Paris Circuit Dynamics Conference
Hippocampal–Cortical Circuits: From structure to computation

Neuromodulators allow neural circuits to alter their biophysical properties in a context-sensitive manner. Beyond their established roles in learning and behavioural state, neuromodulators may also contribute to sensory processing on sub-second timescales, although the computational advantages of such rapid modulation remain unclear.

We developed a model in which a neuromodulatory pathway dynamically adjusts intrinsic and synaptic parameters of a recurrent spiking neural network according to recent circuit activity. Neuromodulation improved performance across auditory and event-based visual tasks while reducing the numbers of neurons and spikes required.

In a speech-in-noise task containing dynamically varying natural and sinusoidally modulated backgrounds, neuromodulated networks were particularly robust near the transition between unsuccessful and reliable recognition. Their activity adapted to momentary noise levels and preferentially allocated processing to quieter, more informative periods, consistent with a “listening in the dips” strategy. This behaviour generalized across noise phases, modulation frequencies and naturalistic backgrounds.

The benefits persisted when modulation was constrained to operate through slow parameter diffusion, a small number of neuromodulator types, spatially dispersed release and a fully spiking modulatory population. Intermediate modulation dynamics were often most effective, suggesting a trade-off between rapid adaptation and temporal integration. Under strong spike regularisation, neuromodulated networks also retained high performance using sparse but distributed population activity, with fewer inactive neurons than unmodulated networks.

These results suggest that rapid neuromodulation can act as a context-dependent gain-control mechanism that dynamically reshapes circuit excitability and temporal integration. Neuromodulation may therefore provide a general means of increasing the flexibility and efficiency of sensory circuits without requiring major changes in circuit size or connectivity.

Norepinephrine (NE) has been primarily characterised as a fast phasic signal that responds to salient events, yet whether slower tonic NE dynamics during learning shape subsequent sleep remains unknown. Using high-density fiber photometry of the genetically encoded NE sensor GRABNE, we longitudinally monitored NE dynamics across multiple frontal cortical regions during an auditory discrimination task and subsequent sleep. We showed that fast phasic NE transients are associated with individual actions, whereas a distinct slower tonic NE elevation tracked prolonged task engagement. Following learning, the first NREM episode exhibited a transient, learning-dependent increase in cortical NE that differentiated early from late NREM physiology. Optogenetic silencing of NE neurons during task performance impaired learning and reorganised post-learning NREM by reducing slow-wave activity and spindle occurrence. These findings reveal a hierarchical, multi-timescale architecture of NE signaling that links sustained behavioral engagement to post-learning sleep and positions the LC–NE system as a state-dependent bridge between online performance and offline sleep.

Dopamine (DA) is a critical neuromodulator that underlies essential brain functions—including movement, reward, learning, and memory, and its dysregulation contributes to disorders such as Parkinson's disease. Deciphering these diverse functions requires the ability to quantitatively measure DA dynamics across physiological and pathological conditions. Although intensity-based fluorescent DA sensors have revolutionized in vivo neurochemical monitoring, they remain vulnerable to intensity-related artifacts caused by fluctuations in expression levels, optical scattering, and tissue movement. Fluorescence lifetime provides a powerful solution to these limitations. Because lifetime is an intrinsic photophysical parameter largely independent of absolute fluorescence intensity, it bypasses intensity-dependent artifacts while preserving high spatial and temporal resolution. Here, we report the development and comprehensive characterization of FLIM-DA, a novel fluorescence lifetime-based dopamine sensor. We demonstrate that FLIM-DA exhibits high sensitivity and accurately reports absolute DA concentrations across a physiological range in vivo. Furthermore, because its lifetime readout is inherently resistant to variations in sensor expression, excitation power, pH fluctuations, and motion artifacts, FLIM-DA delivers stable, reproducible signals across diverse experimental paradigms. We validate the utility of FLIM-DA by capturing both optogenetically evoked and behaviorally driven DA dynamics in mice, and we show that it reliably resolves basal tonic DA levels in pathological models of Parkinson's disease. Together, FLIM-DA provides a robust, artifact-resistant method for the quantitative interrogation of dopaminergic signaling in health and disease.

