Is the idle brain the devil ’ s workshop? Resting state connectivity in schizophrenia

Is the idle brain the devil’s workshop? Resting state connectivity in schizophrenia Judith M. Ford Daniel H. Mathalon Harshad Shanbhag Brian J. Roach FBIRN

Funding and Support • Veterans Administration • VA Merit Review Award Program • NIMH K02 • NIMH R01 • NCRR--FBIRN • NARSAD • UCSF

Is the idle mind busy? • When a mind is free to wander, spontaneous cognition tends to gravitate towards thoughts and feelings—it is constantly active. • Can you think about nothing? • Subjects were asked to think about nothing for 20-25 sec. • They were then asked if they had had any thoughts in that period. • 100% reported ongoing ideation (Flavell et al 2000). • Indeed, there has been biological evidence for a constantly active brain ever since electroencephalography (EEG) was discovered in 1929, by Hans Berger. • EEG did not cease even when the subject was at rest. • PET and fMRI data are consistent with EEG.

The Default Mode Network • A decade ago, a set of brain regions was identified that appeared to show greater activation at rest than during active, goal directed tasks (Raichle, 2001). • These networks have been labeled default mode networks (DMN), resting state networks (RSN), and task-negative networks (TNN). • They involve a set of regions that show a consistent pattern of deactivation during tasks and activation during rest (or stimulus-independent thought) (Raichle, 2001;Greicius, 2003;Fransson, 2005). • Thus, the rest condition identifies a well-defined and quantifiable neural network that is associated with a fundamental aspect of human nature and human experience – free and voluntary thought. • DMN allows a scientific approach for understanding the brain basis of free thought and how it is altered in psychiatric conditions.

Schizophrenia • Schizophrenia is a serious mental illness characterized by • positive symptoms (e.g., hallucinations, delusions), • negative symptoms (e.g., avolition and apathy), • disorganization (e.g., thought disorder), • cognitive deficits. • Except for cognitive deficits, which can be assessed with standard neuropsychological tests, the other symptoms are difficult to study. • Resting states may provide an ideal paradigm for observing where the mind naturally wanders in patients, and thus it may provide insight into the pathways that allow the troubling symptoms of the illness to emerge.

Why study resting state activity in schizophrenia? #1 • It eliminates the confound of task-related performance differences between healthy subjects and schizophrenia patients, who invariably perform even the simplest task with less speed and accuracy than healthy controls. • Some investigators attempt to find tasks that both groups can do perfectly, like the auditory oddball task. • Other groups will titrate task difficulty to minimize performance differences or will analyze only the trials for which performance was accurate. • Neither approach is perfect. • One solution is to adopt a protocol in which no task is used, such as rest. • Because task demands are minimal, the full range of patients can participate.

Why study resting state activity in schizophrenia? #2 • Neuroimaging studies have consistently found hyperactivation (i.e., reduced task suppression) of the DMN in patients with schizophrenia during a broad range of cognitive tasks • Auditory oddball task (Garrity et al. 2007), • Language tasks(Jeong & Kubicki 2010). • Working memory tasks(Meyer-Lindenberg et al. 2005, Pomarol-Clotet et al. 2008, Whitfield-Gabrieli et al. 2009), • When working memory demands are parametrically increased, healthy people exhibit greater suppression of the DMN during working memory tasks, • patients fail to exhibit this pattern (Meyer-Lindenberg et al. 2005, Pomarol-Clotet et al. 2008, Whitfield-Gabrieli et al. 2009). • Greater DMN suppression better cognitive performance in healthy people (Weissman et al 2006; Whitfield-Gabrieli et al 2009). • Failure to suppress DMN is consistent with poor performance on tasks.

Why study resting state activity in schizophrenia? #3 • Resting state can be a good way to study connectivity and dysconnectivity in schizophrenia. • Several lines of research suggest that schizophrenia is characterized by abnormalities in connections between spatially distributed networks (Shenton et al., 2001; Kubicki et al., 2002; Hubl et al., 2004). • Schizophreniahas therefore been described as a dysconnectivitydisorder (Peled, 1999). • Aberrant connectivity may underlie functional abnormalities observed in the disorder.

