Periodic Reporting for period 1 - GradStim (Modeling the perturbational gradients of the human brain)
Período documentado: 2023-06-01 hasta 2025-05-31
Resumen del contexto y de los objetivos generales del proyecto
Context and Overall Objectives
Understanding consciousness, what it is, how it arises, and how it fades, is one of the great scientific and philosophical challenges of our time. While we typically think of the brain as being organized into specialized regions that perform different functions, recent neuroscience has revealed that brain activity is better understood as flowing along continuous gradients: large-scale patterns that smoothly transition from sensory processing areas to regions involved in abstract thought and self-awareness. These “neural gradients” provide a powerful framework for investigating how different states of consciousness such as sleep, anesthesia, or the effects of certain drugs, alter the way the brain functions globally.
This project aimed to study how these gradients change as we move from wakefulness into different unconscious states. More specifically, the project investigated whether the first principal brain gradient, a marker of large-scale cortical organization, contracts systematically as people fall asleep, become sedated, or undergo anesthesia. Such a contraction would indicate that the brain becomes less globally integrated, potentially reflecting both reduced awareness and lower responsiveness to external stimuli.
In addition to studying empirical brain data derived form advanced neuroimaging techniques, the project also developed computational brain models that simulate how these gradients might emerge from the brain’s underlying structure and chemistry. These models were used not only to replicate the patterns seen in real human brain scans but also to test whether “virtual stimulation” could push the brain from one state (e.g. deep sleep) toward another (e.g. light sleep or wakefulness).
Approach and Main Achievements
The project combined three main lines of investigation. First, using functional MRI data from healthy volunteers and patients, we analyzed how the brain’s large-scale organization changes across different states of consciousness, including light and deep sleep, sedation with propofol, and full anesthesia. The data revealed a progressive contraction of the first functional gradient, with each step down in consciousness corresponding to a more localized and fragmented brain network. The ordering of states: light sleep, moderate sleep, sedation, deep sleep, and anesthesia; consistently followed the degree of gradient contraction, revealing a neural “scale” of consciousness. Second, to understand how these gradients arise, we simulated brain activity using mathematical models based on the Hopf bifurcation framework, grounded in principles from physics and neuroscience. These models incorporated anatomical data (structural connectivity) and neurochemical information (such as neurotransmitter receptor densities) to simulate brain dynamics under varying conditions. After optimization, the models closely matched empirical data and successfully reproduced realistic brain gradients. Testing various biological priors further showed that features like resting-state network maps improved simulation accuracy. Third, we explored whether simulated stimulation could shift brain states toward greater consciousness. We applied periodic perturbations to models optimized for low-consciousness states and assessed whether the resulting gradient signatures became more similar to those of more conscious states. In many cases, stimulation induced a meaningful shift toward the desired target, suggesting that brain-wide gradients are not only passive markers of consciousness but may also be useful tools for exploring how consciousness could be modulated through virtual interventions. This opens up new avenues for research into computational neuromodulation and potential applications in disorders of consciousness.
Expected Impact and Significance
This project bridges a crucial gap between neuroimaging, computational modeling, and clinical neuroscience. By identifying the contraction of the brain’s principal functional gradient as a reliable marker of decreasing levels of consciousness, it provides a robust, quantifiable method for monitoring brain states. This has direct implications for fields such as anesthesia monitoring, where assessing the depth of unconsciousness is critical, and sleep research, where transitions between different sleep stages can now be studied through a gradient-based lens. It also holds promise for improving the diagnosis and assessment of patients with disorders of consciousness, such as those in minimally conscious or unresponsive states, where objective biomarkers are urgently needed.
In parallel, the successful reproduction of these gradients using computational brain models demonstrates that large-scale neural dynamics can be predicted, and potentially manipulated, using simulations informed by anatomical and neurochemical data. This not only validates the utility of whole-brain modeling in understanding consciousness but also lays the groundwork for future in silico neuroscience, where hypotheses can be explored through virtual experiments before being tested in clinical or laboratory settings. The broader impact of this work extends beyond neuroscience: by contributing to a mechanistic understanding of conscious awareness, the findings may inform the development of next-generation artificial intelligence systems modeled after the brain’s dynamic organization. All datasets were anonymized and processed in compliance with ethical standards. The simulation code and analysis tools are being prepared for open-access release, ensuring transparency, reproducibility, and broader adoption by the neuroscience research community.
