... 84 4.3.2 Depression Sub-typing Approach ...... 84 4.3.3 Demographic and Clinical Assessment...... 85 4.3.4 Imaging Protocol ...... 85 4.3.5 Naturalistic Stimuli – Film Clips ...... 85 4.3.6 Imaging Acquisition and Pre-processing ...... 86 4.3.7 Dynamic Causal Modelling...... 87 4.3.8 Inter-Subject Correlations ...... 88 4.3.9 Network-Based Statistic...... 88 4.4 Results ...... 90 4.4.1 Inter-Subject Correlations of Hidden Neuronal States ...... 90 4.4.2 Connectivity amongst Brain Modes in Melancholia ...... 92 4.4.3 Network-Based Modelling of Naturalistic Film viewing ...... 94 4.4.4 The Impact of Medication on Sub-Network Scores ...... 95 4.5 Discussion ...... 95

Chapter 5: General discussion and future directions ...... 99 5.1 Evidence of Disrupted Attentional Inference in Melancholia ...... 101 5.2 The Bayesian Brain, Attention and Inference: Towards a Cognitive Neuroscience of Melancholia ...... 102 5.3 Distorted Interoceptive Inference in Melancholia ...... 105 5.4 Validity of the Proposed Model ...... 106 5.5 Limitations ...... 107 5.6 Integrating Neuropsychological and Neuroimaging Data ...... 109 5.6.1 Brain Network Correlates of Set-Shifting Performance ...... 110 5.7 Conclusions ...... 112

Appendix 1 ...... 113

Appendix 2 ...... 118

Appendix 3 ...... 125

Bibliography...... 129

Page | VII

List of Tables

Table 2-1: Clinical and demographic characteristics of melancholic (Mel), non- melancholic (N-Mel) and control groups ...... 48

Table 2-2: Frequencies of hits (H), misses (M), false alarms (FA) and correct rejections (CR) across signal valence conditions and group on the go/no-go task. d′ is presented as a function of hit and false alarm rates ...... 49

Table 3-1: Demographic and clinical characteristics across melancholic, non- melancholic, and control groups ...... 68

Table 3-2: Between group contrasts of incoming and outgoing ‘node degree’ for attentional and insula networks ...... 70

Table 3-3: Between group edge connectivity differences among spatially distributed brain networks ...... 71

Table 4-1: Inter-subject correlations of negative and positive film viewing conditions ...... 91

Table A2-1: Symptoms and signs expressed by melancholic (Mel1-Mel16) and non- melancholic (NMel1-NMel16) participants ...... 119

Table A2-2: ANCOVAs contrasting primary treatment partitions, controlling for diagnostic group, on ‘node degree’ parameters ...... 124

Table A3-1: Prediction of presence or absence of differing drug classes, controlling for clinical group, from interaction of rest and negative film viewing sub-network edge weights ...... 128

Page | VIII

List of Figures

Figure 1-1: Schematic representation of proposed brain attention systems ...... 11

Figure 1-2: ICA estimated resting patterns of multisubject data sets: coronal, sagittal, and axial view of spatial map for each component. Images are z statistics overlaid on the average high-resolution scan transformed into standard space. Black to yellow are z values, ranging from 2.0 to 5.0. The left hemisphere of the brain corresponds to the right side of the image ...... 17

Figure 1-4: Dorsal and ventral attentional systems in the brain defined via intrinsic connectivity analyses ...... 20

Figure 1-5: Areas of significant brain activity in depression. (a) Decreased (blue) and increased (red) activation in depressed patients at-rest compared with controls. (b) Increased activation (red) and decreased activation (blue) with selective serotonin reuptake inhibitor (SSRI) treatment in depressed patients. (c) Increased (red) and decreased (blue) activation in depressed patients compared with controls in response to happy stimuli. (d) Decreased (blue) and increased (red) activation in depressed patients compared with controls in response to sad stimuli ...... 29

Figure 2-1: Overview of task design showing positive signals with negative noise trials and positive signals with neutral noise – comprising the positive signal condition. The same design – with varying noise – was consistent in the negative and neutral conditions ...... 43

Figure 2-2: Graphical model for hierarchical signal detection theory ...... 44

Figure 2-5: Individual parameter estimates for bias and discriminability to positive, negative and neutral signal conditions across groups ...... 53

Figure 3-1: Analysis pipeline illustrating the use of ICA spatial maps to inform sDCMs. Optimised model parameters from the sDCMs were used for between-group comparisons ...... 66

Page | IX

Figure 3-2: Network diagram showing all possible edge connections. Significant between group differences, and trend-level effects, depicted by thick coloured lines, and coloured dashed lines, respectively. Each independent component or mode is depicted in terms of the location of its highest spatial weighting ...... 72

Figure 4-1: Analysis pipeline illustrating the use of ICA spatial maps to inform sDCMs. Directed edge weights derived from the sDCMs (of both rs-fMRI and film viewing conditions) were used in the NBS to test for condition by group effects ...... 87

Figure 4-2: Analysis pipeline for calculating inter-subject correlations of hidden neuronal states. Illustrated schematically for DMN mode. From top: (1) BOLD time series of each subject; (2) DCM inversion of BOLD to give neuronal states for each subject; (3) Inter-subject correlations calculated on hidden neuronal states ...... 89

Figure 4-3: Group comparisons of rank-ordered distributions of all 64 edge weights across positive and negative film viewing and resting state. Left column shows melancholic versus healthy controls: Right column shows melancholia versus non- melancholic MDD...... 93

Figure A1-1: Violin plots (overlaid with box-plots) of posterior distributions of the mean and standard deviation of bias and discriminability across signal conditions and groups ...... 114

Figure A3-1: a) Overall and b) continuous ratings of emotional valence for the two film clips, “Bill Cosby” and “The Power of One”, averaged across 18 healthy participants. Error bars signify standard error of the mean (SEM) ...... 127

Page | X

Abstract

Melancholia has long been positioned as a quintessentially biological depressive condition. Impairments in attention are prominent, particularly in shifting attention away from internal states, but a detailed neurocognitive understanding of these deficits is lacking. This thesis hence sought to clarify the cognitive and neurobiological mechanisms of attention deficits in melancholia. Analytic methods spanning cognitive and brain network modelling were employed to explore the biases and inflexibility of attention in melancholia, compared to non-melancholic depressed and healthy individuals.

The first study (Chapter 2) investigates disrupted attentional inference to emotional stimuli across sub-types of depression and healthy participants. I hypothesised that both depressed groups would show impaired discriminability of emotional signals, and that melancholia would be characterised by decreased sensitivity to emotional stimuli. Signal detection data from an attentional control task were modelled using hierarchical (Bayesian) statistics. Melancholia was associated with disrupted sensitivity of emotional signals, and poorer discriminability of neutral signals, hence likely reflecting distorted attentional inference. The second study (Chapter 3) explores resting state functional brain network effective connectivity across melancholic, non-melancholic and control groups. Interactions between cortical systems corresponding to attention, executive control and interoception – derived from independent component analysis (ICA) – were modelled using dynamic causal modelling (DCM). Analyses supported the hypothesis that relationships amongst networks subserving attention and interoception would be disrupted in melancholia. This study revealed a specific ‘dysconnectivity’ between brain regions underpinning attention and interoception in melancholia. In the third study (Chapter 4), I advanced an in-scanner naturalistic film viewing paradigm to quantify brain networks underling the shifting of attention from rest to dynamic processing of exogenous emotional stimuli, employing the same groups as Chapter 3. I hypothesised that cortical systems would remain in an “at-rest” state in melancholia, reflecting impaired attentional shifting to exogenous stimuli. Surprisingly, neuronal activity in systems supporting attention and interoception were increased in melancholia compared

Page | XI

to controls during negative film viewing. I speculate that these findings reflect ineffective neuronal adaptation during attentional resource allocation to emotional material in melancholia. Preliminary analyses (presented in Chapter 5) highlight that impaired attentional set-shifting performance is associated with disruptions to these neuronal systems, hence pointing to a disorder-specific behavioural analogue of the neurobiological findings. The studies comprising this thesis offer a unique cognitive and neurobiological explanation for attentional deficits in melancholia, and act to explain aspects of its clinical presentation in terms of impaired redirection of attention away from persistent and dysphoric internal states.

Page | XII

Acknowledgements

I am ever grateful to my supervisors. Firstly, to Professor Gordon Parker, who started me on this journey with a simple question, “What would you like to do?” Throughout the last seven years I have many a time thought I might lose the light in my eyes, but with your unwavering support, Gordon, I have felt able to push through. And here I am. My sincerest warm thanks to you. I also express nothing other than immeasurable thanks to Professor Michael Breakspear, as my primary PhD supervisor, for his vast knowledge base, technical skill and thoughtful judgement. Michael, your rock solid support, both personally and professionally, has been remarkable, and I thank you for this. It has been a pleasure working with you throughout. Taking on board a MatLab newbie has, I’m sure, led to episodic hair pulling, and I truly am appreciative of all that you (and your team) have taught me. Being part of your team has been a tremendously enjoyable experience. Thank you.

The opportunity to undertake the studies within this thesis would not have been possible without the support of the National Health and Medical Research Council of Australia (NHMRC): Program Grant’s 510135 and 1037196. Additionally, I acknowledge the support of QLD Health (via a VMO Research Award to Professor Michael Breakspear) and the facilities at QIMR Berghofer Medical Research Institute, which facilitated the analysis phase of my research to continue unabated.

My thanks also go to the many colleagues who I have met along the way, and from whom I constantly feel inspired – it may have been the smallest anecdote, grain of advice or extensive chat in the corridor or coffee shop, but to you all I am indebted for what you’ve taught me. My sincerest thanks to: Kathryn Fletcher, Dusan Hadzi- Pavlovic, Amelia Patterson, Stacey McCraw, Rebecca Graham, Tamara Yuen, Melissa Green, the staff and psychiatrists at the Black Dog Institute Depression Clinic; and also to those at NeuRA who helped keep the imaging study afloat: Professor Caroline Rae, Michael Green, NeuRA IT, Segar, Ness and many others. A huge thank you to my colleagues at QIMR Berghofer – I am so enormously appreciative of all your help and support. In particular, my thanks to Christine Guo, James Roberts, Vinh Nguyen and Anton Lord for all your technical pearls and, of course, the rest of the BNE team

Page | XIII

(Kartik, Leo, Tamara, Justin, Matt A, Sascha and Phil) for your encouraging words, and thoughtful insights in the lab meetings.

Thank you to my parents, sisters and loved ones. I truly don’t know how best to express my gratitude to you – you’ve each been there for me at every step and it is something I cherish every day. I feel blessed to have you in my life. Thank you for believing in me and helping me stay strong through this. Thanks Mum for being so accepting of my late night telephone rants insights into psychiatry and for being straight with me when I needed it most! I’m glad you got the speakerphone. And thanks Dad for always lending an ear, and offering your perspective on life’s tribulations. Your combined wisdom has kept me from sinking many a time. Love you always. To Amanda and Melissa, my wonderful sisters. We may have done the sibling thing by distance for a while now, but I am ever grateful for the hour-long conversations of all sorts, which always keep me grounded. Stay strong. Love you lots. And, of course, thank you to my wonderful friends for their continuous words of encouragement throughout the PhD process.

Finally, I extend my deepest appreciation to the patients and community volunteers who gave their time to this research.

This thesis is dedicated to my sons, William and Max.

Page | XIV

Publications and Presentations

Publications arising/in press

Hyett, M., Parker, G., & Breakspear, M. (2014). Bias and discriminability during emotional signal detection in melancholic depression. BMC Psychiatry, 14, 122.

Hyett, M. P., Breakspear, M. J., Friston, K. J., Guo, C. C., & Parker, G. B. (in press). The insula state of melancholia: disconnection of interoceptive and attentional networks. JAMA Psychiatry (PSY14-0514R accepted 19/09/2014).

Oral presentations

Hyett, M., Parker, G. Zalesky, A., & Breakspear, M. (2014). Modes, movies and networks: A failure of adaptation to negative emotional content in melancholia. Biological Psychiatry Australia Conference, Melbourne, Australia, 13-14 Oct, 2014.

Poster presentations

Hyett, M., Green, M., & Parker, G. (2010). Neurocognitive mechanisms of emotion regulation in depression. 18th EPA European Congress of Psychiatry, Munich, Germany, 27 Feb-2 Mar, 2010. Abstract published in: European Psychiatry, 25, 1429.

Hyett, M. P., Parker, G. B., & Breakspear, M. (2012). Quantifying sub-optimal decision-making in depression. The 3rd Australasian Cognitive Neuroscience Conference, Brisbane, Australia, 29 Nov-2 Dec, 2012. Abstract published in: Frontiers in Human Neuroscience, doi: 10.3389/conf.fnhum.2012.208.00151.

Awards

2010 – Postgraduate Research Scholarship Scheme (Competitive Travel Grants Program). Awarded from the University of New South Wales, Sydney, Australia.

Page | XV

Declaration of Contributions to Publications

I made significant contributions to all papers arising out of this thesis (Chapters 2 and 3) in terms of study design, data collection and its management, design of statistical approach, data analysis, and writing of the first and final drafts of manuscripts.

For the publication arising from Chapter 2, I contributed to the selection of neuropsychological tests in collaboration with Associate Professor Melissa Green (UNSW School of Psychiatry), and undertook neuropsychological evaluation of all participants. I chose the statistical modelling approach and was assisted by Professor Michael Breakspear in implementing this. I took primary responsibility for writing the paper in collaboration with my supervisors, Professor Michael Breakspear and Professor Gordon Parker.

For the publication arising from Chapter 3, I was involved in the design of the fMRI acquisition protocol and oversaw testing of participants (assisted by one other research assistant, Dr Tamara Yuen). I conducted all of the fMRI analyses under the supervision of Professor Michael Breakspear, and acknowledge the analytic support provided by Professor Stephen Smith (University of Oxford), and Professor Karl Friston (University College London). The writing of this paper was a collaborative effort including myself, Professor Michael Breakspear, Professor Gordon Parker, Professor Karl Friston and Dr Christine Guo.

Matthew Hyett

5 January 2015

Page | XVI

Abbreviations

ACC anterior cingulate cortex AGN affective go/no-go AI anterior insula AMPA α-amino-3-hydroxy-5-methyl-4 -isoxazolepropionic acid ANCOVA analysis of covariance AUD auditory mode BOLD blood-oxygen-level-dependent CANTAB Cambridge Neuropsychological Test Automated Battery CI credible interval DCM dynamic causal modelling DLPFC dorsolateral prefrontal cortex DMN default mode network d′ d-prime DSM Diagnostic and Statistical Manual of Mental Disorders DST dexamethasone suppression test ECT electroconvulsive therapy EPI echo planar imaging EXC executive control mode FDR false discovery rate fMRI functional magnetic resonance imaging FSL FMRIB (Functional MRI of the Brain) software library FWE family-wise error FWHM full-width half-maximum GAF Global Assessment of Functioning HPA hypothalamic-pituitary-adrenal HPD highest posterior density HPDd highest posterior density difference ICA independent component analysis IED Intra-Extra Dimensional Set Shift INS insula mode ISC inter-subject correlation LFP left frontoparietal mode

Page | XVII

L-INS left insula mode MCMC Markov chain Monte Carlo MDD major depressive disorder MELODIC multivariate exploratory linear optimized decomposition into independent components MINI Mini International Neuropsychiatric Interview MNI Montreal Neurological Institute MRI magnetic resonance imaging MVP medial visual pole mode NBS network-based statistic NMDA N-methyl-D-aspartate OFC orbitofrontal cortex PET positron emission tomography PFC prefrontal cortex QIDS-SR16 Quick Inventory of Depressive Symptomatology-Self Report – 16 item rCBF regional cerebral blood flow RFP right frontoparietal mode R-INS right insula mode rs-fMRI resting state functional magnetic resonance imaging RVP Rapid Visual Information Processing sDCM stochastic dynamic causal modelling SDT Signal Detection Theory SEM standard error of the mean sgACC subgenual anterior cingulate cortex sgPFC subgenual prefrontal cortex SOC Stockings of Cambridge SPM8 Statistical Parametric Mapping - version 8 SSRI selective serotonin reuptake inhibitor STAI State-Trait Anxiety Inventory vmPFC ventromedial prefrontal cortex VOI voxel of interest WTAR Wechsler Test of Adult Reading

Page | XVIII

Chapter 1: General Introduction

1.1 Background

Melancholic depression is a debilitating, often chronic disorder seen as the ‘classical’ depressive syndrome. It is characterised by marked impairments in psychomotor and cognitive functioning, alongside prominent mood, energy and affective disturbances. The notion that melancholic depression, or ‘melancholia’, represents a distinct form of depression – separable from a residual group of non-melancholic depressions – has its advocates (Parker & Hadzi Pavlovic, 1996; Taylor & Fink, 2006), but consistent identification of its biological substrates has long evaded clinicians and scientists (Hadzi-Pavlovic & Boyce, 2012). The search for objective markers of brain dysfunction in depressive illness has an extensive history, precipitated by reports and clinical observation of cognitive impairments in patients. Neuropsychological impairments observed in depressive disorders cut across multiple functional domains, with deficits commonly observed in attention, episodic (i.e., autobiographical) and working memory, verbal fluency, learning, reaction time under uncertainty and, in more severe presentations, preservative responding and inhibition of motor function (Austin, Mitchell, & Goodwin, 2001). Impairments of attention in particular are representative of the melancholic sub-type of depression, affecting both set-shifting (i.e., shifting attention from one stimulus to another) and sustained attention (Austin et al., 1999; Michopoulos et al., 2008; Politis, Lykouras, Mourtzouchou, & Christodoulou, 2004). Such deficits have been proposed to reflect disrupted fronto-subcortical brain circuitry in melancholia (Austin et al., 2001). However, despite considerable efforts to clarify the cognitive and neurobiological mechanisms of depression there has been little consensus amongst studies to date. Hence, neurobiological models of melancholia have remained largely conjectural.

Page | 1

Thus far, most studies examining the cognitive and neurobiological aspects of depression have focussed on isolated processes and localised brain regions, principally conforming to the ‘lesion-deficit’ paradigm of traditional neuropsychology. However, the utility of this approach in complex psychiatric presentations such as depressive disorders is frequently overstated (Keefe, 1995). Current conceptualisations of cognitive function suggest that it is underpinned by complex brain networks, where interactions between brain regions support the emergence of cognitive and perceptual operations (Jirsa & McIntosh, 2007). This approach has intuitive appeal for modelling psychiatric disorders (Menon, 2011), with value in explaining a range of illnesses where cognitive impairment is prominent, including schizophrenia (Micheloyannis et al., 2006) and autism spectrum disorder (Barttfeld et al., 2011). Here, I propose that the use of cognitive and imaging technologies arising out of contemporary “network” neuroscience may be of substantial benefit for understanding the core cognitive disturbances of melancholia.

Although melancholia is associated with a range of cognitive deficits, attentional impairments appear central to the disorder (Austin et al., 1999; Austin et al., 1992; Cohen, Lohr, Paul, & Boland, 2001; Lemelin et al., 1996; Mialet, Pope, & Yurgelun- Todd, 1996). With this in mind I set out to examine the mechanisms underlying attentional disturbances across refined sub-types of depression. This thesis moves away from the traditional neuropsychological battery approach of multi-domain assessment to one of specificity, with the aim of providing a detailed theory of attention in melancholia. Attention itself has been studied extensively for over 100 years, making it an ideal candidate for such in-depth analysis. One of the earliest definitions of attention came from psychologist William James, who stated that attention reflects:

“… the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought.” (James, 1890)

Page | 2

This definition highlights that the main function of attention is to bring something into conscious awareness. Since the earliest conceptualisations of attention, psychologists and, latterly, cognitive neuroscientists have sought to understand its function, and differing processes, through cognitive and neurobiological modelling. Attention was initially viewed as comprising multiple, orthogonal components (Broadbent, 1954; Posner & Petersen, 1990), but such perspectives have been all but overturned through advances in cognitive neuroscience. In addition to the brain having distinct local specialisation, it is now widely believed that integration occurs across multiple levels of brain organisation (e.g., neuronal, inter-region). This view offers a unified explanation of brain and behavioural function (Friston, 2005; Sporns, Chialvo, Kaiser, & Hilgetag, 2004; Tononi, Sporns, & Edelman, 1994). Integrated, large-scale, brain networks have also been suggested to support the emergence of cognitive operations such as attention (Bressler & Menon, 2010). Further, computational treatments of attention indicate that it adheres to rules of (Bayesian) optimal inference (Feldman & Friston, 2010), which lends itself to dynamics on distributed, hierarchical neuronal systems (Friston, 2008).

The objectives of this thesis are to clarify mechanisms of attentional dysfunction in melancholic depression through application of computational approaches to neuropsychological and neuroimaging data. From a neuropsychological perspective, I seek to elucidate inferential mechanisms of attentional control to emotional stimuli across different forms of depression (melancholic and non-melancholic) and in healthy individuals (Chapter 2). Further, to clarify the neurobiological underpinnings of such processes, I report several studies (Chapters 3 and 4) examining brain networks subserving attentional and interoceptive processing in melancholic, non-melancholic and control groups. These studies utilised a range of novel methodologies spanning acquisition and analysis. Finally, preliminary analyses examining relationships between attentional processes and brain networks are presented in Chapter 5 (Section 5.6), offering a synthesis of the differing data modalities.

Page | 3

Prior to introducing the current studies, I will overview the burden and epidemiology of depression, as well as issues related to its classification. I will then examine cognitive and neurobiological theories of attention, prior to reviewing the literature on neuropsychological and neuroimaging studies of depression. Where possible I will highlight studies in depression that have focused on attentional processes, and those that have studied differing depressive sub-types.

1.2 Depression: Epidemiology and Classification

1.2.1 The Burden of Depression Depression is of substantial public health concern. The 2004 update of the Global Burden of Disease (WHO, 2008) reported that unipolar depression was the leading cause of years lost to disability. The same report estimated that depression was the third leading cause of burden of disease (measured by disability adjusted life years) behind lower respiratory infections and diarrheal diseases, yet more burdensome than diseases such as heart disease, cerebrovascular disease and diabetes. Furthermore, the 2004 report estimated that over 150 million individuals worldwide suffer from unipolar depression, nearly six times that of schizophrenia (26.3 million) and bipolar disorder (29.5 million). Epidemiological prevalence estimates do, however, vary widely across studies depending on diagnostic criteria used. For example, those that have looked at rates of major depressive disorder (MDD) have reported lifetime prevalence estimates of 16.2%, and 12-month estimates of 6.6% (Kessler et al., 2003). When diagnoses of depression were more weighted to melancholia (i.e., using stricter diagnostic criteria), lifetime estimates of between 7.5% and 10% were reported (Kessler, Zhao, Blazer, & Swartz, 1997). In addition to being a common illness, depression is highly disabling. Those who suffer from this illness are much more likely to experience significant functional limitations, including poorer physical, social and role functioning (Wells et al., 1989). These functional impairments have also been shown to be causal factors in increased service utilisation (Johnson, Weissman, & Klerman, 1992), and contribute to

Page | 4

significant economic burden through increases in days lost from work (Broadhead, Blazer, George, & Tse, 1990).

1.2.2 Classifying Depression – From Phenomenology to the Laboratory The definition of depression has been debated for many years. Its classification has traditionally rested upon descriptive psychopathology, which relies on symptom-based reports from patients or observations by clinicians. Early conceptualisations of melancholia (variably termed vital, endogenous, Type A or endogenomorphic depression) positioned it as an illness categorically distinct from reactive/neurotic depressive states (Parker, Hadzi-Pavlovic, & Boyce, 1996). This binary view was widely accepted for some time, but gradually loss traction when multivariate analyses failed to find convincing support for distinct disorders, and hence favour shifted to dimensional definitions. The view that melancholia was simply a severe expression of depression was advanced in 1980 with the publication of the third edition of the Diagnostic and Statistical Manual for Mental Disorders – DSM-III (APA, 1980). This nosological framework saw depression as varying by degree, hence the term “major” depression (“minor” depression also being considered as comprising a valid diagnostic group). Here the number of presenting symptoms was the principal basis for a diagnosis of depression. The DSM approach to depression diagnosis has, however, been contested. It has expanded the boundaries of depression to include sub-syndromal states and normal grief reactions (Parker, 2000, 2005, 2011), which impacts negatively on treatment selection and identifying casual factors in research studies. The consensus- based diagnostic approach also fails to align with emerging efforts to inform nosology by the biological underpinnings of mental illnesses (Insel et al., 2010). In contrast, clinical diagnoses of melancholia are made by observation of symptoms and signs, while also considering additional developmental and aetiological factors that may influence the clinical picture, including personality style, family history, and prior response to treatment. A return to studying more refined sub-types of depression may thus advance understanding of the causal mechanisms of depression, which are

Page | 5

otherwise obscured under the dimensional framework of depression advocated in DSM (Ingram & Siegle, 2009).

The view that depression is not a single entity, but rather consists of two or more types, has prevailed in psychiatry for over 100 years. Jaspers was the first to propose a binary view of depression (Jaspers, 1913), with his original distinction between ‘reactive’ and ‘vital’ (cf. endogenous) depression enduring until 1980 – the year DSM- III was published. Reactive depressive episodes, according to Jaspers, and subsequent proponents of the binary model, were brought on by negative external events. Endogenous depression was, to the contrary, an illness that developed from within, and was without external cause (Shorter, 2013). This distinction was challenged in the lead- up to DSM-III when it was reported that reactive and endogenous depressive patients experienced similar rates of negative life events (Thomson & Hendrie, 1972). The term ‘endogenous’ depression was subsequently abandoned. However, melancholia was subsequently resurrected, independently from DSM, and again positioned as the quintessential biological form of depression, distinct from a set of residual non- melancholic conditions (Taylor & Fink, 2006). Proponents of melancholia have since called for it to be reinstated in diagnostic classification systems as a distinct entity (Parker et al., 2010). A range of key features dominates its clinical picture. Psychomotor slowing and/or agitation are particularly characteristic in melancholia (Sobin & Sackeim, 1997), as are the vegetative signs of initial insomnia, early morning wakening, weight loss/anorexia, loss of libido and diurnal variation of mood (Davidson & Turnbull, 1986). Clinical observational rating of psychomotor disturbance offers superior utility in diagnosing the disorder when compared with symptom-based reports (Parker & Hadzi Pavlovic, 1996), but the self-reported symptoms of feeling physically slowed, anhedonia and mood and energy worse in the morning, are also sensitive indicators of the disorder (Parker et al., 2009). Together, these findings of varying clinical phenotypes suggest that there may be distinct aetiological pathways for melancholia – in particular those of biological origin.

Page | 6

One measure that generated wide interest when first developed was the dexamethasone suppression test (DST), which measures plasma free cortisol non- suppression. This test has shown promise in differentiating psychotic melancholia, non- psychotic melancholia and the non-melancholic conditions. An early study using the DST demonstrated specificity of 96% and sensitivity of 67% in identifying clinically diagnosed melancholia (Carroll et al., 1981). Yet despite the potential utility of the DST, some have argued that over-emphasising associations between pathophysiological processes and clinical presentation may overlook contributions of alternative aetiological factors (Hickie, 1996). The DST does, however, provide an index of hypothalamic-pituitary-adrenal (HPA) axis integrity. As the HPA axis is involved in the regulation of stress and emotions (Herman & Cullinan, 1997), it might be expected that this system is intricately linked to disorders of mood. The DST was largely abandoned in clinical and research applications for depression given its non-specificity to melancholia as assessed by DSM criteria (Mitchell, 1996). The emergence of newer technologies, namely functional neuroimaging, superseded such neuroendocrine methods as the field sought to understand the neurobiological basis of depression. Consequently, advances in brain imaging technologies contributed to a paradigm shift in psychiatry (Andreasen, 1988), based on the prevailing view that mental illness originated from organic (brain) dysfunction.

The 1990s were designated as the “decade of the brain”, on the grounds that “continued study of the brain was needed to combat the large number of debilitating neural diseases and conditions (Jones & Mendell, 1999).” This period marked a significant increase in neuroscience research and was the impetus for a new era of research into the causes of depression. Neuroimaging studies sought to understand what brain regions were partially or wholly responsible for contributing to its symptomatology. Many such studies have been limited in focus by attempting to explain the neurobiology of MDD through identification of localised brain regions. Significant heterogeneity is often introduced in such studies given the near exclusive focus on broad diagnostic categories. Further, the complexity of depression – even in

Page | 7

refined depression sub-types such as melancholia – is at odds with attempts to isolate specific regional dysfunction. A more detailed review of cognitive and neurobiological studies of depression is provided below in Sections 1.6 and 1.7, respectively, predominantly focussing on attention.

The cognitive impairment seen almost ubiquitously in melancholia, particularly of an inability to pay attention and concentrate, is indicative of mood state-related brain dysfunction. I therefore propose that the study of attentional processes, and neural mechanisms underlying these, offers significant potential to progress towards a more refined neurocognitive model of this disorder. Whilst the neural mechanisms of attentional control in MDD have been investigated (Elliott, Rubinsztein, Sahakian, & Dolan, 2002), precise understanding of the behavioural and neuronal manifestations of attention in melancholia remains unclear. Prior to overviewing previous studies of attention and cognitive dysfunction in depression, it is first necessary to provide an overview of attention itself.

1.3 Theories of Attention

There are multiple, competing views regarding the processes that underlie attention. Common to all theories, however, is that attention endows an individual with the ability to focus on a particular aspect of a stimulus, at the expense of disengaging from all other perceptual information (Nobre & Kastner, 2014). As conscious beings, we possess an instinctive understanding of what it means to pay attention, but on closer inspection it stands as one of the most complex cognitive processes. This may be due to attention being involved in a vast array of conscious experiences, and that seemingly broad models of attention can often be subdivided into specific components. Thus, when it is observed that a depressed individual is unable to concentrate or pay attention, what are the underlying candidate processes? Is attention simply not focussed on the external world, and instead remains focussed on an internal state? Examination of the different

Page | 8

models of attention, and advancing their application, will help to answer such questions. The literature on modelling attention is extensive, and thus only a brief overview of theories relevant to depression will be presented.

A variety of methodologies, including behavioural observation, analysis of brain- behaviour relationships, the lesion-deficit method, and use of a range of imaging modalities, have been used to study attention (Posner, 2011). In addition to human research, there has also been widespread use of non-human primates. This section overviews the cumulative contributions that each method has provided towards understanding attention.

1.3.1 Parsing Attention into Distinct Behavioural Components Moray (1970) proposed six different definitions of attention, effectively rendering a universal account of attention problematic. Several theorists have aligned with this notion in attempting to model attention, namely that it comprises a range of differing processes. Following Moray’s proposition, Posner and Boies (1971) put forward a tripartite model of attention through introduction of the concepts of alertness, selectivity, and processing capacity. The first, alertness, involves maintaining optimal sensitivity to expected incoming stimuli (the so-called ‘foreperiod’ of attention) and is most commonly observed in the context of sustained attention or vigilance. The second, selectivity, is defined as the ability to select certain sources of information over other sources, also referred to as selective attention. This is analogous to a filter, functioning to block input from unselected sources, and to attenuate processing of other sources. The third, central processing capacity, refers to the concept of attention being constrained by a limiting mechanism, such that any two (or more) operations requiring attentional processing resources will be in conflict. This third component parallels early views of attention whereby a limited-capacity channel only permits a restricted amount of information to pass through (Broadbent, 1954). The concept of attention being capacity-limited remains a focus of research efforts to this day (Dux & Marois, 2009).

