Advancements in Brain Stimulation Modeling: Integrating Neurophysiology with Connectome Approaches

Several human studies employing intracerebral and transcranial perturbations reported the cortical network as unable to engage in large-scale and complex interactions during non-rapid eye movement (NREM) sleep and anesthesia. In both conditions, cortical bistability, which is the tendency of cortical circuits to fall into an Off-period after an initial activation, is in a key position to hinder large-scale complex interactions. Accumulating evidence suggests that cortical bistability characterizes not only NREM sleep and anesthesia in healthy subjects; it also marks wakefulness in vegetative state patients and the perilesional area surrounding focal cortical lesions in awake stroke patients, leading to loss of consciousness and loss of functions, respectively. Given the important clinical implications, it is crucial to understand the nature of cortical bistability and its impact on cortical responsiveness. In this talk, we will suggest a mechanistic explanation of the experimental findings above by means of mean-field theory and simulations of a cortical-like module endowed with activity-dependent adaptation. First, we will show that fundamental aspects of the local responses elicited in humans by direct cortical stimulations can be replicated by systematically varying the relationships between adaptation strength and excitation level in the network. Then, we will reveal a region in the adaptation-excitation parameter space of key relevance for both physiological and pathological conditions, where spontaneous activity and responses to perturbation diverge in their ability to reveal Off-periods. Finally, we will substantiate through simulations of connected cortical-like modules the role of adaptation mechanisms in preventing cortical neurons from engaging in reciprocal causal interactions, as suggested by empirical studies.
 

The brain is a complex, nonlinear, multiscale, and intricately interconnected physical system, whose laws of motion and principles of organization have proven challenging to understand with currently available measurement techniques. In such epistemic circumstances, the application of systematic perturbations, and measurement of their effects, is a central tool in the scientific armory. We employed a combination stimulus-evoked high-density electroencephalography data along with a whole-brain connectome-based computational modeling approach to answer questions around the physiological origin of the stimulus-evoked potentials. Initially, we employed a `virtual dissection' approach to study the extent to which model-generated stimulus-evoked stimulation patterns at the primary stimulation site relied on recurrent incoming connections from the rest of the brain, and at what times. These in-silico interventions resulted in substantial reductions in late stimulus-evoked responses when pivotal connections were inactivated. This indicates that network properties are essential for shaping late stimulus-evoked activity.
However, the spontaneous activity in the resting human brain exhibits well-organized spatiotemporal patterns which form the so-called resting-state networks (RSNs) and prior research has revealed a hierarchical organization of these networks, ranging from high-order multimodal networks to low-order networks. In this framework, we further characterized the effects observed for late stimulus-evoked responses, demonstrating how they are highly dependent on whole-brain integrity for high-order networks and mainly restricted to intrinsic network properties when the stimulus is delivered to low-order networks. Finally, subject-specific estimation of model neurophysiological parameters underscores the role of local parameters in evoked responses when stimulation is delivered to low-order networks, while multimodal networks rely on global brain properties. Overall our results, and the framework for investigating such questions that we are introducing here, have clear and practical relevance to basic and clinical research, but also have broader implications for the scientific understanding of functional brain organization. 

In recent years, considerable efforts have been invested to reduce the complexity of brain activity recordings through the application of nonlinear dimensionality reduction algorithms. Besides their practical use for data analysis, low dimensional manifold representations emerge as a fundamental principle underlying information processing in neural circuits. This work addresses two interconnected questions concerning the role of manifolds in large-scale brain dynamics. First, we ask whether the neural manifolds exist independently of the dimensionality of the brain activity recordings; second, we investigate whether the interactions between manifold coordinates can be modelled to provide meaningful insights into the dynamics of whole-brain activity during rest and cognition. To tackle these questions, we used deep autoencoders to extract manifolds of reduced dimensionality underlying functional magnetic resonance imaging (fMRI) recordings at different brain parcellations. Next, we constructed computational models of non-linear oscillators coupled according to the effective functional connectivity of the obtained manifolds, taking into account the level of non-equilibrium dynamics quantified by the non-reversibility of the signals. We found that the underlying dimensionality of whole-brain fMRI recordings has an optimal value around 10 regardless of the parcellation. Furthermore, modelling the dynamics of the coupled manifold coordinates allowed us to more faithfully reproduce whole-brain activity during cognition compared to models in the original state space with a higher number of dimensions. We then leveraged the generative capabilities of the model to perform in silico perturbations in the reduced network space. These findings support the key role of manifolds as an organizing principle of brain function at the whole-brain scale, also suggesting that future modelling efforts should strive to address the dynamics of low dimensionality representations. 

