Mapping brain response variability in schizophrenia

Mapping brain response variability in schizophrenia Dissertation zur Erlangung des akademischen Grades des Doktors der Naturwissenschaften an der Un...
Author: Herbert Dixon
0 downloads 0 Views 5MB Size
Mapping brain response variability in schizophrenia

Dissertation

zur Erlangung des akademischen Grades des Doktors der Naturwissenschaften an der Universität Konstanz Fachbereich Psychologie

Vorgelegt von Todor Iordanov

Tag der mündlichen Prüfung: der 11. Januar 2012 Referent/in: Prof. Dr. Thomas Elbert Referent/in: Prof. Dr. Brigitte Rockstroh Referent/in: Prof. Dr. Harald Schupp

Konstanzer Online-Publikations-System (KOPS) URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-201838

1

2

Table of Contents Abbreviations .................................................................................................................................................... 5 Own research contributions........................................................................................................................ 8 Summary .......................................................................................................................................................... 10 Zusammenfassung ....................................................................................................................................... 12 1. General Introduction .............................................................................................................................. 15 2. Local Heschl’s Gyrus-based coordinate system for intersubject comparison of M50 auditory response modeled by single equivalent current dipole .............................................. 22 Abstract ....................................................................................................................................................... 22 Introduction ............................................................................................................................................... 22 Materials and methods .......................................................................................................................... 24 Subjects ................................................................................................................................................... 24 Data acquisition ................................................................................................................................... 25 Data analysis ......................................................................................................................................... 25 Coordinate transformations ........................................................................................................... 26 Results .......................................................................................................................................................... 31 Discussion ................................................................................................................................................... 34 3. Reduced mismatch negativity and increased variability of brain activity in schizophrenia ................................................................................................................................................. 36 Abstract ....................................................................................................................................................... 36 Introduction ............................................................................................................................................... 36 Methods ....................................................................................................................................................... 38 Subjects ................................................................................................................................................... 38 Stimuli and design .............................................................................................................................. 40 Data acquisition and analysis ......................................................................................................... 40 Results .......................................................................................................................................................... 43 Discussion ................................................................................................................................................... 49 4. Dipole Parameter Estimation of M50 Auditory Evoked Fields Applied to the Study of Training-Induced Neuroplasticity in Schizophrenia ...................................................................... 55 Abstract ....................................................................................................................................................... 55 Introduction ............................................................................................................................................... 56 Methods ....................................................................................................................................................... 56 Subjects ................................................................................................................................................... 56 Data acquisition ................................................................................................................................... 58 Paradigm ................................................................................................................................................ 58 Cognitive Training .............................................................................................................................. 58 Data Pre-Processing ........................................................................................................................... 58 Dipole Fits .............................................................................................................................................. 58 3

Gating Ratios ......................................................................................................................................... 58 Asymmetry ............................................................................................................................................ 59 Orientations .......................................................................................................................................... 59 Statistics.................................................................................................................................................. 59 Results .......................................................................................................................................................... 59 Discussion ................................................................................................................................................... 63 5. The Effects of Specific Cognitive Training on the Parameters of the Evoked Responses in Schizophrenia Measured with MEG ................................................................................................. 65 Abstract ....................................................................................................................................................... 65 Introduction ............................................................................................................................................... 66 Methods ....................................................................................................................................................... 67 Participants ........................................................................................................................................... 67 Study Design, Cognitive Assessments, and Training Protocols......................................... 69 MEG Measurement ............................................................................................................................. 70 Data Analysis ........................................................................................................................................ 71 Results .......................................................................................................................................................... 74 Discussion ................................................................................................................................................... 81 6. General Discussion .................................................................................................................................. 84 Appendix A: Development and implementation of user defined MRI rigid-body transformations (with FSL) ...................................................................................................................... 94 Appendix B: Dipole Parameter Estimation of M50 Auditory Evoked Fields Using Three Different Head Models ............................................................................................................................. 115 Abstract .................................................................................................................................................... 115 Introduction ............................................................................................................................................ 115 Methods .................................................................................................................................................... 115 Results ....................................................................................................................................................... 118 Discussion ................................................................................................................................................ 122 Appendix C: The advantage of using coordinate system based on inner brain anatomical landmarks ..................................................................................................................................................... 123 References .................................................................................................................................................... 126

4

Abbreviations AC

Anterior Commissure

ANOVA

ANalysis Of VAriance

BDI

Beck Depression Inventory

BEM

Boundary Element Model

BPRS

Brief Psychiatric Rating Scale

CE

Cognitive Exercise package (PositScience, San Francisco, USA)

CP

Cogpack training package (Marker Software)

CPZ

Chlorpromazine Equivalents

CV

Circular Variance

DD

Duration Deviant

EEG

Electroencephalography

ERP

Event-Related Brain Potential

FD

Frequency Deviant

fMRI

Functional Magnetic Resonance Imaging

FNIRT

FMRIB’s Nonlinear Image Registration Tool

FSL

FMRIB Software Library

fT

Femto Tesla

GAF

Global Assessment of Functioning Scale

HC

Healthy Controls

HG

Heschl’s Gyrus

HGT

HG Transformation

ICD

International Classification of Diseases

ITPL

Inter-Trial Phase-Locking

LPA

Left Pre-Auricular Point

M100

The magnetic counterpart of N100

M50

The magnetic counterpart of P50 5

M±SD

Mean ± Standard Deviation

MEG

Magnetoencephalography

MMN

Mismatch Negativity

MMNm

Magnetic Counterpart of MMN

MNI

Montreal Neurological Institute

MRI

Magnetic Resonance Imaging

ms

Milliseconds

N100

A wave with a negative peak at approximately 100 ms following stimulus onset

NIfTI

Neuroimaging Informatics Technology Initiative

P300

A wave with a positive peak at approximately 300 ms following stimulus onset

P50

A wave with a positive peak at approximately 50 ms following stimulus onset

PAN

Pre-Auricular Points and Nasion

PC

Posterior Commissure

PT

Planum Temporale

RMANOVA

Repeated Measures ANalysis Of VAriance

RPA

Right Pre-Auricular Point

S1

Dipole moment of the click one response in a double click paradigm

S2

Dipole moment of the click two response in a double click paradigm

SD

Standard Deviation

SGR

Sensory Gating Ratio

SNR

Signal-to-Noise Ratio

SQUID

Superconducting Quantum Interference Device

SZ

Schizophrenia Patients

VI

Variability Index

VP(n)

Versuchsperson(en) 6

7

Own research contributions

The realization of the studies for my thesis was only possible by means of cooperation with a number of colleagues. My own research contributions to the four studies are listed below. Study 1: Local Heschl’s Gyrus-based coordinate system for intersubject comparison of M50 auditory response modeled by single equivalent current dipole Authors: Todor Jordanov, Tzvetan Popov, Christian Wienbruch, Thomas Elbert and Brigitte Rockstroh Published in Journal of Neuroscience Methods Own Contributions: Literature and scientific background research, development of the new coordinate system and applying it for all data analysis, writing a detailed description of methods, results and their interpretation. Study 2: Reduced mismatch negativity and increased variability of brain activity in schizophrenia Authors: Todor Jordanov, Tzvetan Popov, Nathan Weisz, Thomas Elbert , Isabella PaulJordanov and Brigitte Rockstroh Published in Clinical Neurophysiology Own Contributions: Literature and scientific background research, hypothesis development and all data analysis, writing a detailed description of methods, results and their interpretation. Study 3: Dipole Parameter Estimation of M50 Auditory Evoked Fields Applied to the Study of Training-Induced Neuroplasticity in Schizophrenia Authors: Todor Jordanov, Tzvetan Popov, Christian Wienbruch, Thomas Elbert and Brigitte Rockstroh

8

Presented as a poster on the 49th annual meeting of the Society for Psychophysiological Research (SPR) in Berlin, Germany Own Contributions: Literature and scientific background research, hypothesis development and data analysis, writing of the manuscript. Study 4: The Effects of Specific Cognitive Training on the Parameters of the Evoked Responses in Schizophrenia Measured with MEG Authors: Todor Jordanov, Tzvetan Popov, Christian Wienbruch, Thomas Elbert and Brigitte Rockstroh Not published yet Own Contributions: Literature and scientific background research, hypothesis development and data analysis, writing of the manuscript.

