Prefrontal – accumbens neural dynamics abnormalities in mice vulnerable to develop food addiction

Pablo Calvé | UPF

Food addiction is closely associated with eating disorders including obesity and binge eating. However, the neural alterations underlying this disorder have not been yet clarified. We have trained mice in an operant model to develop food addiction and we recorded neural activities in the medial prefrontal cortex (mPFC) and the nucleus accumbens (NAc) to identify the dynamic neural abnormalities underlying the development of this behavioral disorder. Interestingly mPFC-to-NAc signaling was disrupted at high frequencies during lever-press decision-making for reinforcement in addicted animals. Moreover, addicted mice exhibited reduced low gamma oscillations and theta-gamma coupling in the NAc during reward expectancy. Disrupted mPFC-to-NAc connectivity and gamma synchrony in the NAc correlated with increased reinforcement levels, underliying the functional relevance of these abnormalities. The CB 1 antagonist rimonabant rescued both neural alterations. These findings suggest that abnormal mPFC-NAc circuit dynamics are candidate mechanisms underlying specific behavioral alterations during food addiction.

Accumulation of evidence during perceptual decision-making in mice

Alexis Cerván | IDIBAPS

During perceptual decision-making, neural circuits must faithfully represent relevant stimulus features, transform them into decision evidence, accumulate this evidence along the stimulus duration and commit to a choice. However, despite the central role that this computation has in decision-making, the dynamics of evidence accumulation are not yet well understood. Here, we have developed a novel auditory two-alternative forced-choice (2AFC) task, in which head-fixed mice listen to two concomitant, intensity-fluctuating stimuli, each presented through a speaker on either side of the head. Mice have to infer which sound (left or right) was louder on average and lick in the associated spout to receive a water reward. We find that mice can learn the task and show a sigmoid probability of a rightward choice. However, the temporal weighting of the evidence is not generally uniform and can be idiosyncratic as revealed by reverse correlation analysis (i.e. psychophysical kernels). Moreover, we find that mice exhibit a tendency to repeat their previous response and that this trial history bias is best explained as repeating lapses rather than a bias term that alters the decision criteria of the categorization, that is an upward rather than horizontal shift of the psychometric curve (repeating lapse = 0.21 ± 0.04, alternating lapse = 0.05 ± 0.01; p=0.001, n=7 mice, data are mean ± s.e.m., two-tailed paired-samples t-test). Together, our results show that evidence integration is heterogeneous across individuals and that is affected by non-sensory biases that limit subjects’ performance.

Regular and sparse neuronal synchronization are described by identical mean field dynamics

Pau Clusella  | UPF

Fast oscillations (30-200Hz) are a prominent component of neural activity. Recently, exact neuronal mean-field theories for heterogeneous networks of quadratic integrate and fire (QIF) neurons have been achieved, which explain that neurons frequency- lock due to feedback inhibition. This gives rise to fast collective rhythms at the frequency of the synchronized neurons. Yet, th suitability of such models to faithfully describe neuronal oscillations is seriously challenged by a remarkable dynamical feature of single neurons during episodes of fast oscillations: Neurons do not fire regularly with a fast rhythm, but sparsely at a very low rate. Theoretical studies have demonstrated that such sparse synchronization emerges in populations of spiking neurons with inhibitory feedback and noise. Here, we exploit the fact that mean field theories for QIF neurons can be extended to include noise, and show that QIF networks with inhibitory feedback and noise robustly display sparse synchronization. Moreover, we demonstrate that fast collective oscillations are largely independent on whether single cell dynamics is periodic and fast, or stochastic with low firing rates.

Modeling repulsive serial dependence in schizophrenia

Joana Coutinho | IDIBAPS

Current working memory contents have been shown to be attracted towards previous recent memories. This effect is called serial dependence and is dependent on the length of the current memory delay length. Interestingly, serial dependence is disrupted in some patient populations, such as in schizophrenia and anti-NMDA receptor encephalitis. In these patient groups, a repulsive instead of an attractive serial effect can often be observed. Understanding the mechanisms by which serial dependence is disrupted in the patient populations can provide insight on the physiopathology of these diseases. Previous modeling work supported by evidence from non-human primates has shown that activity-silent traces are likely to underlie attractive serial dependence. Prefrontal persistent activity during the memory delay encodes the memory item and, by engaging short-term plasticity mechanisms, it leaves a trace of increased spiking probability in the coded location even after persistent activity disappears at the end of the trial. This memory trace then has an attractive effect on subsequent memories. However, the repulsive effect observed in patients has so far been modeled using a shift in the input to the neural population, relaying its generation to another brain region. Here, we ask if a direct alteration in prefrontal cortex can produce the inversion of the serial dependence effect. To produce a delay-dependent serial repulsion, we propose two mechanisms. Firstly, we alter the strength of the activity-silent traces through inverting the ratio of facilitation and depression of the synapses. For a second model, we implement another activity-silent process based on adaptation currents in network neurons that bias neural activity away from the previously perceived item. Both models propose possible mechanisms of how serial dependence can be altered in patient groups through cellular and synaptic changes in prefrontal cortex, and further studies are necessary to disentangle them.

Optimal dynamic allocation of finite resources for many-alternatives decision-making

Francesco Damiani | UPF

The capability of identifying the best in a set of noisy or uncertain options shapes our interaction with the environment, underlying pivotal cognitive processes [1, 2, 3, 4]. In a decision problem, noisy observations are integrated over time, to evaluate, compare, and choose from them. The required computation involves allocating attention or other finite neuronal resources dynamically over the alternatives to make the best choice. The issue of allocating finite resources has been recently studied [5, 6, 7] under the context of the so-called breadth-depth (BD) dilemma, but how they should be divided and dynamically allocated in an optimal manner is not known. Here, we introduce a novel perspective, designing a recurrent neural network (RNN) to deal with the BD dilemma from a dynamical point of view.

LSD-induced increase of Ising temperature and algorithmic complexity of brain dynamics

Giada Damiani | Neuroelectrics

Understanding the fundamental properties of global dynamics of the brain is a topic of growing interest in computational neuroscience. Statistical physics methods can be used to study the brain as a nonlinear dynamical system with a large number of interacting degrees of freedom operating close to a critical point. Near criticality, complex systems generate structured data with power laws and high entropy (apparent complexity) and, notably, display enhanced sensitivity to weak perturbations, maximal information flow, and long timescales, while the dynamics of the system collapse to a lower-dimensional (center) manifold [Jirsa & Sheheitli, 2022]. Experimental data suggest that brains operate close to such critical boundaries, with altered states of consciousness appearing to move the brain away (e.g., anesthesia) or towards (psychedelics) criticality [Carhart-Harris & Friston, 2019]. Prior work by Ezaki et al [2017, 2020] using the generalized Ising model – one of the simplest known systems displaying phase transitions (critical phenomena) – has shown that different brain states may be mapped into a pairwise maximum entropy Ising model at varying distances from criticality. Here, we use this framework to analyze resting-state fMRI BOLD data from fifteen subjects in a control condition (placebo) and during psychedelic ingestion (LSD) collected from a prior study [Carhart-Harris et al., 2016]. To address the limited quantity of data available from typical fMRI studies, we first fit an “archetype” generalized Ising model representative of the entire AAL-parcellated dataset using maximum likelihood estimation with a pairwise maximum entropy model (MEM) [Ezaki et al., 2017]. The archetype model system temperature is then adjusted for each individual and condition using the same method. We analyze the resulting set of temperatures to show first that, at the group level and in both conditions, the model is near criticality in the paramagnetic phase. Second, the individualized Ising temperature increases significantly under the effects of LSD compared to the placebo condition (p = 2 x 10-4). That is, LSD ingestion shifts the system away from the critical point between paramagnetic and ferromagnetic phases, to a more disordered state. Then, we estimate the Lempel-Ziv-Welch (LZW) complexity of the binarized BOLD data (flattened along the spatial dimension first) for each participant and condition, and from the synthetic data generated with the individualized model using the Metropolis algorithm. We find that while LZW complexity computed directly from experimental data does not reveal an interesting statistical relationship, presumably due to the short length of the time series and group size, LZW complexity computed using the archetype model correlates strongly with condition (p = 7 x 10-5) and individualized temperature (p = 3 x 10-6). This study suggests, together with prior ones, that the effects of LSD increase the entropy rate of brain dynamics. In agreement with Ezaki et al., 2020, we find the brain state in the placebo condition is already above the critical point, with LSD resulting in a shift away from criticality into a more disordered state, which is in contrast with previous predictions [Carhart-Harris & Friston, 2019] and studies [Atasoy et al., 2017, Girn et al., 2022].

