This project explores the molecular and cellular mechanisms behind iron and peroxidized lipid clearance via endolysosomal secretion in neurons and astrocytes, with a focus on dopaminergic neurons related to Parkinson's disease. It hypothesizes that dysregulation in secretory ferroptosis, where neurons release toxic lipid-protein particles, leads to neurodegeneration. Using iPSCs and knockout mouse models, the study will investigate key players like VAMP7 and ATG5, which govern endosomal secretion and autophagy. Advanced techniques such as lipidomics and proteomics will be applied to better understand the role of secretory pathways in neurodegeneration and identify potential therapeutic targets for Parkinson's disease.

This thesis project seeks to explore how iron and peroxidated lipids are cleared from brain cells, specifically neurons and astrocytes, by a process known as endolysosomal secretion. The research will focus on a specific type of neuron linked to Parkinson's disease—dopaminergic neurons—and aims to understand how disruptions in this clearance mechanism may lead to the cell death seen in neurodegenerative diseases.
Parkinson’s disease is characterized by the buildup of iron and peroxidated lipids in the brain, contributing to the death of neurons. A major question the project aims to answer is how these harmful substances are normally cleared from the cells and what happens when this process is disrupted. The team believes that secretory ferroptosis—where neurons release toxic particles—is key to understanding this problem. Ferroptosis is a form of cell death that occurs when iron and damaged fats accumulate and cannot be properly removed from the cell. The buildup of these substances can be toxic, leading to cell damage and death, a major feature of neurodegeneration in diseases like Parkinson’s.
To investigate this, the research team will use advanced methods, including stem cell-derived models of neurons and mouse models with specific gene knockouts. They will focus on two critical proteins, VAMP7 and ATG5, involved in the pathways that manage this cellular cleanup. VAMP7 is responsible for the secretion of harmful substances from late-stage endosomes, while ATG5 is involved in autophagy, the process by which cells degrade and recycle damaged components. By studying the impact of removing these proteins from neurons and astrocytes, the researchers aim to understand better how secretory pathways protect against or contribute to neurodegenerative processes.
The project has three key objectives:
1. Identifying the Mechanism of Iron and Peroxidated Lipids Clearance: we will use human stem cell models to study how iron and damaged lipids are expelled from neurons and astrocytes. They will focus on how the absence of VAMP7 and ATG5 impacts this process, potentially leading to harmful buildup inside the cells. These findings could shed light on the biological processes that protect neurons from damage, or alternatively, how the failure of these processes contributes to neurodegeneration.
2. Analyzing the Toxicity of Neuronal Secretions: we will investigate whether the iron and lipids secreted by neurons and astrocytes, especially when VAMP7 and ATG5 are absent, are more or less toxic to surrounding cells. We will study the secreted components in detail, using advanced techniques like proteomics and lipidomics, to understand what these cells release and whether it contributes to cell death and the spread of Parkinson's disease-related proteins, like ?-synuclein.
3. Studying the Role of Ferroptosis in Neurodegeneration: We will explore how defects in the cellular clearance pathways might lead to ferroptosis in the brain. We will investigate how this form of cell death, driven by iron and lipid peroxidation, is linked to the progression of Parkinson’s disease. The role of ?-synuclein, a protein that accumulates in Parkinson's disease, will also be examined to see if its secretion and aggregation are affected by the disruption of these pathways.
By focusing on these mechanisms, the project hopes to provide new insights into how neurons manage the buildup of harmful substances and how failures in these processes contribute to diseases like Parkinson's. This could open up new therapeutic strategies aimed at preventing or slowing the progression of neurodegeneration by targeting these secretory pathways.

Hearing loss has been uncovered to be a major risk factor for dementia. However, the biological processes underlying this link remain poorly understood. The goal of the PhD is to determine how different forms of hearing loss may accelerate the emergence of neurodegenerative diseases, and to decipher the underlying mechanisms. To tackle this question, mouse models of Alzheimer’s disease will be crossed with different cutting-edge mouse models for hearing impairment. By using a very innovative behavioral system, the PhD student will analyze the consequences of hearing loss in mouse models of Alzheimer’s disease on cognitive performances. Based on preliminary results of the hosting lab showing that different forms of hearing loss lead to impairment of the cerebrovascular system, the student will also investigate whether the cerebrovascular deficits linked to hearing impairment could promote dementia, using biological markers of neurodegenerative disorders.

The detection and processing of sounds operated by the peripheral and central auditory systems underlie our abilities to communicate through language and to appreciate music. Dysfunctions of the auditory system underlie a large variety of disorders like for example deafness of peripheral or central origins, speech perception difficulties especially in noisy environments, and dyslexia, and are also associated with more severe psychiatric disorders. According to the World Health Organization (WHO), 466 million people worldwide, including 34 million children, currently have disabling hearing loss. In addition, hearing loss has been uncovered to be a major risk factor for dementia[1]. People with mild, moderate and severe hearing loss have risks of developing dementia later in life two, three and five times higher, respectively than people without hearing loss. However, these observations are purely correlative, because the biological processes linking hearing loss to dementia remain poorly understood. This lack of understanding represents a missed opportunity for the treatment of some forms of dementia, as hearing loss is the most significant risk factor for dementia that is potentially modifiable (9% of patients)[1]. By 2040, more than 80 million people will suffer from dementia. Hearing restoration/rehabilitation could therefore potentially alleviate the symptoms of dementia in more than 7 million people. The key issue is therefore to determine whether hearing restoration can help to prevent cognitive decline and to decipher the underlying mechanisms.
Several mechanisms, which are not mutually exclusive, have been proposed to explain the link between hearing loss and increased dementia risk [2]. A first one is a common pathology that would affect the cochlea and ascending auditory pathways (causing hearing loss) and the cortex (causing dementia). A second possible mechanism is that hearing loss results in a cascade of pathophysiological mechanisms by the impoverishment of sensory inputs to the brain affecting its structure and function. A third mechanism is that people with hearing impairment could use greater cognitive resources for sound processing, making these resources unavailable for other aspects of cognition. The goal of the PhD is to determine how different forms of hearing loss may accelerate the emergence of neurodegenerative diseases, and to decipher the underlying mechanisms.
To tackle this question, the PhD candidate will make use of a mouse model of Alzheimer’s disease that will be crossed with different cutting-edge mouse models for hearing impairment including models of congenital deafness, progressive hearing loss, and auditory neuropathy, the latter mimicking sound intelligibility deficits often present in the elderly [3]. The candidate will mainly focus on two major goals. First, by using a very innovative behavioral system where animals live in colony and are monitored for their neural activity, auditory performance [4] and social interactions, the PhD student will analyze the consequences of hearing loss in mouse models of Alzheimer’s disease on cognitive performances. Behavioral tests including novel object recognition test, memory tests, sleep performances and advanced analysis of social interactions within a mouse colony will be used. Based on preliminary results of the hosting lab showing that different forms of hearing loss lead to impairment of the cerebrovascular system [5], we will investigate whether the cerebrovascular deficits linked to hearing impairment could promote dementia, using biological markers of neurodegenerative disorders associated with dementia as a readout.
This PhD project may, for the first time, reconstitute the chain of mechanistic events potentially linking hearing loss to dementia, by demonstrating a faster onset of cognitive deficits and a faster accumulation of dementia biomarkers in deafened models of Alzheimers disease. Our work will provide a scientific basis for evaluating the use of hearing aids to decrease the risk of dementia in the general population suffering from progressive deafness. This is of major importance because our findings could be rapidly translated into clinical practice, through more systematic screening of hearing deficits in the general population and greater indications for hearing aids instigated by changes in health policy.

The PhD will take place in the “Plasticity of Central Auditory Circuits” team at the Hearing Institute, a research center of Institut Pasteur. The laboratory is composed of 10 members using a combination of approaches including behavior paradigms to test auditory perception, in vivo electrophysiology and in vitro electrophysiology (patch-clamp), and different sets of immunhistochemical techniques.

References:

1) Livingston, Gill et al. The Lancet, 2024; 404(10452):572-628. PMID: 39096926.
2) Griffiths TD, et al. Neuron. 2020. PMID: 32871106. Review
3) Occelli F, …, Gourévitch B. Neuroscience. 2019;404:184-204. PMID: 30769096.
4) Postal O, … , Michalski N, Gourévitch B. Front Behav Neurosci. 2020;14:588834. PMID: 33132864;
5) Kirst C, Skriabine S, …, Michalski N, Tessier-Lavigne M, Renier N. Cell. 2020;180(4):780-795.e25. PMID: 32059781.

Animals and humans process vast amounts of information but retain only the most relevant experiences, with emotions playing a key role in memory consolidation. Emotions, particularly stress, enhancing attention and arousal, influence brain regions like the hippocampus. In neuroscience, it is known that hippocampal reactivations help recall and consolidate experiences. In artificial intelligence (AI), strategies inspired by these reactivations, especially in reinforcement learning, have been shown to accelerate learning. When it comes to emotionally charged circumstances, the relationship between stress, its most prominent hormone, cortisol, and memory functions in the hippocampus, is complex. This project aims to explore the link between stress and hippocampal reactivations via a computational model, and test it in artificial agents. This bio-inspired approach could (a) reveal new insights into the generation of hippocampal replay mechanisms and (b) improve AI's learning efficiency.

Most animals and humans encounter an immense amount of information and experiences throughout their lives, yet they manage to retain and recall only the most relevant, or a partial, but meaningful representation of them. Not everything experienced is stored in long-term memory; emotions play a key role in this memory consolidation process [Bower (1983) Phil. Trans. Royal Soc. London. B, Bio. Sci. 302.1110:387-402]. Emotions strongly influence cognitive processes, by enhancing our level of arousal and attentional resources, modulating the activity of brain regions such as the amygdala, the prefrontal cortex, and the hippocampus [Tyng et al., (2017) Front. Psychol]. Since the work of Scoville and Milner [(1957) Journ. Neurosc. 9.8:2907-2918], the role of the hippocampus as a cognitive mapping center has been studied in neuroscience, particularly through neurophysiological studies on spatial tasks in rodents. This research highlighted the fact that certain patterns of sequential activation of hippocampal neurons, observed during task execution, are then replayed during sleep or periods of calm wakefulness in the animal. These activities have been called hippocampal reactivations and are recognized as a powerful mechanism used, particularly by place cells, to recall, organize, and consolidate past experiences and infer future ones.
From an artificial intelligence (AI) perspective, it is also known that offline replay and updating of values associated with an agent’s actions can accelerate learning after a small number of real interactions with rewarding or punishing events (for example, Lin [(1992) Mach. Learn. 8-293-321]). The great interest in implementing computational strategies inspired by hippocampal reactivations in AI lies in tasks where past experiences and acquired knowledge must be re-evaluated and refined to perform better in future decision-making steps. This is the case with reinforcement learning (RL) paradigms [Sutton and Barto (2018) MIT press], where initially, when no prior knowledge is usually available, the best strategy is to interact with an environment through trial and error, and only when the level of experience increases the agent can exploit its previous knowledge to reach a sequence of actions approaching optimal behavior. In mammals and rodents, this consolidation of knowledge does not depend solely on the animal performing the same actions in the same situations: memories, particularly targeted recalls of experiences, are fundamental for effective learning from a small set of accumulated real experiences. Since the first RL algorithms which exploit strategies inspired by hippocampal reactivations [Sutton (1990) Mach. learn. proceed.], several researchers have proposed RL-based computational models inspired by these neuroscientific findings that are capable of reproducing the major experimental results regarding the quantity and type of reactivations generated [Khamassi and Girard (2020) Bio. Cybern. 114.2:231-248].
One of the major forces that impact the emotional state is stress, induced by the interaction between the animal, other animals, and the environment, both perceived through the animal’s senses. The influences from an external stressful environment and positive social interactions have been recently modeled by Khan and Cañamero [(2022) Front. Rob. AI 9] as the interaction between two hormones, cortisol and oxytocin, and their homeostatic balance. Concerning the role of emotional states related to stress and cortisol, Cañamero’s research has for example modeled their influence on learning and adaptation [Hiolle, Lewis and Cañamero (2014) Front. Neurorob. 8], behavior development [Lones, Lewis and Cañamero (2017) IEEE Trans. Cogn. Devel. Sys. 10.2-445-454], pain perception [L’Haridon and Cañamero (2023) ACII] and the appearance of compulsive behavior in Obsessive-Compulsive Disorder [Lewis, Canamero and Fineberg (2019) Comp. Psych.; Lewis and Canamero (2019) ACII]. Regarding the role of anxiety and stress on the hippocampus memory mechanisms, many key issues still need to be investigated: it has been known from various animal and human studies that stress impairs many memory functions at the level of the hippocampus [Kim, Pellman, and Kim (2015) Learn. Mem. 22.9:411-416] but, recently, Sherman et al. [(2023) Journ. Neurosc. 43.43:7198-7212] found that cortisol could also enhance the hippocampal associative memory functions.
The thesis project aim at a better understanding of the relationship between stress and the generation of hippocampal reactivations by means of a computational model that could be systematically tested in simulation or eventually on a real robot. This will bring the great advantage of testing functional hypotheses we have about the relationship between stress and hippocampal replay, with our computational model, in a very controlled experimental set-up, when we could repetitively simulate different emotional profiles and sensitivities on our agents. An additional point to this research will be to observe and analyzed the proposed model embodied on a robotic platform with the aim to look systematically at what are the effects of such a model in a very controlled context that still present elements of stochasticity and unpredictability that make a robot interacting with the real world a step closer to animal experiments, compared to pure simulations.
As analyzed in Massi et al. [Front.Neurorob. 16], the adoption of RL techniques inspired by hippocampal reactivations for AI has just begun. After validating a strategy that combines offline reactivations generation through a model-based agent with reactivations generated by a model-free method, the question remains of how to optimize the timing of this offline reactivation generation and its quantity. So, the proposed project aims to link the generation of offline reactivations to the internal emotional state of an agent. The idea is that, by following the concept of homeostasis and the regulation of key emotion-related hormones, such as oxytocin and cortisol [Avila-Garcia and Cañamero (2004) SAB], in a learning task, the agent will change its internal emotional state in relation to (a) its performance in task completion (e.g., in a spatial navigation task, effectively avoiding punishments to quickly reach a reward state), (b) egocentric external stimuli (e.g., in a spatial navigation task, proximity to walls or other agents approaching), and (c) a combination of the above two elements. This emotional internal state will allow the agent to trigger reactivations during moments of intense stress, for example, and not at just any moment in the task, where they may prove unnecessary. With this bio-inspired approach to AI, we will test and validate optimal strategies for offline reactivation generation within RL algorithms, with the aim of improving and accelerating artificial learning and disclose new possible driving mechanisms for hippocampal reactivations in neuroscience.
So far, in RL, many strategies inspired by neuroscientific evidence on the hippocampus have been proposed and tested to have a spontaneous and optimal generation of different types of replay-like activities [Mattar and Daw (2018) Nat. Neuro. 21.11:1609-1617; Diekmann and Cheng (2023) ELife 12]. Still a model that bases the generation of such reactivations on emotions and more specifically on the internal emotional state of an agent is lacking. That’s why the proposed project aims to theorize and test such a model which could spontaneously enable the generation of RL-based replay to improve memory consolidation and learning processing over different tasks and help shedding light on our understanding on the functional relationship between emotional states (stress in particular) and hippocampal replay. This research objective can be accomplished also thanks to the validation of our results against experimental behavioural data from rodents that could be provided by our collaborators.

Deficiencies in the regulation of proteostasis have been associated with neurodevelopmental disorders. Ubiquitination is a key posttranslational modification for the regulation of protein trafficking and degradation. In neurons, protein degradation needs to be controlled in an activity-dependent, synapse-specific and target-specific manner for proper circuit function and plasticity. A solution would be the use of adaptor proteins capable of modulating ubiquitination and degradation pathways and whose activity and localization can be regulated in response to neural activity. However, the identities of these adaptor proteins and their regulation in neurons remain largely unknown. The goal of the project is to understand the activity-dependent control of protein ubiquitination and degradation at synapses during synapse maturation and plasticity, and their dependence on NEDD4 ubiquitin ligases and the transmembrane adaptor SUSD4 for synapse-specific regulation.

Synapses are the fundamental units of neuronal circuits. Mutations in genes coding for synaptic proteins are recognized as important contributors to neurodevelopmental diseases such as schizophrenia and autism spectrum disorders (ASDs). Proper neuronal circuit maturation and plasticity relies on the recruitment of a variety of proteins with specific functions and the modulation of their stoichiometry at synapses. For example, at excitatory synapses, the choice between recycling and degradation of glutamate receptors is key for the direction of synaptic plasticity. Proper regulation of proteostasis at synapses is thus fundamental for brain function.
For each synaptic protein, its turnover is finely tuned via regulation of protein synthesis, transport, and degradation. Protein degradation is regulated via several pathways: the proteasome, the endolysosomal pathway and autophagy. One key parameter that needs to be better understood is the regulation of synaptic protein degradation in a protein target-selective, synapse-specific and activity-dependent manner. One solution for this type of very fine tuning in complex cells such as neurons is the use of adaptor proteins that can control degradation pathways and whose localization and function can be regulated via activity. The SUSD4 gene codes for a transmembrane protein containing four complement control protein (CCP) extracellular domains that are also found in synaptic proteins in invertebrates. The human SUSD4 gene is located in the region deleted in the 1q41-42 syndrome and mutations in SUSD4 have been associated with developmental delays, intellectual disabilities and ASDs. Therefore, misregulation of SUSD4-dependent processes might contribute to the etiology of neurodevelopmental disorders. Recently, two studies including one from the Selimi team showed that SUSD4 function is necessary for proper mouse behaviour, in particular motor coordination and learning. Our team discovered that the SUSD4 protein promotes AMPA-type glutamate receptor degradation, binds to key regulators of protein degradation, the NEDD4 ubiquitin ligases, and is necessary for proper synapse plasticity in cerebellar Purkinje cells. Recently, a study performed in cancer cells showed that SUSD4 promotes autophagy in an EGFR dependent manner, further highlighting its capacity to regulate degradation pathways.
Given its structure and its ability to bind and control degradation complexes, SUSD4 could thus function as an adaptor between various substrates and the degradation machinery and be modulated in an activity-dependent and synapse-specific manner. The goal of the proposed project is to provide a detailed analysis of the role of the CCP domain containing transmembrane protein SUSD4 in activity-dependent regulation of protein stoichiometry and synaptic function. We will combine biochemical characterization, dynamic analysis of synaptic proteins, neuron-specific genetic tools and functional analysis in the mouse to identify these molecular mechanisms in the cerebellum. Indeed, cerebellar Purkinje cells are a perfect illustration of the spatiotemporal complexity of proteostasis in neurons since each PC receives more than 150000 parallel fiber synapses whose plasticity needs to be fine-tuned in an input-specific and activity-dependent manner. Using this model
Our study will provide two important advances : 1) unravel the function of SUSD4, a candidate gene for neurodevelopmental disorders such as ASDs; 2) discover a general mechanism for the activity-dependent and target-selective regulation of degradation and proteostasis at synapses.
For this we will have the following specific aims:
1: Activity-dependent control of SUSD4 dynamics in spines: We will use state-of-the art imaging technologies to monitor SUSD4 recruitment in spines of cultured cerebellar Purkinje cells, and it dependence on binding to NEDD4 ligases.
2: Activity-dependent control of protein stoichiometry at synapses by SUSD4: We will combine already established biochemical assays and already characterized genetically modified mouse lines to test the consequences of SUSD4 loss of function on the molecular composition of synapses, in particular on the stoichiometry of neurotransmitter receptors.
3: Role of extracellular interactions in SUSD4 function. CCP domains, aka Sushi domains, are found in multiple proteins with neuronal functions, such as the C. elegans LEV-9 protein known for its role in acetylcholine receptor clustering. We will identify partners interactions with the extracellular domain of SUSD4 using affinity-purification of cerebellar synaptosomes using this domain as a bait and mass spectrometry. We will study the role of these interactions in the maturation and plasticity of Purkinje cell synapses.

Biologically realistic neural network models are crucial for comprehending brain functioning but cannot be verified in humans. Blood Oxygen Level Dependent functional magnetic resonance imaging (BOLD fMRI) is the primary source of information on human brain activity in clinical and basic research. It relies on neurovascular coupling (NVC), the functional link between neurons, astrocytes, and blood vessels. BOLD fMRI is interpreted statistically rather than on the underlying biological mechanisms because a quantitative NVC model is missing. Ultra-high magnetic field fMRI provides a microvascular readout closer to neuronal spatiotemporal activation than the macrovascular-weighted BOLD of standard fMRI. BOLD representation is outdated because it uses the vein-based balloon model. The PhD candidate will build a computational model of NVC recapitulating the interplay between neurons, astrocytes, and blood vessels. Interpreting novel BOLD fMRI data will advance basic and clinical research.

1) Background and significance
Biologically realistic neuronal networks modeling has been the object of intense efforts of bright theoretical neuroscientists [1]. Now, we can model relatively extensive brain networks with remarkable physiological detail [2]. Interestingly, these models are based on microscopic data derived from electrophysiology or live microscopy and produce predictions at the network level that are hard to verify, in particular in the human brain. The lack of experimental verification in humans does not allow for closing the loop: model ? predictions ? experimental data ? model refinement, which would boost the comprehension of human brain functioning.
BOLD fMRI is the technique of choice to gather information on human brain activity, and ultra-high magnetic field (UHF) MR scanners provide invaluable data quality and spatiotemporal resolution. However, BOLD signals are indirectly related to neuronal network activity, and the link is the neurovascular coupling. Our limited understanding of NVC, due to its high physiological complexity, has hampered the development of efficient NVC models [3,4,5,6,7]. Our research team has recently started addressing this need [8,9], and we noticed that:
- An NVC model must account for the principal molecular cascades and the temporal dynamics of the neurotransmitters involved in the coupling.
- Astrocytes must be integrated into the model at the synaptic and NVC level.
- NVC dynamics must be reconsidered in light of the most updated results from animal models and human BOLD fMRI data at UHF.
Astrocytes are key cells for NVC10: they support neurons’ energy demand and ensure efficient synaptic transmission, functions that are impaired in most neurological and neurodegenerative diseases, making them targets for biomarkers and pharmacological therapies. Astrocytes’ participation in the NVC has been historically controversial [11], and a causal demonstration of their involvement in NVC was provided only recently [12]. An increasing body of evidence highlighted the multifaceted role of astrocytes in NVC. Their main cell compartments, namely the cell body [11], the processes [13], and the end-feet [10,14], display heterogeneous spontaneous and stimulated calcium activity, making it difficult to assess a temporal correlation between NVC multiple mechanisms and astrocytes’ participation. In addition, astrocytes drive several NVC mechanisms, and each molecular pathway has not yet been completely dissected. Because of the features mentioned above, an early and late astrocytic contribution in NVC was expected and indirectly reported by several studies using multi-modal techniques and in vivo pharmacology [12,,15]. A 2017 review [16] listed the many key cellular steps of NVC and the participation of cell types; astrocytes are present in every step and have a dual role of actuators and targets of the molecular mechanisms. Altogether, these arguments foster the urgency of a model to recapitulate the multiple astrocytic contributions to each time step of the NVC.
Focusing on the delayed functional hyperemic response, a seminal paper by Helmchen and colleagues [15] showed a temporal correlation between a strong BOLD activation and astrocytes’ calcium transients. Subsequent studies by Gordon and collaborators further showed a neuronal activation threshold to trigger astrocytes’ participation in functional hyperemia [17], and, recently, a study by the same team demonstrated a direct involvement of astrocytes, leveraging the use of genetically modified mice and in vivo pharmacology [12]. In this study, the quantification of vascular dilation clearly shows that silencing astrocytes produced a significant decrease in the late NVC component. Still, it also slightly modifies the amplitude of the early component. Our team already simulated the astrocytes’ effect on the early vascular component, modeling the molecular pathways involving arachidonic acid and prostaglandin [8], providing a framework to link neuronal activity to the BOLD signal, and vice?versa, where neuronal activity can be inferred from the BOLD signal. Our model needs to be extended and completed, adding molecular pathways, such as the NO [18] and implementing the fast hemodynamic response function (HRF) deduced from micro- and mesoscopic data [12,19] and confirmed by BOLD fMRI at ultra-high field [20], instead of the canonical slow HRF. Next, the Ph.D. candidate will fit the model’s parameters on the early and late components of the vascular responses to assess which molecular pathways are crucial for each phase and account for astrocytes’ contribution. The predictions that we will then make with the new model will help design new stimulation/acquisition protocols for fMRI studies in humans [21].
2) Specific aims
Identify key astrocytic functional parameters leading to BOLD fMRI responses and their biological pathways in healthy and diseased subjects through a computational model of NVC fitted to experimental data. Make this model an open-access platform to be further improved and integrated with other features of the NVC and to work as a benchmark for testing proposed neuronal network models on human BOLD fMRI data.
3) Perspectives
The final goal of this Ph.D. project is to produce a model for the NVC that also considers astrocytes' contribution and helps understand its complex interdependent mechanisms. From the point of view of neurophysiology, this model is intended to solve the puzzle made by the many pieces of evidence accumulated on NVC molecular cascades and astrocytes’ contributions. Translating the model prediction to human fMRI will provide new avenues to verify biologically realistic neuronal network models and advance our understanding of brain function in health and disease.
4) Bibliography
1. Chang, YJ. et al. Sci Rep 14, 5145 (2024).
2. Ye C. et al.. Health Data Sci. (2024).
3. Kowia?ski P. et al., Neuroscience Research, 75(3), 171-183 (2013).
4. Denfield, GH. et al., Neuron, 91(5), 954 – 956 (2016).
5. Huneau C. et al., Frontiers in Neuroscience, 9 (2015)
6. Lundengård K. et al., PLoS Comput Biol 12(6): e1004971 (2016)
7. Sten S. et al. PLoS Comput Biol 19(1): e1010818 (2023)
8.Tesler, F. et al. A. Sci Rep 13, 6451 (2023).
9. Garcia, D.W., Jacquir, S. Biol Cybern (2024).
10.Otsu, Y. et al. Nature Neuroscience 18, 210–218 (2014).
11.Nizar, K. et al.J Neurosci. 33, 8411–8422 (2013).
12.Institoris, A. et al. Nat Commun 13, 7872 (2022).
13.Bindocci, E. et al. Science 356, eaai8185 (2017).
14.Rungta, R. L. & Charpak, S.. Nat Neurosci 19, 1539–1541 (2016).
15.Schulz, K. et al.. Nature Methods 9, 597–602 (2012).
16. Ladecola C. Neuron. 96(1):17-42 (2017).
17.Institoris, Á et al. J Cereb Blood Flow Metab. 35, 1411–1415 (2015).
18. Haselden WD. et al. PLoS Comput Biol 16(7): e1008069 (2020).
19.Aydin, A.K. et al. Nat Commun 11, 2954 (2020).
20. Polimeni J.R., Lewis L.D. Progress in Neurobiology, 207, 102174 (2021)
21. Gauthier C.J., Fan A.P. NeuroImage, 187, 116-127 (2019)
22.Boido D. et al. Nat Commun 10, 1110 (2019).
23. Tran C.H., Gordon G.R. Microcirculation. 22(3):219-27 (2015).

Astrocytes in the brain constitute a predominant influence for the cerebrovascular system controlling perivascular homeostasis, blood-brain barrier integrity, crosstalk with the peripheral immune system, endothelial transport, or vessel contractility in response to neuronal activity. Astroglial regulation towards the cerebrovascular system is set at the level of the gliovascular unit, a specialized interface composed by astrocyte perivascular processes that fully sheath the brain vessels. Such morphological and functional polarity is of critical importance for the brain physiology and is altered in megalencephalic leukoencephalopathy with subcortical cysts (MLC), a rare type of leukodystrophy. The goal of the present project is to decipher gliovascular alterations in this specific pathological context.

