Understanding the brain implies the discovery of computational principles of organization and function. I will argue that the search for such principles necessitates comparative experimental work, an appreciation of biology and evolution, and a belief that there may be few algorithmic solutions to the problems that brains evolved to solve. I will present our strategy to tackle problems such as cortical function, sleep dynamics and visual perception using unconventional experimental systems and a comparative perspective.
The abstracts of the conference presentations and working groups (plenary lectures, talks, posters, workshop working groups) are listed here. This page is continuously updated with the submitted abstracts.
Gilles Laurent, a French citizen, is born in 1960 in Casablanca, Morocco. After a PhD in Neuroethology from the University Paul Sabatier of Toulouse (France) and a Doctorate in Veterinary Medicine in 1985, he works as a postdoc and a Locke Research Fellow of the Royal Society at the University of Cambridge/UK. In 1990, he joins the faculty of the Biology Division at the California Institute of Technology (Pasadena, CA, USA), where he becomes Lawrence A. Hanson Professor of Biology in the field of "Computation and Neural Systems” in 2002. In 2008, Gilles Laurent was appointed Director of the Department of Neural Systems and Coding at the Max Planck Institute for Brain Research in Frankfurt am Main (Germany).
His interests are in experimental and computational neuroscience, with an emphasis on coding and circuit computation in the olfactory and visual systems.
Keynote lecture
Jean-Pierre Changeux is a French neuroscientist known for his research in several fields of biology, from the structure and function of proteins, to the early development of the nervous system up to cognitive functions.
He is involved in the Human Brain Project, especially in the Ethics and Society sub-project.
He entered the École Normale Supérieure (ENS) where he obtained a Master's degree. He pursued PhD studies at the Institut Pasteur under the direction of Jacques Monod and Francois Jacob, and gained his doctorate. Changeux then left France for postdoctoral studies first at the University of California Berkeley then at Columbia University College of Physicians and Surgeons, New-York. He returned to France as attaché to the chair of Molecular Biology held by Jacques Monod. In 1972, he became director of the Unit of Molecular Neurobiology at the Institut Pasteur, where he received a professorship. In 1975, Changeux was elected professor at the Collège de France, chair of Cell Communications, position that he held until 2006.
Professor Changeux is author of more than 600 scientific articles and several books, technical or for general audience.
The Human Brain Project (HBP) is a 10-year European initiative launched in 2013 aimed at putting in place a cutting-edge, ICT-based scientific Research Infrastructure for brain research, cognitive neuroscience and brain-inspired computing. In March 2016 HBP announced the release of its 6 ICT Platforms to users outside the project. The Platforms are designed to help the scientific community to accelerate progress in neuroscience, medicine, and computing. They consist of prototype hardware, software tools, databases, programming interfaces, and initial datasets, which will be refined and expanded on an on-going basis in close collaboration with end users.
In this session, Professor Changeux will introduce HBP and its platforms, what can make this new Research Infrastructure an opportunity for the scientific community, and propose an insight into the HBP’s social, ethical and philosophical implications.
Info session
Attachment | Size |
---|---|
CRCNS Conference 2016 - J.-P. Changeux.pdf | 3.94 MB |
Neuronal transmitters are released via the fusion of synaptic vesicles with the plasma membrane. Vesicles dock to the membrane via a protein complex termed SNARE, which contains membrane attached (t-SNARE) and vesicle attached (v-SNARE) proteins. The fusion occurs in response to a calcium inflow, and the vesicle protein Synaptotagmin (Syt) serves as a calcium sensor. A cytosolic protein Complexin (Cpx) interacts with the SNARE complex, restricting the spontaneous fusion. Although molecular interactions of these proteins have been extensively studied, it is still debated how the fusion proteins dynamically interact with each other and with lipid bilayers to trigger lipid merging. To elucidate this mechanism, we combined molecular dynamics (MD) simulations of Syt interacting with the SNARE complex, Cpx, and lipid bilayers and genetic approaches in Drosophila. Basing on MD simulations, we created a model of the molecular “fusion clamp” wherein Cpx dynamically interacts with v-SNARE, preventing full SNARE assembly. The model enabled us to predict new mutations in v-SNARE and Cpx that enhance Cpx ability to inhibit spontaneous fusion. To understand how Syt regulates Ca2+ dependent fusion, we employed MD simulations to investigate Syt conformational ensemble and its dependence on Ca2+ binding. Basing on the results of a genetic screen that revealed new loss-of-function mutations in Syt, we developed a model of the pre-fusion protein complex, including Syt interacting with the SNARE bundle, Cpx, and lipid bilayers. This model creates the basis for a systematic approach to manipulating the fusion machinery based on theoretical predictions derived from MD simulations.
Authors: Maria Bykhovskaia, A. Jagota, J. T. Littleton
Talk
Pyramidal neurons in the rodent hippocampus exhibit spatial tuning during spatial navigation, and they are reactivated in specific temporal order during sharp-wave ripples observed in quiet wakefulness or slow wave sleep. However, analyzing representations of sleep-associated hippocampal ensemble spike activity remains a great challenge. In contrast to wake, during sleep there is a complete absence of animal behavior, and the ensemble spike activity is sparse (low occurrence) and fragmental in time. To examine important issues encountered in sleep data analysis, we constructed synthetic sleep-like hippocampal spike data (short epochs, sparse and sporadic firing, compressed timescale) for detailed investigations. Based upon two Bayesian population-decoding methods (one receptive field-based, and the other not), we systematically investigated their representation power and detection reliability. Notably, the receptive-field-free decoding method was found to be well-tuned for hippocampal ensemble spike data in slow wave sleep (SWS), even in the absence of prior behavioral measure or ground truth. Our results showed that in addition to the sample length, bin size, and firing rate, number of active hippocampal pyramidal neurons are critical for reliable representation of the space as well as for detection of spatiotemporal reactivated patterns in SWS or quiet wakefulness. In contrast to standard paradigm (“meaning first, memory later”) for hippocampal memory, our new analysis paradigm allows us to investigate the hippocampal sleep memory first by decoding intrinsic structures of population codes, and then determine the meaning later (i.e., how those structures might correlate with subsequent or preceding wake behavior).
Author: Zhe (Sage) Chen
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - Z. (S.) Chen.pdf | 1.76 MB |
Introduction: Most studies employ functional magnetic resonance (fMRI) imaging to study brain networks in vivo, using the blood oxygen level dependent (BOLD) effect. However, the BOLD effect is rather unspecific and represents a mixture of cerebral blood flow (CBF), blood volume (CBV) and oxygenation (CMRO2) changes. Therefore, we explore the neurophysiological basis of functional brain connectivity by using simultaneous positron emission tomography and MR imaging (PET/MR).
Methods: We simultaneously measure BOLD-fMRI and PET in small animals. Analysis is performed using ROI- and independent component analysis techniques. Different PET-tracers such as the glucose metabolism tracer [18F]FDG, the perfusion marker [15O]H2O, the serotonin transporter tracer [11C]DASB and the D2 dopamine receptor antagonist [11C]Raclopride are used in this project (DLR-01GQ1415).
Results: When comparing [18F]FDG-PET with BOLD-fMRI, we could show significant differences (p<0.005) in the constitution of a default mode-like network structure in rats. The Euclidean distance between regions does not correlate with PET/MR connectivity measurements. [18F]FDG-PET connectivity information follows a small world characteristic, i.e. measured and randomized clustering coefficient are nearly equal (CM approx. equal CR), whereas the characteristic path length (Lambda M < Lambda R) in the measured metabolic networks is smaller compared to randomized networks.
Conclusion: Brain networks, based on metabolic information, are in some respects complementary to the information derived from BOLD-fMRI. The typical resting state networks appear to be constructed from a multitude of metabolic and receptor specific subsystems, which can specifically be assessed by combined PET/MR imaging. Connectivity based on metabolic information (Cometomics) has a high potential for basic research as well as clinical translation.
Authors: Hans F. Wehrl, André Thielcke, Suril Gohel, Bernd Pichler, Bharat Biswal
Talk
Sleep improves performance in memory tasks. Replay during sleep of spike sequences associated with a previously learned task is thought to be the mechanism of memory consolidation. Human and animal data revealed that augmenting sleep rhythms – spindles (sleep stage 2) or slow oscillation (stage 3) - by electrical, sensory or pharmacological stimulation leads to learning improvements. Here we used computational models developed based on human recordings to reveal mechanistic link between NREM sleep rhythms and spike sequence replay in memory consolidation. The model included effects of neuromodulators, allowing transitions between awake, stage 2 and stage 3 sleep, and synaptic plasticity. Performance was compared before and after a period of sleep following awake training.
We found a significant increase in performance in the model that experienced sleep slow wave activity compare to the model remained in awake state. Spike sequence analysis revealed that the learned sequences were reactivated spontaneously during Up states of slow oscillation. This led to synaptic reorganization and recall improvement after the sleep. When memory interference was introduced by training few competitive sequences, the slow oscillation only helped the strongest memory of the set. Including period of spindles before slow oscillation, as observed during normal sleep cycle, protected the weaker memory and improved performance for both memories. Our study predicts that spontaneous reactivation of the learned sequences during sleep spindles and slow waves represents a key mechanism of memory consolidation and the basic structure of sleep stages provides an optimal environment for consolidation of competing memories.
Authors: Maxim Bazhenov, Y. Wei, G. Krishnan
Talk
Our collaborative study focused on the analysis and modeling of spinal cord circuits that generate the rodent locomotor rhythm, including the central rhythm generators (RGs), the left-right coordination of locomotor activity, and the effects of afferent stimulation. Coordination of activity between left and right sides of the cord is provided by contralaterally projecting commissural interneurons (CINs). We have constructed and analyzed a spinal cord model consisting of left and right RGs interacting bilaterally via three neuronal pathways mediated by different CINs. The CIN populations incorporated in the models include the genetically identified inhibitory (V0D) and excitatory (V0V) subtypes of V0 CINs and excitatory V3 CINs. The model also includes the ipsilaterally projecting excitatory V2a interneurons mediating excitatory drive to the V0V CINs. The proposed network architectures and CIN connectivity allow the models to closely reproduce and suggest mechanistic explanations for speed-dependent contributions of V0D and V0V CINs and V2a interneurons to left–right alternation of neural activity, and switching between left–right alternating and left–right synchronous hopping-like patterns in mutants lacking specific neuron classes. We have also investigated and modeled the phase resetting and speed accelerating effects of dorsal root stimulation on the locomotor activity generated in the spinal cord and the roles of slow modulatory actions of these sensory pathways. The computational models provide insights into the architecture of the spinal networks and the organization of different CIN and afferent pathways controlling locomotor activity, and propose testable predictions about the organization and operation of mammalian locomotor circuits.
Authors: Ilya A. Rybak, R. M. Harris-Warrick, N. A. Shevtsova, S. Dietz, O. Kiehn
Talk
How does the human brain use neural activity to create and represent meanings of words, sentences and stories? Over the past decade we have studied this question by giving people text to read while we image their neural activity with fMRI and MEG. As a result, we have begun to understand answers to questions such as "What is the information flow in the human brain during the 400 msec required to comprehend a single word?" "How is the neural encoding of an adjective and subsequent noun composed into the encoding the the adjective-noun phrase?" "Given a sentence presented as a sequence of words, which word semantics does the brain activate neurally, at which points in time during sentence processing?", "Given a story presented as a sequence of words, which aspects of neural activity encode story context, versus the new word being read?" This poster or talk will both survey the main lessons we have learned to date, and present our newest CRCNS-supported results.
Authors: Tom Mitchell, M. Just, N. Rafidi, M. Toneva, D. Schwartz, O. Stretcu, D. Howarth, E. Laing
Talk
Individual differences in reward learning have elicited increasing interest in the recent years, in particular concerning addiction and craving associated with drug-taking. Animal studies have shown that the same training situation can result in widely differing patterns of conditioned responses. Some individuals, sign-trackers (ST) are attracted to the signal while others, goal-trackers (GT) are directly attracted by the reward. M. Khamassi and colleagues (Paris) developed a computational model which accounts for a large range of ST/GT results in terms of the balance between concurrent learning processes and their dependence on dopamine signaling in the striatum. This has led to a series of experimental predictions, which the current project is testing using rigorous protocols. In Bordeaux, we observed the emergence of ST and GT populations and tested the algorithmic nature of their responses. GT, but not ST responses, were found to be sensitive to the value and the precise identity of rewards in devaluation procedures, as expected from the computational model. Using an anatomically selective pharmacogenetic approach in TH-Cre transgenic rats, we will then examine the role of tonic dopamine levels in several regions of the striatum on ST versus GT learning and expression. In Baltimore, M. Roesch and colleagues assess the role of behavior and dopamine (DA) in determining sign- versus goal-tracker status through fast-cyclic voltammetric measurements of phasic DA signals and optogenetic manipulation of DA neurons. These data are expected to shed new light on the determinants of variations in conditioned behavior.
Authors: Alain R. Marchand, Mehdi Khamassi, E. Coutureau, M. R. Roesch
Poster session I
The cerebellum is implicated in a wide array of functions, most notably in the fine tuning of sensory-motor control. Numerous models have been developed to investigate the functional role of the cerebellar cortex (CC). Generally, the CC is considered as a three layer network where mossy fibres (MFs) and Purkinje cells (PCs) form the input and output layer, respectively, and granule cells (GCs) constitute a hidden layer. Most modelling work investigates how learning of synaptic weights between GCs and PCs determines CC’s computational capabilities. Much less attention is paid, however, to the computational role of the synapses between MFs and GCs.
Recent findings established that synaptic transmission between MFs and GCs is indeed highly diverse, exhibits different forms of synaptic short-term plasticity (STP) and is tuned to the pre-synaptic input modality (Chabrol et al. 2015). Using a simplified rate model, we demonstrate the computational capabilities conferred to the CC by diverse STP of MF-GC synapses. In general, when STP is present, changes in MF inputs elicit transient modulation of MF-GC synaptic strength which, in turn, produce large transient changes in PC firing rates, similar to in vivo findings. We show how PC firing rate transients can be used to perform three distinct computations: i) Detection of novelty in MF input patterns. ii) Distinction of ’meaningful’ MF signals over noise. iii) Learning of a precisely timed signal.
Our simulations highlight novel computational advantages of dynamics synapses in the cerebellum and elucidate in particular the role of their broad diversity.
Authors: Alessandro Barri, M. T. Wiechert, D. DiGregorio
Poster session I
Olfactory perception and behavior relies on the ability of olfactory neural circuits to generate stable representations of odor identity from olfactory cues that fluctuate over a large range of odor concentrations. Odor percepts are thought to emerge in piriform cortex, yet how information about odor identity is encoded in piriform neural networks remains poorly understood. To explore fundamental principles of cortical odor coding we have recorded, using in vivo two-photon calcium imaging of odor-evoked activity from large ensembles of piriform neurons in anesthetized mice. Population coding analyses reveal that odorant identity - at a given odorant concentration - can accurately be decoded from spatially distributed ensembles of piriform neurons. However, piriform response patterns change substantially over a 100-fold change in odor concentration, apparently degrading information about odor identity.
