«DEPARTMENT OF PHYSIOLOGY Unsupervised characterization of human brain networks that are involved in emotional processing and regulation THESIS ...»
TEL AVIV UNIVERSITY
SACKLER FACULTY OF MEDICINE
THE DR. MIRIAM AND SHELDON G. ADELSON
DEPARTMENT OF PHYSIOLOGY
Unsupervised characterization of human brain networks that are
involved in emotional processing and regulation
THESIS SUBMITTED FOR THE DEGREE “DOCTOR OF PHILOSOPHY”
BY ADI MARON-KATZ
SUBMITTED TO THE SENATE OF TEL AVIV UNIVERSITYOctober 2015 This work was carried out under the academic supervision of Talma Hendler Professor of Psychiatry and Psychology School of Psychological Science, Sagol School of Neuroscience, Faculty of Medicine Tel-Aviv University Director, Functional Brain Center, Wohl Institute for Advanced Imaging, Tel-Aviv Sourasky Medical Center AND Ron Shamir Professor of Computer Sciences Blavatnik School of Computer Science Tel-Aviv University Dedicated with gratitude to my parents and entire family for their love and encouragement, and especially to Oded, Shachar and Ofek. Without your love and support I would not have completed this thesis.
I wish to start by thanking my Yoga professor Orit Sen-Gupta, who by beautifully sharing her knowledge of the Yogic philosophy of the mind, inspired me to go on this quest. I wish to thank my advisors Prof. Talma Hendler and Prof. Ron Shamir who made it possible for me.
I also wish to thank my fellow students in the lab at Wohl Institute for Advances Imaging at the Tel-Aviv Sourasky Medical Center as well as in the lab for Computational Genomics at Tel-Aviv University for their help and support all through my PhD. Especially I would like to thank Michal Ballas-Gruberger and Haggai Sharon for their invaluable support and inspiration; Eti Ben Simon, David Amar, Sharon Vaisvase, Tamar Lin and Gadi Gilam for productive collaboration; Gal Raz, Shahar Jamshy, Ilana Podlipsky and Yael Yaakov for helpful discussions.
Thanks This work was supported by the Israeli ministry of Science and Technology
TABLE OF CONTENT1.
2.1The resting Brain
2.2 Experience related fingerprint in resting state fMRI
2.3 Neural traces of Inter-individual differences in handling emotionally-challenging experiences
2.4 Methods for studying variability in resting state functional connectivity
3. Research Objectives
4. General methods and materials
4.1 Functional MRI
4.1.2 fMRI acquisition
4.1.3 fMRI preprocessing and parcellation
4.1.4 Cross correlation functional connectivity analysis
4.2 Behavioral measures of affective response
4.3 Comparing neural measures against behavioral and other physiological measures...... 26 5. Improving interpretation of large-scale changes in resting state networks 27
6.............Characterizing changes in resting-state networks induced by a psychological perturbation
6.2 Case study 1: Characterizing changes in resting-state networks induced by acute social stress
6.2.1 Specific Background
6.2.2 Specific materials and methods
6.2.4 Discussion – study 1
6.3 Case study 2: Characterizing changes in resting-state networks following an anger inducing social interaction
6.3.1 Specific Background
6.3.2 Specific materials and methods
6.3.4 Discussion- study 2
6.4 Joint Discussion
Ch aracterizing changes in resting-state networks following a physiological perturbation 83
7.2 Specific materials and methods
8. Concluding Discussion
8.1 Overview of the results
8.2. Methodological insights and contributions
8.2. Study Limitations
LIST FIGURES TABLESOF AND Figure 5-1: Two examples that demonstrate the difference between the two approaches to connectivity enrichment significance
Figure 5-2: RichMind results visualization for case study 1
Figure 5-3: RichMind results visualization for case study 2
Figure 6-1: An illustration of data-driven univariate rsFC analysis.
Figure 6-2:: Psychological response to stress on experimental timeline.
