CNS*2025 Florence

Hearing Music: A Shared Geometry Governs the Trade-off Between Reliability and Complexity in the Neural Code

Pauline G. Mouawad∗1,Shievanie Sabesan1, Alinka E. Greasley2, Nicholas A. Lesica1
1The Ear Institute, University College London, London, UK
2School of Music, University of Leeds, Leeds, UK


*Email: p.mouawad@ucl.ac.uk





Introduction
Music is central to human culture, shaping social bonds and emotional well-being. Its unique ability to connect sensory processing with reward, emotion, and statistical learning makes it an ideal tool for studying auditory perception [1]. Previous studies have explored neural responses to speech and to simple musical sounds [2, 3], but the neural coding of complex music remains unexplored. We addressed this gap by analyzing multi-unit activity (MUA) recorded from the inferior colliculus (IC) of normal-hearing (NH) and hearing-impaired (HI) gerbils in response to a range of music types at multiple sound levels. The music types included individual stems (vocals, drums, bass, and other) as well as mixtures in which the stems were combined.
Methods
Using coherence analysis, we assessed how reliably music is encoded in the IC across repeated presentations of stimuli and the degree to which individual stems are distorted when presented in a mixture. To explore neural activity patterns at the network level, we implemented a manifold analysis using PCA. This identified the signal manifold, the subspace where reliable musical information is embedded. To model neural transformations underlying music encoding, we developed a deep neural network (DNN) capable of generating MUA from sound, providing a framework for interpreting how the IC processes music. Finally, to assess the impact of hearing loss, we conducted a comparative analysis for NH and HI at equal sound and sensation levels.
Results
We identified strong nonlinear interactions between stems, affecting both the reliability and geometry of neural coding. The reliability of the responses and the dimensionality of the signal manifold varied widely across music types. With increasing musical complexity, the dimensionality of the signal manifold increased, however the reliability decreased. The leading modes in the signal manifold were reliable and shared across all music types, but as musical complexity increased, new neural modes emerged, though these were increasingly unreliable (Figure 1). Our DNN successfully synthesized MUA from music with high fidelity. After hearing loss, neural coding was strongly distorted at equal sound level, but these distortions were largely corrected at equal sensation level.
Discussion

Music processing in the early auditory pathway involves nonlinear interactions that shape the neural representation in complex ways. The signal manifold contains a fixed set of leading modes that are invariant across music types. As music becomes more complex the manifold is not reconfigured; instead, new, less reliable modes are added. These new modes reflect a fundamental trade-off between fidelity and complexity in the neural code. The fact that suitable amplification restores near-normal neural coding suggests that mild-to-moderate hearing loss primarily affects audibility rather than the brainstem’s capacity to process music.
Figure 1. Complexity and Reliability in the Latent Space
Acknowledgements
Funding for this work was provided by the UK Medical Research Council through grant MR/W019787/1.
References
1. Patrik N Juslin and Daniel Västfjäll. “Emotional responses to music: The need to consider
underlying mechanisms”.https://doi.org/10.1017/S0140525X08005293.
2. Vani G Rajendran et al. “Midbrain adaptation may set the stage for the perception of musical
beat”. In: Proceedings of the Royal Society B: Biological Sciences 284.1866 (2017), p. 20171455.
https://doi.org/10.1098/rspb.2017.1455.
3. Shievanie Sabesan et al. “Large-scale electrophysiology and deep learning reveal
distorted neural signal dynamics after hearing loss”. In: Elife 12 (2023), e85108.
https://doi.org/10.7554/eLife.85108.

Balancing Stability and Flexibility: Dynamical Signatures of Learning in In-Vitro Neuronal Networks

Forough Habibollahi*1, Brett J. Kagan1

1Cortical Labs, Melbourne, Australia


*Email: forough@corticallabs.com

Introduction

CL1 is a novel system which bridges biological intelligence and adaptive neuronal traits by integrating in-vitro neuronal networks with in-silico computational elements using micro-electrode arrays (MEAs) [1]. These cultivated neuronal ensembles demonstrate self-organized, biological adaptive intelligence in dynamic gaming environments via closed-loop stimulation and concurrent recordings. While in-vitro neuronal networks are shown to achieve real-time adaptive learning, the underlying network dynamics enabling this learning remain under explored.


Methods
We investigated pairwise causal relationships between recorded channels using Granger causality analysis [2], reconstructing a connectivity network from statistically significant causal interactions. The most influential/influenced nodes were identified as those with highest outgoing/incoming connections. To explore dynamic properties, we reconstructed the phase space of the spiking time series from all recorded channels using state-space reconstruction [3]. Optimal embedding dimensions were determined by minimizing false nearest neighbors, while time delays were selected by detecting the first local minimum of mutual information across different delays. Recurrence plots were generated from the reconstructed phase spaces to analyze temporal patterns.
Results
We analyzed 45-minute spiking recordings at 25 kHz from 23 neuronal cultures, comprising 111 rest sessions and 133 gameplay sessions. Across both rest and gameplay conditions, we observed distinct dynamic patterns between “influential” and “influenced” nodes. Overall, the gameplay sessions exhibited higher recurrence (RR) and determinism (DET) compared to rest (Fig. 1.a). However, in both conditions, the “influenced” nodes displayed lower RR and more negative Lyapunov exponents—indicative of more ordered behavior that lies farther from the edge of chaos. In contrast, the most influential nodes showed higher RR, reflecting recurrent and cyclic dynamics, and had small negative Lyapunov exponents, consistent with behavior near the edge of chaos (Fig 1.b.).
Discussion
Our findings reveal a functional dichotomy in in-vitro neuronal networks. Influential channels exhibit cyclic behavior near the edge of chaos, marked by high RR and near-zero negative Lyapunov exponents, balancing order and chaos. These “near-chaotic” nodes drive network dynamics, enabling rapid influence and adaptability.

In contrast, influenced channels remain more ordered, with lower recurrence and more negative Lyapunov exponents, suggesting stable responsiveness.
This interplay between near-chaotic drivers and stable receivers enables neuronal cultures to balance robustness with adaptability. By defining how distinct dynamical states interact, our results shed light on coordinated neuronal activity and the role of near-chaotic dynamics in flexible behavior.



Figure 1. a) Comparison of dynamic metrics between rest and gameplay sessions. Bar plots show mean values (±SEM) for recurrence rate (RR), determinism (DET), laminarity (LAM), and Lyapunov exponent across all recorded electrodes. b) Dynamic properties of influential vs. influenced nodes across rest and gameplay conditions.
Acknowledgements
F.H. and B.J.K. are employees of Cortical Labs. B.J.K. is a shareholder of Cortical Labs. B.J.K. holds an interest in patents related to this publication.
References
[1]https://doi.org/10.1016/j.neuron.2022.09.001
[2]https://doi.org/10.2307/1912791
[3]https://doi.org/10.1007/BFb0091924

Representational drift as a correlate of memory consolidation

Denis Alevi*+1,2, Felix Lundt+1, Simone Ciceri1, Kristine Heiney1, Henning Sprekeler1,2,3

1Modelling of Cognitive Processes, Technische Universität Berlin, Berlin, Germany
2Bernstein Center for Computational Neuroscience Berlin, Berlin, Germany
3Science of Intelligence, Research Cluster of Excellence, Berlin, Germany
+Equal contribution
*Email: denis.alevi@tu-berlin.de

Introduction

Neural representations – and their population geometry – often change over time despite stable behavior, a phenomenon termed representational drift [1-4]. It is debated if drift is driven by a random process or if it has a directed component, and if it serves a computational function [5]. Systems memory consolidation is a promising candidate [6], because it predicts a temporal reorganization of neural memory engrams. However, it remains unclear how classical theories of consolidation relate to the population-level view of drift and how apparently unstructured drift could be driven by a directed consolidation process.
Methods
We present a computational model for engram dynamics under memory consolidation and explore the resulting representational drift. Assuming that engram changes are driven by reactivations, the model displays recurrent neural network (RNN)-like dynamics, but evolves on long time scales of memory consolidation. This allows us to reinterpret common dynamical phenomena in RNNs in light of memory consolidation and relate them to experimentally observed drift. In simulation, we study how single cell tuning curves and the geometry of neural representations change over time, when not all neurons are observed and develop analytical results for the effect of subsampling, based on Green’s functions and random matrix theory.
Results
Our model redistributes memory engrams across neural populations while maintaining stable memory recall through null-space dynamics [2]. The model can display power-law forgetting without requiring a diversity of learning rates [7]. Low-rank dynamics induce selective consolidation and semantization. In line with experimental findings on representational drift, individual neurons exhibit diverse tuning changes: stability, gradual drift, and abrupt changes of preferred stimulus. Multi-day decoders [2] reveal invariant subspaces on the full population, but degrade quickly under subsampling. A theoretical analysis shows that the dynamics of subsampled populations can be predominantly driven by the unrecorded population, which generates seemingly noise-driven dynamics.

Discussion
Our phenomenological model of engram dynamics bridges the gap between the area-centered perspective of systems consolidation and the population-level perspective of representational drift. Our results show that despite systematic population dynamics, a recorded subset of the neural population can appear to have unstructured dynamics [2]. Recent evidence for stable geometric structure during representational drift in CA1 [7] is consistent with our model of RNN-like engram dynamics, and we hypothesize that unstable population geometry [3] could also be explained by subsampling. Overall, our model offers a functional interpretation of drift as a means to redistribute engrams for improved memory retention.