We now identify a concentration-invariant subnetwork of piriform neurons, which encodes odor identity independent of intensity. Our data suggest that distinct perceptual features of odors are encoded in distinct subpopulations of piriform neurons. In independent experiments we have identified genes selectively expressed in piriform neurons with distinct neural cell type identity and connectivity. Together, our results establish novel experimental and computational tools to decipher the contributions of piriform neural circuit components to the encoding of distinct stimulus features in olfactory cortex.
Author: Alexander Fleischmann
Poster session I
Compared to sensory neocortex, the anterior piriform cortex (APC) lacks a topological representation of odor identity. This raises the question, “Does space play a role in cortical odor processing?” We investigated the rostrocaudal distribution of neural activity in APC following exploration of a novel environment. We found a strong rostral bias in the density of active neurons that was absent in homecage mice. Since pyramidal cells (PC) receive uniform excitation along the RC-axis, but receive stronger inhibition from caudal APC, we propose that inhibitory circuits support spatial patterning of neural activity. Consistent with this hypothesis, we found that L3 PCs have the strongest rostral bias in activity and received the most caudally biased inhibition. However, surprisingly, L3 interneurons received stronger inhibition from rostral APC. We investigated the circuit mechanisms underlying these opposing inhibitory asymmetries. Interneurons that express Calbindin and/or parvalbumin increase in density along the RC-axis, providing a simple direct mechanism for stronger caudal inhibition. In contrast, somatostatin (SST) interneurons decrease in density along the RC axis. Further, selective optogenetic activation SST cells yielded rostrally-biased inhibition onto interneurons. Thus, rostral PCs are potentially disinhibited through SST->Interneuron->PC circuits. Taken together, our findings suggest that two synergistic circuit mechanisms increase neural activity in rostral APC during novel exploration. Interestingly, the former appears “hard-wired” but the latter may be modulated in a context dependent manner producing different activity profiles for novel exploration versus homecage. We are now implementing a theoretical framework to address the role of rostrocaudal asymmetries in odor coding.
Authors: Anne-Marie Oswald, A. Large, P. Schick, S. Mielo
Poster session I
The cortical networks that underlie behavior exhibit an orderly functional organization at local and global scales, an organization that is especially evident in the visual cortex of carnivores and primates. Here, the full range of stimulus orientations is represented in the activity of neighboring columns of neurons contained within a millimeter of cortical surface area, and these responses are shaped by a distributed network that shares similar functional properties and is arranged in an iterated fashion across several millimeters of the cortex. In this study, we use endogenous (spontaneous) cortical activity patterns to explore the fine scale functional architecture of this distributed network and how the coordinated local and global structure of the network arises during development. First, using in vivo imaging of calcium signals in the ferret, we show that in the mature visual cortex, the local structure of orientation columns is accurately predicted by correlations of spontaneous activity with neurons that lie at distances several millimeters away. Next, using chronic in vivo imaging techniques, we show that large scale modular correlation patterns, predictive of the mature organization, are evident at early stages of cortical development, prior to the formation of long range horizontal network connections. The early emergence of long-range modular correlations without long-range connections can be explained by local connections using a neural field model with connectivity consistent with observed local correlation structure. Our results suggest that local connections in early cortical circuits generate structured long-range network correlations that underlie the formation of distributed functional networks.
Authors: Bettina Hein, Matthias Kaschube, David Fitzpatrick, P. Huelsdunk, G. B. Smith, D. E. Whitney
Poster session I
Electrodes placed on the brain surface are predominantly used to study brain intrinsic activity with high spatio-temporal resolution, but have recently been utilized to inject electrical currents in order to modulate brain activity for surgical guidance, rehabilitation and brain-computer interfaces. However, it is only poorly understood how injected currents interact with brain functionality. Moreover, recent advances in manufacturing cortical arrays with high electrode density offer the possibility to simultaneously measure brain activity while injecting currents on a submillimeter scale (micro-electrocorticography, µECoG). Hence, computational modeling has become essential to predict and understand current flow pattern for targeting specific brain regions (ROI) using different µECoG electrode array designs. Our computational framework allows maximizing the current in the target ROI while current flow in non-target ROIs is restricted and safety-related features can be accounted for, e.g., avoiding local concentrations of electrical current density that may harm brain tissue.
Here we present how our multi-national (US-German) collaboration applies these models to validate cortical and subcortical indwelled µECoG arrays for a phantom, animals (ovine) and humans by utilizing computed patterns of low-amplitude currents delivered to specific brain ROIs. In vivo µECoG measurements of stimulation-evoked brain responses will help to guide us to assess the accuracy and efficacy of the computed current patterns for improved designs of cortical arrays.
Authors: C. Alexis Gkogkidis, M. Dannhauer, S. Guler, R. MacLeod, D. Brooks, J. Ojemann, T. Ball
Poster session I
Purpose: In the retina, visual signals from photoreceptors are pooled by bipolar cells, rectified, and pooled again to generate responses of retinal ganglion cells (RGCs). This process creates “subunits” in the RGC receptive field (RF) that mediate nonlinear responses to fine texture and movement. We present a novel method to expose subunit organization based on RGC recordings.
Methods: Light responses of parasol RGCs in isolated primate retina were obtained using multi-electrode recordings. Responses were fitted with a model in which subunit activity is produced by a linear combination of stimulus intensities, then exponentiated and summed to determine RGC firing rate. The likelihood of the model parameters was maximized by alternating between estimating subunit membership via soft-clustering the spike-triggered stimulus ensemble, and estimating the linear weights associated with each subunit. To test the functional relevance of subunits, a “null” stimulus was presented, orthogonal to the linear RF estimated from the spike-triggered average.
Results: In simulated data, this estimation approach accurately recovered subunits asymptotically, and yielded subunit estimates that were local aggregates of actual subunits in the limited data regime. In real data, subunits were spatially localized and non-overlapping, as expected from bipolar inputs. Null stimulation revealed the presence of spatial nonlinearities, some of which could be explained by the estimated subunits.
Conclusions: A novel technique for estimating subunits of primate RGCs shows promising results in recovering putative bipolar cell spatial structure and explaining response nonlinearities.
Authors: E.J. Chichilnisky, N. Shah, N. Brackbill, A. Tikidji-Hamburyan, C. Rhoades, G. Goetz, A. Sher, A. Litke, L. Paninski, E. Simoncelli
Poster session I
Multi-attribute decision making requires integration of attributes to generate a value estimate. This process could occur in parallel by integrating all attributes simultaneously, or sequentially by switching attentional focus between different attributes. To distinguish between these alternatives, we developed a novel gamble task to determine where attention is deployed during the decision process. The task will be used for both monkeys and humans. Subjects choose between gamble options, each characterized by two attributes: reward amount and probability. The amount and probability of a given option are each represented by the length of a line segment on a screen, and the two segments corresponding to the same option are visually connected on the screen. Using eyetracking, a line segment is only revealed when the subject directly looks at it, allowing us to observe which attribute the subject is attending at a given time. Subjects are free to make eye movements and to inspect each cue as long and as often as necessary. They indicate their final choice by a hand movement. Preliminary experiments in humans show that on most trials more than four fixations are made before the final choice. The number of fixations rises with the difficulty of the choice. In the first four fixations, the subjects showed strong idiosyncratic spatial biases that were absent in later fixations. Subjects showed two different search strategies, with one group prioritizing inspection of the different attributes of the same gamble option while the other group compared primarily the same attribute across gamble options.
Authors: Ernst Niebur, Veit Stuphorn
Poster session I
Identifying essential nodes in the brain and manipulating their activity is important to understand brain function and to treat neurological and psychiatric conditions. Selection of targets in most clinical cases is based, however, on a trial-and-error strategy since targeted manipulations are conducted in the absence of guiding theory. Here we use network theory to identify the core essential nodes in system-wide brain networks formed in an experimental model of learning and memory. We first find that long-term potentiation (LTP) of synapses in the hippocampus (HC) induces a functional reorganization of long-range connections into a network of networks (NoN) composed by the HC, prefrontal cortex (PFC) and nucleus accumbens (NAc). Network optimization theory predicts that NAc is the essential core node in this memory related NoN, despite being a low connectivity node. This theoretical prediction is experimentally confirmed by inhibition of a single predicted core node in NAc by a targeted pharmacogenetic intervention which, remarkably, completely eliminates LTP-induced functional reorganization of the whole NoN. Besides unveiling a fundamental role of the NAc in the control of HC-PFC interactions in the context of memory, this result suggests the applicability of network optimization theory to design intervention protocols in the brain.
Authors: Hernan Makse, L. Parra, S. Canals
Poster session I
Synaptic plasticity is a cardinal cellular mechanism for learning and memory. The endocannabinoid (eCB) system has emerged as a pivotal pathway for synaptic plasticity because of its widely characterized ability to depress synaptic transmission on short- and long-term scales. Recent reports indicate that eCBs also mediate potentiation of the synapse. However it is not known how eCB signaling may support bidirectionality. We combined electrophysiology experiments with mathematical modeling to question the mechanisms of eCB bidirectionality in spike-timing dependent plasticity (STDP) at corticostriatal synapses. Our results demonstrate that STDP outcome is controlled by eCB levels and dynamics: prolonged and moderate levels of eCB lead to eCB-mediated long-term depression (eCB-tLTD) while short and large eCB transients produce eCB-mediated long-term potentiation (eCB-tLTP). Moreover, we show that eCB-tLTD requires active calcineurin whereas eCB-tLTP necessitates the activity of presynaptic PKA. Therefore, just like glutamate or GABA, eCB form a bidirectional system to encode learning and memory.
Authors: Ilya Prokin, Y. Cui, H. Xu, B. Delord, S. Genet, L. Venance, H. Berry
Poster session I
We present a novel method for estimation of the fiber orientation distribution (FOD) function based on diffusion-weighted Magnetic Resonance Imaging (D-MRI) data. We formulate the problem of FOD estimation as a regression problem through spherical deconvolution and a sparse representation of the FOD by a spherical needlets basis that form a multi-resolution tight frame for spherical functions. This sparse representation allows us to estimate FOD by an $l_1$-penalized regression under a non-negativity constraint. The resulting convex optimization problem is solved efficiently by an alternating direction method of multipliers (ADMM) algorithm. The proposed method leads to a sharp feature-preserving reconstruction of the FODs. Through extensive experiments, we demonstrate the effectiveness and favorable performance of the proposed method compared with two existing methods. We also apply the proposed method to real 3T D-MRI data sets of healthy elderly individuals. The results show realistic descriptions of crossing fibers with less noise than competing methods even with a relatively small number of gradient directions.
Authors: Jie Peng, H. Yan, O. Carmichael, D. Paul
Poster session I
A hallmark feature of chronic pain is its ability to impact other sensory and affective experiences. It is notably associated with hypersensitivity at the site of tissue injury. It is less clear, however, if chronic pain can also induce a generalized site-nonspecific enhancement in the aversive response to nociceptive inputs. Here we showed that chronic pain in one limb in rats increased the aversive response to acute pain stimuli in the opposite limb, as assessed by conditioned place aversion. Interestingly, ensemble neural activities in the anterior cingulate cortex (ACC) correlated with noxious intensities. In addition, at the single neuron level, machine learning decoding based on ACC recordings accurately predicted pain intensities. Meanwhile, optogenetic modulation of ACC neurons showed bidirectional control of the aversive response to acute pain. Chronic pain, however, altered acute pain intensity representation in the ACC at both ensemble and single neuron levels. Furthermore, activation of ACC had similar effect as chronic pain on increasing the aversive response to acute noxious signals, whereas inactivation of ACC inhibited this pain-magnifying phenotype. Thus, chronic pain likely disrupts cortical circuitry to enhance the aversive experience in a generalized anatomically nonspecific manner.
Author: Jing Wang
Poster session I
The striatum is a major site of learning and memory formation for sensorimotor and cognitive association. One of the mechanisms underlying memory storage is synaptic plasticity - the long lasting, activity-dependent change in synaptic strength. Synaptic plasticity requires an elevation in intracellular calcium. A common hypothesis is that the amplitude and duration of calcium transients can determine the direction of synaptic plasticity. Calcium dependence of striatal plasticity if unclear, because it requires dopamine and because of the diversity in stimulation protocols. To test whether calcium can predict plasticity direction, we developed a calcium-based plasticity rule using a spiny projection neuron model with sophisticated calcium dynamics including calcium diffusion, buffering, and pump extrusion. We utilize three spike-timing-dependent plasticity (STDP) induction protocols, in which post-synaptic potentials are paired with precisely timed action potentials and the timing of such pairing determines whether potentiation or depression will occur. Results show that despite the variation in calcium dynamics, a single, calcium-based plasticity rule, which explicitly considers duration of calcium elevations, can explain the direction of synaptic weight change for all three STDP protocols. Additional simulations show that the plasticity rule correctly predicts the NMDA receptor dependence of long-term potentiation and the L-type channel dependence of long term depression. We further show that the same calcium plasticity rule can predict the outcome of several frequency dependent paradigms. By utilizing realistic calcium dynamics, the model reveals mechanisms controlling synaptic plasticity direction, and shows that the dynamics of calcium, not just calcium amplitude, are crucial for synaptic plasticity.
Authors: Joanna Jedrzejewska-Szmek, S. Damodaran, D. B. Dorman, K. T. Blackwell
Poster session I
High-cervical spinal cord injury can lead to respiratory deficiency due to paralysis of inspiratory muscles. Functional electrical stimulation has been applied to restore ventilatory function in individuals with respiratory deficiency as an alternative to mechanical ventilation. In our approach, stimulation is applied through intramuscular electrodes in the diaphragm causing contraction to increase lung volume. The developed respiratory pacing system requires 2 or 4 stimulation channels, not necessarily synchronized, with both open-loop and closed-loop capabilities. A multi-channel, multi-application electronic stimulation system has been developed, compliant with respiratory pacing experiments. The stimulator is built around a full-custom high voltage integrated circuit (IC). Each of the 8 available stimulation channels provides fully configurable biphasic current waveforms, with up to 1mA amplitude current (set as multiples of 4µA) and over-second pulse width (20µs steps). Different stimulation modes can be utilized, such as bursts mode and continuous stimulation. The stimulator can be used in standalone application or controlled by a wide range of devices, such as computers or FPGAs, through USB or UART. A user friendly GUI and a python based Application Programming Interface allows to fully control each parameter of the stimulator during the experiment (maximum delay 50µs). We present in-vivo experimental results in an uninjured rat model (n=2), using the stimulation system, in open-loop and closed-loop configurations. In the latter, a software-based adaptive controller modulates the amplitude of stimulation pulses delivered to the diaphragm with breath volume as feedback. Results show successful modulation of breath volume to a desired value via stimulation.