Page | 9

1.3.2 Automatic and Controlled Attention A series of studies in the 1970s (Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977) proposed that “automatic” and “controlled” information processing best described attention. Automatic attentional processing can occur when an object or stimulus has been learnt over time, and thus attention to that object on subsequent exposure is relatively automatic. In contrast, controlled processing requires active control of attention by an individual. Modulation of these control processes was suggested to be via other cognitive functions such as working memory. Briefly, working memory is responsible for holding multiple elements of information in the mind, and where each element is able to be manipulated for further processing (Baddeley, 2003). Given such definitions we can see that there are likely a range of reciprocal relationships between attention and other cognitive processes. Consequently, these interrelationships may be a contributing factor in failing to achieve consensus definition of attention.

1.3.3 Neuropsychological Models of Attention Contemporary views of attention moved beyond purely cognitive definitions to a more integrative framework that considered the role of brain structure and function. Mirsky and colleagues (1991) proposed that attention was a consequence of the coordination of divergent cerebral structures. Neuropsychological test scores of focussing, vigilance/sustained attention and attentional set-shifting (i.e., controlled attention) were used to derive data-driven estimates of four broad elements of attention, namely: “focus-execute”, “shift”, “sustain”, and “encode”. Figure 1-1 illustrates the putative neural structures underlying these four elements. From this early neuroanatomical depiction we can see that these elements of attention are broadly distributed across the cortex, extending from prefrontal cortex (PFC) and parietal and temporal cortices, to subcortical structures including the anterior cingulate cortex (ACC), amygdala and hippocampus. Similar neuroanatomical accounts have been proposed providing further support for the view that differing aspects of attention correspond to a distributed array of brain regions (Posner & Petersen, 1990).

Page | 10

Figure has been removed due to Copyright restrictions.

See http://www.ncbi.nlm.nih.gov/pubmed/1844706

Figure 1-1: Schematic representation of proposed brain attention systems.

Adapted from Mirsky et al. (1991). Copyright © Plenum Publishing Corporation.

Posner and Petersen’s model of attention was developed primarily from behavioural and physiological studies (i.e., lesion studies of humans and primates), and initially comprised three independent attentional systems, specifically: orienting, alerting and target detection. The posterior parietal lobe was identified as a key region involved in the covert orienting of attention, and was subsequently referred to as the “posterior attention system”. Midbrain regions, in particular portions of the thalamus and superior colliculus, were identified as being specific to visual orienting. Damage to these regions has been shown to be associated with difficulties in shifting attention away from a previously attended stimulus. Specifically, lesions to the posterior parietal lobe are associated with an inability to disengage attention from stimuli contralateral to the location of the lesion (Posner, Walker, Friedrich, & Rafal, 1984). Brain regions

Page | 11

underlying target detection include the PFC, ACC and supplementary motor area, whereas alerting, or attentional preparedness, is thought to correspond to an “anterior attention system” in the brain. This system is believed to be right lateralised, but knowledge regarding localisation of this attentional sub-process beyond this asymmetry is limited. This brain-based taxonomy of attentional systems has recently been refined given insights into cognitive control processes and the ACC (Botvinick, Braver, Barch, Carter, & Cohen, 2001). In particular, the ‘target detection’ network has been repositioned as an ‘executive control’ network (Petersen & Posner, 2012) given overlap of the former with the concept of ACC-mediated cognitive control (Carter & Krug, 2012).

From such work it has become increasingly apparent that a diversity of brain regions contribute to attention. This is especially evident across differing formulations of attention where it appears highly integrated with a variety of other, apparently distinct, cognitive processes (e.g., working memory). Emerging evidence suggesting that differing cognitive processes map onto near identical cerebral structures (Price & Friston, 2005) casts doubt on the modular, ‘phrenological’ account of mind, which until recently dominated research into cognition-brain relationships. Further, empirical studies of attention typically overlook the contribution of inferential processes that are inherent to many cognitive tasks, which itself relies on orchestrated communication between differing brain regions (Friston, 2005). A range of inferential processes can be interrogated through formal psychophysical modelling of cognitive tasks, which I propose is of significant benefit towards increasing understanding of attentional dysfunction in melancholia.

1.3.4 Cognitive and Computational Modelling of Attentional Processes The majority of studies of attention have been largely informal (Logan, 2004). That is, most have used empirical data to propose qualitative models without regard for underlying psychophysical processes (e.g., perceptual inference). To address this,

Page | 12

quantitative approaches, such as signal detection theory or SDT (Green & Swets, 1966), developed over the last half century, can be applied to understand attentional inference processes. Classically, the signal detection capacity of an individual is estimated through yes/no discrimination tasks, where a decision is made as to the occurrence of a signal (response) in the presence of noise (distractor) stimuli. Response parameters of bias and discriminability can be derived using SDT, providing an approximation of the internal perceptual criterion used, and the discrimination capacity between any given signal and noise distributions, respectively. The more these distributions overlap, the less sensitive an individual is to the value of a true signal, and thus the poorer their discrimination will be. The SDT approach hence provides a unique quantitative description of experimental data in allowing interpretation beyond observable behaviour.

It has been suggested that optimal performance on attentional tasks corresponds to “evidence accumulation” in the brain (Feldman & Friston, 2010). This is based on the assumption that the brain represents uncertainty in a probabilistic (Bayesian) manner, which in turn informs our perceptions of the world and subsequent actions (Knill & Pouget, 2004). Neurobiologically plausible accounts of Bayesian inference have been proposed, with emphasis here placed on the importance of coordination between hierarchically organised brain systems in contributing to perception and action (Friston, 2005). In parallel, methodological advances that seek to unify probabilistic cognitive and perceptual inference with moment-to-moment functional brain signal fluctuations offer unique insights into Bayes optimal responding (Garrett et al., 2013). Thus, while there is a large body of research that has focussed on isolated regions in attention, there is an emerging literature on cortico-cortical interactions, particularly between frontal and parietal brain regions. To appreciate the utility of such advances, I first overview the imaging methods commonly used in cognitive neuroscience.

Page | 13

1.4 Functional Neuroimaging: Background and Analysis Methods

Functional neuroimaging techniques, namely positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), have provided unparalleled insight into brain regions supporting cognitive processes (Cabeza & Nyberg, 1997, 2000). Studies of brain activation are dependent on observing changes in brain states, typically within the same scanning session (Friston, Ashburner, Kiebel, Nichols, & Penny, 2007). PET and fMRI allow quantification of such brain activity through observation of changes in glucose metabolism and haemodynamics, respectively.

PET allows the in vivo measurement of brain region concentrations of positron- emitting radionuclides, or tracers. Depending on the tracer used, a range of different metabolic processes can be measured, including blood flow, oxygen utilisation, glucose utilisation, blood volume, and amino acid transport amongst others (Lammertsma & Frackowiak, 1985). In contrast, fMRI measures blood flow changes through quantification of changes in blood oxygenation (Huettel, Song, & McCarthy, 2009). The ratio of oxyhaemoglobin to deoxyhaemoglobin concentration following blood flow to a given region generates a magnetic signature, detectable through MRI, known as the blood-oxygen-level-dependent (BOLD) signal (Ogawa, Lee, Kay, & Tank, 1990). In short, blood flow observed through this mechanism reflects metabolic processes generated by active neurons. The rise and fall of oxygen concentrations are seen as a proxy of neuronal activity. In addition, an accumulation of evidence indicates that other processes mediate blood flow in the brain. Neurovascular signalling, mainly via glutamatergic transmission, plays a major role in regulating cerebral blood flow, which is predominantly controlled by astrocytes (Attwell et al., 2010). Astrocytes surround synapses and are stimulated by neuronal activity, and also contribute to blood vessel dilatation through control of smooth muscle cells. In addition, blood flow in the brain is affected by its complex microvasculature – including pericytes (cells present along capillaries) – that are able to alter capillary diameter (Peppiatt, Howarth, Mobbs, &

Page | 14

Attwell, 2006). Hence, blood flow changes detected through fMRI are likely the result of a complex interplay of cellular, biochemical and metabolic processes. The widespread application of fMRI in comparison to PET is due to a number of factors; first, PET is significantly more expensive than fMRI due to the cost associated with manufacturing radiotracers; second, fMRI is less invasive and third, has superior spatial and temporal resolution, thus making it more suited for event-related experiments where specific cognitive processes are to be interrogated (Cabeza & Nyberg, 2000). Despite the advantages of fMRI it is remains sensitive to motion and susceptibility artefacts.

Behavioural observation of healthy and neurologically impaired populations has traditionally been used to understand cognitive processes and, further, relate performance to specific brain regions. In the case of brain-damaged patients, impaired behavioural function attendant with infarcts to a given cortical region implicate that area as specific to that function. This is referred to as the lesion-deficit model. Functional imaging studies have, however, shown there to be significant overlap of neural systems for what are thought to be distinct cognitive processes (Price & Friston, 2005). Brain organisation hence appears to adhere to two fundamental principles, namely functional specialisation and functional integration (Tononi et al., 1994). As per the lesion-deficit account, functional specialisation has its roots in ‘localisationism’, where identification of particular brain regions for specific functions was the predominant goal (Friston et al., 2007). This view of brain function holds that specific cortical regions are specialised for some aspects of cognitive and perceptual processes. Functional integration emphasises interactions between distributed brain regions via cortico-cortical connections, that are well established as functioning to unify distinct regions (Friston, 2003). It is thus likely that a combination of functional specialisation and functional integration support the emergence of cognitive and perceptual processes. Numerous methods have been developed that allow modelling of the integrative architecture of the brain – broadly referred to as network analysis (Sporns, 2010). These methods model interregional connectivity through use of covariances and/or correlations between regionally specific time series. Functional connectivity and effective connectivity

Page | 15

(Friston, 1994) are two methods for quantifying such relationships. Functional connectivity measures statistical correlations between distinct brain regions, whilst effective connectivity makes inferences regarding causal interactions between neuronal populations.

1.4.1 Dynamics of Brain Organisation Two methods of relevance to this thesis are briefly overviewed – namely independent component analysis (ICA) and dynamic causal modelling (DCM). ICA allows investigation of the spatio-temporal structure of fMRI data (Beckmann, DeLuca, Devlin, & Smith, 2005). ICA works upon the assumption that the sensor-level data (fMRI voxels) reflects a linear mixing of unknown source (neuronal) processes, and decorrelates components as much as possible by maximising independence and Gaussianity. ICA is thus an example of ‘blind source separation’, which separates the source signals from the mixed signals. The ‘unmixing’ step minimises the mutual information between output sources (spatial maps), decomposing the data into time- varying functional brain imaging maps and their associated time courses, describing the spatial and temporal characteristics of the signals. This approach is also referred to as the ‘cocktail party’ problem – in a crowded room, all simultaneous speech signals initially appear ‘mixed’, but can be separated through selection of specific sources (Bell & Sejnowski, 1995). It has been demonstrated that spatial components (or ‘modes’) derived from resting state fMRI (rs-fMRI) and ICA are consistent across individuals (Damoiseaux et al., 2006) (e.g., see Figure 1-2), and also correspond to established cognitive activation networks (Smith et al., 2009).

Page | 16

Figure has been removed due to Copyright restrictions.

See http://www.ncbi.nlm.nih.gov/pubmed/16945915

Figure 1-2: ICA estimated resting patterns of multisubject data sets: coronal, sagittal, and axial view of spatial map for each component. Images are z statistics overlaid on the average high-resolution scan transformed into standard space. Black to yellow are z values, ranging from 2.0 to 5.0. The left hemisphere of the brain corresponds to the right side of the image.

Adapted from Damoiseaux et al. (2006). Copyright © National Academy of Sciences.

Whereas ICA identifies spatially distributed networks or “modes”, it does not, on its own, reveal information regarding their interactions. Interactions between different brain regions can be inferred from data using DCM. This approach models interactions between neuronal populations using haemodynamic time series (Friston, Harrison, & Penny, 2003). DCM’s are ‘generative models’ and, in essence, provide a quantitative description of how observed (experimental) data are generated. In experimental settings, DCM can be used to model the influence of specific task-driven factors on interactions between neuronal populations. In such models, a series of differential equations prescribe the existence and direction of causal connections between nodes (brain

Page | 17

regions). Inverting these models from data provides information on the strength (edge weights) of causal influences between brain regions. Inverted models also provide estimates of neuronal states. While often used in studies where there are experimental manipulations, DCM has recently been extended for application to rs-fMRI data (Daunizeau, David, & Stephan, 2011; Li et al., 2011). This is achieved by endowing the original “deterministic” DCM models with noisy fluctuations, hence yielding “stochastic” DCMs. Dynamic causal modelling has an established role in cognitive neuroscience, with its first application dealing with attention to motion (Buchel & Friston, 1997), and has more recently been used in combination with ICA {Goulden, 2014 #3535;St Jacques, 2011 #3534}. Hence, this approach is ideally suited to neuroimaging studies of attention.

1.5 Neuroimaging Studies of Attention

There has been a progressive shift towards understanding attention in terms of brain- behaviour relationships, with more recent research focussing heavily on elucidating its neurobiological basis (see Figure 1-3). In a review of PET and fMRI studies of cognition, Cabeza and Nyberg (2000) identified a broadly distributed network of specialised regions underpinning the attentional processes of: i) sustained attention; ii) selective attention; iii) stimulus-response compatibility tasks (e.g., Stroop and those related to conflict resolution); iv) orienting of attention; and v) divided attention. All functions corresponded to brain activation in frontal, cingulate, parietal, temporal and, with the exception of divided attention, occipital and subcortical brain structures. Certain studies, reporting on specific components of attention, showed more consistent activations in given regions than did others for the same function. For example, sustained attention was more likely to be associated with activation in the frontal cortex, relative to the involvement of cingulate, temporal, occipital and subcortical structures. Selective attention showed the opposite pattern. However, across all included studies, both frontal and parietal cortices were consistently involved. Indeed, the frontal and parietal cortices appear so intricately involved in attention that they have been the focus of a significant body of research, both historically and in present day formulations.

Page | 18

Figure 1-3: An historic overview of significant milestones in the study of attention. Personal depiction.

1.5.1 Segregation of Attentional Brain Networks Given the extent of the cortex that could be classified as ‘frontoparietal’, attempts have been made to identify regions within – and indeed beyond – this system that correspond to specific attentional processes. Corbetta and Shulman (2002) reviewed evidence for the existence of two, partially segregated attention networks, organised topographically into ‘dorsal’ and ‘ventral’ systems. Figure 1-4 illustrates these networks (Fox et al., 2005). The dorsal system, involved in preparing and applying goal-directed selection of stimuli, and for associated responding, includes parts of the intra-parietal cortex and superior frontal cortex. This network is also referred to as a “top down” attentional control system. The ventral network is predominantly right lateralised and includes the temporo-parietal junction and inferior frontal cortex. This network is specialised for the detection of behaviourally relevant stimuli, in particular those that are unexpected and/or emotionally salient. These systems are suggested to interact during attentional

Page | 19

reorienting, with the ventral frontoparietal system acting to mediate the dorsal system for the redirection of attention to salient stimuli.

Figure has been removed due to Copyright restrictions.

See http://www.ncbi.nlm.nih.gov/pubmed/15976020

Figure 1-4: Dorsal and ventral attentional systems in the brain defined via intrinsic connectivity analyses

Adapted from Fox et al. (2005). Copyright © National Academy of Sciences.

Frontoparietal brain regions appear to underpin numerous attentional functions, predominantly in the sensory (e.g., visual, auditory) domain (Cabeza & Nyberg, 2000). However, there is evidence to suggest that these brain regions are also involved in a variety of other cognitive processes, including mental imagery (Formisano et al., 2002), and tasks requiring internalised mental effort, such as working memory (Pessoa, Gutierrez, Bandettini, & Ungerleider, 2002). It can thus be appreciated that an extensive network of brain regions, several of which coalesce to support the emergence of seemingly disparate cognitive processes, underpins attention.

Page | 20

1.6 Cognitive Deficits and Attentional Disturbances in Melancholia

The processes of attention and executive functioning feature prominently in studies of depression, and melancholia more specifically. Traditional neuropsychological assessment allows for the assessment of an individual’s cognitive strengths and weaknesses, relative to a given normative metric (e.g., in the case of depression, studies typically make comparison with a control group, matched for confounding factors). As a field, neuropsychology attempts to relate behaviour to brain function, on the premise that “behaviour is generated by complex physical processes in the central nervous system” (Heilman & Valenstein, 2012). A “diminished ability to think or concentrate” is one of the nine symptoms under DSM (APA, 2013) that count towards a diagnosis of MDD. Such deficits are also characteristic of melancholia, as succinctly captured by British naval officer and writer John Custance, who suffered from a mood disorder:

“Instead of the light of ineffable revelation I seem to be in perpetual fog and darkness. I cannot get my mind to work; instead of associations “clicking into place” everything is an inextricable jumble; instead of seeming to grasp a whole, it seems to remain tied to the actual consciousness of the moment.”(Custance, 1952, p. 63)

Melancholia has indeed been likened to dementia in terms of its cognitive deficits – so called “depressive pseudodementia” given the reversibility of impairments (Berrios, 1985). Neuropsychological methods are thus often employed in an attempt to understand such symptoms, and indeed depression more broadly. An early overview of neuropsychological functioning in depression emphasised the need for identification of cognitive strengths and weaknesses across different depressive sub-types (Miller, 1975). It was concluded that cognitive deficits were similar across different ‘non biological’ types of depression (e.g., atypical, reactive), but that differentiation was evident for certain cognitive functions when endogenous depression (cf. melancholia) was considered. Deficits in attention, while not studied directly, were found to be characteristic of melancholic depression, but not those with reactive depression (Miller,

Page | 21

1975). Most subsequent studies shifted focus away from depression sub-typing, towards broader definitions of depression following the introduction of the DSM, but several have investigated the capacity that neuropsychological tests of attention hold for differentiation of depressive sub-types. Such studies have typically focused on one of three broad processes: psychomotor dysfunction as a causal mechanism of impaired attention, selective attention (i.e., attentional set-shifting/inhibition), and controlled versus automatic processing.

1.6.1 Relationships between Psychomotor Disturbance and Attention Deficits Depression with prominent psychomotor disturbance is associated with marked impairments in attention. The primacy of psychomotor disturbance in melancholia has thus emerged as a candidate phenotype for the study of attentional dysfunction in depression. It has been demonstrated that those with endogenous depression with observable psychomotor slowing perform more poorly on measures of attention- dependent psychomotor speed and shifting compared to healthy controls or those with non-endogenous depression (Austin et al., 1992). However, links between psychomotor disturbances and attentional dysfunction have not consistently been observed in melancholia (Austin et al., 1999). That said, associations between increased psychomotor retardation scale scores and attentional impairments have been reported in broadly defined depression (Lemelin & Baruch, 1998). Here, severe psychomotor retardation was correlated with more pronounced attentional dysfunction. Thus, findings to date have been mixed regarding psychomotor dysfunction as being a mediating factor of attention dysfunction in depression. This is further challenged by reports that attentional dysfunction appears to continue into the non-symptomatic phase of MDD (Paelecke-Habermann, Pohl, & Leplow, 2005), where it might be expected that any clinical signs of psychomotor dysfunction resolve. Additionally, few studies have objectively assessed psychomotor speed and attention in parallel, regardless of phenotypic expression. Clinical ratings of psychomotor dysfunction, while useful for assigning melancholic status, may be suboptimal for quantifying relationships with attentional function. Thus, there remains an open question as to whether attentional

Page | 22

deficits are epiphenomena of psychomotor disturbance in melancholia, or whether attentional deficits also occur in those with melancholia without significant psychomotor disturbance.

1.6.2 Selective Attention: Set-Shifting and Inhibition Attentional set-shifting is defined as the ability to shift attention across different stimulus features, and is particularly impaired in melancholic depression compared to non-melancholic depression (Austin et al., 1999; Austin et al., 1992; Michopoulos et al., 2006). However, understanding of the mechanisms underlying such deficits remains unclear. Diurnal variation of mood in depression, and attendant variation in neuropsychological function (Moffoot et al., 1994), has been identified as one potential mechanism. This is based on the premise that increased morning cortisol levels, which have some specificity to melancholia, may influence set-shifting ability. Despite the fact that cortisol hypersecretion has been linked to cognitive deficits in depression (Rubinow, Post, Savard, & Gold, 1984), identification of relationships between set- shifting performance and cortisol levels in melancholia has not been demonstrated (Michopoulos et al., 2008), discounting this as a mechanism of disrupted set-shifting. In addition to set-shifting deficits, those with depression also exhibit inhibitory control impairments. For example, deficits in selective attention have been observed using the Stroop Colour-Word Test (Stroop, 1935) – a test of distractor stimuli inhibition (i.e., naming the colour of an incongruent colour word). Increased Stroop interference scores have been reported in MDD, hence indicating a generalised inhibitory mechanism deficit in this disorder (Lemelin et al., 1996). In this same study selective attention impairments were observed in the absence of psychomotor slowing. When considering the mixed findings regarding psychomotor disturbance and attention, this provides further evidence that attention deficits can occur independently of psychomotor signs.

Additionally, the use of emotional stimuli in studies of attention has been highly informative for understanding cognitive mechanisms underlying depression, especially

Page | 23

given reports of maladaptive cognitive biases towards negative information in this disorder (Mathews & MacLeod, 2005). To this end, go/no-go tasks with an affective component have been employed to understand mechanisms of response inhibition towards emotional stimuli. Such tasks provide unique insight into cognition-emotion interactions by requiring an individual to respond to target emotional words and inhibit responses to non-target (distractor) words. An inability to shift focus away from negative emotional material is characteristic of those with MDD (Murphy et al., 1999), hence providing support for the notion of attentional control impairments towards negative stimuli in depression. Mechanisms underlying the cognitive control of emotion have not been examined across different sub-types of depression. As such it remains unclear as to whether the above findings are of relevance to disorders such as melancholia.

1.6.3 Controlled/Effortful and Automatic Attentional Processing It has been suggested that impaired performance on cognitive tasks in depression, particularly those involving attention, is partially a product of decreased effort (Hartlage, Alloy, Vazquez, & Dykman, 1993). The level of effort required to complete any given task closely aligns with the theory of controlled versus automatic attentional processing (Schneider & Shiffrin, 1977; Shiffrin & Schneider, 1977). Those with depression exhibit more pronounced impairments in attentional control on tasks that are ‘effortful’ (Tancer et al., 1990), thus highlighting that controlled attention may be particularly affected in this disorder. Further, performance on tasks comprising both attentional and decision-making components decreases in depression (Thomas, Goudemand, & Rousseaux, 1999). This suggests that performance deficits may be a result of inefficient resolution of uncertainty under high effort conditions. Despite this, attention deficits have also been observed under low-effort, more automatic conditions in a cohort of severely depressed patients (Cohen et al., 2001). While reference to different sub-types of depression was not specified, the identified dysfunction of automatic processes in severe depression may offer insight into impairments in acute (i.e., often severe) expressions of melancholia.

Page | 24

1.6.4 Summary of the Neuropsychology of Attention in Depression It has been noted that deficits in attentional functioning in depression are one of the few consistent findings arising out of neuropsychological studies of affective disorders (Cohen et al., 2001). Despite the field having made progress towards a neuropsychological deficit model of depression, mechanisms underlying diminished attentional capacity remain unresolved. Specifically, no studies to date have examined the psychophysical properties of attentional control (using approaches such as SDT) across clinically meaningful sub-types of depression. Application of such methods would arguably lead to greater insight into stimulus-response contingencies across different subjects, and hence contribute to more refined understanding of the cognitive processes underlying attentional dysfunction.

1.7 Functional Imaging Studies of Depression

The frontoparietal attention system has been positioned as a central brain system mediating the expression and maintenance of numerous mental illnesses, including depression (Cole, Repovs, & Anticevic, 2014). This system, also known as the cognitive control system, is highly integrated with other distributed systems in the brain (e.g., visual, motor, limbic system) through a rich network of reciprocal connections (Cole, Pathak, & Schneider, 2010). Clarifying relationships between frontoparietal ‘hubs’ and other systems, such as the limbic system, is likely to be essential towards furthering understanding of stress and mood regulation (Herman, Ostrander, Mueller, & Figueiredo, 2005). Despite a vast body of research spanning some three decades there has not been consistent evidence for specific brain dysfunction in MDD, or in depressive sub-types. This may possibly be related to the fact that the majority of studies have attempted to elucidate disorder-level pathophysiological markers (in heterogeneous groups) rather than focussing on specific illness features in refined sub- types (e.g., attentional mechanisms in melancholia).

Page | 25

Nonetheless, advances in functional neuroimaging have proved invaluable in the search for pathophysiological mechanisms of psychiatric illnesses, including depression (Andreasen, 1988). Early applications of PET revealed that depression is associated with regional cerebral blood flow (rCBF) reductions in both dorsolateral prefrontal cortex (DLPFC) (Baxter et al., 1989) and caudate nucleus (Baxter et al., 1985). However, conflicting findings were frequent across initial investigations (Depue & Iacono, 1989). Some indicated regionally specific alterations, as above, while global reductions of cerebral blood flow were identified in subsequent studies (Sackeim et al., 1990) – and still others failed to find differences specific to depression (Maes et al., 1993). Around the same time, the PFC emerged as a candidate brain region of relevance, and was subsequently identified, along with the ACC, as being closely linked to the symptomatology of depression (Bench et al., 1992; George, Ketter, & Post, 1994). The PFC is an extensive cerebral structure intricately involved in cognitive control (including emotional evaluation), executive attention and decision-making (Miller & Cohen, 2001). Despite the PFC being implicated in depressive disorders findings have been inconsistent, with some identifying blood flow reductions (Bench et al., 1992), while others have identified blood flow increases (Tutus et al., 1998). Hence, it remains unclear as to the precise role of the PFC in depression. Further, such studies offer little insight into the neurobiology of melancholia given their focus on broadly defined diagnostic groups. They also lack specificity to MDD given “hypofrontality” was one of the initial pathophysiological hallmarks of schizophrenia (Andreasen et al., 1994; Franzen & Ingvar, 1975; Weinberger, Berman, & Zec, 1986). These early studies were, however, precursors to a large body of research that sought to understand the neurobiological underpinnings of depression.

1.7.1 Emotional Circuitry and Depressive Disorders The PFC is central for reward processing, and exerts a top-down influence over subcortical structures for optimal regulation of behaviour. Contemporary formulations of emotional systems in the brain suggest that the PFC is part of an extensive cortical circuit, referred to as the ‘Papez circuit’ (Dalgleish, 2004). The original model (Papez,

Page | 26

1937) comprised connections between the thalamus (hypothalamus and anterior thalamus), sensory cortex, hippocampus and ACC, and was later extended to include the PFC and amygdala under the triune theory of the brain (or ‘visceral brain’) (MacLean, 1949). This extended system soon became known as the limbic system (Maclean, 1952). Given the central role of these regions in emotion processing (Dalgleish, 2004), their study in disorders like depression has been of intrinsic appeal (Drevets, 2000). Indeed, disruptions to pathways linking prefrontal brain regions and the limbic system have been identified in depression {Kennedy, 1997 #3548;Rogers, 2004 #3553}.

The subgenual area of the PFC (sgPFC) has been implicated in depression after it was identified as being reduced in activity compared to healthy control subjects (Drevets et al., 1997). In contrast, an adjacent region, the subgenual ACC (sgACC), was found to be hyperactive in severely depressed patients (Mayberg et al., 2005). This distinction was interpreted as possibly reflecting differences across sub-types of depression and, thus, rather than being artefactual, may contribute to our understanding of differing pathophysiological processes in disorders such as melancholia. Despite the inconsistent findings, common to each was the identification of limbic structures, which is in turn consistent with earlier suggestions of disrupted limbic-cortical connectivity in depression (Mayberg, 1997). Such work highlights the importance of identifying brain structures responsible for emotional processing – and their reciprocal interactions – and determining their correspondence with differing clinical features. Mayberg’s working model suggested that ‘bottom up’ limbic system processes fail to inhibit PFC activation and vice versa in depressive disorder. This proposed dysregulation of cortico-limbic pathways was one of the first ‘network’ approaches to brain dysfunction in depression, and has subsequently been independently validated in MDD (Seminowicz et al., 2004). Reports of pathophysiological processes in emotion-centric brain regions foreshadowed a range of studies investigating neural responses to differing emotional stimuli, drawing on methodological advances in cognitive activation experiments in neuroimaging (Elliott et al., 2002).

Page | 27

A recent meta-analysis of fMRI studies of depressed subjects established that a range of structures consistently show increased activation, compared to controls, during the presentation of negative stimuli (Hamilton et al., 2012). These included the amygdala, insula and dorsal ACC. In contrast, decreased activation in the dorsal striatum and DLPFC in response to negative stimuli has been consistently identified. Many of these structures form part of the limbic system, and again highlight a range of key brain structures subserving emotional processing and regulation as being of relevance to depression {Rive, 2013 #3552}. This identification of multiple, distributed brain regions is consistent with emerging views of brain function in psychiatric illnesses – namely, that illnesses impact upon a constellation of regions and their interactions (Bassett & Bullmore, 2009). To illustrate this, Figure 1-5 highlights results of a meta- analysis of brain activation studies, revealing that disrupted brain function in depression occurs across a range of brain regions, and which is dependent on differing experimental conditions (Fitzgerald, Laird, Maller, & Daskalakis, 2008).

Page | 28

Figure has been removed due to Copyright restrictions.

See http://www.ncbi.nlm.nih.gov/pubmed/17598168

Figure 1-5: Areas of significant brain activity in depression. (a) Decreased (blue) and increased (red) activation in depressed patients at-rest compared with controls. (b) Increased activation (red) and decreased activation (blue) with selective serotonin reuptake inhibitor (SSRI) treatment in depressed patients. (c) Increased (red) and decreased (blue) activation in depressed patients compared with controls in response to happy stimuli. (d) Decreased (blue) and increased (red) activation in depressed patients compared with controls in response to sad stimuli.

Adapted from Fitzgerald et al. (2008). Copyright © John Wiley & Sons.

Page | 29

1.7.2 Imaging Distinct Depressive Sub-types and Specific Clinical Features The majority of functional imaging studies into depression have used broad diagnostic categories (i.e., MDD), but several have prospectively investigated melancholia in an attempt to model its neurobiological basis. As overviewed above, it has been demonstrated that those with treatment resistant melancholia show decreased baseline rCBF in the sgACC (Mayberg et al., 2005). This aligns with earlier work (Pizzagalli et al., 2004) revealing that those with MDD with melancholic features show relative decreases in glucose metabolism (identified through PET) in the sgACC compared to those with non-melancholic depression. No structural abnormalities were observed in the sgACC in this study. In contrast, Soriano-Mas and colleagues (2011) identified structural brain abnormalities in melancholic patients, in particular grey matter volume reductions in the left insula, and white matter volume increases in the upper brainstem tegmentum. This latter study also observed that reduced left insula volume was predictive of the clinical course of melancholia (increased time to symptomatic improvement), raising the possibility that structural brain pathology may correspond to clinical outcome in melancholia.

Correlations between a range of clinical features and brain dysfunction have also been examined. Reduced psychomotor activity has been identified as being associated with total white matter hyperintensities, with the severity of such pathology predictive of a poorer response to treatment (Hickie et al., 1995). Further, reduced rCBF and structural lesions to white matter pathways are associated with psychomotor slowing, as assessed by reaction time under uncertainty, in older patients with MDD (Hickie et al., 2007). These studies indicate that psychomotor dysfunction is closely linked to structural abnormalities. However, given the advanced age of the samples care should be taken not to generalise the findings to younger individuals with depression: These changes might specifically reflect the melancholic profile of “vascular depression” (Alexopoulos et al., 1997). Furthermore, aside from observed rCBF reductions, functional brain contributions to psychomotor dysfunction remain unknown. The functional underpinnings of cognitive deficits have, however, been investigated in

Page | 30

depression. Specifically, reversible cognitive impairments are associated with deficits to several neural structures, including the left medial anterior PFC and cerebellar vernis (involved in autonomic regulation) (Dolan et al., 1992).