Understanding consciousness, what it is, how it arises, and how it fades, is one of the great scientific and philosophical challenges of our time. While we typically think of the brain as being organized into specialized regions that perform different functions, recent neuroscience has revealed that brain activity is better understood as flowing along continuous gradients: large-scale patterns that smoothly transition from sensory processing areas to regions involved in abstract thought and self-awareness. These “neural gradients” provide a powerful framework for investigating how different states of consciousness such as sleep, anesthesia, or the effects of certain drugs, alter the way the brain functions globally.
This project aimed to study how these gradients change as we move from wakefulness into different unconscious states. More specifically, the project investigated whether the first principal brain gradient, a marker of large-scale cortical organization, contracts systematically as people fall asleep, become sedated, or undergo anesthesia. Such a contraction would indicate that the brain becomes less globally integrated, potentially reflecting both reduced awareness and lower responsiveness to external stimuli.
In addition to studying empirical brain data derived form advanced neuroimaging techniques, the project also developed computational brain models that simulate how these gradients might emerge from the brain’s underlying structure and chemistry. These models were used not only to replicate the patterns seen in real human brain scans but also to test whether “virtual stimulation” could push the brain from one state (e.g. deep sleep) toward another (e.g. light sleep or wakefulness).
Approach and Main Achievements
The project combined three main lines of investigation. First, using functional MRI data from healthy volunteers and patients, we analyzed how the brain’s large-scale organization changes across different states of consciousness, including light and deep sleep, sedation with propofol, and full anesthesia. The data revealed a progressive contraction of the first functional gradient, with each step down in consciousness corresponding to a more localized and fragmented brain network. The ordering of states: light sleep, moderate sleep, sedation, deep sleep, and anesthesia; consistently followed the degree of gradient contraction, revealing a neural “scale” of consciousness. Second, to understand how these gradients arise, we simulated brain activity using mathematical models based on the Hopf bifurcation framework, grounded in principles from physics and neuroscience. These models incorporated anatomical data (structural connectivity) and neurochemical information (such as neurotransmitter receptor densities) to simulate brain dynamics under varying conditions. After optimization, the models closely matched empirical data and successfully reproduced realistic brain gradients. Testing various biological priors further showed that features like resting-state network maps improved simulation accuracy. Third, we explored whether simulated stimulation could shift brain states toward greater consciousness. We applied periodic perturbations to models optimized for low-consciousness states and assessed whether the resulting gradient signatures became more similar to those of more conscious states. In many cases, stimulation induced a meaningful shift toward the desired target, suggesting that brain-wide gradients are not only passive markers of consciousness but may also be useful tools for exploring how consciousness could be modulated through virtual interventions. This opens up new avenues for research into computational neuromodulation and potential applications in disorders of consciousness.
Expected Impact and Significance
This project bridges a crucial gap between neuroimaging, computational modeling, and clinical neuroscience. By identifying the contraction of the brain’s principal functional gradient as a reliable marker of decreasing levels of consciousness, it provides a robust, quantifiable method for monitoring brain states. This has direct implications for fields such as anesthesia monitoring, where assessing the depth of unconsciousness is critical, and sleep research, where transitions between different sleep stages can now be studied through a gradient-based lens. It also holds promise for improving the diagnosis and assessment of patients with disorders of consciousness, such as those in minimally conscious or unresponsive states, where objective biomarkers are urgently needed.
In parallel, the successful reproduction of these gradients using computational brain models demonstrates that large-scale neural dynamics can be predicted, and potentially manipulated, using simulations informed by anatomical and neurochemical data. This not only validates the utility of whole-brain modeling in understanding consciousness but also lays the groundwork for future in silico neuroscience, where hypotheses can be explored through virtual experiments before being tested in clinical or laboratory settings. The broader impact of this work extends beyond neuroscience: by contributing to a mechanistic understanding of conscious awareness, the findings may inform the development of next-generation artificial intelligence systems modeled after the brain’s dynamic organization. All datasets were anonymized and processed in compliance with ethical standards. The simulation code and analysis tools are being prepared for open-access release, ensuring transparency, reproducibility, and broader adoption by the neuroscience research community.
Trabajo realizado desde el comienzo del proyecto hasta el final del período abarcado por el informe y los principales resultados hasta la fecha
During the course of the fellowship, the project pursued a comprehensive approach combining empirical neuroimaging analyses, computational modeling, and perturbational simulations to investigate how brain gradients reflect and shape different states of consciousness. The primary goal was to understand whether and how the brain’s large-scale functional organization, particularly the principal gradient of cortical connectivity, contracts as consciousness diminishes, and whether this behavior could be predicted and manipulated through biologically informed whole-brain models.