Authors: Jonathan Castelli, Sylvie Renaud, Yannick Bornat, F. Kölbl, R. Siu, G. N'Kaoua, A. Mangalore, B. Hillen, J. Abbas, N. Lewis, R. Jung
Poster session I
Parkinsonism is associated with a range of changes in neural activity in the basal ganglia, including enhanced bursting, oscillations, and correlations. While we know that these all stem in some way from depletion of the neurotransmitter dopamine, how the loss of dopamine induces these changes remains to be fully understood. I will discuss collaborative, interdisciplinary work exploring how the cellular architecture and network dynamics of an inhibitory loop in the basal ganglia, namely the pallidostriatal circuit, may yield exaggerated beta-band synchrony and locking to beta oscillations, specifically in the dopamine-depleted state, with possible implications for motor function. Our experiments reveal and characterize a specificity in the striatal targets of a pallidostriatal projection and suggest that the pallidal neurons involved exert a direct functional influence on motor capabilities under dopamine depletion. Our computational modeling predicts that the pallidostriatal pathway influences striatal output preferentially during periods of synchronized activity within globus pallidus (GPe) and that, under dopamine-depleted conditions, this effect becomes a key component of a positive feedback loop between the GPe and striatum that promotes synchronization and rhythmicity. Overall, our results generate novel predictions about the role of the pallidostriatal pathway in shaping basal ganglia activity in health and disease. The project involved in a collaboration with Aryn Gittis at Carnegie Mellon University and members of our groups, supported by a US National Science Foundation CRCNS award.
Authors: Jonathan Rubin, V. L. Corbit, T. C. Whalen, K. T. Zitelli, S. T. Crilly, A. H. Gittis
Poster session I
Increasing biological evidence suggest that topology and structure of cortical areas subtend perception and function. I will present new experimental data and a few theoretical thoughts on the functional organization of cat visual cortex, their possible organization principles and roles in perception. It is well-known that orientation preference shows a continuous organization with punctual singularities taking the form of pinwheels. Using high resolution optical imaging and specifically designed stimuli and data analysis protocols, I will exhibit that spatial frequency preference is also organized into a continuous map, except at punctual singularities, co-localized with the pinwheels, but organized as an electric dipole potential. From the mathematical viewpoint, I will show that pinwheels and dipole are the unique optimally parsimonious topologies allowing exhaustive representation of spatial frequencies and orientations. These topologies also allow tradeoffs in the perception of orientation and spatial frequency, but to achieve this balance, cells shall have sharper selectivity near the singularity, prediction that we validated experimentally. Shall the brain favor exhaustivity and parsimony, similar architectures may be found in the representation of other attributes. I will show preliminary data on the perception of movement and a new type of singularity satisfying exactly the same principles as pinwheels and dipoles, and will conclude based on these on a few hypotheses on how speed could be encoded in primary visual areas.
This talk combines past and current works developed with in particular with J. Ribot, A. Romagnoni, C. Milleret & D. Bennequin.
Authors: Jonathan Touboul, Alberto Romagnoni, J. Ribot, C. Milleret, D. Bennequin
Poster session I
Sleep and awake states feature different levels of neuromodulators, which contribute uniquely to the excitability of neuronal circuits, network firing patterns, and the plasticity of their synapses. We investigated changes in the log-normally distributed firing rates of hippocampal CA1 neurons between different sleep and awake states and within each state individually. Firing-rate changes within non-REM sleep, REM sleep, and state transitions from non-REM to REM favored higher-firing neurons, with either smaller increases or stronger decreases among lower-firing neurons. In contrast, transitions from REM to non-REM sleep reduced variability across the population, resulting in higher firing among lower-firing neurons and vice versa. These dynamics yielded net decreases in firing rates across sleep, with the largest decrease occurring in lower-firing cells, and net increases during waking, with median quantiles showing the greatest firing increases. We previously showed that firing decreases across sleep were predicted by the density of spindles and sharp-wave ripples in non-REM epochs and were integrated over epochs of REM sleep. Overall, these results demonstrate that sleep/wake states affect lower and higher-firing neurons differently, with non-REM sleep playing a normalizing role, and are consistent with competitive interactions that favor higher-firing neurons, and greater plasticity in median and lower-firing cells.
Authors: Kamran Diba, H. Miyawaki
Poster session I
A currently popular view of reinforcement learning is that the brain makes use of both model-based (goal-directed) and model-free (habitual) learning systems. I consider an alternative learning system, whose core is the successor representation (SR), which exhibits some of the operating characteristics of both model-based and model-free learning. The SR compactly encodes the manifold structure of the state transition function, and is closely related to spectral techniques for manifold learning. When applied to spatial environments, the SR learns a map that captures the underlying geometry of the state space. Such a map is consistent with the hippocampal representation of space: a variety of place field phenomena, such as changes in place fields induced by manipulations of environmental geometry and reward, arise naturally from the SR. Moreover, an eigendecomposition of the SR leads to a spatial representation resembling entorhinal grid cells. Recent behavioral experiments support the existence of an SR-based reinforcement learning system.
Authors: Kimberly L. Stachenfeld, M. Botvinick, S. Gershman
Poster session I
Muscle spindles are widely considered length and velocity encoders for our kinesthetic sense. They are a type of sensory organ that is particularly sensitive to stretch located within muscle, but their spike rates non-unique and history-dependent with respect to muscle length information, calling into question their ability to faithfully provide a sense of position to the central nervous system. Here, we hypothesize that muscle spindle spike rates are more closely related to force-related variables of the spindle-bearing muscle during stretch, due to the in-series connections of these sensors with their own muscle fibers. First, we show muscle spindle Ia afferents in cat fire in direct proportion to history-dependent passive muscle force and rate change in force across a range of stretch perturbations, with only a single set of parameters. We used supervised learning algorithms to find optimal summations of measured force and its first time-derivative that minimized the error between the model output and the measured spike rates. Second, we show that a similar force-based model, accounting for passive (i.e. non-contractile) tissues within the musculotendon, is capable of predicting history-dependent spike rates in anesthetized rats. Here, we included a model of passive tissues in the muscle and removed this estimated stiffness from the measured force before performing optimizations to predict spike rates. Our results suggest that multi-scale models of muscles and muscle spindles are necessary for understanding the underlying mechanisms of proprioceptive sensory encoding, and provide a mechanistic explanation for history-dependent phenomena observed at both the cellular and behavioral levels.
Authors: Kyle Blum, B. L. d'Incamps, Paul Nardelli, D. Zytnicki, T. Cope, L. Ting
Poster session I
Dopamine modulates striatal synaptic plasticity, a key substrate for action selection and procedural learning. Thus, characterizing the repertoire of activity-dependent plasticity in striatum and its dependence on dopamine is of crucial importance. While plasticity under prolonged activation is well elucidated, its expression in response to few spikes remains less documented. Using spike-timing dependent plasticity (STDP), a synaptic Hebbian learning rule which depends on the activity on either side of the synapse, we recently reported a new form of plasticity: a striatal spike-timing dependent potentiation (tLTP) induced by few coincident pre- and post-synaptic spikes (5–15), mediated by endocannabinoids (eCB-tLTP). Whether this eCB-tLTP interacts with the dopaminergic system remains to be investigated. We show that eCB-tLTP is impaired in a rodent model of Parkinson’s disease and is rescued by L-DOPA treatment. We found that opto-inhibition of dopaminergic neurons prevent eCB-tLTP induction and that dopamine type 2 (D2R) receptors located at corticostriatal glutamatergic afferents are required for eCB-tLTP expression. Finally, we provide a realistic mathematical model for the dynamics of the implicated signaling pathways. Combining our experimental results and modeling, we show that dopamine allow the emergence of a tLTP in response to few coincident pre- and post-synaptic spikes and control eCB-plasticity polarity (tLTP vs tLTD) via presynaptic D2Rs by modulating the effective eCB thresholds. While usually considered as depressing synaptic function, our results show that eCBs associated to dopamine constitute a versatile system underlying bidirectional plasticity implicated in basal ganglia physiopathology.
Authors: Laurent Venance, Ilya Prokin, Hughes Berry, H. Xu, S. Perez, B. Detraux, A. Cornil, Y. Cui, B. Degos, A. de Kerchove d’Exaerde
Poster session I
Epilepsy is one of the most common neurological syndromes, affecting an estimated 50 million people worldwide. In one-third of these patients, seizures cannot be controlled despite maximal medication management. It is increasingly recognized that epileptic seizures represent an interplay of neural network dynamics. The complexity of these networks interactions which define the epileptogenic cortex and drive seizure initiation and spread makes understanding and treating epilepsy a unique challenge. Utilizing invasive brain voltage recordings from patients with intractable epilepsy, we infer dynamic functional networks during spontaneous seizures based on several standard coupling statistics (correlation, coherence, phase locking value). We apply a recently developed graph theory approach to identify and track in time well-connected subsets of nodes (a.k.a., communities) in the inferred functional networks. We show in simulation - of both abstract and biophysically motivated models - that the network inference and community detection methods perform accurately. As part of this work, we have developed the network inference and analysis procedures within an online version control system, and we will make the repository publicly available, for reuse and further development by the community. The dynamic network analysis and statistical modeling of human seizure data could provide new approaches to improve patient care of medically refractory epilepsy. In our continuing work, through prospective and retrospective studies, we will apply these methods to identify principled surgical targets, and predict which patients will - and will not - benefit from surgery and to suggest alternative surgical and control strategies.
Authors: Louis-Emmanuel Martinet, E. Spencer, E. Kolaczyk, M. Kramer, S. Cash
Poster session I
Receptive fields in the visual system scale such that the width of the receptive field increases linearly with the distance from the center of the fovea. We analyze the theoretical problem of how to optimally distribute receptors to represent information in a world with an unknown scale. The optimal solution closely resembles the organization of the visual system and provides a natural account for the behavioral Weber-Fechner law, a foundational result in psychology. The scaling laws derived lead to several quantitative predictions that can be tested with existing technologies. The generality of these arguments suggests that the same quantitative relationships should hold for other sensory modalities as well as cognitive dimensions. These results suggest a common adaptive principle underlying the distribution of receptors along one-dimensional continua: The organization of the brain enables it to be equally prepared for any possible state of the world.
Authors: Marc Howard, K. Shankar
Poster session I
Neurotransmitter release at a fast chemical synapses is critical for communication of information between neurons in the brain, and depends on the spatio-temporal dynamics of calcium influx, its reaction with binding partners and free diffusion, all within the active zone, an area less than 0.05 µm2. Therefore, understanding of the geometrical relationship between the synaptic vesicles and the Ca2+ channels is critical for understanding the determinants of synaptic strength, time course, and plasticity. Because the spatial scale is in the nanometer range, direct experimental observation of the spatio-temporal dynamics driving synaptic vesicle fusion has not been possible. One strategy is therefore to use a computational approach to simulate the nonlinear dynamics of Ca2+ -triggered vesicle fusion. In our study we built a stochastic model of the presynaptic calcium channel gating, Ca2+ diffusion and binding, and synaptic vesicle fusion. To understand the influence of topography on synaptic diversity, we performed simulations designed to predict the different behavior of inhibitory and excitatory terminals within the cerebellar cortex. Our preliminary results suggest that inhibitory terminals use small clusters of Ca2+ channels to drive release from their periphery, as we described previously for the calyx of Held. While the excitatory terminals require a random placement of Ca2+ channels , as well as random placement of vesicles with an exclusion zone of > 30 nm. We therefore suggest that nanoscale distribution of Ca2+ channels and synaptic vesicles can differ between synapses and is a major factor influencing the diversity of function across synapse types.
Authors: Maria Reva, N. Rebola, D. DiGregorio
Poster session I
Few attempts have been made to model parsing with biological neural networks. The Neural Blackboard Architecture (NBA), proposed by Van der Velde and De Kamps (Velde 2006) is one of them. It was designed to answer many challenges in the neural modeling of sentence processing, including the ones detailed by Jackendoff (Jackendoff 2002). Here we expand on previous simulations of the Blackboard Architecture (Velde 2015) on leaky-integrage-and-fire (LIF) populations with population density techniques(de Kamps 2013) implemented in MIIND (Kamps 2008), to compare simulated time courses of neural activity associated to sentence parsing with functional magnetic resonance data (fMRI) and intracranial recordings (electro-corticography; ECOG).
Preliminary results suggest that the neural dynamics of the simulated circuit of LIF neural populations, without tuning of the circuit parameters, already approximate the qualitative behavior of neuroimaging measurements. In a study of Pallier et al, about manipulation of the size of constituents in sequences of words presented visually to participants, they observed a sublinear increase of the amplitude of hemodynamic responses in language related regions as a function of constituent size. We confirmed that simulating the same phrases under a phrase grammar theory with a simple bottom-up parsing scheme leads to the mentioned pattern. In the case of ECOG recordings, recent work from Nelson et al. (under review), provides evidence the the Local Field Potentials have increase with phrase constituent size and drop after binding of words into constituents. We confirm both qualitative properties from a preliminary simulation of right branched phrases of increasing number of words.
Authors: Martin Perez-Guevara, Christophe Pallier, M. de Kamps
Poster session I
Attachment | Size |
---|---|
CRCNS Conference 2016 - M. Perez-Guevara & C. Pallier.pdf | 1.84 MB |
Widening economic inequity is a key concern for our society, and previous cohort-based studies have also suggested a link between economic inequity and depression. However, little is known about the underlying neural mechanism of the link, due to substantial individual differences. Here, we demonstrate that functional magnetic resonance imaging activity patterns in the amygdala/hippocampus induced by the inequity during an economic game can predict both present and future (measured one year later) depression index (BDI-II). Such predictions were not possible by behavioral measures of the participants. These findings reveal the connection between the sensitivity of the amygdala/hippocampus to economic inequity, and the present and future depression index, and suggest an important involvement of the amygdala and hippocampus into the effect of economic inequity on human mental states.
Author: Masahiko Haruno
Poster session I
Perceptual decisions are thought to depend solely on characteristics of the sensory signal. In this view, the motor response is considered to be a neutral output channel that only reflects the upstream decision. However, in many daily situations, the action that follow our decisions can differ dramatically in terms of the required physical effort (or the motor cost). Indeed, previous literatures have shown that physical effort is used in motor planning, and influences behavioural decisions. However, it remains unclear whether the motor cost is simply integrated with the perceptual decision to optimise the expected reward, or whether the preceding experience of unequal motor costs can recursively influence the perceptual decision itself. Here, we show that manipulating the motor response cost for hand movements during a visual motion discrimination task can change the perceptual decisions, even if the decision is made with a different effector. When participants reported the direction of the visual motion by left or right manual reaching movement with different resistances, their reports were biased towards the direction associated with less effortful option. Importantly, repeated exposure to such resistance on hand during perceptual judgments also biased subsequent judgements using voice. Further analysis using drift diffusion model showed that the motor cost acts to increase the distance between the starting point of the evidence accumulation process and the decision bound for the costly decision. This demonstrates that the cost to act can influence our decisions beyond the context of the specific action, and changes how the sensory inputs are transformed into decisions.