Acknowledgements
Kristine Heiney is funded by a Postdoctoral Research Fellowship from the Alexander von Humboldt Foundation.
References
[1]https://doi.org/10.1016/j.cell.2017.07.021
[2]https://doi.org/10.7554/eLife.51121
[3]https://doi.org/10.1038/s41586-021-03628-7
[4]https://doi.org/10.1007/s00422-021-00916-3
[5]https://doi.org/10.1016/j.conb.2022.102609
[6]https://doi.org/10.1371/journal.pcbi.1003146
[7]https://doi.org/10.1101/2025.02.04.636428

Backpropagation through space, time and the brain

Paul Haider*1, Benjamin Ellenberger1, Jakob Jordan1, Kevin Max1, Ismael Jaras1, Laura Kriener1, Federico Benitez1, Mihai A. Petrovici1

1Department of Physiology, University of Bern, Bern, Switzerland

*Email: paul.haider@unibe.ch
Introduction

Effective learning in the brain relies on the adaptation of individual synapses based on their relative contribution to solving a task. However, the challenge of spatio-temporal credit assignment in physical neuronal networks remains largely unsolved due to the biologically implausible assumptions of traditional backpropagation algorithms. This study aims to bridge this gap by proposing a novel framework that efficiently performs credit assignment in real-time, without violating spatio-temporal locality constraints, driven by the need for biological systems to learn continuously and interact with dynamic environments.

Methods
We introduce Generalized Latent Equilibrium (GLE), a computational framework for fully local spatio-temporal credit assignment in physical, dynamical networks of neurons. GLE is based on an energy function of neuron-local mismatch errors, from which neuronal dynamics are derived using stationarity and parameter dynamics using gradient descent principles. This framework leverages the morphology of dendritic trees and the ability of neurons to phase-shift their output rates relative to their input (see, e.g., [1]), enabling complex information processing. Additionally, the adjoint method is employed to demonstrate that our learning rules approximate gradient descent on the total integrated cost over time, effectively approximating backpropagation through time (BPTT).
Results
The resulting neuronal dynamics can be interpreted as a real-time, biologically plausible approximation of backpropagation through space and time, incorporating continuous-time leaky-integrator neuronal dynamics and continuously active, phase-free, local synaptic plasticity. The corresponding equations suggest a direct mapping to cortical microcircuitry, with L2/3 pyramidal error neurons counter-posing L5/6 pyramidal representation neurons in a ladder-like fashion. We demonstrate GLE's effectiveness on both spatial and temporal tasks, such as chaotic time series prediction, MNIST-1D [2], and Google Speech Commands datasets, achieving results competitive with powerful ML architectures like GRUs and TCNs trained with offline BPTT.

Discussion
This framework has significant implications for understanding biological learning processes in neural circuits and designing neuromorphic hardware. GLE is applicable to both spatial and temporal tasks, offering advantages over existing alternatives like BPTT and real-time recurrent learning (RTRL) in terms of efficiency and biological plausibility. The framework's locality and reliance on conventional analog components make it an attractive blueprint for efficient neuromorphic hardware. This study contributes to a deeper understanding of how physical neuronal systems can efficiently learn and process information in real-time, bridging the gap between machine learning and biological neural networks.



Acknowledgements
This work was supported by the European Union, the Volkswagen Foundation, ESKAS, and the Manfred Stärk Foundation. We also acknowledge the Fenix Infrastructure and the Insel Data Science Center for their support.
References
1. Brandt, S., Petrovici, M.A., Senn, W., Wilmes, K.A., & Benitez, F. (2024). Prospective and retrospective coding in cortical neurons. https://arxiv.org/abs/2405.14810
2. Greydanus, S., & Kobak, D. (2020). Scaling Down Deep Learning with MNIST-1D. International Conference on Machine Learning. https://arxiv.org/abs/2011.14439

Competition between memories for reactivation as a mechanism for long-delay credit assignment

Subhadra Mokashe*1, Paul Miller2


1Neuroscience Graduate Program, Brandeis University, Waltham, USA
2Department of Biology, Brandeis University, Waltham, USA


*Email: subhadram@brandeis.edu

Introduction
Animals learn to associate an event with its outcome, as in conditioned taste aversion, when they gain aversion to a conditioned stimulus (CS, recently experienced taste) if sickness is later induced [1]. Overshadowing arises if another intervening taste (interfering stimulus, IS) gains some credit for the causality of the outcome, thereby reducing the aversion to the CS [2]. The known short-term correlational plasticity mechanisms do not wholly explain how networks of neurons achieve long-delay credit assignment. We hypothesize that reactivation of stimuli during sickness causes specific associative learning between those stimuli and the sickness, and the competition between the stimuli for reactivation could explain overshadowing.
Methods
We build a spiking recurrent network model with clustered connectivity for excitatory neurons and unstructured inhibitory feedback. We assume the recurrent strengths are enhanced at the time of stimulus presentation due to Hebbian mechanisms and then decay in time. Given that the IS is introduced after the CS, the IS ensemble has higher recurrent strength than the CS ensemble. When we simulate the network, we see reactivation of both tastes (Fig 1 A). We calculate the fraction of time the network spends reactivating a stimulus as a readout of association with the outcome (sickness). We vary the interstimulus interval by changing the difference in recurrent strengths (Δ) and vary the delay to sickness by varying the recurrent strengths.


Results
When we look at the time spent in each state as we increase Δ, we see that not only the time spent in the IS increases, but the time spent in the CS decreases (Fig. 1 B). We only changed the recurrent strengths of the IS ensemble; the time spent in the CS ensemble was affected, indicating competition between the memories for reactivation and accounts for overshadowing. When the CS to IS interval is held constant, paradoxically, more conditioning to the CS is shown by a later sickness onset than earlier sickness [2]. We can explain the result via greater time spent in the CS state (Fig. 1 D) with an appropriate decay profile of recurrent weights (Fig. 1 C) such that the reduced overshadowing outweighs the reduction in conditioning with increased delay.


Discussion
How actions are associated with delayed outcomes is not well understood. We explore the reactivation of memories as a mechanism for long-delay credit assignment in conditioned taste aversion (CTA). We show that competition between memories for reactivation could explain how credit is assigned when there is ambiguity about the cause of an outcome. We use theoretical predictions to constrain our model and are able to explain experimental findings for overshadowing [2]. This study could explain credit assignment not only in CTA and overshadowing but also in other forms of long-delay learning and provide insights into how credit is assigned when there is ambiguity in the cause of an outcome.



Figure 1. A. Reactivation of the stimuli. B. Fraction of time spent by the network in stimuli states as a function of Δ. C. Time spent in the CS state as a function of the recurrent strength Δ, specific decay profile of the recurrent weights (red line). D. Rebound seen in the time spent in the CS state as a function of delay to the sickness onset only in the presence of the IS (red line).
Acknowledgements

We acknowledge Donald Katz and Hannah Germaine for discussions about the work. We thank NIH, NINDS for funding via R01 NS104818.
References
●
https://doi.org/10.1037/h0029807
●
https://doi.org/10.3758/s13420-016-0246-x

Global brain dynamics modulates local scale-free neuronal activity

Giovanni Rabuffo*1,2, Pietro Bozzo1, Marco Pompili1, Damien Depannemeacker1, Bach Nguyen2, Tomoki Fukai2, Pierpaolo Sorrentino1, Leonardo Dalla Porta3

1 Institut de Neurosciences des Systèmes (INS), Aix Marseille University, Marseille, France
2Okinawa Institute for Science and Technology (OIST), Okinawa, Japan
3Institute of Biomedical Investigations August Pi i Sunyer (IDIBAPS), Systems Neuroscience, Barcelona, Spain

*Email: giovanni.rabuffo@univ-amu.fr

Introduction

The brain's ability to balance stability and flexibility is thought to emerge from operating near a critical state [1]. In this work we address two major gaps of the “brain criticality hypothesis”:
First, local (between neurons) and global (between brain regions) criticality are often investigated independently, and a unifying framework is lacking.
Second, local neuronal populations do not maintain a strictly critical state but rather fluctuate around it [2]. The mechanisms underlying these fluctuations remain unclear.
To bridge these gaps, we introduce a connectome-based model that allows for a simultaneous assessment of local and global criticality (Fig.1). We demonstrate that long-range structural connectivity shapes global critical dynamics and drives the fluctuations of each brain region around a local critical state.
Methods
Decoupled brain regions are described by a mean-field model [3] which exhibits avalanche-like dynamics under stochastic input (Fig.1, Blue). Brain regions are connected via the Allen Mouse Connectome [4], and simulations are performed for different values of the global coupling parameter [5]. Simulated data consists of fast LFP, and slow BOLD signals (Fig.1, Red). The model results are validated against empirical datasets (Fig.1, Gray), including a mouse fMRI dataset [6] and LFP recordings from the Allen Neuropixel dataset [7]. To quantify the fluctuations around criticality, we identified neuronal avalanches as deviations of the local LFP signals below a fixed threshold (Fig.1, Blue) and measured sizes (area under curve) and durations (time to return within threshold). The magnitude of the fluctuations around criticality is assessed by analyzing the variance of the range of avalanche sizes across 2s-long epochs.
Results
For low global coupling, individual brain regions maintains local criticality (Fig.1, Blue) but remains globally desynchronized. Increasing coupling induces spontaneous long-range synchronization, paralleled by local fluctuations around criticality (Fig.1, Red). Notably, the working point where the simulations match the experiments corresponds to the regime with the largest range of avalanches sizes and durations (Fig.1, Grey). Strongly connected regions exhibit greater fluctuations around criticality, a testable prediction of the model. To verify this, we examined Allen Mouse Brain Atlas ROIs with LFP data and found a significant correlation between empirical critical fluctuations and regional structural connectivity properties (Fig.1, Green).
Discussion
Our results, comparing brain simulations and empirical datasets across scales, support the brain criticality hypothesis and suggest that criticality is not a static regime for a local neuronal population, but it is dynamically up- and down- regulated by large-scale interactions.