The brain continuously integrates distinct sensory modalities to generate adaptive behavior, yet the mechanisms resolving their inherent temporal discrepancies remain poorly understood. In the dorsomedial striatum (DMS), visual responses exhibit significantly longer latencies than tactile inputs, leading to asynchronous processing. We hypothesized that local neuromodulation by dopamine (DA) dynamically resolves this temporal mismatch. We employed a multidisciplinary approach combining in vivo whole-cell optopatch-clamp recordings, targeted optogenetic DA release, acetylcholine fiber photometry, and viral-mediated anatomical tracing in mice. We recorded identified direct (dMSNs) and indirect (iMSNs) pathway medium spiny neurons during independent and simultaneous whisker and visual stimulation. Our biological results demonstrate that DA release selectively enhances bimodal integration in dMSNs by accelerating visual responses, while leaving tactile responses unchanged. Anatomical tracing and photometry revealed that this specificity could relie on the differential recruitment of cholinergic interneurons (ChINs). Visual inputs, driven heavily by pyramidal tract projections from V1, efficiently recruit ChINs, which strongly inhibit MSNs. DA release transiently pauses ChIN firing, effectively disinhibiting, or "unbraking," specifically the visual response. To rigorously validate this circuit hypothesis, we developed an in silico spiking neural network model incorporating both MSN pathways and ChIN dynamics. The computational model successfully replicated our in vivo findings, demonstrating that the DA-mediated blockade of ChINs, counterbalanced by a DA-induced excitability reduction in iMSNs, is mathematically sufficient to synchronize multimodal sensory processing selectively in dMSNs. Together, our empirical and computational results describe a novel neuromodulatory mechanism by which dopamine bridges temporal delays across sensory modalities. This framework provides critical insights into how local circuit neuromodulation drives the precise sensorimotor transformations required for computation and behavior. For more information about our work, check the bioRxiv or contact us! (https://www.biorxiv.org/content/10.1101/2023.09.25.559270v1)

Behavioral state profoundly modulates cortical activity, but the underlying brainwide mechanisms, particularly the role of subcortical structures, remain poorly understood. Neuromodulation is likely implicated in mediating these state shifts, but it is unclear if it acts directly on cortical circuits or by coordinating brainwide activity patterns. We used functional ultrasound imaging (fUSI) to measure brainwide activity during optogenetic manipulations of cholinergic neurons and spontaneous fluctuations in behavioral state in head-fixed mice. Optogenetic activation of cholinergic neurons led to an increase in behavioral arousal, as seen by an increase in whisking and pupil dilation. It also caused changes in brainwide activity, with activation of thalamic and midbrain nuclei followed by decreases in activity, in particular in the hippocampus. These effects were overall stronger when the mouse was in a low-arousal state before stimulation onset. Thus, the cholinergic system can reconfigure whole-brain activity, providing a potential pathway for state-dependent modulation of cortical activity.

Prosocial behaviors, such as rescuing an individual in distress, require coordinated emotional and cognitive processes. We previously identified the dorsal hippocampus (dHPC) as essential for acquisition of prosocial rescue behavior (RB) in mice, in which an "observer" mouse learns to liberate a "victim" conspecific from cold water. However, the neuromodulatory mechanisms supporting rescue learning remain poorly understood. Noradrenaline (NE) regulates arousal, attention, and synaptic plasticity and has been implicated in emotional contagion and social behavior, yet whether noradrenergic signaling in the dHPC contributes to acquisition or emission of RB remains unknown. We investigated real-time NE dynamics in dHPC CA1 during RB training using fiber photometry with GRAB-NE alongside jRGECO1 to monitor neural population calcium activity in freely behaving observer mice. We focused on victim insertion, first door interaction, victim liberation, and first post-reunion social interaction. After validating signals with the tail suspension test, event-locked analysis showed that victim insertion and liberation each evoked robust NE increases, with no significant change in neural activity, while first door interaction produced no significant change in either signal, dissociating the NE increase from door opening itself. Post-reunion social interaction was associated with a significant decrease in population calcium activity and a non-significant downward NE trend. These effects did not differ by sex. Victimless control days did not elicit the same noradrenergic response to the chamber, door opening, or object insertion. Together, these findings show dHPC noradrenergic signaling during RB acquisition is dynamic and event-specific, with victim insertion and liberation, but not door interaction, associated with NE increases. This dissociation suggests that this prosocial behavior may be driven by emotional arousal elicited by observing the trapped victim, offering insight into how noradrenergic modulation shapes hippocampal circuits supporting prosocial learning, memory, and emotional contagion, with relevance to neuropsychiatric disorders marked by impaired empathy.

Understanding how neuronal ensembles drive behavior requires tools that can both record and manipulate brain activity with high precision. All-optical strategies, combining calcium imaging and optogenetics, have revolutionized this field by enabling real-time identification and control of neural circuits.

 To push these capabilities further, we present 2P-FENDO, the first fiber-coupled microscope capable of two-photon (2P) functional imaging and 2P holographic photostimulation in freely moving mice, now in its second and improved version. With a 500 µm field of view and near-single-cell resolution, 2P-FENDO provides an unparalleled combination of flexibility, precision, and large-volume imaging, making it an ideal tool for studying complex neuronal dynamics during natural behavior. Its adaptability allows researchers to explore brain activity in real-world conditions, offering new insights into the intricate relationship between neural circuits and behavior. 