Author: Nobuhiro Hagura
Poster session I
In a neural map, cells of the same subtype perform the same computation in different places of the sensory/visual field. How these different cells code together a complex visual scene is unclear. It is commonly assumed that they will code for a single feature to form a feature map, but this has rarely been observed directly. Using large-scale recordings in the retina, we show that a homogeneous population of fast OFF ganglion cells encode simultaneously two radically different features of a visual scene. Cells close to a moving object coded linearly for its position. More distant cells remained largely invariant to its position and responded non-linearly to speed changes. Cells switched from one computation to the other depending on the stimulus. We developed a model that predicts these results, and determined how it is implemented by the retinal network. Ganglion cells of a single type do not code for one, but two features simultaneously. This richer, flexible neural map might also be present in other sensory systems.
Authors: Olivier Marre, S. Deny, U. Ferrari, E. Mace, P. Yger, R. Caplette, S. Picaud, G. Tkacik
Poster session I
Extracellular field potentials are challenging to interpret due to many contributing sources. We aim at revealing the neural sources of the "neurophonic" that can be recorded in the nucleus laminaris (NL) in the brainstem of the barn owl. Putative generators of the neurophonic are the activity of afferent axons, the synaptic activation onto NL neurons, and spikes of NL neurons.
We recorded the neurophonic in response to binaural high-frequency tones (3-7 kHz), and we varied the interaural time difference (ITD). The mean activity of the monaural inputs to NL does not change with ITD whereas their relative phase does, leading to cancellation or summation of input signals. The activity of the binaurally-sensitive output of NL, i.e., firing rate of NL neurons, strongly depends on ITD. To isolate monaural and binaural contributions to the neurophonic, we analyzed its broad-band power spectrum (0.1-8 kHz).
Extracellularly recorded NL neurons' action potentials have most power in the frequency range 200-700 Hz. In the neurophonic, this low-frequency component (LFc) depended on ITD, but changes were small (below 5%). Thus, the LFc reflects the small contribution that NL neurons' action potentials add to the neurophonic.
Further ITD modulation of the neurophonic spectrum could be observed only at the stimulus frequency. The magnitude of this high-frequency component (HFc) and its ITD modulation were much larger than that of the LFc. Thus, the HFc is related to the inputs of NL and might originate predominantly from afferent axons, which generate considerable power only in this high-frequency range.
Authors: Paula Kuokkanen, Thomas McColgan, C. Carr, R. Kempter
Poster session I
Sounds in natural environments are complex mixtures from many different sources. The human auditory system performs the challenging task of reliably separating and grouping competing sounds in natural environments into percepts called "auditory streams". The present project combines human psychophysics with temporally and spatially high-resolution recordings (electrocorticography, ECoG) to characterize spatio-temporal neural features associated with human auditory perception. We recorded large-scale ECoG data from multiple auditory and auditory-related cortical areas during an auditory streaming paradigm, including stimuli that support auditory perceptual bistability. Auditory stimuli were sequences of pure tones presented in an ABA_ triplet repetition fashion, where we varied the frequencies of A and B tones by semitones. Triplets presented with 6 and 8 semitone differences were perceptually bistable stimuli. We characterized ECoG in the core auditory cortex in the time domain as averaged evoked potentials (AEPs), where responses were time-locked to the onset of each tone in the triplet. We then compared neural activity across all recording sites during the two auditory percepts using dynamic mode decomposition (DMD), uncovering differences in spatial DMD modes between the two percepts throughout cortex outside core auditory cortex. Our results suggest perceptual switching is correlated with neural activity across a network of cortical areas, and that it may be possible to describe these spontaneous switches using dynamic models derived from large-scale ECoG data.
Authors: Rodica Curtu, Bingni W. Brunton, X. Wang, K. V. Nourski
Poster session I
Hearing, vision, touch – underlying all of these senses is stimulus selectivity, a robust information processing operation in which cortical neurons respond more to some stimuli than to others. Previous models assume that these neurons receive the highest weighted input from an ensemble encoding the preferred stimulus, but dendrites enable other possibilities. Non-linear dendritic processing can produce stimulus selectivity based on the spatial distribution of synapses, even if the total preferred stimulus weight does not exceed that of non-preferred stimuli. Using a multi-subunit non-linear model, we demonstrate that selectivity can arise from the spatial distribution of synapses.Moreover, we show that this implementation of stimulus selectivity increases the neuron’s robustness to synaptic and dendritic failure. Contrary to an equivalent linear model, our model can maintain stimulus selectivity even when 50% of synapses fail or when more than 50% of dendrites fail. We then use a Layer 2/3 biophysical neuron model to show that our implementation is consistent with recent experimental observations, of a mixture of selectivities in dendrites, that can differ from the somatic selectivity, and of hyperpolarization broadening somatic tuning without affecting dendritic tuning. Our model predicts that an initially non-selective neuron can become selective when depolarized. In addition to motivating new experiments, the model’s increased robustness to synapses and dendrites loss provides a starting point for fault-resistant neuromorphic chip development.
Author: Romain Cazé
Poster session I
In natural vision, humans can perceive thousands of objects and actions, getting subjective impressions. Revealing how the brain represents these objective and subjective experiences tells us how the brain sees and structures the natural world. Here, I introduce some of our recent studies on visual representation in the human brain via predictive models of brain activity under natural vision. We recorded human brain activity evoked by natural movies using functional magnetic resonance imaging (fMRI). Using high-dimensional latent feature spaces derived from linguistics, we have developed voxel-wise encoding models and fit them to the movie-evoked activity in the human occipitotemporal cortex. Analyzing the fit models revealed semantic spaces that described the representations of objects, actions, and impressions in the human brain. Moreover, by solving an inverse problem, we have also succeeded in decoding objective and subjective experiences from the evoked brain activity.
Author: Shinji Nishimoto
Poster session I
The goal of our CRCNS research is to develop a stochastic frame work using stochastic dynamical operators (SDO) in order to quantify the inter-relations of neural dynamics and neural connectivity in motor control, and to develop predictive models suitable for neuroprostheses. Tests use real neural data collected in spinal frogs and simulated networks.The SDO theory supports investigation of highly nonlinear neural dynamics in a linear domain by using the system state probability distributions rather than dealing directly with actual state values. In addition, it may support deeper understanding of the stochastic nature of neural dynamics for motor control. We aim to apply this advanced technique to (1) predictive motor control and (2) neural dynamics and connectivity analysis. In predictive motor control (1), our aim is to provide a descriptive theoretical model for the dynamics of movement variables where we are able to predict movement and driving muscle signals and thereby control a robot simulating the motion using neural spike recordings. The SDO framework provides a tool to understand and describe the effect of neurons deep within the CNS on the movement behaviour, as well as those neurons closer to motoneurons. We are testing the framework with the Rybak group's spinal cord models and with real spinal data. In neural analysis (2), we aim to understand and quantify the effect of individual neurons in the dynamics of the spinal neural network and we investigate the capacity of the SDO framework to discover and estimate functional and neurophysiological connectivities in the network.
Authors: Simon Giszter, M. Abolfat-Beygi, T. D. Sanger
Poster session I
Large-field visual stimulation such as when looking out of a moving train, causes optokinetic response (OKR) eye movements that are controlled by a sensorimotor circuitry involving a multitude of cortical and subcortical areas. The visual-motion sensitive parietal areas MT and MST have been shown by lesion studies to be critically involved in the control of OKR eye movements. Neural recordings during OKR suggested that the firing rate of MSTd neurons directly encodes stimulus velocity, which might then be used as control signal by brainstem nuclei. Other studies, however, found visual-motion responses modulated by eye movement-related signals in the same area. In our project, we characterize neural tuning in MSTd in the awake monkey using an information-theoretic method accounting for neuronal latencies. We showed that the majority of MSTd neurons exhibit gain-field-like tuning functions rather than directly encoding a single variable. Neural responses revealed a large diversity of tuning to combinations of retinal and extraretinal variables such as retinal speed, eye velocity, and direction of motion. Our results support the idea that signals in MSTd provide the possibility to construct motor output from appropriate combinations of neural responses by subsequent brain areas. The distributed and implicit multimodal coding observed in area MSTd thus seems to provide basis functions for flexible readout rather than coding single variables as previously suggested.
The presentation is based on and extends the paper "Eye Velocity Gain Fields in MSTd During Optokinetic Stimulation" by L. Brostek, U. Büttner, M. J. Mustari, and S. Glasauer.
Authors: Stefan Glasauer, L. Brostek, U. Büttner, M. J. Mustari
Poster session I
The neuromuscular junction is a reliable synapse in which reliability derives from the summed activity of numerous unreliable elements, each consisting of a synaptic vesicle and associated voltage gated calcium channels (VGCCs). Lambert-Eaton myasthenic syndrome (LEMS) is an autoimmune disease that reduces reliability, leading to muscle weakness. This weakeness is due to an autoantibody-mediated removal of some of the VGCCs that are critical for transmitter release, an upregulation of other VGCC types, and a disruption in organization of these VGCCs. We have used a combination of electrophysiological recording and MCell computer modeling to examine structure-function relationships, the disease LEMS, and novel LEMS treatment strategies. We find that the organization of the transmitter release site (especially the number and distribution of VGCCs) is a critical determinant of physiological function. Further, we find that LEMS effects on physiology cannot simply be explained by a loss of VGCCs. Lastly, upon exploring LEMS treatment strategies, we have used MCell modeling to reveal the spatio-temporal dynamics of calcium ion flux into transmitter release sites during exposure to a potassium channel blocker (DAP), used alone, or in combination with a calcium channel gating modifier (GV-58). We find that broadening the presynaptic action potential (using predicted DAP effects on action potential shape) increases the effectiveness of GV-58 (modeled by adding drug bound states to our VGCC kinetic model). We are currently studying the dose-response details of the synergistic effect between these two drugs using our MCell model before moving to pre-clinical testing in animals.
Authors: Stephen Meriney, Christopher Meriney, Rozita Laghaei
Poster session I
The origins of extracellular field potentials in the brain are often unclear. Many modeling studies and interpretations of the EFP focus solely on extracellular potentials induced by synaptic currents on the dendrites and by spikes at the soma of a postsynaptic neuron.
Based on previous recordings of field potentials in the auditory brainstem, we here present a model of extracellular potentials from axon fiber bundles. The aim of the model is to show how a wide array of field potentials may be explained by the axons' anatomical features. In particular, we show that the branchings and terminations of axons in a typical projection area lead to a dipolar EFP structure. Dipoles have a farther spatial reach than the quadrupolar potentials traditionally associated with axons. The model predicts strong contributions of nerve tracts to the extracellular potential, which are currently neglected by other models.
Along with the theoretical description, our multichannel electrode recordings from the barn owl auditory brainstem show several features observed in the model. We recorded responses in the nucleus laminaris (NL) to tones, clicks, and white noise stimuli. The low-frequency (<1kHz) component of the EFP response to auditory click stimulation in NL shows the polarity reversal predicted by the model. The low frequency component has a dipolar structure and extends spatially for at least several millimeters from the nucleus.
Authors: Thomas McColgan, P. Kuokkanen, C. Carr, R. Kempter
Poster session I
Managing neuroscience data requires the integration of information from multiple sources. Information about stimuli, animals, protocols, and other metadata are necessary to interpret the resulting data correctly. Storing such information consistently is an essential part of experimental research and indispensable for reproducibility and re-use of data. The NIX project (www.g-node.org/nix) provides an open format and software tools to facilitate comprehensive annotation and efficient organization of neuroscience data. The format stores recorded or derived data together with the meta-information, including relationships between data items, regions of interest, events, etc. This enables fine-grained access and convenient selection of subsets of the data for analysis. Efficient use of NIX format features is facilitated by software libraries supporting common languages, including C++, Python, Matlab and Java, on all major platforms, which makes it easy to seamlessly integrate data managem ent in the lab data collection and analysis workflow.
Authors: Thomas Wachtler, C. J. Kellner, A. Stoewer, M. Sonntag, A. Koutsou, A. Sobolev, J. Grewe
Poster session I
Attachment | Size |
---|---|
CRCNS Conference 2016 - T. Wachtler (P).pdf | 1.37 MB |
In [Denker et al. 2011, Cereb. Cortex], it was shown by Unitary Event analysis [Grün 2002a, 2002b, Neural Comput.] that the occurence of simultaneous spikes is more strongly locked to the phase of the LFP beta-oscillations than the activity of the other neurons, which was related to the existence of cell assemblies. To study the influence of remote brain areas visible in the LFP on the correlated single neuron activity in a small cortical subnetwork, we examine a balanced network of homogeneously connected binary model neurons [Ginzburg et al. 1994, PRE] receiving input from a sinusoidal perturbation [Kühn, Helias 2016, arxiv]. This simple model serves us to capture the main properties and illustrate mechanisms that cause time-modulated correlations. The Glauber dynamics of the network is simulated and approximated by mean-field theory. Treating the periodic input in linear response theory, the cyclostationary first two moments are analytically computed, which agree with their simulated counterparts over a wide parameter range. The zero-time lag correlations consist of two terms, one due to the modulated susceptibility (via external input and recurrent feedback) and one due to the time-varying autocorrelations. For some parameters, this leads to resonant correlations and non-resonant mean activities. Our results can help to answer the salient question how oscillations in mesoscopic signals and spike correlations interact. An interesting extension of our model is to include cell assemblies [Litwin-Kumar et al., 2012, Nature Neur.] allowing a closer comparison to experimental findings. Supported by the Helmholtz foundation (VH-NG-1028, SMHB); EU Grant 604102 (HBP). Simulations with NEST (www.nest-simulator.org).