Figure 1. (Blue) Local neural mass model displays critical-like avalanche dynamics. (Red) Coupling brain regions via the empirical Allen structural connectivity we simulate fast LFP and slow BOLD global dynamics. (Gray) Simulated LFP displays global critical activity and simulated BOLD data matches fMRI experiments. (Green) The fluctuations around criticality correlate with structural in-strength.
Acknowledgements
We thank the Institut de Neurosciences des Systèmes (INS), Marseille, France, and the Okinawa Institute for Science and Technology, Japan for their generous support and sponsorship of this research. Their contributions have been instrumental in advancing our understanding of brain criticality and its implications.

References
[1] O’Byrne, J., & Jerbi, K. (2022) https://doi.org/10.1016/j.tins.2022.08.007
[2] Fontenele, A. J., et al. (2019) https://doi.org/10.1103/physrevlett.122.208101
[3] Buendía, V., et al., (2021) https://doi.org/10.1103/physrevresearch.3.023224
[4] Oh SW, et al. (2014) https://doi.org/10.1038/nature13186
[5] Melozzi F, et al. (2017) https://doi:10.1523/eneuro.0111-17.2017
[6] Grandjean, J., et al. (2023). https://doi.org/10.1038/s41593-023-01286-8
[7] https://allensdk.readthedocs.io/en/latest/visual_coding_neuropixels.html

Acetylcholine Waves and Dopamine Release in the Striatum: A Reaction-Diffusion Mechanism

Lior Matityahu¹, Naomi Gilin¹, Gideon A. Sarpong², Yara Atamna¹, Lior Tiroshi¹, Nicolas X. Tritsch³, Jeffery R. Wickens², Joshua A. Goldberg¹*
¹Department of Medical Neurobiology, Institute of Medical Research Israel-Canada, The Faculty of Medicine, The Hebrew University of Jerusalem, Jerusalem, Israel ²Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan ³Neuroscience Institute, New York University Grossman School of Medicine, New York, NY, USA
*Email: joshua.goldberg2@mail.huji.ac.il

Introduction

Striatal dopamine (DA) encodes reward and exhibits traveling waves across the mediolateral axis during behavior. However, the mechanism generating these patterns remains unknown. Cholinergic interneurons (CINs) modulate DA release through nicotinic acetylcholine receptors (nAChRs) on DA terminals. We hypothesized that reciprocal interactions between CINs and DA axons might underlie wave generation. Here, we investigated whether acetylcholine (ACh) exhibits wave-like activity, whether nAChRs extend DA release spatial scale, and whether a reaction-diffusion framework can explain these waves' emergence from local interactions.

Methods
We imaged ACh sensors (GRAB-ACh3.0, iAChSnFR) in the dorsal striatum of head-fixed mice through cranial windows and GRIN lenses. To test whether nAChRs extend DA release, we expressed GRAB-DA2m in striatal DA axons and measured electrically-evoked DA release at increasing distances with and without the nAChR antagonist mecamylamine. We combined patch-clamp recordings of individual CINs with two-photon imaging of GRAB-DA2m to test if single CINs trigger DA release. We developed and analyzed activator-inhibitor reaction-diffusion models of CIN-DA interactions, exploring how parameters influence wave behavior.

Results
We observed ACh waves propagating primarily lateral-to-medial at velocities of ±10 mm/s. Mecamylamine reduced DA release spatial scale by approximately 50% (from ~532 µm to ~264 µm). Action potentials in individual CINs induced local DA release. We will present novel in vivo data showing that chemogenetic silencing of CINs reduces the spatial scale of ongoing DA release events in awake mice, directly confirming CINs' role in extending DA release. Our modeling demonstrated that CIN-DA interactions form an activator-inhibitor system generating traveling waves. Phase-nullcline-flow analysis (Fig. 1) revealed that wave properties depend on system parameters, explaining directional biases in behavioral contexts.

Discussion
Our findings provide evidence for striatal ACh waves and establish that local CIN-DA fiber interactions drive endogenous traveling waves. The new in vivo data showing CINs extend DA release validates our model's core assumption. The reaction-diffusion framework explains how waves emerge from local axo-axonal interactions without external pacemakers. Our model predicts: strongly coupled DA-ACh waves, nAChR blockade compromising wave propagation, and interneuron activity influencing wave direction. This mechanism contributes to spatiotemporal coding in the striatum, with implications for reward processing, learning, and movement coordination.




Figure 1. Figure 1. Phase-nullcline-flow analysis of the activator-inhibitor model. (a) Nullclines and flow field showing fixed points. (b) The direction of wave propagation depends on the area between nullclines. β values control the coupling strength between CINs and DA axons, determining whether CIN waves advance (β=1.0) or recede (β=1.8).
Acknowledgements
This work was funded by a Research Grant from the Human Frontier Science Program (RGP0062/2019), an ERC Consolidator Grant (646886), and grants from the National Institutes of Health (DP2NS105553 and R01MH130658) and Dana and Whitehall Foundations.

References
[1] Hamid, A. A., et al. (2021). Wave-like dopamine dynamics as a mechanism for spatiotemporal credit assignment. Cell, 184, 2733-2749.
[2] Threlfell, S., et al. (2012). Striatal dopamine release is triggered by synchronized activity in cholinergic interneurons. Neuron, 75, 58-64.
[3] Matityahu, L., et al. (2023). Acetylcholine waves and dopamine release in the striatum. Nature Communications, 14, 6852.
[4] Liu, C., et al. (2022). An action potential initiation mechanism in distal axons for dopamine release control. Science, 375, 1378-1385.

Mathematical insights into the spatial heterogeneity of extracellular serotonin induced by the geometry and dynamics of serotonergic fibers

Merlin Pelz*1, Skirmantas Janusonis2, Gregory Handy1,3

1School of Mathematics, University of Minnesota, Minneapolis, USA
2Department of Psychological and Brain Sciences, University of California, Santa Barbara, USA

*Email: mpelz@umn.edu
Introduction

All vertebrate brains, from fish to humans, contain dense meshworks of axons (fibers) that release serotonin, a key signaling molecule. The role of this massive system is poorly understood, with no analogs in current AI architectures, but it appears to support neuroplasticity. Its effects on neural networks are exerted through serotonin receptors whose activation depends on serotonin molecules in the local extracellular space. Recent studies have revealed a lack of fundamental understanding of the spatiotemporal characteristics of extracellular serotonin [1]. In particular, its concentration may vary greatly within microscopic volumes and over short time frames. Such sustained heterogeneity may be a key feature of the plastic brain.
Methods
To investigate how the geometry of the spatial arrangement of release/reuptake sites (i.e., fiber varicosities [2,3]; Fig. 1(a), (b)) and the timing of release shape serotonin concentrations in microscopic brain volumes, we extend previous work [4] and consider a 2D compartmental-reaction diffusion system that is analytically tractable. Each varicosity is modeled as a small disk where the kinetics of serotonin release and uptake (adapted from [5]) are implemented. The disks interact with the surrounding diffusive space through an infinitely permeable boundary (Fig. 1(c), (d)). This system can be rigorously reduced to an integro-ordinary-differential system that can be numerically solved efficiently.
Results
Our system highlights precise coupling terms across varicosities that capture the diffusive memory dependence and global coupling and can be solved using arbitrary serotonin reaction kinetics at the varicosities. Using biologically realistic parameters, we observe that the serotonin concentration exhibits large temporal and spatial variation near varicosities, while regions farther away stabilize to a concentration that depends on the surrounding varicosity density (Fig. 1(e), (f), (g)). We are currently investigating the dependence of the serotonin concentration on the spatial distribution of varicosities (with fibers forming a regular lattice, fibers as stochastic paths [6], etc.).
Discussion
Neural tissue shows many features of criticality [7]. While some heterogeneities on the microscopic scale are due to noise which is not amplified by the brain, other heterogeneities may be actively maintained to support phase transitions and symmetry-breaking/pattern formation. In particular, it may be important in cortical oscillations, wakefulness-sleep transitions (e.g., no firing in REM sleep), and neuroplasticity (e.g., some psychedelics act on the serotonergic system with long-lasting therapeutic effects for some mental disorders). Further, our work will extend current reaction-diffusion pattern formation theory if nontrivial symmetry-breaking and oscillatory synchronization properties are found in this one-diffusing-species system.