Amplitude and Phase Dynamics Reconstructed From Experimental Data

Silvia Daun | University of Cologne

Coupled neural systems of large scale, such as brain regions, demonstrate oscillatory activity within different frequency ranges in measured electrophysiological signals. The traditional way of estimating the coupling between such systems is the analysis of the interaction between either the amplitudes or the phases of the signals. However, for a better understanding of the systems and their couplings it is important to consider both the dynamics of the amplitude and the phase. Recent advances in the field of amplitude and phase modeling have demonstrated that an oscillating system can be reduced to a simpler dynamical system consisting of amplitude and phase using a transformation (a parametrization), which is related to the eigenfunctions of the Koopman operator. In this work, we demonstrate the possibility of reconstructing amplitude and phase dynamics directly from experimental data. For this purpose, we use Dynamic Causal Modelling (DCM) and have adjusted the modeling component of DCM to simulate a simple dynamical system of amplitude and phase. By means of Bayesian inference, DCM is able to calculate the coupling parameters of the system by comparing the output of the simulated dynamical system with the phase and the amplitude of the measured signals. For successful convergence, however, DCM requires a good initial approximation of the system’s parameters. Using two coupled radial isochron clocks as examples, we demonstrate a method for obtaining the initial approximation for the parameters of the transformation and the coupling by using the approximation of the eigenfunctions of the Koopman operator. The results of this work can be used to reconstruct the amplitude and phase dynamics of large-scale neural systems using neurophysiological measurements of high temporal resolution, such as EEG and MEG. The reconstructed dynamical system then enables the analysis of the coupling between distant brain regions, as well as the construction of a mathematical model for further simulation and study of various neural pathologies and disorders of the brain.

Synaptic conductance estimation problem: discerning between the excitatory and the
inhibitory contribution in spiking regimes

Rosa Maria Delicado Moll | Universitat de les Illes Balears

Revealing the information that a neuron receives from other neurons and distinguishing between excitatory and inhibitory inputs is a problem that has been investigated in recent years to provide valuable information about local connectivity in the brain. Experimentally, even thought the membrane potential can be extracted, the synaptic conductances are difficult to estimate due to the diversity of synaptic inputs and their unattainable conductances. Hence, inverse methods are required to unveil this information. Most studies focus on the subthreshold regime, where computational methods have been developed that allow a good estimation of the time course of these variables. However, the procedures developed in the spiking regime are scarce due to the nonlinearities presented in the data. In this work, we present a new computational strategy that takes into account two important measures of the membrane potential trace, which are the spike amplitude and the spike frequency. With these measures, the method allows, in addition to estimating synaptic conductances, to separate the excitatory contribution (gE) from the inhibitory one (gI). Our results show that the time course of gE and gI can be well estimated with mean squared errors of order 10^-5.

Machine Learning-based Support Network Extraction for Neural State Characterization

Michael DePass | Universitat de Barcelona

Discriminative support network extraction is a relatively new technique which involves two principal steps. First, machine learning classifiers are employed to distinguish the classes, or neural states, of interest. Second, the most important features are extracted and visualized. This support network of important features serves as a signature of the classification and can be used, considering the feature set upon which the classifiers were trained, to characterize the neural states being analyzed. Though discriminative support networks have been used for neural state characterization in the past, a thorough analysis of the methods used for their extraction as well as their statistical relevance, remains to be performed. Absent such analyses, support networks remain difficult to interpret and require additional analyses to be performed post-extraction to establish their scientific relevance. Thus, we developed a discriminative support network extraction pipeline involving three classifiers and three feature importance algorithms in combination with statistical analyses to elucidate the statistical relevance of the extracted networks. Furthermore, we evaluated the techniques based on their computational complexity, classification accuracy, and replicability over repeated extractions. Multinomial logistic regression (MLR), K nearest neighbors (KNN), and 1D convolutional neural networks (CNN) were used to classify five neuro-motor states associated with a reach-to-grasp task. Subsequently, recursive feature elimination (RFE), beta regression coefficients, and perturbation-based feature importance algorithms were used to derive the support networks. CNN-based methods resulted in the highest classification accuracy though the derived support networks exhibit lower replicability and statistical significance compared to other techniques. MLR+RFE, however, resulted in the highest replicability and statistical relevance.

Neural network dynamics underlying biased perception in motion discrimination tasks

Miguel Donderis | Centre de Recerca Matemàtica

Perception is influenced by past choices. In fine discrimination tasks, a categorical choice biases perceptual reports about the direction of motion away from the decision boundary. This bias has been explained using neural encoding-decoding models1 and Bayesian principles, but a mechanistic explanation in terms of neural circuit dynamics is lacking. Here, we develop a neural network model that addresses this issue and study the integration of spatially modulated input in a bump attractor network. We have previously shown that this attractor network can integrate a direction stimulus and track the stimulus average in the phase of the activity bump almost optimally. To model post-decision biases, we then study analytically and through simulations this network by introducing different spatially structured stimulus inputs (i.e. attention or urgency signals). We find that their different Fourier modes lead to different fixed points that determine the temporal evolution of the phase of the bump. The estimation bias curve shows a maximum bias close to the decision boundary and lowest values for the fixed points. To validate our model, we compare our results with psychophysical data obtained under different experimental conditions (varying stimulus coherence, fine vs. coarse discrimination). Our network model provides a comprehensive computational framework to study the neural mechanisms underlying stimulus estimation and perceptual categorization and their interaction.

Characterizing phase irregularity and phase locking at two spatial scales from elec-
troencephalographic recordings of epileptic seizures

Anaïs Espinoso | UPF

Epileptic seizures are related to abnormal excessive or synchronous neuronal activity. In this work we analyse multi-channel intracranial electroencephalographic (EEG) seizure recordings from sixteen patients with pharmacoresistant focal-onset epilepsy. The EEG allows us to asses synchronization at different spatial scales. Thus, in our study we present two approaches on two different scales, and eventually we combine them. The first approach is local and applies the coefficient of phase velocity variation V to individual EEG signals. Following previous work [1], we found that this measure showed that the epileptic process reduces phase irregularity. The second approach is non-local and analyses the degree of phase locking between the dynamics of measured EEG signals using the normalized multivariate phase locking μ. We perform this non-local analysis to quantify the temporal evolution of the degree of phase locking in three different periods: prior to, during and after the seizure. This analysis is performed among all channels and sets of channels belonging to separate recording electrodes. Moreover, using V we define sets of channels using a degree of participation criteria during the epileptic seizure. Since seizures are not a steady-state dynamics, apart from the degree of phase locking, the EEG frequency content shows a complex temporal evolution across the duration of the seizure. Thus, apart from the analysis of broadband signals, the study of phase locking is also performed in frequency bands. Our results reveal a variety of spatiotemporal patterns of multivariate phase locking among different seizures of the same patient and among patients. In general, there is an increase of synchronization and a decrease of phase velocity variation across all channels when the seizure starts. Furthermore, in some cases, a low level of synchronization is obtained during the seizure. In many cases, after the end of the seizure, there is a pronounced hypersynchronous activity. In conclusion, conceptually simple and easy to compute univariate and multivariate EEG signal analysis techniques can help to study the evolution of seizures.