Authors: Tobias Kühn, M. Denker, P. Mana, S. Grün, M. Helias
Poster session I
Plasticity and memory are thought to occur following pathway-specific changes in synaptic strength that result from spatially and temporally coordinated intracellular signaling events. To better understand how cAMP and PKA dynamically operate in CA1 neurons, we used live two-photon imaging and genetically-encoded fluorescent biosensors to monitor cAMP levels or PKA activity in acute brain slices. Stimulation of Beta-adrenergic receptors (isoproterenol) or combined activation of adenylyl cyclase (forskolin) and inhibition of phosphodiesterase (IBMX) produced cAMP transients with greater amplitude and rapid on-rates in intermediate and distal dendrites compared to somata and proximal dendrites. In contrast, isoproterenol produced greater PKA activity in somata and proximal dendrites compared to intermediate and distal dendrites, and the on-rate of PKA activity did not differ between compartments. Computational models show that our observed compartmental difference in cAMP can be reproduced by a uniform distribution of PDE4 and a variable density of adenylyl cyclase that scales with compartment size to compensate for changes in surface to volume ratios. However, reproducing our observed compartmental difference in PKA activity required enrichment of protein phosphatase in small compartments; neither reduced PKA subunits nor increased PKA substrates were sufficient. Together, our imaging and computational results show that compartment diameter interacts with rate-limiting components like adenylyl cyclase, phosphodiesterase and protein phosphatase to shape the spatial and temporal components of cAMP and PKA signaling in CA1 neurons and suggests that small neuronal compartments are most sensitive to cAMP signals whereas large neuronal compartments accommodate a greater dynamic range in PKA activity.
Author: Vincent Luczak
Poster session I
Odor perception is the result of a complex cascade of events initiated by the interaction of odorants with olfactory receptors (ORs). ORs are G protein-coupled transmembrane receptors (GPCRs) expressed in our olfactory receptor neurons (ORNs). Each ORN expresses only one type of ORs. Thus, OR activation is equivalent, in principle, to the behavior of a stimulated ORN. Although the odor of an odorant is fully encoded within its molecular structure, understanding structure-odor relationships requires decoding the combinatorial activation code of multiple ORs. This represent a fundamental step and major challenge towards our comprehension of odor perception in the brain. Using molecular dynamics simulations, we investigate the molecular mechanism of OR activation in general, as well as as specific OR-odorant recognition in particular. The in silico tool, combined with in vitro and in vivo studies on OR/ORN behavior, is aimed to develop a 'computational nose' that can predict odor perception using merely the molecular structure of an odorant as input. The study will advance current understanding of odor perception, and also benefit GPCR-related pharmaceutical research.
Authors: Xiaojing Cong, Jérôme Golebiowski, H. Matsunami, M. Ma
Poster session I
Contemporary neuroscience is heavily data-driven, but today's data management technologies and sharing practices fall at least a decade behind software ecosystem counterparts. Distributed version control systems, such as Git, facilitate collaborative software development, and turnkey distributions, like NeuroDebian, free researchers from tedious and unreliable maintenance tasks. Based on git-annex (http://git-annex.branchable.com), which provides git-based framework for data logistics and versioning, DataLad project (http://datalad.org) aims to make a rich collection of disjoint neuroscience datasets available through a simple unified interface of a "data distribution", in order to facilitate large and small-scale collaborations. In this demo we will present accomplished milestones, and current state of the project development.
Authors: Yaroslav O. Halchenko, Michael Hanke
Demo
Pain is a common multi-dimensional experience, but pain mechanisms remain poorly understood. Current pain research mostly focuses on molecular and synaptic changes at the spinal and peripheral levels. Our team use multidisciplinary approaches (animal behavior, in vivo electrophysiology, optogenetics, and neural decoding) to dissect central pain circuits. Neuroimaging studies in humans have suggested that the primary somatosensory cortex (S1) is implicated in sensory-discriminative component of pain, whereas the anterior cingulate cortex (ACC) is involved in the affective-aversive component of pain. The ACC receives nociceptive inputs from the medial thalamus and other cortical regions. Individual ACC neurons can respond to noxious stimuli by increasing firing rates. While previous studies have demonstrated that ACC is necessary and sufficient for the acquisition of stable aversive learning inherent in the chronic pain condition, its role in the aversive response to transient acute pain is less well characterized. Here we use rats to address the specific role of ACC in acute thermal pain. Our results showed that (1) ACC population activities correlate with the intensity of noxious stimuli, and individual ACC neuronal activity can encode pain intensity; (2) We develop state-space methods to reliably detect the onset of acute pain signals based on population spike activity; (3) We use machine learning methods to discriminate different pain intensity using ACC ensemble spike activity; (4) Our optogenetic experiments indicate that ACC neurons can bidirectionally control the aversive response to acute pain. Our ultimate goal is to develop a rodent brain-machine interface system to achieve real-time pain modulation.
Authors: Zhe (Sage) Chen, Jing Wang
Poster session I
Synaptic plasticity has been demonstrated to be the main phenomenon underlying memory and learning. Spike-timing dependent plasticity (STDP) is a synaptic hebbian learning rule in which the change in synaptic efficacy depends on the paired activity on either sides of the synapse. Recent experiments have shown that the number of pairings involved in STDP is an important parameter that impacts on the form of plasticity which is expressed. In the dorsal striatum, it has been shown that a low numbers of pairing (~5-10) are able to induce an endocannabinoid-mediated spike-timing-dependent potentiation (eCB-tLTP) whereas at higher number of pairings (~100), anti-hebbian STDP constituted by an NMDAR-mediated tLTP and an eCB-tLTD are observed (Cui and al, eLife).
Here, we developped a phenomenological model of STDP based on the calcium-hypothesis (Graupner and al, PNAS). LTP and LTD are triggered when the calcium transients reach some defined thresholds. This stochastical model can be solved analytically, with a mean-field approach and using Ornstein-Uhlenbeck processes. The present model with its few parameters can predict a wide-range of STDP curves and their dependence on distinct parameters (numbers of pairings or temporal window of the pairings). Numerical simulations were also performed to study the effect of the pairing frequency in the LTP expression and the comparison with the experimental data. This model is constituted with few parameters and equations. Thus, the next step would be to implement them in neural networks, with different populations of neurons. This would help understanding the function and capabilities of STDP for in biological systems.
Authors: Gaetan Vignoud, J. Touboul, L. Venance
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - G. Vignoud.pdf | 5.47 MB |
Brain-computer interfaces (BCI) translate neural activity into movements of a computer cursor or robotic limb. BCIs are known for their ability to assist paralyzed patients. A lesser known, but increasingly important, use of BCIs is their ability to further our basic scientific understanding of brain function. In particular, BCIs are providing insights into the neural mechanisms underlying sensorimotor control that are currently difficult to obtain using limb movements. A key advantage of BCIs is that they simplify the output interface of the brain without (hopefully) simplifying away the complexities of brain processing that we seek to understand. We will demonstrate this advantage of BCI using two examples from our work. The first involves identifying a network-level explanation for why learning some tasks is easier than others. The second involves the use of internal models identified from neural population activity to explain why subjects make movement errors. These findings deepen our understanding of how neurons interact during learning, and suggest ways to accelerate learning of everyday skills.
Authors: Byron Yu, A. Batista, S. Chase
Talk
Contemporary neuroscience is heavily data-driven, but today's data management technologies and sharing practices fall at least a decade behind software ecosystem counterparts. Distributed version control systems, such as Git, facilitate collaborative software development, and turnkey distributions, like NeuroDebian, free researchers from tedious and unreliable maintenance tasks. Based on git-annex (http://git-annex.branchable.com), which provides git-based framework for data logistics and versioning, DataLad project (http://datalad.org) aims to make a rich collection of disjoint neuroscience datasets available through a simple unified interface of a "data distribution", in order to facilitate large and small-scale collaborations. In this talk we will present accomplished milestones, and current state of the project development.
Authors: Yaroslav O. Halchenko, M. Hanke
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - Y. O. Halchenko.pdf | 4 MB |
Volume and complexity of data acquired in neuroscience labs have increased dramatically over the past years. At the same time, scientific progress has become more and more dependent on collaborative efforts, exchange of data, and re-analysis of data. Ensuring that data acquired in the lab stays accessible and that it can be easily shared and re-used – i.e., understood – has become increasingly important, but also challenging. To help scientists deal with these challenges, we are developing methods and tools supporting efficient data management in the laboratory. Key elements are the collection and integration of metadata along with the data, and the consistent representation of data and metadata in open, machine-readable formats. Integrating annotation and organization of data in the laboratory data workflow has the immediate benefit of making data access and analysis efficient and reproducible, and at the same time eliminates the need of additional data preparation for data sharing or publication.
Author: Thomas Wachtler
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - T. Wachtler (T).pdf | 2.97 MB |
Professor Herz is the Chairman of the Bernstein Association for Computational Neuroscience and Coordinator of the Bernstein Center Munich, Germany.
Andreas V. M. Herz is professor at the Ludwig-Maximilians University Munich. His research interests are in computational neuroscience and cellular biophysics, the neural basis of spatial navigation, optimality and adaptation of sensory systems, acoustic communication, auditory processing and sexual selection, olfactory learning, interaction of cell-intrinsic rhythms and large-scale oscillations, and collective properties of neural network. In the past, Professor Herz has also worked in the fields of applied mathematics, statistical physics, game theory and theoretical immunology.
In this talk, Professor Herz will introduce the "Bernstein Network Computational Neuroscience“.
The Bernstein Network arose from a funding initiative of the German Federal Ministry of Education and Research (BMBF) that fosters the research discipline of Computational Neuroscience. By establishing regional centers and nation-wide interconnections, the network aims at establishing the scientific concept of Computational Neuroscience in a sustainable fashion in Germany. The combination of experimental and theoretical approaches, the integration of neuroinformatics as well as the training of young researchers are central structural principles. The innovative potential of the neurosciences is stimulated by links to biomedical and technological application fields.
Info session
Attachment | Size |
---|---|
CRCNS Conference 2016 - A. V. M. Herz (IS).pdf | 6.67 MB |
Dr. James W. Gnadt is Program Director at the NIH National Institute of Neurological Disorders and Stroke (NINDS) for funding related to sensory, sensorimotor and neuroendocrine functions in the brain, and quantitative experimental approaches.
He also serves as Team Lead for the Integrative and Quantitative Approaches Team of the NIH BRAIN Initiative.
Dr. Gnadt came to NINDS in 2008 from a career as a NIH-funded investigator in systems and quantitative neurophysiology since 1986 that spanned appointments at The Salk Institute, University of Alabama at Birmingham (UAB), Stony Brook, Howard and Georgetown Universities. Dr. Gnadt received his PhD in Physiology & Biophysics from UAB in 1985 and was an Alfred P. Sloan Research Fellow in 1993-95. He studied the neurological basis of cognitive, motor and sensory behavior by combining behavioral studies with in vivo & in vitro neurophysiology and systems control analyses. He was also co-inventor of an electrophysiological technique for using adaptive, spectral noise cancellation to allow real-time removal of microstimulation artifacts in neural recordings.
In this talk Dr. Gnadt will introduce the BRAIN Initiative.
The Brain Research through Advancing Innovative Neurotechnologies® (BRAIN) Initiative is part of a new Presidential focus aimed at revolutionizing our understanding of the human brain. By accelerating the development and application of innovative technologies, researchers will be able to produce a revolutionary new dynamic picture of the brain that, for the first time, shows how individual cells and complex neural circuits interact in both time and space. Long desired by researchers seeking new ways to treat, cure, and even prevent brain disorders, this picture will fill major gaps in our current knowledge and provide unprecedented opportunities for exploring exactly how the brain enables the human body to record, process, utilize, store, and retrieve vast quantities of information, all at the speed of thought.
Info session
Attachment | Size |
---|---|
CRCNS Conference 2016 - James W. Gnadt.pdf | 773.83 KB |
The cerebellum aids the learning and execution of fast coordinated movements, with acquired information being stored by plasticity of parallel fibre--Purkinje cell synapses. According to the current consensus, erroneously active parallel fibre synapses are depressed by complex spikes arising as climbing fibres signal movement errors. However, this theory cannot solve the credit assignment problem of using the limited information from a global movement evaluation to optimise behaviour by guiding the plasticity in numerous neurones. We identify the possible implementation of an algorithm solving this problem, whereby spontaneous complex spikes perturb ongoing movements, create an eligibility trace for plasticity and signal resulting error changes to guide plasticity. These error changes are extracted by adaptively cancelling the average error. This framework, stochastic gradient descent with estimated global errors, generates specific predictions for synaptic plasticity rules that contradict the current consensus. However, in vitro plasticity experiments under physiological conditions verified our predictions, highlighting the sensitivity of plasticity studies to unphysiological conditions. Using numerical and analytical approaches we demonstrate the convergence and estimate the capacity of learning in our implementation. Finally, a similar mechanism may operate during optimisation of action sequences by the basal ganglia, where dopamine could both initiate movements and signal rewards, analogously to the dual perturbation and correction role of the climbing fibre outlined here.
Authors: Boris Barbour, J.-P. Nadal, G. Bouvier, C. Clopath, C. Bimbard, N. Brunel, V. Hakim
Talk
The cortical networks that underlie behavior exhibit an orderly functional organization at local and global scales, an organization that is especially evident in the visual cortex of carnivores and primates. Here, the full range of stimulus orientations is represented in the activity of neighboring columns of neurons contained within a millimeter of cortical surface area, and these responses are shaped by a distributed network that shares similar functional properties and is arranged in an iterated fashion across several millimeters of the cortex. In this study, we use endogenous (spontaneous) cortical activity patterns to explore the fine scale functional architecture of this distributed network and how the coordinated local and global structure of the network arises during development. First, using in vivo imaging of calcium signals in the ferret, we show that in the mature visual cortex, the local structure of orientation columns is accurately predicted by correlations of spontaneous activity with neurons that lie at distances several millimeters away. Next, using chronic in vivo imaging techniques, we show that large scale modular correlation patterns, predictive of the mature organization, are evident at early stages of cortical development, prior to the formation of long range horizontal network connections. The early emergence of long-range modular correlations without long-range connections can be explained by local connections using a neural field model with connectivity consistent with observed local correlation structure. Our results suggest that local connections in early cortical circuits generate structured long-range network correlations that underlie the formation of distributed functional networks.
Authors: Matthias Kaschube, David Fitzpatrick, Bettina Hein, P. Huelsdunk, G. B. Smith, D. E. Whitney
Talk
To meet ongoing cognitive demands, the human brain must seamlessly transition from one brain state to another, in the process drawing on different cognitive systems. How does the brain's network of anatomical connections help facilitate such transitions? Which features of this network contribute to making one transition easy and another transition difficult? Here, we address these questions using network control theory. We calculate the optimal input signals to drive the brain to and from states dominated by different cognitive systems. The input signals allow us to assess the contributions made by different brain regions. We show that such contributions, which we measure as energy, are correlated with regions' weighted degrees. We also show that the network communicability, a measure of direct and indirect connectedness between brain regions, predicts the extent to which brain regions compensate when input to another region is suppressed. Finally, we identify optimal states in which the brain should start (and finish) in order to minimize transition energy. We show that the optimal target states display high activity in hub regions, implicating the brain's rich club. Furthermore, when rich club organization is destroyed, the energy cost associated with state transitions increases significantly, demonstrating that it is the richness of brain regions that makes them ideal targets.