9

Summary

It is known that the brain response to auditory stimuli of schizophrenia patients is more variable than that in healthy controls. The goal of this thesis was to investigate the role of this increased variability in the formation of evoked potentials and localizing it in the brain. It was also investigated if the increased brain response variability in schizophrenia could be normalized by a computer based auditory training program. This thesis consists of four studies, which are based on three magneto-encephalographic (MEG) recordings of schizophrenia patients and healthy controls: a) All subjects participated in a resting state measurement without any stimuli representation. Then two experiments were performed: b) The double click paradigm was used where 100 pairs of 3ms square-wave clicks were presented with a 500ms offset-to-onset interval and an 8 s jittered inter-trial interval (offset to onset) for assessment of sensory gating in the M50 brain response component and c) An odd-ball paradigm was used for assessment of the mismatch negativity (MMN) response to rare stimuli in the sequence of standard (frequent) stimuli. Study 1: The M50 brain response localization in the paired-click paradigm was compared between schizophrenia patients and healthy controls. Since it is known that the brain structural variability in schizophrenia patients is greater than that in healthy controls a new coordinate system based on the individual form and position of the primary auditory cortex was defined and used for the localization of the M50 origin and the comparison between the groups. It was shown that the localization of the M50 component was further anterior in schizophrenia patients than in healthy controls in the left hemisphere. This finding was not the result of structural differences between schizophrenia patients and healthy controls. It implies that localization differences between groups are the result of functional processing differences. Study 2: The connection between the MMN amplitude and inter-trial response variability was investigated. It was hypothesized that the reduced averaged MMN amplitude in schizophrenia patients is due to an inconsistent brain response to the same stimulus across trials. Results suggested that amplitude variation across trials does not influence the averaged MMN amplitude in patients but that an inter-trial phase locking impairment in schizophrenia could contribute to the reduced MMN. This means that an impaired auditory sensory memory in schizophrenia is independent from variation in 10

response strength of associated cortical networks, but might be related to an increased temporal response variability of these networks. Study 3: The connection between dipole parameters,

symptoms and cognitive

performance in schizophrenia was investigated. It was shown that not only the amplitude of the reconstructed brain activity correlates with symptoms and cognitive performance but also dipole orientation, peak latency and signal-to-noise ratio. Dipole localizations were not correlated with any symptoms or cognitive scores. This shows that schizophrenia symptoms and associated cognitive deficits are reflected in various cortical parameters. Consequently, these parameters should be considered when investigating auditory brain response impairment in schizophrenia. Study 4: In line with study 2 it was shown that the averaged amplitude of the M50 response strongly depends on the response variability across trials. Therefore, it was of interest, whether two different computer based training programs for schizophrenia would be able to lessen cortical response variability. No improvements of response variability could be determined after the training period for either training. However, it was shown that in those patients who had a lower sensory gating ratio (SGR) after the CE training, the SGR also correlated with inter-trial phase locking values. This finding suggests that the brain of the schizophrenia patients might have used one impairment (poor phase locking) in order to improve another auditory impairment (impaired sensory gating). This might be considered as evidence for neuroplasticity induced by the CE training. In sum, one might conclude that schizophrenia is not only characterized by poor sensory gating, deficient information processing and auditory sensory memory but also by instable and inconsistent attractor networks which cause a more variable response (in the sense of brain signal amplitude and latency) to auditory stimuli and consequently impaired cognitive performance. This increased variability was not influenced by computer programs designed to lessen schizophrenia symptoms. Thus, cortical response instability and variability should be further investigated and taken into account when comparing cognitive scores between schizophrenia patients and healthy controls.

11

Zusammenfassung Das Ziel dieser Dissertationsschrift war zu untersuchen, in wieweit die Gehirnreaktion auf auditorischen Reize bei Schizophrenen „variabler“ als bei den gesunden Personen ist und ob diese erhöhte Variabilität sich durch ein computer-basiertes Training verbessern lässt. Die vorliegende Dissertationsschrift setzt sich aus vier Studien zusammen, die auf 3 magnetoenzephalographische (MEG) Aufnahmen von schizophrenen Patienten und gesunden Versuchspersonen (VPn) beruhen. a) Alle VPn wurden im Ruhezustand (ohne irgendwelche Aufgabestellung) gemessen. b) Im Double-Click Paradigma wurden zwei gleiche 3 ms lange auditorische Stimuli schnell hintereinander (500 ms inter-stimulus Interval) präsentiert. Ziel war, die Fähigkeit von Personen mit und ohne Schizophrenie zu untersuchen, die Gehirnreaktion bei wiederholter Stimuluspresäntation zu unterdrücken. Beim mismatch-negativity (MMN) Paradigma wurde eine lange Reihe gleicher Stimuli (standard Stimuli) präsentiert, die in zufälligen Zeitintervallen durch andere, vom standard Stimulus abweichende Stimuli, unterbrochen wurde. Ziel war, die Fähigkeit des Gehirns von Personen mit und ohne Schizophrenie zu untersuchen, Änderungen in der akustischen Umwelt zu erfassen und bei abweichenden Stimuli Ressourcen zur Verfügung stellen zu können. Studie 1: Es wurde die Quelllokalisationen der M50 Komponente (die Reaktion des Gehirns auf sensorische Reize ca. 50 ms nach der Reizpräsentation im Double-ClickParadigma) untersucht.

Da strukturelle Unterschiede in der Asymmetrie der

Temporalkortizes zwischen Personen mit und ohne Schizophrenie bekannt sind, wurden

anhand

Koordinatensysteme

von

individuellen

basierend

auf

der

strukturellen individuellen

Gehirnbilder Struktur

individuelle

des

primären

auditorischen Kortex konstruiert und für die Quellokalisation verwendet. Es wurde gezeigt, dass die Lokalisation der M50-Komponente in der linken Hemisphäre bei schizophrenen Patienten weiter anterior liegt als bei Kontrollpersonen. Dieses Ergebnis war unabhängig von strukturellen Kortexunterschieden zwischen Patienten und gesunden Versuchspersonen. Dies deutet darauf hin, dass die Lokalisationsunterschiede zwischen den Gruppen auf funktionelle Verarbeitungsunterschiede zurückzuführen sind. Studie 2: Es wurde untersucht, ob die reduzierte Amplitude der MMN bei schizophrenen Patienten tatsächlich eine geringere Aktivierung des auditorischen Kortex widerspiegelt 12

oder ob die erhöhte Variabilität beim Auftreten der MMN relativ zum Stimulus dafür verantwortlich sein kann. Es wurde gezeigt, dass die MMN Amplitude bei Schizophrenie unabhängig von einer erhöhten Amlitudenvariabilität ist. Allerdings liegt nahe dass eine schlechtere Phasensynchronisierung des MMN Peaks wahrscheinlich eine große Rolle für eine reduzierte MMN spielt. Dies könnte bedeuten, dass das sensorische Gedächtnis bei Schizophrenie unabhängig von der erhöhten Variabilität in der Reaktionsstärke verantwortlicher kortikaler Netzwerke ist, nicht jedoch von der erhöhten zeitlichen Variabilität der Reaktion dieser Netzwerke. Studie 3: Es wurde untersucht, welche Dipolparameter neben der M50 Amplitude eine Bedeutung für die Symptomatik und kognitive Leistung bei Schizoprenie haben. Außer der Amplitude wurden auch die Lokalisation, Orientierung, Latenz und das Signal-zuRausch-Verhältnis untersucht. Es wurde gezeigt, dass alle Parameter außer der Quelllokalisation im Zusammenhang mit verschiedenen Krankheitsmerkmalen stehen. Dies zeigt, dass sich Symptome und kognitive Defizite bei Schizophrenie auch in diversen kortikalen Parametern wiederspiegeln. Folglich sollten nicht nur die Amplituden der Gehirnreaktion sondern auch alle anderen verfügbare Informationen über die P50-Quellen mitberücksichtig werden, wenn Schizophrenie untersucht wird. Studie 4: Es wurde zunächst gezeigt, dass die M50 Amplitude stark von der zeitlichen Variabilität der Reaktionsamplituden abhängt. Weiter wurde untersucht, ob die Verbesserung des Sensory Gating Ratios (SGR) nach dem Training mit

zwei

Computerprogrammen mit einer Verbesserung der M50 Phasensynchronisierung zusammenzuhängt. Es wurde gezeigt, dass über alle Versuchspersonen hinweg keine Verbesserung der Phasensynchronisierung erzielt werden konnte. Dies galt für beide Trainingsprogramme. Allerdings zeigte sich, dass nach dem CE-Training bei den VPn, die nach dem Training ein verbessertes SGR hatten, das SGR auch mit der Phasensynchronisierung korrelierte. Dieses Ergebnis könnte darauf hindeuten, dass das Gehirn der schizophrenen Patienten ein Defizit (schlechtere Phasensynchronisierung) nutze, um ein anderes Defizit (schlechtes SGR) zu verbessern. Dies könnte auf Neuroplastizität durch das CE Training hinweisen. Als Endergebnis kann zusammengefasst werden, dass Schizophrenie sich nicht nur durch

eine

schlechtere

Unterdrückung

der

Gehirnreaktion

bei

wiederholter

Stimuluspräsentation und durch die beeinträchtigte Fähigkeit des Gehirns, Änderungen in der akustischen Umwelt zu erfassen auszeichnet, sondern auch durch erhöhte 13

zeitliche Variabilität der Gehirnraktionen und einer geminderten stimulus-bezogenen Phasensinchronisierung. Die zuletzt genannten Merkmale wurden weder durch ein auditives, noch durch ein kognitives Training beeinflusst. Erhöhte zeitliche Variabilität und reduzierte Phasensynchronisierung eröffnen eine neue Perspektive für das Verständnis auditorischer Verarbeitung bei Schizophrenie und sollten in Zukunft vertiefter

untersucht

werden,

bzw.

bei

der

Analyse

evozierter

Reaktionen

mitberücksichtigt werden.