By bridging the gap between high-resolution functional imaging and optogenetic control in freely behaving animals, 2P-FENDO opens exciting possibilities for the precise investigation of how distributed neuronal networks encode and influence behavior.

Visual gamma entrainment using sensory stimuli (vGENUS) is a promising non-invasive therapeutic approach for Alzheimer's disease (AD), showing efficacy in improving memory function. However, its mechanisms of action remain poorly understood. 

Using young AppNL-F/MAPT double knock-in (dKI) mice, a model of early AD, we examined brain dynamics alterations before amyloid plaque onset. High-density EEG recordings and metrics from fields outside neuroscience were used to assess brain dynamics fluidity—a measure of the brain's ability to transition between activity states. We revealed that dKI mice exhibit early, awake state-specific reductions in brain dynamics fluidity associated with cognitive deficits in complex memory tasks. 

Daily vGENUS sessions over two weeks restored brain dynamics fluidity and rescued memory deficits in dKI mice. Importantly, these effects built up during the stimulation protocol and persisted after stimulation ended, suggesting long-term modulation of brain function. Based on these results, we propose a "brain dynamics repair" mechanism for vGENUS that goes beyond current amyloid-centric hypotheses. 

This dual insight - that brain dynamics are both a target for repair and a potential diagnostic tool - provides new perspectives on early Alzheimer's disease pathophysiology.

We perceive our external world through actively sensing it. Active sensing, in the form of eye saccades or of sniffs, play a critical role in enabling and organizing perceptual brain signals. For instance, neural oscillations incorporate saccadic rhythm in visual areas (1-3). Active sensing might further generalize to "active processing". Heartbeat rhythm, which is coupled to micro-saccades (4), facilitates visual detection (5). Further, brain internal rhythms, such as hippocampal sharp-wave ripples (6-8), organize hippocampal neurons activity and contribute to large-scale activation of cortical and subcortical areas (9). However less is known about how these external and internal rhythms interact to organize brain signals in different brain regions and how they influence interactions between brain areas. Synchronization between areas has been proposed as a main way of rendering cognitive processes at the level of the whole brain, and it might well benefit from external and internal clocks. To answer this question, we recorded neural activity from three different brain areas of the monkey: the hippocampus, the secondary visual area TE and the secondary auditory area RM, as well as eye movements and electrocardiograms, while monkeys were watching pictures or hearing sounds, and we analyzed how spiking and oscillatory brain activity was influenced by the occurrence of saccades, heartbeats and ripples.

The ability to maintain and manipulate words in working memory to compose meaningful phrases is a hallmark of human cognition, yet its neural underpinnings remain incompletely understood. Linguistic working memory is known to rely on a cortical-hippocampal circuit, but the exact mechanisms supporting it remains to be elucidated. Using magnetoencephalography (MEG), we investigated the neural dynamics of linguistic working memory as participants read noun phrases of increasing length (one-word, two-word, and five-word phrases) and compared them with a subsequent image. Multivariate decoding of MEG signals revealed three distinct processing stages. First, during phrase comprehension, the neural representation of individual words was sustained for a variable duration depending on phrasal context, indicating that words remained explicitly represented in neural activity longer when they needed to be integrated with upcoming words. Second, during the delay period, these word-specific representations were transformed into a working memory code, whose activation increased with semantic complexity, quantified by the number of unique words in the phrase. Finally, the retrieval speed and accuracy of the memoranda depended on semantic complexity, with faster access to surface-level semantic properties than to deeper ones, suggesting that additional computations were required to reconstruct more complex representations. These findings indicate that the brain initially encodes phrases in an explicit, readily-accessible format but later compresses them within linguistic working memory, necessitating a decompression process for retrieval. Our results highlight the role of a cortical-hippocampal circuit in sustaining and transforming linguistic representations, placing strong constraints on models of phrase processing in the human brain.

It has been shown experimentally that hippocampal theta sequences, generated during active exploration of the environment, play a crucial role in selecting appropriate actions in goal-directed tasks. Additionally, experimental studies have demonstrated that spontaneous task-independent replay of sequences encoding a learned route during sharp-wave–ripple (SWR) events can be beneficial for consolidating path memory and discovering new behavioral strategies, such as forming novel space-action representations, finding shortcuts, and integrating and associating different routes. 

However, the mechanism by which activation of a sequence of upcoming positions in each theta cycle and replay of sequences in SWR alters and shapes action selection remains unclear. In a closed-loop model of spatial navigation based on place cells and action selection cells, we investigate how the emergence of theta code and sequence replay can modulate the action selection strategy. 

The model comprises a recurrent inhibitory-excitatory layer with place cells formed from spatially tuned input received from the environment, connected to the action-selection neuronal population modeled as a ring attractor. The activity of action selection neurons determines the agent's actions in the environment, with the task of finding a reward region and learning the optimal path through synaptic plasticity of place-to-action and place-to-place cell connections. 