1.7.3 Task-Driven Attentional Disturbances in Depression Furthering understanding of the neurobiology of attentional disturbances in depression has been assisted through applications of the ‘cognitive activation’ paradigm. Attenuated neural responses during inhibitory control of emotional words have been identified in ventral and rostral ACC, orbitofrontal cortex (OFC) and medial PFC in MDD (Elliott et al., 2002), pointing to a distributed neural architecture underpinning cognition-emotion interactions. Additionally, increased attentional processing times (on the Stroop) have been linked to DLFPC dysfunction (Wagner et al., 2006). There is also evidence that frontoparietal systems are disrupted in depression. Decreased glucose metabolism in the parietal cortex and DLPFC, and increased metabolism in the OFC, has been identified across both endogenous and non-endogenous depressed groups (Biver et al., 1994). Given the known role of the parietal cortex in supporting attentional processes, it may be possible that dysfunction to such brain regions contributes to the clinical picture of melancholia, and attentional deficits in this illness more specifically.

1.7.4 Alternate Approaches to Studying Brain Function As highlighted above, task-based paradigms are frequently employed in neurobiological studies of depression in an attempt to understand brain-behaviour relationships. This approach can be problematic in psychiatric settings given task completion difficulties that are inherent when studying severely ill patients. The use of resting state imaging paradigms (where intrinsic neuronal fluctuations are recorded at-rest) overcomes these task demands while also providing insight into unconstrained neuronal mechanisms. Resting state acquisitions have the unique advantage of being able to capture task- independent cognition, which is apposite for assessing the internally focussed, dysphoric quality of thought in melancholia.

Page | 31

In addition, an emerging literature indicates that naturalistic stimuli, namely the viewing of movie clips, is able to elicit consistent patterns of cortical activation across individuals (Hasson, Nir, Levy, Fuhrmann, & Malach, 2004). In addition to having high ecological validity, application of this method in depressed individuals would vastly reduce the burden associated with task completion, while still allowing for observation of neuronal processes underlying a range of cognitive functions including attention. I propose that these cutting-edge developments in neuroscience offer an advance over extant modelling approaches towards elucidating cognitive and neurobiological markers of attention deficits in melancholia.

1.7.5 Emerging Network Approaches to Understanding the Neurobiology of Depression The studies overviewed have offered important insights into the neural basis of depressive disorders and, in several instances, have contributed towards positioning melancholia and its prototypic features (e.g., cognitive and psychomotor dysfunction) as having a distinct neurobiological basis. Despite such advances, there is a paucity of research into complex brain networks in melancholia, and a consequent lack of knowledge regarding precisely how key features of the illness correspond to neurobiology. Connectivity disturbances have been identified in MDD amongst brain regions similar to those overviewed in this thesis thus far (Anand et al., 2005; Avery et al., 2014; Davey, Harrison, Yucel, & Allen, 2012; Greicius et al., 2007; Lu et al., 2012; Schlosser et al., 2008), but such relationships are yet to be elucidated across different depressive sub-types. I propose that the application of novel neuroimaging approaches in specific diagnostic groups will contribute to a more refined understanding of pathophysiological processes in depressive disorders. The study of distinct biological phenotypes, such as melancholia, closely aligns with calls for psychiatry to move towards a taxonomy based on observable cognitive and neurobiological markers, rather than relying on symptoms, for diagnosis (Insel et al., 2010). Here, I draw on a range of novel methods from cognitive neuroscience in an attempt to further understanding of the cognitive and neurobiological mechanisms of attention in melancholia.

Page | 32

1.8 Computational Modelling of Attentional Disturbances in Melancholia

There is clear consensus that those with depressive disorders exhibit impaired cognitive functioning including, but not limited to, attention and executive control, and that these disturbances are likely driven by perturbed neuronal function. Aside from diagnostic limitations (e.g., continued study of MDD), there is inherent variability between individuals in cognitive and neurobiological studies, placing further limits on inferences about group-wise processes. In addition, studies that have used depressive sub-typing have reported conflicting findings, rendering it difficult to identify reliable neurobiological and cognitive markers for disorders such as melancholia. Thus, despite consensus that there are some neurobiological and cognitive impairments in melancholia, its status as a distinct neurobiological entity remains inconclusive. Given the limitations faced by studies to date it is clear that a multifaceted approach is necessary for advancing knowledge of mechanisms underlying attentional impairments in depressive disorders.

1.9 An Introduction to the Studies

Here, I present a series of studies employing neuroscience methods aimed at refining understanding of the cognitive and neurobiological mechanisms of attention in melancholia.

Adopting a behavioural perspective, in Chapter 2 I present a detailed investigation of the psychophysical properties of attentional control in melancholic, non-melancholic and control groups. The combined use of SDT and Bayesian hierarchical statistical modelling in this study allowed the testing of hypotheses regarding optimal control of attention to emotional stimuli under conditions of uncertainty.

Page | 33

In Chapter 3, I investigate resting state brain network integrity in melancholic and non-melancholic depression, compared to healthy controls, using stochastic dynamic causal modelling (sDCM). Here, it was hypothesised that melancholia would be associated with disrupted effective connectivity amongst key brain modes subserving attention and interoception.

Chapter 4 builds on work presented in Chapter 3: specifically, I set out to determine brain network properties of attentional reorienting using a novel film viewing paradigm. In particular, I test the hypothesis that those with melancholia were more likely (than non-melancholic and control counterparts) to show disruptions in attentional, interoceptive and sensory brain networks during the shifting of attention from resting state to emotional film viewing.

In Chapter 5, I discuss the implications of the studies and highlight key limitations. This is followed by several analyses, in preliminary form, of relationships between neuropsychological functioning and brain network parameters during emotional film viewing, per Chapter 4. It is proposed that such integrative analyses act to unify the divergent literatures on cognitive and neurobiological mechanisms of attentional dysfunction in melancholia. Finally, I offer suggestions for future research studies, thus allowing refinement and extension of the findings presented herein.

Page | 34

Chapter 2: Bias and discriminability during emotional signal detection in melancholic depression

2.1 Abstract

Background: Cognitive disturbances in depression are pernicious and so contribute strongly to the burden of the disorder. Cognitive function has been traditionally studied by challenging subjects with modality-specific psychometric tasks and analysing performance using standard analysis of variance. Whilst informative, such an approach may miss deeper perceptual and inferential mechanisms that potentially unify apparently divergent emotional and cognitive deficits. Here, we sought to elucidate basic psychophysical processes underlying the detection of emotionally salient signals across individuals with melancholic and non-melancholic depression.

Methods: Sixty participants completed an Affective Go/No-Go (AGN) task across negative, positive and neutral target stimuli blocks. We employed hierarchical Bayesian SDT to model psychometric performance across three equal groups of those with melancholic depression, those with a non-melancholic depression and healthy controls. This approach estimated likely response profiles (bias) and perceptual sensitivity (discriminability). Differences in the means of these measures speak to differences in the emotional signal detection between individuals across the groups, while differences in the variances reflect the heterogeneity of the groups themselves.

Results: Melancholic participants showed significantly decreased sensitivity to positive emotional stimuli compared to those in the non-melancholic group, and also had a significantly lower discriminability than healthy controls during the detection of neutral

Page | 35

signals. The melancholic group also showed significantly higher variability in bias to both positive and negative emotionally salient material.

Conclusions: Disturbances of emotional signal detection in melancholic depression appear dependent on emotional context, being biased during the detection of positive stimuli, consistent with a noisier representation of neutral stimuli. The greater heterogeneity of the bias across the melancholic group is consistent with a more labile disorder (i.e., variable across the day). Future work will aim to understand how these findings reflect specific individual differences (e.g., prior cognitive biases) and clarify whether such biases change dynamically during cognitive tasks as internal models of the sensorium are refined and updated in response to experience.

2.2 Introduction

Melancholia is frequently conceptualised as a biological disorder encompassing disturbances of mood, motor function, thinking, cognition and perception (Parker & Hadzi Pavlovic, 1996; Taylor & Fink, 2006). Whilst cognitive impairments in melancholia have been investigated in detail (Austin et al., 2001; Bench et al., 1992), definitive identification of selective neurocognitive impairments has not been achieved. Given the pressing need to examine underlying perceptual and inferential processes in heterogeneous illnesses such as depression (Paulus & Yu, 2012), it is increasingly recognised that a range of methodological approaches should be utilised in the analysis of neurocognitive data to more accurately capture the nature of disturbances across individuals. Such refined approaches have direct utility in enhancing understanding of group-specific psychophysical processes across sub-types of depression.

There is typically great inter-subject variability on tests of neuropsychological function in the major psychiatric illnesses (Franzen, 2000). Meaningful interpretations of brain function in specific disorders is difficult given such variability. This is further Page | 36

compounded by summarising an individual’s position on a performance continuum (as with summing errors on a task) in order to infer the presence or absence of cognitive impairments. Furthermore, commonly utilised neuropsychological tests in depressive disorders typically rest upon broad construct-level approaches (e.g., tests of ‘executive function’ or ‘attentional control’) that do not facilitate the development of theories regarding specific psychophysical disturbances in individuals. Despite such drawbacks in assessing cognition in psychiatric illnesses, significant advances have been made over the past 20 years in explaining human perceptual inference and action (Dayan, Hinton, Neal, & Zemel, 1995; Friston, 2005; Knill & Pouget, 2004) using probabilistic statistical principles such as those developed through a Bayesian-based approach (Bayes, 1763). The Bayesian statistical modelling approach has been applied to individual and group cognitive data across multiple cognitive domains, including signal detection that is viewed as encompassing the processes of attention, decision-making and executive functioning (Lee, 2008; Lee & Wagenmakers, 2014). Formally, the signal detection capacity of an individual can be influenced by prior beliefs (or internal models of the world) and the incoming sensory stream, generating that individual’s response profiles. This, in turn, provides an ideal platform through which to measure perception and inference. In the analysis of cognitive data, SDT (Green & Swets, 1966) allows modelling of the optimal detection of stimuli, through estimating discriminability and bias (Wickens, 2002). SDT relates discriminability – how easily signal (response) and noise (non-response) trials are distinguished – and bias, a measure of how well the decision-making criterion relates to the optimal criterion. Both constructs reflect an individual’s internal model of the sensorium and their prior contextual beliefs. Signal and noise trials of a task can be represented along a perceptual strength construct in SDT, referring to the strength of inference made to a particular stimulus – that is, the probability that a conclusion (decision/action) is true given its premises. Inferences during streams of trials are continuously monitored through sensory experience and evaluation, and may then be used to update decision criteria for subsequent task performance. Rouder and Lu (2005) suggest it is reasonable to expect that on such tasks there will be significant participant-level variability in signal detection sensitivity, creating a need for statistical models that capture individual processes.

Page | 37

Inter-subject variability is rarely modelled in neuropsychological studies of depressed individuals. Moreover, commonly used aggregation methods have the potential to lead to statistical effect estimates that may poorly represent group heterogeneity (Rouder & Lu, 2005). Bayesian statistics offer the ability to pre-specify prior knowledge through the specification of priors (Lee & Wagenmakers, 2005). A Bayesian approach to data analysis is also appealing in the setting of decision-making in the face of uncertainty because it embodies the same type of assumptions – and hence represents the same constructs – as emerging models of human decision-making (Behrens, Woolrich, Walton, & Rushworth, 2007; Karim, Harris, Langdon, & Breakspear, 2013). When considering group data using SDT, individual subject variability can be modelled using hierarchical Bayesian techniques (Rouder & Lu, 2005), allowing estimation of data-driven posteriors of mean bias and discriminability as well as their variance or precision (the inverse of variance) (Lee, 2008; Lee & Wagenmakers, 2014). When cognition is variably disrupted, as arguably is the case in depression, inter-subject estimates of bias and optimal judgement may be influenced, which can ideally be modelled through hierarchical Bayesian SDT analyses. There are several reasons as to why such an approach may offer significant benefit.

In health, cognitive ‘priors’ can be viewed as personal beliefs that are optimised towards the most likely value of a given percept (Griffiths & Tenenbaum, 2006). In depression, however, such processes may be suboptimal in different ways across individuals, extending variously across perceptual, inferential and performance domains. It has been suggested that depression is associated with distorted inference (e.g., “arbitrary inference”) at certain levels of severity (e.g., psychotic depression) (Beck & Alford, 2009), yet despite recent theoretical research with Bayesian modelling in depression (Huys & Dayan, 2009; Paulus & Yu, 2012) no studies have employed Bayesian statistics to model cognitive capabilities in depressive illnesses such as melancholia. Most studies to date have attempted to delineate underlying mechanisms of negative cognitive biases (Mathews & MacLeod, 2005) based on the notion that depressed individuals have a characteristically negative view of the self, world and

Page | 38

future (Beck, 1979; Beck & Alford, 2009). Studies have shown that attention is selectively drawn to negative information (Mathews, Ridgeway, & Williamson, 1996), and that memory of negative information is enhanced (Gilboa-Schechtman, Erhard- Weiss, & Jeczemien, 2002) in depression. However, few studies have provided a formal quantitative framework for modelling individual level disturbances from empirical psychophysical data. While some studies (Elliott et al., 2002) have established evidence for neurobiological correlates of response bias, it remains to be seen whether cognitive biases extend across depression as a whole or whether they are specific to given individuals or groups. From the findings to date it is evident that there is an unmet need in elucidating basic mechanisms of neurocognitive dysfunction across individuals with depression.

We propose that biases in emotional stimulus processing in depression can be accurately captured through investigation of different depressive sub-types using a hierarchical Bayesian emotional SDT framework. Employing an emotional word ‘go/no-go’ task, which requires responding and inhibition of responding to serially presented, randomly sequenced positive, negative and neutral words, we hypothesised that each depressed sub-set would show less optimal responding (poorer discriminability) across emotional signal conditions as compared to the control group, and that the melancholic participants would show more difficulty in detecting true signal trials from noise trials, particularly on emotional signal conditions (i.e., lower sensitivity) compared with non-melancholic and control participants.

2.3 Methods

2.3.1 Sample Participants consisted of 20 melancholic and 20 non-melancholic depressed individuals, recruited through a specialist depression clinic at the Black Dog Institute in Sydney, Australia. A healthy control group of 20 participants was recruited through the

Page | 39

community. The study was approved by the University of New South Wales Human Research Ethics Committee and all study participants gave informed consent prior to taking part.

2.3.2 Psychiatric and Neurological Screening Exclusion criteria for healthy controls included a lifetime history of a mood and/or psychotic disorder as screened by the Mini International Neuropsychiatric Interview (MINI) (Sheehan et al., 1998). Depressed participants were considered eligible if they had a current major depressive episode, but no (hypo)mania or psychosis identified on the MINI. Those with depression were additionally required to meet a current (past 7 days) level of depression severity of 11 or more on the 16-item Quick Inventory of Depressive Symptomatology – Self Report (QIDS-SR16) (Rush et al., 2003), indicative of at least moderate depression severity. All participants were required to be fluent in English, and the age range for inclusion was between 18 and 75 years. Exclusion criteria for all participants consisted of current or past drug or alcohol dependence, current or past history of neurological disorder (i.e., neurodegenerative conditions, stroke, central nervous system infection, tremor), a history of brain injury with significant impairment (i.e., neurotrauma from haemorrhage, oedema, hypoxia), invasive neurosurgery and/or an estimated full scale IQ score of below 80 as assessed by the Wechsler Test of Adult Reading (WTAR) (Pearson Assessments, 2001). An additional exclusion criterion for depressed participants was having received electroconvulsive therapy (ECT) within the preceding six months. Current medication was recorded. In addition to the above screening, all participants completed the Global Assessment of Functioning (GAF) (APA, 2000) and the State-Trait Anxiety Inventory (STAI) (Spielberger, 1983). Patient groups were assessed for observable psychomotor disturbance using the CORE measure (Parker et al., 1994). Screening was conducted by trained research assistants at the Black Dog Institute.

Page | 40

Delineation of depressive sub-types (melancholic versus non-melancholic depression) proceeded according to the clinical criteria previously reported by Parker and colleagues (2010). Trained psychiatrists at the Depression Clinic assessed these criteria, which include the presence or absence of psychomotor disturbances, concentration and/or decision-making impairment, non-reactive affect, distinct anergia, diurnal mood variation – being worse in the morning, appetite and/or weight loss, early morning wakening, absence of preceding stressors accounting for the depth of the depressive episode, previous good response to adequate antidepressant therapy, and normal personality functioning (Parker & Hadzi Pavlovic, 1996; Taylor & Fink, 2006). The focus of the current chapter on perceptual accuracy but not reaction time tempered circularity between diagnostic assignment (e.g., cognitive slowing in melancholia) and signal detection performance. Previous research (Parker et al., 2009) has shown that the prototypic diagnostic approach used in the current study (involving symptom and non- symptom data) to be more strongly differentiating of melancholic and non-melancholic depression than use of the DSM-derived (APA, 2000) melancholic specifier criteria which considers symptoms only.

2.3.3 Neuropsychological Testing Procedures All participants took part in a brief neuropsychological assessment, with tests taken from the Cambridge Neuropsychological Test Automated Battery (CANTAB) (Robbins et al., 1994), that included the Stockings of Cambridge (SOC), Intra-Extra Dimensional Set Shift (IED), Rapid Visual Information Processing (RVP) and the AGN. All testing took place in a sound-attenuated room that housed a desktop computer running CANTAB eclipse version 3.0 software. Additional computer hardware (touch screen and response/press pad) allowed recording of behavioural responses to the stimuli. The focus of the present chapter is on the AGN.

Page | 41

The AGN is a test of emotional word discrimination that requires responding to ‘go’ trials, and inhibition of responding on ‘no-go’ trials, to negative, positive and neutral stimuli. The task consists of 20 blocks with 18 words in each block. The first two blocks are used for training and are not further analysed. Two word categories are presented within each block. No two consecutive blocks present the same word combinations. For each block there are nine signal trials and nine noise trials. Prior to the onset of each block participants are primed with what word categories to expect. As depicted in Figure 2-1, each block requires detection of signal and noise (positive, negative or neutral emotional word) categories, with six possible signal/noise combinations, in the following order of administration (repeated three times, giving the 18 blocks): positive (signal)-neutral (noise); positive-negative; neutral-positive; neutral-negative; negative- neutral; negative-positive. Words appear for 300 milliseconds one at a time in the centre of the computer screen, followed by a 900 millisecond inter-stimulus interval. Each block hence lasts for 22 seconds. Participants rest between blocks for five seconds, allowing preparation for the following block. Analysis variables from the AGN consisted of hits (correct responses to signal trials), false alarms (incorrect responses to noise trails), misses (incorrect rejections to signal trials) and correct rejections (to noise trials).

Page | 42

Figure 2-1: Overview of task design showing positive signals with negative noise trials and positive signals with neutral noise – comprising the positive signal condition. The same design – with varying noise – was consistent in the negative and neutral conditions.

2.3.4 Hierarchical Bayesian Modelling of AGN Data A hierarchical Bayesian graphical model for SDT (Lee, 2008; Lee & Wagenmakers, 2014) was used for the AGN data. As a statistical technique, such modelling – with the use of Markov chain Monte Carlo (MCMC) sampling – allows for integration of a prior distribution (prior beliefs/knowledge) with data from the behavioural task to obtain approximations of the posterior distributions of the response parameters. AGN blocks were categorised by emotional signal condition (positive, negative, neutral): thus, three specified models were used in this study. Alternating noise conditions were pooled together for each of the signal conditions. For example, a positive signal distribution in the current analysis has a noise distribution that includes both negative and neutral stimuli. We modelled individual participant responses to the task (counts of hits and false alarms) to generate posterior estimates of discriminability and bias from hit and false alarm rates, separately for each of the emotional signal conditions, with uniform prior distributions hence assigning equal probabilities to all possible states. Page | 43

The modelling approach is represented in Figure 2-2. By convention, unobserved variables are nodes without shading and observed variables are ones with shading, while continuous variables are represented as circular nodes and discrete variables as square nodes. It thus follows for the current study that the observed behavioural data are represented by the shaded grey squares, and our estimated variables, hi and fi (hit and false alarm rates), are shown as unshaded circular nodes. Further modelled (unobserved) parameters are also shown in the model – in particular, ci and di are estimates of bias and discriminability, whilst the top level of the hierarchy formally incorporates group- wise mean (μ) and standard deviation (σ) estimates of both ci and di. To derive these estimates, signal and noise trials are denoted by S and N, respectively, to which individual (i) counts of hits (Hi) and false alarms (Fi) are derived and, subsequently, their rates (hit rate = hi and false alarm rate = fi). In the graphical model, ϕ is used to calculate the cumulative distribution function of hi and fi, whilst λ (at the top of the hierarchy) is the precision of the c and d parameters, from which σ estimates are obtained.

Figure 2-2: Graphical model for hierarchical signal detection theory.

Page | 44

For bias estimates, the optimal criterion is centred at zero between two equal- variance Gaussian distributions representing signal and noise distributions. A negative bias value relative to zero indicates a preference towards more ‘yes’ responses (as it is nearer the noise distribution), whereas higher positive values reflect a preference for ‘no’ responses. The location of the response criterion (positioned along a unidimensional strength construct of perceptual accuracy) is derived from the participant’s estimate as to whether a stimulus constitutes a signal or a noise trial. For instance, a participant whose ‘yes rule’ is further from the noise distribution requires a stronger signal to permit detection – thus, the strength of the signal required is determined by the strength of the participant’s internal representation of that signal. In such cases, the available perceptual evidence needs to be substantially greater than the criterion rule to say ‘yes’ (Wickens, 2002). Discriminability is an estimate of the ability to differentiate between signal and noise trials and indicates how well a signal trial can be detected, with higher values indicating an increased capacity to distinguish between signal and noise trials. A previously developed implementation of hierarchical SDT (Lee & Wagenmakers, 2014), as shown in Appendix 1, was modified and implemented in WinBUGS (Lunn, Thomas, Best, & Spiegelhalter, 2000). Posterior distributions of bias and discriminability were estimated using MCMC sampling (10,000 samples), using hit and false alarm rates over uniform (“reference”) priors (i.e., where all responses are equally likely). The posterior means of bias and discriminability, taken as the joint emotional signal detection capacity of individuals in each group, is the focus of our analyses. Convergence of all chains was assessed using (obtained by comparing two parallel chains in WinBUGS) (Gelman & Rubin, 1992), and a method developed by Geweke (1992), which compares the first 10 percent of each chain with the last 50 percent. All chains were found to be at convergence after 10,000 iterations using (all <1.1). However, the MCMC samples for the positive condition in the non- melancholic group did not converge using the criteria of Geweke and, thus, a burn-in of 5,000 samples (i.e., where the first 5,000 samples are discarded, followed by a further 10,000 samples being drawn) for all positive chains across all groups was used.

Page | 45

The modelling approach allows for inspection of a range of posterior distribution statistics. The 95% credible interval (CI) estimates are reported for each of the individual distributions of bias and discriminability, which are compared according to a 95% highest posterior density (HPD) interval difference (Plummer, Best, Cowles, & Vines, 2006). A 95% HPD group difference estimate (HPDd) that does not contain zero is considered “significantly different”. This latter approach is similar to that suggested by Lindley (1965). We also visualise the violin plots for each condition in each group for each signal condition (using the MCMC distributions) (see Appendix 1; Figure A1- 1). Violin plots marry the traditional box plot, representing the interquartile range, with smoothed distributional characteristics of the samples (Hintze & Nelson, 1998).

2.4 Results

2.4.1 Sample Characteristics Group differences on demographic variables were assessed using independent group t- tests for continuous variables and chi-square statistics for categorical variables (α set at 0.05). Group characteristics including age, gender, depression severity, STAI scores, GAF, estimated IQ and medication status (SSRI and/or other medication) are presented in Table 2-1.

There were no group differences for age, gender or estimated IQ. Both depressed groups had fewer years of education than the control group but did not differ from each other. Depression severity as measured by the QIDS, as well as STAI-State and STAI- Trait scores did not differ between melancholic and non-melancholic depression groups but, as anticipated, each clinical group differed significantly from the control group. Consistent with their classification, the melancholic group had significantly higher (CORE-rated) psychomotor disturbance scores compared to the non-melancholic group, as well as lower GAF scores (and also in comparison to controls). Lower rates of current SSRI usage and higher rates of medication other than SSRIs (e.g., tricyclics,

Page | 46

serotonin noradrenaline reuptake inhibitors (SNRIs)) were observed in the melancholic group compared to the non-melancholic group. Raw counts of hits, misses, false alarms and correct rejections across signal conditions in the AGN task are displayed in Table 2- 2, along with signal detection sensitivity values, indexed by d-prime (d’).

Visual inspection of these frequency tables suggests higher hit rates in the non- melancholic and control groups compared with the melancholic group across positive and neutral conditions. The d′ statistics (indexing the separation of the signal and noise response distributions) also point to a reduction in signal detection sensitivity across all signal conditions in the melancholic group compared to the non-melancholic and control groups. To examine AGN task performance more formally we next report on the hierarchical Bayesian modelling.

Page | 47

Table 2-1: Clinical and demographic characteristics of melancholic (Mel), non-melancholic (N-Mel) and control groups. Group Group Contrast Test Variables Melancholic N-Mel Control Mel vs N-Melǂ Mel vs Control‡ N-Mel vs Control† Age 41.7 (13.5) 42.4 (9.1) 38.6 (15.2) ǂ t = −0.19, p = 0.85 ‡ t = 0.68, p = 0.85 † t = 0.96, p = 0.34 2 2 2 % Female 65% 65% 55% ǂ χ = 0.00, p = 1.00 ‡ χ = 0.42, p = 0.52 † χ = 0.42, p = 0.52 Years of education 14.4 (2.7) 14.1 (2.7) 17.1 (3.7) ǂ t = 0.43, p = 0.41 ‡ t = −2.66, p <0.01 † t = −3.00, p <0.01 Estimated IQ 108.1 (8.5) 109.2 (6.4) 114.3 (11.8) ǂ t = −0.46, p = 0.65 ‡ t = −1.91, p = 0.06 † t = −1.70, p = 16 QIDS-SR 16.6 (4.0) 16.6 (4.4) 0.9 (1.2) 0.10ǂ t = 0.00, p = 1.00 ‡ t = 16.92, p <0.01 † t = 15.39, p <0.01 STAI-State 49.1 (15.3) 48.3 (9.7) 28.7 (8.1) ǂ t = 0.19, p = 0.85 ‡ t = 5.28, p <0.01 † t = 7.00, p <0.01 STAI-Trait 59.1 (11.4) 62.9 (8.3) 34.2 (7.4) ǂ t = −1.20, p = 0.24 ‡ t = 8.17, p <0.01 † t = 11.55, p <0.01 CORE (Non-interactiveness) 4.4 (3.3) 0.9 (1.7) - ǂ t = 4.28, p <0.01 CORE (Retardation) 4.9 (4.1) 1.6 (2.9) - ǂ t = 3.00, p <0.01 CORE (Agitation) 1.5 (2.4) 0.2 (0.7) - ǂ t = 2.38, p < 0.05 CORE Total 10.8 (7.6) 2.7 (4.4) - ǂ t = 4.14, p <0.01 GAF 54.3 (15.6) 67.5 (9.7) 94.5 (2.2) ǂ t = −3.23, p <0.01 ‡ t = −11.43, p <0.01 † t = −12.17, p 2 SSRI% yes (n) 30% (6) 40% (8) - <0.01ǂ χ = 9.69, p <0.01 2 Other med%yes (n) 80% (16) 35% (7) - ǂ χ = 27.22, p <0.01

Page | 48

Table 2-2: Frequencies of hits (H), misses (M), false alarms (FA) and correct rejections (CR) across signal valence conditions and group on the go/no-go task. d′ is presented as a function of hit and false alarm rates.

Condition and Group H M FA CR Hit Rate False Alarm Rate d′ Positive Melancholic Depression 911 169 100 979 0.84 0.09 2.33 Non-Mel Depression 979 101 141 939 0.91 0.13 2.47 Control 964 115 134 945 0.89 0.12 2.40 Negative Melancholic Depression 1007 71 88 991 0.93 0.08 2.88 Non-Mel Depression 1041 39 72 1007 0.96 0.06 3.31 Control 1008 70 83 997 0.93 0.07 2.95 Neutral Melancholic Depression 809 271 213 864 0.75 0.20 1.52 Non-Mel Depression 861 216 221 858 0.80 0.20 1.68 Control 908 171 177 903 0.84 0.16 1.99 (NB: within conditions, not all participant’s scores summed to the total number of trials because pre-emptive responses were not recorded - however, d′ was calculated for the total number of trials)

Page | 49

2.4.2 Group Effects of Mean Bias and Discriminability Results from the hierarchical modelling showed a significant difference between melancholic and non-melancholic groups in terms of their mean bias to positive signal conditions (HPDd = 0.022 – 0.554). As shown in Figure 2-3, this difference was driven by the melancholic group favouring ‘no’ responses (positive mean bias values) and the non-melancholic group favouring ‘yes’ responses (negative mean bias values), but neither depressed group differed from the control group on this measure. This observed differential performance between depressed groups is further illustrated in Figure 2-4, with the left panel displaying the mean posterior estimates of bias for all groups and the right panel showing the posterior probability for the difference between melancholic and non-melancholic groups. The violin plots (Figure A1-1) provide additional means for inspecting these differences. Individual parameter estimates for both bias and discriminability across the differing signal conditions are provided in Figure 2-5. Visualising the data in this manner suggests that the group results are a valid representation of the inter-subject variability and are not driven by outliers, clustering or multiple sub-groups.

Page | 50

Figure 2-3: Mean and standard deviation (SD) posterior estimates of bias and discriminability across groups. Legend: Mel = melancholic, Non-Mel = non-melancholic. ** denotes difference between melancholic and non-melancholic groups. *** denotes difference between melancholic and control groups.

Page | 51

In terms of mean discriminability, there was a significant difference between melancholic and control groups during neutral trials (HPDd = -1.020 – -0.042). This difference in posterior estimates of discriminability is consistent with impaired discrimination capacity for neutral signals in the presence of both positive and negative noise trials. The non-melancholic group appeared slightly less optimal than the control group in terms of discriminability to neutral signals, but this effect did not reach significance. Hierarchical analyses of the mean bias for negative and neutral signal conditions, and mean discriminability for positive and negative signal conditions, did not reveal any differences across groups.

Figure 2-4: Left: Posterior distributions of MCMC sampling for the mean bias to positive signal trials for each group. Right: Posterior density of the estimated difference between melancholic and non-melancholic groups for the bias to positive signal trials – dashed line indicating the crossing of the difference distribution at zero.

Page | 52

2.4.3 Comparing Standard Deviation Model Estimates Posterior estimates of the standard deviations for bias and discriminability across all stimulus conditions are also shown in Figure 2-3. Significantly increased standard deviation values were found in the melancholic group compared to non-melancholic and control groups on positive signal trials. A similar pattern was found for the negative signal condition, with larger standard deviation estimates found in the melancholic group compared to the non-melancholic group. No group effects were found for neutral bias or for any of the signal conditions for discriminability. The increased standard deviations in the emotional signal blocks in the melancholic group is indicative of greater variability of mean bias values across participants, thus reflecting a significantly broader distribution of bias values compared to those values observed in the other groups.

Figure 2-5: Individual parameter estimates for bias and discriminability to positive, negative and neutral signal conditions across groups.

2.4.4 Sensitivity and Robustness Analyses of Model Posteriors We conducted a robustness check of the results using narrower priors on mean and gamma (precision) of bias and discriminability and, additionally, a sensitivity analysis by uncollapsing noise conditions. These additional analyses are presented in Appendix 1. Briefly, for the hierarchical model of positive and neutral signal conditions, Page | 53

differences in mean bias and discriminability were more robust to changes in prior distributions than were the standard deviation differences. The majority of effects were robust when the noise conditions were unpooled. The loss of some significant effects is consistent with the loss of power that arises when trials are split and not pooled. Nonetheless, these additional analyses do highlight that bias to positive, and discriminability to neutral, signal trials may be influenced by differing noise conditions.

