CNS*2025 Florence

Organizers: Sara J. Aton1, Michal Zochowski2


1. Department of Molecular, Cellular and Developmental Biology, University of Michigan, Ann Arbor, Mi, USA

2. Department of Physics and Biophysics Program, University of Michigan, Ann Arbor, Mi, USA 

 

Introduction 

State-dependent neural network dynamics have been proposed as an essential component of memory consolidation. Sleep and wake states drive changes in network-wide behavior and brain physiology, including state-specific network oscillations, synaptic plasticity profiles, excitatory-inhibitory balance, and neuronal activity levels. These changes are often mediated via release of neuromodulators that affect different features of network-wide dynamics, Those neuromodulators ( that include but are not limited to acetylcholine, serotonin, norepinephrine and dopamine) on one hand target specific cell populations that express required receptor types, while on the other act as "global regulators" by influencing large neural networks and different brain modalities, adjusting their activity levels and changing computational properties of these networks.

Here, we will bring together an interdisciplinary group of experts whose research focuses on elucidating the role of state dependent neuromodulatory effects on network plasticity, dynamics and overall information processing. The workshop will bring together researchers using experimental as well as theoretical/computational approaches, while the presentations will bridge the scales from molecular and cellular to network effects.

Specific questions that we are aiming to be addressed are:

• How state specific changes in neuromodulatory milieu affect neuronal dynamics and network plasticity,
• How does neural network dynamics change during sleep?
• How are memories transformed during the sleep consolidation process?
• How state-dependent neuromodulatory processes interact with memory consolidation?
• What is the role of REM vs SWS sleep for various forms of memory?
• Can effects of state-dependent memory consolidation be successfully incorporated into AI models for improved performance?

 



Schedule:

Time                                  Speaker                                   Title
9:00 - 9:05              Sara Aton / Michal Zochowski               Introduction

9:05 - 9:30                       Maksim Bazhenov                        Interleaved Replay of Novel and Familiar Memory Traces During Slow-Wave Sleep Prevents Catastrophic Forgetting

9:30 - 9:55                        Cecilia Diniz Behn                       Mathematical modeling of noradrenaline dynamics across behavioral states

9:54 - 10:20                           Victoria Booth                         Modeling infraslow oscillatory brainstem dynamics during NREM-REM sleep cycles

10:25 - 10:50                          Niels Niethard                       VIP interneuron activity during sleep – conveying the cortical infraslow oscillation

10:50 - 11:15                    Abdelrahman Rayan                   Latent Dimensions of Rodent Sleep: A Deep Learning Approach

11:15 - 11:40                    Penny Lewis (virtual)                  Detecting memory reactivation in REM and NREM sleep - towards a generalisable EEG classifier

11:40 - 12:30                                  All                                 Questions and Debate

12:30                                                                                   End of Workshop and Lunch Break

 

Talk Summaries: 

Interleaved Replay of Novel and Familiar Memory Traces During Slow-Wave Sleep Prevents Catastrophic Forgetting
Maksim Bazhenov
Department of Medicine, University of California at San Diego, San Diego, CA, USA

Abstract
Humans and animals can learn continuously, acquiring new knowledge and integrating it into the pool of lifelong memories. Sleep replay has been proposed as a powerful mechanism contributing to interference-free new learning. In contrast, artificial systems suffer from a problem called catastrophic forgetting, where new training damages existing memories. This issue can be mitigated by interleaving training on new tasks with past data; however, whether the brain employs this strategy remains unknown. In this work, we show that slow-wave sleep (SWS) employs an interleaved replay of familiar cortical and novel hippocampal memory traces within individual Up states of the sleep slow oscillation (SO), allowing new memories to be embedded into the existing pool of cortical memories without interference. Using a combination of biophysical modeling and analyses of single-unit activity from the mouse retrosplenial cortex, we found that hippocampal ripples arriving near the Down-to-Up or Up-to-Down transitions of the sleep SO entrain novel memory replay, while the middle phase of the Up state always replays familiar cortical memories. This strategy ensures the consolidation of novel cortical memory traces into long-term storage while minimizing damage to familiar ones. This study presents a novel theory of how the replay of familiar and novel memory traces is organized during SWS to enable continual learning.