Why study resting state activity in schizophrenia? #4 • Northoff and Qin (2011) proposed a resting state hypothesis of auditory verbal hallucinations, suggesting that voices may be • “traced back to abnormally elevated resting state activity in auditory cortex itself, abnormal modulation of the auditory cortex by anterior cortical midline regions as part of the default mode network, and neural confusion between auditory cortical resting state changes and stimulus-induced activity.” • Perhaps the DMN is hyperconnected with auditory cortex in patients who hallucinate.

Auditory Verbal Hallucinations • Experienced by 75% of people with schizophrenia. • Experienced as voices in the absence of external sounds. • Voices commenting or conversing. • Associated with increased distress, morbidity, and mortality (suicide/homicide). • Often the chief presenting complaint. • Cardinal symptom of schizophrenia • ‘Voices’ are unbidden. • Non-self, but often recognizable speakers, family members or celebrities.

Methods • Subjects • 149 patients with schizophrenia • 163 age- and gender-matched healthy controls • Data were acquired from 7 FBIRN sites • Acquisition parameters • 3T magnets • TR/TE=2000/30ms • FOV=22 cm • Flip angle=70 degrees • bandwidth=100 kHz; matrix=64x64 • 24 slices (6mm thick, no gap) • axial-oblique plane, oriented parallel to AC-PC line • prescribed from the midsagittal slice of a SPGR anatomic sequence • Resulting voxel dimensions 3.75 x 3.75 x 6mm. • Task • Subjects were asked to lie quietly and keep their eyes focused on the word RELAX during the scan for 6 minutes.

Pre-Processing • 1. To identify and remove subjects with excessive motion • Calculate SFNR (signal-to-fluctuation-noise ratio) • SFNR =mean/ standard deviation of time series after drift has been removed (high motion=low SFNR) • Subjects with SFNR < 150 were dropped from this analysis • 21 controls • 38 patients • 2. INRIalign to extract motion parameters for use in CONN • 3. Slice time correction (SPM5) • 4. Despiking (AFNI) • 5. Removing noise • Define noise ROI from white matter and CSF • Extract 5 PCA parameters from noise ROI • Linear and quadratic terms • These 7 factors are regressed out of each voxel • 6. Images normalized to the MNI template (SPM5) • 7. Images smoothed using 8x8x8mm kernel (SPM5).

Extracting Functional Connectivity CONN toolbox (http://www.nitrc.org/projects/conn) . 12 motion parameters regressed out 6 motion parameters from INRIalign 6 temporal derivatives Bandpass filtered the data (0.004 - 0.08Hz). Voxel-wise correlation maps r-correlation between all voxels and a seed r-to-z transformed to create connectivity maps for each subject. Connectivity values were tested for significance in SPM5. Cluster level significance (p<.05, corrected) Site (7 FBIRN sites) was included in the models

Goals of this analysis DEFAULT MODE NETWORK. • Because medial prefrontal cortex (MPFC) has been suggested to be involved with “self-generated thoughts, intended speech, and emotions” (for review, see Gusnard, 2001), we focused our DMN analysis on connections with MPFC in the midline core of the DMN. • We also used a posterior cingulate/precuneus (PCC) seed, also dominant in the midline core of the DMN. LANGUAGE PROCESSING NETWORK. • Because of the dominance of hallucinations in schizophrenia, we used a seed in Wernicke’s area, an area known to be active during speech perception and the experience of auditory verbal hallucinations. SYMPTOM CORRELATIONS To determine if symptoms drive the effects, we used Positive, Negative, and Total symptoms as regressors in the connectivity analysis within the patient group.

Medial Pre-Frontal Cortex (MPFC) and Posterior Cingulate/Precuneus Seeds (PCC) Auditory Cortex, Wernicke’s and Thalamus Seeds

DMN with MPFC Seed Patients Controls Correlations Patients Controls Anti-correlations

Group Comparisons: DMN with MPFC Seed Patients>Controls Controls>Patients Caudate Putamen CBL Thalamus MTG PCC MFG n.s. P<.0001, cluster level corrected

MPFC connectivity with these regions is greater in patients with more severe psychotic symptoms (PANSS_POS) (p<.001, cluster corrected) Thalamus Caudate/Putamen Parahippocampal gyrus Hippocampus Globus pallidus Lentiform nucleus PCC To the extent that this cluster includes limbic areas, connectivity with MPFC may reflect the emotional tone to self-reflections. Caudate Ventral Lateral Nucleus Parahippocampal gyrus