2.5 Discussion

The hierarchical Bayesian SDT model adopted in this study revealed that signal detection processes in melancholic and non-melancholic depression are significantly influenced by stimulus type and individual subject variability. Our modelling approach allowed interrogation of the neuropsychological data at two levels: the mean results across individuals in specific groups, and the heterogeneity of the groups themselves, across the psychophysical constructs of bias and discriminability. In terms of mean differences, we observed that the melancholic participants overall were less sensitive to detecting emotional signals, while non-melancholic participants displayed more liberal responding to emotional signal blocks. This provides support for our prediction that the melancholic group would display difficulty in detecting signal trials from noise trials on emotional word blocks. Optimal responding was also found to be reduced in the melancholic group compared to the control group, as evidenced by decreased mean discriminability on neutral signal blocks. In terms of subject heterogeneity, we found that there was greater inter-subject variability of the bias estimates for the emotional signal conditions in the melancholic group, which indicates divergent bias estimates across individuals. Visualising the range of individual participants’ responses (Figure 2- 5) argues against this effect being driven by outliers. Further, changes to the precision of prior distributions had little impact on the observed mean differences, suggesting that these findings are highly robust. Despite this, however, when differing noise conditions were examined, there was a moderate effect, with some previous significant differences for specific signal conditions no longer remaining significant. The observed impact of differing noise conditions thus warrants further consideration in future psychophysical Page | 54

research using differing emotional and non-emotional stimuli. The main findings allow for specific neurocognitive models to be advanced with regard to depression and its sub- typing: namely the potential to gain insight into underlying psychophysical mechanisms across individuals and whether the depression is melancholic or non-melancholic in type, and highlight several important issues in the analysis and interpretation of neurocognitive data.

Our findings align with the commonly held notion (Austin et al., 2001) that those with melancholic depression exhibit cognitive deficits that are observable during task performance. However, we add the observation that those with non-melancholic depression may also be impaired in their ability to perform ‘optimally’ on cognitive tasks such as the AGN. The observed trend of less optimal responding in non- melancholic depression did not, however, reach significance but may benefit from review in future studies. While research using the AGN task in depression (Murphy et al., 1999) positions it as a measure of ‘inhibitory control’, the analytic methods previously employed often prevent interpretation beyond a continuum of impairment (e.g., number of errors on a task). The current study is the first, to our knowledge, to utilise an affective go/no-go task in sub-types of depression. In doing so – and through analysis of the data using hierarchical Bayesian SDT – the findings offer an increased understanding of the sensitivity and discriminability capacity of individuals with depression, and highlight the importance of examining for apparent dysfunction with more refined models. Recently, Schulz and colleagues (2007) examined the convergent validity of emotional and non-emotional go/no-go tasks and concluded that, together, they offer “moderate capacity” for probing the neuropsychological construct of behavioural inhibition. Those authors emphasised the need for testing emotional and non-emotional signal detection mechanisms in affective disorders to clarify underlying cognitive-emotional contributions. The diverging sensitivities across emotional and non-emotional conditions in the depressed groups in this study, along with a lack of discrimination to neutral signals in melancholia, suggests that a range of cognitive mechanisms may be involved in responding to differing stimuli.

Page | 55

Across neuropsychological studies of depression it is evident that no single cognitive deficit model can be applied to specific groups, due to the inherent heterogeneity of the disorder class. However, in light of the current findings of discrepancies in bias between signal conditions, it might be possible that set-shifting impairments – as previously reported in depression (Murphy et al., 1999), and more specifically in melancholic depression (Michopoulos et al., 2008) – play an important role. While not explicitly specified in the current study, it is conceivable that neuropsychological constructs such as disrupted attention set-shifting and perseveration underlie the observed results. The melancholic sub-type has been shown to be differentiated to non-melancholic depression on the basis of response selection performance (Rogers, Bellgrove, Chiu, Mileshkin, & Bradshaw, 2004), where performance on compatible and incompatible trial types (e.g., stimulus-response compatibility tests) reflect cognitive strengths and weaknesses. As a rule these studies have not specified psychophysical functions of stimulus sensitivity and discrimination capacity, and have tended to report broader metrics of performance such as numbers of hits and misses across subjects. From a cognitive standpoint there is likely to be significant benefit in modelling performance-related psychophysical mechanisms (i.e., through SDT and similar analyses) in depressive disorders and then next establishing whether this provides insight towards any neurocognitive disease mechanisms. We argue that the cognitive deficits observed in different types of depression can be conceptualised in such a way so as to explain impairments in emotion-bound optimal decision-making.

Prior research on sensory processing offers further insight into the findings of decreased sensitivity to emotional signals and poorer discrimination to neutral signals in melancholia. Knill and Pouget (2004) suggest that perception of one’s environment is influenced by the likelihood of the presence or absence of relevant stimuli given an individual’s past experience (i.e., perceptual priors) with that stimuli. These factors contribute to the relative uncertainty over one’s environment, and allow inference regarding the causes of percepts. We propose that the low sensitivity to emotional

Page | 56

signals and lack of discriminability observed across melancholic participants may be a result of inefficient sensory integration – possibly resulting from constructs such as inefficient cognitive control mechanisms (Ridderinkhof, van den Wildenberg, Segalowitz, & Carter, 2004). This, in turn, may be a function of ‘inflexible’ priors (e.g., negative cognitive biases) and unsuccessful updating (e.g., such as perseveration due to a failure of emotional inhibition) (Hauser, 1999). The observation that the non- melancholic participants responded more frequently to noise trials on emotional signal blocks also suggests that they too are less sensitive to fluctuations in the emotional environment. Such erroneous judgements could be due to emotional processing biases in depression, a factor that has been acknowledged in accounting for decision-making impairments (Paulus & Yu, 2012). Research into the probabilistic nature of attention and decision-making (Glimcher, 2003; Huys & Dayan, 2009; Yu, Dayan, & Cohen, 2009) suggests diverse mechanisms underlying optimal judgement. Neurobiologically, probabilistic learning paradigms have been used to examine human cognition (Chamberlain et al., 2006), with findings pointing to distinct roles of serotonin and noradrenaline in learning and inhibition. Both neurochemicals have long been implicated in depressive disorders (Schildkraut, 1965) and may be of relevance to understanding the differences observed between and within depressive sub-types.

In addition to the behavioural changes within individuals, research using Bayesian inference has also highlighted the importance of perceptual variation between individuals. Recent theoretical work on perceptual uncertainty advocates the utility of modelling trial-by-trial updating across individuals in a Bayesian framework (Mathys, Daunizeau, Friston, & Stephan, 2011), which we argue to be of significant benefit in conceptualising the current findings from the signal detection task. The hierarchical modelling using MCMC in the current study yielded estimates of the standard deviation of both bias and discriminability performance (i.e., the extent of the differences between measured individuals). The increased standard deviations in the melancholic group on emotional bias suggests differential individual performance profiles when compared to non-melancholic and control groups. This could be due to a range of non-cognitive

Page | 57

factors in an otherwise homogeneous group or may alternatively reflect divergent cognitive strengths and weaknesses, which we now consider.

Several lines of research have indicated that melancholic depression is associated with diurnal variation of mood (Paykel, 1971), with such variation thought to impact on neuropsychological performance across the day (Moffoot et al., 1994). Clinical depression with diurnal variation has also been found to result in differential performance on accessibility and recall of positive and negative (self-related) experiences (Clark & Teasdale, 1982), with positive memories more likely to be retrieved when depression is less severe (i.e., in the afternoon/evening). Varying biological influences such as cortisol hypersecretion – shown to be specific to melancholia (Carroll, 1982) – may play a pivotal role in modulating cognitive function in depression as previously suggested (Rubinow et al., 1984). Such factors are important considerations with respect to the current findings given participants were not all tested at the same time of day. Furthermore, inconsistent medication regimens across individuals within and between groups may contribute to individual differences of emotional processing biases (Harmer, Goodwin, & Cowen, 2009), thus possibly dampening the neurocognitive effects of negative mood and affect in certain individuals. If such factors were found to be unrelated – and individual differences were indeed evident upon replication – it is conceivable that melancholia (due to the observed variation) may be able to be portrayed as comprising several distinct sub-types (e.g., functional and structural melancholia) as suggested by Parker and Hadzi-Pavlovic (1996). Factors such as family history, age-of-onset, presence/absence of neuropathological changes and cardiovascular disease would need to be clarified for any such sub-typing to be put advanced within the current context. Given the increased age of our sample, neuropathological changes in some individuals cannot be excluded. Prior to these issues being clarified, however, the utility the findings from this study are first and foremost in their ability to inform psychophysical models of depression, with several caveats.

Page | 58

As indicated above, there are several study limitations. Firstly, the analyses did not interrogate trial-to-trial variability. Thus, the supposed dynamics of sensory integration (stimulus-response updating), as previously put forward (Mathys et al., 2011), could not be quantified in this study. In addition, our signal detection task did not allow for further analysis of aspects of emotional decision-making (i.e., specific attentional control mechanisms) beyond bias and discriminability functions. Such specific limitations, if overcome in future studies, would provide a more refined model of attention and decision-making in clinical conditions such as melancholia. It is therefore proposed that future work should attempt to examine the utility of dynamic models of attention and decision-making in light of changing emotional environments, along with key clinical variables, to further establish the mechanisms by which perception and action interact in depression.

There has been an upsurge of interest in framing cognitive function in psychiatric conditions in probabilistic terms (Montague, Dolan, Friston, & Dayan, 2012; Paulus & Yu, 2012), precipitated by research in cognitive and computational sciences that, in health, humans respond optimally in their environments. The signal detection approach used in the current study offers insight into the optimal response parameters of those with depression, and extends previous suggestions of ‘emotional response biases’ in depression through psychophysical modelling. Further, the use of hierarchical modelling in the current study allowed estimation of the most probable response distributions, and is an advance on previous (e.g., frequentist) attempts that aim to elucidate neuropsychological dysfunction in depressed groups through group averaging approaches. Future studies should attempt to clarify how different cognitive processes operate across different individuals. Such work should also aim to provide a more detailed characterisation of perceptual (probabilistic) sensory updating across changing environments in depression, whilst recognising the fundamental role pre-existing cognitive biases play in response to environmental demands.

Page | 59

Chapter 3: The insula state of melancholia: Disconnection of interoceptive and attentional networks

3.1 Abstract

Background: Melancholia has long been considered to have a biological weighting but evidence for its specific neurobiological origins is limited. The distinct neurocognitive, psychomotor and mood disturbances observed in melancholia do, however, suggest aberrant co-ordination of frontal-subcortical circuitry, which may best be captured through analysis of complex brain networks. Whilst neuroimaging research has traditionally focussed on task-related activity, patients with melancholia also report a distinct and intrusive dysphoric state during internally generated thought. Here, we investigate the effective connectivity between spontaneous (resting state) brain networks in melancholia, focussing on those networks underlying attention and interoception.

Methods: Resting state fMRI (rs-fMRI) data were acquired from 16 melancholic, 16 non-melancholic, and 16 control participants at a hospital-based research institute. We identified five canonical resting state networks (default mode, executive control, left and right frontoparietal attention and bilateral anterior insula) and inferred spontaneous interactions amongst these using DCM. Graph theoretic measures of brain connectivity, namely in- and out-degree of each network, and edge connectivity between regions, comprised our principal between-group contrasts.

Results: Melancholia was characterised by a pervasive disconnection involving anterior insula and attentional networks, compared to those in the control and non-melancholic depressive groups. Decreased effective connectivity between the right frontoparietal and Page | 60

insula networks was present in those with melancholic compared to non-melancholic depression. Reduced effective connectivity between insula and executive networks was found in those with melancholia compared to healthy controls.

Conclusions: We report reduced effective connectivity in resting state fMRI between key networks involved in attention and interoception in melancholia. We propose that these underlie the impoverished variety and affective quality of internally generated thought in this disorder.

3.2 Introduction

Despite advances in pursuing the neurobiological causes of ‘clinical depressive’ conditions, the literature is characterised by divergent findings, likely reflecting their heterogeneity and varying aetiologies. One, melancholia (previously termed endogenous depression), has long held consistent ascriptions – being genetically weighted, having prominent biological perturbations, evidencing over-represented clinical features, and showing a greater response to physical therapies than to psychotherapy (Parker et al., 1996). As psychiatry strives towards a nosology based upon genetic, behavioural and neurobiological criteria (Insel et al., 2010), melancholia arguably represents a canonical test case (Parker et al., 2010).

Historical failure to identify specific neurobiological correlates of melancholia is consistent with recent advances in cognitive neuroscience that regard the brain as a complex network (Sporns et al., 2004), whereby psychiatric conditions reflect changes in functional integration rather than perturbations within an isolated region (Menon, 2011). Large-scale brain networks supporting mood regulation, interoception and cognition (e.g., concentration and attention) are thus likely candidates for furthering understanding of melancholia’s neurobiology. The “salience network” (Seeley et al.,

Page | 61

2007), encompassing the anterior insula (AI) and dorsal ACC, embodies interactions among interoceptive and cognitive processes that are endowed with an affective or emotional aspect – and is hence a candidate target for studying melancholia. For example, following deep brain stimulation of the sgACC, treatment resistant depressed patients showed increased blood flow in the AI (Mayberg et al., 2005), highlighting its involvement in mood disturbances. The AI participates in visceral and somatic sensory processing underpinning mood regulation (Augustine, 1996), control of interference during working memory updating (Levens & Phelps, 2010), decision-making (Xue, Lu, Levin, & Bechara, 2010) and mediating exchange of salient information to other brain regions, particularly those involved in attention (Menon & Uddin, 2010). Frontoparietal attention networks, and their relationship to emotional and affective processes, have also been well characterised (Lindquist & Barrett, 2012; Markett et al., 2014; Vincent, Kahn, Snyder, Raichle, & Buckner, 2008). As disturbances in concentration and attention are well documented in melancholia (Austin et al., 1999) – the ‘psycho’ component of its key feature of psychomotor disturbance – examining attention networks provides an opportunity to model this disorder’s neurobiology. In addition to the distinctive anergia that disrupts task completion, patients with melancholic depression report a pervasive dysphoria at-rest and difficulties switching between the usual myriad of internal thought processes (e.g., memory, planning, daydreaming). We hence sought to study the integration of resting state cognitive and emotional networks in those with melancholic depression and in appropriate control sub-groups.

Failures of functional integration in the brain can be tested by analyses of functional and effective connectivity. Studies of functional connectivity (statistical correlations between remote regions) in major depression have shown increased coupling between the default mode network (DMN) and the ACC (Greicius et al., 2007; Sheline et al., 2009), which is associated with disrupted cognitive processing (Davey et al., 2012). Analyses of effective connectivity (inferences regarding causal interactions) (Friston et al., 2003) indicate decreased connectivity across divergent regions, particularly those involved in attention, salience and executive processing (Desseilles et al., 2011; Lu et

Page | 62

al., 2012; Schlosser et al., 2008). To date, the specificity of such brain network dysfunction in depressive sub-types has not been addressed.

Here, we examine for distinct neural attributes of spontaneously generated thought in melancholia by analysing rs-fMRI data. We employed ICA to identify cortical systems or modes underlying cognitive processes including attention, salience, executive function and internally generated thought. We then applied sDCM to infer the patterns of directed effective connectivity between these modes, hypothesising that melancholia would be characterised by disturbed effective connectivity in networks subserving attention, salience and interoception.

3.3 Methods

3.3.1 Participants Not all participants studied in Chapter 2 were included in the current chapter. Hence, we here provide full details of recruitment, diagnostic approach and demographic and clinical information for the current cohort. Participants included 16 melancholic and 16 non-melancholic depressed patients, recruited through the Black Dog Institute depression clinic. All consecutively assessed patients aged between 18 and 75 years meeting a diagnosis of unipolar major depression were referred for possible study inclusion. Towards the end of the recruitment process, some sub-sampling of the non- melancholic group was mandated to ensure age and gender matching. Written informed consent was obtained and no monetary incentive provided. A control group of 16 participants, who disavowed any current and/or past mood and/or psychotic illness, was recruited through the community. All participants were screened for current and past mood and psychotic conditions with the MINI (Sheehan et al., 1998). Exclusion criteria included (hypo)mania or psychosis, current or past drug or alcohol dependence, neurological disorder or brain injury, a WTAR score of below 80 and ECT in the preceding six months. Depression severity was quantified with the QIDS-SR16 (Rush et

Page | 63

al., 2003). Participants also completed state and trait versions of the STAI (Spielberger, 1983), and overall functioning was measured by the GAF scale (APA, 2000). We used the CORE measure (Parker et al., 1994) to assess for psychomotor signs in the depressed participants, with this measure having non-interactiveness (i.e., cognitive slowing), retardation and agitation sub-scales.

3.3.2 Depression Sub-typing Approach Clinical diagnoses of melancholic or non-melancholic depression were made by psychiatrists weighting previously detailed criteria (Parker et al., 2010; Taylor & Fink, 2006). For a diagnosis of melancholia, two compulsory criteria were required: (A) psychomotor disturbance (expressed as motor slowing and/or agitation); and (B) an anhedonic mood state. In addition, five of the following nine clinical features were required (and met) in all assigned melancholic patients: (1) concentration and/or decision-making impairment; (2) non-reactive affect; (3) distinct anergia; (4) diurnal mood variation – being worse in the morning; (5) appetite and/or weight loss; (6) early morning wakening; (7) no preceding stressors accounting for the depth of the depressive episode; (8) previous good response to adequate antidepressant therapy; and (9) normal personality functioning. Patient-specific data are provided in Appendix 2 (Table A2-1).

3.3.3 Q-sort Methodology to Derive Prototypic Melancholic Symptom Scores To further characterise our melancholic group, a 32-item Q-sort (Parker et al., 2009) was completed by all clinical participants assessing the relative weighting of melancholic and non-melancholic prototypic features.

3.3.4 Imaging Data Acquisition and Analysis All participants underwent a 6 ¼ minute rs-fMRI scan, with resulting data pre- processed to mitigate head motion and spatially normalised to the Montreal Neurological Institute (MNI) atlas in Statistical Parametric Mapping (SPM8) software Page | 64

(Friston et al., 1995). Full acquisition, pre-processing and analysis details for the current analysis are provided in Appendix 2. We identified key neuronal systems showing spontaneous activity fluctuations in a subject-specific fashion using ICA. High dimensional spatial ICA was performed using MELODIC implemented in FSL, yielding 70 modes of neuronal, physiological and artefactual origin (Beckmann & Smith, 2004). We chose five of these components as being most likely to reflect spontaneous mental activity, namely the DMN, executive control (EXC), bilateral anterior insula (INS), and left (LFP) and right frontoparietal (RFP) attention modes (see Figure 3-1) (Damoiseaux et al., 2006). These were identified by optimising the spatial overlap with pre-published resting state and cognitive maps (Smith et al., 2009) and ensuring appropriate component selection with visual inspection. The inclusion of left and right frontoparietal attentional modes sought to capture ‘cognitive’ attention (i.e., cognitive control systems), which have been shown to correspond to higher-order cognitive domains (Markett et al., 2014), and are distinct from those involved in attending to visual context (e.g., dorsal attention network) (Corbetta, 1998). We then specified and estimated subject-specific sDCMs of effective connectivity among the components or modes (Li et al., 2011) using their associated time series. In brief, DCM assesses effective connectivity among neuronal populations by combining dynamic models of neuronal states and detailed biophysical models of haemodynamics. Inferences about these neuronal states – and their interactions – are obtained by inverting these generative models using standard (variational) Bayesian techniques, allowing for estimation of the influence of one brain region over another. Whilst DCM has been widely employed to study task-driven effects by modelling endogenous fluctuations (Friston et al., 2003), recent innovations now permit inferences on effective connectivity in rs-fMRI (Breakspear, 2013; Friston, Li, Daunizeau, & Stephan, 2011; Li et al., 2011). Our resulting fully connected sDCMs were then optimised to yield a sparse and parsimonious (minimally complex) representation of the network for each participant (Friston et al., 2011). These representations (directed and weighted graphs) were used for between group comparisons, using graphic theoretic measures (average degree and edge strength) as group dependent variables. Group differences were corrected for multiple measures (5 modes; 20 edges) after accounting for inter-subject correlations

Page | 65

(Sankoh, Huque, & Dubey, 1997), with further details of these adjustments provided in the Results.

Figure 3-1: Analysis pipeline illustrating the use of ICA spatial maps to inform sDCMs. Optimised model parameters from the sDCMs were used for between-group comparisons.

3.4 Results

3.4.1 Clinical and Demographic Comparisons The melancholic, non-melancholic and control groups did not differ significantly by age or gender (Table 3-1). Age ranges for the groups were 20-52 (melancholic), 21-56 (non- melancholic), 22-75 (controls) – with two controls aged over 60. Two non-melancholic participants and one healthy control participant were left-handed. The control group

Page | 66

reported more years of education compared to the melancholic group (t = -2.15, p = 0.04), and their estimated IQ was higher than both melancholic (t = -2.69, p = 0.01) and non-melancholic groups (t = -3.12, p < 0.01). The depressed groups did not differ by years of education, WTAR scores, depression severity or state and trait anxiety scores. Consistent with the diagnostic primacy of psychomotor disturbance, the melancholic group had higher scores compared to the non-melancholic group on all CORE sub- scales (non-interactiveness: t = 3.05, p < 0.01; retardation: t = 3.66, p < 0.01; agitation: t = 2.55, p < 0.05) and higher total CORE scores (t = 3.93, p < 0.01). All groups differed on the GAF with the melancholic group having the most severe functional impairment, followed by the non-melancholic and then the control group participants (see Table 3- 1). More non-melancholic participants reported being on a SSRI antidepressant drug compared to melancholic participants (χ2 = 5.24, p < 0.05), while a higher proportion of melancholic participants were on non-SSRI medications (χ2 = 8.13, p < 0.01). More melancholic participants reported being on an antipsychotic medication (χ2 = 4.57, p < 0.05).

Analysis of the Q-Sort symptom profiles showed that the seven items best differentiating melancholic from non-melancholic participants were (in ranking): feeling somewhat paralysed doing basic things (mean difference = 3.67); finding it difficult to do basic things (0.97); feeling physically slowed (0.79); anticipatory anhedonia (0.57); energy worse in the morning (0.45); mood non-reactivity (0.43); and thinking slowed (0.27).

Page | 67

Table 3-1: Demographic and clinical characteristics across melancholic, non-melancholic, and control groups. Group Group Comparison

Test Variables (Mean (SD) reported Non- Melancholic vs. Non- Melancholic vs. Non-melancholic vs. except where denoted as %) Melancholic Melancholic Control melancholic Control Control Age 38.0 (9.94) 40.44 (10.73) 43.75 (14.10) t = -0.68, nsa t = -1.33, ns t = -0.75, ns % Female (n) 50% (8) 62.5% (10) 56.3% (9) χ2 = 0.51, ns χ2 = 0.13, ns χ2 = 0.13, ns Years of education 14.81 (3.31) 15.88 (2.44) 17.44 (3.58) t = -1.03, ns t = -2.15, p < 0.05 t = -1.44, ns Estimated IQ 107.93 (12.40) 108.19 (9.96) 117.94 (7.55) t = -0.06, ns t = -2.69, p < 0.05 t = -3.12, p < 0.01 QIDS-SR16 16.69 (4.22) 15.06 (4.10) 1.19 (1.51) t = 1.10, ns t = 13.82, p < 0.001 t = 12.68, p < 0.001 STAI-State 49.73 (16.18) 46.25 (12.37) 25.44 (6.52) t = 0.68, ns t = 5.42, p < 0.001 t = 5.95, p < 0.001 STAI-Trait 55.00 (13.33) 62.88 (8.43) 31.19 (6.45) t = -1.98, ns t = 6.27, p < 0.001 t = 11.94, p < 0.001 GAF 58.13 (7.04) 69.38 (6.29) 95.00 (0.00) t = -4.77, p < 0.001b t = -20.95, p < 0.001 t = -16.29, p < 0.001 CORE (Non-interactiveness) 3.44 (3.10) 0.75 (1.69) - t = 3.05, p < 0.01 - - CORE (Retardation) 4.88 (3.40) 1.06 (2.41) - t = 3.66, p < 0.001 - - CORE (Agitation) 0.69 (0.00) 0.00 (0.00) - t = 2.55, p < 0.05 - - CORE Total 9.00 (6.12) 1.81 (4.02) - t = 3.93, p < 0.001 - - Current Medications Nil medication % yes (n) 6.3% (1) 31.3% (5) - χ2 = 3.28, ns - - SSRI % yes (n) 12.5% (2) 50% (8) - χ2 = 5.24, p < 0.05 - - Any medication other than SSRI %yes (n) 81.3% (13) 31.3% (5) - χ2 = 8.13, p < 0.01 - - Dual-action antidepressant (e.g., serotonin 50% (8) 31.3% (5) - χ2 = 1.17, ns - - noradrenaline reuptake inhibitor) Tricyclic or monoamine oxidase inhibitor 25% (4) 12.5% (2) - χ2 = 0.82, ns - - Mood stabilizer (e.g., lithium or 6.3% (1) 12.5% (2) - χ2 = 0.37, ns - - valproate/divalproex) Antipsychotic 25% (4) 0% (0) - χ2 = 4.57, p < 0.05 - - a ns not significant; b uncorrected p-values for between group comparisons; significant differences (p < 0.05) in bold.

Page | 68

3.4.2 Overall Network (Node Degree) Effects Effective connectivity among the 5 ICA modes was summarised with their group- average in- and out-degree (Figure 3-1), and assessed for significance after adjusting for correlations between dependent variables (average r = 0.4) corresponding to a threshold of α < 0.019.

The melancholic group had markedly diminished inward connectivity (i.e., average in-degree) of the RFP (attentional) mode from the other four modes (Table 3-2), in comparison to the non-melancholic group (Z = 2.93, p < 0.004). In particular, single- subject networks in the melancholic group contained, on average, only one incoming connection, approximately half that of the non-melancholic group. Melancholic participants had a significantly diminished outgoing influence of the INS component upon other modes compared to healthy controls (Z = 2.38, p < 0.017). The out-degree influence of the INS mode in the non-melancholic participants was closer to that of healthy controls. A trend-level effect was present for the out-degree of the LFP, with melancholic participants showing fewer outgoing connections compared to non- melancholic participants (Z = 2.39, p = 0.021).

Page | 69

Table 3-2: Between group contrasts of incoming and outgoing ‘node degree’ for attentional and insula networks. Group Melancholic Non-Melancholic Control Degree Direction and Mann- Region Mean SD Mean SD Mean SD Whitney U Z pa Degree In RFP 1.00 1.03 2.06 1.00 1.63 0.72 203.00 2.93 0.004b Degree Out INS 1.81 1.28 2.56 1.26 2.81 0.91 189.00 2.38 0.017c Degree Out LFP 1.50 1.09 2.31 0.87 2.06 0.85 188.50 2.39 0.021† a Significant differences after (Sidak) FWE-adjustment and correction for correlated outcome variables (p < 0.02), shown in bold. b Contrast between melancholic vs. non-melancholic groups. c Contrast between melancholic vs. control groups. † Trend-level difference between melancholic and non-melancholic groups.

Page | 70

3.4.3 Specific Effective Connectivity (Group-Averaged Edge Degree) Effects Effective connectivity between each pair of the 5 ICA modes was summarised with the corresponding group-averaged value of the edge degree (the proportion of times each specific edge was present in the subject-specific DCMs). Multiple Pearson chi square tests were performed on these 20 (non-self) edges, and assessed for significance after adjusting for correlations between dependent variables (average r = 0.21) corresponding to a significance threshold of p < 0.0047. There were two significantly reduced connections in the melancholic group (Table 3-3/Figure 3-2), both involving outgoing edges from the INS. The specific influence of INS component on the RFP mode (INS → RFP) was weaker in comparison to non-melancholic participants ( 2 = 8.13, p = 0.0043), while INS → EXC connectivity was weaker in comparison to the control group ( 2 = 8.96, p = 0.0027). Both effects are consistent with the decreased overall outgoing influence of the insula in the melancholic group.

Table 3-3: Between group edge connectivity differences among spatially distributed brain networks. From Region → Region Proportion Present Proportion Absent Statistic Melancholic Non-Mel χ2 p-value LFP → DMN 0.25 0.62 4.571 0.03 RFP → INS 0.12 0.50 5.236 0.02 INS → RFP 0.19 0.69 8.127 0.004a LFP → LFP 0.75 0.37 4.571 0.03 Melancholic Control INS → DMN 0.56 0.87 3.865 0.04 INS → EXC 0.56 1.00 8.960 0.003a EXC → INS 0.75 1.00 4.571 0.03 LFP → RFP 0.25 0.62 4.571 0.03 Non-Mel Control INS → DMN 0.56 0.87 3.865 0.05 INS → EXC 0.75 1.00 4.571 0.03 EXC → INS 0.69 1.00 5.926 0.015 a Significant after (Sidak) FWE-adjustment and correction for correlated outcome variables (p < 0.0047), shown in bold. Additional group differences presented are unadjusted trend-level effects (p < 0.05).

Page | 71

Figure 3-2: Network diagram showing all possible edge connections. Significant between group differences, and trend-level effects, depicted by thick coloured lines, and coloured dashed lines, respectively. Each independent component or mode is depicted in terms of the location of its highest spatial weighting

NB: The locations of nodes are a schematic representation of the ICA modes, with each location representing the peak activation cluster of the corresponding ICA spatial map.

Page | 72

No other specific edge-wise effects surpassed the corrected statistical threshold. Interestingly, all trend-level differences between the melancholic and non-melancholic groups involved the LFP and RFP attention modes, with all most reduced in the melancholic group except for LFP self-connectivity, which was increased (Table 3-3). For completeness, we note that all trend-level effects between melancholic participants and healthy controls were also weaker in the former, and involved either the INS or the LFP components. No differences between the non-melancholic group and the healthy controls exceeded statistical significance. Trend-level effects also indicated weaker involvement of the INS mode in non-melancholic participants compared to controls.

3.4.4 Examining the Impact of Medications on Network Parameters Analysis of covariance (ANCOVA) was used to examine for possible medication effects. Patients were split into two subsets: Firstly, we compared all patients on – or not on – an SSRI. Secondly, we compared patients taking medication(s) other than SSRIs (all broad spectrum antidepressants, antipsychotics, mood stabilisers) with patients not on any such medications. We controlled for clinical group (melancholic versus non- melancholic) and analysed the main network effects observed above (in-degree for RFP, and out-degree for both INS and LFP modes). No significant differences were observed for medication on the differing network parameters. Formal statistics for all six tests are provided in Appendix 2 (Table A2-2).

3.5 Discussion

Whereas disturbances in cognitive and behavioural tasks in those with melancholia have been well documented, much of the illness burden is experienced through unpleasant and dysphoric affects during spontaneous thought. Although the quantitative documentation of these phenomena is incomplete, melancholic patients describe pervasive somatic pre-occupation, perseveration of unpleasant themes and diminished ability to set shift amongst spontaneously arising thoughts. Wernicke proposed a theory

Page | 73

of brain function (Wernicke, 1885, 1906) whereby higher-order cognitive processes were the result of integration between multiple, spatially distributed neuronal systems. Disruption to this integrative architecture (“organs of connection”) hence contributed to disorders such as aphasia and schizophrenia. This has become known as the ‘sejunction hypothesis’ and emphasises an anatomical disconnection of axonal processes. Modern day formulations of psychiatric ‘dysconnections’ emphasise synaptic abnormalities leading to a more functional and context-sensitive disintegration – aligned to our findings. A parallel may be drawn between the current findings and schizophrenia, where undirected thought often exacerbates internal preoccupation with abnormal subjective experiences. Analyses of resting state EEG and fMRI data (Breakspear et al., 2003) have proven fruitful in testing the schizophrenia “disconnection hypothesis”, yielding evidence of a “subtle but pernicious” disconnection (Friston & Frith, 1995). Using stochastic DCM, we studied the neural correlates of spontaneous thought in those with melancholia, leveraging recent advances in computational modelling to highlight a novel “disconnection syndrome” involving attentional and insula-based cortical systems, which differentiated those with melancholic and non-melancholic depression.