14

1. General Introduction According to the current version of the international classification of diseases (ICD-10, Version 2007) schizophrenic disorders are characterised by „distortions of thinking and perception, and affects that are inappropriate or blunted. Clear consciousness and intellectual capacity are usually maintained although certain cognitive deficits may evolve over the course of time. The most important psychopathological phenomena include thought echo; though insertion or withdrawal; thought broadcasting; delusional perception and delusions of control; influence or passivity; hallucinatory voices or discussing the patient in the third person; thought disorders and negative symptoms.“ There are three main clusters of symptoms in schizophrenia: cognitive, negative and positive. Cognitive symptoms include poor attention, working-memory deficits and a deficit in planning and completing tasks (Liddle, 1987; Green, 1996; Mueser and McGurk, 2004; Rolls et al., 2008). Negative symptoms include apathy, lack of emotion, motor retardation, lack of motivation, blunted affect and passive behavior (Liddle, 1987; Mueser and McGurk, 2004). Positive symptoms of schizophrenia include disordered thoughts and speech, hallucinations and delusions (Liddle, 1987; Mueser and McGurk, 2004). Schizophrenic disorders are heterogeneous and can thus be sub-divided into different forms of schizophrenia, paranoid schizophrenia being the most common diagnosis in Europe (Tateyama, M. et al., 1999). Paranoid schizophrenia is dominated by relatively stable, often paranoid delusions, usually accompanied by hallucinations, particularly auditory, and perceptual disturbances. Disturbances of affect, volition, speech and catatonic symptoms are either absent or relatively inconspicuous. The prevalence of schizophrenic disorders is estimated to be 1% (Jablensky , 1997). The first symptoms of schizophrenia typically appear between the ages of 16 and 30 years, and infrequently after the age of 45 years (Almeida et al., 1995). Usually the symptoms accompany schizophrenia patients their entire life, although many patients experience remissions of symptoms later in life (Häfner et al., 2003b). Genetic and environmental factors are known to play a role in the aetiology of schizophrenia. The probability of developing schizophrenia among relatives of patients is higher than in people without any relatives having the disorder (McGuffin et al., 15

1995). The risk of developing schizophrenia is especially high for people with schizophrenic parents and for those having a monozygotic twin with schizophrenia (Cardno et al., 1999). Besides the genetic factors mentioned above some environmental factors might also increase the risk of developing schizophrenia. Prenatal events like maternal influenza, rubella, malnutrition, diabetes mellitus, and smoking during pregnancy have been shown to have some influence on the frequency of occurrence of the disorder (Susser and Lin, 1992; Takei et al. 1996; Thomas et al., 2001). Regarding the neuropathology two main facts characterize schizophrenia: ventricular enlargement (Wright et al., 2000) and decreased cortical volume (Andreasen et al., 1994; Powchik et al., 1998). Brain structures like the frontal lobes, amygdala, hippocampus, parahippocampus, thalamus and medial temporal lobe, cingulate gyrus, and superior temporal gyrus have decreased volumes in patients with schizophrenia compared with controls (Wright et al., 2000; Byne et al., 2002; Lawrie and Abukmeil, 1998, Narr et al., 2000). Schizophrenia is characterised by alterations in brain structure and functionality. It has been shown that certain brain regions of patients with schizophrenia tend to be less asymmetric compared to control participants (Petty, 1999). For example, Planum temporale (PT) asymmetry reversal has been reported in schizophrenia patients (Rossi et al., 1992, 1994; Petty et al., 1995; Barta et al., 1997; Hirayasu et al., 2000; Kasai et al., 2003). In addition to structural alterations, schizophrenic patients have been shown to have difficulties with a wide range of cognitive tasks: short-term and long-term memory (Aleman et al., 1999), maintaining attention and initiating action (Cornblatt and Kellp, 1994), or decision making (Shurman et al., 2005). However, it is still not clear, whether functional abnormalities are the consequence or the cause of structural abnormalities. A lot of electroencephalographic (EEG) and magnetoencephalographic (MEG) research has focused on investigating auditory mechanisms and their disturbance in schizophrenia. The three most popular and most stable functional measures of an auditory perception impairment in schizophrenia are the mismatch negativity (MMN) (Näätänen, 1978), the auditory sensory gating ratio (SGR) derived from the P50 doubleclick paradigm (Adler and colleges, 1982) and brain response variability (also called “cortical noise” (Winterer and colleagues, 2000) or variability index (VI)). A reduced MMN is believed to be evidence for deficits in automatic auditory sensory processing, 16

pre-attentive auditory perception and discrimination (Pekkonen et al., 2002; Näätänen and Kähkönen, 2009; Magno et al., 2008; Turetsky and Moberg, 2009), or decreased recruitment of brain areas during task performance (Bates et al., 2009). Impaired sensory gating (i.e. a high SGR) is generally considered to be a neurophysiological marker of an attention impairment (Thoma et al., 2004). Clinical studies of the M50 component (the MEG equivalence to P50) have shown that sensory gating predicts performance on neuropsychological tests of attention, working memory, general memory, and executive function in schizophrenia. Right-hemispheric SGR is correlated with severity of negative symptoms in schizophrenia (Thoma et al., 2005), whereas lefthemispheric SGR is related to positive symptoms (Irwin et al., 2003). An increased VI is related to an increased amount of slow wave activity, impairment of phase locking during stimulus processing and deficient bilateral temporal lobe coherence in schizophrenia patients (Winterer et al., 1999; 2000; 2004). It is found to negatively correlate with working memory scores (Winterer et al., 2004) and reaction time (Winterer et al., 2000). Various experimental and computational observations of cortical actions of dopamine have supported the notion that diminished mesocortical dopamine signaling and the resulting increased response variability of prefrontal neurons might be involved in cognitive and behavioral deficits in schizophrenia (Weinberger, 1987; Weinberger et al., 1988; Goldman-Rakic, 1994; Goldman-Rakic et al., 2000). The treatment of schizophrenia can be divided into pharmacological and psychosocial treatment. Antipsychotics are mainly used for the reduction of psychotic symptoms and prevention of relapses (Kane and Marder, 1993) though having more modest effects on negative symptoms and cognitive impairment (Greden and Tandon, 1991). Psychosocial intervention aims to enhance functioning in areas such as independent living, relationships, and work. Recently, new alternative strategies for treatment of schizophrenia have been developed. With the growth of the computer industry, computer programs for training the perception in schizophrenia have become available. For example the Cognitive Exercise (CE) package (PositScience, San Francisco, USA) aims to enhance discrimination ability in the auditory system. It is an adaptive training involving different auditory tasks. One task involves listening to pairs of two frequency-modulated tones that can either sweep upwards or downwards. It is then required to reproduce the order of the two sound-sweeps (up-up, up-down, down-up, down-down). Another task 17

involves listening to syllable pairs and judging, which syllable was played first. Other tasks involve identifying arrays of open and closed syllables in spatial and sequential context, discriminating tone frequencies, and remembering details of a short narrative. The training is adaptive in that it varies task difficulty with the performance of the patient. Increase of difficulty level is achieved by shortening the frequency-modulation period of the tones or the formant transition period of the syllables. The inter-stimulusinterval between the tones and the syllables also becomes shortened with increasing task performance. If the patient makes a mistake, the tasks become easier accordingly. The CE training has been shown to improve verbal memory (Fisher et al., 2009a; 2009b) and to normalize the M100 “attenuation’’ that represents the normal suppression of neural activity associated with second syllable presentation due to ongoing first syllable processing (Adcock et al., 2009). Another training which is used in schizophrenia treatment (despite not exclusively aiming at schizophrenia as the sole disorder) is the Cogpack (CP) training package (Marker Software). In contrast to the CE training, it not only targets the auditory modality, but involves 64 exercises of visuomotor skills (manipulate, follow or mark a moving figure with the computer mouse; divide lines or pies; reproduce or mirror a drawing; catch a bouncing ball), vigilance (scanning, catch falling star, continuous performance), comprehension (character recognition), language (word finding upon clues, text-content, authors or titles of quoted poems, place words or syllables in order, anagrams), memory (words, images and labels, patterns, signs, addresses, routes, life scenes), logic (mental arithmetic, geometry, numbers and number words, quantities, simple and deductive comparisons, block or series completion), everyday skills (times and dates, compass, geography, money, weights and measures, road signs, license plates, abbreviations; Olbrich, 1998; Geibel-Jakobs et al., 1998; Sartory et al., 2006). This new way of treating schizophrenia patients may also be considered as another sub-group of the psychosocial treatment methods. This is because computer trainings are often used in combination with role playing and discussion of the program results with therapists. It is likely that a computer-based training alone is not enough to cure schizophrenic symptoms. Rather, only a combination of different treatment methods will result in substantial improvement in schizophrenia patients. However, in order to get an impression of the effectiveness of a computer training it needs to be evaluated when applied separately. 18

