Periodic Reporting for period 4 - NeuroDevo (Spontaneous and sensory-evoked activity shape neural circuits in the developing brain)
Berichtszeitraum: 2023-11-01 bis 2025-08-31
The brain of a developing fetus must build the essential connections needed later in life. A major challenge in neuroscience is understanding how brain circuits transition from weak, imprecise early connections to strong, highly organized ones in adulthood. Initially guided by genetic and molecular mechanisms, this process largely depends on patterned neural activity. In early development, sensory organs are still immature, yet neural circuits across the brain generate their own spontaneous activity. Experiments show that this activity is crucial for refining connectivity, and disruptions can contribute to disorders such as autism or schizophrenia.
Despite technological progress, we still know little about the structure of spontaneous activity and how it guides the formation of neural circuits. This project combined data analysis, theory, and computational modeling to understand how spontaneous activity is generated, how it changes over time, and how it shapes the development of circuits in the sensory cortex. This approach allowed us to test biological assumptions, link models to data, and propose experimentally testable mechanisms. Understanding these mechanisms is also important for improving the diagnosis and treatment of developmental brain disorders.
Our work showed that the brain’s early spontaneous activity is not random “noise,” but contains meaningful patterns that help circuits wire appropriately. We found that different kinds of neurons switch roles as the brain matures: certain inhibitory neurons help synchronize early activity, while others take over later to support the flexible, adult-like patterns needed for computation. We also discovered that small activity events from the sensory periphery help set up accurate maps of the outside world, while larger activity events generated by the network help stabilize and refine connections. At the subcellular level, we showed how these early patterns cluster related inputs together and explain differences across species. At the larger scale, we demonstrated that coordinated spontaneous activity between sensory areas helps build the foundations for multisensory processing. Finally, we developed new learning rules showing how developing brain circuits can naturally organize into stable networks with diverse functions and recover from sensory disruptions. Overall, the work demonstrated how the brain uses its own early activity to build the circuits that later support sensation, learning, and behavior.
Despite technological progress, we still know little about the structure of spontaneous activity and how it guides the formation of neural circuits. This project combined data analysis, theory, and computational modeling to understand how spontaneous activity is generated, how it changes over time, and how it shapes the development of circuits in the sensory cortex. This approach allowed us to test biological assumptions, link models to data, and propose experimentally testable mechanisms. Understanding these mechanisms is also important for improving the diagnosis and treatment of developmental brain disorders.
Our work showed that the brain’s early spontaneous activity is not random “noise,” but contains meaningful patterns that help circuits wire appropriately. We found that different kinds of neurons switch roles as the brain matures: certain inhibitory neurons help synchronize early activity, while others take over later to support the flexible, adult-like patterns needed for computation. We also discovered that small activity events from the sensory periphery help set up accurate maps of the outside world, while larger activity events generated by the network help stabilize and refine connections. At the subcellular level, we showed how these early patterns cluster related inputs together and explain differences across species. At the larger scale, we demonstrated that coordinated spontaneous activity between sensory areas helps build the foundations for multisensory processing. Finally, we developed new learning rules showing how developing brain circuits can naturally organize into stable networks with diverse functions and recover from sensory disruptions. Overall, the work demonstrated how the brain uses its own early activity to build the circuits that later support sensation, learning, and behavior.
The work during the project led to several publications, preprints, and presentations at major conferences. In Aim 1, we developed multilayer circuit models constrained by experimental data to understand how spontaneous activity is generated in the sensory cortex. Our work showed that specific GABAergic inhibitory neurons, first deep-layer and later more local interneurons of different genetically identified subtypes, act in sequence to synchronize early spontaneous events, strengthen recurrent connections, and ultimately enable the transition from early global waves to the desynchronized activity seen in mature cortex. These results, presented at the Gordon Research Conference and the Bernstein Workshop, are currently being prepared for publication.
In Aim 2, we analyzed calcium imaging data from developing visual cortex and identified two core spontaneous activity patterns: low-amplitude retinal events and high-amplitude cortical events. We built multiscale models to study their developmental roles. In Wosniack et al., (2021), we showed that retinal events refine nearest-neighbor connectivity, while cortical events regulate synaptic strengths. In Kirchner & Gjorgjieva (2021), we demonstrated how spontaneous activity organizes synapses into functional clusters and explained species-specific differences in dendritic organization. Additional work (Dwulet et al., 2024) showed how coordinated spontaneous activity across sensory areas supports early multisensory circuit formation.
In Aim 3, we investigated how circuits reorganize with sensory experience. In Eckmann et al., (2024), we developed a biologically grounded learning framework showing how excitatory and inhibitory plasticity together self-organize recurrent networks into stable, functional architectures that reproduce core computations of sensory cortex. In related work (Montangie et al., 2020; Festa et al., 2024), we developed new mathematical tools to analyze plasticity in recurrent networks. In Wu et al. (2020), we used these rules to explain how firing rates and correlations recover after sensory deprivation.
Across the project, results were disseminated through peer-reviewed publications, preprints, invited talks at COSYNE, the Bernstein Conference, multiple Gordon Research Conferences, and public outreach events such as TEDxTUM. The models, code, and analytical tools developed here are openly shared and already used by other research groups, supporting further exploitation of the findings.