We observe that during rest replay of sequences of place cells representing a learned non-optimal path, the plasticity of connections among place cells and action neurons, governed by the spike-timing-dependent plasticity (STDP) rule, leads to the shortening of place cell sequences, resulting in a sparser code. This plasticity also adjusts the connections from place cells to action cells, enabling the formation of representations of shortcuts. 

During active exploration, the activation of upcoming place cells in each theta cycle results in the simultaneous activation of actions associated with each pair of states. This activation pattern causes a bump to form in the ring attractor, directing the agent towards the shortcut path based on the summation of action vectors.

Although the role of hippocampal-cortical dialogue in memory encoding and consolidation has been thoroughly investigated, most mechanistic studies in rodents are based on recordings in the dorsal hippocampus (dHPC), which sends very few projections to the medial prefrontal cortex (mPFC) and therefore does not seem ideally placed to interact with it. In contrast, the ventral hippocampus (vHPC) sends massive projections to the mPFC and integrates spatial and non-spatial information via distributed reciprocal connections (dHPC, amygdala, etc.). To understand how signals from dHPC and vHPC are integrated or segregated in mPFC networks during learning and memory, we recorded simultaneously from dHPC, vHPC and mPFC in freely behaving rats performing a spatial alternation task. As previously described, we show that some vHPC cells are space-modulated even if they do not exhibit clear place fields in our environment. While most ripples are known to be asynchronous between hippocampal regions, our preliminary results indicate that PFC units primarily respond to vHPC ripples suggesting the vHPC plays a key role in the information transfer from the HPC to the mPFC. This highlights the need for further investigation of vHPC properties to better understand the nature and mechanisms of memory encoding.

Adaptive threat responses require both defensive behaviours to minimize danger but also less studied recuperative strategies to recover from the induced physiological stress. 

Here, we identify a novel slow-breathing immobility recovery state in the mouse that emerges when animals identify safe environments after threat avoidance. This immobile state is characterized by a unique 2-4Hz breathing profile and sharp-wave ripple (SWR) hippocampal replay of the aversive experience. 

Suppressing SWRs inhibits its emergence by impairing learning of safe parts of the environment and incurs a subsequent and proportional increase in post-task stress, demonstrating the role of this state in reducing stress after threat. Pharmacological anxiolysis with diazepam promotes the recovery state but suppresses SWRs, revealing a trade-off between immediate stress relief and long-term safety learning. 

Together, these results demonstrate the role of hippocampal replay in gating stress recovery processes and therefore its importance for emotional resilience.

Place and grid cells form a neural system for self-location, firing in sequential patterns within each hippocampal theta cycle as rodents navigate a linear track. These sequences, known as "theta sweeps," represent the animal's decoded location sweeping forward from its current position. Interestingly, recent findings in open-field environments reveal alternating left-right theta sweeps and propose a circuit responsible for their generation (Vollan* & Gardner* et al., 2025, Nature). 

We present a mechanistic computational framework of this circuit, incorporating theta-modulated head-direction (HD) cells, conjunctive grid × direction cells (ConjGCs), and pure grid cells (GCs) (Ji* & Chu* et al., 2025, Curr. Biol.). The framework is based on continuous attractor dynamics, firing rate adaptation, and theta modulation. It first induces bidirectional sweeps in the HD ring attractor, where population activity encodes internal direction. This directional signal is then transmitted via ConjGCs to the GC attractor network, generating left-right positional sweeps in GCs. 

This framework integrates key findings on theta sweeps across different types of spatial tuning cells over the past three decades. For grid cells, it accounts for the left-right sweeps in open fields, alignment of internal position and direction sweeps, and explains how sweep length depends on grid spacing. For place cells, it elucidates the emergence of cyclic sweeps in T-mazes, and forward-backward sweeps in linear tracks (Chu* & Ji* et al., 2024, eLife). Additionally, it successfully predicts precession in theta-modulated HD cells within the anteroventral thalamic nuclei and reconciles diverse empirical findings on HD cell firing, including theta skipping, anticipatory firing, and phase precession (Ji* & Lomi* et al., 2025, Hippocampus). 

Overall, the framework provides important insights into spatial orientation and navigation and may contribute to understanding spatial deficits in the early stages of Alzheimer's disease (Castegnaro* & Ji* et al., 2023, Curr. Biol.).