Authors: Richard Betzel, S. Gu, J. D. Medaglia, F. Pasqualetti, D. S. Bassett
Talk
In this talk we report on the two following abstracts:
Abstract 1:
Humans can easily identify visual objects independent of view angles and lighting, individual words in speech independent of volume and pitch, and smells independent of their concentrations. The computational principles and underling neural mechanisms responsible for invariant object recognition remain mostly unknown. Olfaction, being one of the most genetically tractable, anatomically compact and very relevant to rodent behavior sensory system, presents a great opportunity to explore and model such mechanisms.
Here we propose a new computational principle of concentration invariant odor identification based on temporal ranking of receptor activation. We assume that only a small population of first activated receptor neurons is fully responsible for odor identification. To test this model, which we called ‘primacy coding’, we performed optogenetic masking experiments, and demonstrated the relevance of short temporal interval at the beginning of the sniff cycle for odor identification. We propose a computational model of how such code can be read by the cortex. Furthermore we derive implications of primacy coding for the evolution of olfactory receptor genes and bulbar-cortical connectivity.
Abstract 2:
Sensory systems are constantly facing the problem of computing the stimulus identity that is invariant wrt several features. In the olfactory system, for example, odorant percepts have to retain their identity despite substantial variations in concentration, timing, and background. This computation is necessary for us to be able to navigate in chemical gradients or within variable odorant plumes. How can the olfactory system robustly represent odorant identity despite variable stimulus intensity? We propose a novel strategy for the encoding of intensity-invariant stimulus identity that is based on representing relative rather than absolute values of the stimulus features. We propose that, once stimulus features are extracted at the lowest levels of the sensory system, the stimulus identity is inferred on the basis of their relative amplitudes. Because, in this scheme, stimulus identity depends on relative amplitudes of features, identity becomes invariant with respect to variations in intensity and monotonous non-linearities of neuronal responses. For example, in the olfactory system, stimulus identity can be represented by the identities of p strongest responding odorant receptor types out of 1000. We show that this information is sufficient to ensure the robust recovery of a sparse stimulus (odorant) via l1 norm or elastic net loss minimization. Such a minimization has to be performed under the constraints imposed by the relationships between stimulus features. We map this problem onto a dual problem of minimization of a functional of Lagrange multipliers. The dual problem, in turn, can be solved by a neural network whose Lyapunov function represents the dual Lagrangian. We thus propose that the networks in the piriform cortex computing odorant identity implement dual computations with the sparse activities of individual neurons representing the Lagrange multipliers. Our theory yields predictions for the structure of olfactory connectivity.
Authors: Dmitry Rinberg, Alexei Koulakov
Talk
The meaning, or semantic content, of language is represented in a network of cortical regions that includes much of the temporal, parietal, and frontal lobes. We mapped these representations in detail by having subjects listen to hours of natural stories while we recorded BOLD fMRI data, and then using regularized regression to build encoding models that predicted BOLD responses as a function of the semantic content of the stories. This revealed that the cortex is tiled with areas that represent different types of semantic information. In an earlier study we used similar techniques to map representations of the semantic content of visual scenes using fMRI data collected while subjects watched hours of silent natural movies. Combining this information with our results on language semantics, we found that the two representations are anatomical aligned along a boundary that surrounds visual cortex. For nearly any visual category that is represented on the posterior side of the boundary, we find an adjacent representation of the same linguistic category on the anterior side of the boundary. In many cases, voxels in a narrow region immediately on the boundary of visual cortex represent the same semantic category in both modalities. This suggests that there are shared organizational principles between high-level representations of language and vision.
Authors: Alexander G. Huth, S. F. Popham, N. Y. Bilenko, J. L. Gallant
Poster session II
Sensory systems are constantly facing the problem of computing the stimulus identity that is invariant wrt several features. In the olfactory system, for example, odorant percepts have to retain their identity despite substantial variations in concentration, timing, and background. This computation is necessary for us to be able to navigate in chemical gradients or within variable odorant plumes. How can the olfactory system robustly represent odorant identity despite variable stimulus intensity? We propose a novel strategy for the encoding of intensity-invariant stimulus identity that is based on representing relative rather than absolute values of the stimulus features. We propose that, once stimulus features are extracted at the lowest levels of the sensory system, the stimulus identity is inferred on the basis of their relative amplitudes. Because, in this scheme, stimulus identity depends on relative amplitudes of features, identity becomes invariant with respect to variations in intensity and monotonous non-linearities of neuronal responses. For example, in the olfactory system, stimulus identity can be represented by the identities of p strongest responding odorant receptor types out of 1000. We show that this information is sufficient to ensure the robust recovery of a sparse stimulus (odorant) via l1 norm or elastic net loss minimization. Such a minimization has to be performed under the constraints imposed by the relationships between stimulus features. We map this problem onto a dual problem of minimization of a functional of Lagrange multipliers. The dual problem, in turn, can be solved by a neural network whose Lyapunov function represents the dual Lagrangian. We thus propose that the networks in the piriform cortex computing odorant identity implement dual computations with the sparse activities of individual neurons representing the Lagrange multipliers. Our theory yields predictions for the structure of olfactory connectivity.
Author: Alexei Koulakov
Poster session II
Due to the structure of neural circuits, inhibitory connectivity typically surrounds excited regions with inhibitory activity, protecting the brain from runaway excitation (ictal activity) generated when a seizure forms. However, repeated waves of ictal activity can break down the surround inhibition, allowing a seizure to propagate. This is termed the break-down of the inhibitory surround. Our group is gearing up for a series of volumetric calcium imaging experiments to explore the initiation, propagation and termination of PTZ-initiated seizure activity in the larval zebrafish. Using light-sheet microscopy, we will quantitate seizure activity at neuronal resolution in the intact larval zebrafish central nervous system. In this talk/poster, I will present the current status of the project and discuss recent results in which we predict ictal events from inter- and pre-ictal activity, and perform blind categorization of inter-ictal and control segments extracted from confocal calcium imaging data.
Authors: Andrew Sornborger, F. Hsieh, J. Zheng, J. D. Lauderdale, S. Baraban
Poster session II
A problem of fundamental importance in neuroscience is to understand how properties of synaptic connectivity vary across behavior. A possible, but challenging, approach is to interpret fine (millisecond) timescale correlations among populations of spike trains, which can be recorded extracellularly, in such terms. There are many technical barriers to such an approach, including a paucity of ground truth data available for interpretative calibration. Here, we approach the question indirectly from the perspective of biophysical models. Working with simple models of monosynaptic transmission and varieties of background noise, we sought conditions for precise pre- and post-synaptic spike time relationships. A primary conclusion is that, in addition to connectivity parameters, low variability of subthreshold potentials appears to be an important ingredient. We describe nonparametric and semiparametric models for statistical inference of connectivity in this light, and discuss implications in an experimental context.
This is joint work with Jonathan Platkiewicz, Matthew Harrison, and Gyorgy Buzsaki.
Authors: Asohan Amarasingham, J. Platkiewicz, M. Harrison, G. Buzsaki
Poster session II
The circadian clock temporally coordinates most physiological processes in mammals. The master circadian clock resides in the bilateral suprachiasmatic nuclei (SCN) within the hypothalamus and consists of about 20000 neuronal oscillators. A transcriptional-translational feedback loop involving a collection of 'clock' genes generates rhythms in both gene expression and firing in individual neurons. Highly robust and precise rhythms are produced by synchronization of the neurons via inter-neuronal coupling (synapses and diffusible neuropeptides). The master clock adapts to the external day-night environment by entraining to light input via the retinohypothalamic tract. This creates an interesting dichotomy in the clock between being robust to irrelevant environmental perturbations and being responsive to changes in the environment, which is the focus of our work. We show using computational models that changes in the strength and timing of a single coupling agent can alter the balance between synchrony and entrainment, but not very finely.
However, the SCN is highly heterogeneous with strong and weak oscillators, multiple neuropeptides and neurotransmitters and spatially structured coupling. The precise topology of the network in which synchrony and entrainment emerge is also unknown. We propose that having two coupling agents with opposing actions synergistically allows the SCN to balance the two requirements of robust rhythms and flexible entrainment. We have identified these two coupling agents as VIP and GABA. Being able to manipulate this balance would allow faster recovery from jet-lag and shift-work and combat the deterioration of rhythms with age.
Authors: Bharath Ananthasubramaniam, C. Mazuski, E. Herzog, H. Herzel
Poster session II
The cerebellum aids the learning and execution of fast coordinated movements, with acquired information being stored by plasticity of parallel fibre--Purkinje cell synapses. According to the current consensus, erroneously active parallel fibre synapses are depressed by complex spikes arising as climbing fibres signal movement errors. However, this theory cannot solve the credit assignment problem of using the limited information from a global movement evaluation to optimise behaviour by guiding the plasticity in numerous neurones. We identify the possible implementation of an algorithm solving this problem, whereby spontaneous complex spikes perturb ongoing movements, create an eligibility trace for plasticity and signal resulting error changes to guide plasticity. These error changes are extracted by adaptively cancelling the average error. This framework, stochastic gradient descent with estimated global errors, generates specific predictions for synaptic plasticity rules that contradict the current consensus. However, in vitro plasticity experiments under physiological conditions verified our predictions, highlighting the sensitivity of plasticity studies to unphysiological conditions. Using numerical and analytical approaches we demonstrate the convergence and estimate the capacity of learning in our implementation. Finally, a similar mechanism may operate during optimisation of action sequences by the basal ganglia, where dopamine could both initiate movements and signal rewards, analogously to the dual perturbation and correction role of the climbing fibre outlined here.
Authors: Boris Barbour, J.-P. Nadal, G. Bouvier, C. Clopath, C. Bimbard, N. Brunel, V. Hakim
Poster session II
The suprachiasmatic nucleus (SCN) is the master mammalian pacemaker. The intra-SCN neuronal network maintains synchronized circadian rhythms through synaptic communication involving both fast neurotransmitters and neuropeptidergic signaling. Specifically, release of vasoactive intestinal polypeptide (VIP) from approximately 10-15% of SCN neurons, plays a key role in entraining and synchronizing circadian rhythms. While the effects of VIP are well characterized, little is known about the firing patterns that cause release of VIP. To characterize these firing patterns, we identified multi-day spontaneous VIP firing activity through ‘optical tagging’. To do so, we cultured SCN neurons from mice expressing channelrhodopsin2 only in VIP neurons on multielectrode arrays. Following 3 days of recording, we stimulated the culture and used the ChR2-mediated depolarization to retroactively identify the spontaneous firing patterns of VIP neurons. Our results indicate that VIP+ neurons exhibit characteristic circadian and instantaneous firing patterns compared to VIP- neurons. Specifically, many exhibit complex irregular firing patterns with high-frequency ‘burst-like’ firing. We next tested whether stimulation of VIP neurons with characteristic VIP firing patterns was sufficient to phase shift SCN rhythms in vitro and in vivo. Our results indicate that stimulating VIP neurons with ‘burst-like’ firing phase shifts PER2 rhythms in a cultured SCN slice and entrains circadian locomotor behavior. Interestingly, our results suggest that lower frequency tonic firing is not sufficient to phase shift PER2 rhythms and takes longer to entrain locomotor behavior. Overall, our results suggest that specific firing patterns modulate neurotransmitter release from VIP neurons with different consequences for the circadian system.
Authors: Cristina Mazuski, E. Herzog
Poster session II
Understanding the structure and dynamics of cortical connectivity is vital to understanding cortical function. Experimental data strongly suggest that local recurrent connectivity in the cortex is significantly non-random, exhibiting, among other features, particular patterns of synaptic turnover dynamics, including a heavy-tailed distribution of synaptic efficacies, a tendency for stronger synapses to be more stable over time, and a power-law-like distribution of synaptic lifetimes. While previous work has modeled some of these individual features of local cortical wiring, there is no model that begins to comprehensively account for all of them. We present a spiking network model of a rodent Layer 5 cortical slice which, via the interactions of a few simple biologically motivated intrinsic, synaptic, and structural plasticity mechanisms, qualitatively reproduces these features. In particular, we focus on the distribution of synaptic lifetimes in this model. We note that this distribution is power-law-like, and that its slope remains independent of numerous model parameters, being only strongly affected by the balance between depression and potentiation in STDP. These results suggest that this could provide a measure for the balance between these factors in experimentally observed data. As well, our model suggests that mechanisms of self-organization arising from a small number of plasticity rules provide a parsimonious explanation for numerous experimentally observed features of recurrent cortical wiring dynamics. Interestingly, similar mechanisms have been shown to endow recurrent networks with powerful learning abilities in various models, suggesting that these mechanisms are central to understanding both structure and function of cortical synaptic wiring.
Authors: Daniel Miner, J. Triesch
Poster session II
Neural oscillations serve many purposes in the human brain, including the integration and routing of information across cortical areas. Previous research has also demonstrated the behavioral relevance of pre- and poststimulus neural oscillations in various frequency bands, such as delta, alpha, beta, and gamma oscillations. Thus far, it is not well understood how prestimulus high gamma power (> 80 Hz) relates to behavioral outcomes. In this intracranial EEG study we investigated how pre- and poststimulus gamma and high-gamma power in the sensorimotor cortex affects response speed to visuotactile target stimuli. Three epilepsy patients with subdurally implanted electrode arrays were presented with brief unisensory visual, unisensory tactile, or bisensory visuotactile stimuli via LEDs and solenoid tappers mounted to a handheld foam cube. Alternating between blocks, individuals were instructed to respond either to visual or tactile inputs of unisensory or bisensory stimuli. For each individual, single-trial time-frequency resolved correlations between oscillatory power and response times were calculated using permutation tests with cluster-based correction for multiple comparisons. In all patients, increased prestimulus gamma power (about 60-140 Hz) in the sensorimotor cortex contralateral to the stimulated hand predicted longer reaction times to visuotactile stimuli when attention was directed to the visual modality. Prestimulus gamma power did not consistently correlate with the behavioral responses when paying attention to tactile stimuli. When analyzing the relationships between poststimulus power and response times a more heterogeneous pattern of positive and negative relationships across electrode sites emerged. Interestingly, the patterns of relationships between poststimulus power and behavioral responses were similar for the two attention conditions. We conclude that prestimulus gamma and high-gamma power in the sensorimotor cortex impedes responses to visual inputs of visuotactile target stimuli. Our study provides evidence for the behavioral relevance of prestimulus high gamma power in the processing of multisensory visuotactile stimuli.