Whilst our ICA components were derived from rs-fMRI data, recent work has established close correspondence between resting state networks and those engaged in classic cognitive tasks (Smith et al., 2009). Indeed, we explicitly used this correspondence to identify attentional control (left and right frontoparietal) modes that covary with a variety of cognitive tasks involving modulation of attention, namely attentional disengagement and reorienting (Markett et al., 2014). This focus on dynamic networks relating cognition to spontaneous self-directed thought allows unique insights into the neurobiology of attentional disturbances in melancholia (Austin et al., 1999). For example, Chapter 2 revealed that melancholic participants exhibit a context sensitive bias during detection of emotional signals in the presence of uncertainty. It seems reasonable to assert that a qualitatively similar disturbance operates during self- directed thought, leading to an attentional bias toward, and diminished ability to shift away from, negative internal thoughts.

Page | 74

The involvement of the EXC mode is consistent with disturbances in executive function in melancholia (Austin et al., 1999), and biased decision-making in particular (e.g., Chapter 2). Our ‘executive control’ (EXC) mode was obtained by comparison to previously published ICA-derived maps that are reliably found in resting state data and in tasks requiring inhibitory control (Smith et al., 2009). However, we note that neuroanatomically, our EXC mode overlaps substantially with portions of the ventromedial PFC (vmPFC), which is endowed with a variety of affective control functions, including those related to self-reference, affect and visceromotor regulation (Roy, Shohamy, & Wager, 2012). Hence this mode likely contributes to a range of affective control functions in addition to classic cognitive control.

As noted, the AI is a key region in mood regulation, but also contributes to visceral and somatic sensory processing (Augustine, 1996) and mediates the exchange of salient information to other brain regions, particularly those involved in attention (Menon & Uddin, 2010). Recent work, building upon its central role in interoception (Seth, 2013), identified an association between weaker functional connectivity of the AI and somatic symptoms of MDD (Avery et al., 2014). Our insular component in the current study was chosen specifically to incorporate the more anterior regions implicated in interoceptive processes. We observed a diminished outgoing influence of this AI mode in those with melancholia compared to healthy controls, particularly on the executive mode, which includes key regions (e.g., vmPFC) subserving affective control mechanisms. Taken together with the diminished connectivity of the RFP mode, this provides support for our hypothesis that melancholia is associated with disruptions to key regions underpinning attention and internalised mood state regulation.

Our finding of selective deficits in connectivity from the AI is important in relation to current theoretical neurobiology treatments of emotion and self-awareness in terms of interoceptive inference (Seth, 2013). In these models, the AI is regarded as generating top-down predictions of interoceptive (bodily) states that enslave autonomic

Page | 75

(sympathetic and parasympathetic) reflexes that mediate homoeostasis. As the AI receives top-down predictions from attentional systems (e.g., the frontoparietal mode), these interoceptive predictions are themselves contextualised by high-level representations that become endowed with visceral or affective aspects (Gu, Hof, Friston, & Fan, 2013). In the setting of predictive coding, compromised outflow from the AI to RFP corresponds to a failure of ascending prediction errors to update (emotional) representations in the PFC – that may be manifest as somatic preoccupation.

Our finding of disconnectivity of the AI and frontoparietal modes in those with melancholia was inferred from modelling effective connectivity, and is thus not intended to convey evidence of a neuroanatomical disconnection (in the sejunction sense). Thus, we propose that fundamental disruptions to neuronal dynamics – rather than a coarse anatomical ‘disconnection’ – position melancholia as a disorder of connectivity. DCM rests upon relatively rapid, contextually specific changes in synaptic efficacy, classically due to glutamatergic (AMPA- and NMDA-mediated) synaptic activity (Stephan et al., 2008). This is a more subtle network perturbation than the striking correspondence between resting state networks and gross changes in network anatomy that occur in neurodegenerative disorders (Seeley, Crawford, Zhou, Miller, & Greicius, 2009). Nonetheless, the present approach is predicated on the same proposed correspondence between neurocognitive syndromes and disturbances in underlying functional networks.

Future work could build on the current findings by addressing several study limitations. Although our sample size is well suited to finding group differences of meaningful effect size (Friston, 2012), a larger cohort studied in a longitudinal setting would permit several ambiguities to be addressed. First, it will allow disentangling of more specific (but more subtle) group differences in specific mode-to-mode influences. Second, it may help elucidate the relationship between network changes and illness

Page | 76

expression. Quantifying the subjective experience of the “resting state” through a structured questionnaire could also address the subjective correlates of our observed network effects, particularly against domains such as “discontinuity of mind”, “self” and “somatic awareness” (Delamillieure et al., 2010). Third, our cross-sectional design did not allow determination of whether the observed network changes were evident prior to the onset of depressed mood or were state disturbances, limiting inferences about illness causality. Further, given our patient groups differed from healthy controls on estimated IQ, it will be important to clarify the impact of neuropsychological variables on network changes, given their relationship to resting state connectivity (Cole, Yarkoni, Repovs, Anticevic, & Braver, 2012; Song et al., 2008; van den Heuvel, Stam, Kahn, & Hulshoff Pol, 2009). Fourth, the class of prescribed medications differed between our groups, and may have contributed to the findings. Patients within the melancholic group were more likely to be on a combination of antidepressants, mood stabilisers and antipsychotics, precluding regression of single class dosage-equivalence as a nuisance variable. Post hoc analyses, using ANCOVA to control for diagnosis, suggested that the main effects reported were unlikely to have been driven by the main medication divisions amongst our clinical participants (SSRI versus no SSRI, and the presence or absence of any non-SSRI medication). We acknowledge the caveats of such a post hoc approach, and addressing this in future studies will be essential given the differential effects both antipsychotics (Miller et al., 1997) and antidepressants (Wagner et al., 2010) have on cerebral blood flow. While samples of medication-naive participants would be ideal, the severity of melancholia raises ethical issues regarding the withholding of medication for research purposes. Finally, future analyses of effective connectivity in clinical disorders should incorporate methodological innovations in stochastic DCM, including alternative approaches to group-level inferences.

In summary, we position the neurobiology of the “spontaneous dysphoria” of melancholia as a weakening of interactions amongst regions that subserve attention, mood regulation and interoception. Computational accounts of internally generated

Page | 77

thought highlight the importance of a critical, homeostatic balance between stable self- regulation and dynamic instability (Aburn, Holmes, Roberts, Boonstra, & Breakspear, 2012; Friston, Breakspear, & Deco, 2012). We propose that our findings reflect a loss of this optimal balance, undermining the adaptive role of interoception.

Page | 78

Chapter 4: Scene unseen: Disrupted neuronal adaptation in melancholia during emotional film viewing

4.1 Abstract

Background: Attentional disturbances in melancholia are ubiquitous, and often manifest as an inability to shift focus away from dysphoric internal states. Disorganisation of complex brain networks, particularly those involved in interoception and attention, may underlie impoverished attentional set-shifting between internally generated thought and exogenous emotionally salient signals in this disorder. Here, we used a naturalistic film viewing paradigm to model the change in effective connectivity between brain networks upon engagement with dynamic, emotional material.

Methods: Functional neuroimaging data were acquired from 16 melancholic, 16 non- melancholic, and 16 control participants during a resting state acquisition followed by free viewing of films (of positive, negative and neutral valences). Using independent components analysis (ICA), we identified eight distributed networks (default mode, executive control, left and right frontoparietal attention, left and right anterior insula, visual and auditory). We then inferred causal interactions amongst these brain modes using dynamic causal modelling (DCM). Three separate analyses were undertaken. Consistency of hidden neuronal states between individuals during emotional film viewing was tested using inter-subject correlation (ISC) analyses. We next interrogated effective connectivity strengths between all brain modes for resting state and emotional film viewing conditions as a measure of network integrity in our groups. We then tested for group effects on these edge strengths using network-based statistics.

Page | 79

Results: Consistent and significant ISCs were observed across most cortical systems across groups for both negative and positive film viewing conditions. However, ISCs of the left insula during positive film viewing were decreased in all groups (particularly in melancholia) and not significant. In addition, network-wide increases in effective connectivity were observed in melancholia during emotional film viewing compared to controls. Our melancholic group was further characterised by a striking increase in effective connectivity strength amongst a sub-network of regions involved in attention and interoception when shifting from rest to negative film viewing, compared to control participants who showed the inverse.

Conclusions: We propose that the current findings reflect a failure of neuronal systems in melancholia to adapt to dynamic exteroceptive emotional stimuli. The involvement of attention- and insula-based cortical systems highlights a potential neurobiological mechanism for disrupted attentional resource allocation, particularly in switching between interoceptive and exteroceptive signals, in melancholia.

4.2 Introduction

There is an inherent tendency in melancholia to focus on interoception of a predominantly dysphoric quality (Taylor & Fink, 2006). Overrepresented melancholic features such as anhedonia and attentional dysfunction (Pizzagalli, Jahn, & O'Shea, 2005) are proposed to contribute to this interoceptive focus. For example, disturbances in attention may impair reorienting away from internal emotional material, hence undermining the ability to engage in external perceptual processing. This aligns with notions of ineffective resource allocation in depression (Thomas et al., 1999). Emerging neurobiological formulations suggest that complex brain networks underlying attention may contribute to the phenotypic expression of psychiatric disorders (Cole et al., 2014). The distinct attentional impairments in melancholia (Austin et al., 2001) motivate research into underlying complex brain networks, which underlie disorders such as

Page | 80

schizophrenia (Andreasen, Paradiso, & O'Leary, 1998; Friston & Frith, 1995; Micheloyannis et al., 2006). In the present setting, we hypothesise that attentional disturbances in melancholia are underpinned by a failure to reorganise brain networks in response to emotionally salient exteroceptive signals. This position raises new approaches towards the study of attention- and insula-based attention systems in melancholia. However, to further our understanding of the neurobiology of impaired attentional set-shifting, it is necessary to model the system in response to a changing context. In the present study we designed and implemented a novel in-scanner naturalistic film viewing paradigm to clarify brain networks underlying attentional shifting between internal states and exteroceptive, dynamic emotional stimuli in melancholia.

While inattention is a common feature of melancholia (Parker & Hadzi Pavlovic, 1996), the search for its underlying neurobiological mechanisms is in its infancy (Elliott et al., 2002). This corresponds with challenges in identifying brain correlates of the disorder more broadly. For instance, it is yet to be determined whether increases (Greicius et al., 2007) or decreases (Veer et al., 2010) in connectivity strength between cortical regions best characterise major depressive disorder. Inconsistencies across such neurobiological studies likely relate to diagnostic heterogeneity (e.g., major depression may effectively ‘homogenise’ differing depressive sub-sets with differing causes), so hindering identification of reliable biological markers. Using the more refined depressive sub-type of melancholia, analyses in Chapter 3 revealed disconnectivity between brain modes involved in attention (i.e., frontoparietal), cognitive control (vmPFC) and interoception (anterior insula), thus suggesting that disrupted interactions between a constellation of regions underlie melancholia. Despite such advances, detailed understanding of neurobiological systems underlying attentional dynamics in specific depressive sub-types remains elusive. As neuroscience moves towards more refined understanding of cognition-emotion interactions (Pessoa, 2008), assisted through development of ecologically valid methodologies (Hasson et al., 2004), there exists a unique opportunity to further clarify the neurobiology of melancholia.

Page | 81

Imaging studies of attention have historically relied on task-based studies – the cognitive subtraction approach (Hampshire & Owen, 2006). While providing insight into neural mechanisms of basic cognitive processes, the ecological validity of laboratory-style behavioural tasks has been challenged (Chaytor & Schmitter- Edgecombe, 2003). Furthermore, such tasks can at times be difficult to implement in psychiatric populations where task completion may be compromised (Elliott et al., 1996). Advances in imaging protocol design and analytic techniques, namely the use of naturalistic stimuli (i.e., film viewing) in a scanner environment, overcomes these concerns and hence provides a superior platform for interrogating the role of attention, and its underlying neurobiological processes, in depression. Films have the capacity to manipulate emotion in a manner that closely reflects everyday emotional dynamics (Gross & Levenson, 1995), and have been shown to elicit reliable and consistent neuronal responses across a range of cortical regions (Hasson, Malach, & Heeger, 2010; Hasson et al., 2004; Hasson, Yang, Vallines, Heeger, & Rubin, 2008; Honey et al., 2012). Aside from providing a unique understanding of neural processes under naturalistic settings, the minimisation of task demands is appealing when studying disorders such as melancholia. The advancement of such novel methodologies runs in parallel to progress in our understanding of the brain as a complex network. Together they offer substantial potential to elucidate the neurobiological mechanisms of dynamic attentional and emotional processes in depressive disorders.

As discussed in Chapter 3, there is now a wealth of theoretical reasoning and experimental evidence that positions the brain as a complex network (Sporns et al., 2004). Consistency of functional brain systems has been demonstrated across individuals using resting state fMRI (rs-fMRI) (Damoiseaux et al., 2006). These systems correspond to a range of cognitive and emotional processes including attention and visceral regulation (Laird et al., 2011; Seeley et al., 2007; Smith et al., 2009) – both key factors underlying disturbances of mood (Drevets, 2001). Recent research suggests that frontoparietal attention (Cole et al., 2014) and insula (Avery et al., 2014) brain systems – particularly connectivity to and from these regions – underpin (major)

Page | 82

depression. Despite such findings, understanding of brain systems underlying key phenotypic aspects of refined depressive sub-types (e.g., attention in melancholia) is limited. In light of the “cognitive dysmetria” of schizophrenia (Andreasen et al., 1998; Friston, 1998) – where difficulties in information processing emerge from disrupted neuronal systems – we propose that attentional deficits in melancholia may be underpinned by disturbances of brain network connectivity. The present study leverages these advances in neuroscience to examine whether brain networks related to attention and interoception are disrupted in melancholia when reorienting from an at-rest state to the free viewing of emotional film stimuli.

Here, we sought to examine the neural properties of the redirection of attention from spontaneous internal thought to exogenous, emotional stimuli, in melancholic, non- melancholic and healthy control groups. To this end, we advanced a naturalistic (positive, negative and neutral) film viewing and rs-fMRI paradigm to quantify the change in brain states during the reorienting of attention from rest to exogenous attentional processing. We extend on the analytic approach presented in Chapter 3. Briefly, ICA – combined across rs-fMRI and the film viewing conditions – was used to identify brain modes corresponding to attention, executive control, salience and sensory processing; sDCM (Friston et al., 2003; Li et al., 2011) was then applied to estimate hidden neuronal states, and generate estimates of the strength of relationships between the modes. To these steps, we add use of Network-Based Statistics (NBS) (Zalesky, Fornito, & Bullmore, 2010) to identify sub-networks that differ between depressed and control groups, across resting state and film viewing conditions. We hypothesised that neuronal states during emotional film viewing would remain in an “at-rest” state in melancholia, reflecting continued focus on internal states “indifferent” to the flow of emotionally engaging material.

Page | 83

4.3 Methods

This study utilised the same cohort of participants as in Chapter 3, hence all clinical and demographic details are the same. For completeness, brief details are again overviewed. Full details are provided in Appendix 2. There is also substantial overlap between the current chapter and Chapter 3 in terms of analytic approach for the functional imaging data. Pre-processing and analysis steps specific to the current study are overviewed herein, with additional analytic details provided in Appendix 3.

4.3.1 Participants Thirty-two unipolar depressed patients, 16 of whom met additional criteria for melancholia (see 4.3.2 below), were consecutively assessed and recruited through the specialist depression clinic at the Black Dog Institute in Sydney, Australia. Sixteen matched healthy control participants were recruited through the community. Clinical participants met criteria for a current major depressive episode and disavowed current and/or lifetime (hypo)mania or psychosis, upon questioning on the MINI (Sheehan et al., 1998). The age range for all participants was 18 to 75 years. Exclusion criteria for all participants included: current and/or past drug or alcohol dependence, neurological disorder (e.g., neurodegenerative condition, tremor), history of invasive neurosurgery, traumatic brain injury, ECT in the preceding six months, and/or an estimated IQ of below 80 as estimated with the WTAR. Additional exclusion criteria for control participants were a current or past mood or psychotic disorder, as determined by the MINI.

4.3.2 Depression Sub-typing Approach In addition to meeting DSM criteria for major depression, clinical participants were assigned to a melancholic or non-melancholic class by clinic psychiatrists, weighting previously detailed criteria (Parker et al., 2010; Taylor & Fink, 2006). Full diagnostic

Page | 84

details of the current cohort of participants is provided in Appendix 2 (Table A2-1) as per Chapter 3.

4.3.3 Demographic and Clinical Assessment Depression severity was assessed with the QIDS-SR16 (Rush et al., 2003). Overall functioning and anxiety was measured with the GAF scale (APA, 2000) and STAI-State and STAI-Trait measures (Spielberger, 1983), respectively. Years of education and current medication status were also assessed. The CORE measure of psychomotor disturbance (Parker et al., 1994) was used to quantify psychomotor impairment across the domains of retardation, non-interactiveness and agitation. A 32-item Q-Sort was also administered to the depressed participants (Parker et al., 2009), highlighting individual prototypic symptom patterns ranging from least to most characteristic. Results from these demographic and clinical measures are provided in Chapter 3 (Table 3-1).

4.3.4 Imaging Protocol All participants underwent four separate, temporally consecutive fMRI scans. Participants first undertook a 6 ¼ rs-fMRI scan (per Chapter 3). This was followed by three separate fMRI scans where participants viewed three film clips of differing emotional valences (positive, negative and neutral), each lasting six minutes.

4.3.5 Naturalistic Stimuli – Film Clips For the positive condition, participants watched an excerpt from a stand-up comedy routine – “Bill Cosby – Himself”. For the negative condition, participants watched a scene from the movie “The Power of One”, depicting inhumane treatment of prisoners during the apartheid era. For the neutral condition, participants viewed dynamic footage of landscapes and flowing water. The films were viewed through an MRI-compatible monitor, with matching audio streams provided via an insert earphone system Page | 85

(Sensimetrics Model S14). Prior to the onset of each film clip, participants were given brief instructions displayed as text on the monitor: “Video will begin soon. Please relax and watch”. A separate cohort of 18 healthy participants was recruited through the community to provide continuous ratings of their emotion during the viewing of the positive and negative films. Overall, ratings were consistent with the valence of the film clips. These reliability ratings are provided in Appendix 3 (also see Figure A3-1).

4.3.6 Imaging Acquisition and Pre-processing Full image acquisition, pre-processing and analysis details are provided in Appendix 3. Briefly, functional images were acquired on a Philips 3 Tesla scanner equipped with a 12-channel head coil. rs-fMRI data were acquired first, followed by the three separate film viewing conditions (pseudo-randomly counterbalanced). Each echo planar imaging (EPI) volume was realigned, normalised (unwarped), and smoothed using SPM8 software (Friston et al., 1995). Spatial ICA was used across all subjects and sessions using MELODIC in FSL. The ICA decomposition generated 70 components, or “modes”, of neuronal, physiological and artefactual origin (Beckmann & Smith, 2004). We selected eight neuronal modes representing a variety of emotional, cognitive and perceptual systems, namely auditory (AUD); default mode network (DMN); executive control (EXC); left insula (L-INS); right insula (R-INS); left frontoparietal attention (LFP); medial visual pole (MVP); and right frontoparietal attention (RFP; Figure 4-1) (Damoiseaux et al., 2006). All components, with the exception of L-INS and R-INS, were matched with previously reported cognitive and sensory networks (Smith et al., 2009), using spatial cross-correlation. The insula modes were identified by first determining the coordinates of bilateral anterior insula cortices using PickAtlas, with these coordinates then used to identify ICA components that were the most anatomically specific (i.e., high localisation/local spatial extent). Time series from each of the eight components were then extracted and entered into DCM in order to infer subject-specific patterns of effective connectivity.

Page | 86

Figure 4-1: Analysis pipeline illustrating the use of ICA spatial maps to inform sDCMs. Directed edge weights derived from the sDCMs (of both rs-fMRI and film viewing conditions) were used in the NBS to test for condition by group effects.

4.3.7 Dynamic Causal Modelling As discussed in Chapter 3, DCM assesses effective connectivity between neuronal populations (Friston et al., 2003). As a computational technique, it combines dynamic models of neuronal states with detailed models of haemodynamics. Estimates of the influence of one brain region over another are obtained by making inferences about the neuronal states (and their interactions) by generative model inversion using (variational) Bayesian techniques. DCM returns matrices reflecting connectivity between different inputs (here, regions or modes), which can be expressed as either the strength of a connection or as a binary representation. Such modelling also allows recovery of hidden neuronal states (through model inversion).

Page | 87

4.3.8 Inter-Subject Correlations Consistent haemodynamic responses have previously been identified across individuals during film viewing (Hasson et al., 2004), contributing to the validity of this experimental approach. This is classically achieved through the study of inter-subject correlations (ISC). To our knowledge, stochastic DCM has not been previously used to infer hidden neuronal states during free viewing of dynamic films. To provide validity to this approach, we hence examined ISCs amongst the hidden neuronal states in each of the distributed modes following inversion of sDCM. Multiple comparisons were controlled for using Sidak’s false discovery rate (FDR) correction ( = 0.0055). Figure 4-2 illustrates this approach.

4.3.9 Network-Based Statistic Stochastic DCM returns subject-specific effective connectivity matrices. We then sought to identify any sub-networks in these matrices of node-to-node edge strengths that differed according to condition and group. This was achieved by employing NBS (Zalesky et al., 2010), a permutation-based method allowing for control over family- wise error given the mass univariate testing required when there are multiple edges in a network. This approach allows identification of edges that constitute a topologically connected network that differs between conditions and/or groups. Here, we examined for group x condition differences in network strength across rest and negative and positive film viewing conditions. NBS follows the traditional principles of cluster-based thresholding of statistical parametric maps, by first setting a preliminary height threshold (pair-wise connections) and then imposing a family-wise error (FWE)- adjusted cluster threshold (for networks of connections). We applied a pair-wise threshold of p <0.01 (uncorrected), followed by cluster-wise thresholding of p <0.05 (FWE corrected).

Page | 88

Figure 4-2: Analysis pipeline for calculating inter-subject correlations of hidden neuronal states. Illustrated schematically for DMN mode. From top: (1) BOLD time series of each subject; (2) DCM inversion of BOLD to give neuronal states for each subject; (3) Inter-subject correlations calculated on hidden neuronal states.

Page | 89

4.4 Results

4.4.1 Inter-Subject Correlations of Hidden Neuronal States Overall, ISCs were significant for the neuronal states of most modes across negative and positive film viewing conditions in melancholic and control groups (Table 4-1). Seventy five percent of the ISCs we tested were significant across the differing groups, film viewing conditions and modes. ISCs were particularly robust for the AUD mode across all groups, and particularly for the control (r = 0.1833) and non-melancholic (r = 0.1325) groups during negative film viewing. Interestingly, ISCs of both L-INS and R- INS were significant across all groups during negative film viewing. However, during positive film viewing none of the groups showed significant and consistent synchronisation of the L-INS. There was in fact no discernable correlations of neuronal states across melancholic subjects for this mode during positive film viewing. In addition, ISCs of neuronal states for the R-INS during positive film viewing in melancholia were not significant after adjustment for multiple comparisons. In contrast, ISCs of the R-INS during positive film viewing were significant in non-melancholic and control groups.

Page | 90

Table 4-1: Inter-subject correlations of negative and positive film viewing conditions. Negative Film viewing Positive Film viewing Melancholic Non-Melancholic Control Melancholic Non-Melancholic Control Mean (sig.) Mean (sig.) Mean (sig.) Mean (sig.) Mean (sig.) Mean (sig.) AUD 0.0610 (< 0.001) 0.1325 (< 0.001) 0.1833 (< 0.001) 0.0437 (< 0.001) 0.0755 (< 0.001) 0.0778 (< 0.001) DMN 0.0422 (< 0.001) 0.0156 (0.041) ‡ 0.0318 (< 0.001) 0.0347 (< 0.001) 0.0547 (< 0.001) 0.0438 (< 0.001) EXC 0.0417 (< 0.001) 0.0187 (0.013) ‡ 0.0400 (< 0.001) 0.0210 (0.004) 0.0434 (< 0.001) 0.0359 (< 0.001) L-INS 0.0545 (< 0.001) 0.0347 (< 0.001) 0.0569 (< 0.001) 0.0000 (0.467) ‡ 0.0082 (0.140) ‡ 0.0117 (0.069) ‡ R-INS 0.0456 (< 0.001) 0.0339 (< 0.001) 0.0351 (< 0.001) 0.0131 (0.046) ‡ 0.0234 (0.002) 0.0233 (< 0.001) LFP 0.0678 (< 0.001) 0.0236 (0.006) ‡ 0.0310 (< 0.001) 0.0225 (0.002) 0.0173 (0.026) ‡ 0.0093 (0.139) ‡ MVP 0.0235 (0.005) 0.0107 (0.098) ‡ 0.0379 (< 0.001) 0.0373 (< 0.001) 0.0437 (< 0.001) 0.0357 (< 0.001) RFP 0.0267 (0.007) ‡ 0.0091 (0.161) ‡ 0.0455 (< 0.001) 0.0395 (< 0.001) 0.0529 (< 0.001) 0.0733 (< 0.001) ‡ Bold denotes inter-subject correlation that was not significant after adjusting for multiple comparisons.

Page | 91

4.4.2 Connectivity amongst Brain Modes in Melancholia We next analysed the distribution of the directed edge weights between the eight brain modes, yielding 64 unique edges. This was achieved by taking the average edge strength across all subjects, then ranking these (for visualisation purposes) from most positive to most negative. When illustrated across the differing conditions (Figure 4-3), it is evident that those with melancholia had, on average, stronger connection weights (both positive and negative directions) under both film viewing conditions compared to healthy controls (Fig. 4-3; left column). The extent of this effect was less pronounced at rest. The non-melancholic group are positioned between the distributions of the control and melancholic groups.

We formally tested whether these edge weight distributions differed between groups. Null distributions for each group and condition were derived through random permutation sampling (1000 samples per condition). The group-wise standard deviation of the edge weights was compared to the corresponding null distribution. There was a statistically significant difference between the melancholic and control groups in the breadth of the distribution of connectivity weights across the network of edges during positive film viewing (difference = 0.0223; p < 0.001). An effect was also present for the negative film viewing condition (difference = 0.0198; p = 0.0170), with trend-level differences apparent for the resting state condition (difference = 0.0140, p = 0.038), between melancholic and control groups. No significant differences were found between melancholic and non-melancholic groups for the positive film viewing condition (difference = -0.0060, p = 0.790), negative film viewing condition (difference = -0.0091, p = 0.818), or for resting state (difference = -0.0029, p = 0.640).

Page | 92

Figure 4-3: Group comparisons of rank-ordered distributions of all 64 edge weights across positive and negative film viewing and resting state. Left column shows melancholic versus healthy controls: Right column shows melancholia versus non-melancholic MDD. Page | 93

4.4.3 Network-Based Modelling of Naturalistic Film viewing We then used NBS to investigate between group differences in directed edge weights among the 8 modes (see Figure 4-1 for analysis pipeline). To test our central hypothesis – namely a failure to redirect resources when shifting from rest to film viewing – we examined interaction effects of group by condition (rest vs. film viewing). A sub- network of edges, including LFP L_INS, MVP LFP and LFP RFP, showed a strong and significant effect between rest and negative film viewing conditions for the melancholic versus control group contrast (p < 0.032, FDR corrected, Figure 4-4). Specifically, connectivity strengths for this sub-network of edges were substantially higher in the melancholic group (Figure 4-4, bar graph) during the viewing of the negative film than at rest. In contrast, the control group showed the reverse, with stronger connectivity strength at rest than during the negative film viewing. Interestingly, the non-melancholic group showed a trend towards that of the control group.

Figure 4-4: Sub-network of edges distinguishing melancholic and control groups across resting state and negative film viewing conditions identified using the NBS.

Page | 94

4.4.4 The Impact of Medication on Sub-Network Scores We tested for potential medication effects using logistic regression. Patients were divided into two sub-sets: those in receipt of an SSRI medication, versus those not on any such medication; and those on any other non-SSRI medication (e.g., antipsychotics, mood stabilisers, all broad-action antidepressants) compared to those who were not taking non-SSRI medication. We controlled for clinical group, and examined whether the interaction between resting state and negative film viewing (averaged edge weights across the sub-network identified in 4.4.3) predicted the presence/absence of differing drug classes in two separate logistic regressions. No significant effects were observed in predicting receipt of differing drug classes (on/off SSRI, exp(β) = 0.00, p = 0.25; on/off non-SSRI drug, exp(β) = 0.00, p = 0.26). Formal statistics for these tests are provided in Appendix 3 (Table A3-1).

4.5 Discussion

Cognitive impairments contribute strongly to the burden of illness in melancholia, with difficulties in shifting attention particularly characteristic (Austin et al., 1999). We proposed that the inherent inability to disengage from dysphoric internal states in melancholia could be captured in brain networks known to support attention and interoception. Surprisingly, a specific sub-network of brain regions underpinning attention and interoceptive control showed an increase of effective connectivity strength in melancholia during negative film viewing. This effect was specific to the melancholic sub-type (i.e., not observed in non-melancholic patients), and was in conflict with our initial hypothesis that brain dynamics in those with melancholia during emotional film viewing would resemble an at-rest state, reflecting a failure of resource re-allocation. We further observed that total network effective connectivity strength in melancholia is significantly increased when ‘attending’ to emotionally salient films, suggesting a system-wide regulatory failure of causal influences between cognitive, emotional and sensory cortical systems in this disorder. We position these findings as reflecting several key features of melancholia, namely disrupted attention and non-interactiveness.

Page | 95

Our findings do not support our initial hypothesis. We cautiously advance an alternative interpretation of these findings. It has been demonstrated that effective connectivity decreases in cortical systems with learning (Buchel, Coull, & Friston, 1999), which may be of relevance to the current findings. We speculate that the sub- network of regions increasing in effective connectivity strength in melancholia during negative film viewing may be a result of inefficient neuronal adaptation. Specifically, the current results indicate a failure of cortical systems to adapt to the demands of dynamic exteroceptive emotional stimuli in melancholia. In health, the brain has been conceptualised as a self-organising system (Friston, Breakspear, et al., 2012) – that is, a system that acts to minimise perturbations, and hence maximise ‘homeostasis’. We frame the findings of network-wide increases in effective connectivity strengths in melancholia as reflecting a breakdown of this critical balance between system stability and instability during attention to dynamic, naturalistic emotional material. This aligns with contemporary formulations of brain dysfunction in psychiatric disorders, which are positioned as arising from instability in the brain’s ‘dark energy’ (intrinsic neuronal activity) (Zhang & Raichle, 2010). Together, these factors may contribute to findings of disrupted neuronal processes in melancholia. Consequently, one might expect that such neurobiological perturbations may underpin difficulties in shifting attention between interoceptive signals and exteroceptive emotional stimuli.