Imagining what was there: looking at an absent offer location modulates neural responses in
OFC

Demetrio Ferro | UPF

When making choices, we allocate our fixations to each contemplated option, and tend to look longer at more valued ones. The purpose of fixation during choice remains unknown. Here we examined behavioral and neural activity of rhesus macaques (Macaca mulatta) performing a two-options risky choice task in which offers occurred in sequence, each followed by a long (600 ms) blank screen delay period. As expected, we found that subjects allocated their gaze towards offer presentation locations and spent more time looking at the most valuable offers; after factoring out value, we found that more looking time was devoted to the chosen offer. Surprisingly, we found the same pattern before choice execution when the screen was blank: subjects spent more time fixating to the locations where valuable offers had occurred. Moreover, we found that neural encoding of the offers’ expected values in orbitofrontal cortex (OFC) is modulated by eye position even when the screen is blank. Specifically, when gaze is directed to a former offer location, its value is more strongly encoded in OFC, while the encoding of the alternative offer value is suppressed. The same modulatory effects by gaze on value encoding are observed later in the trial when monkeys were supposed to report their choice while both offers were presented simultaneously. Our results provide evidence that eye position reflects an internal deliberation process that modulates the encoding of imagined content, providing a new window to study the hidden dynamics of decision-making.

Sequential Episodic Control 

Ismael Tito Freire González | UPF

State of the art deep reinforcement learning algorithms are sample inefficient due to the large number of episodes they require to achieve asymptotic performance. In contrast, Episodic Reinforcement Learning (ERL) algorithms, inspired by the mammalian hippocampus, typically use extended memory systems to bootstrap learning from past events to overcome this sample inefficiency. However, such memory augmentations are often used as mere buffers, from which isolated past experiences are drawn to learn from in an offline fashion (e.g., replay). Here, we demonstrate that including a chaining bias in the acquired memory content derived from the order of episodic sampling improves both the sample and memory efficiency of ERL algorithms. We test our Sequential Episodic Control (SEC) model in a foraging task to show that storing and using integrated episodes as event sequences leads to faster learning with fewer memory requirements as opposed to a standard ERL benchmark, Model-Free Episodic Control, that buffers isolated events only. To assess the generality and robustness of SEC, we study the effects of memory constraints and forgetting and compare it to an ablated version of the model that lacks the sequential bias. Furthermore, our model provides a novel perspective on how fast sequential hippocampal episodic memory systems could bootstrap slow cortical and subcortical learning, subserving both habit formation and deliberation in the mammalian brain.

Hidden hearing-loss and the cocktail-party problem

Juan Fuentes | Australian Hearing Hub

Solving the auditory cocktail party problem (ACPP) refers to the ability to isolate sound sources of particular interest in noisy environments. Over-
exposure to loud sounds can produce a lesion called cochlear synaptopathy or hidden hearing-loss (HHL), because it leaves hearing sensitivity intact while significantly damaging high-threshold auditory nerve fibres. Physiological and perceptual effects of HHL haven’t been fully understood, but is likely to impact the ability to listen in noisy environments. Here we adapt a theoretical neuroscience framework, proposing a neural code for sound intensity, designed for contributing to solve the ACPP, in order to yield a continuum of optimal solutions based on maximum-entropy distributions, that are subsequently treated as Bayesian priors. We assessed single- and multi-neuron recordings from the inferior colliculus of mice and gerbils through this framework, datasets were composed of two groups, noise-exposed (NE, or HHL group) and sham-exposed (SH, or controls). We inferred via maximum a-posteriori the parameters of the optimisation state in which these neural populations were likely to operate. Our results are consistent with what has been shown previously with several types of stimuli, i.e., NE group showing higher coding utility (based on mutual information) than controls only for low sound intensities, whereas controls showing consistently higher coding utility for moderate to high sound intensity than NE group. This is as well consistent with an over-representation of low-threshold fibers in NE group, and an homeostatic central gain increase, compensating for the lack of synapses due to the lesion. The information-theory based framework along with the neural code proposed, suggests a potential mathematical primitive for the development of a measure for HHL through utility functions, and a door to understand how brains gather information to ultimately, being able to solve in real-time the ACPP, within negative SNR scenarios.

Generalities and idiosyncrasies characterizing the behavior of humans performing a 2AFC task presenting trial-to-trial correlations

Alexandre Garcia Duran | IDIBAPS

Recent studies have thoroughly characterized the behavior of rats performing a Two-Alternative Forced Choice (2AFC) auditory task, in which the probability to repeat the previous stimulus category is varied in a blockwise fashion. These studies showed that rats exhibit a transition bias: a tendency to alternate/repeat the previous response using an estimate of the probability given the recent trial history. However after error trials, the transition bias was null (Hermoso-Mendizabal et al. 2020). Even though it is suboptimal, this so-called reset strategy has been shown to be highly robust and present in many task variants (Molano-Mazón et al. 2021). On the other hand, the reaction times of rats performing the 2AFC task has been shown to be governed by two independent processes: one that depends on the accumulation of the stimulus evidence and a second, stimulus-independent process that only depends on the time elapsed since the beginning of the trial (Hernández-Navarro et al. 2021). Here we have investigated the behavior of human subjects performing an auditory 2AFC task presenting the same type of correlations experienced by the rats. We found that their strategies were more heterogeneous, with some subjects displaying a clear reset strategy while others developed a more optimal strategy. Furthermore, the reaction times of the human subjects showed evidence of being influenced by the two processes mentioned above, suggesting that the existence of two the different mechanisms described in rats may be a general feature present across species.

Postnatal environmental enrichment enhances memory through distinct neural mechanisms in healthy and trisomic female mice

Thomas Gener | Institut Català de Nanociència i Nanotecnologia

Stimulating lifestyles have powerful effects on cognitive abilities, especially when they are experienced early in life. Cognitive therapies are widely used to improve cognitive impairment due to intellectual disability, aging, and neurodegeneration, however the underlying neural mechanisms are poorly understood. We investigated the neural correlates of memory amelioration produced by postnatal environmental enrichment (EE), a rodent model of cognitive therapy, in diploid mice and the Ts65Dn mouse model of Down syndrome (trisomy 21). The experiments were conducted in females as Ts65Dn females are particularly sensitive to EE. We recorded neural activities in brain structures key for memory processing, the hippocampus and prefrontal cortex, during rest, sleep and memory performance in mice of both genotypes reared in poor or enriched environments. We identified enhanced neural activities in the hippocampus of enriched wild-type animals across different brain states (augmented pyramidal activity and gamma synchrony) and enhanced neural activities associated with memory processing (augmented theta-gamma coupling and sleep ripples). As in Ts65Dn males, trisomic females exhibited hypersynchronous theta and gamma rhythms across different brain states in the hippocampus and prefrontal cortex, along with enlarged ripple activities and disordered hippocampus-to-prefrontal cortex gamma signals associated with memory deficits. These pathological neural activities correlated strongly with memory impairment and were attenuated in their trisomic EE-reared peers. Our results suggest distinct neural mechanisms for the generation and rescue of healthy and pathological brain synchrony, respectively, by EE and put forward hippocampal-prefrontal hypersynchrony and miscommunication as major targets underlying the beneficial effects of EE in intellectual disability.