MLC1 is an astrocyte-specific membrane protein whose defect leads to megalencephalic leukoencephalopathy with subcortical cysts (MLC), a rare type of leukodystrophy characterized by an early-onset macrocephaly and a progressive white matter vacuolation leading to ataxia, spasticity, and cognitive decline. MLC1 forms a membrane complex at the junctions between astrocyte perivascular processes (PvAPs). In pilot studies, we showed that MLC1 is a marker of the PvAP postnatal maturation. We then analyzed the gliovascular unit (GVU) in the MLC mouse model deleted for MLC1 (Mlc1 KO). We determined that absence of MLC1 modifies the development of astrocytic morphology and polarity and the organization of astrocytic perivascular coverage. These changes correlated to an altered development of vascular smooth muscle cells contractility which regulate the cerebral blood flow. We also observed an altered neurovascular coupling as well impaired cerebrospinal/interstitial parenchymal fluid circulation which could contribute to the water accumulation and the megalencephaly observed in the mouse model and in MLC patients. This work suggested that MLC is a developmental pathology of the GVU with PvAP and cerebrovascular defects being primary events in the MLC pathogenesis.

We now propose to further study the physiopathological process of MLC. In particular, we aim to uncover the unknown links between astrocytic perivascular coverage and vascular alterations in MLC. This will be done through a multidisciplinary approach combining transcriptomic analysis of the vascular compartment, gene therapy to restore the expression of MLC1 in astrocytes, imaging analysis of the gliovascular interface and in vivo analysis of cerebrovascular functions such as neurovascular coupling, blood flow and brain drainage. Importantly, this project is based on validated experimental approaches of molecular biology and imaging as well as on an established collaborative network.

This project will be crucial to characterize the cerebrovascular pathology linked to MLC. It will allow to potentially propose novel therapeutic approach to cure MLC. It will also highlight how astrocytes and the cerebrovascular systems work in concert to regulate brain physiology.


References of the laboratory for this project:
Cohen-Salmon, M., L. Slaoui, N. Mazare, A. Gilbert, M. Oudart, R. Alvear-Perez, X. Elorza-Vidal, O. Chever, and A.C. Boulay. (2021). Astrocytes in the regulation of cerebrovascular functions. Glia. 69:817-841.

Gilbert, A., X. Elorza-Vidal, A. Rancillac, A. Chagnot, M. Yetim, V. Hingot, T. Deffieux, A.C. Boulay, R. Alvear-Perez, S. Cisternino, S. Martin, S. Taib, A. Gelot, V. Mignon, M. Favier, I. Brunet, X. Decleves, M. Tanter, R. Estevez, D. Vivien, B. Saubamea, and M. Cohen-Salmon. (2021). Megalencephalic leukoencephalopathy with subcortical cysts is a developmental disorder of the gliovascular unit. Elife. 10.

Gilbert, A., X.E. Vidal, R. Estevez, M. Cohen-Salmon, and A.C. Boulay. (2019). Postnatal development of the astrocyte perivascular MLC1/GlialCAM complex defines a temporal window for the gliovascular unit maturation. Brain Struct Funct. 224:1267-1278.

Slaoui, L., A. Gilbert, A. Rancillac, B. Delaunay-Piednoir, A. Chagnot, Q. Gerard, G. Letort, P. Mailly, N. Robil, A. Gelot, M. Lefebvre, M. Favier, K. Dias, L. Jourdren, L. Federici, S. Auvity, S. Cisternino, D. Vivien, M. Cohen-Salmon, and A.C. Boulay. (2023). In mice and humans, brain microvascular contractility matures postnatally. Brain Struct Funct. 228:475-492.

We investigate brain malformations, revealing perturbed, and clarifying normal, development mechanisms. Focusing on abnormal neuron localisation we aim to shed light on the impact of mutations, e.g. in DCX, a microtubule-associated protein, and in CERS, ceramide synthases, associated with epilepsy and intellectual disability, and to explore ways of correcting defects. Our data suggest important membrane trafficking and organelle defects in several models, which need pursuing.

Aims for the PhD project (progenitors and neurons, control and mutant cells) are to:
a) pinpoint disrupted cell processes, by visualizing mutant organelles and membranes using fluorescent assays, confocal, super-resolution and live microscopy, as well as functional assays.
b) identify abnormal membrane compositions in mutant cells by isolating membrane fractions, followed by omics studies.
c) rescue mutant phenotypes by replenishing the proportions of correctly functioning organelles and membranes.

The strategy of our group is to discover new regulation/dysregulation mechanisms during cortical development, using disease genes and key models as starting points (1). The pathogenic mechanisms for cortical malformations (rare disorders, e.g. heterotopia) and autism spectrum disorder (ASD) are not well-known. Collaborating with clinicians, we identify key gene mutations in patients, aiding classification and diagnosis, as well as often revealing unexpected molecular and cellular patho-mechanisms (2-10). Our projects query cortical progenitors, immature migrating, as well as post-migratory neurons. Projects are translational (collaborations with French & European clinical networks, and expert researchers for studying human neural cells in 2 and 3D (organoid) cultures, 8).

Objectives: Our recent data suggest trafficking and/or structural defects of organelles (ribosomes, mitochondria and primary cilia) in polarised cells in several cortical malformation models (5,11). We hypothesize that disruption of organelle trafficking and homeostasis can greatly perturb cortical development. This is an understudied area and our preliminary data underline the importance of investigating this further. Our projects are based on key neocortical and hippocampal genetic heterotopia models (knockout, knockin, knockdown). Importantly for heterotopia, we identify strong phenotypes in mouse models and also we have begun to identify relevant perturbed mechanisms in human in vitro models (e.g. derived from induced pluripotent stem cells). Approaches include the use of Omics data, single cell and brain analyses, using diverse, adapted imaging approaches.

We focus on:
a) mitochondrial transport and function during neuronal migration (7, R. Belvindrah);
b) membrane movement from the endoplasmic reticulum and mitochondria, particularly studying ceramide containing membranes (present e.g. in primary cilia 12,13, Chinnappa in preparation) in progenitors and neurons.
Investigating these cell processes, we aim to shed light on rare gene mutations, e.g. in DCX, a microtubule-associated protein, and ceramide synthase (CERS) enzymes found in human patients with epilepsy and intellectual disability, and to explore ways of correcting defects in mutant cells. These data should help further decipher cortical malformations, as well as elucidating normal development mechanisms.

The PhD project will hence focus on studying organelle and vesicle trafficking, lipid composition and protein localization to membranes, comparing human in vitro cultures to mouse cells. An important goal is to systematically compare and study the neural membranes and organelles of cells incorporating the different patient mutations. These combined studies will reveal substantial new data for the novel question of the role of membrane exchange and contacts within the cell, protein and lipid trafficking and its consequences in nervous system development and function.

Aims for the PhD project (progenitors and neurons, control and mutant conditions) are:
a) To pinpoint individual disrupted cell processes, by visualizing and characterizing mutant organelles and membranes using selected fluorescent sensors, confocal, super-resolution and live microscopy, as well as functional assays
b) To identify abnormal membrane compositions in mutant cells by isolating different membrane fractions (e.g. vesicles), followed by proteomic and lipidomic studies.
c) To rescue mutant phenotypes by replenishing the proportions of correctly functioning organelles and membranes.

We have dedicated staff in complementary domains to help pursue our questions. We also regularly recruit post-docs, PhD and Master students, including from around the world. With our harmonious and well-balanced structuration, team meetings, mentoring and lab organization, all favorising mutual support and scientific integrity, we are confident that our goals can be appropriately reached. We regularly publish in open access journals, and describe our work in different settings, including to the public (high school students, patient associations, fairs).

References
1) Romero et al. Genetics and mechanisms leading to human cortical malformations.
Semin Cell Dev Biol. 2018;76:33-75. PMID:28951247
2) Belvindrah et al. Mutation of the ?-tubulin Tuba1a leads to straighter microtubules
and perturbs neuronal migration. J Cell Biol. 2017 Aug 7;216(8):2443-2461.
PMID:28687665
3) Bizzotto et al. Eml1 loss impairs apical progenitor spindle length and soma shape
in the developing cerebral cortex. Sci Rep. 2017 Dec 11;7(1):17308. PMID:29229923
4) Khalaf-Nazzal et al. Early born neurons are abnormally positioned in the
doublecortin knockout hippocampus. Hum Mol Genet. 2017 Jan 1;26(1):90-108.
PMID:28007902
5) Uzquiano et al. Mutations in the Heterotopia Gene Eml1/EML1 Severely Disrupt
the Formation of Primary Cilia. Cell Rep. 2019;28:1596-1611. PMID:31390572
6) Penisson et al. Lis1 mutation prevents basal radial glia-like cell production in the
mouse. Hum Mol Genet. 2022 Mar 21;31(6):942-957. PMID:34635911
7) Stouffer et al. Doublecortin mutation leads to persistent defects in the Golgi
apparatus and mitochondria in adult hippocampal pyramidal cells. Neurobiol Dis
2022;168:105702. PMID:35339680
8) Jabali et al. Human cerebral organoids reveal progenitor pathology in EML1-linked
cortical malformation. EMBO Rep. 2022;23(5):e54027. PMID:35289477
9) Romero et al. Novel role of the synaptic scaffold protein Dlgap4 in ventricular
surface integrity and neuronal migration during cortical development. Nat Commun.
2022;13(1):2746. PMID:35585091
10) Klingler, et al. Mapping the molecular and cellular complexity of cortical
malformations. Science. 2021 371(6527):eaba4517. PMID: 33479124.
11) Zaidi et al. Forebrain Eml1 depletion reveals early centrosomal dysfunction
causing subcortical heterotopia. J Cell Biol. 2024 Dec 2;223(12):e202310157. PMID:
39316454.
12) Ehses et al. RNA Transport: From Head to Toe in Radial Glial Cells. Curr Biol.
2016;26(24):R1285. PMID:27997841
13) Tripathi et al. Palmitoylation of acetylated tubulin and association with ceramiderich
platforms is critical for ciliogenesis. J Lipid Res. 2021;62:100021. PMID:33380429

The human brain is a complex, self-organizing system where neurons transmit information via ion channels and communicate through synapses. Mathematical models have helped researchers explore how local population dynamics and brain network topology influence overall brain activity and synaptic/structural neuroplasticity as well as the brain’s ability to reorganize in response to various stimuli. This research aims to create a large-scale spiking neural network (SNN) model to investigate mechanisms of memory formation, storage, and retrieval, focusing on the hippocampus. Using a Hodgkin-Huxley-based model, the study will explore dopamine's effects on synaptic plasticity and test various connectivity patterns to find optimal configurations for memory processes. Additionally, the project will computationally assess the impact of emotional states on the excitation-inhibition balance in hippocampal circuits, aiming to enhance our understanding of neural mechanisms underlying memory.

The human brain is a complex, self-organizing system. Researchers aim to understand its core principles to improve brain models, explore disease neurobiology, and develop treatments. Neurons, the brain's basic units, transmit information via voltage-gated ion channels and communicate through chemical and electrical synapses. Neuronal networks generate brain functions such as perception, thought, and memory. Pathological activity patterns, linked to disorders like epilepsy and migraines, often result from genetic mutations affecting ion channels and synaptic function. [Russell et al. (2013) Annu. Rev. Neurosci. 36:25]. Neuronal activity in individual nodes is typically modelled by a system of ordinary differential equations. Such types of mathematical models provide a theoretical framework for investigating the impact of the localized population dynamics combined with the complex topology of the brain network on the overall brain activity [Deco et al. (2008) PLoS Comput Biol 4:e1000092; Sanz-Leon et al. (2015) NeuroImage 111:385) and its dynamics [Popovych et al. (2021) NeuroImage 236:118201; Manos et al. (2023) Front. Comput. Neurosci. 17:295395; Courson et al. (2024) Front. Comput. Neurosci. 18:1360009].
Neuroplasticity, also known as neural or brain plasticity, refers to the brain’s ability to adapt and reorganize in response to factors such as learning, environmental changes, practice, or stress [Voss et al. (2017) Front. Psychol. 8:1657]. This adaptability occurs through both synaptic plasticity, where neurons modify synaptic transmission, and structural plasticity, involving the formation or retraction of neural connections and changes in cell morphology [Kehayas & Holtmaat (2017) The Rewiring Brain 3-26]. In [Manos et al. (2021) Front. Physiol. 12:716556] the authors computationally explored long-term changes in structural plasticity triggered by external stimulation to optimize stimulation dosage while in [Anil et al. (2023) PLoS Comput. Biol. 19:e1011027] the authors simulated a recurrent neural network with homeostatic structural plasticity, revealing its importance in stimulation protocols.
Recent studies have underscored the hippocampus's susceptibility to neuroplastic changes in neurodegenerative diseases, impacting cognitive and emotional functions [Weerasinghe et al. (2022) Int. J. Mol. Sci. 23:3349]. Synaptic structural alterations, crucial for learning and memory, have been observed under high-resolution time-lapse imaging [Ma and Ta (2022) Semin. Cell Biol. 125:84]. In the prefrontal cortex, spiking activity patterns supporting cognitive tasks like working memory can be learned and replayed using biologically realistic plasticity rules [Sarazin et al. (2021) Front Neural Circuits 15:648538]. Models of hippocampal circuits, including CA1, CA3, and DG regions, provide insights into long-term memory mechanisms [Chua and Tan (2017) AAAI Tech. Rep. SS-17-07]. Additionally, silent cell assemblies contribute to memory persistence, influenced by neuromodulation that balances synaptic connectivity [Gallinaro et al. (2021) PLoS Comput. Biol. 17:e1009593; Fuchsberger & Paulsen (2022) Curr. Opin. Neurol. 75:102558; Tesler et al. (2024) Front Comput Neurosci. 18, 1432593].
The hippocampus has been recognised as a predominant brain structure for the consolidation of new information from short-term to long-term memory [O’Keefe and Dostrovsky (1971) Brain Res. 34:171]. Many studies have been conducted since, given the increasing evidence of the importance of hippocampal cell units for self-localisation and information recall. The current consensus is that the hippocampus displays a particular activity pattern, called sharp wave ripples (SWR), at a frequency of 150-200 Hz, which is temporally compressed with respect to the timescale of the neural activity happening during the real spatial experience, and thus enhances spike-timing dependent plasticity (STDP) [Dan and Poo (2004) Neuron 44:23]. Hippocampal SWR could encode temporally structured spatial patterns and drive the initial storage and the later retrieval of relevant experience [Pfeiffer (2020) Hipp. 30:6]. Models of memory formation, consolidation, and retrieval have been proposed at different levels of brain abstraction, from decision-making [Cazé et al. (2018) Journ. Neurophys. 120:2877] to spiking neural networks (SNN) [Tan et al. (2013) IEEE CIS], tested on artificial agents [Massi et al. (2022) Front. Neurorob. 16: 864380] and compared to animals’ neural recordings [Mattar and Daw (2018) Nat Neurosci. 21:1609]. Exogenous stimuli and the subsequent emotions play an important role in this, by modulating the activity of brain areas such as the amygdala, the prefrontal cortex and the hippocampus [Tyng et al. (2017) Front.Psychol. 8:1454]. In animal and human behavior, the two most studied conditioning stimuli concern sustenance and survival and they can be of opposite emotional strength, with two different neural circuits involved in the process [Yacubian et al. (2006) J Neurosci 26:9530].
From a decision-making and reinforcement learning (RL) perspective, the question was addressed recently in [Bryzgalov, Massi, et al., in preparation], and preliminary results show that higher learning rate and more intense replay activity are needed while learning to avoid than learning to approach. Brain circuits for encoding aversive and appetitive stimuli can be considered as orthogonal in the brain [Tye (2018) Neuron 100:436], but the same neurotransmitters, such as dopamine, could be implied in the learning process. In fact, the release of dopamine is linked to unexpected reward signals during learning, but it is shown that it can be linked to unexpected punishment too [Sands et al. (2023) Sci. Adv. 9:eadi4927]. Dopamine but also other neurotransmitters, has been proved to enhance synaptic consolidation and recall, but the precise dynamics of this mechanism have just started to be investigated [Lehr et al. (2022) Sci. Rep. 12.1:17772]. The dopaminergic influence in a SNN model of the striatum has been modeled for RL tasks [González-Redondo et al. (2023) Neurocomp. 548:126377] and, it has also been implemented as the joint action of short-latency excitation and long-latency inhibition for the modulation of both STDP and neuronal excitability [Chorley and Anil (2011) Front.Comp.Neuro. 5:21]. Finally, [Bono et al. (2023) Elife, 12:e80671] have used a mechanism inspired by hippocampal replay to learn successor representations and cognitive maps in a SNN.
The Research Objectives (ROs) of this proposal can be summarized as follows:
- RO1: Large scale SNN setup and implementation of synaptic/structural plasticity rules and neuromodulation features: Theoretical and computational understanding of the relevant combined relevant mechanisms for memory formation, storage and retrieval.
- RO2: Study the effects in the neural excitation-inhibition balance of positive and negative delivered emotional states in the hippocampus large scale SNN.
We will use a Hodgkin-Huxley-based spiking neuron model that supports a wide range of firing patterns, including tonic spiking, bursting, and seizure-like events [Bandyopadhyay et al. (2021) J. Comput. Neurosci. 50:33]. Such a SNN model allows us to implement and investigate dopamine effects via voltage-gated calcium and potassium channels while will allow us to design large neural networks to explore various structural and synaptic connectivity patterns, aiming to identify configurations that favor memory formation and retrieval. More details can be found in the “Feasibility of the project in 3 years - Specify the steps” section.
In summary, this research proposal will allow us to build a large-scale SNN to (i) explore different mechanisms associated with optimal memory storage and retrieval and (ii) computationally investigate the role of neural excitation-inhibition balance in these processes when positive and negative emotional states are delivered in the hippocampus.

Zebrafish have a special type of radial glia called radial astrocytes (RAs). In contrast to those of mammals, RAs do not degenerate and display analogous functions to those of astrocytes. In a recent study, we found that in the optic tectum of zebrafish, RAs synchronize their Ca2+ transients immediately after the end of a spontaneous escape behavior. The RA synchronizations modulated the direction selectivity and the functional correlations among tectal neurons. This mechanism supports freezing behavior following a switch to an alerted state. Here, we propose to investigate the neuronal pathways and mechanisms underlying the behavioural-state transition. This study will shed light on the neuronal and glia pathways and mechanisms across the entire brain, underlying the behavioral-state transitions.

Zebrafish have a special type of radial glia called radial astrocytes (RAs). In contrast to those of mammals, RAs do not degenerate and display analogous functions to those of astrocytes. Previous studies showed that RAs in the hindbrain enable larvae to switch to a quiescent behavioral state. In the forebrain, perturbations of the glia-neuron interactions lead to the emergence of epileptic seizures?. In a recent study, we found that in the optic tectum of zebrafish, RAs synchronize their Ca2+ transients immediately after the end of a spontaneous escape behavior. Using two-photon Ca2+ imaging, optogenetics and ablations we showed that RA synchronous Ca2+ events are mediated by the locus coeruleus. The RA synchronizations modulated the direction selectivity and the functional correlations among tectal neurons. This mechanism supports freezing behavior following a switch to an alerted state. Here, we propose to investigate the neuronal pathways and mechanisms underlying the behavioural-state transition. For this purpose, we will use transgenic zebrafish expressing the genetically encoded Ca2+ indicator GCaMP7f and the opsin ReaChR (optogenetic stimulation of neurons), in combination with a 3D two-photon system and a digital micromirror device for pattern illumination for opsin activation.
Supported by preliminary results, we will investigate the hypothesis that the medula oblangata gates the behavioral-state transitions, by integrating information from the internal state of the animal (e.g. sensory inputs form the gut and/or oxygen levels). In addition, we will investigate the downstream neuronal and glial pathways following the spontaneous activation of the medula oblangata.
Overall, this study will shed light on the neuronal and glia pathways and mechanisms across the entire brain, underlying the behavioral-state transitions.

The ORGAI project brings together two Vision Institute teams expert in imaging and organoid development, and aims to develop AI and machine learning tools to quantify cell specific data. We aim to validate our novel live imaging approach named DFFOCT for label-free, longitudinal imaging of patient-derived organoid models. Our goal is to demonstrate that DFFOCT, combined with AI-driven analysis, can create relevant numerical twins of organoids to predict which drugs will be most effective for individual patients. The PhD student will develop new machine learning and AI models, enabling automatic high-throughput data analysis. These techniques will make it possible to quantify the morphology and viability of each of the organoid's cells, identify their cell types, and monitor their fate. These digital tools will increase the specificity of label-free microscopes, predict biological results from a minimum number of experimental data, and generalize results.

Inherited retinal dystrophies (IRD) that cause definitive loss of photoreceptors typically result in permanent visual impairment. Retinitis Pigmentosa (RP) is the most common form of IRD, with a prevalence of 1 out of 3500–5000, for more than 1.5 million cases worldwide. Mutations in Rhodopsin (RHO) lead to the death of rod photoreceptors, followed by a secondary death of cone photoreceptors. In France, the P347L (Class I mutation) appears to be the most prevalent[1] while the P23H mutation (Class II) is the most common mutation in the USA[2] . Class I RHO mutants fold normally in cells but affect post Golgi trafficking and outer segment targeting contributing to photoreceptor cell death[3]. In contrast, Class II mutants cause RHO misfolding resulting in retention in the endoplasmic reticulum (ER) causing an ER stress leading to cell death[4]. Although the mechanisms of photoreceptor death carrying a mutation in RHO are not yet fully defined, preventing and rescuing the retina is a major challenge for which pharmacotherapy can be an answer[5,6]. In fact, neuroprotection has emerged as a strategy for delaying photoreceptor death and preserving vision[7]. One advantage of the neuroprotection strategy is that it can have the capacity to slow photoreceptor degeneration regardless of the underlying causative pathway and may even be generalizable to other retinal neurodegenerative diseases[7]. Currently, there are no approved treatments for IRD. Despite the blossoming of recent progress in gene and cell therapies, the use of neuroprotective agents remains a front-line approach for a spectrum of neurodegenerative diseases, including IRD disorders caused by photoreceptor death[8]. The approved neuroprotectants have documented pharmacokinetics, bioactivities and demonstrate minimal adverse effects, facilitating long-term administration and compliance that are normally required for the treatment of chronic neurological diseases[9].

Today, the possibility to make a “Disease-in-a-Dish” with patient-based cell models - using induced pluripotent stem cell (hiPSC)-derived retinal cells - represents a chance for drug discovery. These highly relevant cellular models offer a unique opportunity for studying the effects of specific gene defects in the human context to better understand the disease and find anti-degenerative treatment[5,10]. Sacha Reichman’s team at the Vision Institute is working on identifying neuroprotective compounds in hiPSC-derived retinal organoid disease models[11].

To image these organoids live, Kate Grieve’s research group has pioneered a novel, label-free imaging technique called Dynamic Full-Field Optical Coherence Tomography (DFFOCT)[12-15]. This method detects all living cells within complex samples and measures their local activity, offering valuable insights into cell metabolism[12,13], stress[15], mitosis[13], and apoptosis[12]. DFFOCT has already demonstrated its utility in long-term imaging of retinal organoids over several weeks, without any sign of phototoxic effects[13]. However, while DFFOCT contrast relies on the intrinsic optical and biophysical properties of tissues, its specificity remains limited and interpretation can be challenging. We hypothesize that biological specificity can be enhanced through a multi-scale analysis, combining information on cell morphology, activity, metabolism, and scattering properties. By incorporating machine learning and AI, we aim to achieve virtual staining of samples [16,17], offering contrast similar to fluorescence imaging without the need for labelling. Our team has previously published on AI analysis of DFFOCT data in the context of cancer biopsies, and would now like to translate this to retinal organoid data [17].

The primary scientific objective of the ORGAI project is to validate DFFOCT as a versatile and cost-effective method for label-free, longitudinal imaging of patient-derived organoid models. Our goal is to demonstrate that DFFOCT, combined with AI-driven analysis, can create relevant numerical twins of organoids to predict which drugs will be most effective and least toxic for individual patients.

To achieve this, the ORGAI project will take the following steps:

- Patient-based models of organoids will be developed by Sacha Reichman’s team at the Vision Institute. This step develops retinal organoids from patients with inherited retinal dystrophies and tests neuroprotective molecules identified by the Vision Institute to assess structural and functional restoration.

- High throughput label-free microscopes developed by Kate Grieve’s team at the Vision Institute will be used to image the organoids. DFFOCT has proven useful to follow cell viability and cell stress in retinal cell organoids over several weeks. The retinal organoid models of RP undergoing degeneration and with the drug screenings for neuroprotection will be followed with DFFOCT, forming an image database.

- Data Analysis and algorithm development. With DFFOCT, we can quantify the morphology and viability of all cells in the organoids. But AI and automatic data analysis are required to transfer such data into interpretable metrics and to perform multiscale analysis. Steps will involve segmentation and analysis of 3D spatial interactions to quantify organoid health at different stages; time prediction to predict the outcome of long-lasting toxicity and efficacy drug testing and compressed sensing to improve DFFOCT speed and reduce data volume; and finally aggregation of data from multiple organoids under different conditions to build several models.

We anticipate that digital tools developed in the ORGAI project in the specific context of identifying neuroprotective compounds in hiPSC-derived retinal organoid disease models may be generalizable to other samples imaged with DFFOCT label free live microscopy and could therefore beyond this project be applied to imaging with other groups involved in the DIM C-BRAINS network.

1. Audo, I. et al. Invest Ophthalmol Vis Sci 51, 3687–3700 (2010).
2. Mendes, H. F et al. Trends Mol Med 11, 177–185 (2005).
3. Athanasiou, D. et al.. Prog Retin Eye Res 62, 1–23 (2018).
4. Remondelli, P. & Renna, M. Front Mol Neurosci 10, 187 (2017).
5. Avior, Y. et al. Nat Rev Mol Cell Biol 17, 170–182 (2016).
6. Mendes, H. F. & Cheetham, M. E. Hum Mol Genet 17, 3043–3054 (2008).
7. Lin, J. B., et al. Ophthalmology science 2, (2022).
8. Wubben, T. J., et al. Curr Opin Ophthalmol 30, 199–205 (2019).
9. Mikitsh, J. L. et al. Perspect Medicin Chem 6, 11–24 (2014).
10. Mack, D. L., et al. Am J Phys Med Rehabil 93, S155–S168 (2014).
11. Reichman, S. et al. Stem Cells 35, 1176–1188 (2017).
12. Scholler J, et al. Light Sci Appl. 2020 Aug 17;9(1):140.
13. Monfort T, et al. Commun Biol. 2023 Sep 28;6(1):992.
14. Azzollini S, et al. Biomed Opt Express. 2023 Jul 1;14(7):3362
15. Groux K, et al. Commun Biol. 2022 Jun 10;5(1):575.
16. Bai B, et al. Light Sci Appl. 2023 Mar 3;12(1):57.
17. Scholler J, et al. J Med Imag [Internet]. 2023 Jun 1 [cited 2024 Mar 20];10(03). Available from: https://www.spiedigitallibrary.org/journals/journal-of-medical-imaging/volume-10/issue-03/034504/Automatic-diagnosis-and-classification-of-breast-surgical-samples-with-dynamic/10.1117/1.JMI.10.3.034504.full

We are proposing a comprehensive analysis of the role of CHRFAM7A, a human-specific gene, in biomedical disease models. CHRFAM7A is present in humans for the last 200 000 years, but not in non-human primates, after a fusion between the CHRNA7 nicotinic acetylcholine receptor (nAChR) alpha7 gene, and ULK-4, encoding for a serine/threonine kinase. The importance of this gene became clearly apparent in human genetic studies linking variation in CHRFAM7A to Alzheimer's disease (AD) and schizophrenia (SZ). They have identified copy number variation (CNV) and a two base-pair deletion leading to further truncation of the protein, referred to as the delta2bp variant. In SZ and bipolar disorder, upregulation of CHRFAM7A was observed in the brain and association studies suggest a correlation with the delta2bp variant. We are therefore proposing to carry out a detailed dissection of the underlying mechanisms.