Figure 1. a,b: Serotonergic fibers of a mouse brain with varicosities in dark red and light green (scale bars: 1μm (a), 5μm (b)). c: Mathematical system with well-mixed cyan varicosity neighborhoods and blue diffusing serotonin molecules (concentration). d: Zoomed into a single varicosity neighborhood. e-g: Numerical solutions for different varicosity and thus fiber arrangements (bright ~ high, dark ~ low).
Acknowledgements
-
References
● https://doi.org/10.1111/jnc.15865
● https://doi.org/10.1101/2023.11.25.568688
● https://doi.org/10.3389/fnins.2022.994735
● https://doi.org/10.48550/arXiv.2409.00623
● https://doi.org/10.1016/j.bpj.2021.03.021
● https://doi.org/10.3389/fncom.2023.1189853
● https://doi.org/10.1016/j.tins.2022.08.007

Mechanisms of neurotransmitter driven depolarization in perisynaptic astrocytic processes

Ryo J. Nakatani*1and Erik De Schutter1

1Computational Neuroscience Unit, Okinawa Institute of Science and Technology, Okinawa, Japan

*Email: ryo.nakatani@oist.jp

Introduction


Electrophysiological properties of cells underlie the fundamental mechanisms of the brain. Although astrocytes have typically been considered not electrically excitable, recent studies show depolarization of astrocytes induced by local extracellular potassium changes [1]. Interestingly, astrocytic depolarization is induced within the periphery of cortical somatosensory astrocytes, proposed to be at contact sites between neurons and astrocytes. Astrocytic depolarization is thought to affect the brain’s information processing, as depolarization alters astrocyte neurotransmitter uptake [1, 2]. However, specific mechanisms causing astrocytic depolarization have yet to be confirmed due to the limitations of experimental techniques.

Methods
Therefore, we aimed to construct a computational whole-cell astrocyte model to assess which channels were responsible for astrocyte depolarization. Our model included channels known to depolarize astrocytes, such as Kir 4.1, GLT-1 and GABAAR, and other channels we hypothesized to depolarize the astrocyte such as NMDAR (Fig. 1 top) [1, 3]. The model used a protoplasmic hippocampal astrocyte morphology [4], analogous to a cortical astrocyte, capturing both the soma and fine processes. Our model was also sensitive to extracellular ions, by simulating changes in reversal potential at different locations. This allowed us to create a simplified but accurate astrocyte model, responsive to neuronal activity.
Results
Our simulations show, depolarization by potassium uptake alone was unphysiological, requiring∼20 mM of potassium in physiological channel densities. However, the model reached experimentally observed 20 mV depolarizations in peripheral astrocytes by activating neurotransmitter receptors. Difference in neurotransmitter receptors created different decay dynamics, as well as difference in required channel densities to achieve experimental depolarization amplitudes. Depolarization in our model was mainly driven by the inward current from these receptors, which also induced small outward potassium currents and local increase in extracellular concentration (Fig. 1 bottom). All observed ion/potential changes were spatially confined.

Discussion




We hypothesize the strong attenuation, from high conductance and lack of voltage-dependent sodium channels, are key in isolating responses to local synapses. Moreover, our models show how both excitatory and inhibitory neurotransmitters can contribute to peripheral astrocytic depolarizations, revealing a possible mechanism of how astrocytes control synaptic efficacy through local increase of extracellular potassium (Fig. 1 bottom). Inter-synapse communication via astrocyte may also be possible, with inhibitory neurotransmitter induced depolarization altering diffusion dynamics in adjacent excitatory synapses. These insights suggest new mechanisms of how learning and memory are locally regulated by astrocytic processes.

Figure 1. Figure 1. Top: Schematic of whole-cell astrocyte computational model. Color scale show membrane potential during depolarization. Cartoon depicts channels in the computational model. Bottom: Voltage and currents recorded in PAP. (Left) A comparison of membrane potentials measured in sections marked within morphology. (Right) Individual currents recorded in the PAP for both GABAAR and NMDAR.
Acknowledgements
This research has been funded by OISTGU and by JSPS KAKENHI grant number 24KJ2184.
References
● Armbruster, M. et al. (2022). Neuronal activity drives pathway-specific depolarization of peripheral astrocyte processes.Nature Neuroscience, 25(5).
●
O’Kane, R. L. et al. (1999) Na+-dependent glutamate transporters of the blood-brain barrier: a mechanism for glutamate removal.Journal of Biological Chemistry, 274(45).
●
MacVicar, B. A. et al. (1989) GABA-activated Cl-channels in astrocytes of hippocampal slices.Journal of Neuroscience, 9(10).
●
Savtchenko, L. P. et al.(2018) Disentangling astroglial physiology with a realistic cell modelin silico.Nature communications, 9(1).

The role of gain neuromodulation in layer-5 pyramidal neurons

Alejandro Rodriguez-Garcia*1, Christopher J. Whyte2, Brandon R. Munn2, Jie Mei3,4,5, James M. Shine2, Srikanth Ramaswamy1,6


1Neural Circuits Laboratory, Biosciences Institute, Faculty of Medical Sciences, Newcastle University, Newcastle upon Tyne, United Kingdom
2Brain and Mind Center, The University of Sydney, Sydney, Australia, Center for Complex Systems, The University of Sydney, Sydney, Australia
3IT:U Interdisciplinary Transformation University Austria, Linz, Austria
4International Research Center for Neurointelligence, The University of Tokyo, Tokyo, Japan
5Department of Anatomy, University of Quebec in Trois-Rivieres, Trois-Rivieres, QC, Canada
6Theoretical Sciences Visiting Program (TSVP), Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan

*Email: a.rodriguez-garcia2@newcastle.ac.uk


Introduction
Layer-5 pyramidal neurons exhibit BAC firing, where distal dendritic inputs coincide with somatic backpropagating action potentials (BAPs) to trigger Ca²⁺ spikes, converting isolated spikes into bursts and increasing gain[1]. This mechanism is essential for cognitive functions like attention and perceptual shifts[2, 3]. The ascending arousal system flexibly reconfigures neuronal activity during perceptual shifts while maintaining network stability[4, 5]. Here, we explore the role of gain neuromodulation in learning using a biophysically plausible network of layer-5 pyramidal neurons with dendritic-targeting somatostatin (SOM) and somatic-targeting parvalbumin (PV) interneurons.

Methods
We developed a two-compartment Izhikevich neuron model with separate somatic and apical dendritic compartments. The apical dendritic compartment is a 2D nonlinear system governing Ca²⁺ spike generation[3, 6]. The somatic and dendritic compartments are coupled so that somatic sodium spikes trigger BAPs, while dendritic plateau potentials switch somatic activity from regular spiking to bursting. This shift is achieved by increasing the post-spike reset voltage and reducing the spike adaptation in the somatic compartment. BAP events occur stochastically[7], controlled by a soma‐apical coupling parameter. Neuromodulatory signals modulate apical drive and coupling to adjust somatic gain[8–10]. The model is embedded in a toroidal network geometry that incorporates SOM and PV interneurons. Connectivity follows a Gaussian profile[3, 4], and synapses exhibit plasticity via STDP[11].

Results
Simulations demonstrate that both increased dendritic drive and enhanced somatic-apical coupling effectively elevate the gain of pyramidal neurons, likely due to hysteresis in the apical compartment that generates a transient stable state above the calcium threshold (Fig.1A,B). In contrast, dendritic-targeted inhibition reduces gain, while somatic-targeting inhibition significantly raises the adjacent neurons firing threshold (Fig.1C). Capturing these dynamics at the network level leads to a reconfiguration of activity, as burst-like behavior increases spike frequency and accelerates STDP weight updates, rapidly resetting the network to adapt to changing input streams.

Discussion
Our findings highlight the critical role of neuromodulatory control over pyramidal gain through a biologically-informed framework[12], providing a mechanistic explanation for transitions between flexible and stable network states by evaluating its effects to STDP plasticity, in line with previous studies[2–4]. Dendritic-targeted inhibition reduces gain, while somatic-targeted inhibition raises the firing threshold, following experimental observations[13], providing an inhibitory gating control[14]. Future work will leverage neuromodulatory signals to induce flexible, stable neural processing for adaptive learning in biological and neuromorphic systems.




Figure 1. Study of layer-5 neurons with PV and SOM inhibition. (A) Schematic of the model in isolation. (B) Hysteresis in the apical compartment induced by increasing apical drive. (C) Gain enhancement resulting from soma-apical coupling. (D) Gain reduction achieved through dendritic-targeted inhibition. (E) Elevation of the firing threshold via somatic-targeted inhibition.
Acknowledgements
This work was supported by the Lister Institute Prize Fellowship to S.R.; Newcastle University Academic Track (NUAcT) Fellowship to S.R.; NUAcT PhD studentship to A.R-G. J.M. acknowledges support from the Japan Society for the Promotion of Science (JSPS) and the Japan Science and Technology Agency (JST). J.M.S. was supported by the National Health and Medical Research Council (GNT1193857).