Authors: Daniel Senkowski, M. Krebber, J. Keil, D. Friedman, P. Dugan, W. Doyle, O. Devinsky, E. Halgren, T. Thesen
Poster session II
A hallmark of the respiratory response to hypercapnia is the emergence of active expiratory pattern in the abdominal motor output (AbN). This pattern consists of late-expiratory (late-E) bursts attributed to the recruitment of expiratory neurons in the parafacial respiratory group (pFRG) by mechanisms not completely understood. It has been previously shown that the pFRG late-E activity is suppressed by pons transection and the consequent inhibition of post-inspiratory (post-I) neurons of the Bötzinger complex. Therefore, pontine areas that control post-I activity may be involved in the control of hypercapnia-induced late-E activity. Here, we explored the role of the Kölliker-Fuse (KF) nucleus in modulating late-E respiratory activity during hypercapnia. Based on model simulations, we proposed that the KF area provides excitatory drive to BötC post-I inhibitory neurons, which, in turn, tonically inhibit the late-E neurons of the pFRG. Our model predicted that during hypercapnia: 1) KF inhibition lower the recruitment threshold of late-E activity; 2) KF excitation prevents emergence of late-E activity. These modeling predictions were tested experimentally using the arterially-perfused in situ rat preparation. We found that: i) KF inhibition with isoguvacine (10mM) advanced by 1s the onset of hypercapnia-induced late-E bursts in AbN; and ii) disinhibition of KF with gabazine (100muM) greatly attenuated AbN late-E activity. The model suggests that KF-driven inhibitory inputs to the pFRG, possibly through the post-I BötC neurons, may determine the presence and/or onset timing of AbN late-E bursts during hypercapnia.
Authors: Daniel Zoccal, W. Barnett, A. P. Abdala, J. F. R. Paton, Y. Molkov
Poster session II
The synaptic weight, a central concept in numerous neuroscience studies, is still lacking a proper biophysical foundation at the level of macromolecular assemblies. Given their intrinsic morphological plasticity, dendritic spines are prime candidates to modulate synaptic weights. Understanding the structural dynamics of spines is thus key to understanding synaptic transmission and plasticity. So far, electron microscopy tomography (EMT) is the only imaging method that provides an isotropic resolution that is sufficient to study the ultrastructural anatomy of these ubiquitous cellular bodies. Applying an advanced topological segmentation-algorithm (Günther et al., Computer Graphics Forum 31, 2012) to analyze EMT image stacks, we were able to extract the entire actin cytoskeleton of individual spines from the mouse cerebellar and hippocampal formations. Here, we present results of this methodological pipeline, focusing on the cytoskeletal organization as morphological alterations of dendritic spines are always driven by changes of the underlying macromolecular structure that provides their mechanical stability.
Author: Dinu Patirniche
Poster session II
Humans can easily identify visual objects independent of view angles and lighting, individual words in speech independent of volume and pitch, and smells independent of their concentrations. The computational principles and underling neural mechanisms responsible for invariant object recognition remain mostly unknown. Olfaction, being one of the most genetically tractable, anatomically compact and very relevant to rodent behavior sensory system, presents a great opportunity to explore and model such mechanisms.
Here we propose a new computational principle of concentration invariant odor identification based on temporal ranking of receptor activation. We assume that only a small population of first activated receptor neurons is fully responsible for odor identification. To test this model, which we called ‘primacy coding’, we performed optogenetic masking experiments, and demonstrated the relevance of short temporal interval at the beginning of the sniff cycle for odor identification. We propose a computational model of how such code can be read by the cortex. Furthermore we derive implications of primacy coding for the evolution of olfactory receptor genes and bulbar-cortical connectivity.
Author: Dmitry Rinberg
Poster session II
Deciding in an uncertain and changing environment involves a balance between accumulating rewards through well-known actions (exploitation) and keeping track of less rewarding but potentially adaptive actions (exploration). It is likely that the balance between these two strategies is dependent on the uncertainty of the current situation: we might reasonably expect increased exploration in uncertain environments. This question is addressed in rats using a non-stationary 3-armed bandit task with two distinct levels of uncertainty. We find that uncertainty does in fact increase the probability of shifting choice after a rewarded trial (win-shift), an indicator of exploration. Additionally, this indicator slowly decreases within and between blocks of trials, demonstrating that rats dynamically modulate their tendency to explore. A computational meta-learning model which adequately replicates the behaviour of the animals is proposed, and points to an average squared error signal (estimating the variance in rewards of the current policy with regards to the environment) as a strong candidate for setting the exploration-exploitation trade-off. According to recent results, tonic levels of dopamine may be responsible for this control and to assess this hypothesis, we tested rats under flupenthixol, a dopamine D1 and D2 receptor antagonist, in the same task. Systemic flupenthixol dose-dependently increases the percentage of win-shift, irrespective of the level of uncertainty or of the degree of learning within a block. Modelling reduced sensitivity to the squared error signal in the computational model captures these changes, supporting the hypothesis that dopamine is the biological substrate of this signal.
Authors: François Cinotti, Mehdi Khamassi, B. Girard, A. Marchand, E. Coutureau
Poster session II
The project “Model-driven single-neuron studies of cortical remapping in the dorsal and ventral visual streams” aims at investigating the neural circuits linking vision, attention and oculomotor planning by a combination of neuro-computational and experimental neurophysiological studies. We here report the developed neuro-computational model of perisaccadic perception based on predictive remapping and corollary discharge signals. Our proposed model rests on the assumption that parietal areas such as LIP receives two different kinds of eye position information. A proprioceptive information about eye position and a preparatory corollary discharge about the intended saccade displacement. We demonstrate that this model can explain two recently discovered types of perisaccadic spatial attention shifts: (i) attention lingers after saccade at the (irrelevant) retinotopic position, that is, the focus of attention shifts with the eyes and updates not before the eyes land to its original position (Golomb et al., 2008, J Neurosci.; Golomb et al., 2010, J Vis.). (ii) shortly before saccade onset, spatial attention is remapped to a position opposite to the saccade direction, thus, anticipating the eye movement (Rolfs et al., 2011, Nat Neurosci.). We show that both observations are not contradictory and emerge through the model dynamics. The former is explained by the proprioceptive eye position signal and the latter by the corollary discharge signal. Interestingly, both eye-related signals are core ingredients of the model and are required to explain data from mislocalization and displacement detection experiments. Thus, our model provides a comprehensive framework to discuss multiple experimental observations that occur around saccades.
Authors: Fred Hamker, J. Bergelt, J. Mazer
Poster session II
Synaptic plasticity has been demonstrated to be the main phenomenon underlying memory and learning. Spike-timing dependent plasticity (STDP) is a synaptic hebbian learning rule in which the change in synaptic efficacy depends on the paired activity on either sides of the synapse. Recent experiments have shown that the number of pairings involved in STDP is an important parameter that impacts on the form of plasticity which is expressed. In the dorsal striatum, it has been shown that a low numbers of pairing (~5-10) are able to induce an endocannabinoid-mediated spike-timing-dependent potentiation (eCB-tLTP) whereas at higher number of pairings (~100), anti-hebbian STDP constituted by an NMDAR-mediated tLTP and an eCB-tLTD are observed (Cui and al, eLife).
Here, we developped a phenomenological model of STDP based on the calcium-hypothesis (Graupner and al, PNAS). LTP and LTD are triggered when the calcium transients reach some defined thresholds. This stochastical model can be solved analytically, with a mean-field approach and using Ornstein-Uhlenbeck processes. The present model with its few parameters can predict a wide-range of STDP curves and their dependence on distinct parameters (numbers of pairings or temporal window of the pairings). Numerical simulations were also performed to study the effect of the pairing frequency in the LTP expression and the comparison with the experimental data. This model is constituted with few parameters and equations. Thus, the next step would be to implement them in neural networks, with different populations of neurons. This would help understanding the function and capabilities of STDP for in biological systems.
Authors: Gaetan Vignoud, Jonathan Touboul, Laurent Venance
Poster session II
Dendritic spines are highly plastic and may partake in learning and memory storage processes governed at the ultra-structural level. In order to gain a quantitative understanding of the influence of the spine’s micro-scale architecture on the electro-chemical signals transferred to dendrites, we are developing a numerical framework to solve an electro-diffusion model described by the Poisson-Nernst-Planck (PNP) equations. Computational complexity is addressed by introducing a novel hybrid-dimensional discretization approach and developing specialized numerical methods based on Finite Volume discretization and geometric multigrid solvers. Ultra-structural reconstructions generated by our collaborators (A. Herz, D. Patirniche, M. Stemmler, LMU Munich, Germany) from 3D electron microscopy images (M. Ellisman, UCSD, USA) can be integrated into the numerical simulation framework in order to investigate the interplay between ultra-structural morphology on electro-chemical signals.
Authors: Gillian Queisser, Markus Breit, D. Patirniche, A. V. M. Herz
Poster session II
Neuronal arbors develop from growth processes regulated by complex molecular interactions. The convergence of extracellular, intracellular, and activity-dependent events on the cytoskeletal effectors, primarily actin filaments and microtubules, facilitates neural development as well as maintenance of mature morphology. Several effective techniques to digitally reconstruct and analyze neuronal morphologies exist, but the quantification of their structural dynamics remains challenging. Current descriptions of neuron morphology are static and do not contain precise representations of intracellular components. Recent advances in tissue labeling and imaging techniques necessitate the co-evolution of the standard SWC format for representing digital arbor tracings. Generation of time-varying reconstructions is required, co-registering subcellular information with neuronal morphology. Additionally, large numbers of augmented reconstructions are required to develop data-driven mechanistic models of neuronal development and of structural plasticity.
Here we present the definition of a novel multichannel file structure and corresponding Vaa3D plug-in to handle this new type of data. We also introduce a design to tag dynamic structural changes in a time-coded manner. Next, we illustrate ongoing progress in using the multichannel/time-lapse system on developing neurons in the Drosophila larva. Time-varying images of overall neuronal morphology along with fluorescently labeled subcellular cytoskeletal components are digitally traced into the aforementioned file structures. These new reconstructions enable complete statistical analysis of the structural changes and the underlying molecular processes. Lastly, we demonstrate how stochastic computational simulations of neuronal growth, statistically constrained by and validated against these novel reconstructions, can help select the most experimentally promising genetic alterations to gain additional biological insight.
Authors: Giorgio Ascoli, S. Nanda, H. Chen, R. Das, H. Peng, D. N. Cox
Poster session II
Introduction: Most studies employ functional magnetic resonance (fMRI) imaging to study brain networks in vivo, using the blood oxygen level dependent (BOLD) effect. However, the BOLD effect is rather unspecific and represents a mixture of cerebral blood flow (CBF), blood volume (CBV) and oxygenation (CMRO2) changes. Therefore, we explore the neurophysiological basis of functional brain connectivity by using simultaneous positron emission tomography and MR imaging (PET/MR).
Methods: We simultaneously measure BOLD-fMRI and PET in small animals. Analysis is performed using ROI- and independent component analysis techniques. Different PET-tracers such as the glucose metabolism tracer [18F]FDG, the perfusion marker [15O]H2O, the serotonin transporter tracer [11C]DASB and the D2 dopamine receptor antagonist [11C]Raclopride are used in this project (DLR-01GQ1415).
Results: When comparing [18F]FDG-PET with BOLD-fMRI, we could show significant differences (p<0.005) in the constitution of a default mode-like network structure in rats. The Euclidean distance between regions does not correlate with PET/MR connectivity measurements. [18F]FDG-PET connectivity information follows a small world characteristic, i.e. measured and randomized clustering coefficient are nearly equal (CM approx. equal CR), whereas the characteristic path length (Lambda M < Lambda R) in the measured metabolic networks is smaller compared to randomized networks.
Conclusion: Brain networks, based on metabolic information, are in some respects complementary to the information derived from BOLD-fMRI. The typical resting state networks appear to be constructed from a multitude of metabolic and receptor specific subsystems, which can specifically be assessed by combined PET/MR imaging. Connectivity based on metabolic information (Cometomics) has a high potential for basic research as well as clinical translation.
Authors: Hans F. Wehrl, André Thielcke, Suril Gohel, Bernd Pichler, Bharat Biswal
Poster session II
There is a growing interest in joint multi-subject fMRI analysis. The challenge of such analysis comes from inherent anatomical and functional variability across subjects. One approach to resolving this is a shared response factor model. This assumes a shared and time synchronized stimulus across subjects. The basic goal of this model is to jointly factorize the data matrices X into a product of the “mixing matrices” W and a “source matrix” S, where W is subject-specific and S is shared across subjects, so as to minimize some constrained or regularized objective function. Such a model can often identify shared information, but it may not be able to pinpoint with high resolution the spatial location of this information. In this work, we provide an effective searchlight based shared response model for locating shared information in small contiguous regions (searchlights) across the whole brain while keeping the quality of the found shared information. So our method can serve as a first step in multi-subject fMRI analysis to help identify regions that worth further investigation in a neuroscience experiment. Validation using classification tasks demonstrates that we can pinpoint informative local regions. In the framework of shared response models, we also report an interesting variant of ICA. Compared with the state-of-art methods, this method has improved and robust performance in our validation experiments while using fewer factors. This suggests that this method is capturing informative factors in a concise way.
Authors: Hejia Zhang, P. Chen, J. Chen, X. Zhu, J. S. Turek, T. L. Willke, U. Hasson, P. J. Ramadge
Poster session II
Centrifugal feedback projections from higher to lower brain areas are pervasive in the mammalian brain. They can provide the receiving brain area with information that is not available in the feedforward stream and thereby modify the information processing by the lower brain area in a task-dependent manner. These functions depend essentially on the network connectivity, but it is only poorly understood what controls its formation and specificity.
The rodent olfactory system is well suited to investigate the mechanisms controlling the evolution of network connectivity and functionalities arising from it. Even in adult animals the olfactory bulb, which directly receives the sensory information from the nose, is persistently rewired through the addition and controlled removal of granule cells (GCs), which comprise the bulb's dominant interneuron population, and through highly dynamic fluctuations of the dendritic spines connecting the inhibitory GCs with the bulb's principal neurons, mitral/tufted cells (MC/TC).
We developed computational models for the network evolution via spine fluctuations and adult neurogenesis of the GCs. Our modeling shows that Hebbian-type spine stability can explain the experimentally observed increase in spine stability with spine age. Our neurogenesis model focuses on the role of centrifugal projections from olfactory cortex onto GCs. The experimentally supported activity-dependence of GC survival leads to the formation of subnetworks that provide bidirectional connections between the bulbar and the cortical representations of learned stimuli. The resulting specificity of cortical inhibition of MC/TC via GCs allows, e.g., context-enhanced discrimination of occluded stimuli and cortical switching of bulbar processing.