Nicotinic acetylcholine receptors (nAChRs) are important transmembrane neurotransmitter receptors for acetylcholine in the mammalian brain (1). They modulate key neuronal functions such as excitability, synaptic transmission and plasticity. By coordinating the firing of groups of neurons, they change neuronal network states and consequently whole-brain responses to internal and external inputs, and have been linked to disease (2). They are also exogenously activated by nicotine, and are responsible for the events leading to nicotine addiction (3). The predominant receptors in the brain are alpha4 beta2* containing hetero-, and (alpha7)5 homo-pentamers (4). Being particularly expressed in the cortex and hippocampus (5), these receptors are considered to be important drug targets (6).
Owing to the high frequency and complexity of the CHRFAM7A CNV in the human population, understanding its functional impact is imperative for interpreting clinical trials targeting alpha7 nAChR. Despite its clear association to neuropsychiatric diseases, functional studies are sparse. In CHRFAM7A-transfected Xenopus oocytes, CHRFAM7A is a stoichiometric dominant-negative regulator of alpha7 nAChR.
We will use human induced pluripotent stem cells (hiPSC) to model CNV and coding variation using CRISPR-Cas9 mediated genome modification on an isogenic background. The cells will be differentiated into glutamatergic and GABAergic neurons, to extensively characterize their phenotypes in vitro, and in vivo. This is based on our initial published work using hiPSC in vitro and in vivo after transplantation (7–10).
This approach will be complemented by the detailed analysis of a transgenic mouse model expressing the human CHRFAM7A gene in the immune and nervous system. Based on initial behavioural findings, we will carry out detailed two-photon imaging in defined cell types expressing CHRFAM7A. This will be combined with the localized expression of CHRFAM7A and the delta2bp variant (11), using inducible lentiviral vectors.
Our preliminary data point to, and we will pursue further the findings that human CHRFAM7A containing receptors are significantly less responsive compared to rodent alpha7 nAChRs when activated by a combination of an alpha7 agonist and an alpha7 Positive Allosteric Modulator (PAM), like PNU-120596 (12). These dramatic differences in the pharmacological properties of the "human" receptors can explain failures in clinical trials targeting the rodent alpha7 when applied to human subjects. We will therefore define a novel pharmacology to fill this translational gap.
For the in vitro part of the project, we are using human induced pluripotent stem cell (hiPSC) lines from the Wellcome Trust Sanger Centre collection: (https://www.sanger.ac.uk/collaboration/hipsci/). These lines are fully sequenced at the genomic and transcriptomic level. All genetic engineering and potential genetic drifts during long-term cultures can therefore be validated also with respect to the starting material. The currently used line for our preliminary experiments outlined below contains two alleles each of CHRFAM7A and CHRNA7. The validation of hiPSC lines will be performed as we have described previously (13). This more laborious approach ensures that the iPSC starting material for our research program also allows at later stages to be transplanted into mouse brains for long-term development of mature neuronal networks as we have shown recently (9).


1. Neuron 76, 116–129
2. Curr. Opin. Neurobiol. 29, 88–95
3. Nat Rev Neurosci 11, 389–401
4. Annu Rev Pharmacol Toxicol 40, 431–458
5. Proc Natl Acad Sci U S A 104, 8155–8160
6. Nat Rev Drug Discov 8, 733–750
7. FASEB J. 31, 828–839
8. Sci. Rep. 9, 94
9. Dev. Biol. 461, 86–95
10. Sci. Rep. 10, 1–13
11. Neuropharmacology 96, 274–288
12. J. Neurosci. 25, 4396–4405
13. JCI 129, 2145–2162

The purpose of the project is to develop a novel diagnostic tool to identify older adults at risk of falling. With ageing, there is an increasing incidence of balance impairments, with dramatic impact on health and quality of life. Despite extensive study, laboratory and clinical measures of balance are poorly predictive of fall risk. In typical study designs, balance is assessed by exposing participants to repeated perturbations. The first response is often discarded from the analysis, as it is very different (often much stronger) than the subsequent, “habituated” response. The novelty in our approach is to study how subjects adapt, from the first trial response to subsequent habituated responses. Our hypothesis is that balance impairments are due to an inability to adapt postural control to changing environmental constraints. The project results will lay the foundation work for developing rehabilitation strategies to improve postural adaptation in impaired populations.

– Project’s objectives and research hypotheses –
The purpose of the project is to develop a novel diagnostic tool to identify older adults at risk of falling.
The first objective is to determine how young, healthy participants adapt sensorimotor control to alterations in the context. The second objective is to evaluate the effect of postural adaptation on daily balance and mobility in ageing. For this, we will study a population with large inter-individual differences in balance and mobility, namely healthy adults over 60 years of age. Our hypothesis is that, in healthy older adults, balance and mobility impairments do not result from isolated sensory or motor deficits, but from an inability to adapt postural control to changing environmental constraints. The project results will lay the foundation work for developing rehabilitation strategies to improve postural adaptation in impaired populations.

– Motivation and state of the art –
With ageing, there is an increasing incidence of balance [1] and mobility [2] impairments, causing dramatic impact on health and quality of life [3]. Despite extensive study in the last thirty years, the laboratory and clinical measures of balance that have been developed are poorly predictive of fall risk, with very disparate reported levels of sensitivity, specificity and accuracy [4]. Our previous work shows that this may be because such studies neglect older adults’ inability to adapt postural control [5], [6]. Indeed, in typical study designs, balance is assessed by exposing participants to repeated perturbations in the laboratory. The first response is often discarded from the analysis, as it is very different (often much stronger) than the subsequent, “habituated” response [7]. This habituated response is not different between older fallers and non-fallers [8]. In daily living however, falling may occur after a single unexpected perturbation, if the person’s sensorimotor control was not adapted to the environment. The novelty in our approach is to study how participants adapt, from the first trial response to subsequent habituated responses.

– Methodology –
Participants will perform dynamic standing balance in two environments, designed to induce either maximal or minimal sensorimotor gains. Standing still requires the centre of pressure (CoP) and centre of mass (CoM) to remain vertically aligned. In the first task, participants will experience forwards and backwards translations of the support surface. In this context, maintaining balance requires a large and immediate movement of the CoP, which benefits from large ankle stiffness (achieved through co-contraction of ankle muscles) and feedback gain. In the second task, participants will experience toe-up and toe-down rotations of the support surface. This task requires maintaining the CoP immobile, and therefore minimal ankle stiffness and feedback gain. Participants will be randomly assigned to experience first a block of translations or a block of rotations, using a 6 degree of freedom perturbation platform available in the host laboratory.
In Experiment 1, 20 healthy young participants will be recruited. In Experiment 2, we will recruit healthy community-dwelling participants aged over 60. Balance [1] and mobility [2] impairments affect a third of this population. To ensure sufficient statistical power, we will recruit 45 participants to have 15 fallers and 15 participants with mobility impairment. Participant’s mobility will be evaluated using questionnaires [2], self-selected walking speed, and inertial sensors worn in daily life. They will then perform the experiment described above. To develop a diagnostic tool for detecting fall risk, participants will then be followed over a year to determine whether they experience a fall.

– Impact –
With ageing, impaired motor coordination results in an increasing occurrence of balance [1] and mobility [2] impairments. These are associated with a loss of independence, reduced quality of life, and an increased risk of falling, hospitalization, and premature death [3]. Falls in older adults pose a high economic burden for acute health care and rehabilitation [9]. To identify risk factors for falling, most studies use a normative approach: they compare sensorimotor control between a group at high risk (e.g. older adults) and a control group, typically healthy adults aged around 25 years [6]. The normative approach neglects a crucial aspect of sensorimotor coordination: neural learning. Indeed, altered postural control is not simply the result of an underlying deficit, but also of ongoing neural learning which allows the person to adapt to changes in their biomechanical and neural capacities, and compensate their deficit. To identify balance impairments, it is therefore essential to identify people who are unable to adapt their postural control to changing environmental constraints. We will determine how participants adapt postural adaptation across the adult lifespan. This will provide a novel diagnostic tool to identify older adults at risk of falling.

– References –
[1] T. Masud and R. O. Morris, ‘Epidemiology of falls’, Age Ageing, vol. 30 Suppl 4, pp. 3–7, Nov. 2001.
[2] J. Holmes, E. Powell-Griner, M. Lethbridge-Cejku, and K. Heyman, ‘Aging differently: Physical limitations among adults aged 50 years and over: United States, 2001-2007’, NCHS Data Brief, no. 20, pp. 1–8, Jul. 2009.
[3] J. A. Stevens, P. S. Corso, E. A. Finkelstein, and T. R. Miller, ‘The costs of fatal and non-fatal falls among older adults’, Injury Prevention, vol. 12, no. 5, pp. 290–295, 2006.
[4] L. Montesinos, R. Castaldo, and L. Pecchia, ‘Wearable Inertial Sensors for Fall Risk Assessment and Prediction in Older Adults: A Systematic Review and Meta-Analysis’, IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 26, no. 3, pp. 573–582, Mar. 2018, doi: 10.1109/TNSRE.2017.2771383.
[5] C. Le Mouel, R. Tisserand, T. Robert, and R. Brette, ‘Postural adjustments in anticipation of predictable perturbations allow elderly fallers to achieve a balance recovery performance equivalent to elderly non-fallers’, Gait & Posture, vol. 71, pp. 131–137, Jun. 2019.
[6] R. Dubbeldam, Y. Y. Lee, J. Pennone, L. Mochizuki, and C. Le Mouel, ‘Systematic review of candidate prognostic factors for falling in older adults identified from motion analysis of challenging walking tasks’, Eur Rev Aging Phys Act, vol. 20, no. 1, p. 2, Feb. 2023, doi: 10.1186/s11556-023-00312-9.
[7] J. H. J. Allum, K.-S. Tang, M. G. Carpenter, L. B. Oude Nijhuis, and B. R. Bloem, ‘Review of first trial responses in balance control: influence of vestibular loss and Parkinson’s disease’, Hum Mov Sci, vol. 30, no. 2, pp. 279–295, Apr. 2011, doi: 10.1016/j.humov.2010.11.009.
[8] R. W. Baloh, S. Corona, K. M. Jacobson, J. A. Enrietto, and T. Bell, ‘A Prospective Study of Posturography in Normal Older People’, Journal of the American Geriatrics Society, vol. 46, no. 4, pp. 438–443, 1998.
[9] S. Heinrich, K. Rapp, U. Rissmann, C. Becker, and H. H. König, ‘Cost of falls in old age: A systematic review’, Osteoporosis International, vol. 21, no. 6, pp. 891–902, 2010, doi: 10.1007/s00198-009-1100-1.

"Number sense", the ability to quickly perceive approximate object quantities in the visual scene, is a crucial cognitive capacity of humans and other animals thought to depend on higher-level brain regions. But how responses to number are derived from visual images and the role of lower-level brain structures in this is not well understood. This project proposes to investigate subcortical areas which have been largely neglected in the study of human cognition, although recent findings in other animals suggest they could contribute to visual number sense. Unlike most non-invasive cognitive neuroscience techniques, ultra-high field functional imaging can measure activity in small and deep subcortical structures. We will use 7 or 11.7T fMRI to study how subcortical visual regions distinguish numbers of objects and their non-numerical properties, compared to cortical regions. Our results could potentially on the longer term provide novel explanations for disturbed numerosity processing.

Numbers, an essential underpinning of human science and technology, do in their most concrete meaning refer to quantities of objects. The ability to rapidly perceive these in an approximate fashion, without serial counting relying on language, is found across human cultures and animal species (1). More advanced numerical abilities of humans, and their impairments in math learning difficulties such as dyscalculia, can to some extent be predicted by differences in the precision of this basic non-verbal “number sense” (2–4). As a consequence, substantial interest in the cognitive neuroscience community concerns the neural bases of visual number sense. Human brain regions associated with number processing encompass a particular fronto-parietal cortical network, and individual neurons selectively responsive to different numbers of items have been described in related brain areas of other animals (1, 5).
A key unresolved question is how such neuronal responses to numerosity are derived from the signals impinging onto our retina, where continuous physical variables co-vary with numbers of objects and a given number can lead to profoundly different stimulation depending on object types and their surroundings. Another question is how numerical responses in high-level brain areas of the human cerebral cortex fit with the presence of number sense in many animal species that lack similarly developed higher brain structures.
Visual inputs in vertebrate species are processed by a series of evolutionarily conserved regions in the midbrain and thalamus before reaching higher cognitive areas (6). The midbrain’s tectum which corresponds to the superior colliculus in mammals, while considered less important for conscious visual recognition in primates, can represent a map of the locations of salient objects in the environment with some degree of independence of those objects’ defining features (7, 8), which could make it particularly suitable as an intermediate step in the extraction of discrete numerosity. Although findings supporting a role of such early regions in numerosity processing have been absent or scarce until recently, results from optical imaging in zebrafish now show that number responsive neurons are indeed not limited to the forebrain but also found in the tectum (9), and specific early visual circuits that according to some views are comparable to those of the vertebrate tectum (10) are required for behavioral numerosity discrimination in flies (11). In humans, behavioral results suggesting that numerosity discrimination may partly rely on monocular parts of the visual system have been interpreted as pointing to an involvement of the subcortex (12), although they cannot determine the region of origin of these effects without ambiguity. In a reanalysis of human fMRI data from subcortical structures from one of our previously published studies (13), we find that pattern signals in regions of the thalamus which receive inputs from the superior colliculus do also to some extent distinguish between different numbers of visual items, although this study was not optimized to achieve best signal in the subcortex and did not precisely focus on distinguishing discrimination of numbers of items per se from differences in other visual features.
In this project, we therefore aim to benefit from the enhanced signal-to-noise ratio offered by functional imaging at ultra-high magnetic fields (14), to conduct more focused high-resolution acquisitions in subcortical areas of interest (superior colliculus and different parts of the thalamus) in addition to parietal cortex, and to understand how those regions represent numbers of objects across changes in non-numerical quantities and types of features defining the objects. The experiments will make use of fMRI at 7 or 11.7 Tesla (15). Given the prevalent focus on the cerebral cortex in human cognitive neuroscience, the idea that evolutionarily older subcortical areas should still have functional relevance for numerical abilities in humans may appear unconventional, but could also be seen as timely given recent reports of a role of regions like the superior colliculus in aspects of higher-level cognition (16). If confirmed, our hypothesis has the potential to profoundly change current ways of thinking about human number cognition, and possibly on the longer term contribute novel explanatory accounts of disturbed number processing (as in dyscalculia).
This project takes place within the Cognitive Neuroimaging Unit (UNICOG) located at the NeuroSpin center in the greater Paris area. The UNICOG laboratory has been a leader in the cognitive neuroscience of number processing and other areas of human cognition for many years and offers a stimulating international work environment. NeuroSpin is France’s most advanced neuroimaging center housing a multi-disciplinary combination of researchers, from experts in data acquisition methods to neuroscientists and clinicians. The center has in recent years developed the Iseult 11.7T human MRI scanner, the highest field strength currently available for human studies in the world. This project will be one of the first to attempt to exploit this system to answer relevant questions about the neural basis of cognition.

References
1. A. Nieder, Neuroethology of number sense across the animal kingdom. Journal of Experimental Biology 224, jeb218289 (2021).
2. J. Halberda, M. M. M. Mazzocco, L. Feigenson, Individual differences in non-verbal number acuity correlate with maths achievement. Nature 455, 665–668 (2008).
3. M. Piazza, et al., Developmental trajectory of number acuity reveals a severe impairment in developmental dyscalculia. Cognition 116, 33–41 (2010).
4. E. Castaldi, A. Mirassou, S. Dehaene, M. Piazza, E. Eger, Asymmetrical interference between number and item size perception provides evidence for a domain specific impairment in dyscalculia. PLoS ONE 13, e0209256 (2018).
5. M. Piazza, E. Eger, Neural foundations and functional specificity of number representations. Neuropsychologia 83, 257–273 (2016).
6. T. Isa, E. Marquez-Legorreta, S. Grillner, E. K. Scott, The tectum/superior colliculus as the vertebrate solution for spatial sensory integration and action. Curr Biol 31, R741–R762 (2021).
7. R. Veale, Z. M. Hafed, M. Yoshida, How is visual salience computed in the brain? Insights from behaviour, neurobiology and modelling. Phil. Trans. R. Soc. B 372, 20160113 (2017).
8. B. J. White, et al., Superior colliculus neurons encode a visual saliency map during free viewing of natural dynamic video. Nat Commun 8, 14263 (2017).
9. P. Luu, et al., Neural Basis of Number Sense in Larval Zebrafish. [Preprint] (2024). Available at: https://www.biorxiv.org/content/10.1101/2024.08.30.610552v1 [Accessed 4 September 2024].
10. J.-S. Joly, G. Recher, A. Brombin, K. Ngo, V. Hartenstein, A Conserved Developmental Mechanism Builds Complex Visual Systems in Insects and Vertebrates. Current Biology 26, R1001–R1009 (2016).
11. M. Bengochea, et al., Numerical discrimination in Drosophila melanogaster. Cell Rep 42, 112772 (2023).
12. E. Collins, J. Park, M. Behrmann, Numerosity representation is encoded in human subcortex. Proc. Natl. Acad. Sci. U.S.A. (2017).
13. S. Czajko, A. Vignaud, E. Eger, Human brain representations of internally generated outcomes of approximate calculation revealed by ultra-high-field brain imaging. Nat Commun 15, 572 (2024).
14. W. van der Zwaag, A. Schäfer, J. P. Marques, R. Turner, R. Trampel, Recent applications of UHF-MRI in the study of human brain function and structure: a review. NMR Biomed (2015).
15. N. Boulant, et al., In vivo imaging of the human brain with the Iseult 11.7-T MRI scanner. Nat Methods 1–4 (2024).
16. B. Peysakhovich, et al., Primate superior colliculus is causally engaged in abstract higher-order cognition. Nat Neurosci 1–10 (2024).

Patients with cerebellar lesions have ataxic uncoordinated movement due in a large part to their inability to adapt movements to their changing environment. We propose to take advantage of recent advances in our lab to determine the cellular and network basis for how the cerebellum adapts simple motor commands to a changing environment. The Stell team has recently developed a new behavioral paradigm (Nguyen and Stell 2024) that allows for perturbation of simple motor commands. Since one of the main functions of the cerebellum is to coordinate motor adaptation to perturbations, it is essential to study its role in this context, using a behavior paradigm that allows for both perturbed movement and access to the brain regions involved in movement coordination. Overcoming this barrier with our new behavioral paradigm and cutting edge imaging technology allows us to examine how neurons in the cerebellum interact to process, adapt, and refine routine motor actions.

Cerebellar neuronal dynamics during adaptation to movement perturbations

The ability to learn precise, coordinated motor programs and adapt them to a changing environment is essential for the survival of all animals. A leading theory for how mammalian brains quickly adapt motor programs to perturbations from the environment proposes that the cerebellum uses forward models to compare perceived sensory stimuli (perceived reafference) to a prediction of sensory stimuli (expected reafference) that is based on previous experience (Medina, 2011). Any differences between predicted and perceived reafference are then relayed to the motor cortex so it can adapt future motor programs to the perturbation. This ability to adapt motor commands to the environment is one of the major functions of the brain and there have been recent advances in our understanding of how it is performed (Li and Mrsic-Flogel, 2020). However, gaps remain: Where and how does the cerebellum store the expected reafference that it uses to predict and compare to the perceived reafference? How is activity of the network of cells in the motor cortex affected by differences in perceived and expected reafference that is output by the cerebellum? We propose to use a novel behavior paradigm that we have recently developed (Nguyen and Stell, 2024, bioRxiv) to further investigate, at a cellular level, how cerebellar processing affects motor learning.

Understanding the cellular, and network interactions during motor adaptation requires a unique behavioral paradigm that permits simultaneous observation of neuronal and behavioral adaptation to changes between perceived and expected reafference. In a classic experiment in humans, perceived reafference is decoupled from expected reafference using prism goggles that shift the visual field by several degrees while people use ballistic movements (throws) to locate a target in space (Martin et al., 1996a). In healthy subjects, initial failure to locate a target informs future attempts, and subjects adapt motor programs to reliably locate targets extremely quickly (within 10-20 attempts). However, patients with cerebellar deficits fail to adapt motor programs to the new sensory reafference produced by the goggles—presumably because the comparison between the perceived and expected reafference cannot be achieved without the cerebellum. Interestingly, in this paradigm the motor cortex must also be involved in the adaptation because movements are ballistic and subjects do not get feedback that they can use to adjust their performance until after the movement is executed. Therefore, the motor cortex is obliged to use the feedback from previously perturbed commands/trials to modify subsequent commands before they occur.

Our recently developed behavioral task in head-fixed mice allows us to observe activity of individual cells in both the cerebellum and recapitulates the essential components of prism goggles experiment in humans: Head-fixed mice use a forelimb to control a joystick that moves a water-port. The predicted and perceived reafference is decoupled by changing the rate at which the water-port moves with respect to the joystick, and mice must use previously perturbed trials to modify future trials.

The hypothesis is that the expected reafference associated with a motor command is saved in synaptic weights of synapses onto the cells in the cerebellar cortex that are activated by an efferent copy of the motor command. When the motor command is consistently paired with a novel sensory input, the weights of these synapses change to report the sensory input as a known expectation. The net sum of activity through those synapses is then compared to the net sum of activity in the synapses that relay the perceived reafference into the cerebellum. The cerebellar output then reports any differences between motor (expected reafference) and sensory (perceived reafference) inputs as a modulation of firing rate of the cerebellar nuclei cells that are the output of the cerebellum.

Specifically, This project proposes to test these hypothesis using a combination of in vivo electrophysiology, 2-photon imaging, and optogenetics to determine:

- Whether a comparison is made in the cerebellum between learned and perceived reafference that is output as a change in spike rates of cerebellar neurons.

- How various cell-types in the cerebellum perceive, learn, and store expected reafference associated with a simple motor program.

Salla disease is a rare autosomal recessive disorder caused by mutations in the SLC17A5 gene, leading to defective sialin, a lysosomal transporter responsible for the export of sialic acids. This results in the accumulation of sialic acids in lysosomes and contributes to severe motor and cognitive deficits due to brain hypomyelination. The pathophysiology of this disease remains poorly understood and no treatment exists. We are interested in the development of therapeutic molecules. For this purpose, we intend to develop 2D-models related to this central nervous system pathology. We will differentiate induced Pluripotent Stem Cells from patients to neurons and oligodendrocytes. These models will be used to assess the complete cascade of sialin dysfunction and the potential therapeutic effects of candidates, from substrate export and cell surface sialylation, to functional downstream effects, such as improvement on cell-cell interactions and myelination of neurons by oligodendrocytes.

Free Sialic Acid Storage disorders, FSASD, are rare, autosomal recessive diseases caused by mutations in the SLC17A5 gene, which encodes sialin, the lysosomal transporter responsible for exporting sialic acids and acidic hexoses, such as glucuronate from lysosomes. Defective sialin leads to accumulation of sialic acid in lysosomes and its excretion in urine. Although not fully understood, sialin dysfunction results in brain hypomyelination, leading to severe motor and cognitive deficits. Approximately 200 cases of FSASD have been reported worldwide, though underdiagnosis is likely due to limited awareness, non-specific clinical symptoms, and the lack of routine biochemical testing. The clinical spectrum ranges from early lethal forms to slower-progressing variants that can extend into adulthood. The severe infantile form, sialic acid storage disorder (ISSD), is caused by mutations that abolish sialin expression or function, leading to substantial sialic acid accumulation. The more common and milder Salla disease is associated with the hypomorphic Arg39Cys (R39C) missense mutation in sialin, which reduces its transport activity and partially impairs its lysosomal delivery. Currently, the pathophysiology of this disease remains poorly understood and there is no treatment for this devastating disease.
Our team aims to develop a pharmacological chaperone therapy for Salla disease by utilizing small-molecule allosteric ligands of sialin to correct the trafficking and/or transport defects caused by the R39C mutation. However, in order to validate the potential of a molecule in the treatment of the disease, it is necessary to have more complex models, reproducing the main characteristics of this central nervous system disease. The PhD objective will be to develop live-cell 2D models of Salla disease, to perform physiopathological studies on these models and to assess the efficiency of some drugs. To achieve these goals, the work will be divided into four tasks.
The live-cells models of Salla disease are fibroblasts or specialized cells derived from human induced Pluripotent Stem Cells (hiPSCs) provided by the members of the FSASD consortium (Drs. Malicdan and Huizing, NIH, USA; Dr. Christensen, CHOC USA).
In a first task, the PhD student will differentiate iPSCs from patients and isogenic controls or neural stem cells (which we derived from iPSCs) to cells more related to a CNS pathology, including neurons and oligodendrocytes. Considering the critical interaction between neurons and oligodendrocytes in neural network function, he will also develop conduct co-cultures of neurons and oligodendrocytes, possibly with some astrocytes. As an alternative, in parallel to this first approach that can take time, he can also generate cultures from the R39C knock-in mice (from NIH).
These models will be used to assess the complete cascade of sialin dysfunction and the potential therapeutic effects of candidates, from substrate export and cell surface sialylation, to functional downstream effects, such as improvement on cell-cell interactions and myelination of neurons by oligodendrocytes.
In a second task, he will collaborate with another team, “Chemical Glycobiology”, UMR 8576 UGSF, Université de Lille, the PhD student. He will develop specific functional cellular assays. He will monitor the chronic accumulation of sialic acid by LC-MS in control cells and patient cells treated with candidate molecules. The experiment on cells will be performed in the laboratory and the mass spectrometry analysis done in the other laboratory.
In a third task, he will also utilize metabolic oligosaccharide engineering (MOE), a powerful strategy for tracking sialin lysosomal activity developed by the “Chemical Glycobiology” team. MOE introduces alkyne-labeled sialic acid analogs into the cells, which are endocytosed and delivered to lysosomes, where they are exported by sialin. Once exported into the cytosol, they are incorporated into glycoconjugates, specifically gangliosides. The detection of these gangliosides, modified with the alkyne-labeled sialic acids, occurs through click chemistry—a technique that allows the covalent attachment of fluorescent probes to alkyne groups. The reagents that are necessary will be provided by the Chemical Glycobiology” team. The experiment will be made and analyzed in our laboratory. This approach will help unravel the pathology of sialin underlying Salla disease and allow the validation of novel candidate drugs.
The fourth task will be divided in subparts relative to the cell-models, neurons, oligodendrocytes or co-cultures. In this task, he will aim to link the molecular and cellular preliminary dysfunction of sialin in Salla disease with the physiopathology and assess its correction by our pharmacological strategy. Previous works proposed that defective sialic acid egress from lysosomes could alter the sialylation status of membrane proteins and glycolipids. A decreased turnover of sialoglycoconjugates, such as gangliosides, has been described, which could affect neuronal myelination and brain function. Additionally, sustained expression of the heavily sialylated form of a cell-surface adhesion protein, PSA-NCAM, was observed in sialin knock-out mice compared to WT mice and proposed to cause reduced myelination, through decreased maturation of oligodendrocytes.
Thus, to better understand the physiopathology and assay potential treatments, in collaboration with the other team, he will examine the ganglioside profile in neural stem cells and neurons derived from patients’ iPSC or isolated from R39C knock-in mice, quantifying either endogenous gangliosides or gangliosides sialylated with the sialic acid analog described above (MOE) to focus on the role of sialin as sialic acid source. The readout is an analysis by mass spectrometry. He will proceed the same way as in the second task : the samples will be prepared in our laboratory and sent for analysis in Lille.
In the laboratory, on neurons, he will also observe the expression of PSA-NCAM by western-blotting. He will perform the same assays with candidate molecules to see if they can reverse the potentially observed effect of the Salla disease phenotype.
As the myelination of neurons involves their neural counterparts, oligodendrocytes, it is planned to examine if the physiopathology could be explained by defective Oligodendrocyte Precursor Cells (OPC) proliferation and differentiation. OPCs will be generated from iPSCs or isolated from mouse brains and monitored for developmental markers over time using immunolabeling.
Finally, as there is a crosstalk between neurons and oligodendrocytes, he will to examine neuronal activity, metabolism, and oligodendrocyte maturation on co-cultures developed in task 1 and asses the effect of potential drug candidates.