References
●
https://doi.org/10.1093/cercor/bhh065
●
https://doi.org/10.7554/eLife.93191.2
●
https://doi.org/10.1101/2023.07.13.548934
●
https://doi.org/10.1038/s41467-023-42465-2
●
https://doi.org/10.1098/rsfs.2022.0079
●
https://doi.org/10.1073/pnas.1720995115
●
https://doi.org/10.1152/jn.00800.2016
●
https://doi.org/10.2174/157015908785777193
●
https://doi.org/10.1016/j.neuron.2018.11.035
●
https://doi.org/10.1016/j.celrep.2018.03.103
●
https://doi.org/10.1162/neco.2007.19.6.1468
●
https://doi.org/10.48550/arXiv.2407.04525
●
https://doi.org/10.1002/phy2.67
●
https://doi.org/10.1073/pnas.2311885121

Single-cell optogenetic perturbations reveal stimulus-dependent network interactions

Deyue Kong*1, Joe Barreto2,Greg Bond2, Matthias Kaschube1, Benjamin Scholl2

1Frankfurt Institute for Advanced Studies, Frankfurt am Main, Germany
2University of Colorado Anschutz Medical Campus, Department of Physiology and Biophysics, Aurora, Colorado, USA

*Email: kong@fias.uni-frankfurt.de


Introduction
Cortical computations arise through neuronal interactions and their dynamic reconfiguration in response to changing sensory contexts. Cortical interactions are proposed to engage distinct operational regimes that either amplify or suppress particular neuronal networks. A recent study in mouse primary visual cortex (V1) found competitive, suppressive interactions between nearby, similarly-tuned neurons, with exception of highly-correlated neuronal pairs showing facilitatory coupling [1]. It remains unclear whether such feature competition generalizes to cortical circuits with topographic organization, where neighboring neurons within columns exhibit similar tuning to visual features, and distal excitatory axons preferentially target similarly-tuned columns.
Methods
We investigated interactions between excitatory neurons in the ferret V1 and how network interactions depend on stimulus strength (contrast). We recorded the responses of layer 2/3 neurons to drifting gratings of eight directions at two contrast levels using 2-photon calcium imaging, while activating individual excitatory neurons with precise 2-photon optogenetics. We statistically quantified the effect of target photostimulation on neural activity (inferred spike rate) during visual stimulation using a Poisson generalized linear model (GLM). We then used our model to estimate a target’s influence on the surrounding neurons’ activity and their stimulus coding properties.
Results
Our analyses revealed interactions that depended on cortical distance, stimulus properties, and functional similarity between neuron pairs. Influence of photostimulated neurons strongly depended on cortical distance, but overall exhibited net suppression. Suppression was weakest between nearby neurons (<100µm), but was found across large cortical distances. Distant-dependent suppression was reduced when visual stimuli were low contrast. Examining functional-similar neurons, we found that noise correlations between neuron pairs were most predictive of measured interactions, showing a strong shift from amplification to competition: at low contrast, we observed local amplification between noise-correlated excitatory neurons, but increasing contrast led to a predominantly suppressive influence across all distances.
Discussion
Our data support predictions from theoretical models, such as stabilized supralinear networks (SSN), in which networks amplify weak feed-forward input, but sublinearly integrate strong inputs [2,3]. Furthermore, decoding analyses suggest that the contrast-dependent shift from facilitation to suppression correlates with improved decoding accuracy of direction. These findings demonstrate that stimulus contrast dynamically modulates recurrent interactions between excitatory neurons in ferret V1, likely by differentially engaging inhibitory neurons. Such dynamic modulation supports optimal encoding of sensory information within columnar cortices.




Acknowledgements

References
[1] Chettih, SN, Harvey, CD. Single-neuron perturbations reveal feature-specific competition in V1. Nature (2019).doi:10.1038/s41586-019-0997-6
[2] Rubin DB, Van Hooser SD, Miller KD. The stabilized supralinear network: a unifying circuit motif underlying multi-input integration in sensory cortex. Neuron(2015) doi: 10.1016/j.neuron.2014.12.026. PMID: 25611511; PMCID: PMC4344127.
[3] Heeger DJ, Zemlianova KO. A recurrent circuit implements normalization, simulating the dynamics of V1 activity. PNAS(2020). doi: 10.1073/pnas.2005417117. . PMID: 32843341; PMCID: PMC7486719.

Predicting neural responses to intra- and extracranial electric brain stimulation by means of the reciprocity theorem

Torbjørn V. Ness*¹, Christof Koch²,Gaute T. Einevoll¹,³
¹ Department of Physics, Norwegian University of Life Sciences, Ås, Norway
² Allen Institute, Seattle, WA, USA
³ Department of Physics, University of Oslo, Oslo, Norway

*Email: gaute.einevoll@nmbu.no



Introduction
Neural activity can be modulated through electric stimulation (ES), which is extensively used in both science and the clinic, including deep brain stimulation and temporal interference stimulation. While ES is grounded in well-established biophysics, it has proven difficult to gain a solid understanding of ES and its sensitivity to features like location, orientation, different cell types, and the ES frequency-content. This represents a major obstacle to the applications of ES.
Here, we show that the reciprocity theorem (RT) can be applied more broadly than previously recognized [1], offering a whole new perspective on ES which reproduces known features, explains surprising observations, and makes new predictions.



Methods
The effect of ES on different biophysically detailed cell models is simulated with NEURON [2] and LFPy [3]. The ES is treated as a current point source which sets up an extracellular potential that is used as a boundary condition at each cellular compartment. The somatic membrane potential-responseVmis calculated. In the RT-based approach the current is inserted intracellularly in the soma, and the resulting extracellular potentialVecalculated. According to the RT the two approaches should give identical results for passive cell models (Vm=Ve, Fig. 1). For transcranial electric stimulation (tES), we used a detailed head model to estimate membrane potential responses to tES deep in the brain.
Results
In all tested cases the RT-based approach to simulating ES introduces zero error for passive cell models, and below a few percent error for subthreshold active cell models [1].
By leveraging the RT, we show that the effect of ES has a 1/rdecay for nearby neurons and 1/r² for distant neurons. Furthermore, for nearby neurons the ES response is approximately cell-type and frequency-independent, while for distant neurons (e.g., tES), pyramidal neurons are most strongly targeted at low frequencies, and interneurons at high frequencies, but with a less synchronous effect [1]. Finally, tES at conventional safety limits (<4 mA) induces subthreshold potential changes of ~40-200 µV, far below the threshold for direct neuronal firing [1].


Discussion
By applying RT, we provide a framework for understanding neural responses to ES, by leveraging our good understanding of extracellular potentials [4]. Our results indicate that conventional tES primarily affects neural activity via subtle subthreshold effects, suggesting indirect network-level mechanisms such as synchronization or stochastic resonance. The weak frequency dependence of subthreshold responses explains recent experimental findings [5], reinforcing RT as a powerful tool for modeling ES. Future work should incorporate network-level dynamics to assess the broader implications of these findings for neuromodulation and brain stimulation therapies.




Figure 1. Reciprocity theorem in the context of electrical brain stimulation: The somatic membrane potential response to an extracellular current injection at position r (panel A) corresponds to the extracellular potential at location r from the same current source injected into the soma .
Acknowledgements
T.V.N. and G.T.E. received funding from the European Union Horizon 2020 Research and Innovation Programme under Grant Agreement No. 101147319 [EBRAINS 2.0]. C.K. thanks the Allen Institute founder, Paul G. Allen, for his vision, encouragement, and support.
References
[1] Ness et al. (2025) bioRxivhttps://doi.org/10.1101/2024.08.04.603691
[2] The NEURON bookhttps://doi.org/10.1017/CBO9780511541612
[3] Hagen et al. (2018)https://doi.org/10.3389/fninf.2018.00092
[4] Halnes et al. (2024)https://doi.org/10.1017/9781009039826
[5] Lee et al. (2024) Neuronhttps://doi.org/10.1016/j.neuron.2024.05.009

Time precise closed-loop protocols for non-invasive infrared laser neural stimulation

Alicia Garrido-Peña¹,Pablo Sanchez-Martin¹, Irene Elices¹, Rafael. Levi¹, Francisco B. Rodriguez¹, Javier Castilla², Jesus Tornero², Pablo Varona¹
1. Grupo de Neurocomputación Biológica, Departamento de Ingeniería Informática, Escuela Politecnica Superior, Universidad Autónoma de Madrid, Madrid, Spain
2. Center for Clinical Neuroscience, Hospital los Madroños, Brunete, Spain
*Email: alicia.garrido@uam.es

Introduction

In the context of the increasing interest in noninvasive neural stimulation techniques, infrared (IR) laser has shown its potential to achieve an effective and spatially localized stimulation. In our recent publication [1], we demonstrated that it is possible to modulate neural dynamics with a Continuous-Wave (CW) Near IR laser in terms of firing rate and spike waveform. We analyzed the biophysical cause of this effect in a computational model, observing a combined alteration of ionic channels and significant action of temperature change. We also assessed the illumination effect at different stages of the action potential generation with a closed-loop protocol. Here we extend these results and stimulation protocols.

Methods

We used a CW-IR laser focused with a micromanipulator in the ganglia of Lymnaea stagnalis and Carcius maenas, while recording membrane potential intracellularly. For the closed-loop protocols, we employed the RTXI software [2], designing algorithms in modules for spike, and burst prediction, target-driven stimulation and neuronal digital-twin experiments built with conductance-based models. The laser was triggered based on the ongoing activity with a precise electro-optical shutter (range of µs). Our software is open-source and available atgithub.com/GNB-UAM.

Results

We show the effectiveness of laser stimulation with high spatial resolution (by the nature of this technique) and temporal precision (by our real-time closed-loop protocols and an electro-optical shutter) on two different neural systems. For both cases, sustained CW laser illumination accelerates neural activity. We also report on the efficacy of activity-dependent illumination with a time-precise shutter. We extend our spike prediction algorithms to neural bursting activity and a target-driven stimulation. To leverage this closed-loop neurotechnology, we present a biohybrid circuit where a model neuron acts as a digital twin of the recorded cell. This enables real-time automatic decision-making based on the model tested response.