There are already several studies reporting computer-training effects using clinical and behavioural measures as dependent variables (Twamley et al. 2003) but there are almost no studies investigating the effects of computer trainings using functional brain imaging techniques like MEG or EEG. In order to investigate, whether a training or therapy has an effect on brain activity, several things need to be considered. For example, a training-program can change the function of a cortical structure, i.e. it can improve its activity, without changing the morphological characteristics of that structure. Alternatively it might cause brain structural changes and consequently also changes in the functionality of these brain structures. Thus, in order to clarify which effect a training program has on the brain, both the structure and the function of a cortical area need to be considered. Another challenge for uncovering training effects on cortical measures is noise in MEG/EEG data. This can be caused by various sources. In the case of MEG, participants rest their head in the MEG-sensor, where there is the chance that the position of the head changes slightly throughout the experiment. Additionally every test person lies in a slightly different way in the MEG sensor compared to another subject. The different position of the head in the MEG-sensor can be determined and corrected with the help of the digitized fiducial points and the digitized head shape. Nevertheless, the variability caused by head movement during the measurement is not possible to be corrected because there is no possibility to detect every movement of the test subject during the measurement. There are also other sources of variability during an MEG measurement – the influence of external non-physiological magnetic sources like the Earth’s magnetic field, magnetic fields caused by electric devices, train vibrations, car engine disturbance etc. Other physiological but not brain-related magnetic fields like eye-blinks, muscle contractions, heart beat etc. can contribute to greater signal variability in the MEG data as well. Even after elimination of the variability sources not originating from the brain there still remains a great amount of variability due to inter-individual differences in the brain structure and functionality. The brain of every individual person is unique. This is an obstacle for uncovering potential training effects, as the same cortical area might be located at slightly different regions, might be oriented in a different way, or might differ in size between different people. Apart from inter-personal variability, there is also variability within one person. For example, the brain does not always respond in exactly 19

the same way at exactly the same time to identical stimuli. The cortical reaction might vary in magnitude and may also jitter in time after a stimulus. Patterson and colleges (2000) manifested that the schizophrenia patients have significantly more intraindividual temporal variability than normal controls in the P50 peak which contributes to the SGR differences between both groups. These could imply that the SGR depends on the VI which is measure of the signal variability (both – latency and magnitude variability). It is not clear, whether the same applies for the MMN as well, i.e. it is not clear, whether a reduced MMN in schizophrenia is a consequence of higher VI or a standalone effect. It is very important to clarify this, as a reduced MMN as the consequence of a higher VI would mean that the automatic signalling of changes in the acoustic stream is not necessarily deficient but the result of less orchestrated neural firing. The present thesis aimed to (1) identify, whether functional differences between people with and without schizophrenia can be attributed to (i) anatomical differences or (ii) increased cortical noise; (2) investigate, whether cortical correlates of schizophrenia can be altered by training. The first study (p. 10, Jordanov et al., 2010) of the present thesis tried to determine, whether the typically reported asymmetric localisation of the primary auditory sources (Heim et al., 2004; Teale et al., 2003) is the result of a reduced anatomical asymmetry, or whether auditory processing takes place in a different cortical region. If the latter were the case, it would mean that schizophrenia is not only characterised by structural cortical differences but that cortical processing is organised in a different way than in healthy control participants. It was then investigated, whether training could change the localisation of the M50 (p. 40, Jordanov et al., Posterpresentation, SPR, 2010). The training procedure will be described below. The second study of the present thesis (p. 24, Jordanov et al., in press) investigated, whether the reduced MMNm in schizophrenia is related to an increased variability of brain activity. For this purpose, MMNm and VI were compared between patients with and without schizophrenia. VI was determined during auditory processing and during resting state in different frequency bands. If MMNm amplitude was related to increased VI in schizophrenia it might imply that the disturbed automatic auditory processing is a result of less orchestrated neural firing. The third study (p. 40, Jordanov et al., Posterpresentation, SPR, 2010) looked for dependencies between the dipole parameters of an auditory M50 response model (e.g. 20

localization, orientation and latency) in a paired-click paradigm and the symptomatic and functional characteristics in schizophrenia. Additionally, the chosen dipole parameters were investigated for training-induced changes after four weeks computerbased cognitive training. The fourth study (p. 51) investigated, whether VI can be influenced by the CE and CP training programs. Matching the subjectively described effort of the training (based on a pilot trial) rather than the total training time, the more game-like CE-training comprised 60-min sessions on 20 consecutive workdays (4 weeks), whereas Cogpacktraining followed the standard instructions of three 60-min sessions per week over 4 weeks yielding in approximately similar absolute training hours. Given that the CE training focuses more specifically on auditory processing, it is expected that it has a greater effect on the localization of the auditory M50 and the VI in auditory areas than the CP training, which trains more wide-spread cognitive skills.

21

2. Local Heschl’s Gyrus-based coordinate system for intersubject comparison of M50 auditory response modeled by single equivalent current dipole

Abstract Allocating electromagnetic auditory responses to active regions in the human auditory cortex can be difficult because of high interindividual variability of the relevant structures. Location and orientation of the primary auditory cortex (Heschl’s Gyrus) and the temporal plane vary with individual features such as age, gender, handedness, or between healthy subjects and patients with a psychiatric disorder (e.g., schizophrenia). Here, we propose a reference coordinate system that considers the individual MRIbased position, orientation and length of the primary auditory cortex to account for interindividual variability. Transformation of the M50 dipole localizations in this new HG-(Heschl’s-Gyrus)-coordinate system, accomplished for 10 healthy subjects and 10 schizophrenia patients, confirmed group difference more precisely than other registration methods. We suggest to use the HG-coordinate system for localization of functional measures and evaluation of brain activity differences between groups or measurement conditions. Keywords: Auditory cortex, Heschl’s Gyrus, Local coordinate system, Intersubject comparison, MEG, Single equivalent current dipole

Introduction The comparison of anatomical areas and functional activities between individual human brains encounters the problem of the high interindividual variability of brain structures. Three commonly used categories of methods to deal with this problem include (a) the registration within a global intrinsic coordinate system using, for instance, the anterior and posterior commissures (AC and PC) as reference landmarks, (b) global normalization of individual data with reference to 3D brain volume (e.g., Talairach and Tournoux, 1988; Evans et al., 1993; Friston et al., 1995), or (c) global normalization with respect to gyral and sulcal patterns on the cortical surface (Thompson and Toga, 1996; Van Essen and Drury, 1997; Van Essen et al., 1998; Fischl et 22

al., 1999; Dickson et al., 2001). All of these methods can be used to assign activity obtained from functional magnetic resonance imaging (fMRI), electroencephalography (EEG) and magnetoencephalography (MEG) to anatomical structures. (a) leads to more accurate results for anatomical structures near to the reference points (AC and PC) than for structures more distant to the reference points or for structures that differ in localizations between the hemispheres, such as the temporal plane (planum temporale, PT). A major limitation of the volume based registration methods (b) is that the average brain loses its representative surface structure by obscuring gyral and sulcal organization (Kang et al., 2004). The third group of methods (c) seeks to consider the natural organization of the brain. Given that the neurons of the human cerebral cortex are arranged in a densely folded sheet forming the surface all over the brain the more natural modeling of the cortex should be a surface instead of 3D volume. Although the surface is a good model of the human cerebral cortex it is not necessarily suitable for intersubject comparisons of brain activity modeled by a single equivalent current dipole using volume conductor models with typically three spatial degrees of freedom. Furthermore, the source localization inaccuracy due to the uncertainty pertaining to the inverse algorithm, a low signal to noise ratio and a functional to structural dataset coregistration errors can cause a dipole localization displacement up to few millimeter in an arbitrary direction within the volume conductor model. The projection of that error afflicted dipole localization on the cortical surface can cause an additional error in the brain activity localization. Therefore, the surface-based co-registration methods are more suitable for group comparison of cortical activity using distributed source models rather than single equivalent current dipole models. One common characteristic of all the methods linked above is that they act in a global manner, i.e. they register entire brains with all sulci and gyri. This can be considered a disadvantage if the goal is to compare structures small in size and variable in position and form such as Heschl’s Gyrus (HG). In that case methods based on local landmarks and structures are required. Several approaches have been published to locally compare auditory cortices. Among these are surface-based (Kang et al., 2004), also local parcellation methods (Kim et al., 2000; Engel et al., 2009), volume based automated HG extraction method (Tahmasebi et al., 2009) and nonlinear volume based registration of individual HG with a standard brain’s HG (Viceic et al., 2009). All these methods are useful and precise but they may not be the perfect choice for intersubject 23

comparison of functional activity modeled by single equivalent current dipole due to an error affliction of the dipole localization and the nonlinear error propagation caused by these methods. That is, dipole localizations will always be subject to errors caused by e.g., low signal to noise ratio or uncertainty pertaining to the inverse algorithm. Thus, the dipole fit will likely be displaced from the true source by a certain distance. Nonlinear transformation approaches can augment this inherent localization error even further. Precise mapping of functional activity is particularly important when aiming to relate regional functional activity to specific group characteristics, to curtail dysfunctional regions when comparing specific patient groups, or to assess functional regional activity between measurements within the same subject, for instance, as a function of intervention. Here a reference coordinate system is described that should compensate for interindividual variability in the comparison of structural and functional measures between and within subjects. The coordinate system was developed and evaluated for Heschl’s Gyrus (HG) as a crucial structure of the primary auditory cortex in the human brain, which is located anterior to the PT and exhibits hemispheric asymmetry. HG is extended more anterior in the right than in the left hemisphere, this asymmetry varying (between individuals) with handedness, gender, or age. Moreover, reduced asymmetry has been reported in schizophrenia patients (Rossi et al., 1992, 1994; Petty et al., 1995; Barta et al., 1997; Hirayasu et al., 2000; Kasai et al., 2003). The proposed coordinate transformation serves to account for this interindividual variance. It combines the simplicity and speed performance of the linear with the high accuracy of the nonlinear registration methods. Its validity was probed in the study of reduced hemispheric asymmetry in schizophrenia patients compared to healthy subjects.