Empirically, we analyzed fMRI data from several datasets encompassing a broad range of consciousness states, including wakefulness, light and deep sleep (N1, N2, N3), pharmacologically-induced sedation and anesthesia, and patients with disorders of consciousness. Across all datasets, we consistently observed a progressive contraction of the first functional gradient as participants transitioned into deeper unconsciousness. This contraction (a reduction in the dynamic range of the gradient) was statistically significant and enabled the reliable ordering of conditions from most to least conscious. Notably, this gradient-based continuum interlocked the transitions across sleep and anesthesia in a consistent trajectory: N1, N2, sedation, N3, and anesthesia.
Beyond overall contraction, we examined how the first gradient projected onto canonical resting-state networks (RSNs) across different conditions. In the sleep dataset, we observed differences in the distribution of gradient loadings in N1 and N2 compared to wakefulness, particularly in the auditory and executive control networks. Interestingly, these networks appeared more engaged in early sleep stages, possibly reflecting residual environmental monitoring or transitional dynamics on the threshold of unconsciousness. In the propofol dataset, the gradient distribution changed in a different manner, showing reductions in the executive control and default mode networks in deeper sedation and anesthesia, while auditory network contributions increased in anesthesia. Rather than interpreting these effects in isolation, we emphasize the observed state-dependent redistribution of gradient loading across networks, suggesting shifts in the functional role or integration of specific systems under different conditions.
In addition, we extended our analysis to higher-order gradients (Gradients 2–10) to explore whether reorganization patterns generalized beyond the principal gradient. These gradients displayed diverse trajectories of contraction and expansion across states, highlighting complex, non-monotonic reconfigurations in functional architecture. This suggests that while the first gradient reliably tracks the loss of consciousness, higher gradients may capture orthogonal axes of cognitive disengagement or sensory decoupling specific to particular states.
To better understand the origin and controllability of these gradients, we implemented a supercritical Hopf bifurcation model to simulate resting-state brain activity. The model incorporated individual brain region dynamics, structural connectivity (DTI-based), and biological priors derived from resting-state networks and neurotransmitter receptor density maps. We began by optimizing the global coupling parameter (G) through repeated simulations with fixed and homogeneously varying bifurcation parameters (a), and evaluated the alignment of simulated gradients with different empirical references. Once a robust baseline was established, we introduced spatial heterogeneity in the bifurcation parameter based on biologically plausible priors such as neurotransmitter receptor maps and RSN templates and applied a genetic algorithm for optimization. This yielded improved reproduction of empirical gradient structure, particularly when prior maps reflected known functional organization.
A key innovation of the project was testing whether simulated stimulation could induce transitions between modeled brain states. Using the optimized models, we applied in silico periodic perturbations and measured whether the resulting gradient signatures approached those of more conscious target states. Transitions were consistently successful from sleep states: from N1, the model could reach wakefulness (W), N2, sedation (S), and anesthesia (A); from N2, transitions to W, N3, S, and A were also achievable; and from N3, transitions to W, N2, S, and A were again possible. In contrast, transitions from pharmacologically induced unconsciousness were far more constrained. From sedation, only transitions to N3 and anesthesia could be induced, and from anesthesia, only N3 was reachable. No upward transitions toward conscious states were possible from anesthesia, indicating that these states are more stable and resistant to perturbation. This asymmetry highlights an important functional difference between physiological and pharmacologically induced unconsciousness, with potential implications for clinical monitoring and intervention.
Altogether, the project met and exceeded its main scientific objectives: (1) it demonstrated the contraction of the first functional gradient as a robust and generalizable marker of reduced consciousness; (2) it revealed state-specific reconfigurations in RSN distribution and higher-order gradients; (3) it successfully reproduced empirical gradient structures using neurobiologically grounded whole-brain models; and (4) it established a novel simulation platform to explore in silico transitions between brain states, offering a powerful framework to probe the mechanisms and controllability of consciousness.
Empirically, we analyzed fMRI data from several datasets encompassing a broad range of consciousness states, including wakefulness, light and deep sleep (N1, N2, N3), pharmacologically-induced sedation and anesthesia, and patients with disorders of consciousness. Across all datasets, we consistently observed a progressive contraction of the first functional gradient as participants transitioned into deeper unconsciousness. This contraction (a reduction in the dynamic range of the gradient) was statistically significant and enabled the reliable ordering of conditions from most to least conscious. Notably, this gradient-based continuum interlocked the transitions across sleep and anesthesia in a consistent trajectory: N1, N2, sedation, N3, and anesthesia.