Discussion

CW-IR laser illumination is a novel noninvasive neurotechnology with high temporal and spatial resolution. CW-IR laser effectively modulated neural activity in two different neural systems with distinct closed-loop protocols. These protocols employ a library for time-precise stimulation adaptable to different neural dynamics (e.g. tonic spiking or bursting activity) and different target goals (a specific firing rate, burst duration...). The digital-twin model also enables online adjustment by the combined modification of the stimulation parameters and the simulated neuron. Exploiting the advantages of closed-loop stimulation and real-time tools, expands the possibilities of this neurotechnology to novel research and clinical applications.
AcknowledgementsWork funded by PID2024-155923NB-I00, CPP2023-010818, PID2023-149669NB-I00 and PID2021-122347NB-I00.
References[1] Garrido-Peña, A., Sanchez-Martin, P., Reyes-Sanchez, M., Levi, R., Rodriguez, F. B., Castilla, J., Tornero, J., & Varona, P. (2024). Modulation of neuronal dynamics by sustained and activity-dependent continuous-wave near-infrared laser stimulation. Neurophotonics, 11(2), 024308.doi.org/10.1117/1.NPh.11.2.024308

[2] Patel, Y. A., George, A., Dorval, A. D., White, J. A., Christini, D. J., & Butera, R. J. (2017). Hard real-time closed-loop electrophysiology with the Real-Time eXperiment Interface (RTXI). PLOS Computational Biology, 13(5), e1005430.doi.org/10.1371/journal.pcbi.1005430

LTP-induced changes in spine geometry and actin dynamics on the timescale of the synaptic tag

Mitha Thomas*1, Cristian Alexandru Bogaciu2, Silvio Rizzoli2, Michael Fauth1

1Third Physics Institute, Georg-August University, Goettingen, Germany
2Department of Neuro- and Sensory Physiology, University Medical Center, Goettingen, Germany
*Email: mitha.thomas@phys.uni-goettingen.de

Introduction

Long-term potentiation of synapses can occur in two phases: an early phase which constitutes a transient increase in synaptic strength, and a late phase which sustains this increase for a longer duration. According to the synaptic tagging and capture hypothesis [1,2], a necessary condition for the late phase is the formation of a transient memory of the stimulation event - the ‘synaptic tag’ - which enables the synapse to capture newly synthesized proteins later on. What implements this transient memory on the timescale of hours remains elusive [2,3]. We follow the hypothesis that it is implemented by actin dynamics in interaction with spine geometry and test this using computational modelling and FRAP experiments.

Methods
Actin forms filaments in the spine which belong to distinct pools (dynamic and stable) with different turnover rates. To study the relation of actin pools with synaptic tagging, we derived a computational model of the interactions between the spine membrane and the actin pools undergoing plasticity. Dynamic actin is modelled as a Markov chain that considers several processes related to actin binding proteins, e.g., branching, capping and severing, which are modulated upon LTP [4, 5]. Stable actin is modelled as a low-pass filter of the dynamic pool with filter coefficients following binding and unbinding of crosslinking proteins. The spine membrane deforms according to the balance between the actin-generated force and the forces resulting from the physical properties of the membrane (Fig 1A-C).
Results
We first test whether can support memory on a timescale of hours without stable actin. At the onset of LTP, there is a rapid increase in dynamic actin, which increases the outward-directed force and, consequently, also the spine volume. However, these changes only last as long as the actin dynamics is modulated. When we introduce the stable pool, it exhibits an overshoot that persists on the timescale of hours, and hence, the synaptic tag. As more stable actin significantly increases the actin-generated force, this also transfers to a long-lasting spine volume increases (Fig 1D-H). To validate these model predictions experimentally, we perform chemical LTP on hippocampal spines and use FRAP to assess stable actin content after LTP. Also here, stable actin shows a significant increase agreeing with the overshoot in the model (Fig 1H).
Discussion
Using a combination of experiments and simulations, we have demonstrated that the dynamics of the stable actin pool after LTP-inducing stimulation leads to a long lasting alteration of actin dynamics and spine geometry. These dynamics fulfil the fundamental criteria for the tag, that is synapse specificity, independence from protein synthesis, and decay within hours. Thus, we present evidence the biophysical implementation of the synaptic tag may be based on the complex interaction of actin with spine membrane.



Figure 1. Actin-spine membrane interactions. A-B: membrane deformation from imbalance between actin-generated force and membrane counter force. C: time evolution of actin dynamics with several associated processes/proteins. D: simulated spine at different time instants. t0: time of stimulation. E: Control and LTP spine volumes. F: amount of dynamic actin. G: amount of stable actin. H: stable actin fraction
Acknowledgements
This work was funded by the German Science Foundation under CRC1286 ”Quantitative Synaptology”, projects C03 and A03. We would like to thank Simon Dannenberg, Stefan Klumpp, Jannik Luboeinski, Francesco Negri, Christian Tetzlaff and Florentin Woergoetter for fruitful discussions on the project.
References
1.https://doi.org/10.1038/385533a0
2.https://doi.org/10.1038/nrn2963
3.https://doi.org/10.1002/iub.2261
4.https://doi.org/10.1016/j.neuron.2014.03.021
5.https://doi.org/10.3389/fnsyn.2020.00009

Identifying Dynamic-based Closed-loop Targets for Speech Processing Cochlear Implants

Cynthia Steinhardt*1,Menoua Keshishian2, Kim Stachenfeld1,3, Larry Abbott1
1 Center for Theoretical Neuroscience, Zuckerman Brain Science Institute, Columbia University, New York, New York USA
2 Department of Electrical Engineering, Columbia University, New York, New York USA
3 DeepMind, Google, London, United Kingdom


*Email: cs4248@columbia.edu



Introduction
Since the development of the first cochlear implant (CI) in 1957, over one million people have used these devices to regain hearing. However, CIs have a number of deficits, such as low efficacy in noise, and these deficits remain poorly understood [1]. CI algorithm research has focused on optimizing single-neuron voltage-driven activations in the cochlea, based on low-level auditory modeling but little work has focused on capturing known features of hierarchical speech processing across the brain [2]. We create a model system to investigate how CI-encoded speech affects phoneme and word comprehension, uncovering a dynamics-based signature for potential closed-loop CI applications.
Methods
We trained a DeepSpeech2 [3] model to convert spectrograms to phonemes using CTC Loss. Speech inputs were sourced from the LibriSpeech dataset. Speech was processed via the AB Generic Toolbox [4] to generate electrodograms, creating CI-transformed inputs or directly given to the model to simulate natural hearing. The model, trained on natural spectrograms, was then tested on CI-transformed inputs. Behavioral experiments were performed and compared to human results. We analyzed phoneme processing dynamics, using a distance metric to determine convergence patterns and tested dynamic signatures for feedback control [5].
Results
Our model exhibited human-like increases in phoneme reaction time with CI-transformed inputs and noise; phoneme confusion and word errors mirrored human behavior, as well [5]. Analysis revealed a specific time window per layer where correct phoneme comprehension dynamics converged for all phonemes, with increasing delays deeper in the network. We create a representation distance metric, measured via a Wasserstein metric between dynamics during comprehension and found it correlated (up to 0.78) with behavioral confusion of the model while processing these phonemes in sentences. Using a linear closed-loop controller, we then successfully pushed dynamics toward correct phoneme perception using this converged representation at a target.
Discussion
This study presents a plausible model for speech perception with and without a CI, validated against human data. We identify a dynamic signature predicting comprehension or confusion within 100 ms—a feasible intervention window. We demonstrate its use for closed-loop feedback and find evidence of human EEG evoked responses with similar dynamics [6], suggesting a potential EEG-based CI parameter selection method. We show plausibility here for a new cochlear implant paradigm, instead of mimicking cochlear processing, we determine pulse parameters that drive desired population-level neural representations of speech. This approach may generalize to other neural implants, as we understand those systems better.



Acknowledgements
We thank the Simons Society of Fellows (965377), Gatsby Charitable Trust (GAT3708), Kavli Foundation, and NIH (R01NS110893) for support.
References
● Boisvert, I., et al. (2020). CI outcomes in adults.PLoS One,15(5), e0232421.
● Rødvik, A. K., et al. (2018). CI vowel/consonant ID.J Speech Lang Hear Res,61(4), 1023-1050.
● Amodei, D., et al. (2015). Deep Speech 2.arXiv:1512.02595.
● Jabeim, A. (2024). AB-Generic-Python-Toolbox.GitHub.
● Steinhardt, C. R., et al. (2024). DeepSpeech CI performance.arXiv:2407.20535.
● Finke, M., et al. (2017). Stimulus effects on CI users.Audiol Neurotol,21(5), 305-315.

Automated identification of disease mechanisms in hiPSC-derived neuronal networks using simulation-based inference

Nina Doorn*1, Michel van1,2, Monica Frega3

1Department of Clinical Neurophysiology, University of Twente, Enschede, The Netherlands

2Department of Neurology and Clinical Neurophysiology, Medisch Spectrum Twente, The Netherlands

3Department of Informatics, Bioengineering, Robotics and System Engineering, University of Genova, Italy


*Email: n.doorn-1@utwente.nl


Introduction
Human induced pluripotent stem cells (hiPSCs)-derived neuronal networks on multi-electrode arrays (MEAs) are a powerful tool to study neurological disordersin vitro[1]. The electric activity patterns of these networks differ between healthy and patient-derived neurons, reflecting underlying pathology (Fig. 1A). However, elucidating the underlying molecular mechanisms is challenging and requiresextensive, costly, and hypothesis-driven additional experiments.Biophysical models can link observable network activity to underlying molecular mechanisms by estimating model parameters that simulate the experimental observations. However, parameter estimation in such models is difficult due to stochasticity, non-linearity, and parameter degeneracy.