Modeling the effects of transcranial electrical stimulation in the context of epilepsy

Maria Guasch-Morgades | Neuroelectrics

Epilepsy is a chronic brain condition that poses significant strain on the quality of life of patients. Despite continuous advances, current therapies for epilepsy are still inefficient or with major adverse effects for many patients. In this context, non-invasive neuromodulation methods such as transcranial electric stimulation (tES) have recently produced promising results by reducing seizure frequency or mitigating its effects [1]. An increasingly popular way to personalize tES is through individualized computational models of the patient. In this direction, the aim of the current work is to develop an optimization strategy for tES in epileptic patients using personalized brain models created from SEEG/EEG/MEEG, dMRI and MRI data. In particular, we will use realistic, personalized biophysical models of the brain with embedded networks of neural mass models (NMM) connected using the structural connectivity of each patient, also referred to as hybrid brain models. In this study, we present the first step towards this goal. We have examined how different stimulation tES strategies applied directly to the personalized model of an epileptogenic focus or to nearby connected nodes can be used to mitigate, and even prevent, seizure initiation and propagation. For this purpose, we have modeled the focus and nearby regions as personalized NMM that can generate and propagate realistic epileptic seizures [4]. We then applied different electric fields targeting different combinations or nodes. Finally, the effectiveness of the different strategies has been evaluated in terms of specially designed methods to quantify seizure frequency and epileptic spike density reduction. In our first models, we show that tDCS can reduce the probability of initiation and propagation of seizures. Further work will focus on the effects of other modalities of stimulation such as tACS.

Phase-locking patterns underlying effective communication in exact firing rate models of neural networks

Gemma Huguet Casades | Universitat Politècnica de Catalunya and Centre de Recerca Matemàtica

Macroscopic oscillations in the brain have been observed to be involved in many cognitive tasks but their role is not completely understood. One of the suggested functions of the oscillations is to dynamically modulate communication between neural circuits. The Communication Through Coherence (CTC) theory (Fries, 2005, 2015) proposes that oscillations reflect rhythmic changes in excitability of the neuronal populations and input volleys must arrive at the peaks of excitability of the receiving population to communicate effectively. Here, we present a modeling study to explore synchronization between neuronal circuits connected with unidirectional projections. We consider an Excitatory-Inhibitory (E-I) network of quadratic integrate and fire neurons modeling a Pyramidal-Interneuronal Network Gamma (PING) rhythm. The network receives an external periodic input from either one or two sources, simulating the inputs from other oscillating neural groups. We use recently developed mean-field models which provide an exact description of the macroscopic activity of the spiking network (Montbri ́o et al., 2015; Dumont & Gutkin, 2019). This low-dimensional mean field model allows us to use tools from bifurcation theory to identify the phase-locked states between the input and the target population as a function of the amplitude, frequency and coherence of the inputs. We identify the conditions for optimal phase-locking and effective communication in term of the response of the target network. We find that faster oscillatory inputs than the intrinsic network gamma cycle show more effective communication than inputs with similar frequency and are more robust to distractors.

Edge-centric analysis of stroke patients: An alternative approach for biomarkers of lesion recovery

Sebastian Idesis | UPF

Most neuroimaging studies of post-stroke recovery rely on analyses derived from standard node-centric functional connectivity to map the distributed effects in stroke patients. Here, given the importance of nonlocal and diffuse damage, we use an edge-centric approach to functional connectivity in order to provide an alternative description of the effects of this disorder. These techniques allow for the rendering of metrics such as normalized entropy, which describes the diversity of edge communities at each node. Moreover, the approach enables the identification of high amplitude co fluctuations in fMRI time series. We found that normalized entropy is associated with stroke lesion severity and continually increases across the time of patients’ recovery. Furthermore, high amplitude co-fluctuations not only relate to the lesion severity but are also associated with patients’ level of recovery. The current study is the first edge-centric application for a clinical population in a longitudinal dataset and demonstrates how a different perspective for functional data analysis can further characterize topographic modulations of brain dynamics.

Evaluation strategies in realistic planning scenarios

Leonie Lambertz | UPF

In everyday life, we often need to commit to potential future actions even when direct feedback is not available. In these cases, the estimated reward of an action will depend not only on a single decision but on a sequence of them, and planning is required to forecast the outcome of sequential decisions. Several studies have highlighted common human strategies to reduce the computational costs of planning (Callaway et al., 2021; Huys et al., 2015, Moreno-Bote & Mastrogiuseppe, 2021). How humans nevertheless plan under novel real-world scenarios remains poorly understood. We have developed a novel task to investigate human evaluations of options in scenarios resembling real life planning. Participants perform the task on a computer screen while their gaze position is recorded. Each trial begins with a written planning scenario (e.g. “plan your birthday party.”). Afterwards 9 images are presented which are grouped into 3 categories using a colored frame around them (e.g., 3 birthday cakes, 3 party locations and 3 decorations). Options within one category are spatially grouped whereby horizontal and vertical orientation of categories is randomized between trials. Participants are asked to select one option per category to make the best possible plan – the one with the highest subjective value – under the provided context. After a 5s delay from stimulus onset, subjects can make the decisions without time pressure, arbitrarily choosing the first option among categories. The images disappear from the screen when the 3 options have been selected. At the end of each trial participants rate the difficulty of the choice within a category and the relevance of a category of options in the specific planning situation. As expected, our preliminary results indicate a positive correlation between fraction of looking time at options within a category and the perceived difficulty of choosing among them in 3 out of 4 subjects (Spearman’s rank correlation, rho-value = 0.38 – 0.5, p-value < 0.05). Interestingly, perceived importance of a category of options predicts a lower fraction of looking time at the respective options in 2 out of 4 subjects (Spearman’s rank correlation, rho-value = -0,23 – -0.33, p-value < 0.05) suggesting subjects’ strong initial preference for the chosen option. We have quantified gaze switching as the number of times when the participant shifts the gaze from one option to another. We differentiated whether the switch occurs between options from the same category or different categories. The results reveal that 63% of switches were performed intra-category, suggesting a high tendency to evaluate options in the same category before moving to another one. Conversely, a significantly lower fraction of of the switches were performed inter category (63% intra, 36% inter; t-test, statistic=9,747, p-value< 0.05). These results suggest that, while most of the evaluation is realized by considering and comparing single within category options, a still significant contribution to the subjective evaluation comes from the out-of-category options. This may indicate that participants are involved in a process of imagining the options in the context of potential future choices. Although still preliminary, the employed task seems to provide a new approach to investigate real life planning strategies as participants evaluate options subjectively through imagination and memory and can freely move between all levels of the decision tree.

Generation of slow BOLD signals from coupled neural mass models in the gamma range

Èlia Lleal | Neuroelectrics

Neural mass models (NMM) provide a useful framework for constructing whole-brain models and have been used extensively to model electrophysiological activity (EEG/MEG/LFPs) in different frequency bands with success. If the appropriate forward models mapping state variables to physical measurables are available, they can be used to personalize whole-brain models assimilating multimodal data such as EEG and fMRI-BOLD. Here we explore how to generate fMRI-BOLD signals from NMMs operating at fast electrophysiological frequencies. In earlier work, BOLD signals have been generated from NMM signals (typically the membrane potential or synaptic activity) by applying a Hilbert transform to retrieve the power envelope in this band. This has been motivated by the observation of a correlation between BOLD and gamma power activity [1]. Another more direct, physiologically-grounded route is using a realistic model such as the Balloon model to transduce synaptic activity into BOLD. This raises an interesting point: the Balloon model acts as a nonlinear low pass filter (<0.5 Hz) of synaptic activation, which drastically reduces the power of frequencies above a few Hz, with an effect in the gamma range. Here we show that low-frequency components can be generated in the spectrum of signals of whole-brain models with coupled NMMs. This is due to the phenomenon of wave beating with nonlinear transduction in the population voltage to firing rate transfer function (Freeman sigmoid in our case). We demonstrate this first with a simple model of two coupled PING NMMs tuned to oscillate in the gamma band and then in a whole-brain model with a realistic connectome. Finally, we show that these slow oscillatory components in synaptic activity are related to the high-frequency power envelopes.