Beyond overall contraction, we examined how the first gradient projected onto canonical resting-state networks (RSNs) across different conditions. In the sleep dataset, we observed differences in the distribution of gradient loadings in N1 and N2 compared to wakefulness, particularly in the auditory and executive control networks. Interestingly, these networks appeared more engaged in early sleep stages, possibly reflecting residual environmental monitoring or transitional dynamics on the threshold of unconsciousness. In the propofol dataset, the gradient distribution changed in a different manner, showing reductions in the executive control and default mode networks in deeper sedation and anesthesia, while auditory network contributions increased in anesthesia. Rather than interpreting these effects in isolation, we emphasize the observed state-dependent redistribution of gradient loading across networks, suggesting shifts in the functional role or integration of specific systems under different conditions.
In addition, we extended our analysis to higher-order gradients (Gradients 2–10) to explore whether reorganization patterns generalized beyond the principal gradient. These gradients displayed diverse trajectories of contraction and expansion across states, highlighting complex, non-monotonic reconfigurations in functional architecture. This suggests that while the first gradient reliably tracks the loss of consciousness, higher gradients may capture orthogonal axes of cognitive disengagement or sensory decoupling specific to particular states.
To better understand the origin and controllability of these gradients, we implemented a supercritical Hopf bifurcation model to simulate resting-state brain activity. The model incorporated individual brain region dynamics, structural connectivity (DTI-based), and biological priors derived from resting-state networks and neurotransmitter receptor density maps. We began by optimizing the global coupling parameter (G) through repeated simulations with fixed and homogeneously varying bifurcation parameters (a), and evaluated the alignment of simulated gradients with different empirical references. Once a robust baseline was established, we introduced spatial heterogeneity in the bifurcation parameter based on biologically plausible priors such as neurotransmitter receptor maps and RSN templates and applied a genetic algorithm for optimization. This yielded improved reproduction of empirical gradient structure, particularly when prior maps reflected known functional organization.
A key innovation of the project was testing whether simulated stimulation could induce transitions between modeled brain states. Using the optimized models, we applied in silico periodic perturbations and measured whether the resulting gradient signatures approached those of more conscious target states. Transitions were consistently successful from sleep states: from N1, the model could reach wakefulness (W), N2, sedation (S), and anesthesia (A); from N2, transitions to W, N3, S, and A were also achievable; and from N3, transitions to W, N2, S, and A were again possible. In contrast, transitions from pharmacologically induced unconsciousness were far more constrained. From sedation, only transitions to N3 and anesthesia could be induced, and from anesthesia, only N3 was reachable. No upward transitions toward conscious states were possible from anesthesia, indicating that these states are more stable and resistant to perturbation. This asymmetry highlights an important functional difference between physiological and pharmacologically induced unconsciousness, with potential implications for clinical monitoring and intervention.
Altogether, the project met and exceeded its main scientific objectives: (1) it demonstrated the contraction of the first functional gradient as a robust and generalizable marker of reduced consciousness; (2) it revealed state-specific reconfigurations in RSN distribution and higher-order gradients; (3) it successfully reproduced empirical gradient structures using neurobiologically grounded whole-brain models; and (4) it established a novel simulation platform to explore in silico transitions between brain states, offering a powerful framework to probe the mechanisms and controllability of consciousness.
Avances que van más allá del estado de la técnica e impacto potencial esperado (incluida la repercusión socioeconómica y las implicaciones sociales más amplias del proyecto hasta la fecha)
The project produced several results that go beyond the current state of the art in systems neuroscience and consciousness research. While the contraction of the brain’s principal functional gradient had previously been reported in various unconscious states such as sleep, anesthesia, or disorders of consciousness; these findings were typically studied in isolation. In this project, we demonstrated for the first time a generalized, robust, and systematic contraction of the first gradient across a broad range of unconscious states of different origins, including physiological sleep, pharmacological sedation, and clinical conditions. This contraction was not only consistent but followed a remarkably clear and reproducible order, interlocking all conditions into a unified, continuous axis of decreasing consciousness. The resulting gradient trajectory can be interpreted as a neural correlate of an “arrow of consciousness,” providing a powerful conceptual and quantitative framework for tracking the depth and stability of different brain states.
Additionally, the project revealed that state-specific reorganization of resting-state networks (RSNs) occurs along this principal gradient. In early sleep stages, changes were observed in the executive control and auditory networks compared to wakefulness, while anesthesia showed marked redistribution involving the default mode, executive, and auditory systems. These differences reflect state-dependent reweighting of sensory and cognitive systems across the gradient dimension. Further analysis of higher-order gradients (2–10) revealed non-monotonic and diverse patterns of reorganization across conditions, indicating that functional connectivity reconfigures along multiple axes (not just the principal one) during transitions in consciousness. These findings provide a multidimensional view of how brain architecture changes in altered states, going beyond static or modular descriptions typically used in clinical or cognitive research.
