Wang, Sheng Hua
(Helsingin yliopisto, 2021)
Neurophysiological dynamics of the brain, overt behaviours, and private experiences of the mind are co-emergent and co-evolving phenomena. An adult human brain contains ~100 billion neurons that are hierarchically organized into intricate networks of functional units comprised of interconnected neurons. It has been hypothesized that neurons within a functional unit communicate with each other or neurons from other units via synchronized activity. At any moment, cascades of synchronized activity from millions of neurons propagate through networks of all sizes, and the levels of synchronization wax and wane. How to understand cognitive functions or diseases from such rich dynamics poses a great challenge. The brain criticality hypothesis proposes that the brain, like many complex systems, optimize its performance by operating near a critical point of phase transition between disorder and order, which suggests complex brain dynamics be effectively studied by combining computational and empirical approaches. Hence, the brain criticality framework requires both classic reductionist and reconstructionist approaches. Reconstructionism in the current context refers to addressing the “Wholeness” of macro-level emergence due to fundamental mechanisms such as synchrony between neurons in the brain. This thesis includes five studies and aims to advance theory, empirical evidence, and methodology in the research of neuronal criticality and large-scale synchrony in the human brain.
Study I: The classic criticality theory is based on the hypothesis that the brain operates near a continuous, second order phase transition between order and disorder in resource-conserving systems. This idea, however, cannot explain why the brain, a non-conserving system, often shows bistability, a hallmark of first order, discontinuous phase transition. We used computational modeling and found that bistability may occur exclusively within the critical regime so that the first-order phase transition emerged progressively with increasing local resource demands. We observed that in human resting-state brain activity, moderate α-band (11 Hz) bistability during rest predicts cognitive performance, but excessive resting-state bistability in fast (> 80 Hz) oscillations characterizes epileptogenic zones in patients’ brain. These findings expand the framework of brain criticality and show that near-critical neuronal dynamics involve both first- and second-order phase transitions in a frequency-, neuroanatomy-, and state-dependent manner.
Study II: Long-range synchrony between cortical oscillations below ~100 Hz is pervasive in brain networks, whereas oscillations and broad-band activities above ~100 Hz have been considered to be strictly local phenomena. We showed with human intracerebral recordings that high-frequency oscillations (HFOs, 100−400 Hz) may be synchronized between brain regions separated by several centimeters. We discovered subject-specific frequency peaks of HFO synchrony and found the group-level HFO synchrony to exhibit laminar-specific connectivity and robust community structures. Importantly, the HFO synchrony was both transiently enhanced and suppressed in separate sub-bands during tasks. These findings showed that HFO synchrony constitutes a functionally significant form of neuronal spike-timing relationships in brain activity and thus a new mesoscopic indication of neuronal communication per se.
Studies III: Signal linear mixing in magneto- (MEG) and electro-encephalography (EEG) artificially introduces linear correlations between sources and confounds the separability of cortical current estimates. This linear mixing effect in turn introduces false positives into synchrony estimates between MEG/EEG sources. Several connectivity metrics have been proposed to supress the linear mixing effects. We show that, although these metrics can remove false positives caused by instantaneous mixing effects, all of them discover false positive ghost interactions (SIs). We also presented major difficulties and technical concerns in mapping brain functional connectivity when using the most popular pairwise correlational metrics.
Study IV and V: We developed a novel approach as a solution to the SIs problem. Our approach is to bundle observed raw edges, i.e., true interactions or SIs, into hyperedges by raw edges’ adjacency in signal mixing. We showed that this bundling approach yields hyperedges with optimal separability between true interactions while suffers little loss in the true positive rate. This bundling approach thus significantly decreases the noise in connectivity graphs by minimizing the false-positive to true-positive ratio. Furthermore, we demonstrated the advantage of hyperedge bundling in visualizing connectivity graphs derived from MEG experimental data. Hence, the hyperedges represent well the true cortical interactions that are detectable and dissociable in MEG/EEG sources.
Taken together, these studies have advanced theory, empirical evidence, and methodology in the research of neuronal criticality and large-scale synchrony in the human brain. Study I provided modeling and empirical evidence for linking bistable criticality and the classic criticality hypothesis into a unified framework. Study II was the first to reveal HFO phase synchrony in large-scale neocortical networks, which was a fundamental discovery of long-range neuronal interactions on fast time-scale per se. Study III raised awareness of the ghost interaction (SI) problem for a critical view on reliable interpretation of MEG/EEG connectivity, and for the development of novel approaches to address the SI problem. Study IV offered a practical solution to the SI problem and opened a new avenue for mapping reliable MEG/EEG connectivity. Study V described the technical details of the hyperedge bundling approach, shared the source code and specified the simulation parameters used in Study IV.