DMN with PCC Seed Patients Controls Correlations Patients Controls Anti-correlations

Group Comparisons: DMN with PCC Seed Patients>Controls Controls>Patients MPFC Thalamus CBL Caudate Putamen n.s. P<.0001, cluster level corrected No symptom correlations

Auditory Mode Network with Wernicke’s Seed Patients Controls Images: p<.01,unc; k=100

Group Comparisons: Wernicke’s Seed Controls>Patients Patients>Controls Images: p<.01,unc; k=100 Images: p<.01,unc; k=100 All significant at p<.0001, cluster level corrected P<.05

Hoffman and colleagues showed similar connectivity when using a seed in Wernicke’s area (bilateral). Putamen and thalamus are involved in initiating and organizing language representations (Booth et al, 2007; Bhatnagar & Mandybur, 2005; Price, 2010). Putamen and thalamus are activated coincident with AVH occurrences in two reports (Shergill et al, 2000; Silbersweig et al, 1996). The connectivity with thalamus was not dependent on hallucinations. There was stronger connectivity between Wernicke’s and thalamus in patients who had a better response to rTMS treatment for hallucinations (Hoffman, personal communication.) Ford and Hoffman, 2012

Areas where connectivity with Wernicke’s is greater in patients with more severe hallucinations. Nothing is significant at a corrected level.

Thalamus Seed Controls>Patients Patients>Controls Images: p<.01,unc; k=100 Cluster corrected, p<.001 Thalamus has stronger connectivity within itself in controls compared to patients suggesting less variability among the voxels in the thalamus. Also controls had stronger connectivity between thalamus and superior frontal lobe, ACC, BA9. Thalamus has stronger connections with the rest of the brain in patients compared to controls.

Summary • DEFAULT MODE NETWORK CONNECTIVITY: • We predicted patients would show hyperconnectivity within the DMN. • Confirmed. Patients with schizophrenia have: • Hyperconnectivity between MPFC and PCC. • Hyperconnectivity between DMN and other regions -- thalamus, middle frontal gyrus, middle temporal gyrus and cerebellum. • There was no area that was more correlated in controls than in patients. • We predicted that DMN hyperconnectivity would be related to clinical severity. • Confirmed for MPFC but not for PCC seed. • Positive symptoms of schizophrenia were related to connectivity between the DMN and thalamus, caudate, putamen, PCC, parahippocampal gyrus, hippocampus. • LANGUAGE NETWORK CONNECTIVITY: • Patients had hyperconnectivity between the Wernicke’s area and thalamus. • Patients had hyperconnectivity between thalamus and most of the brain. • As far as we can tell, hyperconnectivity with the language network is not related to any symptoms of the illness.

Discussion/Speculations The impact of sensory input on self-organization of thalamo-cortical activity may be generally reduced in schizophrenia. As a result, processes underlying perception can become uncoupled from sensory input, particularly at times of hyperarousal, leading to the emergence of hallucinations (Behrendt 2003). Self-referential thought originating in medial pre-frontal cortex might arrive in limbic areas with greater ease in patients than in controls, via hyper-connections between these regions. Negative self-referential thoughts may contribute to paranoia and social isolation. The consequences of hyperconnectivity with thalamus in schizophrenia demands a more refined analysis of thalamus connectivity. Behrens et al, 2003, Nat Neurosciences

Discussion/Speculations Antoine Ritti (1844-1920) used the anatomical-functional discoveries of his teacher to explain that an automatic activity in the thalamus, by stimulating the cortex without reception of sensory information, created autonomous representations, perceived by the patient but not by his entourage, a process occurring spontaneously to some degree. Hence, Ritti seems the first author to introduce the concept of sensory deprivation and release of subcortical function into the pathophysiology of hallucinations. This innovative theory, which gave subcortical structures a role in high-level cognitive function, is very resonant today but was ignored for several decades after Ritti published his work.