Maintaining balance during walking is a challenging task after an injury, such as a lower limb amputation. It requires the nervous system to coordinate the contraction of more than 600 muscles to adjust the forces acting on the body. The goal of the project is to model how the nervous system learns efficient muscle coordination strategies, or “synergies”. Our research hypothesis is that the brain integrates visual and vestibular signals to evaluate the (in)stability of the locomotor pattern, then uses this instability criterion to drive the adaptation of spinal locomotor synergies. To inform and validate the model, we will record how able-bodied participants adapt their motor synergies when walking with a device simulating a lower-limb prosthesis. The model, we will develop, can then be used to improve prosthesis design, by simulating how participants adapt to different lower-limb prosthesis designs, and selecting designs to which amputees can adapt rapidly and intuitively.

– Motivation and state-of-the art –
Balance impairments are widespread after an injury to the musculoskeletal system, such as an ankle sprain [1], or a lower-limb amputation [2]. The recovery of a correct dynamic balance is at the heart of gait rehabilitation. In the biomechanics literature, a novel metric has recently been proposed to quantify dynamic balance during locomotion related to the rotational momentum around the whole body Center of Mass (CoM) [3]. To minimise it, subjects must control the force exerted by the legs onto the ground. In particular, they must adjust the angle of this force so that it points to the CoM. To achieve this, the nervous system must coordinate more than 600 muscles of the human body. During movement, individual muscles do not contract independently, but work together as “synergies” resulting in meaningful motor outputs [4]. These muscle synergies are organized at the level of the spinal cord, that can generate complex, adaptable motor patterns [5]. This has led to the theory that the nervous system composes complex movements by combining such fixed building blocks. Improvements in sensorimotor coordination during learning are thought to originate from the brain finding a better way of combining these fixed building blocks. However, more recent experimental results have shown that spinal synergies themselves change during sensorimotor learning [6], [7]. This adaptation of spinal synergies is however dependent on intact input from the brain [7]. It is however not known how the brain drives the spinal learning of motor synergies, and how this improves walking balance.
– Project’s objective and research hypothsesis –
The purpose of the project is to model how the nervous system learns novel motor synergies to improve walking balance after an alteration of the body biomechanics. Our research hypothesis is that the brain integrates visual and vestibular signals to evaluate the (in)stability of the locomotor pattern, then uses this instability criterion to drive the adaptation of spinal locomotor synergies. To inform and validate the model, we will record how able-bodied participants adapt their motor synergies when walking with a device simulating a lower-limb prosthesis.
– Methodology –
Work Package 1: Experimental measurement of adaptation to altered biomechanics
To determine how the nervous system adapts to altered biomechanics, we will collect experimental data on able-bodied participants equipped with a device simulating a lower-limb prosthesis, which was previously developed in our laboratory [8]. This device is roughly an orthosis which can be linked either at the level of the foot or at the level of the thigh to reproduce the loss of functions experienced after a transtibial or a transfemoral amputation. When walking with the device, participants loose both fine motor control of the ankle (and knee if the device is attached at the thigh) and lower limb proprioceptive input. Preliminary results indicate that, when walking with the ankle prosthesis simulator, participants can adapt within minutes and maintain balance. With the ankle-and-knee prosthesis simulator however, it takes several hours of practice to learn to maintain balance. In the first experiment, we will record novice participants during their first ten minutes of practice with the ankle prosthesis simulator. In the second experiment, we will record participants learning to walk with the ankle-and-knee prosthesis simulator. Participants will practice walking for one hour over five consecutive days, and we will record the first five and last five minutes of each practice session. In both experiments, we will record muscle contractions of the “impaired” and intact limb, whole-body kinematics and ground reaction forces, and analyse the change in muscle synergies and balance (i.e. rotational momentum) over the course of practice.
Work Package 2: Modelling neural learning of motor synergies
The goal is to develop a model of how the brain drives the adaptation of spinal motor synergies to improve balance. Our hypothesis is that the brain integrates visual and vestibular signals to determine trunk rotation, and uses this as an indicator of walking instability. We will develop a learning algorithm allowing the brain to minimise this instability by adjusting spinal synergies.
We will first model the adaptation to the ankle prosthesis. We will consider a simplified biomechanical model with a rigid trunk and a massless leg actuated at the knee and hip, attached to the ankle prosthesis. According to our preliminary modeling results, maintaining balance in this case requires appropriately scaling the hip and knee torques. We will test how the trunk instability can be used to drive the scaling of hip and knee torques. We will then extend this model to include multiple antagonist and bi-articular muscles acting at each joint, to determine how the redundancy of the human muscular system affects learning.
We will then model the adaptation to the ankle-and-knee prosthesis. In this case, participants can only adjust hip torque. A simple scaling of hip torque is not sufficient to solve the task. Instead, subjects must learn a complex temporal profile of hip torque. To model how participants learn this temporal profile, we will build on a model of spinal generation of complex locomotor patterns, which we have previously developed [9].
– Impact –
Balance during standing and walking is a key skill of the human being, and balance impairments have a dramatic impact on health and quality of life. Assistive devices, which improve balance, are associated with a high cognitive cost, which jeopardizes their added value for functional mobility. The goal of this project is to determine how subjects adapt to a device simulating a lower-limb prosthesis. The model we will develop can be used to simulate how amputees adapt to different lower-limb prosthesis designs, and select designs that may be easier or more intuitive for the person to adapt to.
– References –
[1] C. Doherty, C. Bleakley, J. Hertel, B. Caulfield, J. Ryan, and E. Delahunt, ‘Recovery From a First-Time Lateral Ankle Sprain and the Predictors of Chronic Ankle Instability: A Prospective Cohort Analysis’, Am J Sports Med, vol. 44, no. 4, pp. 995–1003, Apr. 2016.
[2] W. C. Miller, M. Speechley, and B. Deathe, ‘The prevalence and risk factors of falling and fear of falling among lower extremity amputees’, Arch Phys Med Rehabil, vol. 82, no. 8, pp. 1031–1037, Aug. 2001.
[3] N. Al Abiad, H. Pillet, and B. Watier, ‘A Mechanical Descriptor of Instability in Human Locomotion: Experimental Findings in Control Subjects and People with Transfemoral Amputation’, Applied Sciences, vol. 10, no. 3, p. 840, Feb. 2020,
[4] E. Bizzi and V. C. K. Cheung, ‘The neural origin of muscle synergies’, Front Comput Neurosci, vol. 7, Apr. 2013.
[5] S. Giszter, F. Mussa-Ivaldi, and E. Bizzi, ‘Convergent force fields organized in the frog’s spinal cord’, Journal of Neuroscience, vol. 13, no. 2, pp. 467–491, 1993.
[6] L. J. Bouyer, P. J. Whelan, K. G. Pearson, and S. Rossignol, ‘Adaptive locomotor plasticity in chronic spinal cats after ankle extensors neurectomy’, J. Neurosci., vol. 21, no. 10, pp. 3531–3541, May 2001.
[7] L. Carrier, E. Brustein, and S. Rossignol, ‘Locomotion of the Hindlimbs After Neurectomy of Ankle Flexors in Intact and Spinal Cats: Model for the Study of Locomotor Plasticity’, Journal of Neurophysiology, vol. 77, no. 4, pp. 1979–1993, Apr. 1997.
[8] A. Louessard, ‘Conception et prototypage de pied prothétique obtenu par fabrication additive’, These de doctorat, Paris, HESAM, 2023. Available: https://theses.fr/2023HESAE076
[9] M. L. de Graaf, L. Mochizuki, F. Thies, H. Wagner and C. L. Mouel, ‘Motor pattern generation is robust to neural network anatomical imbalance favoring inhibition but not excitation’, Apr. 22, 2022, bioRxiv. doi: 10.1101/2022.04.21.489087

This project aims to develop a multiscale atlas of the brain by integrating 2D and 3D multimodal data from both normal and pathological samples. Using datasets such as the Allen Brain Atlas, digital pathology, high-resolution microscopy, and brain MRI, we will explore the correlation between cellular, spatial, and anatomical features across different scales. Novel methods will be developed for 2D to 3D registration, focusing on aligning histological images with MRI data to localize cellular details within a 3D anatomical context.
The atlas will be applied to brain tumor datasets, including TCGA, EBRAINS, and private collections, to capture complex spatial and molecular relationships. By integrating advanced graph-based models, this project will enhance understanding of both normal brain structure and tumor heterogeneity, with potential applications in personalized treatments for neuro-oncological diseases.

The proposed PhD project aims to develop a comprehensive multiscale atlas of both normal and pathological brains through the integration of multimodal datasets. The primary goal is to establish a unified framework that accurately represents anatomical and molecular correlations across different spatial scales, from cellular features to volumetric anatomical structures. This atlas will provide a foundational tool for understanding the complex organization of the brain, bridging cellular-level insights with broader anatomical features. The project will initially focus on studying normal brain anatomy using high-resolution datasets such as the Allen Brain Atlas, which contains detailed molecular, anatomical, and spatial data across multiple species. The Allen Brain Atlas includes high-resolution in situ hybridization (ISH) data, gene expression patterns, and corresponding anatomical mappings, making it an ideal starting point for understanding the baseline organization of the brain.
The project will then expand into pathological conditions, particularly focusing on brain tumors. We will incorporate large-scale, publicly available datasets such as The Cancer Genome Atlas (TCGA), which includes multi-omics data (genomics, transcriptomics, epigenomics) alongside histopathological and MRI images from thousands of tumor samples. Additionally, the EBRAINS platform will provide access to various imaging datasets that offer a macroscopic view of brain anatomy and function, including MRI data. The project will also leverage private collections from collaborating institutions, encompassing over 5,000 patient samples, providing both multi-modal imaging and associated clinical metadata, such as survival outcomes, treatment regimens, and molecular profiles. These resources will allow us to create a multiscale model that can be applied to the study of tumor heterogeneity, disease progression, and response to therapy.
The challenge of integrating such diverse modalities lies in their varying spatial resolutions and data structures. For instance, digital pathology images captured at the cellular level (2D) operate at a vastly different resolution than MRI scans (3D) that provide a volumetric view of the entire brain. Additionally, the dimensionality of these datasets varies significantly, with molecular profiles containing thousands of features per sample. Therefore, one of the key objectives of this project is to develop novel methods for 2D to 3D registration, which is critical for aligning the various modalities and integrating them into a coherent multiscale atlas. This will involve aligning high-resolution microscopic images from digital pathology and high-resolution microscopy with volumetric MRI data, which captures macroscopic brain structures.
We will also develop methods for reconstructing 3D representations from stacked 2D Whole Slide Images (WSIs), which provide rich details at the cellular and subcellular levels. By stacking these WSIs, we can create a volumetric representation that bridges the gap between cellular and anatomical scales. These innovations will enable us to spatially localize cellular features—such as specific cell types, gene expression patterns, and tumor microenvironments—within a 3D anatomical context, providing a more detailed and comprehensive understanding of both normal brain organization and pathological changes that occur in diseases like glioblastoma.

The methodological foundation of this project will focus on graph-based models for understanding spatial relationships between cells and structures in both physiological and tumor contexts. Specifically, we will apply Graph Neural Networks (GNNs) to encode spatial correlations between cells within the 3D brain atlas. This will enable us to capture complex spatial interactions that are difficult to detect using conventional approaches. The graph-based representations will support understanding how cellular architecture and spatial patterns change in disease states, such as tumor heterogeneity.
A key aspect of the project will also involve the use of Graph Attention Networks (GATs) for self-supervised learning, particularly in the context of multimodal data registration. GATs will be leveraged to integrate different data types during the 2D to 3D registration process, specifically to learn and preserve important spatial relationships between 2D histological images and their corresponding 3D MRI volumes. These networks will help refine the registration process by improving feature matching between modalities, facilitating a more accurate integration of cellular and structural information.
In addition to registration, another major goal of the project is the clinical validation of the developed atlas. This will involve cross-referencing the atlas with real-world clinical outcomes and molecular data, specifically focusing on brain tumor datasets. By analyzing spatial and molecular data within this multiscale framework, we aim to identify important correlations between tissue morphology and molecular phenotypes, such as gene expression profiles and mutations. The use of high-dimensional tabular data will be incorporated to provide a holistic view of how molecular changes at the cellular level are reflected in tissue morphology. Techniques such as saliency mapping and feature attribution will aid in the explainability of these models, highlighting key features that correlate with clinical outcomes.
The validation of the atlas across normal and pathological samples will allow us to assess the robustness and applicability of our models. By incorporating molecular data, we will aim to refine our models to better predict patient outcomes and treatment responses. This project will not only contribute to neuro-oncology research but also provide a framework applicable to broader neurological diseases and other complex biological systems.

PhD Scope Summary:
1. Data Gathering and Preprocessing: Gather and preprocess multimodal datasets from publicly available sources (e.g., Allen Brain Atlas, TCGA, EBRAINS), ensuring consistency and preparing them for further analysis.
2. 2D-3D Registration: Develop novel methods for aligning 2D histopathological images with 3D MRI data, focusing on multimodal registration and feature preservation.
3. Graph-Based Self-Supervised Models: Implement GNNs and GATs for spatial correlation modeling and use them to guide 2D-3D registration as well as multimodal integration.
4. Clinical Validation: Validate the generated atlas by correlating its predictions with clinical outcomes and molecular data, particularly focusing on brain tumors.

Quelles sont les bases cérébrales de l’acquisition de la production du langage? Pour répondre à cette question, nous comparerons les représentations du cerveau d’enfants entre 2 et 12 ans à celles des algorithmes d'apprentissage profond. L'étude utilisera des enregistrements intracrâniens de patients implantés avec des électrodes pour traiter leur épilepsie. Trois protocoles d'écoute et de production linguistique seront développés pour étudier l'activité neuronale élicitée à chaque mot écouté ou produit. Ces activations cérébrales seront enfin modélisées par un alignement avec les algorithmes de langage récents, tels que Llama 3 et Mixtral. Une métrique de corrélation sera utilisée pour quantifier la similarité entre représentations corticales et représentations algorithmiques. Enfin, l'analyse fine de ces représentations permettra d'identifier le type d’attributs (acoustiques, phonétiques, lexicaux, syntaxiques) et leur évolution au cours du développement du cerveau de l'enfant.

I. Introduction

1. Motivation
Les modèles de langage de grande taille (LLMs) ont montré des capacités impressionnantes en traitement du langage naturel. Cependant, leurs entrainement requière beaucoup plus de mots que le cerveau des enfants lors de l’acquisition du langage. Cette observation re-soulève une question historique : quelle architecture permet l'acquisition efficace du langage chez les jeunes enfants ?

2. Revue de la littérature
De nombreuses études utilisent l'IRMf et l'EEG (e.g. Dehaene-Lambertz et al Science 2002, PNAS 2021), limitées par le bruit et la sensibilité aux mouvements. Au cours des trois dernières années, notre équipe a utilisé des enregistrements intracrâniens (iEEG) chez l’enfant pour décoder la hiérarchie neuronale de la perception de la parole. Cette recherche, actuellement en revue à Nature Neuroscience, montre que le cortex temporal, frontal et pariétal une séquence hiérarchique des représentations linguistiques lors de l'écoute d'histoires (Evanson et al & Bourdillon & King, under review).

3. Identification des défis
Les recherches sur les bases cérébrales de l'acquisition de la *production* de la parole sont, en revanche, quasi inexistantes. Tout d’abord, les artefacts de mouvement, particulièrement nombreux chez les enfants, empêche l'utilisation de l'IRMf et l'EEG. Par ailleurs, la production du langage est un objet complexe à analyser, en particulier chez les enfants. Enfin, Il manquait jusqu'à récemment de modèles computationnels précis pour modéliser ces processus.

4. Approche proposée
Techniquement, les défis liés aux données cérébrales peuvent être surmontés par l'iEEG, qui offre une meilleure résolution et est s'accomode dans difficultés aux movements élicités lors de la production du langage (Mesgarni & Chang Science 2014). Conceptuellement, les avancées des LLMs et des modèles d'apprentissage profonds multimodaux offrent de prédictions précises des activations neuronales attendus lors de la perception ou de la production du langage (Millet et al 2022).


II. Objectifs de la thèse
Cette thèse vise à modéliser les bases neuronales et computationnelles de la production du langage lors du développement (2-12 ans), en intégrant des données neuronales unique au monde (iEEG chez l’enfant) et le développement des derniers modèles du langage développés en intelligence artificielle.


III. Hypothèses de recherche
Nous postulons que :
(1) il existe une hiérarchie anatomiquement et temporellement définie du traitement du langage, et ce même chez les enfants les plus jeunes de notre cohorte
(2) les calculs de cette hiérarchie évoluent au cours du développement de l'enfant: les représentations du contrôle moteur de la parole de seront seront quasiment identiques à travers les âges, alors que les représentations des mots -- et a fortiori des phrases -- seront de plus en plus stables avec l'âge.
(3) l'architecture computationnelle permettant la production de la parole suit une maturation similaire entre cerveau et modèles d’IA, notamment ceux entrainés sans supervision mais de manière multimodale, plus ou moins un facteur multiplicatif (les modèles nécessitent beaucoup plus de mots)

IV. Méthodologie

1. Données Neuronales
- L'étude inclura, sur 3 années, 40 enfants diagnostiqués avec une épilepsie, âgés de 2 à 12 ans, et nécessitant la pose d’électrodes intracrâniennes pour une durée d’une semaine, dans le cadre de la prise en charge de leur pathologie.
- Nous utiliserons des électrodes stéréotaxiques EEG spécifiquement conçues pour un enregistrement précis et localisé de l'activité cérébrale.

2. Tâches
- Tâche 1: Écoute passive : Les enfants écouteront 1 heure d’histoire pour identifier les représentations de la perception de la parole dans le cerveau, comme décrit dans notre étude précédente (Evanson et al., under review at Nature Neuroscience).
- Tâche 2: Réponse à des stimuli verbaux : les enfants seront invités à répéter des mots et des phrases, ainsi qu’à nommer ou décrire des images.
- Tâche 3: Production de langage spontanée : Les enfants engageront un dialogue guidé par l'expérimentateur pour encourager la production naturelle de langage, permettant l'analyse de la parole spontanée.

3. Prétraitement
- Les données seront traitées en utilisant MNE-Python, un outil spécialisé dans le traitement des signaux EEG/MEG, afin d’obtenir l’activité large fréquence et gamma sur chaque point du cortex enregistré par des dipôles d’électrodes.
- Ces réponses cérébrales seront alignées au stimuli et au comportement.

4. Modélisation
- Modèles: Des modèles de deep learning tels que Llama 3.2, Mistral (pour le texte), Wav2Vec2, Hubert, et Seamless (pour la parole) seront utilisés pour comparer les activations cérébrales aux prédictions des modèles.
- Techniques d'alignement des données neurales avec les modèles : L'alignement sera réalisé en utilisant le 'brain score': une technique désormais classique et qui permet d'évaluer la similarité entre modèle et cerveau avec une régression linéaire.
- Comparaison intra et inter-modèles: l’alignement models - cerveau sera réalisé pour chaque couche de modèle. Les résultats permettront d’évaluer si certains composants ou certaines architectures sont plus similaires au cerveau d’un point de vue fonctionnel. Cette comparaison sera ensuite étendue au cours du développement du modèle (training steps) et du cerveau (indexé par l'âge des enfants).

V. Résultats attendus

1. Nous prévoyons d'identifier un alignement significatif entre les activations cérébrales des enfants et les prédictions des modèles de langage, indiquant une correspondance entre les processus neuronaux et les architectures d'apprentissage profond.
2. L'analyse montrera des motifs distincts de production du langage, reflétant une hiérarchie corticale dans le cortex préfrontal, et une succession des représentations hiérarchiques (contexte -> mot -> syllabe -> articulation) avec des différences au cours du développement. Par exemple, nous nous attendons à ce que les représentations motrices soient similaires à travers les âges, là où les représentations de contexte et de mots seront mieux maintenues chez les enfants les plus âgés.
3. Les résultats mettront en lumière les forces et les limites des modèles de deep learning actuels en comparaison avec les processus neuronaux des enfants, offrant des pistes pour améliorer l’architecture et l’optimisation de ces modèles.

VI. Impact

1. Neuroscience
Ce projet enrichira notre compréhension de la neurobiologie de l'acquisition du langage. Les résultats devraient démontrer l'existence d'une hiérarchie corticale chez les enfants dès l'âge de 2 ans, impliquant des régions spécifiques des cortex fronto-temporo-pariétaux dans la production du langage. Ces découvertes pourraient notamment permettre de valider l'hypothèse d’une maturation hiérarchique du langage au cours du développement.

2. Intelligence Artificielle
La comparaison systématique entre les données cérébrales et les modèles de langage permettra d'identifier les architectures de deep learning les plus efficaces pour simuler les processus neuronaux chez l'enfant. Cette analyse révélera quelles règles d'apprentissage sont les plus adaptées pour modéliser et maîtriser l’acquisition du langage.


VII. References
- Dehaene-Lambertz et al (2002) Functional neuroimaging of speech perception in infants, Science
- Evanson, et al & Bourdillon and King (under review Nature Neuroscience) Decoding the neural language hierarchy in the child’s brain
- Gennerani, et al & Dehaene-Lambertz (2021) Orthogonal neural codes for speech in the infant brain, PNAS
- Mesgarani, et al & Chang (2014) Phonetic feature encoding in human superior temporal gyrus, Science
- Millet, et al & King (2022) Toward a realistic model of speech processing in the brain with self-supervised learning, Neurips

This project aims to develop novel multimodal tools to assess consciousness in patients with disorders of consciousness (DoC). Diagnosis of DoC is challenging due to the lack of overt motor or verbal responses, making it crucial to detect covert cognitive processing. We will integrate EEG, electrospinography (ESG), and somatosensory stimulation to explore spinal-cortical interactions. ESG captures somatosensory evoked potentials (SEPs), revealing how cognitive states modulate spinal responses. In healthy participants, we will test the impact of stimulus predictability on spinal responses (Experiment 1) and examine how EEG dynamics influence these modulations (Experiment 2). Finally, we will apply these tools clinically at Pitié-Salpêtrière Hospital to assess their relevance for DoC patients (Experiment 3). This research aims to enhance diagnosis and neuroprognosis by detecting concealed responses, bridging brain-body interactions, and advancing bedside assessment methods.

Background
Disorders of consciousness (DoC) refers to a group of pathological states in which consciousness is affected due to injury or trauma to the nervous system. The accurate diagnosis of patients with DoC remains a challenge due to the difficulty of assessing awareness in patients without explicit motor or verbal responses. Finding markers of concealed command-following and preserved cognitive processing in these patients is crucial to characterize and neuroprognose these patients. Our research team specializes in developing novel multimodal markers to characterize the state of consciousness of these patients. Our tools range from neurophysiological markers (using EEG), brain dynamical patterns (using fMRI), and brain-body measures (using cardiac, respiratory, or olfactory signals). In this project, we propose to develop novel assessment tools involving somatosensory stimulation, surface electromyography (EMG), electrospinography (ESG), and EEG. ?Although spinal responses and muscular activity measured with a generally available tools such as EMG and ESG have shown promising results (Bekinschtein, et al., 2009; Lesenfants et al., 2016), its potential to detect covert responses and preserved cognitive processing remains unexplored.

The use of ESG to assess cognitive processing centers on the ability of this non-invasive technique to capture spinal cord signals relevant to sensory, motor, and cognitive functions. The spinal cord is crucial in transmitting and modulating sensory and motor information between the brain and the body. However, imaging this structure non-invasively has historically posed significant challenges (Bailey et al., 2024; Kaptan et al., 2024; Nierula et al., 2024). ESG emerges as a promising method due to its ability to record somatosensory evoked potentials (SEPs) from surface electrodes placed along the vertebral column. SEPs provide insight into neural responses to sensory stimuli and allow researchers to infer cognitive processing that involves somatosensory integration.

Recent results highlight that spinal cord signals are not merely reflexive but are modulated by cognitive and emotional states. For instance, cognitive load or emotional context can alter the amplitude of SEPs recorded via ESG, revealing that higher-order brain functions influence spinal processing. In another example, ESG can differentiate between self-generated and externally applied touch (Boehme et al., 2019).

Objective
The primary objective is to explore the interplay between spinal and cortical processing of somatosensory stimulation under different states of consciousness. We hypothesize that top-down regulation will modulate the medullar responses to somatosensory stimulation according to the predictability of the sequence of stimulus (experiment 1). Following the results of our team demonstrating that the dynamics of the brain can predict perceptual processing (Turker et al., 2024), we also hypothesize that the ongoing dynamics of EEG will determine the level of this top-down modulation (experiment 2). Finally, we propose translating the developed tools to the patient’s bedside to test their clinical relevance to characterize DoC patients (experiment 3).

Methods

This project's first stage involves optimizing EEG and ESG recordings in healthy individuals. We will adapt the methodology presented in (Niedernhuber et al., 2022) to deliver two different cognitive protocols designed to assess the subjects’ statistical learning capacity (Bekinschtein et al., 2009; Benjamin et al., 2024). ESG/EEG recordings will be obtained using high-density EEG (256 channels EGI) and ESG with the compatible polybox system. As previously done in previous research in our lab (Bekinschtein et al., 2009; Pérez et al., 2021), to test the interaction of the hypothesized effects with conscious processing, we will record healthy individuals in two conditions: fully attentive (e.g. counting deviant trials) and distracted (e.g. performing in parallel a cognitively demanding task). Finally, DoC recordings will be performed at the Pitie Salpetriere hospital (assisted by the clinical staff of our lab).

Bailey, E., … & Eippert, F. (2024). bioRxiv. https://doi.org/10.1101/2024.09.05.611423
Bekinschtein, T. A., … & Naccache, L. (2009). PNAS, 106(5), 1672–1677. https://doi.org/10.1073/pnas.0809667106
Bekinschtein, T. A., … & Sigman, M. (2009). Nat Neuro, 12(10), 1343–1349. https://doi.org/10.1038/nn.2391
Benjamin, L., … & Dehaene-Lambertz, G. (2024). bioRxiv. https://doi.org/10.1101/2024.01.08.574591
Boehme, R., … & Olausson, H. (2019). PNAS, 116(6), 2290–2299. https://doi.org/10.1073/pnas.1816278116
Kaptan, M., … & Mackey, S. (2024). Front Hum Neuro, 18. https://doi.org/10.3389/fnhum.2024.1339881
Lesenfants, D., … & Noirhomme, Q. (2016). Neurology, 87(20), 2099–2107. https://doi.org/10.1212/WNL.0000000000003333
Niedernhuber, M., Raimondo, F., Sitt, J. D., & Bekinschtein, T. A. (2022). J Neuro, 42(46), 8729–8741. https://doi.org/10.1523/JNEUROSCI.0658-22.2022
Nierula, B., … & Eippert, F. (2024). bioRxiv. https://doi.org/10.1101/2022.12.05.519148
Pérez, P.,... & Sitt, J. D. (2021). Cell Reports, 36(11), 109692. https://doi.org/10.1016/j.celrep.2021.109692
Turker, B., … & Sitt, J. (2024). Comm Bio. https://doi.org/10.1101/2024.01.18.576171

Sleep plays a crucial role in the consolidation of episodic and emotional memories, engaging a complex network of brain structures, including the hippocampus (HPC), prefrontal cortex (PFC), and basolateral amygdala (BLA). While NREM sleep’s role in memory consolidation via neural reactivations is well-established, REM sleep’s contribution is less clear. This PhD project aims to explore neural reactivation and cross-structure dynamics during REM sleep, focusing on the dorsal and ventral HPC, BLA, and PFC. Leveraging large-scale sleep recordings from rodents undergoing spatial and emotional learning, the project will develop novel analytical tools to uncover structure-specific patterns in REM sleep and their variation across brain regions and timescales. The study will provide insights into REM sleep's role in processing emotional and spatial memories and contribute to understanding its relevance for conditions such as PTSD, depression, and anxiety.