The entire auditory system downstream of the cochlea features pronounced offset responses, which follow the termination of sounds. Because of their ubiquity, it is still an unsolved question whether offset responses are generated early in the auditory system and then propagated or recomputed at each processing stage. Here, we analysed large-scale sound responses datasets acquired in the cochlear nucleus, inferior colliculus, medial geniculate nucleus and auditory cortex of awake mice. In all brain regions, offset responses were found to be largely distributed across neurons, and are often combined with onset and sustained response in the same neuron. However, using population activity decoders, we observed that neural representations after the sound offset show a three-fold increase in sounds encoding accuracy in the cortex relative to subcortical areas. This result indicates that cortical offsets encode a more precise short-term memory of the elapsed sound than subcortical offsets and that they likely result from specific computational steps.

Dopamine plays a crucial role in spatial learning and memory, with its effects mediated through various neural pathways. The stellate cells in layer II of the medial entorhinal cortex exhibit specific firing patterns essential for cognition and perception, largely regulated by calcium dynamics within the axon initial segment (AIS). Recent studies suggest a potential coupling between dopamine D2 receptors (D2R) and T-type Ca²⁺ channels, which may influence neural excitability. This study employs computational modeling to explore how D2R activation modulates resting membrane potential (RMP) and action potential (AP) plasticity, particularly in pathological conditions. The model was developed in three stages. First, biophysical parameters of various ion channels in the AIS of stellate cells were adapted from experimental data. Second, a GPCR-mediated cAMP signaling model was incorporated to regulate T-type Ca²⁺ channel conductance using a modified Boltzmann equation. Third, pharmacological simulations were performed to investigate dopamine's effects. Using NEURON software, we conducted current-clamp and voltage-clamp experiments on a single-compartmental model to analyze RMP, APs, and T-type Ca²⁺ currents. Simulations of 10 µM dopamine agonist bromocriptine (Bromo) revealed altered T-type Ca²⁺ channel activation without affecting inactivation. The half-activation potential shifted from –36 mV to –32 mV, reducing excitability. AP frequency decreased under 400 pA current injection (1s), indicating D2R activation dampens excitability. Furthermore, window currents maintaining RMP were reduced, counterbalanced by A-type K⁺ channel regulation. These findings suggest cAMP antagonists and K⁺ channel agonists may serve as potential therapeutic alternatives for dopamine in pathological conditions, with implications for spatial memory research and neurophysiological treatments.

Selective pressures adapt the hippocampus to changing information processing requirements. At the same time, the structure of the hippocampus may also reflect phylogenetic history. To test phylogenetic and ecological signal in the hippocampus, we assessed neuron numbers in five hippocampal neuron populations in 65 species, representing 11 mammalian orders using the Optical Fractionator. We used z-scored data to account for large differences in species' absolute neuron numbers. Species separations by neuron numbers were explored in a phylogenetic PCA (pPCA). Along PC1, rodents, primates and ungulates form the extremes of a continuum and are linked by insectivores and carnivores (loadings, positive: CA3, CA1, subiculum, negative: hilus, granule neurons). In a pPCA of ratios of neuron populations, insectivores (heaviest loading: PC1, granule neurons-> hilus), primates (PC2, granule neurons-> CA3), and bats (PC3, CA1-> subiculum) disperse in nearly orthogonal directions, linked by rodents where the axes meet. Phylogenetic comparative methods were used to study evolution of neuron populations and its correlation to ecological factors. Phylogenetic signal was present in all populations, strongest in hilar and CA3, and weakest in CA1 neurons. The best-fitting model for phylogenetic changes of granule, CA1 pyramidal and subicular neurons was a constrained configuration model, which was still a viable alternative to an unconstrained model that performed best for hilar and CA3 pyramidal neurons. Among ecological factors tested in a PGLS (activity cycle, diet breadth, habitat breadth, home range, social group size and tropic level), diet breadth correlated with granule, hilar and subicular neuron numbers, while species home ranges impacted weakly on CA1 pyramidal neurons and strongly on the ratio between granule and hilar neurons. In conclusion, our results show that CA1 exhibits the lowest level of evolutionary constraint among the species in our sample. Furthermore, our findings underscore the impact of diet breadth on the evolutionary trajectory of the hippocampus.

Mean-field models have been developed to replicate key features of epileptic seizure dynamics. However, the precise mechanisms and the role of the brain area responsible for seizure onset and propagation remain incompletely understood. In this study, we employ computational methods within The Virtual Brain framework and the Epileptor model to explore how the location and connectivity of an Epileptogenic Zone (EZ) in a mouse brain are related to focal seizures (seizures that start in one brain area and may or may not remain localized), with a specific focus on the hippocampal region known for its association with epileptic seizures. We then devise computational strategies to confine seizures (prevent widespread propagation), simulating medical-like treatments such as tissue resection and the application of an anti-seizure drugs or neurostimulation to suppress hyperexcitability. Through selectively removing (blocking) specific connections informed by the structural connectome and graph network measurements or by locally reducing outgoing connection weights of EZ areas, we demonstrate that seizures can be kept constrained around the EZ region. We successfully identified the minimal connections necessary to prevent widespread seizures, with a particular focus on minimizing surgical or medical intervention while simultaneously preserving the original structural connectivity and maximizing brain functionality.