Materials and methods Subjects Structural magnetic resonance images and magnetoencephalographic data obtained from 10 healthy subjects (1 female, 23–29 years) and 10 inpatients meeting ICD-10 Classification of Mental and Behavioral Disorders of schizophrenia (F20.0, 1 female, 20– 49 years) served for the development and testing of the HG-coordinate system 24

transformation. All subjects were informed about the measurement procedures and gave written consent. Data acquisition T1 weighted structural magnetic resonance images (MRI) were obtained from each participant using a Philips Gyroscan 1.5 Tmagnet resonance imaging system (Philips Medical Systems, Gyroscan ACS-T). The resolution of the MRIs was 256×256×200 voxels, with 200 slices covering the left-to-right direction. For identification of head orientation a marker was attached on the right cheek of every subject before starting the MRI scan. For further analysis raw data files were converted with the r2aGUI tool V.2.5.1 (Medical Center and Helmholtz Institute, Utrecht University, Netherlands) in Neuroimaging Informatics Technology Initiative (NIfTI) format. Functional activity of the auditory cortex was measured using magnetoencephalography (MEG). MEG was recorded while the subject was lying in a supine position, using a 148-channel magnetometer (MAGNESTM 2500 WH, 4D Neuroimaging, San Diego, USA) inside amagnetically shielded room (MSR, Vacuumschmelze, Hanau). Data were continuously recorded with a sampling rate of 678.17 Hz and a real bandpass filter of 0.1 to 200 Hz. For artifact control, the vertical and horizontal electro-oculogram (EOG) was recorded from four electrodes placed near the left and right temporal canthus and above and below the right eye using a SynAmps amplifier (NEUROSCAN Laboratories, Sterling, VA, USA). The subject’s nasion, left and right ear canal, and head shape were digitized with a Polhemus 3Space® Fasttrack prior to measurement. Subjects passively listened to pairs of clicks following the standard paired click design for the assessment of M50 (Adler et al., 1982): 100 pairs of 3ms square-wave clicks were presented with a 500ms offset to- onset interval and an 8 s jittered inter-trial interval (offset to onset). Clicks were presented at 50 dB above the subjective hearing level, which was determined separately for each ear, and delivered via plastic tubes to the subject within the shielded MEG recording chamber. Subjects had no task but were asked to keep their eyes focused on a small fixation point at the ceiling throughout the measurement. Data analysis

25

Source localization with Curry Software V.6 (Compumedics Neuroscan USA Ltd.) was based on 68 sensors, 34 over the left and 34 over the right hemisphere, with the aim in mind to exclude magnetic fields generated from non-auditory sources and consequently to increase the goodness of fit for the source localizations. Each 1500-ms epoch (beginning 500ms before the first click) was referred to a 100-ms pre-stimulus interval. Epochs were filtered using functions with Gaussian shaped window slopes: high pass frequency 5Hz (lower edge frequency width 8 Hz) and low pass frequency 55 Hz (upper edge frequency width 11 Hz). Co-registration of the MEG data with the individual MRI comprised two steps: (1) the nasion point and the left and right auricular points (LA and RA) were used for rough alignment; (2) the subject’s digitized headshape was aligned to the headshape drawn from the MRI. A three-layer boundary element model (BEM) generated for every subject was served as volume conductor in the dipole fit procedure using two fixed equivalent current dipoles with seed points in the middle of the left and right HG. TheM50to the first click was identified as the first upward oriented dipole before the M100 in the time interval 35–80 ms. The dipole with the best goodness of fit within a 6-ms interval around the peak of the M50-wave was considered as location of the M50 generator. Only dipole fits with goodness of fit greater than 80% were considered as reliable. Dipole localizations and orientations were determined for ACPCaligned individual MRI datasets. Coordinate transformations FMRIB Software Library (FSL) (FMRIB Analysis Group, Oxford, UK) (Smith et al., 2004; Woolrich et al., 2009) was used for coordinate transformations. The new coordinate system based on the individual Heschl’s Gyrus will be labeled HG-coordinate system. Dipole localizations obtained from Curry V.6 software were superimposed to the transformed MRI datasets and localizations were read out in millimeter-coordinates for further transformations. The subsequent transformation into the HG-coordinate system was constructed as follows. The starting point for the transformation in HG-coordinate system is an ACPCaligned MRI dataset. Using the axial plane touching the superior edge of the AC and the inferior edge of the PC (XY-plane in an ACPC-based coordinate system) a first step to determine the HG’s landmarks is to locate the most posterior point of the insula’s gray matter at the border of the temporal lobe (Fig. 1, left). Then, the sagittal plane is 26

determined which is parallel to the mid-sagittal plane and cuts the posterior point selected before (Fig. 1, left). In that sagittal plane the origin of the HG can be localized which extends transverse on the supratemporal plane (Fig. 1, right).

Fig. 1. Landmarks definition—step 1. The first step for the coordinate system definition is to determine the position of the right HG with respect to the reference structures AC and PC.

After the identification of the HG the most anterior and inferior (labeled HG1) as well as the most posterior and inferior (labeled HG2) points of the HG are located. Drawing first a line connecting the points HG1 and HG2 and another line perpendicular to the middle of that line allows to determine the point where this perpendicular line crosses the superior edge of the HG and label it as HG3. The middle of the distance between HG3 and the center of (HG1, HG2), labeled HG4, is defined as the origin of the new HG-based coordinate system (Fig. 2, left). Furthermore the most distant point with respect to HG4 on the most anterior convolution of the right HG is determined as HG5 (Fig. 2, right). Thereby, the landmarks needed for the definition of the new coordinate system are completed.

27

Fig. 2. Landmarks definition—step 2. HG4 has been chosen as an origin for the new coordinate system. HG5 was determined as the most right-lateral point on the anterior convolution of HG.

The next step accomplishes the construction of the HG coordinate system. Starting from the plane formed by the three points HG1, HG2 and HG3, a line parallel to the (HG1,HG2)-line was drawn to cross the HG4 point. The intersection of that line with the anterior edge of the HG was labeled temporary point one (T1) and the intersection with the posterior edge with temporary point two (T2) (Fig. 3, left). In the plane created by T1, T2 and HG5, a line perpendicular to the line (HG4, HG5) that crossed the HG4 (Fig. 3, right) is used as a Y-axis with a positive direction towards left anterior. The line connecting HG4 with HG5 was chosen as an X-axis with positive direction towards HG5. The Z-axis is chosen to be perpendicular to the XY-plane with positive superior direction (right-handed coordinate system). The coordinate system for the left HG was chosen to be left handed with X-axis (HG4,HG5), Y-axis pointing in right-anterior direction and Z-axis with positive superior direction (Fig. 4).

28

Fig. 3. Heschl’s Gyrus coordinate system construction—Part 1. Step one (left) and step two (right) for the construction of the HG-coordinate system.With these steps the XY-plane of the new coordinate system is completely defined.

Fig. 4. Heschl’s Gyrus coordinate system construction—Part 2. The construction of the HG-coordinate system is complete. For the left HG the X-axis is pointing left anterior, the Y-axis right anterior and Z-axis in superior direction (left handed coordinate system). The coordinate system for the right HG have switched X and Y axes, hence, right-handed coordinate system.

An additional step for taking the length of the individual HGs into account is to normalize with respect to the distance between the points HG4 and HG5. Dividing the Xaxis values with that normalization factor results in a HG’s length that is equal one for all subjects and hemispheres. In order to get the length again in millimeter one has to multiply the X-axis values with a reference value. Possible choices for such a reference value may be the mean value over all subjects and hemispheres or the mean length over all subjects calculated separately for each hemisphere. In this paper the first method is used—the average over subjects and hemispheres (mean length 36 mm). The hypothesis of reducing the interindividual structural variability by the HG transformation (HGT) was evaluated by comparing the distribution of a points chosen by a scientist naïve to the HGT, separately for each dataset, one point in the middle of the 29

HG for each HG using the individual’s MRI dataset in ACPC-coordinate system. Then, both methods, HGT and FNIRT (FMRIB’s Nonlinear Image Registration Tool), were used with these points with the standard deviation (SD) in three-dimensional space

SDx2  SD y2  SDz2 , where SDx, SDy, SDz are the standard deviations in one dimensional space for the axes X, Y and Z, respectively) for each hemisphere as dispersion measure after the transformations. Table 1 Standard deviation comparison between HGT and FNIRT for the points chosen by a naive rater.