The project also advanced the state of the art in modeling neural dynamics by developing and validating a biologically grounded whole-brain simulation framework. Using the supercritical Hopf model, we reproduced empirical gradients by tuning global coupling parameters and spatially varying bifurcation parameters using biological priors, including receptor density maps and resting-state network distributions. Simulated gradients closely matched empirical data, confirming that observed cortical organization can emerge from underlying anatomical and neurochemical heterogeneity. This is, to our knowledge, the first successful demonstration of gradient emergence from biologically constrained whole-brain models, bridging structural and functional domains.
A major innovation was the use of these models to explore the perturbational dynamics of consciousness. By applying simulated periodic stimulation to different nodes, we tested whether one unconscious state could be driven toward another. Transitions were consistently achievable between sleep states and even from sleep to anesthesia, but anesthesia-based states were significantly more resistant to perturbation, showing minimal capacity to revert to wakefulness. This asymmetric controllability provides new evidence that pharmacologically induced unconsciousness is more locked-in and stable than natural sleep, supporting its clinical characterization and raising important implications for therapeutic strategies in disorders of consciousness.
To ensure further uptake and success of these findings, several developments are needed. Future research and validation across larger datasets and clinical populations will be crucial, particularly in disorders of consciousness and anesthesia. Translational efforts could focus on developing clinical monitoring tools based on gradient metrics, potentially using fMRI or EEG approximations. Open-access release of the simulation code and analysis pipelines will help enable broader research use and reproducibility. Extending the framework to include pharmacological modeling will allow for the exploration of receptor-targeted interventions, while collaboration with machine learning and clinical communities may support predictive modeling, brain state classification, and personalized therapeutic strategies.
Altogether, this project not only advances the fundamental understanding of consciousness and brain organization, but also delivers actionable tools and conceptual frameworks that pave the way for translational applications in neuroscience, anesthesiology, and computational psychiatry.
Additionally, the project revealed that state-specific reorganization of resting-state networks (RSNs) occurs along this principal gradient. In early sleep stages, changes were observed in the executive control and auditory networks compared to wakefulness, while anesthesia showed marked redistribution involving the default mode, executive, and auditory systems. These differences reflect state-dependent reweighting of sensory and cognitive systems across the gradient dimension. Further analysis of higher-order gradients (2–10) revealed non-monotonic and diverse patterns of reorganization across conditions, indicating that functional connectivity reconfigures along multiple axes (not just the principal one) during transitions in consciousness. These findings provide a multidimensional view of how brain architecture changes in altered states, going beyond static or modular descriptions typically used in clinical or cognitive research.
The project also advanced the state of the art in modeling neural dynamics by developing and validating a biologically grounded whole-brain simulation framework. Using the supercritical Hopf model, we reproduced empirical gradients by tuning global coupling parameters and spatially varying bifurcation parameters using biological priors, including receptor density maps and resting-state network distributions. Simulated gradients closely matched empirical data, confirming that observed cortical organization can emerge from underlying anatomical and neurochemical heterogeneity. This is, to our knowledge, the first successful demonstration of gradient emergence from biologically constrained whole-brain models, bridging structural and functional domains.
A major innovation was the use of these models to explore the perturbational dynamics of consciousness. By applying simulated periodic stimulation to different nodes, we tested whether one unconscious state could be driven toward another. Transitions were consistently achievable between sleep states and even from sleep to anesthesia, but anesthesia-based states were significantly more resistant to perturbation, showing minimal capacity to revert to wakefulness. This asymmetric controllability provides new evidence that pharmacologically induced unconsciousness is more locked-in and stable than natural sleep, supporting its clinical characterization and raising important implications for therapeutic strategies in disorders of consciousness.
To ensure further uptake and success of these findings, several developments are needed. Future research and validation across larger datasets and clinical populations will be crucial, particularly in disorders of consciousness and anesthesia. Translational efforts could focus on developing clinical monitoring tools based on gradient metrics, potentially using fMRI or EEG approximations. Open-access release of the simulation code and analysis pipelines will help enable broader research use and reproducibility. Extending the framework to include pharmacological modeling will allow for the exploration of receptor-targeted interventions, while collaboration with machine learning and clinical communities may support predictive modeling, brain state classification, and personalized therapeutic strategies.
Altogether, this project not only advances the fundamental understanding of consciousness and brain organization, but also delivers actionable tools and conceptual frameworks that pave the way for translational applications in neuroscience, anesthesiology, and computational psychiatry.