Integrating empirical and computational approaches to drive brain towards healthy dynamical regimes

Neural activity and functional interactions are naturally variable from moment to moment,
resulting in various dynamic configurations of brain activity. Contemporary theories of brain structure and function emphasize systematic variations in cortical neurobiological properties (cell type composition, receptor density) that are also reflected in spontaneous neuronal activity at multiple temporal and spatial scales. The brain’s dynamical regime can be characterised at macroscale using non-invasive neuroimaging recordings. However, existing studies focus on hand-picked properties such as synchrony, power spectrum, or amplitude of slow fluctuations. In contrast, the time-series literature is vast and interdisciplinary, spanning thousands of potential time-series literature from neuroscience but also physics, economics, and beyond. This talk will outline how this vast time-series literature can be combined with the complementary strengths of electromagnetic imaging (MEG) and functional magnetic resonance imaging (fMRI) to comprehensively characterise the dynamical regime of the brain (Shafiei et al., 2020 eLife).

We can then systematically examine the relationship between slow- and fast-oscillating neural activity and heterogeneous cortical microarchitecture using measures such as intracortical myelin, excitatory and inhibitory receptors and transporters derived from in vivo PET, and transcriptomically-derived cell type composition (Shafiei et al., 2023 Nature Communications). Having identified neurobiological determinants of neural dynamics in the human brain, we examine how pharmacological manipulations impact neural activity. Systematically and reversibly perturbing brain function with anaesthesia while recording neural activity provides a unique opportunity for causal manipulation of the brain’s dynamical regime. We apply massive temporal feature extraction to generate more than 6000 dynamical features that comprehensively characterise local neural activity from functional MRI signals in humans and non-human primates, comparing wakefulness against a wide range of anaesthetics. We identify an evolutionarily conserved dynamical profile of anaesthesia that is underpinned by a phylogenetically conserved axis of gene expression pertaining to regulation of arousal and sleep-wake cycles. The dynamical signature of anaesthesia is reversed upon re-awakening induced by deep-brain stimulation (DBS) of the central thalamus in macaques. This work demonstrates that the global dynamical regime supporting consciousness is amenable to bi-directional control by pharmacology and local stimulation of the central thalamus, reconciling local and global views of brain function. More broadly, the combination of invasive brain stimulation and extensive dynamical phenotyping of neuroimaging recordings provides an exceptional opportunity for translational discovery, leveraging the greater experimental accessibility of animal models to obtain causal insight. Given the increasing evidence that anaesthesia and coma manifest as similar dynamical patterns in the human brain, this work holds promise for thalamic DBS as a potential treatment avenue for chronic disorders of consciousness. 

Deep Brain Stimulation (DBS) is a successful symptom-relieving treatment for Parkinson’s disease (PD). However, the introduction of advanced directional DBS electrodes significantly expands the programming parameter space, rendering the traditional trial-and-error approach for DBS optimization impractical and demonstrating the need for computational tools. From a dynamical systems perspective, DBS can be modeled as a strategy to shift patients’ brain dynamics closer to the healthy state. Our recently developed DBS model using The Virtual Brain simulation tool was able to reproduce multiple biologically plausible effects of DBS in PD (Meier et al., 2022, Experimental Neurology). In the current work, we extend our virtual DBS model toward higher resolution for the stimulus input, now sensitive to the exact 3D location of the activated contact, incorporating streamline activations and the electric field.

We leverage empirical DBS data from N=14 Parkinson’s patients with a total of N=392 contact activations of different electrode settings and corresponding motor task outcome. We then use our Virtual Brain simulation engine to model DBS in these patients. A model based on the principal component involvement of the simulated dynamics demonstrated a correlation between predicted and empirically observed motor task improvements due to DBS of r=0.386 (p<10-4) in a leave-one-out cross-validation. Benchmarking revealed better predictions with our computational dynamics than imaging-based static methods such as the sweet spot (r=0.16, p<0.05) and its recently introduced generalisation, termed the “sweet streamline” (r=0.26, p<10-4) approaches (Hollunder et al., 2024, Nature Neuroscience). Furthermore, our model outperforms the traditional trial-and-error method in predicting optimal clinical settings for individual patients, e.g. achieving over a 60% likelihood of identifying the optimal contact within the first two suggested contacts. Thus, our work enables to transition the brain dynamics of individual DBS-implanted PD patients toward a “healthier” state by identifying a sweet-spot analogue for dynamics, i.e. the target state, and linking these dynamics directly to clinical improvement.

In the future, the identified sweet-spot dynamics can be used to optimize the electrode placement and settings in silico in individual patients, showcasing the potential benefit of whole-brain simulations for improving clinical routine.
 