A growing body of work has demonstrated that consistent patterns of signal fluctuation arise across individuals when viewing the same (well directed) films (Hasson et al., 2004). Both lower and higher cortical regions, including occipital, temporal, parietal and frontal cortices, exhibit this pattern (Bartels, Zeki, & Logothetis, 2008; Jaaskelainen et al., 2008). All existing fMRI studies in this field have reported ISCs of haemodynamic signals in healthy individuals, whereas the current analyses identified ISCs in hidden neuronal states, as recovered through sDCM, thus providing important validation of the methodological approach we used. Largely consistent ISCs across differing brain modes were observed in the melancholic and control groups for both film viewing conditions. However, there was reduced inter-subject consistency of

Page | 96

the RFP during negative film viewing in melancholia, possibly indicating divergent attentional processing across individuals in this group. Similarly, a lack of inter-subject consistency was observed for the left insula mode in all groups during positive film viewing. This was particularly pronounced in the melancholic group, where there was no apparent ISC for the L-INS mode. In addition, there was reduced synchrony of the R-INS in melancholia during positive film viewing. These findings may highlight a disengagement from positive emotional content in melancholia, and potentially reflect differential redirection of attention to interoceptive signals across individuals. Despite these suggestions, inferring cognitive states from neuroimaging data is prone to the fallacy of “reverse inference” (Poldrack, 2006). In the current discussion, we have briefly entertained the notion of whether attentional and insula mode dysfunction relates to cognitive processes. However, we acknowledge the limitations of such logic. Even with the observed inter-subject consistency of neuronal states to naturalistic stimuli, limitations must be placed on the extent to which this relates to notions of “interoception”, “attentional redirection” or “ineffective learning”. Measuring other physiological variables, including heart rate, eye movements and skin conductance would provide a more refined index of the level of engagement and physiological responses whilst viewing dynamic, naturalistic stimuli.

Further research could address several limitations. While the sample size was sufficient for identifying significant group effects, larger samples would assist in identifying relationships between neurobiology and other key attributes of the illness (e.g., behavioural markers, clinical trajectory). Further, recruitment of a non-medicated cohort was not ethically justifiable given the severity of depression experienced by patients in the tertiary referral clinic. Despite our patient groups differing on class of prescribed medications, post hoc analyses suggested that our main effects were unlikely a result of the primary medication divisions (SSRI versus no SSRI, and presence or absence of any non-SSRI medication). Nonetheless, addressing the impact of medications a priori in future studies will be essential given the role SSRIs (Wagner et al., 2010) and antipsychotic medications (Miller et al., 1997) have on cerebral blood

Page | 97

flow. However, as discussed in Chapter 3, whilst desirable, capturing medication-free melancholic participants is a major logistic challenge, and the withholding medications for the purposes of research raises ethical concerns.

As psychiatry aligns itself to computational neuroscience (Montague et al., 2012), it is positioned to benefit from the much-need departure away from traditional views of brain function (i.e., representing cognition and emotion in increasingly reductionistic terms) to one that considers observable behaviour (e.g., mood, affect) as emerging from interacting brain systems. This current study advances the neurobiology of melancholia through such a lens.

Page | 98

Chapter 5: General discussion and future directions

In this thesis I sought to understand the cognitive and neurobiological mechanisms of attentional impairments in melancholic depression. In doing so, I additionally sought to investigate the validity of melancholia as a distinct diagnostic entity. The studies revealed that the melancholic phenotype does indeed appear distinct to non-melancholic depressive conditions, and from healthy individuals. The three main studies comprising this thesis position the cognitive impairments in melancholia as arising from disrupted inference processes across cognitive and neuronal systems. Specifically, I identified that melancholia was associated with disrupted psychophysical processing during attentional control of emotion, which suggests that perceptual inference mechanisms may be disrupted in this condition. Perturbations in attention and insula-based brain systems also lend support to the notion of disrupted inference processes in melancholia, which may contribute to difficulties in shifting attention between spontaneous internal thought and the external environment. In this discussion I first provide an overview of the main findings and, in light of these, revisit contemporary cognitive and neuronal models of attention and interoception, noting their particular relevance to melancholia. The utility of emerging neuroscientific models is addressed, before highlighting study limitations and directions for future research. Finally, I offer a brief preliminary analysis in alignment with the proposed future research directions.

In Chapter 2 I demonstrated that those with melancholia show disrupted psychophysical processes underlying sensitive detection of emotionally salient stimuli, and poorer discrimination during detection of neutral stimuli. These effects reveal that melancholic, but not non-melancholic, depressed individuals may be suboptimal in their attention to, and thus inference of emotional stimuli. The use of an attentional control task in Chapter 2 indicates that mechanisms of attentional inference may be specifically disrupted in melancholia. Page | 99

In Chapter 3, I used stochastic dynamic causal modelling (sDCM) to investigate effective connectivity between resting state brain modes known to underpin attention and interoception. In this study it was shown that melancholia is characterised by a pervasive ‘dysconnectivity’ between bilateral anterior insula, frontoparietal attention and executive control brain modes. This ‘sejunction’ of causal influences between brain regions was specific to melancholic depressed subjects, providing robust evidence that this disorder is underpinned by disrupted neuronal integration.

In Chapter 4 I sought to study causal influences between these key neuronal systems during the shifting of attention. To achieve this, I conducted an fMRI experiment, whereby participants viewed naturalistic, emotional films after a resting state acquisition. Using sDCM again, I observed significant increases in global effective connectivity strength during the free viewing of emotional stimuli in melancholia. This suggests a disturbance in homeostasis amongst neuronal systems in this disorder. Further, inter-subject correlations of neuronal states were observed across most modes in melancholia, with the exception of the right frontoparietal mode during negative film viewing and both insula modes during positive film viewing. The lack of any noticeable correlation amongst melancholic participants for the left insula during positive film viewing is consistent with a disjunction of subject-specific neuronal states underpinning interoceptive and visceral regulation. Such findings offer unique insights into disrupted neuronal systems known to support the shifting of attention and interoception in melancholia, and may help to explain the prominence of somatic complaints in this disorder.

Together, the three study chapters demonstrate that perceptual and inferential processes to emotional stimuli are disrupted in melancholia, and provide a neurobiologically realistic framework through which its prototypic features may be viewed.

Page | 100

5.1 Evidence of Disrupted Attentional Inference in Melancholia

Chapter 2 examined perceptual inference mechanisms underlying attentional control of emotion in melancholic and non-melancholic depression through psychophysical modelling. Optimal perceptual inference can be viewed as depending upon the precision of predictions of an uncertain (i.e., stochastic) sensorium. This Bayesian approach to perception and cognition suggests that higher-level cortical systems function to predict lower-level sensory inputs to create models of the causal structure of the world (Friston, 2005). Erroneous prediction of lower-level signals is thought, over time, to be self- correcting, such that higher-level models adapt to minimise discrepancies between inputs and inferences. Following Helmholtz, such probabilistic inference allows an agent to build accurate models of the myriad of uncertain sensory cues, selected from multiple, competing internally held probability distributions over the causes of one’s environment (Dayan et al., 1995).

In this framework, attentional responding has been posited to be a result of active sampling of probabilistic representations of the sensorium (Vul, Hanus, & Kanwisher, 2009), where the precision of the most likely distribution is dynamically updated to support optimal perception (Vossel, Geng, & Friston, 2014). In such treatments, one’s prior beliefs correspond to the expected value of a stimulus. Narrow priors therefore relate to more precise internal representations, and uncertainty increases as they become flatter. Bayesian theories of attention have been proposed for sensory integration, namely as a mechanism of attentional enhancement (Yu & Dayan, 2005), and for top- down integration of lower-level saliency signals (Borji, Sihite, & Itti, 2012). Computational models of learning under uncertainty have been proposed, also referred to as “Bayesian learning” (Mathys et al., 2011), and applied to aberrant learning states, such as learned helplessness, in depression (Huys & Dayan, 2009). Chapter 2 frames the attentional disturbances in melancholia as arising from distorted inference processes to emotional stimuli, as evidenced by suboptimal psychophysical processing on the

Page | 101

affective go/no-go task. However, it is important to note, and as formalised by Huys and Dayan, that such responding may not be suboptimal in the general sense. Rather, such responses may be ‘optimal’ responses to one’s own (internal and external) environment. For example, negative cognitive biases in a depressed individual may be optimal in the sense that they correspond to the depressed state (likely reflecting negatively biased priors). However, such maladaptive cognitive content, when combined with disrupted affect and behaviour, likely play a significant role in contributing to attentional dysfunction in disorders such as melancholia.

5.2 The Bayesian Brain, Attention and Inference: Towards a Cognitive Neuroscience of Melancholia

Chapters 3 and 4 highlighted a breakdown in causal influences between candidate brain regions known to be involved in attention and interoception in melancholia. A detailed overview of the neurobiology of attention was provided in Chapter 1, but attempts have been made more recently to understand the neurobiology of attention in computational terms (Feldman & Friston, 2010). Contemporary formulations of the neurobiology of perception, which could be extended to attention in future research, theorise that the brain operates under the free-energy principle, which in essence prescribes that all action and perception function to minimise surprise upon active sampling of the sensorium (Friston, 2009). This provides an account of how the brain creates a model of the world that is optimised through active sampling of data (e.g., sensory inputs). In essence, this model sees perception as a special case of hypothesis testing (Friston, Adams, Perrinet, & Breakspear, 2012; Gregory, 1980). Such hypothesis testing compares the predictions of sensory signals to the evidence of those signals – with the observed difference being the prediction error (Friston, 2009). Under the free-energy principle, the neurobiology of attention can be framed as minimising prediction error (“free-energy minimisation”) through gain modulation, which functions to match the weight accorded to sensory inputs to the precision of their prior predictions.

Page | 102

It is proposed that the dysconnectivity of attentional networks observed in Chapter 3 relates to notions of synaptic plasticity and dysconnection as previously applied to schizophrenia (Stephan, Baldeweg, & Friston, 2006). This work highlighted aberrant synaptic plasticity, resulting from abnormal modulation of neurotransmitter systems, as one potential mechanism of brain dysconnectivity in schizophrenia. In the present context, it is conceivable that frontoparietal attention systems in melancholia become disrupted through inefficient gain control at the synaptic level, thus leading to dysconnectivity from other brain systems. This was most evident in Chapter 3 where functional dysconnections between the insula, attentional and executive control modes were observed. From a “Bayesian brain” perspective, disrupted causal influences between the insula and right frontoparietal cortices may relate to inefficient bottom up prediction error passing. Specifically, such disruptions may impact upon the updating of emotional representations in frontoparietal and executive control cortical systems from lower-level representations (e.g., insula) of internal feeling states (e.g., somatisation). Research into decision-making provides an empirical basis for such reasoning. Cognitive control regions (e.g., ACC) involved in the updating of probabilistic judgements under uncertainty (Behrens et al., 2007) have been suggested to underpin altered value computation and decision-making in depression (Paulus & Yu, 2012). Given the role of frontoparietal attention systems in supporting cognitive control (Vincent et al., 2008), it is possible that they act to adjust the precision of predictions over lower-level signals emerging from brain regions such as the insula.

Attention and insula-based cortical systems were again implicated in Chapter 4. These regions were part of a sub-network of brain regions underlying the (attempted) redirection of attention to negative film content. I was surprised by this finding as the direction of difference (stronger in melancholia) was opposite to that I had anticipated. It is conceivable that such findings relate to ineffective allocation – rather than a failure of redirection – of attentional resources in melancholia. While not explicitly studied, those with melancholia may remain focussed on – or at least have difficulty shifting attention away from – internal dysphoric states (somatisation): that is a common

Page | 103

symptom reported in the clinical setting (Ebert & Martus, 1994). I argue that the putative interoceptive focus in melancholia aligns with the observed involvement of the insula modes. Few studies have investigated the role of the insula in depression. Those that have (Avery et al., 2014) reported that the insula was decreased in activity during attention to interoception and, at rest, the insula showed increased functional connectivity to limbic regions including the amygdala, sgPFC and OFC. The involvement of the insula was interpreted as aligning with the somatic symptoms of MDD. I suggested in the discussion of Chapter 4 that increases in effective connectivity during film viewing may have been driven by disrupted neuronal adaptation. I extend this here with a complementary explanation. Given melancholia is associated with altered attention and decision-making (Chapter 2) it is possible that altered probabilistic representations – as previously theorised as reflecting depression (Paulus & Yu, 2012) – also occur during negative film viewing. This may also be evident at the neuronal level, leading to disruptions in hierarchical neuronal systems that prevent successful updating of internally held prior beliefs. Notably, I also observed a broader distribution of effective connectivity (stronger positive and negative weights) in melancholia, which we interpreted as a disturbance in homeostasis (the process of dynamically updating a biological system to keep it in a stable flux of energy). This change in auto-regulation is precisely compatible with alterations in error coding and model updating (Friston, Breakspear, et al., 2012).

I offer several explanations in relation to the reduced inter-subject correlations of left and right insula neuronal states (during positive film viewing) between melancholic participants in Chapter 4. First, it is possible that patients may disengage from the film stimuli, and instead remain focussed on their spontaneous, internally generated thought. Second, patients may attempt engagement with the positive stimuli but fail to obtain hedonic reward in a way that continues to capture their attention. The latter explanation is consistent with the primacy of anhedonia in melancholia, and corresponds to evidence that those with this disorder exhibit impairments in modulating behaviour as a function of prior reward (Pizzagalli et al., 2005) – which itself has been positioned along the

Page | 104

lines of Bayesian inference (Schultz, Dayan, & Montague, 1997). In Bayesian models of reward, neurotransmitters such as dopamine act to signal reward-based prediction errors. Predictions in a noisy, unpredictable environment are modulated by their expected salience or likely reward. Dopamine also modulates task switching in the vmPFC to salient cues (Shiner et al., 2014), which aligns with findings in Chapter 3 of EXC mode (which included portions of the vmPFC) dysconnectivity with the insula. Further, the insula has been shown to integrate information regarding the perceived significance of a stimulus with actual perceptual decision-making (Wiech et al., 2010). By analogy, disrupted integration of the expected reward value of a situation with the actual lived experience of that situation may be driven by perturbed insula function in melancholia. Rather than the above two propositions being mutually exclusive, it is possible that they both operate in melancholia under related principles of disrupted neuronal (attention- and insula-based) inference mechanisms. Emerging theories of interoceptive inference also fit with the findings across study chapters, and are indeed of significance given the involvement of the anterior insula in melancholia.

5.3 Distorted Interoceptive Inference in Melancholia

Interoceptive predictive coding proposes that emotional responses are dependent on the continuous updating of predictions of sensory signals emerging from the insula (Seth, 2013). This approach – in line with existing Bayesian brain models – holds that an individual makes sense of bodily feeling states and sensations through applying generative models of what is “most likely to be me” (Apps & Tsakiris, 2014). In a similar vein to prediction error minimisation, individuals likely apply rules of interoceptive inference to maintain autonomic homeostasis (Gu et al., 2013). Those with melancholia experience unremitting internally generated dysphoria, and frequently report somatic complaints (Taylor & Fink, 2006), that may reflect failures of insula- mediated interoceptive inference. The identification of the AI in Chapters 3 and 4 demonstrate that this structure is linked to the neurobiology of melancholia; However, relationships between the insula and disrupted interoceptive inference in melancholia are largely conjectural given that I did not directly probe interoception in my study Page | 105

design. Despite this, it is possible that attention is maladaptively redirected to abnormal interoceptive states in melancholia, which are enslaved by the AI. Indeed, attention plays a key role in modulating message passing in a hierarchically distributed system containing the insula, as evidenced in studies of motor control (Adams, Shipp, & Friston, 2013). In the context of interoceptive inference, low rates of error signalling – related to more precise predictions – may decrease attentional demands on the system (Seth, 2013). High rates of interoceptive error signalling, as is possible in melancholia, may thus place increased demands on attentional systems. Such a model aligns with the results observed in Chapter 4, where increased effective connectivity between the LFP and left insula mode was observed.

5.4 Validity of the Proposed Model

Whilst I have proposed that that several aspects of Bayesian computation offer a unifying framework through which to model neurobiological and cognitive impairments in melancholia, several caveats are acknowledged. Evidence for the existence of a hierarchically organised neural system that accords fully to Bayesian principles remains in its infancy (Clark, 2013). Despite being a relatively new approach towards understanding the brain, it has been heralded as a necessary departure from pre-existing models, and has the capacity to unify psychological and biological theories of perception and emotion (Clark, 2013). Additionally, it is unlikely that melancholia can be reduced to just insula and/or attentional brain region dysfunction. To further develop a neurobiological model of a complex disorder such as melancholia it will be necessary to investigate these regions in conjunction with other cortical systems. For instance, it has recently been suggested that motor processes – and by extension the brain regions underlying motor output – are partly modulated by insula signalling (Seth, 2013). Such suggestions render the insula as a central mechanism for regulating willed engagement with the environment. These emerging perspectives should be integrated into validating the current cognitive and neurobiological findings, especially given the primacy of psychomotor dysfunction in melancholia (Parker, 2007). Further, sustained amygdala activity has been observed in depressed individuals after presentation of emotionally Page | 106

salient material (Siegle, Steinhauer, Thase, Stenger, & Carter, 2002). Such work aligns with the notion I have raised regarding perturbed neuronal homeostasis in melancholia. Hence, brain regions such as the amygdala should also ideally be a focus for future research studies into melancholia. Such theoretical explanations are, at this time, purely speculative, but may offer additional insights at the proposed nexus of disrupted inference, attentional dysfunction and interoception in melancholia.

5.5 Limitations

The studies in this thesis have several limitations. The relatively modest sized, medicated sample of depressed patients recruited for the studies may impact on the generalisability of the findings. The recruitment of a larger sample, especially in a longitudinal setting, would enable further hypotheses to be tested. This could include determining whether attentional deficits and/or disrupted neurobiological processes are present prior to the onset of early episodes, and whether any given deficits continue into remission or contribute towards risk of relapse. Given the wide variety of medications used to treat depression (Beck & Alford, 2009), it remains a challenge to recruit patients on similar medication regimes – or preferably on no medication – which may contribute to difficulties in interpreting both cognitive (Harmer et al., 2009) and neuroimaging findings (Phillips, Travis, Fagiolini, & Kupfer, 2008) in affective disorders. In this thesis both melancholic and non-melancholic groups were on a range of different medications. Whilst I demonstrated that differing medication classes (on/off SSRI, and on/off non-SSRI medication) did not influence the significance of the main neuroimaging findings, recruitment of non-medicated patients would provide for more controlled estimates of neurobiological changes inherent to the disorder.

This discussion has considered distorted inference processes as an explanatory framework for the findings, but the design of the studies limits the extent to which these theories can be formally applied. For instance, measurement of inefficient updating in

Page | 107

the face of uncertainty, which likely underpins the findings presented in Chapter 2 (given the use of SDT), was unable to be directly assessed. Such a limitation may be overcome through the use of tasks that explicitly interrogate trial-to-trial updating (Mathys et al., 2011). This would thus enable further understanding of probabilistic processes in depression, and help determine with greater certainty whether melancholia is indeed associated with distorted attentional inference to emotional stimuli. Integrating such a paradigm into a functional imaging study would, moreover, allow dynamically varying computational variables (such as volatility) to be directly used as explanatory variables in the imaging analysis (Behrens et al., 2007).

In Chapter’s 3 and 4, I advanced the notion that disrupted neurobiological processes in brain regions known to support attention and interoception may contribute to distorted cognitive processes in melancholia. Specifically, I raised the possibility that brain network perturbations may underpin aspects of the disorder’s phenomenology (i.e., attentional focus on dysphoric internal states, and difficulties switching attention between internal states and the external environment). Inferring cognitive states from neuroimaging data is, however, not deductively valid (Poldrack, 2006). Such ‘reverse inference’ occurs in instances where an investigator attempts to infer cognitive processes from activation in a given brain region. This differs from traditional approaches in cognitive neuroscience where brain activation concurrent to a specific task is seen as being directly relevant to the neuropsychological function that the task assesses. Without direct manipulation of cognitive processes in Chapter’s 3 and 4, caution is exercised in relating the neurobiology of melancholia to specific aspects of “attention” and “interoception”. However, these concerns are partially offset given the cortical systems investigated in the current studies. All selected brain components, with the exception of the insula modes, were consistent with previously published cognitive and perceptual spatial brain maps (Smith et al., 2009). Such maps robustly correspond to cognitive activation studies as shown in analyses using the BrainMap database (Laird et al., 2011). The insula modes were also judged as closely resembling portions of the brain’s ‘salience’ network (Seeley et al., 2007). I also note that the use of sDCM applied

Page | 108

to time series obtained from ICA components in resting state data is a novel contribution of this thesis. Whilst it has precedents in prior task-driven analyses, future work is required to further validate this new application. Following in the tradition of DCM development, this could be achieved using simulated data, where the ground truth is known by construction {Friston, 2003 #1690;Stephan, 2008 #1873}.

Additional measures could also be used in future studies to help quantify physiological parameters that may influence emotional processing under ‘passive’ experimental conditions, including resting state and film viewing. In addition to directly questioning participants about the content of their thought following resting state scanning sessions (Delamillieure et al., 2010), physiological sensing equipment can be used to monitor autonomic signals (Critchley, 2009), and would hence allow a more integrative account of brain-body relationships to be advanced. Tracking of eye movements would also refine understanding of the level of engagement with dynamic stimuli both within and outside the scanner environment (Spiers & Maguire, 2007). Furthermore, recent advances highlight the utility of measuring cognitive performance concurrently with in-scanner naturalistic film viewing (Naci, Cusack, Anello, & Owen, 2014), which hence provide a more direct measure of dynamic cognitive processes under naturalistic settings. All such approaches would be invaluable to further clarifying mechanisms of attentional dysfunction in melancholia.

5.6 Integrating Neuropsychological and Neuroimaging Data

While I acknowledge that the ‘ground truth’ of disrupted attention in melancholia will likely be elucidated through future application of probabilistic behavioural paradigms (Mathys et al., 2011), insights may also be gained regarding brain-behaviour relationships through integration of neuropsychological and neuroimaging data. The synthesis of such data sources has been suggested to provide more relevant clinical information than either approach alone in evaluating neurobehavioural disorders (Bigler, 2001). Here, I present a brief analysis aimed at integrating the neuropsychological and neuroimaging approaches presented thus far in this thesis.

Page | 109

5.6.1 Brain Network Correlates of Set-Shifting Performance As overviewed in the first study (Chapter 2), the current cohort of participants underwent a comprehensive neuropsychological assessment, including a range of tests probing attentional and executive functioning. In addition to the affective go/no-go task (AGN) analysed in detail in Chapter 2, participants completed tasks of sustained attention (RVP), spatial planning/executive functioning (SOC) and attentional set- shifting on the intra-extra dimensional set shift (IED) task. Clinical participants completed these tasks on the same day as their MRI scan, and control participants within seven days of the scanning session. As highlighted throughout this thesis, those with melancholia exhibit impairments in shifting attention (Austin et al., 2001). Melancholia is associated with poorer set-shifting performance on the IED compared to non-melancholic MDD (Michopoulos et al., 2008; Michopoulos et al., 2006). The IED task begins with trials requiring simple stimulus discrimination (i.e., two shape elements of the same colour) allowing an individual to establish an attentional ‘set’ (i.e., consistency of responding). Additional stimulus features (e.g., lines of a different colour) are gradually introduced as distractor stimuli. An attentional set must be maintained to the original stimulus elements in the presence of such distractors and, once established, an attentional shift is introduced. Individuals are then required to shift attention to a stimulus of the same type (e.g., perform an “intra-dimensional” shift to a different shape), and finally shift attention to a previously irrelevant stimulus (e.g., perform an “extra-dimensional” shift to lines in the presence of shapes) (Owen, Roberts, Polkey, Sahakian, & Robbins, 1991). The task consists of nine stages, with the extra- dimensional shift occurring at the eighth stage. Deficits at the extra-dimensional shift stage of the IED are characteristic of melancholia (Michopoulos et al., 2006). I propose that such deficits in attentional shifting in melancholia may relate to the neuroimaging findings, in particular the inability of neuronal systems to adapt to task demands (e.g., shifting attention from rest to attending to negative naturalistic film content).

Page | 110

I conducted a series of linear regressions to examine whether the sub-network of effective connectivity edge weights, identified in Chapter 4 (distinguishing melancholic and control groups between rest and negative film viewing conditions), predicted neuropsychological performance on the IED within each group. Averaged sub-network scores were created for each participant separately for resting state and negative film viewing conditions. I selected the following outcome measures from the IED: number of stages completed; total errors; and total errors at the extra-dimensional shift stage. A trend emerged, with the melancholic group evidencing a relationship between the sub- network score for negative film viewing and total errors at the extra-dimensional shift stage of the IED (β = 28.63, p = 0.069). The presence of a moderately strong positive correlation between these measures (r = 0.50, p < 0.05) suggests that as negative film- viewing sub-network scores increase, so too do errors of attentional set-shifting. There was no predictive utility of either resting state or negative film viewing sub-network scores on the remaining IED outcome measures across the remaining group-specific regressions.

These analyses suggest a trend towards correspondence of attentional set-shifting impairments in melancholia and perturbed neurobiological function, in particular across brain regions known to support attention and interoception. The lack of such relationships across groups for the remaining outcome measures, and low predictive utility of resting state network scores, suggests specificity to melancholia in linking set- shifting impairments with brain network dysfunction. However, in light of the marginal significance of the effects, future research in this area would likely benefit from larger sample sizes. This would allow multivariate models to be advanced and assist in clarifying causal relationships between key clinical, cognitive and neurobiological factors.

Page | 111

5.7 Conclusions

Psychiatry has long sought to position melancholia as a distinct depressive illness of biological origin (Parker & Hadzi Pavlovic, 1996; Taylor & Fink, 2006). This thesis set out to address whether methodological advances in cognitive neuroscience could contribute to furthering understanding of disrupted attentional and neurobiological processes in melancholia, and hence align this disorder with emerging diagnostic perspectives that seek to unify objective markers of illness with clinical phenotypes (Insel, 2014). The findings in this thesis lend support to the notion of melancholia as a distinct biological depressive condition. Melancholic, but not non-melancholic, depression was associated with disruptions to psychophysical and neurobiological processes underlying attention and interoception. In light of emerging neuroscientific models the findings broadly implicate melancholia as a disorder of distorted perceptual and attentional inference, reflected predominantly in an inability to shift attention away from dysphoric internal states. I suggest that the current thesis findings offer an advance over existing cognitive and neurobiological models of attentional dysfunction in depressive disorders. If judged as reliable, the current findings may offer substantial benefit in informing novel treatment and prevention approaches for melancholia.

Page | 112

Appendix 1

Page | 113

Figure A1-1: Violin plots (overlaid with box-plots) of posterior distributions of the mean and standard deviation of bias and discriminability across signal conditions and groups.

Page | 114

Hierarchical Signal Detection Theory model{ # Relating observed counts to underlying Hit and False Alarm rates for (i in 1:n) { HR[i] ~ dbin(h[i],S[i]) FA[i] ~ dbin(f[i],N[i]) S[i] <- HR[i]+MI[i] N[i] <- FA[i]+CR[i] } # Reparameterization Using SDT for (i in 1:n) { h[i] <- phi(d[i]/2-c[i]) f[i] <- phi(-d[i]/2-c[i]) } # Group Distributions for (i in 1:n) { c[i] ~ dnorm(muc,lambdac) d[i] ~ dnorm(mud,lambdad) } # Priors muc ~ dnorm(0,.001) mud ~ dnorm(0,.001) lambdac ~ dgamma(.001,.001) lambdad ~ dgamma(.001,.001) sigmac <- 1/sqrt(lambdac) sigmad <- 1/sqrt(lambdad) }

Page | 115

Sensitivity and Robustness Analyses of Model Posteriors

Two separate sensitivity analyses were conducted to determine whether the significant group differences were upheld under differing prior assumptions. The first analysis was a respecification of the hierarchical models using different priors, namely narrower (more precise) priors for muc and mud (mean of zero and precision of 0.01) and also for lambdac and lambdad (set at 0.01,0.01). Whilst there was no clear prior knowledge to inform the choice of these parameter values – hence the original use of uniformative priors – these are important checks of robustness. For the second sensitivity analysis we uncollapsed the noise conditions for those results that showed a between group effect.

Impact of Modifying Precision of Prior Distributions on Posterior Differences

The main effects of positive mean bias between melancholic and non-melancholic groups (HPDd = 0.024 – 0.559) were robust to the precision of the priors, as was the neutral mean discriminability between melancholic and control groups (HPDd = -1.034 – -0.041). As the precision of the priors increased, the group differences for the standard deviation estimates crossed below statistical significance. This indicates that, for the hierarchical model of positive and neutral signal conditions, differences on mean bias and discriminability are more robust to changes in the precision of the prior distributions than estimates of their variability.

The Effect of Differing Noise Conditions

When the uncollapsed noise conditions for the positive mean bias condition were examined across melancholic and non-melancholic groups, the positive/negative (signal/noise) condition remained significant (HPDd = 0.043 – 0.624), but there was no such significant effect for positive/neutral. For mean discriminability of neutral signal trials, the difference for the neutral/positive condition (HPDd = -1.109 – -0.098) remained significant, but not the group effect for the neutral/negative condition. Model

Page | 116

differences for standard deviation of bias estimates between melancholic and non- melancholic groups were upheld across uncollapsed conditions, for both positive/negative (HPDd = 0.022 – 0.624) and positive/neutral (HPDd = 0.064 – 0.640). However, for negative signal trials the melancholic vs. non-melancholic group difference for the standard deviation diminished when noise trials were not collapsed. When comparing melancholic and control groups the positive/negative contrast retained significance (HPDd = 0.028 – 0.610), but the positive/neutral condition did not show group differences. These effects highlight that bias to positive, and discriminability to neutral, signal trials may be influenced by differing noise conditions.

Page | 117

Appendix 2

Detailed Diagnostic Approach to Depression Sub-typing

Clinical diagnoses of melancholic or non-melancholic depression were made by psychiatrists weighting previously detailed criteria (Parker et al., 2010; Taylor & Fink, 2006). For a diagnosis of melancholia, two compulsory criteria were required (see Table A2-1): (A) psychomotor disturbance (expressed as motor slowing and/or agitation); and (B) an anhedonic mood state. In addition, five of the following nine clinical features were required (and met) in all assigned melancholic patients: (1) concentration and/or decision-making impairment; (2) non-reactive affect; (3) distinct anergia; (4) diurnal mood variation – being worse in the morning; (5) appetite and/or weight loss; (6) early morning wakening; (7) no preceding stressors accounting for the depth of the depressive episode; (8) previous good response to adequate antidepressant therapy; and (9) normal personality functioning. Whilst respecting the DSM diagnostic approach to melancholia, these have been customised to take into account criteria that has historically characterised melancholia (Parker et al., 2010; Taylor & Fink, 2006). Table A2-1 shows specific criteria for each patient, as cross-checked against each patient’s clinical assessment material (i.e., clinical notes, assessment letters and referral material to the Black Dog Institute Depression Clinic).

Page | 118

Table A2-1: Symptoms and signs expressed by melancholic (Mel1-Mel16) and non-melancholic (NMel1-NMel16) participants.

Essential Symptoms (Both Required) Specifiers (Five of Nine Required) Mood Diurnal Appetite Early No Response Normal Psychomotor Distinct Concentration Non- Mood /Weight Morning Preceding Physical Personality Patient Disturbance Anhedonia Impairment reactivity Anergia Variation Loss Wakening Stressors Treatment Function Mel1 + + + + + - - - + - + Mel2 + + + + + + - + - + + Mel3 + + + + + + - + - + - Mel4 + + + + + + - - + + + Mel5 + + + + + + - - + + + Mel6 + + + + + + - - + - - Mel7 + + + + + - + + + + + Mel8 + + + + + - + + + + - Mel9 + + - - + + - - + + + Mel10 + + + + + - + + + + + Mel11 + + + + + + - + - - + Mel12 + + + - + + - - + + + Mel13 + + + - + + - + + + + Mel14 + + + + + + - - + + + Mel15 + + + - + - + - + - + Mel16 + + + + + + + + + - - NMel1 ------NMel2 ------NMel3 ------NMel4 ------NMel5 - + ------NMel6 ------NMel7 - + ------NMel8 ------NMel9 ------NMel10 + + + - + - + - - - - NMel11 ------NMel12 + ------NMel13 ------NMel14 - + ------NMel15 + + + - + + - - - - - NMel16 ------‘+’ Indicates presence of symptom/sign; ‘-’ Indicates absence of symptom/sign. Page | 119

Neuroimaging Protocol, Pre-processing and Analysis Steps fMRI Image Acquisition

All participants underwent a 6 ¼-minute resting state fMRI scan (186 volumes) and were instructed to keep their eyes shut for the duration of the scan. Resting state fMRI was acquired at the beginning of a lengthy scanning session (i.e., prior to additional fMRI and structural scanning). All participants explicitly reported remaining awake for the duration of the scan. Scanning was conducted using a Philips 3.0-T scanner (Philips Medical Systems; Best, Netherlands). Functional data were acquired using T2*- weighted gradient echo-planar sequences (33 axial slices; repetition time/echo time: 2000/30 msec; 76° flip angle; reconstruction matrix size: 128 × 128; field of view (anterior-posterior): 240 mm; voxel size: 3.0 × 3.0 × 3.0 mm; no gap).