Sleep is important for the consolidation of episodic and emotional memories1, which rely on a complex network of structures including the dorsal and ventral hippocampus, the prefrontal cortex and the baso-lateral part of the amygdala2,3. The two main phases of sleep, rapid-eye movement (REM) and non-REM (NREM) sleep, are characterized by different neuronal and oscillatory activities throughout the brain4. NREM sleep is associated with sharp-wave ripples (SWRs) in the hippocampus and slow wave activity/spindles in the neocortex. These local-field potential rhythms coordinate neural reactivations across the hippocampus, cortex and amygdala that sustain the sleep-dependent reinforcement of spatial memories and their gradual transfer towards cortical areas. REM sleep, on the other hand, is associated with theta activity throughout the hippocampus-prefrontal cortex-BLA network, with different profiles of power and cross-structure coordination associated with changes in fear memory5,6. In humans, REM sleep is involved in the regulation of emotional reactivity7 but its role in emotional memory processing is still debated8.
The reactivation of neural activity during NREM sleep is believed to support memory consolidation. The patterns of neural activity established during a given experience are reinstated during NREM sleep, so that the type of wakefulness experience can be decoded during epochs of coordinated neural activity orchestrated by NREM sleep rhythms9. Despite its hypothesized role in emotional memory consolidation, neural reactivation has seldom been shown in REM sleep10. It is possible that the methods used in NREM sleep were not appropriate to detect reactivation during REM sleep due to the difference in dynamics. For instance, SWRs in NREM sleep provide a precise time window to look for reactivations that is absent in REM sleep. Alternatively, it is possible that the importance of REM sleep for memory and emotions is based on a mechanism other than reactivation. For example, we have shown that the activity in the BLA during REM is characterized by a heightened activity compared to NREM and wakefulness, but with decreased overall synchrony. This decreased synchrony in REM sleep could play a complementary role to the reactivation of NREM sleep by regulating the BLA network. Overall, the common and structure-specific neural patterns in the dorsal and ventral hippocampus, BLA and mPFC, as well as the mechanisms underlying how these regions interact during REM sleep remain largely unexplored. This PhD project will leverage a rich dataset of large-scale sleep recordings following spatial learning and emotional experiences in order to develop new analytical methods that will uncover patterns of neural reactivation during REM sleep within and across structures. The project also aims to identify how these patterns vary across different timescales and brain regions, with a particular focus on the dHPC, vHPC, BLA, and mPFC. By understanding these dynamics, the study will provide insights into how specific types of experiences are processed during REM sleep and further our understanding of the role of REM sleep.
Aim 1. Dataset completion
In collaboration with a doctoral student of the laboratory, the PhD candidate will acquire simultaneous large-scale recordings in the dorsal hippocampus, ventral hippocampus and BLA of rodents undergoing a spatial learning task under normal or negative conditions. In the cheeseboard maze task, the rats have to learn daily the location of 3 rewards out of 177 possible location on a circular platform11. In the negative condition signaled by a single light cue, they will receive unpredictable eyelid-shocks. The training session will be preceded and followed by long periods of recording in the homecage allowing for the collection of NREM and REM sleep data. The data will then be pre-processed (sleep-scoring, extraction of the behavioral position, spike sorting) and analyzed.
Aim 2. Investigate REM-sleep patterns and reactivation under various learning conditions in the dHPC, vHPC, BLA and mPFC.
The previous dataset and 3 other datasets of the laboratory2,12 will be used to investigate the changes of neural patterns within each structure as a function of the training conditions. Dataset 2 contains recordings in the dorsal HPC and the BLA during sleep before and after training on a rewarded linear track with an aversive element at a single location. Dataset 2 contains recordings in the dorsal and ventral HPC during sleep preceding and following a linear track experience motivated by reward or avoidance (eyelid shock). Dataset 3 consists of LFP and multiunit recordings in the mPFC, dorsal HPC and BLA during sleep before and after contextual fear conditioning under normal or heightened stress conditions induced by a stress hormone injection.
Temporal-spectral decomposition of the LFP in the three datasets will be used to classify sleep epochs into different stages (e.g.: REM and NREM sleep). Oscillation landmarks of the different regions will be extracted in an unsupervised manner using empirical mode decomposition13 and used to segment REM sleep into different (sub) states using tools like hidden markov models (HMMs14). This investigates the possibility of temporally precise network changes happening during REM sleep that could support neural reactivations similar to the ones happening in NREM sleep during SWRs. In addition to that, we will explore novel methods of reactivation detection at the level of units. More specifically, we will focus on methods that do not require a known reactivation of a template sequence or a temporal landmark of the reactivation will be tested. Namely, we will explore previous unsupervised methods like SpikeShip15, HMMs and mutual information-based sequence index14,16. We hypothesize that this analysis approach, tailored to account for specificity of REM dynamics, will help unveil neural reactivations in this sleep stage as well.
Aim 3. Investigate cross-structure REM sleep patterns and reactivation under various learning conditions in the dHPC, vHPC, BLA and mPFC.
Segmenting REM sleep into different neural-driven sub-states might also help uncover new cross-structure dynamics. We will use different population vector analyses2 to look for different connection signatures at the different REM (sub)states. We hypothesize that REM sleep is not an homogeneous state with a single connectivity dynamic, and that dynamics can be extracted using the unsupervised approach devised above. In addition, we hypothesize that reactivations happening during REM sleep (if any) might also happen across the different regions in the aforementioned circuit, impacting how memory is consolidated, and the subsequent performance during the recall of the memory.
This project will contribute to the growing body of research on sleep, memory, and emotional processing by developing novel analytical methods and expanding our understanding REM sleep. The findings may have important implications for conditions such as PTSD, depression, and anxiety, where both memory and emotional regulation are disrupted17.
References
1.Klinzing, et al.. Nat. Neurosci. (2019). 2.Girardeau, G. et al. (2017). 3.Girardeau, G., et al. Nat. Neurosci.(2009) 4.Girardeau, G. & Lopes-dos-Santos, Science (2021) 5.Popa, D. et al. Proc. Natl. Acad. Sci. U. S. A. (2010) 6.Hong, J. et al. Curr. Biol. (2024) 7. Van der Helm, E. et al. Curr. Biol. (2011) 8.Lerner, I.et al. Neurobiol. Learn. Mem.(2021) 9.Pfeiffer, B. E. Hippocampus(2018) 10.Louie, K. & Wilson, M. A. Neuron (2001) 11.Dupret, D.,et al. Nat. Neurosci. (2010) 12.Morici, J. et al. BioRxiV (2024) 13.Quinn, A. J.et al. J. Open Source Softw. (2021) 14.Recanatesi, S.et al. Neuron (2022) 15.Sotomayor-Gómez, B.et al. PLOS Comput. Biol.(2023) 16.Souza, B. C., et al. BioRxiV (2022) 17.Murkar, A. L. A. & De Koninck, J. Sleep Med. Rev. (2018)

This thesis project aims to explore the neural and inflammatory mechanisms underlying corneal neuropathic pain (CNP), a chronic condition affecting the ocular surface and significantly impacting patients' quality of life. While spinal pain networks are well understood, the neural network involved in processing corneal pain remains unclear, leading to limited therapeutic options. Chronic pain conditions, such as CNP, are driven by maladaptive plasticity in the brain and changes in functional connectivity, alongside contributions from neuroinflammation, especially involving microglial activation.
By combining anatomical, functional, and neuroinflammatory network analysis using innovative imaging approaches, this project will advance the understanding of chronic corneal pain mechanisms, potentially leading to more effective treatments for CNP and other trigeminal pain conditions.

Specific cognitive functions are supported by the coordinated activity of a distributed network of brain regions. Pain, especially, results from a remarkably complex interplay of various peripheral and central mechanisms. While the network of brain regions implicated in spinal pain are well studied, trigeminal pain and particularly corneal pain has received only little attention, leading to unsatisfactory therapeutic treatments The ocular surface is the site of numerous pathologies both acute and chronic. Corneal innervation arises from the primary sensory neurons located in the trigeminal ganglion, which transmit nociceptive signals to the sensory trigeminal complex in the brainstem, a pivotal hub in the processing of trigeminal pain. However, beyond the trigeminal ganglion, the neural network involved in ocular pain remains poorly understood. Corneal neuropathic pain (CNP) is a chronic condition associated with painful symptoms that dramatically affect the quality of life of patients. Chronic pain disorders linked to maladaptive plasticity in both peripheral and central nervous systems, affecting functional connectivity networks. Inflammation, especially microglial dysfunction, also contribute to CNP pathophysiology. Understanding brain (neuronal and microglial) networks and their plasticity in chronic corneal pain, using animal models, is crucial for uncovering these mechanisms.
The objective of the proposed project is to use network-based approaches to identify the corneal pain large-scale structural (anatomical) and functional (co-activation) connections in rodent brains. Identifying key hub regions within these networks will improve our understanding of the chronic corneal pain mechanisms and improve treatment strategies. The candidate will identify and characterize the first anatomical and functional corneal pain matrix in mice, as well as the inflammatory mechanisms. This involves building anatomical and functional brain-wide networks engaged in corneal pain in mice, using activity-regulated gene expression (c-Fos expression, tissue analysis, sectioning, immunohistochemistry, confocal analysis) induced by noxious corneal stimulation to identify functional network signature and hub nodes central to chronic pain. This will be combined with a cutting edge imaging approach, functional ultrasound imaging, to observe specific functional alterations of subnetworks of the pain matrix via functional connectivity maps and correlation matrices. The impact of neuroinflammation on corneal pain network will be examined by building a brain-wide microglial connectome and inhibiting microglial activation by pharmacological approaches. Hub regions thus identified will be silenced using a chemogenetic approach to assess their role corneal pain.
This multimodal, innovative project will provide unprecedented insights into the brain circuitry, connectivity and plasticity underlying chronic corneal pain, offering valuable knowledge for future research and potential therapeutic strategies.

The inter-individual behavioral variability arises even when the animals are genetically identical and reared in same environments. The biological mechanisms underlying this behavioral variability remain incompletely understood. The proposed project explores mechanisms underlying variability in genetically identical individuals, examining whether behavioral differences stem from neural connectivity variations. It builds on the team's previous work on mapping neural networks underlying sensorimotor decisions in Drosophila larva and establishing causal relationships between circuit motifs and behavior. Furthermore, it seeks to analyze individuality traits and the underlying neural underpinnings. The research employs a multidisciplinary approach, combining behavioral experiments, quantitative behavioral analysis, and functional and structural neural imaging in Drosophila melanogaster larva. This work could reveal conserved mechanisms shaping individuality across species.

I. Background and context
An open question in neuroscience is what makes us unique in our behavior and where does this individuality come from? Across species, individuals exhibit variable behavior often due to a confluence of environmental and genetic effects (1, 2). In addition, increasing evidence in a variety of species show that even genetically identical individuals that were reared in identical conditions can exhibit variability in a range of behaviors (from predator avoidance and foraging in aphids (3) odor choice, handedness and grooming behavior in Drosophila (1, 4, 5); to exploratory and grooming behavior in mice (6)). The mechanisms by which genetically similar individuals, raised in identical conditions can develop idiosyncratic behaviors, are not well known. It has been shown that invertebrates can display an individual variation in the development of neuronal morphology and synaptic connectivity (7, 8) that can give rise to differences in behavioral phenotypes in genetically identical individuals. This suggests that factors at the level of neuronal circuits contribute to behavioral variability. However, the basis of these behavioral variations and to which extent the different factors and stochastic biological processes (i.e. neuronal wiring…) contribute to the observed variability in behaviors that individuals exhibit and the detailed mechanisms involved, remain incompletely understood. In addition, individual variation in both wiring and behavior may prove to be a very general feature of neural circuits. Thus understanding the biological bases of inter-individual variability in genetically identical individuals is key for understanding not only how circuits develop, but also how they function and give rise to specific behavioral phenotypes. Additionally, this may have implications for our understanding of neuropsychiatric diseases and the way they affect different individuals (9).

II. Project Aims

To address the question of the non-genetic basis of behavioral varibility, several requirements need to be fulfilled. First, one needs to have a very large sample size to be able to quantify idiosyncrasies in behavior. Second, to explore to which extent behavioral variability results from circuit changes, one needs to be able to have, on one hand, a behavioral paradigm that is characterized by inter-individual variability and on the other, a tractable model system which combined, enable to determine the causal links between circuit (wiring) properties and idiosyncratic differences in behavior. Finally, automated assays are essential to reduce any external sources of variability (10).
Here we propose to examine the interactions between behavioral and neural circuit wiring variation using a simple genetic model organism, Drosophila melanogaster larva, which has a rapid life cycle allowing us to obtain a large number of individuals very quickly. Furthermore, thanks to our work over the last decade we have access to a behavioral paradigm of variable behavioral response of larvae to an aversive mechanical stimulus for which we have mapped the underlying circuitry (11–13) and established causal relationships between specific circuit motifs and behavioral outputs (19, 21). We are thus in a unique position to tackle with unpreceded detail the biological bases of inter-individual behavioral variability. Furthermore, thanks to the different high-throughput and automated behavioral assays we have developed and our multidisciplinary quantitative approach, we will be able to assess the traits of individuals that constitute their "personalities".

Aim 1. Identifying mechanisms that control the variability in genetically identical individuals
Our previous work (12, 13) has characterized the larval behavioral responses to a mechanical stimulus and shown that larvae exhibit probabilistic behaviors that vary across trials and across individuals. The inter-trial variability also varies across individuals, with some individuals exhibiting the same behavior over trials. These individuals were genetically homogenous and raised in identical conditions. This variability could be observed even upon optogenetic activation of sensory neurons (12) suggesting that the observed variability is dependent on non-genetic and non-environmental causes. Our hypothesis is, that the differences could emerge from differences in neuronal connectivity.

Given that in our system, we have access to causal links between synaptic and functional connections between pairs of neurons, we will be able to test whether the differences in behavior between individuals result from differences in synaptic wiring. To do this we will perform high-throughput experiments in individual chambers for each larva (to keep identities). We will select different individuals based on their behavioral phenotypes (“Hunchers”, “Benders”, etc.). We will subsequently perform calcium imaging experiments in these individuals to test the strength of connections functionally and immunoshistochemistry combined with expansion microscopy to determine the structural changes in connections between pairs of neurons in the previously identified circuits that control these behaviors (11–13). This is possible due to the short duration of behavioral and imaging experiments (30s –2 min/experiment on average). We will thus determine circuit differences that give rise to the different behavioral tendencies.

Aim 2. Characterizing individuality traits and the underlying mechanisms
We will use the different individuals with characteristic behavioral profiles as in Aim 1. These larvae will be evaluated using various automated behavioral assays (anemotaxis, chemotaxis, phototaxis, responses to sound etc) to determine whether individuals display consistent traits across experimental paradigms. We hypothesize that the variations in behavioral tendencies in response to mechanical stimuli may indicate more general behavioral trait differences such as risk-taking propensity, overall activity levels and adaptability. Performing multiple behavioral tests with the same individuals will assess the persistence of behavioral traits in different contexts and over time. To analyze the data, we will employ unsupervised clustering techniques in a shared latent space (15) as well as supervised classification methods (based on expert annotations(13)), and higher-order sequence analysis (15). This multi-faceted approach will help us identify which behavioral traits covary across different individuals or variants, potentially constituting what is known as a behavioral syndrome (16). Building upon insights from previous work on candidate neurons we will also investigate the role of neuromodulators, such as dopamine and serotonin, proposed to contribute to drive individuality differences due to their capacity to regulate behavior at multiple scales and affect neuronal activity across circuits (17). By integrating these diverse approaches, we aim to determine the biological underpinnings of behavioral individuality, potentially revealing conserved mechanisms that shape personality traits across species.

.
III. References

1. J. M. Mueller et al., Front. Behav. Neurosci.(2022)
2. B. S. Baker et al., Cell. (2001)
3. W. Schuett et al., Dev. Psychobiol. (2011)
4. S. M. Buchanan et al., Proc. Natl. Acad. Sci. (2015)
5. K. Honegger et al., Curr. Biol., (2018)
6. J. Freund et al., Science. (2013)
7. Y.-H. Chou et al., Nat. Neurosci. (2010)
8. G. A. Linneweber et al., Science (2020)
9. P. Faure et al., Front. Behav. Neurosci. (2022)
10. B. de Bivort et al., Front. Behav. Neurosci. (2022)
11. T. Jovanic et al., Curr. Biol. (2019)
12. T. Jovanic et al., Cell (2016)
13. J.-B. Masson et al., Plos Genet. (2020)
14. E. De Tredern et al. BioRxiv (2023), in revision
15. A. Blanc et al., BioRxiv (2024), accepted in Plos. Comp. Biol
16. J. A. Moretz et al., Behav. Ecol. (2007)
17. R. T. Maloney

Our research project aims to advance the understanding of how brains without a cerebral cortex can generate higher-order cognitive abilities, such as problem-solving and imitation. Outside of primates, these abilities are observed in a few species of birds, such as parrots and crows. By combining behavioral studies and MRI techniques, this thesis project will examine the brain organization underlying these higher cognitive abilities, using medium-sized parrots as a model. Thanks to the collaboration between NeuroPSI and NeuroSpin, a parrot brain atlas is currently under construction. As a next step, we will perform functional MRI (MEMRI and BOLD fMRI) to identify the brain areas involved in problem-solving and imitation. We hypothesize that enlarged associative areas in parrots (e.g. mesopallium) are key to these abilities. Comparing these data with those from primates will illuminate the evolution of associative brain areas linked to higher-order cognition in vertebrates.

Background and significance :
Abilities of problem-solving and imitation form the basis of complex behaviors such as tool-use and culture transmission. We are interested in understanding how such abilities have appeared multiple times independently during evolution. The higher-order cognitive functions underlying such behaviors have mostly been investigated in humans and other primates that possess a large neocortex. However, some birds such as parrots also exhibit problem-solving and imitation abilities, despite lacking a six-layered neocortex.
The knowledge in comparative neuroanatomy between the avian and mammalian brains has suggested that functions and neural circuitry of the avian pallium (dorsal telencephalon) are surprisingly similar to those of the mammalian cortex, although their morphology is very different (Stacho et al., 2020). Importantly, such similarities have evolved independently in the mammalian and avian lineages by convergent evolution. The basic organization of the avian pallium is shared by all bird species, but on top of that, some pallial areas such as the associative areas are particularly elaborated in parrots and crows (von Eugen et al., 2020; Ströckens et al., 2022), much like they are in primates. Thus, comparative analyses of the associative pallium between mammals and birds will help identify critical factors necessary for evolving higher-order functions.

Specific aims :
By combining anatomical and behavioral analyses, the thesis project aims to characterize associative brain areas involved in problem-solving and imitation abilities in medium-sized parrots. We chose the green-cheeked conure (Pyrrhura molinae) as a parrot model, which is optimal for combining anatomical and behavioral examinations.
In terms of connectivity, two pallial structures called the mesopallium and the nidopallium caudolaterale (NCL) appear to be the avian associative areas (Ströckens et al., 2022), integrating sensory and motor information. The mesopallium is also proposed to be a "bridge" between sensory and motor regions in the songbird pallium. In addition, similarly to the primate associative areas, the mesopallium and NCL also receive dopaminergic inputs from the mesencephalon. An intriguing question is whether these structures are involved in higher-order cognitive functions such as problem-solving or immitation in birds. The mesopallium and the NCL are particularly well developed in parrots and crows compared to pigeons and chickens, making them a good candidate for the seat of "intelligence" in the avian pallium.
Thus, we will observe neuronal activation of these brain areas in parrots performing or observing other individuals perform problem-solving tasks, by using the highest magnetic field preclinical MR scanner available today.

Research design and methods :
The project contains three main aims consisting of behavioral and MRI studies. The behavioral tasks will be performed in NeuroPSI under supervision of Kei Yamamoto, and MR imaging will be performed in NeuroSpin under supervision of Luisa Ciobanu.

Aim 1) Behavioral trainings of problem-solving and object manipulation in parrots : It has been clearly shown that different species of parrots are capable of performing problem solving tasks (Chen et al., 2019; O’Neill et al., 2021). Puzzle box tasks are a type of approach to assess the cognitive capacity for problem solving, and several different types of puzzle boxes, baited with food, have been used in mammals and birds. To this end, we have designed a “multiple-steps” puzzle box that is applicable in a wide range of animals.
Notably, we designed this box to be applicable to humans too. Our personal observations indicate that: 1) average adult human subjects can open the box without exploring the box (self problem-solving), 2) human babies (one- to two-year-old) could not open the box by themselves even after a few minutes’ exploration. However, once the experimenter showed them how to open, babies could understand the logic very quickly and open it right away (guided problem-solving). Using the same puzzle box, our first goal for the behavioral study is to test the ability of problem solving in parrots, by either self problem-solving or guided problem-solving.

Aim 2) Manganese-Enhanced MRI (MEMRI) : In collaboration with NeuroSpin, we have successfully established a MEMRI protocol to reveal brain regions involved in an object manipulation task in cichlid fish (Estienne et al., in preparation). A modified version of this protocol will be applied in parrots.
The parrots will be injected intraperitoneally with a MnCl2 solution (50 µg/g), and the behavioral tests (Aim 1) in parrots will be performed 24 hours after the injection. After the puzzle box task, the animals will be anesthetized and placed in the MRI scanner for imaging. We will acquire high-resolution T1-weighted images (100 µm isotropic), which will highlight the active brain areas presenting increased accumulation of Mn2+.
Thus, this will allow us to visualize the brain areas activated during problem-solving performance in parrots.

Aim 3) : BOLD functional MRI : Lastly, we will develop BOLD fMRI in awake parrots. This methodology will measure brain activity underlying cognitive functions with precise temporal and spatial resolutions, offering an unprecedented window into complex cognition in a non-mammalian species.
As visual stimuli, we will use a video demonstrating problem-solving behaviors (Aim 1) of another parrot individual. Based on studies on human and non-human primates, it has been suggested that the neural circuits activated while observing an action performed by another individual are similar to circuits that are activated when one performs the same action (Tkach et al., 2007; Lepage et al., 2008). These neural circuits are thought to contribute to imitation (Rizzolatti & Craighero, 2004). Thus, BOLD fMRI in parrots will provide a unique opportunity to decipher the brain circuitry responsible for imitation, which has so far only been studied in primates.

Perspectives :
The functional study will reveal the brain structures that are involved in problem-solving and imitation. Our hypothesis is that parts of the mesopallium and NCL that are well-developed specifically in parrots would be involved in these functions.
Beyond this, the development of these methodology will enable future comparisons between primates and parrots. Since parrots are cognitively on par with primates, identical tasks can be performed by both species, allowing for a comparison of brain activity.
Such an experimental setup would greatly advance our understanding of how evolution produced different neuroanatomical organizations capable of similar behaviors in mammals and birds.

The main objective of this PhD project is to assess the role of axon-oligodendrocyte synaptic interactions in central nervous system (CNS) myelination. Oligodendrocytes are the myelinating cells of the CNS. These cells sense neuronal activity through synaptic and extra-synaptic communications with axons, which play critical roles in myelination. These axon-oligodendrocyte synaptic interactions involve transsynaptic adhesion molecules and postsynaptic density proteins. Interestingly, we recently showed that oligodendrocytes express the postsynaptic density protein PSD-95 at myelin ensheathment sites. However, the fucntion of PSD-95 in myelination remains unknown. The main objectives of this project are to determine: 1) the role of oligodendroglial PSD-95 in myelin formation and growth, and 2) PSD-95 downstream signaling pathways regulating myelination. This project should provide better insights into the mechanisms regulating myelination in physiological and pathological conditions.

The main function of oligodendrocytes in the central nervous system (CNS) is to produce myelin sheaths, insulating axons and increasing the conduction velocity of action potentials. These cells derive from oligodendrocyte precursor cells (OPCs), a class of progenitor cells in the CNS highly abundant at birth, but also persistent in the adult (Franklin and ffrench-Constant, 2017). By their capacity to generate oligodendrocytes, OPCs play a key role in developmental myelination and myelin regeneration following various CNS injury. It has been established, mainly in rodent, that OPCs are synaptically innervated by glutamatergic and GABAergic axonal fibers throughout the CNS (Bergles et al., 2000). Synaptic neurotransmitter release onto oligodendroglial cells may regulate OPC proliferation and differentiation in an activity-dependent manner (Sahel et al., 2015). Synapses on OPCs are lost upon OPC lineage progression into myelinating oligodendrocytes and replaced by synaptic-like interactions formed by trans-synaptic adhesion molecules and post-synaptic density proteins (for review see Masson and Nait-Oumesmar, 2023; Hughes and Appel, 2019). Strikingly, we recently showed that oligodendrocytes expresses the post-synaptic density protein PSD-95, a major scaffold protein clustering glutamatergic receptors at neuronal excitatory synapses (Kennedy, 1997). In oligodendrocytes, PSD-95 expression increases with their differentiation into myelinating cells and is specifically targeted to ensheathment sites, suggesting a potential function of this PDZ protein in myelin sheath growth.

To decipher the functional role of PSD-95 in oligodendrocyte development and myelination, we are currently using the zebrafish as a model system. The zebrafish larva is ideal for such studies as they develop externally and rapidly, are small, transparent and their CNS is well characterized. This PhD project will address the following specific aims:

1) To decipher the functional role of PSD-95 in myelin sheath growth, using CRISPR/Cas9-mediated deletion of dlg4 (the gene encoding for PSD-95) specifically in oligodendroglia, combined with in vivo live imaging of myelination in the zebrafish spinal cord. Oligodendroglia cells’ proliferation, differentiation, and myelin internode length will be quantified in dlg4 mutant and control strains.

2) To determine whether PSD-95 axon-oligodendroglia microdomains are hotspots of calcium signaling that trigger myelin sheath growth. For this aim, the PhD candidate will analyze calcium transients in dlg4 KO and control zebrafish lines, using the calcium indicator GCaMP6 (Baraban et al., 2019; Krasnow et al., 2019). Frequency and amplitude of calcium transients will be correlated with myelin internode growth.

3) To define the downstream signaling pathways mediated by PSD-95 in oligodendrocytes. The candidate will define PSD-95 signaling pathways controlling myelin sheath growth. More specifically, she/he will examine the calcium calmodulin/NAT, Fyn tyrosine kinase and actin cytoskeleton signaling pathways in dlg4 mutant and control zebrafish lines.

Overall, this PhD project should provide better insights into the mechanisms of activity-dependent regulation of myelination under physiological and pathological conditions.