Critical networks and biomarkers of disease progression in early psychosis 

Ludovica Mana | UPF

Psychotic disorders are characterized by heterogeneity in etiopathology, clinical presentation and individual trajectory, complicating diagnosis and treatment. Despite the increasing amount of studies investigating significant alterations, the mechanisms underlying the emergence and progression of these disorders remains unclear. It is therefore fundamental to improve our ability to differentiate between clinical subgroups and to move towards a more personalised approach. Here, we aim to highlight critical neural correlates in early stages of psychosis, and to identify relevant biomarkers correlating with clinical staging. In this work in progress study, we include resting-state fMRI and DTI data from a cohort of 129 healthy controls and 94 patients with early psychosis stratified into four distinct groups based on the severity of their condition and their ability to recover after the first episode. We investigate whether global and local measures of functional connectivity and network properties such as measures of integration and segregation significantly differ in the pathological brain as compared to the healthy controls. When patients were  compared to the healthy condition as a single group, disregarding for disease stages, no significant differences could be found between the measures explored so far. This is not surprising, considering the high heterogeneity of the group. Most interestingly, we explore whether measures of functional connectivity and network properties could be used as a biomarker of progression. To this aim we are currently investigating whether any of these empirical measures can detect relevant differences between stages, and whether they correlate with symptoms and manifestations. Finally, we use a whole-brain model to fit the, empirical data and to extract hidden properties of the network, and the correlates underlying the disruptions in the interplay of brain dynamics and connectomes. Specifically, the model combines functional dynamics and anatomical structure and describes local dynamics of single brain regions using the normal form of a Hopf bifurcation. This model allowed us to identify critical networks involved in damaging mechanisms, relevant for, emergence of the disease as well as in potential compensatory mechanisms involved in the recovery process. In particular, the disruption of local, dynamical properties in areas previously identified as hubs of connectivity emerged as a critical biomarker for progression of disease. These preliminary results allow us to progress in understanding some of the mechanisms underlying the emergence and the progression of psychosis, and could open the way for possible future therapeutic applications.

Deep imagination is a Close Optimal Policy for Planning in Decision Trees under Limited Resources

Chiara Mastrogiuseppe | UPF

Many decisions involve choosing an uncertain course of actions in deep and wide decision trees, as when we apply for a PhD position: we have to choose a research topic, then an expert researcher, a specific project to pursue, and so on. In these cases, exhaustive search for the best sequence of actions is not tractable due to the large number of possibilities and limited time or computational resources available to make the decision. Therefore, planning agents need to balance breadth – considering many actions in the first few tree levels — and depth – considering many levels but few actions in each of them — to allocate optimally their finite search capacity. This so-called breadth-depth dilemma has been studied before in the framework of single-level decisions (Moreno-Bote et al. 2020, Ramirez-Ruiz et al. 2021, Vidal et al, 2021). We provide efficient analytical solutions and numerical analysis to the problem of allocating finite sampling capacity in one shot to infinitely large decision trees, both in the discounted and the undiscounted case. We characterize the optimal policies in model-based planning for the allocation of finite resources on large, stochastically and binary rewarded decision trees. The expected rewards are unknown to the agent and can be learnt by first allocating a finite search capacity of C samples over the nodes of an infinite decision tree. We assume that samples need to be allocated at once, modelling situations where feedback from samples is delayed. We think of the allocation as an internal mental process through which the agent updates their belief of what they would expect from a node if they actually visited it. Once information is acquired, the best expected course of actions can be chosen. We have developed a ‘diffusion-maximization’ backward algorithm, which allows to compute the exact value of playing trees for agents with average capacity C in different environments (whose richness is defined by the probability p of getting a positive outcome). The required number of operations for playing a tree with b branches and depth d is O(bd^2), way more efficient than the O(d^b”) of standard dynamic programming (Browne et al., 2012). Consistently for the many possible families of allocation strategies studied, we find that the optimal policy is sampling few (b ∼ 2) branches in the very first levels in order to reach the deeper leaves of the tree. Exceptions are found only in poor environments, although deep allocations perform very close to the optimality even in those regions. We test the performance of a very-deep heuristics by allocating samples in two not-branching paths in all the parameter space and we found agents incurring in relatively low loss in value. A bias towards the optimality of deep allocations was expected, as the reward accumulated over a path is bounded by its length, and thus exploring more deeply leads to potentially large expected cumulative rewards. Surprisingly, we found that the optimality is stable with respect to the introduction of a discount factor exponentially reducing the rewards in the deeper levels of the tree. All together, our results indicate that deep allocations are as a close-to-optimal strategy in the ‘one-shot’ allocation of finite resources over large decision trees. These findings can provide a theoretical foundation for why human reasoning is pervaded by imagination-based processes. These results are consistent with the observed human bias to mentally simulating few and long courses of actions (Simon, 1972) before making a decision.

Dynamics of top-down feedback axons during wakefulness and sleep

Pedro Mateos-Aparicio | Universitat Internacional de Catalunya/Institut de Neurociències Universitat Autónoma de Barcelona

During wakefulness, information from the outside world reaches primary sensory areas and then is integrated in more complex representations in higher order association cortical areas via bottom-up feedforward connections. Superimposed on the feedforward pathways, top-down feedback connections from higher order areas carrying contextual information and internally generated complex representations shape neural representations and information processing in primary sensory areas. It has been shown that top-down influences in the primary visual cortex (V1) modulate the gain of visual responses during visual stimulation, for example. However, little is known about the dynamics of feedback connections during states in which the feedforward drive is greatly reduced such as sleep. To address this question, we used 2-photon calcium imaging in awake head-fixed mice to investigate the activity of retrosplenial cortex axons in V1 during wakefulness and sleep. In parallel, we performed electrocorticography (EcoG) and electromyography (EMG) recordings coupled with pupil and locomotion tracking to accurately score awake and sleep states into active wake (AW), quiet wake (QW), NREM, and REM sleep. Using a visual stimulation paradigm during wakefulness, a subset of retrosplenial axons showed stimulus position and orientation tuning. In addition, we compared the spontaneous activity during wakefulness and sleep. Our results indicate that in a subset of axons, both mean ΔF/F and frequency of spontaneous Ca2+ events increased during REM sleep compared to awake or NREM periods. Finally, we further explored the relationship between subsets of axons that showed visual tuning responses and those that increased their activity during REM sleep. These results provide experimental evidence of increased drive in top-down feedback axons during REM sleep. Therefore identify the retrosplenial cortex as one of the areas targeting layer 1 that may provide key input during dreaming in the form of increased apical drive during REM sleep.

Neural correlates of multi-timescale behavior

Manuel Molano-Mazón | Laboratorie de Neurosciences Cognitives

Rats can quickly learn to capitalize on the trial sequence correlations of two-alternative forced choice (2AFC) tasks after correct trials, but consistently deviate from optimal behavior after error trials, when they waive the accumulated evidence (Hermoso Mendizabal et al. 2020 Nat. Comm., Molano-Mazón et al. 2021 bioRxiv). We have recently developed a pre-training protocol for Recurrent Neural Networks (RNNs) in a naturalistic task presenting more than two possible choices that recovers this so-called reset strategy (Molano-Mazón et al. 2021 bioRxiv). Here, we perform population analyses of the activity of these pre-trained networks and show that they form an accurate representation of the trial sequence statistics independently of the outcome in the previous trial. After error trials, the reset is implemented by a change in the network dynamics which temporarily decouples the categorization of the stimulus from the across-trial accumulated evidence. We tested these predictions in neural recordings obtained from the Dorsomedial Striatum (DMS) of rats performing the 2AFC task and found that neurons in this area encode the sequence statistics both after correct and after error trials and that the reset strategy is mainly implemented by discarding the information about the previous choice when it was incorrect.