Data Pre-processing

Resting state fMRI images were pre-processed using statistical parametric mapping (SPM8) software (http://www.fil.ion.ucl.ac.uk/spm/) (Friston et al., 1995). For each subject, each image was realigned to the first acquired image, normalised (unwarped) to standard Montreal Neurological Institute (MNI) space and smoothed with a full-width half-maximum (FWHM) kernel of 4 mm. Pre-processed functional data were then used as inputs for probabilistic concatenated independent component analysis (ICA) using the MELODIC (Multivariate Exploratory Linear Decomposition into Independent Components) toolbox in the FMRIB Software Library (FSL) (http://www.fmrib.ox.ac.uk/fsl/) (Beckmann & Smith, 2004). For the ICA, non-brain voxels were masked with voxel-wise demeaning of the data and normalisation of the voxel-wise variance. Pre-processed data were next whitened and projected into a 70- dimensional subspace using Principle Components Analysis providing for a reasonably fine-grained decomposition of functionally relevant brain regions (Smith et al., 2009). These whitened observations were decomposed into sets of vectors that describe signal variation across the temporal domain (giving time courses), the session/subject domain

Page | 120

and across the spatial domain (giving spatial maps). This was implemented through a non-Gaussian spatial source distribution using a fixed-point iteration technique (Hyvarinen, 1999). Estimated component maps were divided by the standard deviation of the residual noise, with a threshold of 0.5 set (the probability that needed to be exceeded by a voxel to be considered ‘active’ in the component of interest) by fitting a mixture model to the histogram of intensity values (Beckmann & Smith, 2004).

Node Selection

Following ICA, nodes were identified and subsequently specified from the group-level spatial maps of five components with the aim of explaining the functional anatomy of attention and interoception in depressive disorders. The components selected were: (A) Default mode network (DMN); (B) Executive control (EXC); (C) Bilateral anterior insula (INS); (D) Left frontoparietal attention (LFP); and (E) Right frontoparietal attention (RFP; see Figure 3-1 in Chapter 3). All components were checked for accuracy by cross-correlating with previously identified cognitive networks (Smith et al., 2009) except for the INS. The INS component was identified from the ICA maps by, i) first obtaining the centre coordinates of the anterior insula using PickAtlas, and then, ii) using these coordinates to identify the most illustrative spatial map from the 70 components of the ICA. These maps were then used in specifying and estimating dynamic causal models (DCMs) across all study subjects.

DCM Specification

DCM infers effective connectivity amongst neuronal populations by combining dynamic models of neuronal states and detailed biophysical models of haemodynamics. Traditionally DCM has been employed to provide generative models of task-related data, where stimulus or task manipulations are introduced as known inputs to regions identified through use of the general linear model (Friston et al., 1995). With stochastic DCM (sDCM), the system perturbations are modelled as unknown system fluctuations

Page | 121

arising endogenously (Friston et al., 2011; Li et al., 2011). Broadly speaking, sDCM otherwise proceeds in a similar vein to classic DCM, namely: 1. The user specifies a model (or models) through the choice of nodes and inputs; 2. The empirical data are introduced as time series of each node; 3. The model evidence (the probability of observing the data given the model) is maximised using a variational scheme to minimise an objective function (the free energy). This also yields posterior model parameter estimates as well as estimates of the unknown state fluctuations (Daunizeau et al., 2011; Li et al., 2011); and 4. If more than one model is specified, model comparison is performed using the evidence for each model. This Bayesian model evidence (aka marginal likelihood) penalises the accuracy of each model by a measure of its complexity (Marreiros, Stephan, & Friston, 2010). In the present setting, we implemented a single, fully connected bilinear DCM with unknown fluctuations at every node. No external inputs were specified, corresponding to the no-task resting state acquisition. This model was estimated in all participants.

For each of the components, the peak activation voxel was identified, with MNI coordinates of these peaks used to define a regionally specific voxel of interest (VOI) to allow initial estimation of the DCMs. For visualisation purposes (e.g., see Figure 3-1, Chapter 3), a 6 mm sphere was used to represent the spatial location of the peak weight of the corresponding ICA mode. Dual regression was used to extract subject-specific time series from the group-level spatial maps (representing an average of the voxels within each map, weighted by their relative expression in that map), with the corresponding component time series used as inputs for subject-specific sDCMs (Daunizeau et al., 2011; Li et al., 2011). In specifying the DCMs, no inputs were selected for the first and second levels, and a fully connected model was chosen for the search space.

Page | 122

Model Optimisation

The functional architecture of the distributed systems from the DCMs was optimised for each subject using a network discovery algorithm (Friston et al., 2011). This algorithm yields a sparse network representation of the original fully connected graphs that are optimal in the sense of having the greatest conditional probability relative to other possible sparse networks. These networks are characterised in terms of their connectivity or adjacency matrices and conditional distributions over the directed (and reciprocal) effective connections between nodes or regions.

The Impact of Medications on Network Parameters

Analysis of covariance (ANCOVA) was used to examine for possible medication effects. We split all patients into two subsets: Firstly, we compared all patients on – or not on – an SSRI. Secondly we compared patients taking medication(s) other than SSRIs (all broad spectrum antidepressants, antipsychotics, mood stabilisers) with patients not on any such medications. We controlled for clinical group (melancholic versus non-melancholic) and analysed the main network effects (in-degree for RFP, and out-degree for both INS and LFP modes). No significant differences were observed for medication on the differing network parameters (see Table A2-2).

Page | 123

Table A2-2: ANCOVAs contrasting primary treatment partitions, controlling for diagnostic group, on ‘node degree’ parameters.

On Any Other Drug Than SSRI vs. Those On SSRI vs Not on SSRI Not on Other Drug Type III Type III Node Degree Direction and Region SS df F Sig. SS df F Sig. Degree In RFP 1.09 1 1.06 0.31 0.29 1 0.28 0.60 Degree Out INS 0.61 1 0.37 0.55 0.32 1 0.19 0.66 LFP 1.09 1 1.11 0.30 0.72 1 0.73 0.40

Page | 124

Appendix 3

Neuroimaging Protocol, Pre-processing and Analysis Steps fMRI Image Acquisition

All participants underwent a 6 ¼-minute resting state fMRI scan (186 volumes) and were instructed to keep their eyes shut for the duration of the scan. Resting state fMRI was acquired prior to three separate fMRI sequences during which participants viewed positive, negative and neutral films (pseudo-randomly counterbalanced across subjects). All participants explicitly reported remaining awake across the scanning sequences. Scanning was conducted using a Philips 3.0-T scanner (Philips Medical Systems; Best, Netherlands). Functional data were acquired using T2*-weighted gradient echo-planar sequences (33 axial slices; repetition time/echo time: 2000/30 msec; 76° flip angle; reconstruction matrix size: 128 × 128; field of view (anterior-posterior): 240 mm; voxel size: 3.0 × 3.0 × 3.0 mm; no gap).

Data Pre-processing

Resting state and film viewing fMRI images were pre-processed using SPM8 (Friston et al., 1995) as per Chapter 3 (see Appendix 2). All pre-processed functional data (across subjects and conditions) were then used as inputs for probabilistic concatenated ICA in MELODIC (Beckmann & Smith, 2004) using precisely the same methods as in Chapter 3 (again see Appendix 2 for technical details).

Node Selection

We selected eight ‘modes’ from the group-level spatial maps as generated by ICA, with the aim of capturing emotional, cognitive and perceptual systems relating to attention

Page | 125

and interoception. The components selected were: (A) Auditory (AUD); (B) Default mode network (DMN); (C) Executive control (EXC); (D) Left insula (L-INS); (E) Right insula (R-INS); (F) Left frontoparietal attention (LFP); (G) Medial visual pole (MVP); and (H) Right frontoparietal attention (RFP; see Figure 4-1). All components were checked for correspondence with previously identified cognitive and sensory networks using spatial cross-correlation (Smith et al., 2009), except for the L-INS and R-INS modes. These components were identified from the ICA maps by, i) first obtaining the centre coordinates of the bilateral anterior insula using PickAtlas, and then, ii) using these coordinates to identify the spatial maps with the most specificity from the 70 components of the ICA. These maps were then used in specifying and estimating stochastic dynamic causal models (sDCMs) across all conditions for all study subjects. The DCM specification and estimation in Chapter 4 used the same approach as outlined in Appendix 2 (for Chapter 3), with the only difference being that they modelled interactions between eight VOIs (i.e., the modes just detailed) for resting state and positive and negative film viewing conditions.

Emotion Ratings During Film viewing

An independent cohort of 18 healthy participants was recruited to provide emotion ratings of the Bill Cosby and The Power of One films. While viewing the film, participants provided continuous ratings of their emotion using rating software custom- built in LabView. They were instructed to continuously report their emotion while moving a mouse while they viewed the film. They were instructed to move the mouse cursor all the way to the left if they felt completely sad, depressed, disgusted or unpleasant; and move the mouse all the way to the right if they felt completely happy, joyful and pleased. A vertical bar, indicating their current rating (between -1 and 1), provided visual feedback. Negative ratings corresponded to values towards -1, whilst positive ratings corresponded to values towards +1. In addition, participants provided an overall rating of the film immediately after the viewing. The order of film presentation was counterbalanced between participants. Overall, emotion ratings of the films were consistent with their purported valences (see Figure A3-1). Page | 126

Figure A3-1: a) Overall and b) continuous ratings of emotional valence for the two film clips, “Bill Cosby” and “The Power of One”, averaged across 18 healthy participants. Error bars signify standard error of the mean (SEM).

The Impact of Medication on Averaged Sub-Network Scores

We used logistic regression to test whether our observed interaction differences across a sub-network of edges (see section 4.4.3) were confounded by differing medication classes. Clinical participants were divided into two sub-sets: those in receipt of an SSRI medication, versus those not on any such medication; and those on any other non-SSRI medication (e.g., antipsychotics, mood stabilisers, all broad-action antidepressants) compared to those who were not taking non-SSRI medication. We controlled for clinical group and showed that sub-network scores across rest and negative film viewing were not predictive of differing medication classes (see Table A3-1).

Page | 127

Table A3-1: Prediction of presence or absence of differing drug classes, controlling for clinical group, from interaction of rest and negative film viewing sub-network edge weights. Dependent Variables Predictors SSRI (Yes/No) Non-SSRI Drug (Yes/No) Rest * Negative Film Viewing† Exp (β) Wald Sig. Exp (β) Wald Sig. 0.000 1.331 0.249 0.000 1.261 0.262 † Controlling for clinical group (melancholic/non-melancholic)

Page | 128

Bibliography

Aburn, M. J., Holmes, C. A., Roberts, J. A., Boonstra, T. W., & Breakspear, M. (2012). Critical fluctuations in cortical models near instability. Front Physiol, 3, 331.

Adams, R. A., Shipp, S., & Friston, K. J. (2013). Predictions not commands: active inference in the motor system. Brain Struct Funct, 218(3), 611-643.

Alexopoulos, G. S., Meyers, B. S., Young, R. C., Campbell, S., Silbersweig, D., & Charlson, M. (1997). ‘Vascular depression’ hypothesis. Arch Gen Psychiatry, 54(10), 915-922.

Anand, A., Li, Y., Wang, Y., Wu, J., Gao, S., Bukhari, L., et al. (2005). Activity and connectivity of brain mood regulating circuit in depression: a functional magnetic resonance study. Biol Psychiatry, 57(10), 1079-1088.

Andreasen, N. C. (1988). Brain imaging: applications in psychiatry. Science, 239(4846), 1381-1388.

Andreasen, N. C., Flashman, L., Flaum, M., Arndt, S., Swayze, V., II, O’Leary, D. S., et al. (1994). Regional brain abnormalities in schizophrenia measured with magnetic resonance imaging. JAMA, 272(22), 1763-1769.

Andreasen, N. C., Paradiso, S., & O’Leary, D. S. (1998). “Cognitive dysmetria” as an integrative theory of schizophrenia: a dysfunction in cortical-subcortical- cerebellar circuitry? Schizophr Bull, 24(2), 203-218.

APA. (1980). Diagnostic and Statistical Manual of Mental Disorders, 3rd ed. (DSM- III). Washington, DC: American Psychiatric Association.

APA. (2000). Diagnostic and Statistical Manual of Mental Disorders, 4th ed., Text Revision (DSM-IV-TR). Washington, DC: American Psychiatric Association.

APA. (2013). Diagnostic and Statistical Manual of Mental Disorders, 5th ed. (DSM-V). Washington, DC: American Psychiatric Association.

Page | 129

Apps, M. A., & Tsakiris, M. (2014). The free-energy self: a predictive coding account of self-recognition. Neurosci Biobehav Rev, 41, 85-97.

Attwell, D., Buchan, A. M., Charpak, S., Lauritzen, M., Macvicar, B. A., & Newman, E. A. (2010). Glial and neuronal control of brain blood flow. Nature, 468(7321), 232-243.

Augustine, J. R. (1996). Circuitry and functional aspects of the insular lobe in primates including humans. Brain Res Brain Res Rev, 22(3), 229-244.

Austin, M. P., Mitchell, P., & Goodwin, G. M. (2001). Cognitive deficits in depression: possible implications for functional neuropathology. Br J Psychiatry, 178(3), 200-206.

Austin, M. P., Mitchell, P., Wilhelm, K., Parker, G., Hickie, I., Brodaty, H., et al. (1999). Cognitive function in depression: a distinct pattern of frontal impairment in melancholia? Psychol Med, 29(1), 73-85.

Austin, M. P., Ross, M., Murray, C., O’Carroll, R. E., Ebmeier, K. P., & Goodwin, G. M. (1992). Cognitive function in major depression. J Affect Disord, 25(1), 21- 29.

Avery, J. A., Drevets, W. C., Moseman, S. E., Bodurka, J., Barcalow, J. C., & Simmons, W. K. (2014). Major depressive disorder is associated with abnormal interoceptive activity and functional connectivity in the insula. Biol Psychiatry, 76(3), 258-266.

Baddeley, A. (2003). Working memory: looking back and looking forward. Nat Rev Neurosci, 4(10), 829-839.

Bartels, A., Zeki, S., & Logothetis, N. K. (2008). Natural vision reveals regional specialization to local motion and to contrast-invariant, global flow in the human brain. Cereb Cortex, 18(3), 705-717.

Page | 130

Barttfeld, P., Wicker, B., Cukier, S., Navarta, S., Lew, S., & Sigman, M. (2011). A big- world network in ASD: dynamical connectivity analysis reflects a deficit in long-range connections and an excess of short-range connections. Neuropsychologia, 49(2), 254-263.

Bassett, D. S., & Bullmore, E. T. (2009). Human brain networks in health and disease. Curr Opin Neurol, 22(4), 340-347.

Baxter, L. R., Jr., Phelps, M. E., Mazziotta, J. C., Schwartz, J. M., Gerner, R. H., Selin, C. E., et al. (1985). Cerebral metabolic rates for glucose in mood disorders. Studies with positron emission tomography and fluorodeoxyglucose F 18. Arch Gen Psychiatry, 42(5), 441-447.

Baxter, L. R., Jr., Schwartz, J. M., Phelps, M. E., Mazziotta, J. C., Guze, B. H., Selin, C. E., et al. (1989). Reduction of prefrontal cortex glucose metabolism common to three types of depression. Arch Gen Psychiatry, 46(3), 243-250.

Bayes, T. (1763). An essay towards solving a problem in the doctrine of chances. Philosophical Transactions of the Royal Society of London, 53, 370-418.

Beck, A. T. (1979). Cognitive Therapy of Depression. New York, NY: Guildford Press.

Beck, A. T., & Alford, B. A. (2009). Depression: Causes and Treatment (2nd ed.). Philadelphia, PA: University of Pennsylvania Press.

Beckmann, C. F., DeLuca, M., Devlin, J. T., & Smith, S. M. (2005). Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc Lond B Biol Sci, 360(1457), 1001-1013.

Beckmann, C. F., & Smith, S. M. (2004). Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging, 23(2), 137-152.

Behrens, T. E., Woolrich, M. W., Walton, M. E., & Rushworth, M. F. (2007). Learning the value of information in an uncertain world. Nat Neurosci, 10(9), 1214-1221.

Page | 131

Bell, A. J., & Sejnowski, T. J. (1995). An information-maximization approach to blind separation and blind deconvolution. Neural Comput, 7(6), 1129-1159.

Bench, C. J., Friston, K. J., Brown, R. G., Scott, L. C., Frackowiak, R. S., & Dolan, R. J. (1992). The anatomy of melancholia: focal abnormalities of cerebral blood flow in major depression. Psychol Med, 22(3), 607-615.

Berrios, G. E. (1985). “Depressive pseudodementia” or “Melancholic dementia”: a 19th century view. J Neurol Neurosurg Psychiatry, 48(5), 393-400.

Bigler, E. D. (2001). Neuropsychological testing defines the neurobehavioral significance of neuroimaging-identified abnormalities. Arch Clin Neuropsychol, 16(3), 227-236.

Biver, F., Goldman, S., Delvenne, V., Luxen, A., De Maertelaer, V., Hubain, P., et al. (1994). Frontal and parietal metabolic disturbances in unipolar depression. Biol Psychiatry, 36(6), 381-388.

Borji, A., Sihite, D. N., & Itti, L. (2012). An object-based Bayesian framework for top- down visual attention. Paper presented at the 26th AAAI Conference on Artificial Intelligence, Palo Alto, CA.

Botvinick, M. M., Braver, T. S., Barch, D. M., Carter, C. S., & Cohen, J. D. (2001). Conflict monitoring and cognitive control. Psychol Rev, 108(3), 624-652.

Breakspear, M. (2013). Dynamic and stochastic models of neuroimaging data: a comment on Lohmann et al. Neuroimage, 75, 270-274; discussion 279-281.

Breakspear, M., Terry, J. R., Friston, K. J., Harris, A. W., Williams, L. M., Brown, K., et al. (2003). A disturbance of nonlinear interdependence in scalp EEG of subjects with first episode schizophrenia. Neuroimage, 20(1), 466-478.

Bressler, S. L., & Menon, V. (2010). Large-scale brain networks in cognition: emerging methods and principles. Trends Cogn Sci, 14(6), 277-290.

Page | 132

Broadbent, D. E. (1954). The role of auditory localization in attention and memory span. J Exp Psychol, 47(3), 191-196.

Broadhead, W. E., Blazer, D. G., George, L. K., & Tse, C. K. (1990). Depression, disability days, and days lost from work in a prospective epidemiologic survey. JAMA, 264(19), 2524-2528.

Buchel, C., Coull, J. T., & Friston, K. J. (1999). The predictive value of changes in effective connectivity for human learning. Science, 283(5407), 1538-1541.

Buchel, C., & Friston, K. J. (1997). Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI. Cereb Cortex, 7(8), 768-778.

Cabeza, R., & Nyberg, L. (1997). Imaging cognition: An empirical review of PET studies with normal subjects. J Cogn Neurosci, 9(1), 1-26.

Cabeza, R., & Nyberg, L. (2000). Imaging cognition II: An empirical review of 275 PET and fMRI studies. J Cogn Neurosci, 12(1), 1-47.

Carroll, B. J. (1982). The dexamethasone suppression test for melancholia. Br J Psychiatry, 140(3), 292-304.

Carroll, B. J., Feinberg, M., Greden, J. F., Tarika, J., Albala, A. A., Haskett, R. F., et al. (1981). A specific laboratory test for the diagnosis of melancholia. Standardization, validation, and clinical utility. Arch Gen Psychiatry, 38(1), 15- 22.

Carter, C. S., & Krug, M. K. (2012). Dynamic cognitive control and fronto-cingulate interactions. In M. I. Posner (Ed.), Cognitive Neuroscience of Attention (2nd ed.). New York, NY: Guildford Press.

Chamberlain, S. R., Muller, U., Blackwell, A. D., Clark, L., Robbins, T. W., & Sahakian, B. J. (2006). Neurochemical modulation of response inhibition and probabilistic learning in humans. Science, 311(5762), 861-863.

Page | 133

Chaytor, N., & Schmitter-Edgecombe, M. (2003). The ecological validity of neuropsychological tests: a review of the literature on everyday cognitive skills. Neuropsychol Rev, 13(4), 181-197.

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behav Brain Sci, 36(3), 181-204.

Clark, D. M., & Teasdale, J. D. (1982). Diurnal variation in clinical depression and accessibility of memories of positive and negative experiences. J Abnorm Psychol, 91(2), 87-95.

Cohen, R., Lohr, I., Paul, R., & Boland, R. (2001). Impairments of attention and effort among patients with major affective disorders. J Neuropsychiatry Clin Neurosci, 13(3), 385-395.

Cole, M. W., Pathak, S., & Schneider, W. (2010). Identifying the brain’s most globally connected regions. Neuroimage, 49(4), 3132-3148.

Cole, M. W., Repovs, G., & Anticevic, A. (2014). The frontoparietal control system: A central role in mental health. Neuroscientist, 20(6), 652-664.

Cole, M. W., Yarkoni, T., Repovs, G., Anticevic, A., & Braver, T. S. (2012). Global connectivity of prefrontal cortex predicts cognitive control and intelligence. J Neurosci, 32(26), 8988-8999.

Corbetta, M. (1998). Frontoparietal cortical networks for directing attention and the eye to visual locations: identical, independent, or overlapping neural systems? Proc Natl Acad Sci U S A, 95(3), 831-838.

Corbetta, M., & Shulman, G. L. (2002). Control of goal-directed and stimulus-driven attention in the brain. Nat Rev Neurosci, 3(3), 201-215.

Critchley, H. D. (2009). Psychophysiology of neural, cognitive and affective integration: fMRI and autonomic indicants. Int J Psychophysiol, 73(2), 88-94.

Page | 134

Custance, J. (1952). Wisdom, Madness and Folly: The Philosophy of a Lunatic. New York, NY: Pellegrini & Cudahy.

Dalgleish, T. (2004). The emotional brain. Nat Rev Neurosci, 5(7), 583-589.

Damoiseaux, J. S., Rombouts, S. A., Barkhof, F., Scheltens, P., Stam, C. J., Smith, S. M., et al. (2006). Consistent resting-state networks across healthy subjects. Proc Natl Acad Sci U S A, 103(37), 13848-13853.

Daunizeau, J., David, O., & Stephan, K. E. (2011). Dynamic causal modelling: a critical review of the biophysical and statistical foundations. Neuroimage, 58(2), 312- 322.

Davey, C. G., Harrison, B. J., Yucel, M., & Allen, N. B. (2012). Regionally specific alterations in functional connectivity of the anterior cingulate cortex in major depressive disorder. Psychol Med, 42(10), 2071-2081.

Davidson, J., & Turnbull, C. D. (1986). Diagnostic significance of vegetative symptoms in depression. Br J Psychiatry, 148(4), 442-446.

Dayan, P., Hinton, G. E., Neal, R. M., & Zemel, R. S. (1995). The Helmholtz machine. Neural Comput, 7(5), 889-904.

Delamillieure, P., Doucet, G., Mazoyer, B., Turbelin, M. R., Delcroix, N., Mellet, E., et al. (2010). The resting state questionnaire: An introspective questionnaire for evaluation of inner experience during the conscious resting state. Brain Res Bull, 81(6), 565-573.

Depue, R. A., & Iacono, W. G. (1989). Neurobehavioral aspects of affective disorders. Annu Rev Psychol, 40, 457-492.

Desseilles, M., Schwartz, S., Dang-Vu, T. T., Sterpenich, V., Ansseau, M., Maquet, P., et al. (2011). Depression alters “top-down” visual attention: a dynamic causal modeling comparison between depressed and healthy subjects. Neuroimage, 54(2), 1662-1668.

Page | 135

Dolan, R. J., Bench, C. J., Brown, R. G., Scott, L. C., Friston, K. J., & Frackowiak, R. S. (1992). Regional cerebral blood flow abnormalities in depressed patients with cognitive impairment. J Neurol Neurosurg Psychiatry, 55(9), 768-773.

Drevets, W. C. (2000). Functional anatomical abnormalities in limbic and prefrontal cortical structures in major depression. Prog Brain Res, 126, 413-431.

Drevets, W. C. (2001). Neuroimaging and neuropathological studies of depression: implications for the cognitive-emotional features of mood disorders. Curr Opin Neurobiol, 11(2), 240-249.

Drevets, W. C., Price, J. L., Simpson, J. R., Jr., Todd, R. D., Reich, T., Vannier, M., et al. (1997). Subgenual prefrontal cortex abnormalities in mood disorders. Nature, 386(6627), 824-827.

Dux, P. E., & Marois, R. (2009). The attentional blink: a review of data and theory. Atten Percept Psychophys, 71(8), 1683-1700.

Ebert, D., & Martus, P. (1994). Somatization as a core symptom of melancholic type depression. Evidence from a cross-cultural study. J Affect Disord, 32(4), 253- 256.

Elliott, R., Rubinsztein, J. S., Sahakian, B. J., & Dolan, R. J. (2002). The neural basis of mood-congruent processing biases in depression. Arch Gen Psychiatry, 59(7), 597-604.

Elliott, R., Sahakian, B. J., McKay, A. P., Herrod, J. J., Robbins, T. W., & Paykel, E. S. (1996). Neuropsychological impairments in unipolar depression: the influence of perceived failure on subsequent performance. Psychol Med, 26(5), 975-989.

Feldman, H., & Friston, K. J. (2010). Attention, uncertainty, and free-energy. Front Hum Neurosci, 4, 215.

Page | 136

Fitzgerald, P. B., Laird, A. R., Maller, J., & Daskalakis, Z. J. (2008). A meta-analytic study of changes in brain activation in depression. Hum Brain Mapp, 29(6), 683- 695.

Formisano, E., Linden, D. E., Di Salle, F., Trojano, L., Esposito, F., Sack, A. T., et al. (2002). Tracking the mind’s image in the brain I: time-resolved fMRI during visuospatial mental imagery. Neuron, 35(1), 185-194.

Fox, M. D., Snyder, A. Z., Vincent, J. L., Corbetta, M., Van Essen, D. C., & Raichle, M. E. (2005). The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci U S A, 102(27), 9673- 9678.

Franzen, G., & Ingvar, D. H. (1975). Abnormal distribution of cerebral activity in chronic schizophrenia. J Psychiatr Res, 12(3), 199-214.

Franzen, M. D. (2000). Reliability and Validity in Neuropsychological Assessment (2nd ed.). New York, NY: Plenum Publishers.

Friston, K. (2005). A theory of cortical responses. Philos Trans R Soc Lond B Biol Sci, 360(1456), 815-836.

Friston, K. (2008). Hierarchical models in the brain. PLoS Comput Biol, 4(11), e1000211.

Friston, K. (2009). The free-energy principle: a rough guide to the brain? Trends Cogn Sci, 13(7), 293-301.

Friston, K. (2012). Ten ironic rules for non-statistical reviewers. Neuroimage, 61(4), 1300-1310.

Friston, K., Adams, R. A., Perrinet, L., & Breakspear, M. (2012). Perceptions as hypotheses: saccades as experiments. Front Psychol, 3, 151.

Friston, K., Breakspear, M., & Deco, G. (2012). Perception and self-organized instability. Front Comput Neurosci, 6, 44. Page | 137

Friston, K. J. (1994). Functional and effective connectivity in neuroimaging: A synthesis. Hum Brain Mapp, 2(1-2), 56-78.

Friston, K. J. (1998). The disconnection hypothesis. Schizophr Res, 30(2), 115-125.

Friston, K. J. (2003). Functional integration. In R. S. J. Frackowiak, K. J. Friston, C. D. Frith, R. J. Dolan, C. J. Price, S. Zeki, J. T. Ashburner & W. D. Penny (Eds.), Human Brain Function (2nd ed.). London, UK: Academic Press.

Friston, K. J., Ashburner, J. T., Kiebel, S. J., Nichols, T. E., & Penny, W. D. (2007). Statistical Parametric Mapping: The Analysis of Functional Brain Images. London, UK: Academic Press.

Friston, K. J., & Frith, C. D. (1995). Schizophrenia: a disconnection syndrome? Clin Neurosci, 3(2), 89-97.

Friston, K. J., Harrison, L., & Penny, W. (2003). Dynamic causal modelling. Neuroimage, 19(4), 1273-1302.

Friston, K. J., Holmes, A. P., Worsley, K. J., Poline, J.-P., Frith, C. D., & Frackowiak, R. S. J. (1995). Statistical parametric maps in functional imaging: A general linear approach. Hum Brain Mapp, 2(4), 189-210.

Friston, K. J., Li, B., Daunizeau, J., & Stephan, K. E. (2011). Network discovery with DCM. Neuroimage, 56(3), 1202-1221.

Garrett, D. D., Samanez-Larkin, G. R., MacDonald, S. W., Lindenberger, U., McIntosh, A. R., & Grady, C. L. (2013). Moment-to-moment brain signal variability: a next frontier in human brain mapping? Neurosci Biobehav Rev, 37(4), 610-624.

Gelman, A., & Rubin, D. B. (1992). Inference from iterative simulation using multiple sequences. Statistical Science, 7(4), 457-511.

George, M. S., Ketter, T. A., & Post, R. M. (1994). Prefrontal cortex dysfunction in clinical depression. Depression, 2(2), 59-72.

Page | 138

Geweke, J. (1992). Evaluating the accuracy of sampling-based approaches to calcualting posterior moments. In J. M. Bernardo, J. O. Berger, A. P. Dawid & A. F. M. Smith (Eds.), Bayesian Statistics 4: Proceedings of the Fourth Valencia International Meeting. Oxford, UK: Clarendon Press.

Gilboa-Schechtman, E., Erhard-Weiss, D., & Jeczemien, P. (2002). Interpersonal deficits meet cognitive biases: memory for facial expressions in depressed and anxious men and women. Psychiatry Res, 113(3), 279-293.

Glimcher, P. W. (2003). Decisions, Uncertainty, and the Brain: The Science of Neuroeconomics. Cambridge, MA: The MIT Press.

Green, D. M., & Swets, J. A. (1966). Signal Detection Theory and Psychophysics. New York, NY: Wiley.

Gregory, R. L. (1980). Perceptions as hypotheses. Philos Trans R Soc Lond B Biol Sci, 290(1038), 181-197.

Greicius, M. D., Flores, B. H., Menon, V., Glover, G. H., Solvason, H. B., Kenna, H., et al. (2007). Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus. Biol Psychiatry, 62(5), 429-437.

Griffiths, T. L., & Tenenbaum, J. B. (2006). Optimal predictions in everyday cognition. Psychol Sci, 17(9), 767-773.

Gross, J. J., & Levenson, R. W. (1995). Emotion elicitation using films. Cognition & Emotion, 9(1), 87-108.

Gu, X., Hof, P. R., Friston, K. J., & Fan, J. (2013). Anterior insular cortex and emotional awareness. J Comp Neurol, 521(15), 3371-3388.