Traditionally, in resting-state functional MRI, differences between brain states are detected by a characterisation of the brain's dynamical regime. Depending on the a-priori assumptions about the underlying data, fMRI features are extracted such as regional activations, functional connectivity or fractional occupancy of dynamic functional connectivity approaches. However, to assess how brain states transition between each other remains a challenge in current paradigms. One of the solutions has been proposed in seminal work by Deco and colleagues, where the authors analysed transitions from awake state to sleep state and back, by analysing the regional perturbations that drive the brain dynamics to the sleep state from awake state and vice versa (Deco et al., 2019 PNAS). Taking this approach further, we introduce “dynamic sensitivity analysis”, an approach that quantifies transitions between brain states in terms of regional ability to rebalance spatio-temporal brain activity. In practice, it means building a whole-brain model based on dynamic functional connectivity, and, by stimulating brain regions, assessing the impact on the brain dynamics (Vohryzek et al., 2023 Computational and Structural Biotechnology Journal). To demonstrate the rationale, we show how this approach can be valuable for enhancing cognitive intervention in Alzheimer’s disease in paradigms using non-invasive transcranial electrical stimulation (tES).

One of the potential and promising adjuvant therapies for Alzheimer’s disease is that of tES to potentiate cognitive training interventions. Conceptually, this is achieved by driving brain dynamics towards an optimal state for an effective facilitation of cognitive training interventions. We used whole-brain models to design non-invasive stimulation strategies for re-establishing healthy brain dynamics, and to facilitate effective cognitive interventions in Alzheimer’s disease. We quantified the empirical differences between the healthy (HC, N = 58), Mild Cognitive Impairment (MCI N = 19) and Alzheimer’s Disease (AD, N = 16) groups in terms of changes to their dynamic functional connectivity profiles, and fitted subject-specific whole-brain models to the altered brain states of MCI and AD. Using an in-silico stimulation protocol, we were able to rank brain regions according to their proclivity to drive an increase in patient’s brain dynamics detected in MCI and AD brain states towards that of healthy controls. These were mainly along the brain’s medial axis both in the anterior and posterior parts. The required intensity of stimulation for successful outcome was lower for the MCI brain state suggesting less extensive protocols compared to the AD group. Crucially, the proclivity of a region to rebalance the dynamics to healthier brain state was correlated with the structural nodal degree. In the future, the selected brain regions informed by the in-silico models, can be used for an electrode placement optimization of tES interventions in clinical context. 

To support the diversity of human cognitive functions, brain regions flexibly form and dissolve coalitions on the fly. How is the brain’s capacity to transition between different functional configurations shaped by brain network architecture? This talk will argue that a powerful approach is to use engineering principles of network control to simulate transitions between behaviourally-derived brain states, guided by the anatomical connectivity of the human brain as reconstructed from diffusion tractography. We identified >100 cognitively relevant brain states in a data-driven manner, corresponding to meta-analytic activation patterns aggregated over 14,000 fMRI studies from the NeuroSynth database. We discovered that the network architecture of the human connectome enables transitions between brain states at lower energetic cost than alternative wiring schemes - even after accounting for geometric constraints. These computational predictions were then validated with in-scanner task performance. We then systematically modelled how transitions could be impacted by changes in cortical thickness associated with 11 neurological, psychiatric and neurodevelopmental disorders from 17,000 patients in the ENIGMA database. We found systematic relationships between cortical abnormality and transitions towards brain states supporting memory and language, providing a mechanistic link between anatomical changes and cognitive symptoms. Finally, we leveraged the largest available database of neurotransmitter receptor expression in the human brain in vivo (18 receptor and transporter maps from >1,200 PET scans) to predict the effects of pharmacological interventions. Dopamine transporters and D1 receptors emerged as well positioned to facilitate many desired brain transitions - consistent with their engagement by drugs used to treat attention deficit such as modafinil and methylphenidate. Crucially, our model also highlighted other receptors such as Mu and H3 as suitable targets to achieve specific brain states (Luppi et al 2024, Nature Biomedical Engineering).

Up to this point, we lacked a comprehensive ‘look-up table’ charting how brain network organization and chemoarchitecture interact to manifest cognitively relevant brain states. By jointly leveraging large-scale databases of network structure, functional activation and neurotransmitter systems, the present work provides an integrative framework for the systematic exploration of the full range of possible transitions between experimentally defined brain states. This systematic approach allowed us to discover the key role of the brain’s wiring diagram in supporting flexible transitions with high energetic efficiency, and how this efficiency can be disrupted by disease and restored by targeted pharmacology. The predictions generated by our model highlight numerous potential clinical and non-clinical applications. We anticipate that future work will combine different facets of our computational framework to evaluate in silico which potential pharmacological treatments may best address the specific cognitive difficulties associated with a given disorder or brain tissue lesion. Altogether, we established a principled foundation for interrogating transitions between specific configurations of brain activity, and designing interventions that promote selective transitions between cognitive states.