References
1. Baraban M, Koudelka S, Lyons DA. Ca2+ activity signatures of myelin sheath formation and growth in vivo. Nat Neurosci. 2018 Jan;21(1):19-23.
2. Bergles DE, Roberts JD, Somogyi P, Jahr CE. Glutamatergic synapses on oligodendrocyte precursor cells in the hippocampus. Nature. 2000;405(6783):187-91.
3. Franklin RJM, Ffrench-Constant C. Regenerating CNS myelin - from mechanisms to experimental medicines. Nat Rev Neurosci. 2017;18(12):753-769.
4. Hughes AN, Appel B. Oligodendrocytes express synaptic proteins that modulate myelin sheath formation. Nat Commun. 2019;10(1):4125.
5. Kennedy MB. The postsynaptic density at glutamatergic synapses. Trends Neurosci. 1997;20(6):264-8
6. Krasnow AM, Ford MC, Valdivia LE, Wilson SW, Attwell D. Regulation of developing myelin sheath elongation by oligodendrocyte calcium transients in vivo. Nat Neurosci. 2018;21(1):24-28.
7. Masson MA, Nait-Oumesmar B. Emerging concepts in oligodendrocyte and myelin formation, inputs from the zebrafish model. Glia. 2023;71(5):1147-1163.
8. Sahel A, Ortiz FC, Kerninon C, Maldonado PP, Angulo MC, Nait-Oumesmar B. Alteration of synaptic connectivity of oligodendrocyte precursor cells following demyelination. Front Cell Neurosci. 2015;9:77.

This project aims to explore how neurocognitive systems interact in the early stages of reading acquisition, focusing on two key processes: 1) learning letter-sound correspondences and 2) visual letter recognition. In fact, reading requires overcoming a perceptual invariance (mirror invariance), which can hinder literacy development. Our research has shown that inhibiting mirror invariance (distinguishing mirror letters like "b" and "d") enhances reading fluency. This project will test how multimodal learning (combining visual, auditory, motor, and sensory systems) modulate these processes. Three work packages (WPs) will address: 1) the impact of multimodal training on learning letter-sound associations through controlled behavioral studies; 2) the neural correlates of these learning processes using fMRI; and 3) applying these insights to improve classroom teaching through randomized controlled trials with children. The outcomes will offer evidence for optimizing educational practices.

1. Context and Objectives

Reading acquisition is a complex process involving diverse cognitive skills, ranging from language-specific abilities like lexical and syntactic knowledge to domain-general functions such as memory, attention, and executive control. Among the first essential steps is learning arbitrary letter-sound correspondences, which is crucial for decoding and underpins future reading fluency and comprehension. This project focuses on this foundational phase, exploring how neurocognitive systems interact to facilitate 1) the audio-visual mapping needed for reading, and 2) the visual representation of letters. In addition, it seeks to apply these findings to educational contexts to enhance early reading instruction.

Visual recognition of letters requires the brain to overcome natural perceptual invariances (size, shape, and mirror) typically beneficial for recognizing objects in various perspectives. However, in alphabetic systems like the Latin alphabet, mirror invariance—which treats mirrored objects as identical—hinders literacy acquisition, as distinguishing between "d" and "b" is critical for reading. Previous research, including our own, has shown that literacy training inhibits mirror symmetry for letters and that teaching children to distinguish mirrored letters accelerates reading fluency. However, this inhibition is not limited to mirrored letters but extends to most asymmetric letters, which may explain the two-fold increase in reading speed we observed in first graders after specific training for mirror discrimination.

In addition to visual processing, other cognitive systems interact deeply with one another during learning, as demonstrated by phenomena such as the McGurk effect, Stroop effect, and rubber-hand illusion. Multimodal learning, particularly combining audio-visual cues, has been shown to enhance learning outcomes, and engaging multiple systems like motor (writing, speaking) and sensory systems (touch) could eventually boost memory consolidation. However, the specific mechanisms of how multimodal training affects reading acquisition, and under what circumstances it might hinder learning, remain unclear. A better understanding of these interactions can help optimize educational practices.

This project will address three key questions: 1) How can sensory-motor systems be engaged optimally to produce synergies for letter-sound association learning? 2) How do non-visual systems (e.g., auditory, motor) modulate the visual representation of letters? 3) How can these insights be applied in real-world classroom settings to benefit early literacy acquisition?

The project's primary aim is to refine our multisystem synergy hypothesis in learning to read, which posits that multimodal representations of letters can enhance visual letter recognition, ultimately improving reading fluency.

2. Methodology

The project will unfold in three work packages (WPs):

WP1: Behavioral Studies on Multimodal Learning of Letter-Sound Associations

In WP1, two controlled experiments will investigate the impact of multimodal training on letter-sound association learning. Participants (30 adults per experiment) will be trained to map pseudoletters to syllables using different methods: audio-visual (AV), audio-visual-writing (AVW), and audio-visual-writing-speaking (AVWS). Each participant will be exposed to six pairs of pseudoletter-syllables, including two mirror pseudoletters. After training, participants will perform tasks to assess audio-visual map recognition, visual perception of pseudoletters, writing, and pronunciation.

We hypothesize that increasing the number of engaged systems will strengthen letter-sound associations and modulate visual representations of letters. Specifically, we predict enhanced visual letter recognition with increased size and shape invariances but decreased mirror invariance, particularly for AVWS training. However, when multiple systems are engaged close in time (experiment 2), we expect reduced learning performance due to divided attention and working memory overload.

Participants will be retested one day and one week later to examine memory consolidation, which we predict will improve with multimodal training.

WP2: Neuroimaging Studies on Multimodal Training Effects on Visual Representations

WP2 will examine the neural correlates of multimodal learning using functional magnetic resonance imaging (fMRI). Forty adult participants will undergo fMRI recordings following AV and AVWS training. We predict that multimodal training will enhance functional connectivity between the visual word form area (VWFA), responsible for letter recognition, and other brain areas associated with auditory, writing, and speech production systems. Additionally, we expect AVWS training to result in greater size and shape invariance and reduced mirror invariance in visual letter representations.

WP3: Translating Insights into Educational Practice

WP3 will test the application of multimodal learning strategies in a real-world classroom setting. Kindergarten children (~30 per group) will participate in randomized controlled trials, with baseline measures ensuring comparable groups. Children will be trained on real letters in either a multimodal (AVWS) or audio-visual (AV) condition, with the active control group using symmetric letters (e.g., “x”). The training will last three weeks, 25 minutes per day.

The training's impact on visual letter recognition will be assessed using a same/different judgment task for letter pairs presented in different sizes and orientations, a writing task, and a reading fluency task.

3. Expected Outcomes

This project will deepen our understanding of how neurocognitive systems interact during early stages of reading acquisition. It will provide insights into how multimodal training can optimize letter-sound association learning, leading to more efficient reading instruction. The neuroimaging component will shed light on the underlying brain mechanisms, particularly in the VWFA, while the educational studies will validate these findings in real-world settings, directly impacting literacy education practices.

These findings could inform in fine curriculum design, instructional methods, and interventions for children struggling with reading, ultimately promoting more effective literacy acquisition.

References:

Pegado, F., Nakamura, K., Braga, L. W., Ventura, P., Filho, G. N., Pallier, C., Jobert, A., Morais, J., Cohen, L., Kolinsky, R., & Dehaene, S. (2014). Literacy breaks mirror invariance for visual stimuli: A behavioral study with adult illiterates. Journal of Experimental Psychology. General, 143(2), 887–894. https://doi.org/10.1037/a0033198

Pegado, F., Nakamura, K., Cohen, L., & Dehaene, S. (2011). Breaking the symmetry: Mirror discrimination for single letters but not for pictures in the Visual Word Form Area. NeuroImage, 55(2), 742–749. https://doi.org/10.1016/j.neuroimage.2010.11.043

Pegado, F., Nakamura, K., & Hannagan, T. (2014). How does literacy break mirror invariance in the visual system? Frontiers in Psychology, 5, 703. https://doi.org/10.3389/fpsyg.2014.00703

Torres, A. R., Mota, N. B., Adamy, N., Naschold, A., Lima, T. Z., Copelli, M., Weissheimer, J., Pegado, F., & Ribeiro, S. (2021). Selective Inhibition of Mirror Invariance for Letters Consolidated by Sleep Doubles Reading Fluency. Current Biology, 31(4), 742-752.e8. https://doi.org/10.1016/j.cub.2020.11.031

The auditory system relies on internal knowledge to extract regularities and predict what comes next to rapidly process only what cannot be inferred. The field often distinguishes between prediction in time and prediction of content. However, there is good evidence that these two dimensions interact. The aim of this PhD will be: 1) to identify how mechanisms of temporal predictions, such as sensorimotor oscillations, contribute to speech comprehension in highly naturalistic and variable speech, and 2) to assess how these mechanisms interact with predictions of content to support higher order processing. We will use EEG and iEEG recordings to study oscillatory processes during natural speech listening, while using dynamical and transformer models to develop candidate theories of these interactions. The outcome will integrate oscillator and hierarchical models of prediction to serve communication and will have bearing on future work in the effects on perceptual disorders.

The auditory system is constantly tasked with processing sequences of information with remarkable speed. The ability to do so relies on anticipation. Our brain relies on internal knowledge to extract regularities and infer what comes next to rapidly process only what cannot be inferred (1). The aim of this project is to uncover the underlying mechanisms of predictions in hyper realistic contexts. In particular, the role of temporal prediction in the comprehension of complex natural stimuli like speech remains unclear. While listeners of basic sequences like tones process isochronous sequences more readily, speech is best understood when it is less regular. This dichotomy is in part because the variable timing associated with syllable production and prosodic phrasing provides information which can be used to disambiguate syntax and structure. Are the mechanisms that support this temporal processing shared between speech and more basic rhythms? And if so, how do they ladder up to natural comprehension?

In recent work, our team has developed an Adaptive Frequency Oscillator model to explain how a system of Excitatory and Inhibitory neural populations could be wired to generate human behavior with regards to a temporal prediction task in uncertainty (2). In this project, we plan to build on this work extending the model to handle greater complexity that would typically be reflected in sentences, segmenting speech while maintaining robust predictions in the face of rhythmic deviations. At the same time, state-of-the-art Automatic Speech Recognition models (3), solve temporal parsing through a series of continuous and overlapping integration windows. This sort of brute force method is less plausible in human biology due to large energy costs of continuous high excitability. We will investigate how applying such constraints to AI models will lead to similar oscillator behavior to the AFO model and how its dynamics lead to effective segmentation and temporal prediction errors.

Project 1: Temporal prediction errors as meaningful speech markers.
We will first test the hypothesis that deviations from temporal expectations in speech contain linguistic meanings: deploying Bayesian and AFO Models on corpora of linguistically annotated speech to read out how prediction errors of the timing of individual syllables relate to linguistic groupings (Fig. 6). We will validate our models in the read out of temporal cues to sentence meaning and syntactical structure: assessing how speech patterns are evaluated and judged for natural prosodic patterns. We will follow up this work with a comparative EEG study to see if the prediction errors in timing assessed by our model elicit similar neural responses across populations, and whether these responses can predict effective speech comprehension and hierarchy building.

Project 2: Temporal prediction errors as drivers of sequence memory
During real-time speech, prediction errors in both content and time lead to segmentation in comprehension which can be used for structure building. We will draw on memory literature to assess how predictions in time and content support the encoding of sequential information present in speech and their subsequent binding, in an asynchronic fashion, into separable, hierarchical representations. We will design speech stimuli with congruent and incongruent prediction errors in time and content to assess how later memory is affected. We will quantify prediction errors using information theoretic measures leveraging Large Language Models for content and Bayesian models for temporal information. We will conduct EEG experiments combined with behavioral memory experiments to assess how behavioral memory effects are driven by changes in surprisal at hierarchical levels within sentences and across content and time to assess how the two interact to support successful encoding of information. Finally, we will leverage Representational Similarity Analysis to assess whether changing neural states reflect ongoing tradeoff, as the brain negotiates between storing sequentially presented units and the need to abstract away from the sequential order to allow for future assembling.

References:
1. Knill, D. C. & Pouget, A. The Bayesian brain: the role of uncertainty in neural coding and computation. Trends in Neurosciences 27, 712–719 (2004).
2. Doelling, K.B., Arnal, L.H., & Assaneo, M.F. Adaptive oscillators support Bayesian prediction in temporal processing. PLoS Computational Biology 19 (11), e1011669
3. Baevski, A., Zhou, Y., Mohamed, A. & Auli, M. wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations. in Advances in Neural Information Processing Systems vol. 33 12449–12460 (Curran Associates, Inc., 2020).

Navigation in space is a fundamental cognitive capacity that relies on a brain wide circuit of Head Direction (HD) neurons. Vestibular based HD signals reach the Presubiculum via the anterior thalamus. This internally generated compass signal is combined with visual landmark signals from retrosplenial cortex, to be useful for navigation. Our proposal aims to understand the differential contributions of multi-sensory, input-related and voluntary movement-related determinants of HD signals in thalamo-cortical and parahippocampal networks. We will test the hypothesis that the HD system is not merely a passive responder to visuo-vestibular inputs, but integrates information from ideomotor systems, using a predictive model of movement.
The results of this project hold the possibility of creating a new form of translational neuroscience, that can inform better brain machine interfaces and neural prosthetics for humans.

Main objective. Spatial navigation depends on multi-sensory integration to generate internal representations of space and direction, such as head direction (HD) signals. The aim of this thesis is to dissect multi-sensory, input-related and voluntary movement-related determinants of HD signals in the mouse model. The candidate will study thalamo-cortical and parahippocampal networks containing HD cells, building on the idea that the HD system is not merely a passive responder to visuo-vestibular inputs, but also integrates information from ideomotor systems. We hypothesize that self-initiated changes in direction can update the HD signal, and that HD codes inform navigational decisions. Findings should help understand spatial orientation coding in active and passive conditions, and may inspire novel strategies using decoded signals as a command for brain machine interfaces.

Background. Our sense of orientation in an environment relies on cells and circuits in specialized brain areas. HD neurons function as the brain’s compass. Together with place and grid cells, they form an internal representation of the external world. Single HD neurons increase their firing when the head of an animal is oriented in a preferred direction [1], and the population of HD cells functions as an ensemble with attractor dynamics [2]. HD signals originate in the vestibular system and reach the Presubiculum (PrS), a parahippocampal area, via the Anterior Thalamic Nuclei (ATN) [3–6]. For spatial navigation, this internally generated compass signal is combined with visual landmark signals [7,8]. Our team has shown that multi-sensory vestibular and visual information are integrated in the PrS, where HD input projected via the ATN converges with information on visual landmarks from the retrosplenial cortex (RSC) [9]. In this PhD project, we will focus on these three areas of the HD system, the PrS, and its two main input areas, ATN and RSC. HD signals can be anticipatory [10], depending on the experimental paradigm, the degree to which movements are active, volitional or predictable, and the area of interest. However, current research has not systematically decoupled differential contributions of visual inputs and self-generated, vestibular inputs to the HD circuit. This disambiguation is of particular interest when visual inputs from the environment and motion-related signals jointly inform HD coding.

This thesis will combine large-scale neural recordings across the HD system in controlled visuo-vestibular environments to study multi-sensory contributions to HD codes. The guiding hypothesis is that visual, vestibular, and also predictive volition- or movement-related information jointly but differentially impact HD coding in distinct circuits of the HD system. HD tuning is expected to be sharpened by goal-directed navigation, with enhanced movement-related contributions and predictive HD codes. Active vs. passive movement conditions and visual perturbation experiments will disentangle different sources of information converging in HD circuits and we will test their impact on HD coding. Manipulating presubicular interneurons will address circuit mechanisms for flexible HD updating.

Methodology. High density Neuropixels probes [13] are routinely used in our team, to record action potentials in multiple regions associated with spatial circuits in awake head-fixed mice. Mice are installed at the center of a motorized platform, such that rotating movements over 360° activate the vestibular system. A visual projection onto a surrounding dome creates coherent or mismatched visuo-vestibular stimulation. Typically 40-60 well identified units can be recorded per session in the PrS with angular (directional) tuning evident for >50% of units. Two probes may be inserted to assess coordination of HD signals between areas. Importantly, the mouse can turn a steering wheel with its front paws, in order to autonomously rotate the platform, as if driving a car. HD activity will be compared in head-fixed ‘driving’ mice, or in passively turning ‘passenger’ mice. Chronically implanted probes can record HD activity during goal-directed navigation. Stereotaxic viral injections in transgenic mice are used to ablate or optogenetically silence local interneurons.
Data analysis will compare single-cell HD tuning, HD population coding, and geometrical and topological properties of the HD manifold across conditions. Linear decoders of vestibular, visual, and predictive motion-related information will be used to relate external variables to neural activity, whereas topological data analysis [14] will explore its geometric properties.

The candidate will investigate HD activity in a novel head-fixed paradigm that combines real vestibular and matched or mismatched visual stimuli. The aim is the disambiguation of multisensory and voluntary movement-related inputs to the HD system. HD areas are hierarchically organized, where PrS combines vestibular inputs from ATN with visual landmark information from RSC. Despite this theoretical separation of sources, evidence for mixed coding and early integration has been found [15]; The candidate will disentangle the degree of differentiation between ATN, RSC, and PrS, and test the site of integration with simultaneous recordings across different sensory and motor conditions. Moreover, the PrS also receives orbitofrontal input and may integrate information from ideomotor systems, using a predictive model of future HD. For the first time, active, self-induced rotation will let us test the impact of volitional and motor processes on reactive and predictive HD tuning in head fixed mice. We hypothesize that self-induced movements will lead to more reliable HD codes, due to enhanced predictive HD tuning and as a consequence, more precise coding of the current HD. We will analyze tuning curves, manifold geometry [14], and optimal decoding lags of visual, vestibular, and motor variables across areas of interest. We will compare the directionality of multi-area information flow between different conditions. We will look for the origin of predictive coding in the HD system. On a mechanistic level, we hypothesize that VIP interneurons in PrS may be critically involved in HD updating by weakening bump attractor states and making the circuit more receptive to incoming information. Manipulating VIPcre cells [15] will let us address their role during uncertainty or surprise, when predictions conflict with sensory input.

This thesis project will for the first time clarify the relationship between HD coding and volitional processes. Results will identify the neural origin and distributed nature of efficient volitionally controlled spatial orientation. The outcome may be impactful for translational research for better brain machine interfaces and neural prosthetics for humans.

References
1. Taube JS. J Neurosci. 1995. 2. Peyrache A et al. Nat Neurosci. 2015. 3. Simonnet J, Fricker D. Cell Tissue Res. 2018. 4. Nassar M et al. J Neurosci. 2018. 5. Simonnet J, et al. Nat Commun. 2017. 6. Boccara CN et al. Nat Neurosci. 2010. 7. Yoder RM et al. J Neurosci. 2015. 8. van der Goes M-SH et al. Elife. 2024 9. Richevaux L et al. Elife. 2023;12 10. Bassett JP et al. J Neurophysiol. 2005. 11. Butler WN et al. Curr Biol. 2017. 12. Dillingham CM, Vann SD. Front Neural Circuits. 2019. 13. Steinmetz NA et al. Science. 2021. 14. Chaudhuri R et al. Nat Neurosci. 2019. 15. Cullen KE, Taube JS. Nat Neurosci. 2017. 15. Nassar M et al. Neuroscience. 2024 doi: 10.1016/j.neuroscience.2024.09.0322024

This proposal examines the principles linking developmental trajectories and neural function in motion detection circuits, using Drosophila melanogaster as a model. The current view suggests a genetic program generates T4 and T5 neurons tuned to four motion directions, matching the eye’s visual columns. However, this model faces challenges. Our lab found that lobula plate progenitors undergo transient amplification, likely producing more neurons than expected. Collaborators also found that T4/T5 neurons show broad directional tuning and form six, not four, subtypes. We propose that Drosophila's motion detection system has six T4/T5 subtypes tuned to six motion patterns from the animal’s self-motion. Additionally, we suggest T4/T5 neurons are generated by a plastic genetic program that adjusts the number of motion detection columns to visual inputs through functional plasticity.

Background
The development of any organ or organism sets the stage for its later function. In the brain, neurons accomplish highly specific tasks in neural computation. To achieve this, they have to be born, divide, specialize into distinct cell types, migrate and/or project their processes to their correct target and form synapses with their designated partners. As a result, neural networks of the brain are highly intricate structures that perform computation that for example allow to execute highly specific behaviors in response to specific sensory inputs. These sensory inputs arrive from an intrinsically unpredictable environment and must be processed to produce behavioral responses that to maximize the animal’s chances of survival by appropriately exploring and exploiting this environment.
Broadly speaking, brain research is divided into two major themes: the study of brain development that seeks to understand how the brain comes to be, and the study of brain function that seeks to understand how the brain controls behavior, cognition and emotion. There has historically been relatively little overlap between the two fields mainly because they have been considered to deal with two distinct aspects of the biology of the brain. More recently however, there is an increased awareness that many aspects of brain function are in part rooted in the molecular history of different neuronal subtypes as well as the specificity and variation of the structural organization of brain networks that emerges during development. In other words, there is now an appreciation that to a non-negligible degree, brain function can well be understood in the context of how the brain develops. This is particularly true for instinctive behaviors that animals are “hard-wired” to perform to maximize their chances of survival in an inherently heterogenous environment full of sensory cues to which the animal will be exposed.
One of the key sensory modalities is vision, and many animals rely on their visual system to navigate the world. One important feature of visual systems is the detection of visual motion information, which is generated both by moving objects as well as by self-motion of the animals which generates motion cues all across the retina. Local motion computation has been considered a paradigmatic neural computation. It relies on a comparison of contrast changes at neighboring points in visual space over time, which manifests at the neuronal level as the emergence of direction-selective (DS) responses. The underlying mechanisms have been worked out in great detail(1,2,3). Global motion patterns generated by self-motion have been shown to be encoded by a population of local direction-selective ganglion cells in the vertebrate retina(4). In flies, wide field neurons downstream of the local direction-selective cells encode global motion patterns, but how they achieve these properties from their presynaptic local direction-selective inputs is not known. The overarching objective of this proposal is to investigate the organizational principles that govern the relationship between developmental trajectories and neural function of motion detection circuits, ultimately tuned to animal behavior, using the fruit fly Drosophila melanogaster as a model system.
The current dogma in the field of motion detection circuit development and function in Drosophila is that a rigid deterministic genetic program a priori pre-configures a precise number of neural stem cells to generate exactly the correct number direction-selective neurons – called T4 and T5 neurons – tuned to precisely 4 cardinal directions of motion in exactly the correct number as to match the number of visual columns of the eye (3,5). According to this model, each of the 800 visual columns is home to two classes of neurons (T4 and T5) that each fall into four subtypes processing motion in four direction(6,7).

Hypothesis and objectives
This simple and attractive model nevertheless faces significant challenges in that it fails to account for an increasing number of conceptual considerations and experimental evidence, notably form our labs but also from many other labs in the field, that clearly show that it cannot account for many observations at both the developmental and functional levels. Specifically, work from our labs has shown that lobula plate progenitors undergo transient amplification and likely generate more neurons than can be accounted for by the current simple model (8), and that at the population level T4/T5 neurons show broad directional tuning and fall into 6 and not 4 functional subtypes (9). Therefore, a significant knowledge gap remains in understanding how developmental and functional mechanisms combine to generate a motion detection circuit that subserves animal navigation. To fill this knowledge gap, and based on our significant published and unpublished evidence (see section Ib below) we hypothesize that the Drosophila motion detection system in fact contains 6, and not 4, functional subtypes of T4/T5 neurons, tuned to 6 global motion patterns generated by self-motion of the animal and that T4/T5 neurons are generated by a developmentally plastic genetic program that matches a posteriori the number of motion detection columns to visual column input via functional plasticity dependent mechanisms.
To test this hypothesis our objectives are to
Objective 1. Understand the developmental mechanisms that give rise to six functional T4/T5 subtypes
Objective 2. Determine how six T4/T5s distribute across four layers while covering retinotopic space
Objective 3. Understand how neurons downstream of T4/T5 compute optic flow patterns relevant for behavior from six T4/T5 subtypes
Impact
How brain neurons self-organize during development to form functional circuits that compute sensory signals and generate behavior remains a key challenge in neuroscience. Until recently, neural development and sensory circuit function were largely separate fields of study. Advances in technology and growing understanding of circuit wiring are now bridging these areas. Using the fruit fly Drosophila melanogaster as a model, we propose a project to dissect how developmental mechanisms shape the anatomy and function of visual motion detection circuits. We aim to uncover the rules linking form and function in neural circuits through genetically encoded developmental steps. These rules likely generalize across brain circuits.
References
1. Yang, H. H. & Clandinin, T. R. Elementary Motion Detection in Drosophila: Algorithms and Mechanisms. Annu. Rev. Vis. Sci. 4, 143–163 (2018).
2. Mauss, A. S., Vlasits, A., Borst, A. & Feller, M. Visual Circuits for Direction Selectivity. Annu. Rev. Neurosci. 40, 211–230 (2017).
3. Borst, A., Haag, J. & Mauss, A. S. How fly neurons compute the direction of visual motion. J. Comp. Physiol. A. Neuroethol. Sens. Neural. Behav. Physiol. 206, 109–124 (2020).
4. Sabbah, S. et al. A retinal code for motion along the gravitational and body axes. Nature 546, 492–497 (2017).
5. Courgeon, M. & Desplan, C. Coordination of neural patterning in the Drosophila visual system. Curr. Opin. Neurobiol. 56, 153–159 (2019).
6. Maisak, M. S. et al. A directional tuning map of Drosophila elementary motion detectors. Nature 500, 212–216 (2013).
7. Fischbach, K. F. & Dittrich, A. P. M. The optic lobe of Drosophila melanogaster. I. A Golgi analysis of wild-type structure. Cell Tissue Res. 258, 441–475 (1989).
8. Mora, N. et al. A Temporal Transcriptional Switch Governs Stem Cell Division, Neuronal Numbers, and Maintenance of Differentiation. Dev. Cell 45, 53-66.e5 (2018).
9. Henning, M., Ramos-Traslosheros, G., Gür, B. & Silies, M. Populations of local direction-selective cells encode global motion patterns generated by self-motion. Sci. Adv. 8, (2022).

This research proposal aims to investigate the relationships among Attention-Deficit/Hyperactivity Disorder (ADHD), sleep bruxism, and psychiatric comorbidities in pediatric populations, focusing on iron deficiency. ADHD is a common neurodevelopmental disorder that often co-occurs with sleep disturbances, including bruxism, which can exacerbate symptoms and negatively impact health. We will examine motor activity patterns and sleep disturbances, assessing their associations with iron deficiency in children aged 6-10 years diagnosed with ADHD per DSM-5 criteria. Using a prospective, cross-sectional design, the research will involve a cohort of children with ADHD and matched controls. Various assessments, including actigraphy, polysomnography, and biomarker tests for iron status, will be employed. By exploring these connections, the project seeks to enhance diagnostic precision and develop innovative therapeutic approaches, ultimately improving the quality of life for affected children.