Hadzi-Pavlovic, D., & Boyce, P. (2012). Melancholia. Curr Opin Psychiatry, 25(1), 14- 18.

Page | 139

Hamilton, J. P., Etkin, A., Furman, D. J., Lemus, M. G., Johnson, R. F., & Gotlib, I. H. (2012). Functional neuroimaging of major depressive disorder: a meta-analysis and new integration of base line activation and neural response data. Am J Psychiatry, 169(7), 693-703.

Hampshire, A., & Owen, A. M. (2006). Fractionating attentional control using event- related fMRI. Cereb Cortex, 16(12), 1679-1689.

Harmer, C. J., Goodwin, G. M., & Cowen, P. J. (2009). Why do antidepressants take so long to work? A cognitive neuropsychological model of antidepressant drug action. Br J Psychiatry, 195(2), 102-108.

Hartlage, S., Alloy, L. B., Vazquez, C., & Dykman, B. (1993). Automatic and effortful processing in depression. Psychol Bull, 113(2), 247-278.

Hasson, U., Malach, R., & Heeger, D. J. (2010). Reliability of cortical activity during natural stimulation. Trends Cogn Sci, 14(1), 40-48.

Hasson, U., Nir, Y., Levy, I., Fuhrmann, G., & Malach, R. (2004). Intersubject synchronization of cortical activity during natural vision. Science, 303(5664), 1634-1640.

Hasson, U., Yang, E., Vallines, I., Heeger, D. J., & Rubin, N. (2008). A hierarchy of temporal receptive windows in human cortex. J Neurosci, 28(10), 2539-2550.

Hauser, M. D. (1999). Perseveration, inhibition and the prefrontal cortex: a new look. Curr Opin Neurobiol, 9(2), 214-222.

Heilman, K. M., & Valenstein, E. (2012). Clinical Neuropsychology (5th ed.). New York, NY: Oxford University Press.

Herman, J. P., & Cullinan, W. E. (1997). Neurocircuitry of stress: central control of the hypothalamo-pituitary-adrenocortical axis. Trends Neurosci, 20(2), 78-84.

Page | 140

Herman, J. P., Ostrander, M. M., Mueller, N. K., & Figueiredo, H. (2005). Limbic system mechanisms of stress regulation: hypothalamo-pituitary-adrenocortical axis. Prog Neuropsychopharmacol Biol Psychiatry, 29(8), 1201-1213.

Hickie, I. (1996). Issues in classification: III. Utilising behavioural constructs in melancholia research. In G. Parker & D. Hadzi-Pavlovic (Eds.), Melancholia: A Disorder of Movement and Mood (pp. 38-56). New York, NY: Cambridge University Press.

Hickie, I., Scott, E., Mitchell, P., Wilhelm, K., Austin, M. P., & Bennett, B. (1995). Subcortical hyperintensities on magnetic resonance imaging: clinical correlates and prognostic significance in patients with severe depression. Biol Psychiatry, 37(3), 151-160.

Hickie, I. B., Naismith, S. L., Ward, P. B., Little, C. L., Pearson, M., Scott, E. M., et al. (2007). Psychomotor slowing in older patients with major depression: Relationships with blood flow in the caudate nucleus and white matter lesions. Psychiatry Res, 155(3), 211-220.

Hintze, J. L., & Nelson, R. D. (1998). Violin plots: A box plot-density trace synergism. Am Statistician, 52(2), 181-184.

Honey, C. J., Thesen, T., Donner, T. H., Silbert, L. J., Carlson, C. E., Devinsky, O., et al. (2012). Slow cortical dynamics and the accumulation of information over long timescales. Neuron, 76(2), 423-434.

Huettel, S. A., Song, A. W., & McCarthy, G. (2009). Functional Magnetic Resonance Imaging (2nd ed.). Sunderland, MA: Sinauer Associates.

Huys, Q. J., & Dayan, P. (2009). A Bayesian formulation of behavioral control. Cognition, 113(3), 314-328.

Hyvarinen, A. (1999). Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans Neural Netw, 10(3), 626-634.

Page | 141

Ingram, R. E., & Siegle, G. J. (2009). Methodological issues in the study of depression. In I. H. Gotlib & C. L. Hammen (Eds.), Handbook of Depression (2nd ed.). New York, NY: The Guildford Press.

Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D. S., Quinn, K., et al. (2010). Research domain criteria (RDoC): toward a new classification framework for research on mental disorders. Am J Psychiatry, 167(7), 748-751.

Insel, T. R. (2014). The NIMH Research Domain Criteria (RDoC) project: Precision medicine for psychiatry. Am J Psychiatry, 171(4), 395-397.

Jaaskelainen, I. P., Koskentalo, K., Balk, M. H., Autti, T., Kauramaki, J., Pomren, C., et al. (2008). Inter-subject synchronization of prefrontal cortex hemodynamic activity during natural viewing. Open Neuroimag J, 2, 14-19.

James, W. (1890). The Principles of Psychology. New York, NY: Henry Holt.

Jaspers, K. (1913). Allgemeine Psychopathologie. Berlin: Springer.

Jirsa, V. K., & McIntosh, A. R. (2007). Handbook of Brain Connectivity. Berlin: Springer.

Johnson, J., Weissman, M. M., & Klerman, G. L. (1992). Service utilization and social morbidity associated with depressive symptoms in the community. JAMA, 267(11), 1478-1483.

Jones, E. G., & Mendell, L. M. (1999). Assessing the decade of the brain. Science, 284(5415), 739.

Karim, M., Harris, J. A., Langdon, A., & Breakspear, M. (2013). The influence of prior experience and expected timing on vibrotactile discrimination. Front Neurosci, 7, 255.

Keefe, R. S. (1995). The contribution of neuropsychology to psychiatry. Am J Psychiatry, 152(1), 6-15.

Page | 142

Kessler, R. C., Berglund, P., Demler, O., Jin, R., Koretz, D., Merikangas, K. R., et al. (2003). The epidemiology of major depressive disorder: results from the National Comorbidity Survey Replication (NCS-R). JAMA, 289(23), 3095-3105.

Kessler, R. C., Zhao, S., Blazer, D. G., & Swartz, M. (1997). Prevalence, correlates, and course of minor depression and major depression in the National Comorbidity Survey. J Affect Disord, 45(1-2), 19-30.

Knill, D. C., & Pouget, A. (2004). The Bayesian brain: the role of uncertainty in neural coding and computation. Trends Neurosci, 27(12), 712-719.

Laird, A. R., Fox, P. M., Eickhoff, S. B., Turner, J. A., Ray, K. L., McKay, D. R., et al. (2011). Behavioral interpretations of intrinsic connectivity networks. J Cogn Neurosci, 23(12), 4022-4037.

Lammertsma, A. A., & Frackowiak, R. S. (1985). Positron emission tomography. Crit Rev Biomed Eng, 13(2), 125-169.

Lee, M. D. (2008). BayesSDT: software for Bayesian inference with signal detection theory. Behav Res Methods, 40(2), 450-456.

Lee, M. D., & Wagenmakers, E.-J. (2014). Bayesian Cognitive Modeling: A Practical Course. Cambridge, UK: Cambridge University Press.

Lee, M. D., & Wagenmakers, E. J. (2005). Bayesian statistical inference in psychology: comment on Trafimow. Psychol Rev, 112(3), 662-668.

Lemelin, S., & Baruch, P. (1998). Clinical psychomotor retardation and attention in depression. J Psychiatr Res, 32(2), 81-88.

Lemelin, S., Baruch, P., Vincent, A., Laplante, L., Everett, J., & Vincent, P. (1996). Attention disturbance in clinical depression. Deficient distractor inhibition or processing resource deficit? J Nerv Ment Dis, 184(2), 114-121.

Page | 143

Levens, S. M., & Phelps, E. A. (2010). Insula and orbital frontal cortex activity underlying emotion interference resolution in working memory. J Cogn Neurosci, 22(12), 2790-2803.

Li, B., Daunizeau, J., Stephan, K. E., Penny, W., Hu, D., & Friston, K. (2011). Generalised filtering and stochastic DCM for fMRI. Neuroimage, 58(2), 442- 457.

Lindley, D. V. (1965). Introduction to Probability and Statistics from a Bayesian Viewpoint. Part 2: Inference. Cambridge, UK: Cambridge University Press.

Lindquist, K. A., & Barrett, L. F. (2012). A functional architecture of the human brain: emerging insights from the science of emotion. Trends Cogn Sci, 16(11), 533- 540.

Logan, G. D. (2004). Cumulative progress in formal theories of attention. Annu Rev Psychol, 55, 207-234.

Lu, Q., Li, H., Luo, G., Wang, Y., Tang, H., Han, L., et al. (2012). Impaired prefrontal- amygdala effective connectivity is responsible for the dysfunction of emotion process in major depressive disorder: a dynamic causal modeling study on MEG. Neurosci Lett, 523(2), 125-130.

Lunn, D. J., Thomas, A., Best, N., & Spiegelhalter, D. (2000). WinBUGS - a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and Computing, 10(4), 325-337.

MacLean, P. D. (1949). Psychosomatic disease and the “visceral brain”: Recent developments bearing on the Papez theory of emotion. Psychosom Med, 11(6), 338-353.

Maclean, P. D. (1952). Some psychiatric implications of physiological studies on frontotemporal portion of limbic system (visceral brain). Electroencephalogr Clin Neurophysiol, 4(4), 407-418.

Page | 144

Maes, M., Dierckx, R., Meltzer, H. Y., Ingels, M., Schotte, C., Vandewoude, M., et al. (1993). Regional cerebral blood flow in unipolar depression measured with Tc- 99m-HMPAO single photon emission computed tomography: negative findings. Psychiatry Res, 50(2), 77-88.

Markett, S., Reuter, M., Montag, C., Voigt, G., Lachmann, B., Rudorf, S., et al. (2014). Assessing the function of the fronto-parietal attention network: Insights from resting-state fMRI and the attentional network test. Hum Brain Mapp, 35(4), 1700-1709.

Marreiros, A. C., Stephan, K. E., & Friston, K. J. (2010). Dynamic Causal Modelling. Scholarpedia, 5(7), 9568.

Mathews, A., & MacLeod, C. (2005). Cognitive vulnerability to emotional disorders. Annu Rev Clin Psychol, 1, 167-195.

Mathews, A., Ridgeway, V., & Williamson, D. A. (1996). Evidence for attention to threatening stimuli in depression. Behav Res Ther, 34(9), 695-705.

Mathys, C., Daunizeau, J., Friston, K. J., & Stephan, K. E. (2011). A Bayesian foundation for individual learning under uncertainty. Front Hum Neurosci, 5, 39.

Mayberg, H. S. (1997). Limbic-cortical dysregulation: a proposed model of depression. J Neuropsychiatry Clin Neurosci, 9(3), 471-481.

Mayberg, H. S., Lozano, A. M., Voon, V., McNeely, H. E., Seminowicz, D., Hamani, C., et al. (2005). Deep brain stimulation for treatment-resistant depression. Neuron, 45(5), 651-660.

Menon, V. (2011). Large-scale brain networks and psychopathology: a unifying triple network model. Trends Cogn Sci, 15(10), 483-506.

Menon, V., & Uddin, L. Q. (2010). Saliency, switching, attention and control: a network model of insula function. Brain Struct Funct, 214(5-6), 655-667.

Page | 145

Mialet, J. P., Pope, H. G., & Yurgelun-Todd, D. (1996). Impaired attention in depressive states: a non-specific deficit? Psychol Med, 26(5), 1009-1020.

Micheloyannis, S., Pachou, E., Stam, C. J., Breakspear, M., Bitsios, P., Vourkas, M., et al. (2006). Small-world networks and disturbed functional connectivity in schizophrenia. Schizophr Res, 87(1-3), 60-66.

Michopoulos, I., Zervas, I. M., Pantelis, C., Tsaltas, E., Papakosta, V. M., Boufidou, F., et al. (2008). Neuropsychological and hypothalamic-pituitary-axis function in female patients with melancholic and non-melancholic depression. Eur Arch Psychiatry Clin Neurosci, 258(4), 217-225.

Michopoulos, I., Zervas, I. M., Papakosta, V. M., Tsaltas, E., Papageorgiou, C., Manessi, T., et al. (2006). Set shifting deficits in melancholic vs. non- melancholic depression: preliminary findings. Eur Psychiatry, 21(6), 361-363.

Miller, D. D., Andreasen, N. C., O’Leary, D. S., Rezai, K., Watkins, G. L., Ponto, L. L., et al. (1997). Effect of antipsychotics on regional cerebral blood flow measured with positron emission tomography. Neuropsychopharmacology, 17(4), 230- 240.

Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annu Rev Neurosci, 24, 167-202.

Miller, W. R. (1975). Psychological deficit in depression. Psychol Bull, 82(2), 238-260.

Mirsky, A. F., Anthony, B. J., Duncan, C. C., Ahearn, M. B., & Kellam, S. G. (1991). Analysis of the elements of attention: a neuropsychological approach. Neuropsychol Rev, 2(2), 109-145.

Mitchell, P. (1996). Validity of the CORE: I. A neuroendocrinological strategy. In G. Parker & D. Hadzi-Pavlovic (Eds.), Melancholia: A Disorder of Movement and Mood (pp. 138-148). Cambridge, UK: Cambridge University Press.

Page | 146

Moffoot, A. P., O’Carroll, R. E., Bennie, J., Carroll, S., Dick, H., Ebmeier, K. P., et al. (1994). Diurnal variation of mood and neuropsychological function in major depression with melancholia. J Affect Disord, 32(4), 257-269.

Montague, P. R., Dolan, R. J., Friston, K. J., & Dayan, P. (2012). Computational psychiatry. Trends Cogn Sci, 16(1), 72-80.

Moray, N. (1970). Attention: Selective Processes in Vision and Hearing. New York, NY: Academic Press.

Murphy, F. C., Sahakian, B. J., Rubinsztein, J. S., Michael, A., Rogers, R. D., Robbins, T. W., et al. (1999). Emotional bias and inhibitory control processes in mania and depression. Psychol Med, 29(6), 1307-1321.

Naci, L., Cusack, R., Anello, M., & Owen, A. M. (2014). A common neural code for similar conscious experiences in different individuals. Proc Natl Acad Sci U S A, 111(39), 14277-14282.

Nobre, A. C., & Kastner, S. (2014). The Oxford Handbook of Attention. Oxford, UK: Oxford University Press.

Ogawa, S., Lee, T. M., Kay, A. R., & Tank, D. W. (1990). Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci U S A, 87(24), 9868-9872.

Owen, A. M., Roberts, A. C., Polkey, C. E., Sahakian, B. J., & Robbins, T. W. (1991). Extra-dimensional versus intra-dimensional set shifting performance following frontal lobe excisions, temporal lobe excisions or amygdalo-hippocampectomy in man. Neuropsychologia, 29(10), 993-1006.

Paelecke-Habermann, Y., Pohl, J., & Leplow, B. (2005). Attention and executive functions in remitted major depression patients. J Affect Disord, 89(1-3), 125- 135.

Page | 147

Papez, J. W. (1937). A proposed mechanism of emotion. Arch Neur Psych, 38(4), 725- 743.

Parker, G. (2000). Classifying depression: should paradigms lost be regained? Am J Psychiatry, 157(8), 1195-1203.

Parker, G. (2005). Beyond major depression. Psychol Med, 35(4), 467-474.

Parker, G. (2007). Defining melancholia: the primacy of psychomotor disturbance. Acta Psychiatr Scand Suppl(433), 21-30.

Parker, G. (2011). Classifying clinical depression: an operational proposal. Acta Psychiatr Scand, 123(4), 314-316.

Parker, G., Fink, M., Shorter, E., Taylor, M. A., Akiskal, H., Berrios, G., et al. (2010). Issues for DSM-5: whither melancholia? The case for its classification as a distinct mood disorder. Am J Psychiatry, 167(7), 745-747.

Parker, G., Fletcher, K., Hyett, M., Hadzi-Pavlovic, D., Barrett, M., & Synnott, H. (2009). Measuring melancholia: the utility of a prototypic symptom approach. Psychol Med, 39(6), 989-998.

Parker, G., & Hadzi Pavlovic, D. (1996). Melancholia: A Disorder of Movement and Mood. New York, NY: Cambridge University Press.

Parker, G., Hadzi-Pavlovic, D., & Boyce, P. (1996). Issues in classification II: Classifying melancholia. In G. Parker & D. Hadzi-Pavlovic (Eds.), Melancholia: A Disorder of Movement and Mood (pp. 20-37). New York, NY: Cambridge University Press.

Parker, G., Hadzi-Pavlovic, D., Wilhelm, K., Hickie, I., Brodaty, H., Boyce, P., et al. (1994). Defining melancholia: properties of a refined sign-based measure. Br J Psychiatry, 164(3), 316-326.

Paulus, M. P., & Yu, A. J. (2012). Emotion and decision-making: affect-driven belief systems in anxiety and depression. Trends Cogn Sci, 16(9), 476-483. Page | 148

Paykel, E. S. (1971). Classification of depressed patients: a cluster analysis derived grouping. Br J Psychiatry, 118(544), 275-288.

Pearson Assessments. (2001). Wechsler Test of Adult Reading™ (WTAR™). San Antonio, TX: Pearson.

Peppiatt, C. M., Howarth, C., Mobbs, P., & Attwell, D. (2006). Bidirectional control of CNS capillary diameter by pericytes. Nature, 443(7112), 700-704.

Pessoa, L. (2008). On the relationship between emotion and cognition. Nat Rev Neurosci, 9(2), 148-158.

Pessoa, L., Gutierrez, E., Bandettini, P., & Ungerleider, L. (2002). Neural correlates of visual working memory: fMRI amplitude predicts task performance. Neuron, 35(5), 975-987.

Petersen, S. E., & Posner, M. I. (2012). The attention system of the human brain: 20 years after. Annu Rev Neurosci, 35, 73-89.

Phillips, M. L., Travis, M. J., Fagiolini, A., & Kupfer, D. J. (2008). Medication effects in neuroimaging studies of bipolar disorder. Am J Psychiatry, 165(3), 313-320.

Pizzagalli, D. A., Jahn, A. L., & O’Shea, J. P. (2005). Toward an objective characterization of an anhedonic phenotype: a signal-detection approach. Biol Psychiatry, 57(4), 319-327.

Pizzagalli, D. A., Oakes, T. R., Fox, A. S., Chung, M. K., Larson, C. L., Abercrombie, H. C., et al. (2004). Functional but not structural subgenual prefrontal cortex abnormalities in melancholia. Mol Psychiatry, 9(4), 325, 393-405.

Plummer, M., Best, N., Cowles, K., & Vines, K. (2006). CODA: Convergence Diagnosis and Output Analysis for MCMC. R News, 6(1), 7-11.

Poldrack, R. A. (2006). Can cognitive processes be inferred from neuroimaging data? Trends Cogn Sci, 10(2), 59-63.

Page | 149

Politis, A., Lykouras, L., Mourtzouchou, P., & Christodoulou, G. N. (2004). Attentional disturbances in patients with unipolar psychotic depression: a selective and sustained attention study. Compr Psychiatry, 45(6), 452-459.

Posner, M. I. (2011). Cognitive Neuroscience of Attention (2nd ed.). New York, NY: The Guilford Press.

Posner, M. I., & Boies, S. J. (1971). Components of attention. Psychol Rev, 78(5), 391- 408.

Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annu Rev Neurosci, 13, 25-42.

Posner, M. I., Walker, J. A., Friedrich, F. J., & Rafal, R. D. (1984). Effects of parietal injury on covert orienting of attention. J Neurosci, 4(7), 1863-1874.

Price, C. J., & Friston, K. J. (2005). Functional ontologies for cognition: The systematic definition of structure and function. Cogn Neuropsychol, 22(3), 262-275.

Ridderinkhof, K. R., van den Wildenberg, W. P., Segalowitz, S. J., & Carter, C. S. (2004). Neurocognitive mechanisms of cognitive control: the role of prefrontal cortex in action selection, response inhibition, performance monitoring, and reward-based learning. Brain Cogn, 56(2), 129-140.

Robbins, T. W., James, M., Owen, A. M., Sahakian, B. J., McInnes, L., & Rabbitt, P. (1994). Cambridge Neuropsychological Test Automated Battery (CANTAB): a factor analytic study of a large sample of normal elderly volunteers. Dementia, 5(5), 266-281.

Rogers, M. A., Bellgrove, M. A., Chiu, E., Mileshkin, C., & Bradshaw, J. L. (2004). Response selection deficits in melancholic but not nonmelancholic unipolar major depression. J Clin Exp Neuropsychol, 26(2), 169-179.

Rouder, J. N., & Lu, J. (2005). An introduction to Bayesian hierarchical models with an application in the theory of signal detection. Psychon Bull Rev, 12(4), 573-604.

Page | 150

Roy, M., Shohamy, D., & Wager, T. D. (2012). Ventromedial prefrontal-subcortical systems and the generation of affective meaning. Trends Cogn Sci, 16(3), 147- 156.

Rubinow, D. R., Post, R. M., Savard, R., & Gold, P. W. (1984). Cortisol hypersecretion and cognitive impairment in depression. Arch Gen Psychiatry, 41(3), 279-283.

Rush, A. J., Trivedi, M. H., Ibrahim, H. M., Carmody, T. J., Arnow, B., Klein, D. N., et al. (2003). The 16-Item Quick Inventory of Depressive Symptomatology (QIDS), clinician rating (QIDS-C), and self-report (QIDS-SR): a psychometric evaluation in patients with chronic major depression. Biol Psychiatry, 54(5), 573-583.

Sackeim, H. A., Prohovnik, I., Moeller, J. R., Brown, R. P., Apter, S., Prudic, J., et al. (1990). Regional cerebral blood flow in mood disorders. I. Comparison of major depressives and normal controls at rest. Arch Gen Psychiatry, 47(1), 60-70.

Sankoh, A. J., Huque, M. F., & Dubey, S. D. (1997). Some comments on frequently used multiple endpoint adjustment methods in clinical trials. Stat Med, 16(22), 2529-2542.

Schildkraut, J. J. (1965). The catecholamine hypothesis of affective disorders: a review of supporting evidence. Am J Psychiatry, 122(5), 509-522.

Schlosser, R. G., Wagner, G., Koch, K., Dahnke, R., Reichenbach, J. R., & Sauer, H. (2008). Fronto-cingulate effective connectivity in major depression: a study with fMRI and dynamic causal modeling. Neuroimage, 43(3), 645-655.

Schneider, W., & Shiffrin, R. M. (1977). Controlled and automatic human information processing: I. Detection, search, and attention. Psychol Rev, 84, 1-66.

Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593-1599.

Page | 151

Schulz, K. P., Fan, J., Magidina, O., Marks, D. J., Hahn, B., & Halperin, J. M. (2007). Does the emotional go/no-go task really measure behavioral inhibition? Convergence with measures on a non-emotional analog. Arch Clin Neuropsychol, 22(2), 151-160.

Seeley, W. W., Crawford, R. K., Zhou, J., Miller, B. L., & Greicius, M. D. (2009). Neurodegenerative diseases target large-scale human brain networks. Neuron, 62(1), 42-52.

Seeley, W. W., Menon, V., Schatzberg, A. F., Keller, J., Glover, G. H., Kenna, H., et al. (2007). Dissociable intrinsic connectivity networks for salience processing and executive control. J Neurosci, 27(9), 2349-2356.

Seminowicz, D. A., Mayberg, H. S., McIntosh, A. R., Goldapple, K., Kennedy, S., Segal, Z., et al. (2004). Limbic-frontal circuitry in major depression: a path modeling metanalysis. Neuroimage, 22(1), 409-418.

Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends Cogn Sci, 17(11), 565-573.

Sheehan, D. V., Lecrubier, Y., Sheehan, K. H., Amorim, P., Janavs, J., Weiller, E., et al. (1998). The Mini-International Neuropsychiatric Interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry, 59(Suppl 20), 22-33.

Sheline, Y. I., Barch, D. M., Price, J. L., Rundle, M. M., Vaishnavi, S. N., Snyder, A. Z., et al. (2009). The default mode network and self-referential processes in depression. Proc Natl Acad Sci U S A, 106(6), 1942-1947.

Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychol Rev, 84, 127-190.

Page | 152

Shiner, T., Symmonds, M., Guitart-Masip, M., Fleming, S. M., Friston, K. J., & Dolan, R. J. (2014). Dopamine, salience, and response set shifting in prefrontal cortex. Cereb Cortex, Sep 21 (Epub ahead of print).

Shorter, E. (2013). How Everyond Became Depressed: The Rise and Fall of the Nervous Breakdown. New York, NY: Oxford University Press.

Siegle, G. J., Steinhauer, S. R., Thase, M. E., Stenger, V. A., & Carter, C. S. (2002). Can’t shake that feeling: event-related fMRI assessment of sustained amygdala activity in response to emotional information in depressed individuals. Biol Psychiatry, 51(9), 693-707.

Smith, S. M., Fox, P. T., Miller, K. L., Glahn, D. C., Fox, P. M., Mackay, C. E., et al. (2009). Correspondence of the brain’s functional architecture during activation and rest. Proc Natl Acad Sci U S A, 106(31), 13040-13045.

Sobin, C., & Sackeim, H. A. (1997). Psychomotor symptoms of depression. Am J Psychiatry, 154(1), 4-17.

Song, M., Zhou, Y., Li, J., Liu, Y., Tian, L., Yu, C., et al. (2008). Brain spontaneous functional connectivity and intelligence. Neuroimage, 41(3), 1168-1176.

Soriano-Mas, C., Hernandez-Ribas, R., Pujol, J., Urretavizcaya, M., Deus, J., Harrison, B. J., et al. (2011). Cross-sectional and longitudinal assessment of structural brain alterations in melancholic depression. Biol Psychiatry, 69(4), 318-325.

Spielberger, C. D. (1983). Manual for the State-Trait Anxiety Inventory (Form Y). Menlo Park, CA: Mind Garden.

Spiers, H. J., & Maguire, E. A. (2007). Decoding human brain activity during real- world experiences. Trends Cogn Sci, 11(8), 356-365.

Sporns, O. (2010). Networks of the Brain. Cambridge, MA: The MIT Press.

Page | 153

Sporns, O., Chialvo, D. R., Kaiser, M., & Hilgetag, C. C. (2004). Organization, development and function of complex brain networks. Trends Cogn Sci, 8(9), 418-425.

Stephan, K. E., Baldeweg, T., & Friston, K. J. (2006). Synaptic plasticity and dysconnection in schizophrenia. Biol Psychiatry, 59(10), 929-939.

Stephan, K. E., Kasper, L., Harrison, L. M., Daunizeau, J., den Ouden, H. E., Breakspear, M., et al. (2008). Nonlinear dynamic causal models for fMRI. Neuroimage, 42(2), 649-662.

Stroop, J. R. (1935). Studies of interference in serial verbal reactions. J Exp Psychol, 18(6), 643-662.

Tancer, M. E., Brown, T. M., Evans, D. L., Ekstrom, D., Haggerty, J. J., Jr., Pedersen, C., et al. (1990). Impaired effortful cognition in depression. Psychiatry Res, 31(2), 161-168.

Taylor, M. A., & Fink, M. (2006). Melancholia: The Diagnosis, Pathophysiology and Treatment of Depressive Illness. New York, NY: Cambridge University Press.

Thomas, P., Goudemand, M., & Rousseaux, M. (1999). Attentional resources in major depression. Eur Arch Psychiatry Clin Neurosci, 249(2), 79-85.

Thomson, K. C., & Hendrie, H. C. (1972). Environmental stress in primary depressive illness. Arch Gen Psychiatry, 26(2), 130-132.

Tononi, G., Sporns, O., & Edelman, G. M. (1994). A measure for brain complexity: relating functional segregation and integration in the nervous system. Proc Natl Acad Sci U S A, 91(11), 5033-5037.

Tutus, A., Simsek, A., Sofuoglu, S., Nardali, M., Kugu, N., Karaaslan, F., et al. (1998). Changes in regional cerebral blood flow demonstrated by single photon emission computed tomography in depressive disorders: comparison of unipolar vs. bipolar subtypes. Psychiatry Res, 83(3), 169-177.

Page | 154

van den Heuvel, M. P., Stam, C. J., Kahn, R. S., & Hulshoff Pol, H. E. (2009). Efficiency of functional brain networks and intellectual performance. J Neurosci, 29(23), 7619-7624.

Veer, I. M., Beckmann, C. F., van Tol, M. J., Ferrarini, L., Milles, J., Veltman, D. J., et al. (2010). Whole brain resting-state analysis reveals decreased functional connectivity in major depression. Front Syst Neurosci, 4.

Vincent, J. L., Kahn, I., Snyder, A. Z., Raichle, M. E., & Buckner, R. L. (2008). Evidence for a frontoparietal control system revealed by intrinsic functional connectivity. J Neurophysiol, 100(6), 3328-3342.

Vossel, S., Geng, J. J., & Friston, K. J. (2014). Attention, predictions and expectations, and their violation: attentional control in the human brain. Front Hum Neurosci, 8, 490.

Vul, E., Hanus, D., & Kanwisher, N. (2009). Attention as inference: selection is probabilistic; responses are all-or-none samples. J Exp Psychol Gen, 138(4), 546-560.

Wagner, G., Koch, K., Schachtzabel, C., Sobanski, T., Reichenbach, J. R., Sauer, H., et al. (2010). Differential effects of serotonergic and noradrenergic antidepressants on brain activity during a cognitive control task and neurofunctional prediction of treatment outcome in patients with depression. J Psychiatry Neurosci, 35(4), 247-257.

Wagner, G., Sinsel, E., Sobanski, T., Kohler, S., Marinou, V., Mentzel, H. J., et al. (2006). Cortical inefficiency in patients with unipolar depression: an event- related fMRI study with the Stroop task. Biol Psychiatry, 59(10), 958-965.

Weinberger, D. R., Berman, K. F., & Zec, R. F. (1986). Physiologic dysfunction of dorsolateral prefrontal cortex in schizophrenia. I. Regional cerebral blood flow evidence. Arch Gen Psychiatry, 43(2), 114-124.

Page | 155

Wells, K. B., Stewart, A., Hays, R. D., Burnam, M. A., Rogers, W., Daniels, M., et al. (1989). The functioning and well-being of depressed patients. Results from the Medical Outcomes Study. JAMA, 262(7), 914-919.

Wernicke, C. (1885). Some new studies on aphasie. Fortschr Med, 824-830.

Wernicke, C. (1906). Grundrisse der Psychiatrie. Thieme.

WHO. (2008). The Global Burden of Disease: 2004 Update. Geneva, Switzerland.

Wickens, T. D. (2002). Elementary Signal Detection Theory. Oxford, UK: Oxford University Press.

Wiech, K., Lin, C. S., Brodersen, K. H., Bingel, U., Ploner, M., & Tracey, I. (2010). Anterior insula integrates information about salience into perceptual decisions about pain. J Neurosci, 30(48), 16324-16331.

Xue, G., Lu, Z., Levin, I. P., & Bechara, A. (2010). The impact of prior risk experiences on subsequent risky decision-making: the role of the insula. Neuroimage, 50(2), 709-716.

Yu, A. J., & Dayan, P. (2005). Uncertainty, neuromodulation, and attention. Neuron, 46(4), 681-692.

Yu, A. J., Dayan, P., & Cohen, J. D. (2009). Dynamics of attentional selection under conflict: toward a rational Bayesian account. J Exp Psychol Hum Percept Perform, 35(3), 700-717.

Zalesky, A., Fornito, A., & Bullmore, E. T. (2010). Network-based statistic: identifying differences in brain networks. Neuroimage, 53(4), 1197-1207.

Zhang, D., & Raichle, M. E. (2010). Disease and the brain’s dark energy. Nat Rev Neurol, 6(1), 15-28.

Page | 156