It has long been hypothesised that working memory could be supported by sustained neural activity. By bridging the gap between the presentation of stimuli, such activity could maintain a trace of the initial stimulus to guide behavioural responses to subsequent stimuli. 

However, sustained neural activity is energetically inefficient and recent empirical studies have struggled to find evidence in support of this hypothesis. An alternative proposal suggests that short-term memory traces could be maintained by Calcium-mediated short term synaptic facilitation (STSF), whereby the initial stimulus produces short-term increases in synaptic efficacy that shape the network response to subsequent stimuli. This mechanism has primarily been discussed in relation to working memory maintenance in frontal cortex. 

Here, we aim to establish whether this mechanism could also account for the role of hippocampal circuits in structural learning. Specifically, we model STSF at the CA3-CA1 Schaffer Collaterals, simulating activity during a contextual inference. 

Using a multi-layer perceptron to classify network responses in CA1, we show that Calcium-mediated STSF allows for activity-silent delay period maintenance of information, necessary for acquisition of the task. In addition, we demonstrate that learning depends on the overlap of stimulus representations in CA3 (but not CA1). 

In sum, this model suggests that STSF might support the short-term maintenance of hippocampal memory traces that is required for structural learning and contextual inference, making specific predictions for future experiments in this area.

How does the hippocampus determine when to store a new memory? We hypothesize that when animals transition into a novel environment, a brief increase in Dentate Gyrus (DG) activity triggers downstream circuits to form new neuronal representations. To test this empirically, we investigate how neuronal population activity in hippocampal subregions responds during transitions from a familiar (F) to a novel (N) environment. Using an immersive virtual reality setup for head-fixed rodent navigation (n = 9 mice), we trained animals in a familiar environment for five consecutive daily sessions (30 min each). Performance, measured by anticipatory licking before reward delivery, improved significantly (hit-rate: 0.37 ± 0.15 hits/lap in session 1 vs. 0.95 ± 0.03 hits/lap in session 5, p < 0.05), indicating task learning. In the sixth session, we introduced a novel environment and compared performance across conditions. While performance in the familiar environment continued to improve, introducing the novel environment caused a drop to levels comparable to untrained animals (hit-rate: 0.65 ± 0.07 hits/lap in F vs. 0.49 ± 0.09 hits/lap in N, p < 0.05; no significant difference between session 1 in F and N, p > 0.05). In an independent session, we recorded hippocampal activity during the first F-to-N transition using Neuropixels probes. Based on probe placement, we identified 25 putative DG and 128 putative CA1 principal cells. DG firing rates increased slightly within a 1-second window before and after the transition (Friedman test: χ² = 10.63, p = 0.01) but remained unchanged in CA1 (χ² = 2.94, p = 0.4). Our findings provide a framework for studying how environmental novelty influences hippocampal population dynamics.

A key question in neuroscience is to unravel causal relations between neuronal circuits and behavior. The precise study of neuronal circuits requires to measure and manipulate neuronal activity with high spatial (single-cell) and temporal resolution within large ensembles (Marshel et al., Science 2019). All-optical experiments using two-photon (2P) calcium imaging and optogenetic photostimulation offer a promising approach to studying neuronal circuits in vivo in mice (Emiliani et al., Nat. Rev. 2022). However, they have so far focused on experiments in head restrained mice (Robinson et al., Cell 2020). We recently developed a flexible 2P microendoscope (2P-FENDO) capable of all-optical brain investigation at near cellular resolution in the L2/3 of the barrel and visual cortex of freely moving mice. The system performs fast two-photon (2P) functional imaging and 2P holographic photostimulation of single and multiple cells using axially confined extended spots (Accanto et al., Neuron 2023). Here, we exploited the characteristics and advantages of the 2P-FENDO in the hippocampus, a crucial brain structure being involved in spatial navigation and memory formation. Proof-of-principle experiments were performed in freely moving mice co-expressing GCaMP8s and the opsin ChRmine in the CA1 region of the hippocampus. In a field of view of 250 µm in diameter, we demonstrate functional imaging at a frame rate of up to 100 Hz and precise photostimulation of single and multiple pyramidal neurons. 2P-FENDO enables single-cell resolution imaging and control of hippocampal neurons in freely behaving animals, helping to reveal networks involved in memory and spatial navigation, essential for adaptive behavior and survival.