Mathematical modeling of noradrenaline dynamics across behavioral states
Cecilia Diniz Behn
Department of Applied Math & Statistics, Colorado School of Mines, Golden, CO, USA

Abstract
Wake, rapid eye movement (REM) sleep, and non-REM (NREM) sleep are characterized by differential release of neuromodulators such as noradrenaline (NE). The noradrenergic locus coeruleus (LC) promotes wakefulness and exhibits state-dependent changes in neuronal activity and NE release. Recent experimental advancements have described the LC-NE system with high temporal resolution and refined our understanding of how LC activity changes with behavioral state. Specifically, LC activity and NE release are highest during wakefulness, but they vary with attention and activity level. In NREM sleep, LC activity is phasic and drives infraslow extracellular oscillations in NE. During REM sleep, LC activity ceases, and extracellular NE decays slowly. However, at the offset of REM sleep, extracellular NE increases rapidly suggesting that there is an asymmetry in the dynamics of NE rise and fall. We previously developed a firing rate model formalism that describes both firing rate and associated neurotransmitter release. In current work, we refine this formalism to account for the fine details of LC-NE dynamics and explore the implications of this refinement in the context of sleep-wake network models.


Modeling infraslow oscillatory brainstem dynamics during NREM-REM sleep cycles
Victoria Booth
Department of Mathematics, University of Michigan, Ann Arbor, MI, USA

Abstract
During sleep, the mammalian brain alternates between two major brain states, rapid eye movement (REM) and non-REM (NREM) sleep. Recent work has shown that the electroencephalogram (EEG) during NREM sleep in both humans and mice shows a pronounced infraslow (~50 s) modulation in the σ (10-15 Hz) power range. This infraslow σ power (ISP) rhythm is strongly reflected in the firing activity of the wake-promoting locus coeruleus (LC) and its expression of norepinephrine, as well as in the firing activity in brainstem REM sleep-regulatory areas such as the dorsomedial medulla (dmM) and the periaqueductal gray (PAG) areas. Importantly, transitions from NREM to REM sleep and spontaneous awakenings are synchronized with the ISP rhythm, suggesting that the rhythm plays a crucial role in shaping NREM-REM cycles and overall sleep architecture. Recent recordings of brainstem neural activity have identified subpopulations of REM sleep- activated and -inhibited neurons with opposing infraslow oscillatory activity suggesting mutually inhibitory interactions. We are developing computational models for these REM-regulatory brainstem circuits to analyze how infraslow rhythmic dynamics and slow ultradian processes that promote REM sleep combine to govern NREM-REM cycling and the temporal architecture of sleep.


Latent Dimensions of Rodent Sleep: A Deep Learning Approach
Abdelrahman Rayan, Anumita Samanta, Irene Navarro-Lobato, Ashutosh Narang, Camille Revol, Geoffroy Peirs, Amirsadra Khodadadi, Lisa Genzel
Donders Institute for Brain, Cognition, and Behavior, Radboud University, Nijmegen, Netherlands

Abstract
Rodents are a widely used model for studying the functions of sleep and memory consolidation. However, traditional classification methods—relying on manual scoring of sleep stages into NREM and REM—may oversimplify the complexity of rodent sleep, limiting translational insights across species. In this work, we present a deep learning framework based on a Boltzmann machine to explore the latent structure of rodent sleep beyond conventional classifications. We compare automatic scoring with manual methods, outline their respective strengths and limitations, and highlight how data-driven approaches can reveal finer-grained substates. Our findings suggest that rodent sleep exhibits a richer archite...