Phasic Activation of Dorsal Raphe Serotonergic Neurons Increases Pupil Size

João Morais |  Institut Català de Nanociència i Nanotecnologia

Transient variations in pupil size (PS) under constant luminance are coupled to rapid changes in arousal state, which have been interpreted as vigilance, salience, or a surprise signal. Neural control of such fluctuations presumably involves multiple brain regions and neuromodulatory systems, but it is often associated with phasic activity of the noradrenergic system. Serotonin (5-HT), a neuromodulator also implicated in aspects of arousal such as sleep-wake transitions, motivational state regulation, and signaling of unexpected events, seems to affect PS, but these effects have not been investigated in detail. Here we show that phasic 5-HT neuron stimulation causes transient PS changes. We used optogenetic activation of 5-HT neurons in the dorsal raphe nucleus (DRN) of head-fixed mice performing a foraging task. 5-HT-driven modulations of PS were maintained throughout the photostimulation period and sustained for a few seconds after the end of stimulation. We found no evidence that the increase in PS with activation of 5-HT neurons resulted from interactions of photostimulation with behavioral variables, such as locomotion or licking. Furthermore, we observed that the effect of 5-HT on PS depended on the level of environmental uncertainty, consistent with the idea that 5-HT could report a surprise signal.These results advance our understanding of the neuromodulatory control of PS, revealing a tight relationship between phasic activation of 5-HT neurons and changes in PS.

You don’t always forget: Mechanisms underlying working memory lapses

Tiffany Oña-Jodar | IDIBAPS

Working memory (WM) is central for cognition and is impaired in many brain disorders including those hypothetically mediated by NMDA receptor (NMDAR) hypofunction. Evidence suggests that network attractor states underlie WM maintenance and that failures are mostly caused by fluctuation-driven transitions, but direct evidence for this is still lacking. To investigate what makes WM fail we used a simple two-alternative delayed response task in which mice listen to a lateralized auditory stimulus and, after a variable delay (duration D=0-10 s), they have to lick the associated lateral port. Mice accuracy decreased with delay showing that there were forgetting errors. They also showed a repeating bias, i.e. a tendency to repeat the previous choice, which was however independent of delay. Inactivation of NMDAR caused a decrease in accuracy and an increase of the repeating bias, but critically did not affect the forgetting rate, i.e. the decay of accuracy with delay. We recapitulate these findings using a hidden Markov Model that switches between (1) a WM state describing a stimulus-based strategy that requires memory maintenance, and (2) a history-based state (HB) which elicits lapse responses determined by previous choices. A shift in the transition probabilities towards the HB-state reproduces the effect of NMDAR blockade. Electrophysiological recordings in the anterolateral motor cortex (ALM) supported our model by showing that the encoding of the stimulus differs between inferred HB-state and WM-state trials: neurons showed similar encoding during the stimulus presentation or during the licking response, but only in the WM- state the encoding of the upcoming choice persisted during the delay period. Preliminary analysis of population activity fluctuations suggests that brain state varies along the sessions, with epochs of higher synchrony seemingly coinciding with choice epochs driven by the HB-module. Our results show that task performance is heavily limited by lapse epochs in which subjects stop using WM and suggest that deficits caused by NMDAR hypofunction could be related to an increased sensitivity to the cognitive effort associated with WM rather than by a decrease in its stability.

Forced synchronization of Neural Populations for Communication Through Coherence

Michael Orieux | Universitat Politènica de Catalunya

The Role of Excitatory-Inhibitory Homeostasis in the Recovery of Functional Connectivity after Focal Lesion – A Computational Account

Francisco Páscoa dos Santos | Eodyne Systems SL / Universitat Pompeu Fabra

Stroke-related disruptions in functional connectivity (FC) often spread beyond lesioned areas. Thus, it is unclear how the recovery of FC can be orchestrated on a global scale. Given that post-stroke recovery is accompanied by long-term increases in excitability, we propose excitatory-inhibitory homeostasis as a mechanism of recovery and aim to model its role in restoring FC properties, relating it to observed changes in excitability. In this study, we modeled stroke in connectome-based networks of Wilson-Cowan masses with homeostatic scaling of local inhibition. Activity was recorded pre-lesion (T0), post-lesion (T1) and post-lesion after stabilization of inhibitory weights (T2). Effects were quantified through changes in FC and local inhibitory weights were used as a proxy for excitability. Although FC significantly departed from baseline values at T1, its similarity with baseline was significantly improved from T1 to T2. Interestingly, post-lesion FC dynamics could not be recovered to baseline levels through EI homeostasis. We further observed long-term increases in local excitability, correlated with structural connectivity to lesioned areas, with severe lesions requiring larger and more widespread changes. Importantly, predicted long-term increases in excitability in particular areas (e.g. temporal cortex) may be tied to the emergence of late-onset side-effects of stroke such as epilepsy. Thus, our model not only showed an ability to recover FC towards pre-stroke levels through regulation of local excitability, but also replicated known long-term changes in excitability, possibly tied to the emerge of negative side-effects. Therefore, we present excitatory-inhibitory homeostasis as a key driver of stroke recovery, tying long-term changes in local excitability to the restoration of FC.

High metacognitive efficiency reverses the negative effect of confirmation bias in repeated perceptual decision-making tasks

Alexis Pérez-Bellido | Universitat de Barcelona

Humans are “suboptimal” decision makers. When presented multiple times with similar information, we have a natural tendency to repeat our previous choices despite of being mistaken. This repetition bias does not exclusively depend on decision-level biases. In fact, in line with what the confirmation bias hypothesizes (i.e. human tendency to discount evidence against one’s current position), it has been shown that in perceptual decision- making tasks, we integrate information asymmetrically by weighting more information that is consistent with our previous beliefs. In this study, we used reverse correlation to describe how choice dependent consistency biases are determined by participants metacognitive skills. We presented participants with a maximum of three repeated sequences of six differently oriented gratings. Participants categorized whether the mean orientation of each sequence was closer to the cardinal or the diagonal axis and reported their decisional confidence. Reverse correlation analyses helped to characterize 1) a stimulus independent bias to repeat previous choices from 2) a stimulus dependent bias to weight more information that is consistent with participant’s previous decisions (i.e. confirmation bias). Interestingly, we found that on average, participant’s with stronger confirmation bias improved less their performance after each repetition. However, those participants with high metacognitive skill reversed the negative impact of confirmation bias on repeated choices. Our results demonstrate that differences in consistency between previous decisions and sensory information induce “suboptimal” biases in evidence accumulation. However, confirmation bias might be adaptive when paired with high metacognitive efficiency.

How phase resetting curves influence excitation-inhibition-based synchronization

Pau Pomés  | UPF

Adaptive plasticity in the healthy reading network investigated through combined neurostimulation and neuroimaging 

Franz Schmid | UPF

The human reading network consists of various areas, among which the left inferior frontal cortex (IFC), the ventral occipito-temporal cortex (vOTC) and the dorsal temporo-parietal cortex (TPC) are the most notably ones (Pugh et al., 2001; Rueckl et al., 2015). Within this network the left TPC is crucially involved in phonological decoding, that is, it plays a vital role in both learning and retaining sound-letter mappings (Linkersdörfer et al., 2012). In the present study, the causal contribution of said region for reading was investigated with repetitive transcranial magnetic stimulation (rTMS) and functional magnetic resonance imaging (fMRI). After application of effective or sham TMS over the left TPC participants (N=28; healthy adult readers) overtly read both simple and complex words as well as matching pseudowords. On a behavioral level, effective TMS was found to slow down pseudoword reading. On a neural level, a shift in activity patterns in the left IFC for pseudoword reading was found by applying a multivariate pattern analysis for effective relative to sham TMS. Additionally, an increased effective connectivity from the left vOTC to the left TPC following active TMS was observed. The findings indicate that a TMS-induced disruption of the left TPC causes a compensatory reorganization in the reading network. Hence, the results provide evidence for the left TPC’s causal role for overt pseudoword reading and further highlight the important role of functional interactions in the unimpaired reading network for successful pseudoword processing.