Methods
Here, we address this challenge using simulation-based inference (SBI), a machine-learning approach that allows efficient statistical inference of biophysical model parameters using only simulations [2]. We apply SBI to our previously validated biophysical model of hiPSC-derived neuronal networks on MEA[3], which includesHodgkin-Huxley-type neurons and detailed synaptic models (Fig. 1B). To train SBI, we simulated 300,000 network configurations, varying key parameters governing synaptic and intrinsic neuronal properties (Fig. 1C). We used a neural density estimator to infer posterior distributions of these model parameters given experimental MEA recordings from healthy, pharmacologically treated, and patient-derived networks (Fig 1D).

Results
SBI accurately inferred ground-truth parameters in synthetic data and successfully identified known disease mechanisms in patient-derived neuronal networks. In networks from patients with the genetic epilepsies Dravet Syndrome and GEFS+, SBI predicted reduced sodium and potassium conductances and increased synaptic depression, which was experimentally verified. InCACNA1Ahaploinsufficient networks, SBI correctly identified impaired connectivity. Additionally, SBI detected drug-induced changes, such as prolonged synaptic depression following Dynasore treatment.
Discussion
SBI enables automated and probabilistic inference of biophysical parameters, offering advantages over traditional parameter estimation methods, which can be time-consuming, lack uncertainty quantification, or cannot deal with parameter degeneracy. Our results show how SBI can be used with biophysical models to identify possible disease mechanisms from patient-derived neuronal data. Ourproposed analysis pipeline enables researchers to extract crucial mechanistic information from MEA measurements in a systematic, cost-effective, and rapid manner, paving the way for targeted experiments and novel insights into disease.






Figure 1. Figure 1. A) The activity of in vitro neuronal networks cultured from hiPSCs of healthy controls and patients is measured using MEAs. B) The computational model with biophysical parameters in blue. C) A Neural density estimator is trained on model simulations. Afterward, experimental data is passed through the estimator to approximate the D) posterior distributions. Adapted from [4].
Acknowledgements
This work was supported by the Netherlands Organisation for Health Research and Development ZonMW; BRAINMODEL PSIDER program 10250022110003 (to M.F.). We thank Eline van Hugte, Marina Hommersom, and Nael Nadif Kasri for providing MEA recordings from patient-derived and genome-editedin vitroneuronal networks.
References
● 1.https://doi.org/10.1016/J.STEMCR.2021.07.001
● 2.https://doi.org/10.7554/ELIFE.56261
● 3.https://doi.org/10.1016/J.STEMCR.2024.09.001
● 4.https://doi.org/10.1101/2024.05.23.595522

Linking hubness, embryonic neurogenesis, transcriptomics and diseases in human brain networks

Ibai Diez*1,2, Fernando Garcia-Moreno*3,4,5, Nayara Carral-Sainz6, Sebastiano Stramaglia7, Alicia Nieto-Reyes8, Mauro D’Amato5,9,10, Jesús Maria Cortes5,11,12,Paolo Bonifazi5,11,13


1Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, MA, USA.
2Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA, USA.
3Achucarro Basque Center for Neuroscience, Scientific Park of the University of the Basque Country (UPV/EHU), Leioa, Spain.
4Department of Neuroscience, Faculty of Medicine and Odontology, UPV/EHU, Barrio Sarriena s/n, Leioa, Bizkaia, Spain.
5IKERBASQUE: The Basque Foundation for Science, Bilbao, Spain.
6Departamento de Ciencias de la Tierra y Física de la Materia Condensada, Facultad de Ciencias, Universidad de Cantabria, Santander, Spain.
7Dipartimento Interateneo di Fisica, Università degli Studi di Bari Aldo Moro, and INFN, Sezione di Bari, Italy.
8Departamento de Matemáticas, Estadística y Computación, Facultad de Ciencias, Universidad de Cantabria, Santander, Spain.
9Department of Medicine and Surgery, LUM University, Casamassima, Italy.
10Gastrointestinal Genetics Lab, CIC bioGUNE - BRTA, Derio, Spain.
11Computational Neuroimaging Lab, Biocruces-Bizkaia Health Research Institute, Barakaldo, Spain.
12Department of Cell Biology and Histology, University of the Basque Country (UPV/EHU), Leioa, Spain.
13Department of Physics, University of Bologna, Italy.


* These authors contributed equally to this work; Corresponding author:paol.bonifazi@gmail.com

Intro. The human brain is organized across multiple spatial scales, where micro-scale circuits integrate into macro-scale networks via long-range connections. Understanding the connectivity rules shaping networks is key to deciphering brain function and the effects of neurological damage. Previous studies have explored brain network maturation, but a link between adult connectivity and the sequential evolutionarily preserved neurogenesis remains unestablished. Inspired by the preferential attachment model in network science shaped by the “rich gets richer” principle 1, this study2hypothesizes that brain network topology follows an "older gets richer" principle, where earlier-developed circuits play central roles in adult connectivity. Our hypothesis extrapolates on the macro-scale level previous evidence that hippocampal hubs are early born GABAergic neurons3-5.
Methods. Brain circuits were categorized by their First neurogenic Time (FirsT), determined from developmental neuromeres. Eighteen macro-circuits (MACs) were identified based on available neurodevelopmental data. Structural and functional brain networks were reconstructed using 7-Tesla dMRI and resting-state fMRI from 184 subjects. Connectivity metrics were assessed at high (2,566 ROIs) and low (18 MACs) resolutions. Eigenvector centrality was calculated for each ROI and MAC, with correlations between FirsT and connectivity patterns. Brain transcriptomic data were mapped to connectivity metrics, and enrichment analysis identified associated biological processes and disease relevance.
Results. Significant correlations between structural connectivity and FirsT supported the "older gets richer" principle, with early-born circuits exhibiting higher structural hubness. In contrast, functional centrality was positively correlated with FirsT, highlighting late-maturing circuits' functional prominence. Connectivity strength was stronger among circuits with similar neurogenic timing, supporting a "preferential age attachment" mechanism. Gene expression analysis revealed correlations with FirsT and connectivity metrics, with enriched pathways linked to neurodevelopment, synaptic function, and neuropsychiatric disorders. Disease-associated genes (e.g., APOE for Alzheimer’s, SCN1A for epilepsy) showed significant enrichment at correlation extremes, suggesting differential genetic influences on brain network organization and pathology susceptibility.

Discussion. The study examines adult brain networks reconstructed from MRI to analyze how early neurogenesis affects structural and functional connectivity. Structural findings confirm that older brain regions act as stronger hubs ("older gets richer"). Functional and structural networks follow a "preferential age attachment" rule, linking neurogenesis timing to network topology. Genetic analysis ties neurodevelopmental disorders to network centrality, highlighting disease-linked transcriptional alterations.



Acknowledgements
We thank M De Pittá, D Papo, D Marinazzo, A Mazzoni, Y Ben-Ari for comments. Funds: ANR by MCIN/AEI/10.13039/501100011033 and “ERDF”; PB by Ikerbasque, the Ministerio Economia, Industria y Competitividad (MICINN, Spain) and Maratoia EITB (grant PID2021-127163NB-I00 and BIO22/ALZ/010/BCB);FGMby Ikerbasque, MICINN(grant PID2021-125156NB-I00) andBasque Gov (grant PIBA_2022_1_0027).
References
1.Barabási, A.-L. & Albert, R. Emergence of Scaling in Random Networks.Science286, 509–512 (1999).
2.Diez I et al, https://doi.org/10.1101/2022.04.01.486541
3.Bonifazi, P.et al.GABAergic Hub Neurons Orchestrate Synchrony in Developing Hippocampal Networks.Science326, 1419–1424 (2009).
4.Picardo, M. A.et al.Pioneer GABA Cells Comprise a Subpopulation of Hub Neurons in the Developing Hippocampus.Neuron71, 695–709 (2011).
5.Bocchio, M.et al.Hippocampal hub neurons maintain distinct connectivity throughout their lifetime.Nat Commun11, 4559 (2020).

Back to the Future: Integrating Event-Based and Network Diffusion Models to Predict Individual Tau Progression in Alzheimer's Disease

Robin Sandell*1, Justin Torok1, Kamalini Ranasinghe1, Srikantan Nagarajan1, Ashish Raj1

1Department of Radiology and Biomedical Imaging, University of California, San Francisco, United States

*Email: robin.sandell@ucsf.edu


Introduction

This paper presents a novel method combining an Event-Based Model (EBM) and a Network Diffusion Model (NDM) to predict individual tau protein progression in Alzheimer's disease. Statistical EBMs can infer longitudinal progression from cross-sectional data but lack mechanistic understanding, while biophysical NDMs provide mechanistic clarity but require data on a longitudinal timescale. Our hybrid approach overcomes these limitations. Using only single-visit data, our model can go back in time to infer initial seeding patterns and predict future progression. Analysis reveals high initial heterogeneity in seeding patterns that converges over time with two main seed archetypes correlating with distinct clinical presentations.



Methods


We analyzed data from 650 patients from the Alzheimer’s Neuroimaging Disease Initiative, including tau-PET, MRI, and cognitive scores. EBM assigned a disease stage to each patient based on their biomarker values, enabling a common timescale across subjects [1,2,]. NDM simulated tau progression on the brain’s structural connectivity with two rate parameters: τ accumulation rate and spread rate[3,4]. We optimized NDM parameters and tau seed pattern to accurately predict each subject’s empirical tau map at their EBM assigned stage. Applications include prediction of individuals’ future tau patterns, analysis of inter-subject heterogeneity over time, and identification of tau seed archetypes through clustering analysis.