Group Comparisons: BA41 Seed Patients>Controls Controls>Patients Thalamus Thalamus Visual cortex Cerebellum Visual cortex Thalamus Images: p<.05,unc; k=100 Images: p<.001,unc; k=100 n.s. All significant at p<.0001, cluster level corrected

Auditory Mode Network with BA41 Seed Patients Controls Images: p<.01,unc; k=100

MPFC connectivity with these regions is greater in patients with more severe psychotic symptoms (PANSS_POS) Yellow shows areas in the limbic region cluster (p<.001), perhaps reflecting an emotional tone to self-reflections. Red shows areas in the cerebellum (Crus 1 and 2; p=.028) involved in executive control, salience detection, and episodic memory/self-reflection (Habas et al., 2009).

General PANSS and BA41connectivity Patients with greater connectivity between BA41 and BA 6,8,9, 10 have higher scores on the General Psychopathology scale of the PANSS Somatic concern Anxiety Guilt feelings Tension Mannerisms and posturing Depression Motor retardation Uncooperativeness Unusual thought content Disorientation Poor attention Lack of judgment and insight Disturbance of volition Poor impulse control Preoccupation Active social avoidance BA 6,8,9,10 P<.027, cluster level corrected

Structure • Van Dijk et al (2009) noted that functional connectivity suggests anatomic connectivity. • Studies combining resting state functional connectivity data with structural connectivity data have shown strong correlations between both adjacent and more distal regions within the brain [Greicius, 2009]. • Thus, abnormal connectivity in certain clinical populations could reflect underlying neuroanatomical and even neurovascular abnormalities. • Indeed, patent pathways may be a necessary, but not a sufficient, condition for functional connectivity. • A third region can connect two disconnected regions, resulting in apparent functional connectivity [Damoiseaux, 2009].

Results:Within Group Positive Connectivity with MPFC Figure 1.This image was constructed by (1) making a correlation map (p<.05, FDR corrected) for positive correlations for each group, (2) saving the maps as masks, (3) overlaying the 2 masks. Controls=yellow, Patients=red The overall “orange” impression shows that MPFC time series is highly positively correlated with itself and other DMN regions (PCC, IPL), in both controls and patients. Red indicates that patients have more areas positively correlating with MPFC than controls, and this is especially obvious in the temporal lobe, the focus of our analysis.

Connectivity and Auditory Hallucination Severity (substantianigra seed).Positive connectivity between substantianigra and all these regions is greater in patients with more severe hallucinations (SAPS-hall) p=.005, k=5 MTG PHC MPFC MTG PHC IPL PCC

Areas showing greater connectivity between substantia nigra in patients with more severe AVH Bilat PHC,HC, MTG, temp pole MPFC: BA10,11,9,32 RIGHT: MTG, ITG, STG, fusiform PCC cerebellum

Results:Within Group Positive Connectivity with substantianigra Figure 1.This image was constructed by (1) making a correlation map (p<.05, FDR corrected) for positive correlations for each group, (2) saving the maps as masks, (3) overlaying the 2 masks. Controls=yellow, Patients=red Red indicates that patients have more areas positively correlating with substantia nigra than controls, and this is especially obvious in the temporal lobe, the focus of our analysis.

To understand the pathophysiology of brain-based diseases, clinical neuroscientists use neuroimaging methods: • electroencephalography (EEG) • positron emission tomography (PET) • functional magnetic resonance imaging (fMRI) • FMRI studies typically involve contrasting task activation to • a baseline task or • a “rest” condition. • Rest was popular because we believed that the resting brain was quiescent. We now know that some brain regions are more active during rest than during a wide range of attention-demanding tasks. • These brain regions are considered to be the brain’s default mode network (DMN) (Raichle et al 2001). • The DMN offers a new window into the neural basis of spontaneous human thoughts and feelings, and correspondingly, how this differs in neuropsychiatric diseases.

Instantaneous phase was calculated from that analytic signal, and is plotted in radians (Figure 14B, blue line). The rectified radian values (Figure 14B, red line) are plotted to illustrate the location of peaks (rectified values near 0 radians) and troughs (rectified values near π radians). The component that most resembled a DMN activation map was selected, and the ICA time course (smoothed and unsmoothed) was extracted for the passive listening run (Figure 14A). How does the brain respond to external stimuli when it is in idle? In Figure 14D, we show the auditory ERP from one patient who listening passively. In this experiment, subjects were presented with tones in one run and checkerboards in another, presented randomly every .5-2s. Figure 14C shows axial images depicting a DMN component of BOLD data from all 4 subjects in this analysis. The red and blue lines in Figure 14D result from sorting the single trials according to whether the tone was presented during a peak or a trough of the DMN time series.