In my PhD project, I study the processes that enable the long-term storage of newly acquired and initially fragile information, a phenomenon known as memory consolidation. It is well established that this crucial step in memory formation occurs primarily during slow-wave sleep, where it is thought to rely on a finely regulated dialogue between the hippocampus and the cortex. One possible "conductor" facilitating this dialogue could be the Claustrum, a still poorly understood brain structure that is ideally connected to coordinate these interactions. We have demonstrated that stimulating the Claustrum during post-learning sleep effectively enhances memory consolidation. Unexpectedly, this effect does not rely on a modulation of the classic hippocampo-cortical dialogue but rather on a global orchestration of cortical synchronization. This discovery suggests the existence of a complementary mechanism of consolidation. Our multidisciplinary approach, combining experimental and theoretical neuroscience, will allow us to gain a deeper understanding of this process and its potential clinical implications.

To fully understand how information is processed in complex cell circuits, we need techniques able to target and control specifically the activity of the elements involved, including astrocytes. Currently, available tools like optogenetics and chemogenetics affect the entire astrocytic population, making it difficult to target specific subsets. This lack of specificity hinders our understanding of the role of astrocytes in behavior. 

We present a tool to translate the activity-mediated calcium signals of astrocytes into gene expression in a light-dependent manner, i.e. AstroLight. Using AstroLight in parallel with electrophysiology, pharmacology, fiber photometry and behavioral techniques, we demonstrate mice astrocytic involvement in motivated behaviors of the Nucleus Accumbens (NAc). 

First, we designed AstroLight vectors, validating their expression after viral infection. Then, using fiber photometry we monitored in vivo the astrocyte calcium dynamics in the NAc, demonstrating their involvement during a goal-directed behavioral task. 

Afterwards, we used AstroLight to express channelrhodopsin (ChR2) in the active astrocytes related to reward consumption or exploratory behaviors. Three-dimensional quantification of the ChR2/AstroLight ratio across the NAc revealed astrocytic ensembles related to these different behavioral features. 

Finally, we modulated those ensembles through optogenetic stimulation, showing that activating the astrocyte ensemble related to a specific reward recalls the direction of behavior towards that precise option. 

These results show AstroLight as a powerful tool to study astrocyte-neuron interaction's function with precise spatial and temporal control and reveal that astrocytic ensembles can impact neuronal activity and shape behavioral responses positioning them as relevant component of the circuit.

Connectomics seeks to map the intricate wiring of the brain to elucidate how neural circuits process information and underpin brain function in both health and disease. Traditional imaging techniques such as electron microscopy, while highly detailed, are slow, destructive, and impractical for larger brains. In contrast, X-ray tomography (XRT) offers a fast, scalable, and non-destructive alternative to capture high-resolution maps of neural connectivity. We demonstrate the potential of this method using the mouse olfactory system as a model to investigate input-output transformations that illuminate the organization of sensory processing. XRT achieves high spatial isotropic resolution—from tens of nanometers to micrometers—and is ideally suited for correlative multimodal imaging. By integrating XRT with functional data from calcium imaging and ultra-high resolution structural data from focused ion beam-scanning electron microscopy (FIB-SEM), it is possible to combine complementary modalities to generate comprehensive maps of brain circuitry. While challenges remain, effective integration is achievable.

Multiple Sclerosis (MS) is a neuroinflammatory disease of the central nervous system and is the leading cause of non-traumatic neurologic disability in the young adult population. The structural damage present in MS leads to a rearrangement in the communication between brain areas. Furthermore, the general loss of synapses and neurons leads to a disruption in excitation inhibition balance. In the first part of this project, we explore the hybrid functional magnetic resonance imaging (fMRI)-informed resting-state structural connectivity (rsSC) to simultaneously combine structural and functional information. The sign of the entries in these hybrid matrices can be interpreted as a measure of excitation or inhibition between brain areas. We compute the rsSC matrices in a cohort of 13 healthy controls (HC) and 24 people with MS (pwMS) whose data was collected at the UZ Brussel hospital. In the second part, we explore the functional dynamics of the processed fMRI signals using the Leading Eigenvector Analysis (LEiDA). LEiDA performs temporal clustering to functional time series and allows the extraction of functional states that can help to identify differences in the time evolution of states between HC and MS. Our goal is to study the overlap of the identified states with the reference resting state networks (e.g., the Default Mode Network and the Somatomotor Network) and investigate the switching behavior and time evolution of different brain functional states. For both approaches, we employ different types of brain atlases, namely atlases designed on structural features (structural atlases) or calculated from functional neuroimaging data (functional atlases). We found that the atlas' choice is important in revealing significant differences in Excitation Inhibition Balance in the Default Mode and Somatomotor networks between the HC and MS groups studied and in the functional dynamical patterns of relevant time series.

Sharp-wave ripples (SWRs) are distributed oscillatory events in the CA1 region of the hippocampus, known to be involved in memory consolidation through reactivation of neural assemblies.