Standard Deviation of the Chosen Points (in mm) TRANSFORMATION

ACPC

HGT

FNIRT

LEFT

6.77

3.83

3.84

RIGHT

6.08

2.29

2.59

Asymmetry of source localization between both groups in all coordinate systems was analyzed with a repeated measures ANOVA using group (controls, patients) as a between-group and hemisphere (left, right) as a within-group factor. Validity of the HGT for functional measures was demonstrated by comparing the anterior–posterior asymmetry of the M50 response between healthy control subjects and schizophrenia patients (Fig. 5). At last error propagation for HGT and FNIRT is considered. A point in the middle of the right HG was chosen in a single subject’s coordinate system based on the anterior and posterior commissures (ACPC-coordinate system) and transformed using both—HGT and FNIRT. Then, the same point was shifted with 1mm towards posterior direction (in ACPC-aligned MRI) and transformed again with both methods. The resulting points were compared with the original point by measuring the Euclidean distance between the new shifted points to the original ones (absolute error). Those procedure of shifting was repeated with 2, 3, and 4mm, up to 15mm from the original point.

30

Fig. 5. Average M50 dipole locations for both groups with FNIRT. Mean values of M50 dipole localizations for schizophrenia patients and healthy controls after FNIRT registration.

Results The SD for the points labeling the middle of each HG do not differ between both methods—HGT and FNIRT. For the left hemisphere the values are almost identical for both methods (Table 1, 3rd row). For the right hemisphere results differ by 0.3mm (advantage for the HGT) (Table 1, 4th row), whereas both methods have a lower SD than the ACPC-based alignment (Table 1, 2nd column). Further results about the characteristics of the HGT were demonstrated by asymmetry differences between schizophrenia patients and healthy controls. In an ACPC-based coordinate system a significant interaction hemisphere×group (F(1,18) = 5.76, p < 0.05) confirmed that M50 was localized more anterior in the right than in the left hemisphere in controls (post hoc Fisher LSD-test p = 0.04), whereas the patient group did not show hemispheric differences in M50-localization (Table 2, 2nd column). When transformed into the HG-coordinate system (Fig. 6), a significant interaction hemisphere×group (F(1,18) = 8.23, p < 0.02) confirmed not only hemispheric asymmetry in controls (post hoc Fisher LSD-test: p < 0.03) but also indicated more 31

posterior M50-localization in the left hemisphere in controls compared to patients (p < 0.02) (Table 2, 3rd column).

Fig. 6. Average M50 dipole locations for both groups with HGT. Mean values of dipole localizations in HGcoordinate system for schizophrenia patients and healthy controls. Table 2 p-values for the asymmetry in the different coordinate systems.

Statistical Group Differences in the Dipole Positions (p values) Transformation

ACPC

HGT

FNIRT

Interaction

0.03

0.01

0.01

Patients vs. Controls Left

0.11

0.02

0.03

Controls Left vs. Controls Right

0.04

0.02

0.01

Table 3 Standard deviation of the dipole localizations for both methods - HGT and FNIRT.

Standard Deviation of Dipole Localizations (in mm)

32

TRANSFORMATION

HG

FNIRT

LEFT

14.67

16.24

RIGHT

12.11

12.56

Fig. 7. Error propagation. Absolute error for HGT and FNIRT.

The interaction was confirmed also for FNIRT (F(1,18) = 7.47, p < 0.02). For betweengroup differences in the left hemisphere (post hoc Fisher LSD-test: p < 0.03) FNIRT resulted in higher p-value compared toHGT(Table 2, 4th row). FNIRT yielded better score than HGT in the between hemispheres comparison for the controls (post hoc Fisher LSD-test: p < 0.01) (Table 2, 5th row). The SD of the dipole localizations for both hemispheres suggests better results using HGT compared to FNIRT. The SD for the left hemisphere in HGT was 1.5mm smaller than the same in FNIRT, for the right hemisphere HGT was better by 0.4mm than FNIRT (Table 3).

33

In addition both methods differ in sensitivity for error afflicted points (such points are e.g., dipole localizations). For HGT linear relationship between the input error and the absolute error was assessed in contrast to FNIRT, which provides a nonlinear increase for points with more than 3mm input error (Fig. 7).

Discussion Development of a HG-coordinate transformation aimed to (a) account for interindividual variability of structure, (b) relate functional activity to a reference structure and (c) improve the precision of functional measures in the primary auditory cortex for group comparisons. Results indicate that the HG-coordinate transformation met the first aim. The dispersion in the HG-coordinate system was considerably smaller than the one in the ACPC-coordinate system (Table 1, 2nd column). Comparison with FNIRT supported the accuracy performance of the HG-coordinate system transformation (Table 1, 3rd and 4th columns). Moreover, the HG-coordinate system transformation allows the assignment of functional measures to a specific structure. Being entirely based on the position and orientation of the HG in the individual brain, localization of functional measures (such as M50) in the HG-coordinate system can be assumed to reflect the function of this reference structure. The comparison of M50-asymmetry between controls and schizophrenia patients in ACPC-coordinate system did not allow a conclusion whether the differences in asymmetry resulted from structural or functional differences. Group-specific auditory source localization relative to the anterior commissure does not allow to assume activity in the HG. The asymmetry of M50-localization may have resulted from structural asymmetry, since the right HG in patients is located more posterior relative to the AC compared to the right HG in controls. As a consequence the dipoles measuring their functional activity may have varied. Differences in the auditory cortical activity map can only be assumed after transformation of functional source localizations into a coordinate system based on the brain structure contributing to that function. In the present study comparison of asymmetry of activity localizations after transformation into the HG-coordinate system confirmed significant differences between healthy controls and schizophrenia patients. This indicates that the asymmetry differences may depend not only on the anterior–posterior asymmetry of the HG. They also show differences in the functional map. Moreover, transformation into the HG-coordinate 34

system provided information not manifest in the ACPC-coordinate system, in that healthy subjects exhibit more posterior sources in the left hemisphere than patients. It should be noted, however, that structural differences between schizophrenic patients and healthy control participants may exist, but are transformed away by the HGT. Thus, the finding that functional differences exist between patient groups does not rule out anatomical group differences. The importance of the current method lies in its ability to differentiate between anatomical and functional group differences. Another advantage of the HGT compared to FNIRT can be recognized by looking at the absolute error plot (Fig. 7) and the SD for the dipole localizations (Table 3). All nonlinear approaches augment the input error if it exceeds a certain amplitude (for FNIRT this limit is 3mm for a concrete MRI dataset). The same phenomenon is evident in the standard deviation of the dipole localizations in the left hemisphere. A standard deviation value larger than 10mm caused a greater dispersion for the nonlinear approach (here FNIRT) compared to HGT. Since single equivalent current dipole localizations are always afflicted with errors due to inverse problem uncertainty, low signal to noise ratio or subjects’ movement during the measurement, it is likely that these errors increase additionally even more when using nonlinear deformation methods. Finally, the practicality of the method should be commented on. As long as the MRI images are of good quality, identification of the points needed for the HGT is straight forward after a short amount of practice. In sum, the present results clearly indicate the advantage of referring functional measures to a coordinate system that is based on the brain structure corresponding for that function. Considering structural variability increases sensitivity and, hence, validity of functional analysis. Whereas this conclusion refers to functional activity in the primary auditory cortex and a HG-based coordinate system, the present method and results may encourage the development of similar coordinate system transformation for other cortical structures.

35

3. Reduced mismatch negativity and increased variability of brain activity in schizophrenia Abstract Objectives: Schizophrenia patients commonly exhibit smaller amplitudes of mismatch negativity (MMN) than in controls. It remains unclear whether this results from deficient processes indexed by MMN or ‘normally’ though more variable processing. The present magnetoencephalographic study addressed this question by analyzing intra-individual trial-by-trial variability and MMN amplitude. Methods: Twenty inpatients meeting ICD criteria for schizophrenia and eighteen healthy controls participated in an auditory oddball experiment. The neuromagnetic mismatch field (MMNm) was defined as the difference waveform deviant minus standard tone response. Variability index (VI) in different frequency bands was quantified as trial-by-trial variation of stimulus-evoked responses and epoch-by-epoch variation of signal amplitude during a resting condition. Results: Patients displayed a smaller MMNm amplitude and higher VI during the oddball experiment and during the resting condition than in controls. VI and MMNm amplitude were correlated in controls, but not in patients. Conclusion: Reduced MMN in schizophrenia cannot be explained by augmented variability of brain activity; deficient auditory sensory memory and stimulus related phase-locking may characterize the disorder. Significance: Understanding the contribution of diminished temporal stability of neuronal network dynamics to schizophrenia is crucial in modeling the impact of such instability on performance and thus for understanding deviant attention and memory functions. Keywords: MEG, schizophrenia, mismatch negativity, noise, signal variability, trial-bytrial variability

Introduction

36

Event-related brain potentials (ERP) have been used to understand perceptual or cognitive abnormalities in schizophrenia patients relative to healthy subjects.