Results

Our hybrid model successfully predicted empirical tau using individual tau seed patterns (mean R=0.85) (Fig. 1b). Longitudinal validation confirmed the model's predictive ability (mean R=0.81)(Fig. 1b). Analysis of tau patterns revealed decreasing heterogeneity over disease progression (Fig. 1c). Two primary seed archetypes emerged: focal entorhinal (typical AD) and diffuse temporal (Fig. 1d). The diffuse temporal pattern correlated with earlier disease onset, higher APOE4 carrier frequency, younger age, and faster tau accumulation rates, suggesting a more aggressive disease variant despite similar cognitive impairment levels at diagnosis (Fig. 1e,f).


Discussion

This paper presents a novel hybrid approach combining an Event-Based Model and a Network Diffusion Model to predict individual tau progression in Alzheimer's disease. The method infers initial seeding patterns from single-visit data to forecast future progression. Surprisingly, heterogeneity across subjects was highest at disease onset and decreased over time, suggesting convergence rather than divergence of pathology. Two primary seed archetypes emerged: focal entorhinal (typical AD) and diffuse temporal (associated with earlier onset, higher APOE4 frequency, and faster progression). The hybrid model outperformed prior work while providing mechanistic insights into tau progression that could inform personalized therapeutic strategies[2].








Figure 1. a. Project flow chart. b. Illustration of NDM model fitting and validation for a single patient. c. Distribution of pairwise R correlations between subjects model predicted tau at each stage indicating a process of convergence in tau patterning as disease progresses. d. Emergent tau seed archetypes. e. Demographic variables for each archetypes. f. Total tau over time for each archetypes.
Acknowledgements
We thank ADNI for making their data available to us.

References


● Aksman, L.M., et al. (2021). pySuStaIn: Python implementation of SuStaIn. SoftwareX, 16.https://doi.org/10.1016/j.softx.2021.100811
● Vogel, J.W., et al. (2021). Four trajectories of tau deposition in AD. Nature Medicine, 27(5).https://doi.org/10.1038/s41591-021-01309-6
● Raj, A., et al. (2012). Network diffusion model of disease progression. Neuron, 73(6).https://doi.org/10.1016/j.neuron.2011.12.040
● Anand, C., et al. (2022). Microglia effects on tauopathy using nexopathy models. Scientific Reports, 12(1).https://doi.org/10.1038/s41598-022-24687-4

The Virtual Parkinsonian Patient: the effects of L-dopa and Deep brain Stimulation on whole-brain dynamics

Marianna Angiolelli*1,2, Gabriele Casagrande1, Letizia Chiodo2, Damien Depannemaecker1, Viktor Jirsa1, Pierpaolo Sorrentino1,3


1Aix Marseille Univ, INSERM, INS, Inst Neurosci Syst, Marseille, France
2Department of Engineering, Università Campus Bio-Medico di Roma, Rome, Italy
3Department of Biomedical Sciences, University of Sassari, Sassari, Italy


*Email: marianna.angiolelli@unicampus.it
IntroductionParkinson’s disease is a progressive neurodegenerative disease characterized by the loss of dopaminergic neurons in the substantia nigra. The primary treatment for PD involves the administration of levodopa, but long-term use of it is associated with complications, necessitating alternative therapeutic strategies. One approach is deep brain stimulation (DBS), a neuromodulatory treatment that delivers electrical stimulation to specific brain regions (most often,the subthalamic nucleus). While DBS can be an effective therapy, optimal stimulation parameters are specific to each patient, and finding them can be challenging. Today, parameter tuning is based on a trial-and-error process, which is time-consuming, exhausting for the patient, requires a highly skilled dedicated team, and has very high chances of missing the optimal setting.
MethodsTo predict the effects of DBS on large-scale brain dynamics, we employed a mean-field neural mass model based on the adaptive quadratic integrate-and-fire (aQIF) framework [1], where Dopamine is included. The model was extended to incorporate an external current simulating a biphasic stimulation, mimicking DBS effects. Each brain region was modeled as a neural mass, with connectivity based on individual structural connectomes. The model includes excitatory, inhibitory, and neuromodulatory connections. EEG and deep electrode recordings in the STN validated the predictions. A Bayesian inversion with DNN inferred the neural state in ON/OFF conditions, quantifying parameter uncertainty.
ResultsFirst, we investigate different conditions varying simulations of L-Dopa administration and then changing DBS parameters analyzing large-scale brain activity and its impact on neural avalanche (spontaneous bursts of activations) topological properties. For all patients, we correctly infer that the dopaminergic tone is higher given the dynamics observed after administration of L-Dopa, and lower before the administration of L-Dopa [2]. The same approach enables us to tell apart pre- and post-DBS states across multiple patients, quantifying the effects of stimulation on large-scale brain dynamics.
DiscussionThis work provides a framework to understand how L-Dopa and DBS influence large-scale neural activity, offering insights into mechanisms and optimization for PD treatment. Unlike most PD models focusing on beta-range activity [3], we emphasize aperiodic activities instead, which have only received limited attention in Parkinson’s disease thus far. Furthermore, we focus on large-scale dynamics and efficient parameter estimation with uncertainty, rather than fitting a cost function. This approach explicitly accounts for Dopamine levels and stimulation amplitude, bridging pathophysiology, and personalized clinical predictions of clinical effectiveness.



Acknowledgements
The project leading to this publication has received funding from the Excellence Initiative of Aix-Marseille Université - A*Midex, a French “Investissements d’Avenir programme” AMX-21-IET-017
References
[1] Depannemaecker, D., Duprat, C., Casagrande, G., Saggio, M., Athanasiadis, A. P., Angiolelli, M., ... & Jirsa, V. (2024). A next generation neural mass model with neuromodulation. bioRxiv, 2024-06.
[2] Angiolelli, M., Depannemaecker, D., ... & Sorrentino, P. (2024). The Virtual Parkinsonian Patient. medRxiv, 2024-07.
[3] Meier, J. M., Perdikis, D., Blickensdörfer, A., ... & Ritter, P. (2022). Virtual deep brain stimulation: Multiscale co-simulation of a spiking basal ganglia model and a whole-brain mean-field model with The Virtual Brain. Experimental Neurology, 354, 114111.

Cortical Oscillatory Dynamics in Parkinsonian Networks: Biomarkers and the Potential of Theta Frequency Stimulation

June Jung1, Donald W Doherty1, Adam Newton1, Adriana Galvan5, Thomas Wichmann5, Salvador Dura-Bernal1, Hong-Yuan Chu4, Samuel Neymotin3, William W Lytton1,2

1Department of Physiology and Pharmacology, SUNY Downstate Medical Center, NY,2Kings County Hospital, Brooklyn, NY, USA3Nathan Kline Institute, Orangeburg, NY, USA,4Georgetown University, DC, USA,5Emory University, Atlanta, GA, USA
Introduction

Parkinson’s disease (PD) is marked by characteristic motor symptoms including tremors, stiffness, slowed movement, and balance issues, along with non-motor symptoms like cognitive challenges and difficulty making decisions. Symptom onset and severity vary among individuals. While dopaminergic neuron degeneration in the substantia nigra pars compacta (SNc) is a primary cause of motor dysfunction, recent studies highlight the critical role of disrupted oscillatory activity in the primary motor cortex (M1) in PD pathology. Mouse models of PD, including 6-OHDA mouse, and mitoPark mouse have shown reduced excitability in corticospinal pyramidal tract (PT) neurons.
Methods
We adapted an established mouse primary motor cortex (M1) framework, to make a Parkinsonian motor cortex (PD M1) computational model to investigate changes in neural oscillatory activity. The model incorporated experimental observations from the MitoPark mouse including decreased PT intrinsic excitability and reduced thalamocortical synapse strength to PT neurons (decreased 25% at 16–18 weeks; 50% at 25–28 weeks), correlating with disease progression. Multiple oscillation measures were analyzed as potential biomarkers for tracking disease severity.
Results
In vitro results were used to simulate in vivo Parkinsonian cortical activity, revealing progressively disrupted neuronal firing and increased beta oscillations (~20 Hz) with disease progression.Beta-gamma coupling and modulation index were two oscillatory measures that were significantly reduced under Parkinsonian conditions, with the modulation index progressively declining as the disease advanced. Theta-frequency stimulation suppressed beta bursts, enhanced beta-gamma coupling, and partially restored disrupted cortical network activity caused by PD pathophysiology.
Discussion
These findings suggest that the modulation index may serve as a biomarker for tracking Parkinsonian disease severity. Moreover, theta-frequency stimulation of inhibitory interneurons may help restore imbalanced cortical oscillations and could offer an alternative or complementary strategy to current high-frequency deep brain stimulation (DBS) of the subthalamic nucleus (STN) in PD patients.





AcknowledgementsThis research was funded in part by Aligning Science Across Parkinson’s [ASAP-020572] through the Michael J. Fox Foundation for Parkinson’s Research (MJFF). For the purpose of open access, the author has applied a CC BY public copyright license to all Author Accepted Manuscripts arising from this submission.
References
1. Chen, L., Daniels, S., Kim, Y., & Chu, H.-Y. (2021). Decreased excitability of motor cortical neurons in parkinsonism. Journal of Neuroscience, 41(25), 5553–5565.https://doi.org/10.1523/JNEUROSCI.2694-20.2021

2. Dura-Bernal, S., et al. (2023). Multiscale model of M1 circuits. Cell Reports, 42(6), 112574.https://doi.org/10.1016/j.celrep.2023.112574