Response biases in a visuospatial delayed response task

Balma Serrano Porcar | IDIBAPS

In decision-making tasks, animal responses must sometimes be maintained in short-term memory before their execution, a period during which they are presumably prepared but can also undergo alterations. NMDA receptors have been linked to such mnemonic maintenance. Aiming to better understand neural mechanisms of action selection, maintenance in short-term memory, and execution, we developed a visuospatial delayed response task in mice. Subjects were trained to look at a visual stimulus briefly displayed on a touchscreen, maintain its position during a short mnemonic delay, and execute a response by touching at the remembered position. We found that animals’ errors increased as a function of delay showing that, at least partially, mice made memory maintenance errors. Forgetting errors were idiosyncratically biased, suggesting that memory stability is not homogenous across possible responses but mice exhibit side preferences for one of the memorized choices which increase with delay.  Responses in memory trials were also biased by subjects’ tendency to repeat the previous choice, however, its magnitude did not vary with delay length.  This means that, in some trials, animals use their previous choices instead of a stimulus-guided strategy to resolve the task. Finally, we pharmacologically blocked NMDA receptors systemically and found that animals increased their repeating bias, although their memory stability and overall accuracy were not affected. Our results suggest that mice accuracy is limited by biases in the stability of the different responses and that NMDA receptors might impact how often responses are memory-guided rather than memory stability.

Relating local connectivity and global dynamics through low-rank structure in excitatory-inhibitory networks 

Yuxiu Shao | Ecole normale supérieure, Paris, LNC2

One of the key questions in neuroscience is how the cortical connectivity structure determines the collective dynamics of neural activity and thus controls behaviour. Two complementary approaches have been developed to address this question: (i) representing connectivity in terms of local statistics of excitatory-inhibitory motifs [Trousdale et al. 2012, Aljadeff et al. 2015] ; (ii) representing connectivity through a global low-rank structure that determines the low-dimensional dynamics [Mastrogiuseppe and Ostojic 2018, Beiran et al. 2021, Dubreuil et al. 2020]. It is however currently unclear how local connectivity statistics are related to the global structure and shape the low-dimensional activity. To bridge this gap, we map local E-I statistics onto global statistics of low-rank connectivity and examine the dynamics. We consider a randomly connected, block-like network composed of excitatory and inhibitory subpopulations. Connections in each block are specified by cell-type-dependent statistics and consist of independent and reciprocal parts. We first determine the dominant eigenvalues and eigenvectors of the connectivity matrix, and show that the statistics of their entries universally obey a mixture of Gaussian distribution. We then approximate random EI networks by Gaussian-mixture low-rank networks [Beiran et al. 2021] and show that the mean connectivity determines the dominant low-rank structure, which the reciprocal motifs further modify by modulating the dominant eigenvalue. Comparing the dynamics in the original E-I network and their low-rank approximations, we find that the firing rates of individual neurons closely match. Importantly, we show that two distinct sources of recurrent feedback contribute to the population dynamics: (i) feedback due to the mean component of connectivity; (ii) feedback due to reciprocal connections between neurons. Altogether, our analytical mapping of the local E-I statistics to low-rank description provides an intuitive picture of how local connectivity statistics determine global low-dimensional dynamics and resulting computations.

Impact of linguistic rhythm on brain activity in populations of different native language 

Silvana Silva Pereira | UPF

Speech rhythm constitutes an important part of language-based communication and is required by the processes underlying the construction of intelligible speech (Peelle et al., 2013, Bosker and Ghitza, 2018,K ̈osem et al., 2018, Poeppel & Assaneo, 2020). Whereas the ability to differentiate between languages of different rhythms is already present at birth, a skill we share with some animals (Ramus et al., 2000, Toro et al., 2005), at around five months of age humans start to discriminate between languages of the same rhythmic category as their native language (Bosh & Sebastian-Galles, 1997, Nazzi et al., 1998). Undoubtedly, the progressive acquisition of language must lead to a tuning of the auditory cortex with the consequence of a change in perception of the prosody, a bottom-up vs. top-down interaction. To investigate if natives of languages of different linguistic rhythm show differences at the level of the syllable (bottom-up approach), we performed an experiment were natives of Spanish and natives of English listened to resynthesized speech (avoiding the confounds of comprehension) from languages of different linguistic rhythm: Spanish, English and Japanese. Neural activity was measured with EEG and analyzed in the θ-frequency range, corresponding to the syllable level, where we expect that oscillatory activity accounts for by the periodicities present in the signal (bottom-up component). Indeed, phase-locking values to the stimuli show a similar pattern for participants in both groups, mimicking the regularity patterns observed in the three languages (Ozer et al., 2022). Time-frequency EEG data was then projected onto source space using a standard head model and the resulting 3D grid data was projected onto an AAL atlas for classification using machine-learning techniques. Whereas classification using data from the whole brain does not show a strong tendency, localized activity in the superior temporal gyrus (Yi et al., 2019) shows differences among conditions, particularly in the posterior part, supporting the results found at sensor space.

Neural Correlates of Binocular Rivalry

Melina Timplalexi | Universitat International de Catalunya

Binocular rivalry occurs in human and non-human primates when the two eyes are presented with conflicting images. Instead of generating a subjective experience of a merged image as might be expected, the image perceived alternates coherently between one image or the other. Electrophysiological and imaging studies have provided evidence that the image which is dominant during binocular rivalry is determined by both bottom-up mand top-down factors. Here we study binocular rivalry as a paradigm to investigate top-down regulation of the visual representations in the sensory cortex. Until now, binocular rivalry had never been investigated in rodents which, if they exhibited rivalry, would allow a more in mdepth understanding of the circuit mechanisms of this phenomena. In this study we develop an experimental approach to assess binocular rivalry in mice while functionally imaging and decoding the activity of the sensory cortex. We replicate previous findings regarding the matching of stimulus preference between the contralateral and the ipsilateral eye and we provide the first insights into how the binocular rodent visual cortex responds to and represents mismatches in sensory input to the two eyes.

Dynamics of interhemispheric prefrontal coordination in working memory

Melanie Tschiersch | IDIBAPS

Previously perceived working memory (WM) items have been shown to attract current WM reports. This so-called serial dependence relies on the interaction of active neural representations and long-lasting activity-silent mechanisms in prefrontal cortex (PFC). Furthermore, WM representations are more frequent for contralateral than ipsilateral memorized locations in PFC, and can transfer between hemispheres when midline-crossing saccades occur in the delay. This indicates the consistent specialization of each hemisphere for the corresponding visual hemifield in WM. However, serial dependence challenges this view as it is unclear how it can emerge when consecutive stimuli appear in different hemifields, which engage independent neural substrates. Here, we investigate the transfer of serial dependence between visual hemifields and the associated prefrontal correlates across hemispheres, in order to shed light on the mechanisms of integration of lateralized WM storage. We analyzed behavioral responses and population coding in neural data in relation to serial dependence in simultaneous multi-unit recordings in monkey bilateral PFC during an oculomotor visuospatial delayed response task. We found that serial dependence of stimuli presented across hemifields was diminished in comparison to trials within the same hemifield. However, when decoding target locations from neural activity, we found evidence for reactivations of activity associated with serial dependence in both hemispheres. To reconcile these findings, we looked at the correlations of the hemispheres during the working memory task. We found strong hemispheric correlations during the delay, but a decrease during reactivation time. Jointly, these results hint towards strong hemispheric interactions during working memory engagement but weaker interactions between tasks, leading to a disrupted continuity of serial dependence across hemifields.