A

common finding is that mean amplitudes of various ERP components are smaller in schizophrenia patients than in controls (Callaway et al., 1970; Winterer et al., 2000, 2004). However, it is still discussed, whether smaller mean amplitudes result from smaller activity (as a function of less efficient processing) or larger trial-by-trial variability of activity, which would point to less stable, but otherwise unimpaired processing. Response variability across trials has been examined for amplitude and latency of some ERP components such as N100, P300, and P50 gating ratio (Roth et al., 2007; Iyer and Zouridakis, 2008; Jansen et al., 2010), and augmented inter-trial temporal variability in schizophrenia patients have been related, for instance, to sensory gating deficits (Iyer and Zouidakis, 2008) or increased ‘cortical noise’ (Roth et al., 2007), which might be considered a function of less orchestrated neuronal firing (Rolls et al., 2008). However, substantiating the contribution of trial-by-trial or epoch-by-epoch variability to mean response amplitude as a general phenomenon should be beneficial for a better understanding of perceptual and cognitive abnormalities in schizophrenia. The Mismatch Negativity (MMN) is another example of an ERP with (frequently reported) reduced amplitudes in schizophrenia patients. The MMN is described as a negative deflection in the ERP and prominent at fronto-central and central scalp electrodes in the difference wave obtained by subtracting the ERP to frequent “standard” stimuli from that to rare “deviant” stimuli (Näätänen et al., 2007). The MMN is considered as a measure for automatic context-dependent information processing and auditory sensory memory (Umbricht and Krljes, 2005). Smaller mismatch fields in schizophrenia

patients

than

in

controls

have

been

confirmed

for

the

magnetoencephalographic (MEG) counterpart of the MMN (MMNm; KreitschmannAndermahr et al., 1999; Pekkonen et al., 2002; Kircher et al., 2004). Smaller-than-normal mean MMN amplitude in schizophrenia has been discussed as evidence of deficient automatic auditory sensory processing, pre-attentive auditory perception and discrimination (Pekkonen et al., 2002; Magno et al., 2008; Näätänen and Kähkönen, 2009; Turetsky and Moberg, 2009), or insufficient recruitment of brain areas during task performance (Bates et al., 2009). However, such interpretations do not seem justified as long as the contribution of augmented trial-by-trial response variability to the mean MMN has not been specified. If a substantial contribution is established, 37

reduced MMN amplitudes might indicate deficient stimulus discrimination processes but also efficient though more variable processes. The present magnetoencephalographic (MEG) study addressed this discussion by analyzing trial-by-trial variability of event-related activity in the MMN paradigm. MEG seems advantageous for these analyses as (a) MEG measurements are less affected by conductivities of the skull and the scalp than EEG measurements, so that interpretation of MEG signals does not require preliminary knowledge of the thicknesses and conductivities of the tissues in the head (Hämäläinen et al., 1993); (b) the magnetic field drops rapidly with distance (1/r2), which improves the separation of simultaneous activities in the left and in the right brain hemispheres (Reite et al., 1999). An auditory oddball design was used to measure MMNm in schizophrenia patients and healthy controls using. Average MMNm amplitude and trial-by-trial variation of stimulus-evoked responses were analyzed in addition to epoch-by-epoch variation of brain activity in a resting condition. Both variability indices (VI) were tested for potential relationships with the MMNm amplitude. Resting state data were included in order to determine whether a prominent VI in schizophrenia patients was linked to stimulus-related brain activation or whether it would reflect a more general variability of brain activity across time. We hypothesized that (1) a correlation of small MMNm and large VI measured in the oddball design indicates that MMNm could result from deficient auditory processing and/or from more variable processing; (2) a correlation of small MMNm and large epoch-by-epoch variability (VI) during resting state indicates that deficient auditory stimulus processing is related to temporal variability of brain activity in general, and (3) if the two measures were uncorrelated, at least two processes, poor stimulus discrimination and adequate though variable stimulus discrimination must be considered when discussing dysfunctional processing in schizophrenia.

Methods Subjects Twenty inpatients meeting the ICD-10 criteria for paranoid-hallucinatory schizophrenia (F20.0) and 18 psychiatrically healthy subjects participated in the oddball experiment (see Table 1 for demographic characteristics). Resting state data were available for 16 38

patients and 14 controls (see Table 1). Patients and controls did not differ with respect to age, but controls had more years of education. Table 1: Demographic characteristics for the group of schizophrenia patients and the group of healthy controls for measurement conditions (oddball design and resting state) Patients

Controls

Group difference

N

20

18

Age (M±SD)

31±8.5

27.7±4.8

0.17a

Gender: Females/Males

1/19

7/11

0.01b

Years of education (M±SD)

12±2.1

17.4±2

0a

Handedness Left/Right

4/16

2/16

0.45b

N

16

14

Age (M±SD)

30.3±7.3

30.4±7.1

.97a

Gender: Females/Males

1/15

4/10

0.10b

Years of education (M±SD)

12.1±2.1

18±2

0a

Handedness: Left/Right

3/13

2/12

0.74b

Oddball Design

Resting state

Note: M±SD: mean ± standard deviation a One-way ANOVA; b Pearson’s Chi-square test Table 2: Clinical characteristics of schizophrenia patients Characteristic

M±SD

Range

BPRS

47.3±9.0

26 - 65

GAF

34.1±8.3

20 - 60

BDI

12.5±11.2

1 - 48

Medication: CPZ

843±583

245 - 2700

Medication, number of patients receiving: typical neuroleptics

2

atypical neuroleptics

13

both

5

Note: 39

BPRS: Brief Psychiatric Rating Scale, BPRS (Lukoff et al., 1986); GAF: Global Assessment of Functioning Scale; DSM-IV-TR (American Psychiatric Association, 2000); BDI: Beck Depression Inventory (Beck et al., 1996); CPZ: chlorpromazine equivalents. M±SD: mean ± standard deviation Symptom severity in patients was assessed by the BPRS (Brief Psychiatric Rating Scale, BPRS; Lukoff et al., 1986), Beck Depression Inventory (BDI-II; Beck et al., 1996), and GAF (Global Assessment of Functioning Scale; DSM-IV-TR; American Psychiatric Association, 2000; see Table 2). All patients were on psychoactive medication (see Table 2). Control subjects were included, if they did not meet criteria for a lifetime diagnosis of mental illness (screened with the MINI interview; Ackenheil et al., 1999) and were free of psychoactive medication. For all subjects, exclusion criteria included any history of head trauma with loss of consciousness. All subjects were informed about the measurement procedures and gave written consent prior to measurements.

Stimuli and design The study design was approved by the Ethics Committee of the University of Konstanz. The MMNm was determined in an oddball design, which included a random sequence of 1800 standard tones (500 Hz, 20 ms duration, called S-evoked from hereon), 200 frequency-deviant tones (550 Hz, 20 ms duration, called FD-evoked from hereon), and 200 duration-deviant tones (500 Hz, 60 ms duration). Responses to duration-deviant stimuli were not analyzed in the present study. Stimuli were presented binaurally with 270±15 ms offset to onset interval. A minimum of three and a maximum of six standards were presented between two deviants. Tones were delivered via plastic tubes to the subject within the shielded MEG recording chamber and presented at 50 dB above the subjective hearing level, which was determined separately for each ear. No task was involved, but participants were asked to keep their eyes focused on a small fixation point throughout the measurement.

Data acquisition and analysis MEG was recorded while participants were in a laying position, using a 148-channel magnetometer (MAGNES™ 2500 WH, 4D Neuroimaging, San Diego, USA). Data were continuously recorded with a sampling rate of 678.17 Hz and a real bandpass filter of 40

0.1 to 200 Hz. For artifact control the vertical and horizontal electro-oculogram (EOG) was recorded from four electrodes placed near the left and right temporal canthus and above and below the right eye using a SynAmps amplifier (NEUROSCAN Laboratories, Sterling, VA, USA). The subject’s nasion, left and right ear canal, and head shape were digitized with a Polhemus 3Space® Fasttrack prior to measurement. Global noise was filtered from the MEG data offline by subtracting non-biological external noise that was recorded by 11 MEG reference sensors. Prior to subtraction, reference channels were multiplied with individually calculated fixed weight factors. This noise reduction procedure has little or no influence on biological signals, as the distance between the reference sensors and the subject’s head is large (M±SD 25.8±6.0 cm, min 15.5 cm, max 36.5 cm) relative to the distance between the head and adjacent sensors. MEG data were analyzed using the Matlab-based FieldTrip toolbox, developed at the Donders Institute for Brain, Cognition and Behavior (http://fieldtrip.fcdonders.nl/). Data segments containing eye blinks, muscle artifacts or superconducting quantum interference device (SQUID) jumps were rejected using an artifact rejection function. For each trial the variance across all sampling points was calculated separately for each channel. Whenever the maximum value across channels exceeded a threshold value of 1×10-24 fT2, the trial was classified as artifact-contaminated and rejected from further analysis. The threshold value was empirically determined by computing the variance for random samples of trials with and without artifacts. Variance of artifact-free trials did not exceed 1×10-24 fT2, whereas the variance on trials with eye-movement artifacts ranged from 0.5×10-24 to 6.4×10-24 fT2. Analyses of averaged responses were based on planar gradients of the MEG field distribution determined by nearest-neighbor method (Bastiaansen and Knösche, 2000). The horizontal and vertical components of the estimated planar gradients approximate the signal measured by MEG systems with planar gradiometers (see Figure 1). As the maximal activity of planar gradients is typically located above the source, this analysis can be used to estimate loci of activity sources (Hämäläinen et al., 1993).