Deep brain stimulation (DBS), which modulates dysfunction in the brain network by applying chronic high-frequency electrical stimulation to a specific location in the brain, is being explored as a groundbreaking therapy for drug-resistant neurological and neuropsychiatric diseases. DBS has been applied to several brain diseases such as Parkinson's disease, epilepsy, obsessive compulsive disorder, and major depressive disorder, and has demonstrated positive effects in improving symptoms. However, despite many efforts to provide personalized treatment through advanced neuroimaging techniques and accumulated clinical expertise, the therapeutic effects still vary from patient to patient. A potential reason for these inconsistent findings may be the individual variation in the brain structure and functional network organization. Because the stimulation response depends not only on the external conditions including its location, type, and parameters, but also on the dynamic state of the brain network being stimulated, a systematic approach to investigate individualized impacts of stimulation is required. Therefore, we propose a virtual brain modeling approach that enables personalized exploration of DBS. In particular, we examine the feasibility of this modeling approach in the context of treatment-resistant depression (TRD).
Virtual brain models are constructed from patient-specific structural data including brain anatomy and connectome, and then equipped with computational neural mass models for each brain region, thereby reproducing the functional dynamics of the brain. Given the recent neuroimaging findings reporting that the engagement of specific fiber tracts at the stimulation site is associated with the efficacy of DBS, we extend the existing virtual brain modeling framework to incorporate the geometry of fiber tracts and investigate stimulus-induced network effects. In particular, we combine two factors: high resolution and explicit fiber tract modeling. A high-resolution brain model at a mm-scale is built by placing neural mass models at the vertices of the brain surface mesh, regardless of brain parcellation (region). Through explicit modeling of fiber tracts considering their locations and geometries, neural mass models located at both terminals of each fiber tract are coupled taking into account connection strengths and transmission delays. This approach also allows the stimulation of segments of the fiber tract, including signal propagation along its entire length in both directions.
As a proof-of-concept study, we construct high-resolution virtual brain models for three patients diagnosed with TRD and undergoing DBS treatment and simulate brain response patterns as a function of stimulation location, parameterized by the contact location of electrodes implanted in each patient. In other words, we simulate the effects of local DBS on the patient-specific large-scale brain networks and demonstrate individual explanatory power by predicting the spatiotemporal response pattern due to stimulation. Regional activities triggered by activation of fiber tracts at the stimulation location are propagated to other regions through interactions based on the brain connectome, thereby generating a unique spatiotemporal response pattern. The simulated signals derived at each vertex are projected into the EEG sensor space and compared with empirical data measuring cortical evoked responses time-locked to stimulation at 2 Hz for each patient. Simulation results demonstrate that, despite some limitations in validation against empirical data (i.e., the need for individual parameter adjustment, and differential accuracy across stimulation locations), the virtual brain model with high resolution, in both network nodes and links, is capable of investigating the functional network effects depending on stimulation location in DBS.
This study provides evidence for the capacity of personalized high-resolution virtual brain models to explain the individual variation in up- and downstream effects of DBS and to quantify functional network contributions to DBS efficacy. The further extended modeling approaches, incorporating the longitudinal changes in the functional network induced by long-term DBS, may elucidate the modulation mechanisms of DBS and pave the new ways in the personalized optimization of brain stimulation.