These events mainly occur in two distinct states, quiet awakeness and slow-wave sleep. Interestingly, in both states, SWR content can be biased by sensory stimuli, increasing the likelihood of replaying certain memories. However, the precise mechanisms of this cortical influence on hippocampal activity are still unknown.

Here, we leveraged high-density recordings in hippocampal subfields to point out the importance of the dentate gyrus in ripples. We found evidence for strong and diverse activity in the DG during ripples, but also 100 ms before. This pre-ripple activity, originating from entorhinal cortex-like inputs, was predictive of future ripple amplitude and found in a large subset of granule cells. Importantly, this increase in activity in DG was also seen in CA1 but took a non-oscillatory form.

This pre-ripple activity provides a potential mechanistic explanation of replay content biasing by cortices and also a promising tool to detect ripples before they occur, for total termination or precise enhancement.

Spiking Neural Networks (SNNs) offer a biologically inspired framework for modeling brain-like computations, particularly in tasks such as digit classification. 

This study presents a novel SNN architecture based on the Adaptive Exponential Integrate-and-Fire (AdEx) neuron model, enriched with astrocyte modulation to investigate the role of glial-neuronal interactions in enhancing network performance. 

Using Spike-Timing Dependent Plasticity (STDP) as the learning rule, the network is evaluated on digit classification tasks to assess the computational benefits of astrocyte inclusion. Preliminary results show the comparison of SNN using Leaky Integrate-and-Fire model and using AdEx, and with the inclusion of astrocytes. 

Future work will transition from STDP to a predictive coding framework, aligning more closely with theories of efficient brain computation and hierarchical inference.

Substantial evidence supports a critical role for sleep in memory consolidation. In the hippocampus, place cells encode spatial information and are reactivated during sleep in the same sequence experienced during exploration. This replay occurs during sharp-wave ripple (SWR) oscillations, which are thought to provide optimal timing for synaptic plasticity, particularly long-term potentiation. Previous work demonstrated that disrupting SWRs impairs performance on spatial memory tasks, establishing their causal role in memory consolidation. However, the cellular and molecular mechanisms that regulate this network activity in relation to learning remain poorly understood.

While neurons have long been considered the only drivers of information processing, mounting evidence suggests a key role for glial cells - particularly astrocytes - in modulating neural circuits and behaviour. Though considered electrically silent, astrocytes display dynamic intracellular calcium signaling in response to local and global brain states, including sleep and wakefulness. Recent studies show that astrocytic calcium activity can encode expected reward locations in spatial contexts. Yet, how astrocytes interact with hippocampal network rhythms - especially SWRs - during sleep-dependent memory consolidation remains unexplored.

Here, we hypothesized that astrocytic calcium dynamics are temporally coupled to SWRs during post-learning sleep and play an active role in memory consolidation. Preliminary results suggest that locally disrupting astrocytic calcium in CA1 impairs reference and working memory, associated with an alteration of SWRs. Furthermore, by using simultaneous recordings of large scale somatic astroglial calcium dynamics and extracellular recordings of neuronal activity in freely behaving mice, we studied hippocampal average astroglial calcium signal in relation to hippocampal rhythms during spatial learning tasks and subsequent sleep. We notably identified specific temporal astroglial calcium dynamics following SWRs during sleep.

These findings suggest that CA1 astrocytes actively respond and modulate hippocampal rhythms critical for memory consolidation, revealing a novel glia-neuron interaction mechanism in the encoding of spatial memory.

Memory consolidation, the process by which newly acquired information is stabilized for long-term storage, relies on hippocampal sharp wave ripples (SWRs). SWRs are brief episodes of neuronal activity characterized by CA1 pyramidal dendrite depolarization and the simultaneous occurrence of high-frequency oscillations, the ripples. 

Our laboratory has recently shown that disrupting astrocytic calcium (Ca²⁺) signaling impairs SWRs in mice. This finding highlights a potential role for astrocytic Ca²⁺ dynamics in the generation or modulation of SWRs. 

To explore this further, we performed simultaneous recordings of SWRs and astrocytic Ca²⁺ signals by combining two-photon (2P) microscopy and multi-electrode arrays (MEA), to characterize the properties and dynamics of astrocytic Ca²⁺ events during SWRs. 

Our results reveal a bidirectional interaction between astrocytic Ca²⁺ events and neuronal SWRs: astrocytic Ca²⁺ activity modulates SWRs, while SWRs, in turn, dynamically influence astrocytic Ca²⁺ transients, particularly within the soma. We further investigated astrocytic Ca²⁺ signals during gamma oscillations—a distinct form of rhythmic brain activity.

Strikingly, we found that the characteristics and temporal dynamics of astrocytic Ca²⁺ transients are specific to the underlying neuronal activity, suggesting a context-dependent astrocyte response tuned to different brain states.