41

Figure 1. Grand average mismatch fields averaged separately for healthy controls (left) and schizophrenia patients (right) for axial gradiometers transformed into planar gradiometers. The color bar indicates field strength in femto Tesla/cm (fT/cm). Continuous data were filtered with a 1 Hz high-pass zero-phase and a 45 Hz low-pass zero-phase Butterworth filter of second order. For MMNm analysis epochs of 300 ms duration following stimulus onset relative to a 100 ms prestimulus interval were determined separately for standards and deviants. Resting state data were segmented into 1-sec epochs resulting in an average of 250 epochs. The same number of trials per subject were selected for further analyses by selecting for each subject the first 1000 artifact-free responses to standards, the first 100 artifact-free responses to deviants, and the first 150 artifact-free resting state epochs. The neuromagnetic mismatch field (MMNm) was calculated by subtracting the Sevoked from the FD-evoked field per sensor. Variability in the oddball design was defined as trial-by-trial variation of stimulus-evoked responses (Möcks et al., 1988) calculated for each MEG sensor, each time point, and each stimulus type (standards and deviants). The VI (variability index) estimate is the mean magnitude of the difference between every single trial (or epoch in the resting data) and the average across trials. VI is calculated as a function of time over the entire trial/epoch length ( )



∑( ( )

̅ ( ))

with N representing the number of trials/epochs, t indicating the current time point, xi(t) the value of the i-th trial/epoch in the time point t, and

̅ ( ) the average over all 42

trials or epochs in the time point t. Results will be reported as VI in femto-Tesla (fT), the square root of the noise power (fT2). VI(t) in both oddball and resting state data, was also estimated for frequency bands after band-pass filtering the data: delta-theta (1 to 7 Hz), alpha (8 to 12 Hz), beta (13 to 24 Hz) and gamma (25 to 45 Hz). The mean MMNm amplitude, the VI(t) for oddball, and the VI(t) for resting state were first averaged over time (100 to 240 ms for MMNm; 0 to 250 ms for VI(t) in the oddball design; 0 to 250 ms for the VI(t) during resting state) and then compared between groups (schizophrenia patients versus controls). Comparisons were corrected for family-wise error rate by a non-parametric, t-test based randomization test (Maris and Oostenveld, 2007). This procedure effectively controls for multiple comparisons and allows the identification of sensor clusters of significant group differences. A cluster was defined as a set of adjacent sensors (defined as sensors at less than 3.2 cm distance, yielding on average 3 neighbors per sensor) that exhibited similar differences between groups in t-value and magnitude. Group differences were considered statistically robust for a sensor cluster, whenever the significance level exceeded 95%. Signals of significant sensor clusters were averaged and subjected to a one-way analysis of variance (ANOVA) for group comparison. Since VI has been reported to be lower in females than in males (Winterer et al., 2004), separate ANOVAs were calculated (i) for the entire sample comparing patients and controls with the between-subject factor ‘Group’, (ii) for the control group comparing male and female subjects with the between-subject factor ‘Gender’, and (iii) for the male sample with the between-subject factor ‘Group’ comparing male patients and male controls. The relationship between average event-related responses (MMNm) and VI values was probed by Pearson correlation coefficients. Since the number of subjects in the different groups was quite small regarding statistical power, the size of all statistical effects was calculated. Hedges’ g (Hedges and Olkin, 1985) was used as a somewhat more accurate version of Cohen's d as it adds a correction factor for small samples.

Results As illustrated in Figures 1 and 2, MMMm in the time interval 100 to 240 ms after stimulus onset was smaller in schizophrenia patients than in controls for both frequency and duration deviants. MMNm did not differ between frequency and duration deviants, as indicated by a repeated measures ANOVA with the between factor Group and within factor Deviant, which did not confirm a significant interaction Group × Deviant (F(1,36)= 43

1.05, p> .1) or main effect Deviant (F(1,36)= 0.35, p> .5), but only a significant main effect Group (F(1,36)= 8.1, p< .01). Therefore, results are reported for the amplitude averaged across MMNm types. The group difference was confirmed when only male subjects were considered (F(1,28)= 11.7, p< .01; see also Table 3). As evident in Figure 2 and confirmed by a randomization cluster, statistics differences were prominent at bilateral fronto-temporal sensors.

Figure 2. Left: Scalp distribution of MMNm group differences in the time window 100240 ms after stimulus onset. The range of t-values is represented by color shading (color bar indicates t-values). Sensors contributing to a cluster of significant group difference (p< .025) are represented by asterisks. Right: The quadratic mean of the mismatch response across significant sensors is plotted over time (-100 to 300 ms) for healthy controls (solid line) and schizophrenia patients (dashed line). The MMNm is evident as a prominent deflection between 100 and 240 ms after tone onset.

44

Table 3. Statistical effects for the analysis of variance with the factors Group and Gender on MMNm and VI. The main effects Group of Gender are indicated with F and p values, effect sizes are denoted by Hedges’ g. scores

Statistical

All subjects

Male subjects only

Controls only

effect

Patients vs. controls

patients vs. controls

Female vs. male

Sample size

20 vs. 18

19 vs. 11

7 vs. 11

Group

F(1,36)= 9.18, p = 0.0045

F(1,28)= 12.7,p= 0.0013

F(1,16)= 1.84, p = 0.194

Effect size

g = -0.96

g = -1.33

g = -0.63

Variability index

Sample size

20 vs. 18

19 vs. 11

7 vs. 11

oddball

Group

F(1,36)

F(1, 28) = 5.08, p = 0.03

F(1,16)= 3.21,p = 0.092

Effect size

0.00037

g = 0.83

g = -0.82

MMNm

planar

gradient (fT/cm)

design

(fT)

=

15.44,

p=

g = 1.26 Variability index

Sample size

16 vs. 14

15 vs. 10

4 vs. 10

resting state (fT)

Group

F(1,28)= 13.51, p= 0.0001

F(1,23)= 7.49,p= 0.012

F(1,12)= 0.64, p = 0.44

Effect size

g = 1.31

g = 1.08

g = -0.44

VI in the oddball design was larger in patients than in controls (Figure 3). Differences were not confined to stimulus onset but evident across the entire time interval for responses to standards and deviants. The less variable VI time course for standards may be explained by the larger number of 1000 averages relative to 100 averages selected for deviants. Differences were significant at bilateral temporo-frontal sensor clusters. As for MMNm group differences remained when considering male subjects only.

45

Figure 3. Left: Scalp distribution of group differences of variability indices (VI) following standard (S, top row) and frequency-deviant (FD, bottom row) stimuli. The range of tvalues is represented by color shading and the color bar indicates t-values. Sensors contributing to a cluster of significant group difference are marked by asterisks (for p< .025) or X (for p< .05). Right: The quadratic mean of VI(t) across significant sensors is plotted for patients (dotted lines) and controls (solid lines), as well as for standard tones (top), and deviant tones (bottom) for the time interval -100 to 300 ms.

46

Figure 4. Left: Scalp distribution of group differences of variability indices (VI) during resting state. The range of t-values is represented by color shading and the color bar indicates t-values. Sensors contributing to a cluster of significant group difference are marked by asterisks (p< .025). Right: The quadratic mean of VI(t) across significant sensors is plotted for patients (dotted lines) and controls (solid lines) for a 1 sec time interval. VI determined from resting state MEG (Figure 4) was also more pronounced in schizophrenia patients than in controls. Group differences were mainly confined to left temporo-frontal sensor clusters. No effect of gender was found. VI determined for the different frequency bands (Figure 5) showed group differences of resting state VIs in the delta-theta band and mainly in a left temporal sensor cluster (p0 (rising slope)

Figure 2. Graphical representation of the steps needed for the calculation of the rotation angle in the case when m>0. Calculation of the cross point with the Y-axis For the calculation of the cross point with the Y-axis the equation of the linear function is needed. With its help all points lying on the line can be determined. The equation of the line determined by two points A and B with coordinates in the 2D space (x1, y1) and (x2, y2), respectively, is given by the following equation: ( ) 99

Consequently the value of the function in x=0 is the wanted point (Figure 2), i.e. ( )

Shift the line along the Y-axis into the origin The line should be shifted in the origin O by subtracting the f(0) from the Y-coordinates of the points A and B, i.e. the new points A’ and B’ have the coordinates (x1, y1-f(0)) and (x2, y2-f(0)), respectively (Figure 2). Determine the angle of rotation After the line was shifted to the origin of the coordinate system it is possible to easy calculate the angle of rotation. It is just needed to calculate the angle at the origin O in a triangle determined by the coordinate system’s origin O, the shifted point B’ and the orthogonal projection of B’ on the X-axis. The equation for the angle of rotation is given below: (

( ) ) ‖ ‖

Where y2-f(0) is the Y-coordinate of the point B’ and ||B’|| denotes the Euclidean norm in the 2D space. Shift the line back again Before applying the rotation the line have to be shifted back to the original position. Otherwise, the transformation is going to be wrong. Second case: m

Suggest Documents