State of the art: Neurodevelopmental and mental health disorders are among the most common non-communicable diseases, frequently emerging in early childhood. These conditions significantly disrupt functioning and are leading contributors to morbidity in children, highlighting the urgent need for enhanced diagnostic and therapeutic strategies in pediatrics. Their effective management requires an interdisciplinary approach, combining expertise from neurodevelopment, psychiatry, sleep medicine, and more recently, dentistry, to address the complex and overlapping challenges faced by affected children.
Background: Attention-Deficit/Hyperactivity Disorder (ADHD), as defined by the Diagnostic and Statistical Manual of Mental Disorders, is characterized by excessive motor activity (restlessness), inattention, and impulsivity. With a global prevalence of 5.6-7.6% among children aged 3-18 years (Salari et al 2023), approximately 77.9% of youth with ADHD present comorbidities(Danielson et al 2022), complicating diagnosis and management. The relationship between ADHD and sleep disturbances is noteworthy, as sleep disorders can exacerbate ADHD symptoms, while ADHD may also contribute to poor sleep quality.(Migueis DP et al 2023) The forms of disturbed sleep are varied, yet our understanding of how these sleep problems manifest early in neurodevelopmental and psychiatric disorders remains limited.
Sleep bruxism—defined as rhythmic (phasic) or non-rhythmic (tonic) masticatory muscle activity during sleep—has gained attention for its association with sleep-related disorders. Although bruxism is now considered a (motor) behavior rather than a disease, its prevalence among children ranges from 5% to 49.6% (Machado E et al2014). Studies indicate that sleep bruxism is more common in children with ADHD, suggesting a significant correlation between the two (Souto-Souza D et al 2020). Both ADHD and bruxism may share neurobiological mechanisms involving neurotransmitters such as dopamine and norepinephrine, indicating a potential shared etiology. This association warrants exploration for several reasons: they frequently co-occur, with evidence showing a higher prevalence of bruxism in children with ADHD than in the general population.(Souto-Souza D et al 2020) Moreover, ADHD is often linked with other sleep disorders, including insomnia, sleep-disordered breathing and restless legs symptoms (ranging from 11%-42.9%), which may overlap with or exacerbate bruxism.
Restless behaviors, including ADHD, Restless Legs Syndrome (RLS, a movement condition where you have the urge to move your legs when you rest), and the recently identified restless sleep disorder(DelRosso LM et al 2021), frequently present with psychiatric comorbidities and varied medication use. Recent research highlights iron's role in sleep disorders and wake behaviors associated with mental health and neurodevelopmental conditions such as ADHD and autism spectrum disorder. Iron is a trace element crucial for neurotransmitter synthesis, including dopamine, epinephrine, norepinephrine, and serotonin, which regulate emotions, the sleep/wake cycle, movement, attention, memory, and learning. Consequently, perinatal and postnatal iron deficiency may significantly influence behavioral health, affecting both sleep and wake behaviors.
Investigating these interrelationships could yield insights into shared sleep disturbances, enhancing clinical approaches and holistic care for affected patients. These findings may support both transdiagnostic and disorder-specific prevention and treatment targets. Early symptom recognition may facilitate timely interventions, potentially improving the quality of life for children impacted by this “trifecta”.
Objectives: This proposal aims to investigate the relationships among ADHD, sleep disorders (motor activity patterns), psychiatric comorbidities, and iron deficiency in pediatric populations. Specifically, we aim to: a) examine motor activity patterns and sleep disturbances, including bruxism and restlessness, in children with ADHD and their association with iron deficiency; b) explore the role of psychiatric comorbidities in children with ADHD, as well as the impact of family history of iron deficiency on these disturbances. By focusing on the "trifecta of mis(se)diagnosis"—the intersection of ADHD, sleep problems, and restlessness—this study will assess whether sleep bruxism and related sleep disorders exacerbate ADHD symptoms or vice versa. Understanding these connections may lead to more precise diagnostic tools and innovative treatments.
Methodology: This prospective, cross-sectional study will involve 70 children (35 with ADHD and 35 controls) aged 6-10 years, evaluated at the Academic Hospital Robert Debré in Paris (HRD, France). The ADHD group will consist of medication-naïve children diagnosed according to DSM-5 criteria, with subtype classification based on ADHD-C and ADHD-I. Psychiatric comorbidities will be assessed, particularly focusing on children exhibiting the trifecta of ADHD, sleep disturbances, and restlessness. In addition to our clinical neurodevelopmental assessment, we will utilize actigraphy, overnight polysomnography, dental assessment, biomarker tests to evaluate iron status, and gather dietary and family history data on iron deficiency and sleep-related disorders. Primary evaluation criteria include circadian motor activity patterns (measured by accelerometer) and their association with ADHD subtypes; the presence and severity of the ADHD-sleep-restlessness trifecta, including sleep bruxism; and the association of iron deficiency with restlessness, sleep bruxism, and ADHD severity. Secondary evaluation criteria will focus on the role of family history of iron deficiency in predicting restlessness and motor activity changes within the ADHD-sleep-restlessness trifecta, as well as the interactions between psychiatric comorbidities, sleep bruxism, and motor activity disruptions.
Statistical Analysis :Descriptive statistics will summarize demographic and clinical characteristics. Structural equation modeling will be employed to explore the complex relationships among ADHD, sleep disturbances, bruxism, psychiatric comorbidities, and iron deficiency in a comprehensive model.
Ethics and Consent: This study will adhere to the ethical principles outlined in the Declaration of Helsinki, ensuring the protection of participants' rights, dignity, and welfare. Informed consent will be obtained from all participants (or their legal guardians) prior to enrollment, and a thorough assessment of risks and benefits will be conducted. The study protocol will be reviewed and approved by the APHP/INSERM independent ethics committee to ensure compliance with ethical standards and the safeguarding of participant confidentiality. Participation will be voluntary, with confidentiality maintained throughout the study.
Impact: Our current understanding is that ADHD is a 24-hour disorder. By investigating the interplay among ADHD, the ADHD-sleep-restlessness trifecta, psychiatric comorbidities, and iron deficiency, this study seeks to enhance our understanding of these interconnected conditions. Examining the trifecta of mis(se)diagnosis will provide valuable insights into the neurobiological mechanisms underlying ADHD and its related sleep disorders, ultimately improving diagnostic and therapeutic strategies.

Neurodevelopmental disorders (NDDs) affect 10% of children, causing lifelong cognitive and behavioral challenges. The clinical variability of NDDs complicates the search for effective treatments. Our project targets a common feature across NDDs: disruption of stress response pathways, specifically involving Heat Shock Factors (HSFs). HSF2, which regulates proteostasis during stress, also plays a key role in brain development under normal conditions. Using two patient-derived iPSC models of NDDs—both exhibiting HSF2 dysregulation—and their isogenic models with restored HSF2, we aim to pinpoint the specific role of the HSF pathway in NDDs. Through advanced single-cell analyses applied to cerebral organoids derived from these models, we will identify HSF-dependent genes, pathways, and cell populations. These discoveries will be validated through independent approches and will serve as a “readout” for future drug screening.

1/Context & objective

Neurodevelopmental disorders (NDDs) arise from both genetic and environmental factors and affect approximately 10% of children (1), showing broad clinical spectrum that challenges treatment. Regardless of their diverse origins, NDDs share a common dysregulation of cellular stress responses such as the endoplasmic reticulum and the oxidative stress response pathways (2,3). These pathways are interconnected and potentially linked to Heat Shock Factors (HSFs, stress sensitive transcription factors) (4,5), which control proteostasis upon various stresses, with a potential impact on phenotypic outcomes during early development
The involvement of HSF2 in neural progenitor proliferation and radial neuronal migration underlines its importance in normal brain development (6,7), aligning with the understanding that some cell types, especially those with high proteostasis demands, are particularly susceptible to stress (8). Given their dual functionality, HSFs act as integrators of cell stress and development, bridging pathways previously seen as separate between cellular stress and physiological processes (9).
Altered HSF activation, HSF2 in particular, has notably been detected in NDDs like Fetal Alcohol Syndrome (6) Rubinstein-Taybi syndrome (RSTS) (10) and Angelman-like Syndrome (AGS-L) (11). RSTS is caused by mutation in CBP/EP300 genes coding acetyl-lysine transferases, however in RSTS neural models (induced neural progenitor cells and human Cortical Organoids (hCOs)), decreased levels of HSF2 were observed, which led to impaired stress response and dysregulations of cell-adhesion and chaperon proteins crucial for neurogenesis, all known to depend on HSF2 (10). Disorganization of the neurogenic niche and premature neuronal differentiation were also observed in RSTS hCOs compared to healthy donors (HD) and an ongoing multi-omic single-cell analysis (sc-MO), in the host laboratory, already confirms stress factors and cell adhesion genes dysregulations. AGS-L iPSCs that carry a mutation affecting the HSF2 gene expression is another model of NDD supporting the hypothesis that HSF2 is directly involved in the pathogenesis (11).
These promising results set the basis for the further development of HSF-rescue models that will allow to specifically isolate the HSF contribution in each NDD model, RSTS and AGS-like. We have developed novel iPSC models to be derived in hCOs and iNPCs and expand our SC omics analysis to encompass these new isogenic models.
The project aims thus to elucidate the specific role of the HSF2 pathway and its potential interaction with other stress pathways in the ontogeny of NDDs, using RSTS and AGS-L models as paradigms.

2/Deciphering the impact of the restoration of the HSF pathway in RSTS and AGS-like.

- Task1/Phenotyping AGS-L hCOs and isogenic HSF2-rescued RSTS-iPSCs and AGS-L iPSCs
Task1.1: While our results converge toward a role of HSF2 in driving adhesion and stress alterations in RSTS (10), it is also likely that RSTS pathology relies on an intermingling between strictly HSF2-dependent processes, and CBP/EP300 driven mechanisms that could indirectly be also influenced by HSF2. Therefore, to be independent of the genetic background, we will implement an AGS-L neural model using iPSCs derived from a patient carrying a mutation in the HSF2 locus, recently identified (11).
The severe neuropathological deficits observed in AGS-L pathology arise during development. Hence, conducting an initial exploratory characterization of the AGS-L hCOs will provide the basis for the future study aiming at restoring the phenotype. Based on previous team research, we will analyze three differentiation stages (DIV15, DIV25, and DIV45) and assess hCO cellular organization (size, structure, neuroepithelial arrangement) as well as early neurogenesis (neuroprogenitor cells and neuronal marker analyses in hCO sections).

Task 1.2: An HSF2-rescued RSTS model, with stabilized HSF2 protein in patient-derived iPSCs via CRISPR-Cas editing, has been established in the host laboratory, showing initial restoration of RSTS phenotypes. A similar editing strategy has been applied to AGS-L iPSCs.
HSF2-Rescue AGS-L Model: We will advance the HSF2-rescue AGS-L model by generating hCOs. The iPSC-derived AGS-L cells, in which the HSF2 mutation was corrected using CRISPR-Cas, have undergone quality control and assessments of HSF2 transcript and protein levels. Our next focus will be on generating hCOs and characterizing the restoration of phenotypes identified in Task 1.1.
*Expected results
he laboratory's expertise and access to the EnScore platform create a strong foundation for successfully generating hCOs.

- Task 2/ Generating and analysing sc-MO data
Based on the ongoing sc-MO analysis on RSTS & HD hCOs, we will delineate critical steps of the AGS-L and HSF2-rescue AGS-L hCOs analyses.
Task2.1: hCO production and sc-MO
We will produce and collect HD, AGS-L and HSF2-rescue AGS-L hCO (derived from 2-3 clones) at 2 developmental stages. Some hCOs will be used for sc-MO, and some hCOs will be prepared for Task3. 10xGenomics Multiome technology, will allow to follow sc gene expression and chromatin accessibility in the very same cell. It will be conducted on freshly prepared hCOs at the GENOM’IC Platform at the Cochin Institute in coordination with the WISCI Platform (EDC, Labex), following the procedure which was successfully applied on HD and RSTS hCOs.

Task2.2: sc-MO analyses
Comparative scRNA-seq analyses between HD and AGS-L will identify neural cell populations with cell-fate issues, while pseudotime analyses will uncover defective developmental trajectories. Subsequent comparisons of AGS-L and HSF2-rescue AGS-L will reveal HSF2-dependent gene deregulations. Gene Ontology analyses will identify biological pathways impacted by AGS-L and dependent on HSF2. By integrating scATAC-seq and scRNA-seq data, we will link chromatin dynamics and gene expression, refining developmental trajectories and identifying key chromatin regions related to transcriptional changes.
Our analysis will focus on: i) Impaired cell diversity and HSF2-dependent neural populations, ii) Dysregulated proteostasis in HSF2-related stress pathways, iii) Chromatin accessibility patterns correlated with gene expression.
*Expected Results:
This approach will provide a unified view of gene expression and epigenomic landscape at the single-cell level in hCOs, enabling identification of HSF2-dependent gene regulatory networks and affected subpopulations. The iPop-Up platform and IFB bioinformatic cluster are assisting in data analysis.

- Task3/AGS-L hCOs phenotyping and scMO validations
Task3.1: Validation of sc-MO data
To validate the results from Task 2, candidate genes will be analyzed in hCOs from the same and different production batches. We will perform RT-PCR arrays (Anygene) and immunofluorescence to assess the extent of disturbances, relevant signaling pathways, and the most affected cell populations. Our focus will be on genes associated with the phenotypes identified in Task 1 and those revealed in an ongoing parallel analysis of the RSTS models.
*Expected results
Promising results from the RSTS model have shown restored neural marker expression following HSF2 restoration (sc-MO & RT-QPCR). Cross-analysis of data from both NDD models may reveal potential drug screening targets, leading to the next step: replicating HSF2 rescue in AGS-L and RSTS hCOs using drugs from Ksilink (ANR_PRCE) to demonstrate their therapeutic potential in targeting the HSF pathway.
These comparative analyses will help isolate the HSF contribution to the RSTS/AGS phenotype, highlighting its role in driving pathological traits. This study will provide cellular and molecular tools for large-scale drug screening in NDDs with stress-related features.

1. 21606152 2. 25933971 3. 28770435 4. 25465722 5. 35687139 6. 25027850 7. 32147500 8. 36449150 9. 20628411 10. 36385105 11. 34653234 12. 25188634

Stroke is a leading global cause of death and a major factor in dementia. Cerebral small vessel disease (cSVD), a significant contributor to stroke, is also common in aging populations, even in those who haven't experienced a stroke, and plays a role in cognitive decline and dementia. Research suggests that both stroke and cSVD influence the development of neurodegenerative diseases (ND), but the precise mechanisms remain unclear. The proposed study aims to use advanced genomic and multiomic approaches to investigate how vascular conditions contribute to ND, by identifying new genetic risk factors and exploring their role in disease progression.

Stroke is the second leading cause of death worldwide and a major cause of dementia. Cerebral small vessel disease (cSVD), one of the main causes of stroke, is also highly prevalent in older stroke-free persons in the general population (“covert cSVD”), and a major contributor to cognitive decline and dementia. Mounting evidence suggests that stroke and covert cSVD contribute to the pathogenesis and progression of neurodegenerative diseases (ND), but the mechanisms underlying this intricate relation are poorly understood. We propose to use a suite of state-of-the-art genomic and multiomic approaches applied to diverse datasets (on complex and rare forms of vascular brain disease and ND) to explore the molecular underpinnings of the vascular contribution to ND. The PhD students will be involved in the following tasks:
1- Discover novel genetic risk variants (common and rare) and circulating proteins associated with the risk of stroke and covert cSVD through genome-wide association studies in the context of large international collaborations using cross-ancestry approaches
2- Explore how identified genetic variants and proteins predisposing to vascular brain disease modulate the risk of ND occurrence and progression in large publicly available summary statistics and in-house individual level datasets.
3- In the context of rare monogenic forms of vascular brain disease and ND, for which unique datasets are hosted at ICM, explore how common genetic risk variants of stroke, covert cSVD, and ND, combined in integrative polygenic risk scores, modulate the phenotype spectrum of the disease, in particular with respect to risk of cognitive impairment and dementia

Astrocytes form extended intercellular networks, displaying complex calcium activity. However, the specific organization of these astrocytic networks and the precise extent of their functional connectivity in different brain areas remain unexplored. To unveil the functional architecture of astrocytic networks, we propose to develop a data-driven methodology, modeling, statistics and ML that uses as input one-photon invivo calcium imaging to map temporal correlations in activation events among neighboring astrocytes and neurons. Our approach will involve reconstructing functional both networks by organizing individual activation events chronologically and also synchronously. This chronological order creates activity paths that enable the extraction of local astrocyte and neuronal functional correlations.

Summary:
we will construct a graph that mirrors the intricate astrocyte& neuronal functional network. We will apply this method to identify and quantify the possible differences in local network organizations during behavioral task. A deep-learning component will allow to classify the graphs in subgroups. We hope that our innovative method will enable the identification of activation paths among astrocytes and neurons and will allow the characterization of local network functional connectivity, and will quantify distinct connectivity patterns . We will further analyze the refined properties of the present graphs. This approach could shed light on the heterogeneous functional organization of astrocytic& neuronal networks within the brain during behavioural task

Background: Glial and neuronals cells form intricate networks interconnected by gap junctions, facilitating the exchange of ions and small molecules (Giaume et al., 2010; Dallerac et al., 2018). Among these cells, neurons and astrocytes play a crucial role in various functions, including the regulation of potassium flows during neuronal activity (Dao Duc et al., 2015; Breslin et al., 2018), ensuring metabolic needs, influencing the formation of synapses, and modulating neuronal circuit activity (Verkhratsky et al., 1996; Kettenmann et al., 2008; Poskanzer et al., 2011; Poskanzer and Yuste, 2016). Despite their extensive structural connectivity, little is known about the motifs of astrocyte calcium signaling at the network level and their specific paths during spontaneous activity. In neuronal networks, signal transmission occurs through chemical synapses, initiating a chain of events leading to the propagation of action potentials. However, in astroglial networks, only a subset of astrocytes are activated following a calcium burst initiated in an astrocyte (Rusakov et al., 2014; Zheng et al., 2015; Poskanzer and Yuste, 2016; Wu et al., 2019). We posit that the activated network’s path could serve as a basis for reconstructing synchronized astrocytes in parallel to neurons belonging to a preferentially connected network. This hypothesis provides a potential avenue for understanding the selective activation patterns in astrocytic and neuronal networks during various physiological processes.
Calcium imaging analysis has unveiled recurrent activations within neuronal networks, including Up-Down state activity (Cossart et al., 2003). These networks can range in size, from a few cells in cultured environments to extensive populations in brain slices or in-vivo settings (Yuste et al., 2015; Reynolds et al., 2019). In contrast, much less is known about such recurrent activity in astrocytes, particularly regarding the local connectivity patterns. Reconstructing connectivity in dissociated cultured neurons has leveraged spike dynamics sorting during spontaneous activity and numerical simulations (Stetter et al., 2012). Techniques like constructing correlation matrices for place cells in maze-running mice (Villette et al., 2015), or using calcium fluorescence imaging with statistical inference to discern precise spike timings between neurons and reconstruct networks (Mishchencko et al., 2011), have been successful in the neuronal domain. Meanwhile, studies have tracked individual neurons across multiple sessions using spatial and temporal metrics (Johnston et al., 2022). Recent advances in image segmentation have disentangled both astrocytic and neurotransmitter fluorescence dynamics at single-cell and population levels (Wang et al., 2019). While a segmentation method has been developed to study calcium bursting events in astrocytes (Savtchenko et al., 2018), it falls short in extracting causal relationships between neighboring cells and reconstructing local network connectivity.

In this study, we will analyze spontaneous activation patterns exhibited by astrocytes and neurons using calcium fluorescent imaging with GCaMP6f in vivo. Through the development of a computational method, DL and algorithms, we will use time series data from calcium imaging of astrocytes to reconstruct local network organizations. Our analysis will involve examining calcium activity in hundreds of cells across various recording sessions. Following the segmentation and time ordering of calcium bursts, we will map the temporal causal correlation between successively activated cells into paths, revealing levels of connectivity. This critical step will enable the construction of a local astrocytic and neuronal network structure represented in a graph. We will apply this method to recover connectivities during certain task in freely behaving mice. Finally, we will derive various statistics associated with these graphs, such as the number of connected cells, the count of highly connected astrocytes and neurons (hubs) and the properties of the connectivity matrices.

Work to be done :
WP1: identify the position of co-active astrocytes and neurons displaying calcium signaling
To reconstruct the local network functional connectivity, we will first identify the position of active cell displaying calcium signaling detected in brain via the selective expression of the GCaMP6 genetically-encoded calcium indicator. From each active astrocyte region of interest (ROI), we will extract the individual fluorescence time-series. We will then segment these time-series into calcium bursting events. A possible pipeline will be used to reveal the unknown underlying network dynamics. Briefly, the ROIs will be detected by averaging over time light intensity of the whole recording session and then finding the active regions of the image, corresponding to the most luminous ones. We will extract the average Calcium time-series for each ROI and segment them to detect the activation events by extracting local Calcium peaks. To define sequences of activation, that we will call co-activation paths, we will segment the recording session. To extract the co-activations paths, we will order the individual activation peaks in time, to construct a sequence (path) from the first activated astrocyte to the last one present in the subsection. Finally, the ensemble of co-activation paths will be used to construct a graph which represents the connectivity of the underlying network, where each node of the graph is a cell and an edge between two nodes carries a weight equal to the number of times they were consecutively activated in the recording section.
WP2: to build the co-activation path.
Astrocytes and neurons exhibit spontaneous calcium activity, leading to the successive activation of various neighboring cells. In order to study whether these successive activations follow a repetitive order over realizations, we will extract from the Calcium time-series the successive activated astrocytes and neurons that define a co-activation path. Due to the Calcium baseline fluctuation, we will first correct these slow fluctuations to extract the individual dynamics. Once the individual dynamics have been identified, we will collect the sequential activated cells, to build the co-activation path. This procedure will allow us to construct the activation path followed during this subperiod. Finally, we will obtain N activation paths per recording session, that we will use to reconstruct the graph of the astrocytic network.
To conclude this research will allow us to evaluate the level of astrocytic connectivity, we will focus on astrocytes with high connectivity levels.

Les femmes sont les principales victimes des violences conjugales. Le langage joue un rôle central, Il sert à la fois d'outil de manipulation pour l'agresseur et d’expression émotionnelle pour les victimes. Le conditionnement, souvent inconscient, banalise les comportements abusifs et engendre une dépendance émotionnelle, malgré la souffrance. Cette étude adopte les sciences de la complexité pour aborder ces phénomènes sociaux interdisciplinaires. L’objectif est de relier comportements et activations cérébrales, en se concentrant sur le réseau de la peur et celui de la valence. Le réseau de la peur provoque des comportements d’évitement (barrages sémantiques ou physiques), tandis que le réseau de la valence, analysé par des outils d’IA, permet de comprendre la grammaire comportementale des dépendances affectives.

Introduction et Contexte Sociétal
Notre société est confrontée à des risques psychosociaux, des violences sociales et des phénomènes addictifs qui touchent particulièrement les populations les plus vulnérables. Les violences faites aux femmes sont une crise mondiale, avec des chiffres accablants : en 2020, environ 47 000 femmes ont été tuées par leur conjoint, représentant 58 % des homicides liés aux violences domestiques. Plus d’une femme sur quatre, entre 15 et 49 ans, aurait été exposée à des violences conjugales, souvent avant l’âge de 24 ans. En France, près de 300 000 victimes de violences domestiques sont enregistrées chaque année, dont 70 % sont des femmes. Malgré des avancées sociologiques, politiques et philosophiques, les réponses institutionnelles restent insuffisantes pour endiguer ce fléau.
Objectifs du Projet
Le projet vise à :
1. Comprendre les mécanismes sous-jacents qui maintiennent les victimes dans des relations abusives.
2. Améliorer le diagnostic précoce grâce à des outils simples d’analyse de la parole, basés sur une compréhension fine des mécanismes cognitifs et émotionnels, afin d’agir dès les premières phases de violence.
Cadre Scientifique : Conditionnement et Dépendance Émotionnelle
Les violences domestiques sont souvent liées à des processus de conditionnement émotionnel où la victime développe des réponses automatiques, limitant sa flexibilité cognitive. Ce projet repose sur l’étude des réseaux cérébraux impliqués dans la peur et la soumission :
• Le réseau de la peur déclenche des comportements d’évitement physique ou verbal, révélant une emprise psychologique.
• Le réseau de valence traduit l’attachement affectif à l’agresseur et sera étudié via l’analyse automatisée du langage avec des outils d’intelligence artificielle.
Le Syndrome de Stockholm et les Obstacles au Départ
Le syndrome de Stockholm illustre la complexité du lien entre l’agresseur et la victime. Ce phénomène d’attachement traumatique rend difficile pour les victimes de quitter leur agresseur, même après plusieurs agressions. La séparation est souvent entravée par :
1. La stigmatisation sociale, la peur du jugement et des normes culturelles.
2. Les contraintes économiques et la dépendance financière.
3. La persistance du traumatisme et le risque de revictimisation.
Le projet s'intéresse à la transition émotionnelle nécessaire pour briser ce lien et aux processus cognitifs liés à la culpabilité, à la honte et à l’empathie envers l’agresseur.

Méthodologie : Approche Systémique et Sciences de la Complexité
Les sciences de la complexité sont au cœur du projet, car elles permettent d’aborder les phénomènes comportementaux et psychologiques en tenant compte des interactions non linéaires entre les systèmes biologiques et sociaux. L’analyse du comportement et du langage des victimes sera réalisée à l’aide d’outils d’intelligence artificielle comme LinkRData et LinkRBrain. Ces outils permettront d’identifier les régularités linguistiques et les marqueurs cognitifs associés aux syndromes de stress post-traumatique (TSPT).
Stratégie de Recherche
La première étape consistera à analyser l’activation cérébrale des victimes de TSPT, en cherchant des différences entre celles exprimant de la honte et de la culpabilité, et celles qui n’en présentent pas. L’objectif est de valider un modèle théorique reliant ces activations aux comportements observés.
La seconde étape portera sur l’analyse automatique du langage spontané des victimes, afin d’identifier les marqueurs de la soumission et de la peur dans leurs discours. Cette analyse permettra de développer un outil de détection précoce utilisable en clinique.
Implications Éthiques et Santé Publique
Le projet s’inscrit dans une démarche éthique, en se concentrant sur les femmes victimes de violences conjugales qui restent sous emprise psychologique. L’enjeu est de fournir des outils permettant aux professionnels de santé d’intervenir plus tôt et de mieux accompagner les victimes vers la sortie de la violence. En l’absence de soins, ces femmes courent le risque de développer des troubles psychiques (anxiété, dépression, risque suicidaire) et des complications physiques (douleurs chroniques, maladies auto-immunes).
Conclusion
Ce projet propose une approche innovante et multidisciplinaire pour comprendre et diagnostiquer les mécanismes de la violence domestique. En combinant intelligence artificielle et sciences de la complexité, il vise à améliorer le soutien apporté aux victimes et à promouvoir une intervention précoce, essentielle pour briser le cycle de la violence.

This proposal examines the interplay between neuronal and vascular/metabolic activities in absence seizures, marked by brief behavioral cessations linked to intense EEG discharges. Our team has studied these seizures in the GAERS rat model, finding they cluster during activity-rest transitions. We use in vivo sensors like electro-oculography (EOG) and electromyography (EMG) to monitor behavioral markers of seizure susceptibility. Additionally, we employ functional ultrasound imaging (fUS) for real-time tracking of cerebral blood volume in freely moving animals, assessing hemodynamic variations during seizures. Preliminary results indicate significant heterogeneity in vascular responses. Our goal is to map dynamic changes in perfusion and connectivity related to seizure bursts and explore the metabolic implications, particularly the unexpected hypoperfusion of the striatum. Ultimately, our study aims to clarify the neurometabolic mechanisms underlying absence seizures.