What is the basis of the correlations we see? • Functional connectivity reflects structural connectivity in the DMN. • (a) Task-free functional connectivity in the DMN is shown in a group of 6 subjects. • (b) DTI fiber tractography demonstrates the cingulum bundle (blue) connecting the PCC to the MPFC. Orange tracts connect bilateral medial temporal lobe to PCC. • (c) Although not easily detected with DTI, tracts also connect MPFC to medial temporal lobe (Barbas et al, 1999). Greicius et al, 2009

Components of the DMN • The DMN has emerged as its own field of research [Buckner, 2008], and it is rapidly evolving. • Nevertheless, the function of various regions in the DMN is not fully understood. • There appears to be a cortical midline structure (CMS) that is active when people make self-relevant, affective decisions (Northoff and Bermpohl, 2006). • It primarily involves the posterior cingulate (PCC) and medial pre-frontal cortex (MPFC). • PCC itself has been suggested to be involved with theevaluation of information gathered in internal and external environments [Raichle, 2001] and the retrieval of episodic memories [Greicius, 2003]. • MPFC has been suggested to be involved with “self-generated thoughts, intended speech, and emotions” (for review, see [Gusnard, 2001]).

Modern neuroimaging has revealed, unexpectedly, that a specific network of brain regions is active during rest periods without tasks. Stimulus-dependent or task-dependent conditions used in fMRI studies are designed to provoke a common mental operation across people (e.g., perception of a face or recognition of a prior experience) so as to discover the brain basis of that mental operation. One might imagine that if people were engaged in unconstrained stimulus-independent thought during a “rest” period without any task and with the mind free to wander, then thoughts would be extremely variable as different people thought about different things or one person thought about different things from moment to moment. Correspondingly, with such unconstrained variation in what is being thought about, one might expect a great diversity in what brain regions are engaged (i.e., where activations are measured). Surprisingly, research has shown that such stimulus-independent thought during rest recruits a remarkably consistent neural network – the DMN (Raichle et al 2001; Shulman et al 2007). Thus, the rest condition identifies a well-defined and quantifiable neural network that is associated with a fundamental aspect of human nature and human experience – free and voluntary thought. Study of the DMN allows, therefore, a neuroscientific approach towards understanding the brain basis of free thought, and also the opportunity to study how that brain basis of free thought is altered in psychiatric conditions.

Correlated MPFC activity Auditory cortex activity • A few words about these signals: • They are hemodynamic, not neural. • They are only indirect reflections of underlying neural activity • They are very slow, taking 6-10 seconds, or longer, to resolve. • Neural activity resolves on a millisecond time scale. • They may indirectly reflect neural activity in the gamma range (30-100 Hz). • There is a lot of information that could/can be extracted from these signals: • phase information about which signal is leading • frequency information about the dominant frequency in different areas and in the different groups • information about how frequency changes across time Anti-Correlated MPFC activity Auditory cortex activity Un-Correlated MPFC activity Auditory cortex activity

Results:Within Group Positive Connectivity with MPFC with temporal lobe mask This image was constructed by (1) making a correlation map (p<.05, FDR corrected) for positive correlations for each group, (2) saving the maps as masks, (3) overlaying the 2 masks. Controls=yellow, Patients=red

Group Differences in Positive Connectivity with MPFC with temporal lobe maskBetween-group differences in positive connectivity between MPFC and temporal lobe (SZ>HC: red; HC>SZ: yellow). Red indicates clusters of voxels where patients have more positive correlations between MPFC and temporal lobe than the controls (shown in yellow). The significant cluster is indicated by white arrows. This cluster includes 422 voxels in the left temporal lobe (MTG=288, ITG=19, STG=17). Calculations using AlphaSim indicate that for this temporal lobe ROI, we needed 305 voxels to achieve a p<.05 level of significance (corrected for multiple comparisons using 3dClustSim).

Functional connectivity • While initial observations of the DMN were based on noticing “joint” activation of the cortical midline structures, subsequent investigations have focused on functional connectivity. • If activities in different regions are temporally correlated, these regions are considered to be functionally connected (for review, see Fox, 2007) and part of a network.

Is the idle brain the devil ’ s workshop? Resting state connectivity in schizophrenia