Consistency or fluctuation? How humans manage limited search capacity over consecutive uncertain choices

Alice Vidal | UPF

Many everyday life situations involve making uncertain choices using limited resources (investing money, allocating attention in exams). In these cases, both choices and the resource allocation fluctuate over time. Yet, imposing variability on choice behaviour has been shown to have contradictory effects on performance (Cohen et al., 2007; Wyart &amp; Koechlin, 2016) and the sources of this variability remains up for debate (Beck et al., 2012; Findling et al., 2019) . Here, we manipulate environmental constrains to control for the amount of variability which maximises utility and investigate whether humans favour an equal distribution of resources among the choices (homogenous allocation), or fluctuate over time intentionally. We measure how these strategies impact optimality and propose cognitive processes driving them. We used a variation of the Breadth-Depth Apricot Task (Vidal et al., 2021) to test human behaviour (N=40) on sequential multiple choice scenarios manipulating the number of consecutive choices (block length), the average capacity available per choice in a block (capacity ratio), and the environment richness (success probability). At each choice (trial), participants had first to decide on the amount of resources (samples) to be spent from the capacity available, and their distribution amongst the alternatives. Once allocated, the sampling outcome would be revealed and one of the sampled alternatives had to be selected for the final choice. We compared the empirical results to a model of optimal resource allocation, which predicts a homogenous allocation (spending a number of samples equal to the ratio) except in the low-capacity ratio condition, and formalised an extended model identifying sources of sub-optimality. Overall, we found fluctuations in the resource allocation which were larger than optimal (mean±sd: 0.40±0.28, Wilcoxon test, ). We observed two strategies enabling these fluctuations: skipping a trial (no resources allocated) or spending a capacity inferior to the ratio. Additionally, we found that when available resources are scarce (low capacity), allocation fluctuates more from trial to trial, which corresponds to the optimal strategy. This leads to more skipped trials than in higher capacity ratio conditions (ANOVA, ), a strategy that allows more extensive search on the remaining trials. These fluctuations are amplified in longer blocks (Wilcoxon test, ), resulting in a significantly higher proportion of skipped trials than optimal, and suggesting that participants try to take advantage of the larger resources available to heighten exploration. Our results also indicate that these behavioural fluctuations owe to an intentional strategy, since the great majority (86%) of trials with less than the ratio or no capacity spent occur while capacity is still available, and they are followed by trials with high resource allocation. We developed an extension of the optimal model, and observed that, in addition to maximizing the expected reward, participants follow strategies that seek for informative sampling, favour exploration and are defined by a particular aversion for skipping. Finally, we found that, when sampling in poor environments, participants spread their resources across options (breadth), while in richer environments they focus on relatively fewer options (depth). These trade-offs between breadth and depth reflect close-to-optimal strategies and are not perturbed by fluctuations in resource allocation across trials. To conclude, this is the first time that choice behaviour variability is studied using an ecological paradigm combining many-alternative sequential decisions under limited resources. Results demonstrate that humans follow complex search strategies with fluctuations in resource allocation which overall strike a balance between the obtention of relevant information and the risk to leave some choices to chance. Albeit deviating from statistical optimality, these fluctuations have little impact on performance and prioritise anticipation and gathering of information, both beneficial behaviours in a constantly changing world.

State and context-dependent decision making in C. elegans

Maria Sol Vidal-Saez | UPF

The coupling between the brain, its body and environment is key to understanding behaviors that are strongly affected by the present and past contextual situations (context-dependent behaviors). We have chosen the model organism C. elegans to investigate these behaviors. This worm is ideal to explore in detail the link between neural connectivity and behavior, since its connectome, comprising only 302 neurons, is by far the most complete to date. Despite its simple nervous system, C.elegans displays complex behaviors, among which there are context-dependent decision making examples. We have focused on one of them, namely, chemotaxis in a gradient of NaCl. C. elegans, cultured in the presence of food and NaCl, will move up NaCl gradients. However, worms exposed to NaCl in the absence of food, avoid any NaCl concentration. We have developed a neural network model that explains the 2 opposite behaviors. The connections between neurons are taken from the known worm’s connectome. Also, we have used biologically-grounded models for the neurons, based on the available neural activity data. The model sheds some light on the basic principles of this adaptive behavior.

Comparing connectivity-based and power-based biomarkers of epileptogenic activity with
intracranial EEG

Manel Vila-Vidal | UPF / Universitat Politècnica de Catalunya

The only effective treatment of pharmacoresistant epileptic patients is nowadays surgery. Over the last decade, computational approaches have characterized the epileptogenicity of brain regions based on the spectral power of their recorded intracranial EEG (iEEG) signals to improve pre-surgical diagnosis. More recently, a few studies have resorted to functional connectivity measures with the aim to strengthen existing biomarkers. However, the interplay between power and connectivity has not been thoroughly studied in this setting. Specifically, we here aim to investigate the redundancy/synergy of a variety of connectivity measures with respect to signals’ power in identifying clinically validated seizure foci. For a cohort of 9 patients, we determined the level of redundancy/synergy of linear (e.g., Pearson) and non-linear (e.g., Phase-locking value) connectivity measures with respect to signal’s power in identifying the seizure focus during ictal epochs, exploring different physiological frequency bands. In each case, we evaluated redundancy/synergy via the increase of variance explained in a logistic regression model where a connectivity variable was added to the signals’ power predictor. The connectivity contribution to the seizure focus identification was rather heterogeneous across patients. In particular, for those patients with sufficiently large increased variance explained values (&gt;0.2), this increase was typically higher for non-linear than linear measures across frequency bands. Overall, the largest contribution of connectivity measures was found in the gamma band (30-70Hz). Taken together, these results suggest that integrating frequency-limited connectivity measures into power-based biomarkers might improve seizure focus identification for a substantial proportion of patients.

Dynamic sensitivity analysis: Assessing brain state transitions via whole brain modelling

Jakub Vohryzek | UPF

Traditionally, in resting-state functional MRI, a model-free analysis is used to find significant differences between clinical populations via detection theory or more colloquially p-value testing. Dependent on the a-priori assumptions about the underlying data, different fMRI features can be extracted for the statistical analysis such as regional activations, functional connectivity or fractional occupancy of state-based approaches (Jenkinson et al. 2012; Bullmore and Sporns 2009; Preti, Bolton, and Ville 2016) . Alternatively, model-based techniques infer features from the data and compare significance from model parameters (Baker et al. 2014; Deco and Kringelbach 2014) . However, to assess how brain states transitions between each other remains a challenge in the current paradigms. Here, I argue for dynamic sensitivity analysis that quantifies transitions between various brain states. In practice, this means building a whole-brain model to the spatio-temporal activity, and by stimulating brain regions, assessing the impact on the brain dynamics (Deco et al. 2019) . By doing so, it is possible to quantify the transitions between various brain states described by the Probability Metastable Substates. As a proof of concept, I apply this analysis in a clinical context to depressive patients treated with psilocybin.

Simultaneous disruption of integrative and broadcasting cortical circuits in disorders of consciousness 

Gorka Zamora-López | UPF

Traditionally, in resting-state functional MRI, a model-free analysis is used to find significant differences between clinical populations via detection theory or more colloquially p-value testing. Dependent on the a-priori assumptions about the underlying data, different fMRI features can be extracted for the statistical analysis such as regional activations, functional connectivity or fractional occupancy of state-based approaches (Jenkinson et al. 2012; Bullmore and Sporns 2009; Preti, Bolton, and Ville 2016) . Alternatively, model-based techniques infer features from the data and compare significance from model parameters (Baker et al. 2014; Deco and Kringelbach 2014) . However, to assess how brain states transitions between each other remains a challenge in the current paradigms. Here, I argue for dynamic sensitivity analysis that quantifies transitions between various brain states. In practice, this means building a whole-brain model to the spatio-temporal activity, and by stimulating brain regions, assessing the impact on the brain dynamics (Deco et al. 2019) . By doing so, it is possible to quantify the transitions between various brain states described by the Probability Metastable Substates. As a proof of concept, I apply this analysis in a clinical context to depressive patients treated with psilocybin.