Figure 6-3:: Significant rsFC changes following stress
Figure 6-4: A graph representation of the enrichment analysis results
Figure 6-5: Results of inter-group rsFC change comparison
Figure 6-6: Experimental procedure of case study 2
Figure 6-7: An illustration of the global rsFC analysis steps
Figure 6-8: rsFC changes identified flowing UG
Figure 6-9: Association between overall rAmy rsFC and behavioural measures.................. 80 Figure 7-1: Results of pairwise univariate rsFC analysis seeded in parcels 165 and 148...... 93 Figure 7-2: association between SD –induced rsFC change and behavioral measures........ 94 Figure 7-3: Accuracy rates of LOOCV analysis presented as a function of number of used features (k)
Figure 7-4: LOOCV top-ranking features
Figure 7-5: Annotations used in enrichment analysis
Figure 7-6: SD-induced changes in modular organization.
Table 5-1 – RichMind results for case study 1
Table 5-2 - RichMind results for case study 2
Table 6-1 – Lobe-based enrichment analysis results
Table 6-2 – ENRICHMENT-inducing parcel pairs:
Table 7-1 – Results of univariate overall rsFC analysis
Table 7-2 – Results of pairwise univariate rsFC analysis seeded in parcels 165 and 148..... 91 Table 7-3 - Parcel pairs that were selected as features in all 17 iterations of LOOCV (k=24) analysis
Table 7-4: Modules identified via modularity analysis on group-level rsFC matrices............. 97
LIST OF E QU A TI O N S
(4-1) Pearson Correlation Coeficient
(4-2) Fisher Transformation
(4-3) Spearman Correlation Coeficient
(5-1) Hyper-Geometric score
(7-1) Classificationn success
(7-2) Classification accuracy
(7-3) Classification accuracy p-value
(7-4) Jaccard Score
Abbreviations and symbols
Brain regions: Anterior Cingulate Cortex (ACC); Amygdala (Amy);
Basolateral Amygdala (BLA); Inferior Frontal Gyrus (IFG); Prefrontal Cortex (PFC); dorsolateral PFC (DPLFC); Superior temporal sulcus (STS); Posterior cingulate cortex (PCC); Visual network (VN); Default-mode network (DMN);
Fronto-parietal control network (FPCN); Sub-lobular (SL); Auditory network (AN); Executive control network (ECN); Ventral attention network (VAN);
Dorsal attention network (DAN); Sensory-motor network (SMN) Brain imaging and physiology: Functional Magnetic Resonance Imaging (fMRI); Magnetic Resonance Imaging (MRI); Electroencephalogram (EEG);
General Electric (GE); repetition time (TR); Region Of Interest (ROI); field potential (LFP); Heart rate (HR); Heart rate variability (HRV);
Others: Post-traumatic Stress Disorder (PTSD); Functional Connectivity (FC), Resting State (rs); Leave-one-out cross validation (LOOCV); Ultimatum-game (UG); Trier Social Stress Test (TSST); Degree preserving permutation (DPP);
Hypergeometric (HG); inter-subject correlation (ISC); Automated Anatomic Labeling (AAL)
1. ABSTRACT Background In the past decade and a half, the neuroscience community has learned that during rest, ongoing energy consuming activity takes place in the brain. When focusing on low frequencies, this activity is highly correlated within known functional networks. Although these correlated fluctuations are generally maintained over time, they were shown to vary with changes in cognitive and emotional states, and were suggested to hold information on individual history of interaction with the world as well as a priori cognitive and emotional biases.
Exploration of variability in resting-state (rs) neural functional connectivity (FC) through functional magnetic resonance imaging (fMRI) has been traditionally performed using hypothesis-driven analysis while focusing on one or a few predefined seed regions. This approach can reveal only a fraction of the actual phenomenon as it relies on prior knowledge of the putative functional network structure. An alternative approach is to conduct a whole-brain voxel-wise analysis, which is computationally expensive and sensitive to noise. A possible compromise is to define a set of regions of interest (ROIs) that provide good coverage of the brain. Such dimensionality reduction allows conducting wholebrain rsFC analysis while treating the data as a collection of independent connections and performing statistical analyses of each connection separately.