There is a growing recognition that addressing major neurophysiological questions requires comprehensive approaches that simultaneously capture neuronal and vascular/metabolic activities — two critical facets of brain function. This proposal focuses on absence seizures, which manifest not as convulsions but as brief (tens of seconds) cessations of behavior associated with a train of intense EEG discharges, distributed across cortical and subcortical areas. While a body of studies has clarified the fine temporal distribution (of the order of seconds) among brain areas of the EEG signature associated with a single absence seizure, elucidating their emergence in the context of an otherwise normally behaving animal (of the order of hours and days) have remained an experimental challenge.
Our team has a long-standing interest in epilepsy and has recently characterized the occurrence of absence seizures in the GAERS rat model, leveraging the association of these seizures with behavioral arrest. By employing multiple in vivo sensors to identify periods of activity and rest, we have revealed a specific organization of absence seizures, which tend to cluster in bursts over a 4-minute period at the transition between activity and rest [Tran 2024]. This aligns with findings suggesting a role for sleep/wake pathways in controlling seizures [Toplu 2023].
These sensors, or activity markers, include monitoring of eye movements (electro-oculography, EOG), neck tone (electromyography, EMG) and linear head displacements (accelerometer, ACCEL), alongside the recording of intra-cortical neuronal activity (electroencephalography, EEG) to identify epileptic seizures (Tran et al., 2023). This technical arsenal has enables us to identify activity-rest transitions as periods of susceptibility to the onset of epileptic seizures. The emerging idea is that seizure periods, characterized by hypersynchronous activity, may correspond to the circuits' attempt to modify their faulty developmental connectivity (Plutino et al., 2024), leading to reconfigurations of the neuronal assemblies involved.
In this study, we will couple multimodal behavioral monitoring (EOG, EMG, ACCEL, EEG) with measurements of hemodynamic variations associated with neuronal structures involved in the production of hypersynchronous activity. To this end, we will use a technique, developed within our team, of functional ultrasound imaging (fUS) on mobile animals [Sieu et al., 2015; Bergel et al., 2020]. Unlike fMRI, which requires an immobile subject and the repetition of multiple identical sequences to extract a signal related to the level of O2 in the blood, fUS tracks the local volume of blood in the brain (Cerebral Blood Volume, CBV), in real time from freely moving subjects. This technique provides instantaneous CBV variations across the entire height of the rat brain (15mm), with a spatial resolution of the order of 150 ?m2 for ultrasound shots at 15MHz.
Preliminary results from our team indicate significant heterogeneity in the vascular response during EEG seizures, with fast kinetics occurring in phase with the seizures and slower kinetics on the timescale of seizure bursts. We will employ a combination of electrophysiological recordings, which allow for behavioral monitoring, and imaging techniques to localize the structures specifically involved in the transition between activity and rest. Our goal is to characterize transition-related variations in the amplitude and timing of hypo- or hyper-perfusion in these regions, as well as to explore changes in connectivity between these structures, ultimately constructing a dynamic spatiotemporal map of vascular activity related to seizure bursts.
While we have demonstrated a high degree of concordance in the power of activity markers, this systematic study will detail the temporal sequence of behavioral criteria —such as neck contraction, oculomotor activity, animal movement, and head orientation—providing insights into the degree and timing of motor and cognitive impairment during activity-rest transitions, especially when these transitions are accompanied by seizure bursts.
An additional part of the project will build on findings from Sieu et al. (2015), which show hypoperfusion of the striatum during known activation/hyperperfusion of the thalamocortical loop. This counterintuitive result, which challenges the principle of neurovascular coupling, raises questions about the metabolic activity associated with hypo- and hyperperfusion. While ultrasound imaging in mobile rats reveals changes in blood volume, it does not measure blood oxygen levels. Therefore, this part of the project will involve combining specially designed ultrasound probes with optic fibers to simultaneously access fUS imaging and functional near-infrared spectroscopy (fNIRS), which measures blood oxygen levels. We will synchronize fUS-EEG data with near- and far-infrared signals that assess blood oxygen levels at varying depths in the functional imaging plane, focusing on structures in the motor cortex, thalamus and striatum, which show distinct profiles during epileptic seizures.
Ultimately, this metabolic information could bridge the gap between the variability of vascular dynamics and the existence of massive electrographic seizures, paving the way for a detailed description of the neurometabolic and vascular reconfigurations underlying absence seizure episodes.

Tran H, Mahzoum RE, Bonnot A, Cohen I. Epileptic seizure clustering and accumulation at transition from activity to rest in GAERS rats. Front Neurol. 2024 Jan 24;14:1296421. doi: 10.3389/fneur.2023.1296421. PMID: 38328755; PMCID: PMC10847272
Toplu A, Mutlu N, Erdeve ET, Sariyildiz Ö, Çelik M, Öz-Arslan D, Akman Ö, Molnár Z, Çarçak N, Onat F. Involvement of orexin type-2 receptors in genetic absence epilepsy rats. Front Neurol. 2023 Nov 30;14:1282494. doi: 10.3389/fneur.2023.1282494. PMID: 38107640; PMCID: PMC10721972.
Plutino S, Laghouati E, Jarre G, Depaulis A, Guillemain I, Bureau I. Barrel cortex development lacks a key stage of hyperconnectivity from deep to superficial layers in a rat model of Absence Epilepsy. Prog Neurobiol. 2024 Mar;234:102564. doi: 10.1016/j.pneurobio.2023.102564. Epub 2024 Jan 19. PMID: 38244975.
Sieu, L.-A., Bergel, A., Tiran, E., Deffieux, T., Pernot, M., Gennisson, J.-L., Tanter, M., Cohen, I., 2015. EEG and functional ultrasound imaging in mobile rats. Nat. Methods 12, 831–834.
https://doi.org/10.1038/nmeth.3506
Bergel, A., Deffieux, T., Demené, C., Tanter, M., Cohen, I., 2018. Local hippocampal fast gamma rhythms precede brain-wide hyperemic patterns during spontaneous rodent REM sleep. Nat.
https://doi.org/10.1038/s41467-018-07752-3

Le sommeil lent profond (SLP) intervient dans la mémorisation et la clairance des protéines toxiques via le système glymphatique. Dans la maladie de Parkinson (MP), le sommeil est fréquemment altéré mais quelles sont les répercussions sur les fonctions du sommeil ? Cette thèse étudiera le lien entre SLP et progression clinique et cognitive dans la MP, ainsi que l’effet d’un traitement nocturne des symptômes parkinsoniens sur le SLP. Les caractéristiques électrophysiologiques des ondes lentes en SLP et leur dynamique enregistré au cours d’une nuit de sommeil seront analysées afin de les corréler à des données cognitives et d’imagerie dans une première cohorte longitudinale. Elles seront aussi comparées entre sujets avec et sans traitement nocturne dans une deuxième cohorte. Nous émettons l'hypothèse que la diminution de la profondeur du SLP est liée à une dégradation cognitive plus marquée, et qu'un traitement nocturne pourrait améliorer la qualité du SLP.

Contexte
Dans la maladie de Parkinson (MP), le sommeil est fréquemment altéré, affectant à la fois la qualité de vie des patients par des troubles comme l’insomnie, et la structure même du sommeil. Certaines anomalies, telles que le trouble du comportement en sommeil paradoxal (TCSP), sont associées à un pronostic défavorable pour les fonctions motrices et cognitives. Par ailleurs, ce trouble peut être prodromal d’une MP. Ces altérations du sommeil sont liées aux dépôts d’agrégats anormaux d’alpha-synucléine dans les circuits neuronaux régulant le sommeil et l’éveil, et reflètent leur distribution dans le tronc cérébral et le diencéphale. Ainsi, l’étude du sommeil par polysomnographie chez les patients atteints de MP apparaît particulièrement prometteuse pour explorer de façon innovante la dynamique des réseaux neuronaux et leurs fonctions.
Des recherches récentes ont souligné l’importance du sommeil lent profond (SLP) dans diverses fonctions, notamment la consolidation de la mémoire et la potentielle élimination des protéines toxiques via le système glymphatique. Une caractéristique marquante du SLP est la présence d’ondes lentes, larges et synchronisées dans l’ensemble du cerveau. La quantité de SLP est étroitement liée à la qualité et à la durée de l’éveil. Selon le processus homéostatique : une période d’éveil prolongée induit un rebond du SLP lors de la nuit de récupération, visible par une augmentation de l’activité en ondes lentes. Au cours de la nuit, on observe un déclin physiologique de cette activité, marqué par une diminution de l’amplitude et de la pente des ondes lentes. Cette régulation homéostatique joue un rôle essentiel dans les bienfaits physiologiques et cognitifs du sommeil. Premièrement, les ondes lentes contribuent à l’élimination des métabolites du système nerveux central via le système glymphatique. Deuxièmement, elles favorisent la consolidation de la mémoire par des modifications synaptiques spécifiques. À l’inverse, la privation de sommeil induit des ondes lentes durant l’éveil, perturbant les fonctions cognitives et réduisant la vigilance.
Les ondes lentes semblent donc constituer un marqueur pertinent pour explorer les processus physiologiques impliqués dans la mémoire et la clairance du liquide céphalo-rachidien. Malgré cet intérêt potentiel, les études exhaustives sur les ondes lentes et leur dynamique au cours de la nuit chez les patients parkinsoniens restent rares. On sait que l’activité en ondes lentes est positivement corrélée avec l’amélioration de la mémoire de travail après une nuit de sommeil chez les patients traités par des médicaments dopaminergiques mais on ne connaît pas leur lien avec la progression cognitive. Par ailleurs, bien que la prise en charge des troubles du sommeil améliore la qualité subjective du sommeil, son impact sur les caractéristiques des ondes lentes n’est pas encore établi, et pourrait représenter une piste thérapeutique prometteuse.

Objectifs
L’objectif de cette thèse est double : i. explorer le lien entre le SLP via une analyse précise des ondes lentes et les atteintes cognitives et leur progression dans la MP et dans le trouble du comportement en sommeil paradoxal isolé (TCSP) qui est la phase prodromale de la MP ; ii. évaluer l’impact d’un traitement nocturne par apomorphine sur le SLP.
Méthode
- Population
Deux sets de données déjà collectés (étude ICEBERG et étude APOMORPHEE) seront utilisés. Les sujets de la cohorte ICEBERG (168 patients avec MP, 60 patients avec TCSP et 60 témoins sains) ont réalisés annuellement pendant une période de suivi de quatre ans : i) une évaluation clinique du sommeil et des questionnaires ; ii) des évaluations motrices ; iii) des évaluations neuropsychiatriques et cognitives (tous les 2 ans) ; iv) une évaluation des symptômes non-moteurs ; v) une vidéo-polysomnographie d'une seule nuit (première année) avec 3 à 8 électrodes ; et vi) une imagerie cérébrale (IRM et datscan) tous les deux ans. L'étude APOMORPHEE (46 patients avec une MP) est une étude randomisée, en double aveugle, contrôlée avec placebo, multicentrique, en cross-over. L'étude en cross-over comprenait deux périodes de traitement (soit apomorphine, soit placebo) séparées par une période de sevrage de 14 jours. Les sujets avaient un enregistrement du sommeil d'une seule nuit (8 électrodes) à la fin de chaque période.
- Analyse du signal
Le SLP sera analysé en utilisant de l’analyse spectrale et une caractérisation semi-automatisée des ondes lentes (amplitude, pente, distribution pendant la nuit, durée et densité). Chaque nuit sera divisée en 2 parties (début/fin de nuit) pour évaluer le changement dynamique de la quantité de temps de sommeil lent profond, des caractéristiques de la SWA et des ondes lentes pendant la nuit.
- Corrélation clinico-électrophysiologique
Objectif 1 :
Dans la cohorte ICEBERG, on corrélera l’analyse du SLP (définie précédemment) avec l’évaluation cognitive en cross-sectional (données de la baseline) et en longitudinal (données du suivi) ainsi qu’avec des marqueurs d’imagerie d’atteinte cognitive.
Objectif 2 :
Dans la cohorte APOMORPHEE, on comparera l’analyse du SLP chez les patients avec MP avec ou sans perfusion nocturne d’apomorphine afin d’évaluer l’effet de la thérapie sur le SLP.
Hypothèse
Hypothèse 1 : On attend une diminution des amplitudes des ondes lentes, et une altération du sommeil lent profond associés aux scores cognitifs et peut-être un marqueur dans le SLP de mauvais pronostic.
Hypothèse 2 : On attend d’améliorer les indicateurs d’un SLP de bonne qualité (plus d’activité en ondes lentes, plus large amplitude, …) chez les patients sous apomorphine.
Points forts et limites
Le point fort de ce projet est : 1) d’utiliser des cohortes déjà constituées, prospectives, homogénéisées avec une taille plus importante par rapport aux études précédemment faites, 2) de développer des techniques utilisables dans le futur pour l’analyse du SLP, 3) de contribuer de façon rigoureuse à comprendre le rôle des altération du sommeil sur le développement et la progression cognitive de la MP et de la forme prodromale de TCSPi, 4) évaluer l’impact d’une thérapie nocturne sur la structure du sommeil.
Les limites tiennent à la qualité du signal et à son homogénéité même si lees données polysomnographiques pour la cohorte ICEBERG ont déjà été utilisées pour une autre étude sur le signal et sont donc analysables.
Résultats attendus
Une meilleure caractérisation de l'altération du sommeil dans la MP est essentielle pour explorer le lien entre les troubles du sommeil et sa physiopathologie. Une analyse complète et dynamique du SLP fournira des informations supplémentaires sur le système glymphatique, l'homéostasie synaptique et la consolidation de la mémoire dans la MP. La découverte de nouveaux marqueurs permettant de prédire la conversion ou la progression du phénotype peut permettre une sélection plus précise des patients pour des interventions thérapeutiques spécifiques visant à prévenir ou à ralentir la progression de la MP. L’utilité de la perfusion nocturne d'apomorphine sur l’amélioration des anomalies électrophysiologiques du sommeil, fournissant ainsi des preuves en faveur de son administration nocturne systématique et précoce afin de réduire l'impact délétère d'un sommeil de mauvaise qualité sur la progression de la MP.

This thesis aims at quantifying cerebral functional connectivity (fC) during early neurodevelopment in human neonates using functional ultrasound (fUS). fC is envisioned as a biomarker candidate for early detection of neurodevelopmental disorders. Currently, diagnoses often occur around age 2-3, delaying treatment.
fUS is a new neonatal-suitable method enabling high-resolution imaging of brain activity by detecting small blood flow changes, for a more comprehensive fC assessment. fC data at this age (24 weeks of gestation to term) are scarce due to challenging implementation. The PhD candidate will longitudinally assess fC using fUS in 76 neonates to quantitatively characterize their neurodevelopmental trajectories. To do so, they will develop new fUS-fC analysis methods, such as dynamic fC and homotopic fC, and will seek to identify potential biomarkers for abnormal development. Data will represent an unprecedented collection on human early development and emergence of fC networks.

This thesis project focuses on the quantitative characterization of the cerebral functional connectivity (fC) during early neurodevelopment in human neonate, using a new imaging modality called fUS (functional ultrasound) with preterm and term neonate patients. The potential impact is the early identification of signs of neurodevelopmental disorders, especially those linked to perinatal pathologies, addressing a significant societal and public health concern (~ 1 million newborns/year in Europe). To date, diagnosis is often delayed until the child reaches 2 or 3 years of age when cognitive or behavioral alterations become evident, hindering the deployment of effective therapies.

fC, which quantitatively measures how different brain regions co-activate, is considered a potential biomarker for early diagnosis because fC deficits is likely to precede observable structural connectivity lesions or later behavioral alterations. However, the translation of fC into neonatal clinical practice has been hampered by the limitations of existing modalities, such as low-resolution surface measurements (EEG, NIRS) or unsuitability for neonatal use (fMRI).

Recently we have introduced fUS to the field of neuroimaging, and made successful translational proof of concepts for its use in human neonates1–3. This imaging technique leverages ultrafast imaging framerates (several thousands images/second) to detect small cerebral blood flows which thanks to neurovascular coupling, enables imaging of cerebral activity. The modality is practicable on neonates, and gives excellent temporal (<1s) and spatial resolution (~250µm). We have shown2 that this technique can assess the thalamo-cortical connectivity and discriminate stages of neurodevelopment (preterm vs term), but with a very partial quantification assessing a small field of view for technical reasons and therefore a limited number of functional areas.

The goal of this PhD project is to draw a much broader picture of fC assessment using fUS during the early development, in order to: 1/ to define a neurodevelopmental trajectory through the quantitative characterization of the evolution of fC in the neonatal period (including prematurity) and 2/ potentially determine in this quantification the markers (biomarkers) of an abnormal neurodevelopmental trajectory identified by a neurodevelopmental assessment at 2 years. This will be done using a first-of-its-kind fUS neonatal neuroimager resulting from a 3-year joint development between the Physics for Medicine Paris institute and the startup Iconeus. This fUS neuroimager will be used to acquire 3D scans of the neonates’ brain, and to record fUS data along with simultaneous EEG data, with a largely improved brain-wide field of view for assessment of cerebral structures involved in somato-sensory, auditory and visual connectivity networks. It will enable longitudinal monitoring (4 to 5 time points per patients) of neonates in different conditions (preterm, term, anoxo-ischemic encephalopathy, intra-uterine growth restriction)., The PhD student will elaborate on fUS dynamic fC analysis and homotopic fC that we have shown to be relevant2. Dynamic fC enables to derive brain states from the data, whose temporal occurrence is modified by neurodevelopment. Homotopic fC reveals spatial functional heterogeneities that are linked to the underlying structural organization (such as the grey/white matter boundary). Those markers will be used to characterize the modification of fC with increasing gestational age for the preterm neonate and compare the fC at term equivalent for term-born and preterm-born neonates. We will also compare the early fC assessment with the neurodevelopmental assessment at 2 years of age.

This PhD project leans on two ongoing research projects that enable its concrete implementation. The first one is the ANR JCJC project CONEXUS ending in March 2025. Its aims were to build a whole-brain-capable fUS neuroimager for neonatal use with a miniaturized wearable probe, compliant with medical devices regulations, to test it in terms of imaging performance and acoustic safety, and to design and get the approval for a small clinical trial. This project is completed: the neuroimager is built, compliant for safety, the clinical protocol has been accepted and the ANSM approved the study. The second research project is the 3-year clinical trial itself, which is funded both by the CONEXUS ANR and by a research collaboration agreement between Physics for Medicine and Iconeus. It will be co-lead by Pr Valerie Biran and Dr Charlie Demené, in line with a longstanding collaboration between Robert Debré Hospital and Physics for Medicine. This clinical trial will recruit 76 neonates. A post doc is funded by the CONEXUS ANR up to March 2025. Therefore a PhD position spanning the 3 years of the clinical trial will be ideal to make the most of the unique dataset that will be acquired during that time.

The acquisition of this neuroimaging dataset is a world premiere, and therefore high risk high gain in terms of scientific conclusions. However the publication roadmap for the PhD student can include both technical papers detailing the technological advances clinically tested during the project and life science oriented papers reporting assessment of the fC in neonates.
Robert Debré Hospital is a central center for neonatal intensive care with the patient capacity required for the success of the clinical trial, and Physics for Medicine is a world leader for biomedical ultrasound with all the expertise needed in ultrasound and fUS imaging for the scientific leadership needed for that project.

1. Demene, C. et al. Functional ultrasound imaging of brain activity in human newborns. Sci. Transl. Med. 9, eaah6756 (2017).
2. Baranger, J. et al. Bedside functional monitoring of the dynamic brain connectivity in human neonates. Nat. Commun. 12, 1080 (2021).
3. Faure, F. et al. Quantification of brain-wide vascular resistivity via ultrafast Doppler in human neonates helps early detection of white matter injury. J. Cereb. Blood Flow Metab. (2024).

The cerebellum plays a central role in the learning and execution of complex, coordinated movements. Such movements can only be learnt through repetition. Our group is working to understand the synaptic, network and algorithmic mechanisms of this key function. Projects at several levels are available:
- in vitro investigation of the central synapse of the cerebellar network between granule cells and Purkinje cells, focusing particularly on the prorperties of silent synapses
- in vivo recording and optogentic interventions to test competing hypothesised learning algorithms
- theoretical modelling to analyse possible algorithms and generate experimental predictions

Animals can learn complex, rapid movements during which feedback cannot operate, not least because the result of the movement may only be known after the corresponding motor command has ended. Such movements can only optimised iteratively - by practice. The cerebellum is thought to play a key role in such learning. It is thought that the information acquired during learning is stored as modifications of the hugely abundant synapses between granule cells and Purkinje cells in the cerebellar cortex. These synapses transform the massive amounts of sensory and contextual information arriving in the cerebellum via mossy fibres into motor control contributions that return to multiple levels of the motor system. Modifications of these synapses are believed to be implemented by synaptic plasticity guided by the climbing fibres, whose activity reflects information about movement errors. The consensus theory for the underlying algorithm suggests that granule cell synapses that are active during commands that lead to motor errors undergo long-term depression. As part of an interdisciplinary collaboration with theoreticians, we have shown that this algorithm is extremely limited and have proposed an alternative, which is akin to a trial-and-error scheme (Bouvier et al, 2018). A key element of our algorithm is that the evaluation signal - climbing fibre activity - can also function as a probe of altered motor commands. Our theory predicted novel plasticity rules that we were able to verify in vitro. The current project aims to extend on-going in vivo recordings during motor learning as well as implement optogenetic interventions to test the competing theories of cerebellar learning. At a cellular level, information storage is effected by synaptic plasticity of granule cell-Purkinje cells. Surprisingly, a large majority of these synapses appear to be silent (Isope and Barbour, 2002). Another axis of our work is therefore the study of the extreme variation of properties of this type of synapse, combining patch-clamping in slices, super-resolution and two-photon microscopy. Finally, identifying plausible theories, generating predictions and comparing these with the experimental data requires analytical and theoretical work that constitutes the final axis of our project. We have are currently extended spike-sorting analyses in preparation for this analytical work (Wyngaard et al).

Bouvier et al (2018) Cerebellar learning with perturbations. eLife 7:e31599. https://elifesciences.org/articles/31599
Isope and Barbour (2002) Properties of Unitary Granule Cell -> Purkinje Cell Synapses in Adult
Rat Cerebellar Slices. J. Neurosci. 22:9668
Wyngaard, Llobet and Barbour: https://github.com/BarbourLab/lussac

Recent research suggests a shared cognitive foundation for space, time, and number. Temporal Momentum Effect (TME), where participants overestimate additions and underestimate subtractions of temporal intervals, suggests a potential spatial representation of time.
This project aims to investigate the cognitive origins of the TME, exploring whether it arises from estimation biases or motor production processes. EEG recordings will be used to examine neural correlates of temporal arithmetic, focusing on beta power and its relation to the TME and internal variable coding. Bayesian modeling will be employed to assess whether the TME can be explained by statistical decision theory principles.
Finally, the project will directly probe the spatial-temporal link by manipulating the spatial presentation of temporal operands. By investigating these aspects, the study aims to deepen our understanding of the cognitive mechanisms underlying time perception and its potential spatial representation.

It has been proposed “that space, time and number are part of the essential toolkit that adult humans share with infants and with many other nonhuman animals (p. ix; (1)).” Both, time and non-symbolic number processing have been found to be largely organized according to Weber’s law. Against this background, it is not surprising that various conceptual frameworks suggest common coding principles for numerical and temporal information (6, 7). Spatial coding has been proposed as a pivotal principle in both domains.
According to the ‘mental number line’ (MNL) hypothesis smaller numbers are located left from larger numbers. The spatial organization of the MNL gives rise to cognitive biases such as the Operational Momentum Effect (OME). The OME refers to the tendency to overestimate the outcome of additions and underestimate the outcome of subtractions (Prado & Knops, 2024) and originates from attentional movements along the MNL. Addition deviates attention towards the right side of space, while subtraction deviates attention to the left side of space.
Time has also been described as being intricately associated with space. For example, Buzsaki and Llinas (2017) propose that at the neural level sequence generation in parietal cortex and Hippocampus serve to (re)construct the ordered sequence of events in space and time. Paton and Buonomano (2018) distinguish between temporal information that is inherent in sensory events (sensory timing) versus motor production (motor timing). Accordingly, Wittmann and colleagues (2010) found neural correlates of duration encoding in two separate cognitive networks that were devoted either to (a) the perceptual encoding of the duration or to (b) the motoric reproduction of the encoded durations.
A recently described phenomenon in the temporal domain (Bonato et al, 2021) resembles the OME: Participants were sequentially presented with two white noise stimuli (between 150 ms and 1650 ms) that served as operands. Participants were asked to either combine the temporal duration of the white noise (addition) or to estimate the temporal difference between them (subtraction). Participants then produced the estimated outcome via button press. Compared to a baseline reproduction condition without addition or subtraction, participants overestimated the duration of addition and underestimated the duration of subtraction – comparable to the OME in the numerical domain. The authors interpret these results as a momentum-like effect in the temporal domain and suggest “that some aspects of time representation (either sensory or conceptual) are spatial in nature.” This would nicely integrate the temporal momentum effect (TME) into the family of representational momentum biases and be suggestive of common underlying mechanisms that operate on space, time and number.
In two recent experiments, we replicated the TME when participants produced the estimated outcomes for combinations of two intervals, but found no TME with a psychophysical staircase comparison paradigm. This tentatively points to the idea that the TME emerges at the motor production stage. Since the production method capitalizes on an ‘omnibus’ response measure that comprises estimation, motor preparation and execution, we cannot draw any firm conclusion about the nature of the actual estimate.
In three studies, the current project aims at pinpointing the cognitive origin of the TME (estimation bias vs. motor bias), using EEG as well as Bayesian modelling, and probes the idea that time is represented spatially.

To better understand the cognitive dynamics underlying the TME, Study 1 aims at testing the neurocognitive correlates during the mental combination of two temporal intervals by recording voltage changes from the scalp. Participants will be asked to combine two temporal intervals and produce an estimate of their sum or difference (temporal arithmetic task). In line with previous studies (e.g. Kononowicz & van Rijn, 2015), we will analyse whether trial-to-trial beta power positively correlates (predicts, postdicts) with the length of produced duration, and whether it reflects the TME (i.e. the subjective difference from the correct outcome). Additionally, participants will engage in a second order judgment of their produced estimate on a 5-point scale to indicate whether they think their produced duration was way too small, exact, or too long (Kononowicz, Roger, van Wassenhove, 2019). This will allow us probe whether neural markers of the internal variable coding of the target duration reflects both first and second order judgments.

Assuming a well-known regression to the mean across the stimulus range (including the operands and the outcomes), the use of different ranges for both operations by Bonato et al. (2021) would automatically produce the observed difference between addition and subtraction and the underestimation of subtraction compared to baseline. It does not, however, explain the overestimation for addition compared to the baseline. We will explore whether the TME can emerge from the principles of statistical decision theory. Unfortunately, the hitherto conducted studies on the TME do not allow to reliably assess the biases in a Bayesian ideal observer model. Experiment 2.1 of Study 2 aims at providing the data for such a model. In a temporal arithmetic task, participants produce estimates of the outcome but also of the first and second operand (on separate trials, respectively). The arithmetic operation (addition/subtraction) and the to-be-reproduced element (operand 1, operand 2, outcome) on a given trial will be determined in a pseudorandomized fashion.
Experiment 2.2 will be identical to experiment 2.1 but will use a larger operand range to explore the impact that the regression to the mean will have on the estimates and the model parameters (priors, posteriors).

In a series of experiments, Study 3 will probe more directly the idea that time and space interact by manipulating the spatial position of visually or auditorily presented operands along the horizontal axis. If the mental representation of temporal duration entails spatial references along this axis with longer durations being associated with the right side of space and shorter durations with the left side of space, we would expect the congruence between presented durations and internal (spatial) representation to modulate performance. This may be reflected in either the accuracy of produced outcome or its precision. Operands 1 and 2 will be presented in a lateralized fashion from left to right or vice versa, producing a potential congruence of their perceived motion direction with the displacement along the mental time line. Spatial distance and temporal interval between operands will be manipulated.

Bonato M, D'Ovidio U, Fias W, & Zorzi M (2021) A momentum effect in temporal arithmetic. Cognition 206:104488.
Buzsaki G & Llinas R (2017) Space and time in the brain. Science 358(6362):482-5.
Dehaene S & Brannon E eds (2011) Space, time and number in the brain: Searching for the foundations of mathematical thought (Elsevier Academic Press), p 374.
Kononowicz TW, Roger C, van Wassenhove V (2019). Temporal Metacognition as the Decoding of Self-Generated Brain Dynamics, Cerebral Cortex, 29(10), 4366–80.
Kononowicz TW, van Rijn H (2015). Single trial ? oscillations index time estimation, Neuropsychologia, 75:381–9.
Paton JJ & Buonomano DV (2018) The Neural Basis of Timing: Distributed Mechanisms for Diverse Functions. Neuron 98(4):687-705.
Prado, J., & Knops, A. (2024). Spatial attention in mental arithmetic: A literature review and meta-analysis. Psychonomic bulletin & review.
Wittmann M, Simmons AN, Aron JL, & Paulus MP (2010) Accumulation of neural activity in the posterior insula encodes the passage of time. Neuropsychologia 48(10):3110-20.