In Aim 2, we analyzed calcium imaging data from developing visual cortex and identified two core spontaneous activity patterns: low-amplitude retinal events and high-amplitude cortical events. We built multiscale models to study their developmental roles. In Wosniack et al., (2021), we showed that retinal events refine nearest-neighbor connectivity, while cortical events regulate synaptic strengths. In Kirchner & Gjorgjieva (2021), we demonstrated how spontaneous activity organizes synapses into functional clusters and explained species-specific differences in dendritic organization. Additional work (Dwulet et al., 2024) showed how coordinated spontaneous activity across sensory areas supports early multisensory circuit formation.
In Aim 3, we investigated how circuits reorganize with sensory experience. In Eckmann et al., (2024), we developed a biologically grounded learning framework showing how excitatory and inhibitory plasticity together self-organize recurrent networks into stable, functional architectures that reproduce core computations of sensory cortex. In related work (Montangie et al., 2020; Festa et al., 2024), we developed new mathematical tools to analyze plasticity in recurrent networks. In Wu et al. (2020), we used these rules to explain how firing rates and correlations recover after sensory deprivation.
Across the project, results were disseminated through peer-reviewed publications, preprints, invited talks at COSYNE, the Bernstein Conference, multiple Gordon Research Conferences, and public outreach events such as TEDxTUM. The models, code, and analytical tools developed here are openly shared and already used by other research groups, supporting further exploitation of the findings.
We achieved several advances beyond the state of the art by combining experimental data with theory and computational models to explain how spontaneous activity shapes developing neural circuits across scales.
For Aim 1, we developed network models that generate spontaneous activity through intrinsic neuronal properties and immature inhibition, rather than imposing prerecorded activity. We built multilayer models where deep-layer inhibitory interneurons initially synchronize activity and later hand over to a different class of interneurons as activity becomes desynchronized, characteristic of mature cortex. These models reproduced developmental transitions observed experimentally and explained how spontaneous activity and connectivity co-emerge.
For Aim 2, we linked two processes previously studied separately: connectivity refinement and changes in spontaneous activity (Wosniack et al., 2021). Our model predicted that spontaneous activity should sparsify during development, with large events becoming less frequent, which we confirmed by analyzing experimental data. At the subcellular scale, in Kirchner & Gjorgjieva (2021), we built a unified theoretical framework explaining how synaptic clustering emerges on dendritic branches based on spontaneous activity. The model reconciled differing findings across species and generated predictions for species where these properties have not yet been measured.
For Aim 3, we advanced understanding of how circuits continue to develop after sensory experience onset, and in particular how they recover from sensory perturbations. Using plastic spiking network models in Wu et al. (2020), we showed that different homeostatic mechanisms, such as synaptic scaling and intrinsic plasticity, have distinct and complementary roles in restoring normal activity patterns. This work provided a mechanistic explanation of how cortical circuits regain function after disruption. In Aim 3, further progress was achieved through new learning rules for recurrent networks (Eckmann et al., 2024) and new mathematical tools (Montangie et al., 2020; Festa et al., 2024) that explain how circuits self-organize under correlated activity.
Overall, the expected results have been fully achieved: we developed models that both produce spontaneous activity and use it to refine connectivity, linked activity patterns across scales, and established a comprehensive mechanistic framework for early brain development and recovery.
For Aim 1, we developed network models that generate spontaneous activity through intrinsic neuronal properties and immature inhibition, rather than imposing prerecorded activity. We built multilayer models where deep-layer inhibitory interneurons initially synchronize activity and later hand over to a different class of interneurons as activity becomes desynchronized, characteristic of mature cortex. These models reproduced developmental transitions observed experimentally and explained how spontaneous activity and connectivity co-emerge.
For Aim 2, we linked two processes previously studied separately: connectivity refinement and changes in spontaneous activity (Wosniack et al., 2021). Our model predicted that spontaneous activity should sparsify during development, with large events becoming less frequent, which we confirmed by analyzing experimental data. At the subcellular scale, in Kirchner & Gjorgjieva (2021), we built a unified theoretical framework explaining how synaptic clustering emerges on dendritic branches based on spontaneous activity. The model reconciled differing findings across species and generated predictions for species where these properties have not yet been measured.
For Aim 3, we advanced understanding of how circuits continue to develop after sensory experience onset, and in particular how they recover from sensory perturbations. Using plastic spiking network models in Wu et al. (2020), we showed that different homeostatic mechanisms, such as synaptic scaling and intrinsic plasticity, have distinct and complementary roles in restoring normal activity patterns. This work provided a mechanistic explanation of how cortical circuits regain function after disruption. In Aim 3, further progress was achieved through new learning rules for recurrent networks (Eckmann et al., 2024) and new mathematical tools (Montangie et al., 2020; Festa et al., 2024) that explain how circuits self-organize under correlated activity.
Overall, the expected results have been fully achieved: we developed models that both produce spontaneous activity and use it to refine connectivity, linked activity patterns across scales, and established a comprehensive mechanistic framework for early brain development and recovery.