Alternatively, multivariate techniques evaluate the relationship between the entire connectome matrices and their associated phenotypic variables in a single statistical test. While powerful, such analysis does not reveal information on the involvement of individual connections.
Despite the methodological progress in studying variability in rsFC patters, to date findings that are obtained from large-scale rsFC analysis are mostly interpreted by a qualitative comparison to known neural maps which are is based on existing literature. Such methodology does not use statistical tools for interpretation, and holds the risk of reporting false positive results and missing important findings. In order to perform this interpretation rigorously one must address it statistically. A natural way to do this is by testing whether a link between two known brain regions or functional networks is significantly more prevalent in the results than would be expected by chance (i.e. enriched).
Analysis of enrichment has long been used in the field of Bioinformatics, and is the acceptable way for characterizing large sets of genes that emerge from data driven genomic analysis.
This work uses rsfMRI to examine the way in which different types of emotionally challenging experiences affect patterns of neural coactivation in subsequent resting periods. This is done using several approaches for largescale rsFC analysis in combination with enrichment analysis as an established manner for interpretation. Assuming that rsFC holds information on individual tendencies as well as on prior experience, we examined inter-individual differences in these patterns and their relation to various behavioral measures of emotional reactivity and regulation. We hypothesized that emotionally challenging experiences will induce large scale changes in patterns of fMRI rsFC. We further expected these changes to be associated with subjective measures of emotional experience.
1) Develop improved means for characterizing and interpreting large-scale changes in FC patterns of rsfMRI data.
2) Data-driven investigation of rsFC modulations following several different types of “emotional challenge”.
3) Identify inter-individual differences in rsFC modulations that correspond to inter-individual differences in measures of affective experience.
Methods To evaluate enrichment within sets of neural positions we adopted the statistical hyper-geometric test. We used the same test to evaluate enrichment within sets of neural connections, however for this case we added an additional nonparametric permutation test, which accounts for uneven levels of FC in different brain areas. Both tests were integrated into the RichMind Matlab package.
Given a collection of findings (neural positions or connections), and a known neural mapping, RichMind tests for enrichment in the input, and provides both statistical reports and brain visualization of the identified enrichments. The software was validated on two previously published studies, the first conducted on healthy participants viewing emotion-inducing film clips, and the second on participants with amnestic mild cognitive impairment.
Next, we analyzed data recorded before and after three different emotionally– challenging paradigms: a social-stress induction task, an anger-provoking interpersonal conflict task (the ultimatum game; UG) and a night without sleep (i.e.
sleep deprivation; SD). In all three cases a pre-defined functional parcellation was applied on the data before analysis for dimensionality reduction. We used a univariate analysis approach to identify rsFC changes. In the sleep deprivation study we used a combination of univariate analysis with multivariate approaches of leave-one-out cross validation (LOOCV) and modularity analysis, due to the small sample size. Large-scale findings were characterized using enrichment analysis. Emotional experience was measured using a number of self-reported questionnaires, and in some cases also using a physiological measure of heart-rate and heart-rate variability.
Main results Following the Trier social stress test we identified a large-scale rsFC change across the brain, which included strengthening of thalamo-cortical connectivity alongside a weakening of cross-hemispheral parieto-temporal connectivity.
These alterations were associated with change in subjective stress reports.
Integrating report-based information on stress sustainment 20 minutes post induction revealed a single significant rsFC change between the right basolateral amygdala (BLA) and the precuneus, which inversely predicted the level of subjective recovery. A parcel centered in the right amygdala demonstrated differential rsFC also following the inter-personal conflict task.
Specifically, it showed increased rsFC with a single parcel centered in the right inferior frontal gyrus. Baseline levels of overall rsFC of that parcel were positively correlated with subsequent subject gain in UG as well as reported anger following the game.