Authors: Hermann Riecke, Kurt A. Sailor, Pierre-Marie Lledo, M. Wiechert
Poster session II
Psychophysical experiments on visual processing generally involve a pre-trial eye fixation period which occurs after additional randomly sized pauses between trials. Yet, in real life, the oculomotor system operates continuously. Although paced paradigms have already found impressively fast and accurate saccades towards faces occurring only 100ms after the image onset, the continuous detection of faces by the visual system remains largely unexplored. In a large study containing three experiments on 72 subjects total, we found evidence that “ultrafast” eye movements in humans can also be continuously launched towards faces starting at only 100ms after the presentation of a scene containing a small face. Subjects were able to rapidly saccade towards 4000 faces continuously at rates approaching 5 to 6 faces a second (which included eye movement times towards the target). Upright faces were found quicker and more accurately compared to inverted faces, both with and without a cluttered background. Hiding the faces by blending them into the background had little effect on detection rates or accuracy. In addition, we found limited evidence of fatigue in terms of saccade reaction time over the course of the experiment for upright faces, whereas inverted faces showed slower saccadic reaction times at the end of the experiment. Our study therefore provides evidence for advantages in continuous visual search for upright faces; which can presumably be effectuated by shortcuts in the visual hierarchy.
Authors: Jacob G. Martin, M. Riesenhuber, S. J. Thorpe
Poster session II
We present a new algorithm, functional connectivity hyperalignment, for building a common model of representational and connectivity spaces in the human cortex. Basing hyperalignment on functional connectivity makes it possible to hyperalign brains based on fMRI data obtained in the resting state as well as during viewing of a naturalistic movie. We show its application to both movie data and resting state data from the Human Connectome Project. Functional connectivity hyperalignment affords increases in intersubject correlation of representational geometry and between-subject multivariate pattern classification that approach that of time-series hyperalignment. Increases in intersubject correlation of functional connectivity vectors exceeds those obtained with time-series hyperalignment. Analysis of the spatial point spread function for connectivity vectors reveals a fine-grained structure with a spatial scale of a single voxel.
Authors: J. S. Guntipalli, James V. Haxby
Poster session II
During vocal communication, the auditory system performs fundamental tasks: perceiving and interpreting signals, such as spoken language, and guiding the production of one’s own vocalizations. Here we took advantage of the rich vocal repertoire of a songbird, the zebra finch, to investigate the neural computations that extract relevant information in vocalizations. We focused on the information related to the meaning of communication calls (intent, emotional status and vocalizer identity) and to the vocal gestures necessary for their production.
Using our unique large library of zebra finches vocalizations and a decoding approach of neural responses, we first demonstrate that individual auditory neurons encode information about meaning. Then, we decipher the acoustic code of zebra finch vocalizations and show compelling evidence that while spectral properties of the sound, determined by the filtering properties of the vocal tract, encode call types (intent, emotional status), the parameters related to the sound fundamental frequency, that are driven by the vocal organ (the syrinx), code the vocalizer ID within a particular call type. Using encoding and decoding models, we then find that neural population responses are tuned to the spectral parameters discriminative of call types. Finally, we obtain a motor description of the vocalizations, the motogram, which is a model of the syrinx contractions and vocal tract filtering properties. Using the motogram to synthetize new calls that independently explore the modulations due to vocal tract filtering and syrinx contraction dynamics, we show that neurons are tuned to acoustical features that correspond to distinct motor commands.
Authors: Julie E. Elie, Hedi A. Soula, F. E. Theunissen
Poster session II
Odors activate large populations of olfactory bulb mitral cells. These cells project to piriform cortex where they activate sparse and distributed cortical odor ensembles. We obtained simultaneous recordings from large populations of mitral cells and piriform cortex cells in awake, head-fixed mice to understand how the dense odor responses in bulb are transformed in cortex. Individual piriform cells could be activated or suppressed by different odors and a linear classifier could accurately decode odors from the population using single sniff-spike counts, indicating that these representations are reliable and robust. At the population level, odors evoked a sustained increase in net bulb output. By contrast, there was almost no net change in total cortical output, with a small, brief increase in spiking immediately followed by sustained suppression. We developed a spiking network olfactory bulb-piriform cortex model to reveal the origins of this transformation. In the model, odors activated distinct sets of glomeruli and associated mitral cells at specific phases throughout the sniff cycle. In piriform cortex, a small subset of pyramidal cells were activated by the earliest bulb inputs, which helped recruit other pyramidal cells that only received subthreshold bulb input through their recurrent collateral connections. However, this ramping cascade of cortical activity then recruits strong feedback inhibition that suppresses cortical spiking and discounts the impact of later bulb input. Thus, using a combination of experimental and computational approaches, we provide a mechanistic explanation for how the sparse cortical odor responses are largely defined by the earliest activated bulb inputs.
Author: Kevin M. Franks
Poster session II
Throughout adulthood the mouse olfactory bulb is repopulated with new granule cells (GC) neurons through a process termed adult neurogenesis. It is unknown whether this persistent remodeling can potentially drive synaptic plasticity, as the pre-existing circuitry needs to adapt to allow new neuron integration. In this study, we tracked the dendritic and spine development of adult-born GCs using 2-photon in vivo imaging. Overall, adult born GC dendritic structure stabilized at one month. In contrast, adult born GC distal apical dendritic spine dynamics plateaued at two months, but remained highly dynamic. We compared the spine dynamics of the adult born GCs with pre-existing, early-postnatal born GCs and found matching spine turnover. To test whether the GC spine dynamics correlated with synapse turnover of their synaptic partners, mitral/tufted cells (MC), we labeled mitral/tufted cells with fluorescently labeled gephryn, a GABA receptor structural protein. In vivo imaging of the gephryn puncta demonstrated dynamics matching GC spines. Using a computational model of GC-MC structural plasticity, we found that these dynamics enable the network to rapidly optimize its output in response to changes in the odor environment. Surprisingly, odor representations quickly settled into a steady state despite continued rapid remodeling of the circuit. GC to mitral/tufted cell synapses have so far been demonstrated to lack synaptic strength plasticity, therefore the olfactory bulb appears to be unique in the adult brain, where persistent structural remodeling plasticity may be its dominant form of plasticity.
Authors: Kurt A. Sailor, Hermann Riecke, Pierre-Marie Lledo, M. T. Valley, M. T. Wiechert, G. J. Sun, H. Song
Poster session II
We present an automated solution for detecting bad trials in magneto-/electroencephalography (M/EEG). Bad trials are commonly identified using peak-to-peak rejection thresholds that are set manually. This work proposes a solution to determine them automatically using cross-validation. We show that automatically selected rejection thresholds perform at par with manual thresholds, which can save hours of visual data inspection. We then use this automated approach to learn a sensor-specific rejection threshold.
Finally, we use this approach to remove trials with finer precision and/or partially repair them using interpolation. We illustrate the performance on three public datasets. The method clearly performs better than a competitive benchmark on a 19-subject Faces dataset.
Authors: Mainak Jas, D. Engemann, Y. Bekhti, A. Gramfort
Poster session II
Accurate estimation of neuronal receptive fields is essential for understanding sensory processing in the early visual system. Yet a full characterization of receptive fields is still incomplete, especially with regard to natural visual stimuli and in populations of cortical neurons. Recently, we have introduced a new method for estimating receptive fields simultaneously for a population of V1 neurons, using a model-based analysis incorporating knowledge of the feed-forward visual hierarchy (Antolik et al 2016). Unlike previous studies, the model assumes that a population of V1 neurons shares a common pool of thalamic inputs, and consists of two layers of simple and complex-like V1 neurons. When fit to two-photon recordings of a local population of mouse layer 2/3 V1 neurons, our model offers an accurate description of their response to natural images and significant improvement of prediction power over the current state-of-the-art methods. We have shown that the responses of a large local population of V1 neurons with locally diverse receptive fields can be described with surprisingly limited number of thalamic inputs. We are currently working to generalize this model to be applicable to multi-electrode extracellular recordings in cat V1 in conjunction with motion-cloud and natural image movie stimuli. Our long-term aim is to study contextual modulation in V1 by examining its projection onto biological-like architecture of our model.
This project is led by Margot Larroche, Jan Antolik, Yannick Passarelli, Luc Foubert, Jonathan Vacher, Yves Frégnac and Cyril Monier.
Authors: Margot Larroche, Jan Antolik, Y. Passarelli, L. Foubert, J. Vacher, Y. Frégnac, C. Monier
Poster session II
Neuronal transmitters are released via the fusion of synaptic vesicles with the plasma membrane. Vesicles dock to the membrane via a protein complex termed SNARE, which contains membrane attached (t-SNARE) and vesicle attached (v-SNARE) proteins. The fusion occurs in response to a calcium inflow, and the vesicle protein Synaptotagmin (Syt) serves as a calcium sensor. A cytosolic protein Complexin (Cpx) interacts with the SNARE complex, restricting the spontaneous fusion. Although molecular interactions of these proteins have been extensively studied, it is still debated how the fusion proteins dynamically interact with each other and with lipid bilayers to trigger lipid merging. To elucidate this mechanism, we combined molecular dynamics (MD) simulations of Syt interacting with the SNARE complex, Cpx, and lipid bilayers and genetic approaches in Drosophila. Basing on MD simulations, we created a model of the molecular “fusion clamp” wherein Cpx dynamically interacts with v-SNARE, preventing full SNARE assembly. The model enabled us to predict new mutations in v-SNARE and Cpx that enhance Cpx ability to inhibit spontaneous fusion. To understand how Syt regulates Ca2+ dependent fusion, we employed MD simulations to investigate Syt conformational ensemble and its dependence on Ca2+ binding. Basing on the results of a genetic screen that revealed new loss-of-function mutations in Syt, we developed a model of the pre-fusion protein complex, including Syt interacting with the SNARE bundle, Cpx, and lipid bilayers. This model creates the basis for a systematic approach to manipulating the fusion machinery based on theoretical predictions derived from MD simulations.
Authors: Maria Bykhovskaia, A. Jagota, J. T. Littleton
Poster session II
Rats navigate efficiently between rewarded sites in an open field. From trial to trial, rats explore both inefficient path segments as well as segments that contribute to the final path. During inter-trial pauses, recently traversed place fields reactivate in short sequences (“snippets”) in the hippocampus. We hypothesize that this replay exposes prefrontal cortical circuits to subsequences of the final sequence that should be generated. We test the hypothesis that by selecting efficient segments that include a reward, the model can learn to reconstruct the efficient final trajectory.
Our PFC reservoir model with feedback is capable of concatenating snippets that have place fields with sufficient spatial overlap, in order to generate the target sequence, even if it has never been seen in the training data. When snippets are generated from two sequences of different lengths, with more overlap between subsequences, the system selects and generates the shorter sequence. We then generated snippets from multiple sequences that were not the target sequence, but that contained components of the target sequence. Random snippets were drawn between random pairs of rewarded sites. Thus, trajectory components being a part of a rewarded path were favored, and the PFC model discovered the efficient sequence. Introduction of a Bayesian spatial filter that reconstitutes the output signal before feedback to the reservoir yields significant improvements.
These results indicate that (1) recombination capabilities of the cortical reservoir contribute to target sequence identification, and that (2) a spatial reconstitution filter also contribute to the generation of the target sequence.
Authors: Nicolas Cazin, Peter F. Dominey
Poster session II
Legged locomotion involves various gaits. It has been observed that fast running animals (e.g. cockroaches) employ the tripod gait (three legs lifted off the ground simultaneously) while slow walking animals (e.g. stick insects) use the tetrapod gait (only two legs lifted off the ground simultaneously). Fruit flies use both gaits over a typical speed range. Central pattern generators (CPGs) are networks of neurons in invertebrate thoracic ganglia, responsible for generating locomotive activities. In this work, we use bursting neuron models of CPGs to study the effect of stepping frequency on the transitions from tetrapod gaits at low speed to tripod gaits at higher speeds. To this end, we first derive 6-phase oscillator equations, each corresponding to a neural network controlling one leg. Then by assuming that the left legs maintain a half period phase difference with the right legs (bilateral symmetry), we reduce the 6 equations to 3 equations. Finally, we consider a dynamical system with 2 phase differences defined on a torus. We show that bifurcations occur from multiple stable tetrapod gaits to a unique stable tripod gait as speed increases. We discuss these results in relation to those of Yeldesbay, Tóth, and Daun on a model of stick insect gait transitions.
Authors: Z. Aminzare, V. Srivastava, Philip Holmes
Poster session II
Attachment | Size |
---|---|
CRCNS Conference 2016 - P. Holmes.pdf | 620.52 KB |
Understanding how assemblies of neurons encode information requires recording large populations of cells in the brain. In recent years, multi-electrode arrays and large silicon probes have been developed to record simultaneously from hundreds or thousands of electrodes packed with a high density. However, these new devices challenge the classical way to do spike sorting. Here we developed a new highly automated algorithm to extract spikes from extracellular data, and show that this algorithm reached near optimal performance both in vitro and in vivo. The algorithm is composed of two main steps:
- a "template-finding" phase to extract the cell templates, i.e. the pattern of activity evoked over many electrodes by the spikes of one neuron ;
- a "template-matching" phase where the templates were matched to the raw data to find the spike times.
The time spent on manual curation did not scale with the number of electrodes. We tested our algorithm with large-scale data from in vitro and in vivo recordings, up to 4225 electrodes, and performed simultaneous extracellular and patch recordings to obtain "ground truth" data, where the solution to the sorting problem is at least partially known. The performance of our algorithm was always close to the best expected performance. We thus provide a general solution to sort spikes from large-scale extracellular recordings.
Authors: Pierre Yger, G. L. B. Spampinato, E. Esposito, B. Lefebvre, S. Deny, C. Gardella, M. Stimberg, F. Jetter, G. Zeck, S. Picaud, J. Duebel, O. Marre
Poster session II
Cortical sensory responses are highly variable. This variability can be correlated across neurons (due to some combination of dense intracortical connectivity, cortical activity level, and cortical state), with fundamental implications for population coding. Yet the interpretation of correlated response variability (or “noise correlation”) has remained fraught with difficulty, in part because of the restriction to extracellular neuronal spike recordings. Here, we measured response variability and its correlation at the most microscopic level of electrical neural activity, the membrane potential, by obtaining dual whole-cell recordings from pairs of cortical pyramidal neurons during visual processing. We found that during visual stimulation, correlated variability adapts towards an intermediate level and that this correlation dynamic is mediated by intracortical mechanisms. A model network with external inputs, synaptic depression, and structure reproduced the observed dynamics of correlated variability. These results establish that intracortical adaptation self-organizes cortical circuits towards a balanced regime at which network coordination maintains an intermediate level.
Authors: Ralf Wessel, N. Wright, M. Hoseini
Poster session II
Functional electrical stimulation of respiratory muscles is a viable approach for ventilatory support following spinal cord injury (SCI). Current systems implement open-loop stimulation, which requires manual stimulation parameter tuning and cannot alter stimulation parameters to account for muscle fatigue. Our US-French collaborative team has designed and developed a novel computation-enabled adaptive ventilatory control system (CENAVEX) to address these limitations.
To facilitate control system development, a computational, biomechanical model of the respiratory system was developed. Using the model, we identified controller parameters that followed the respiratory waveform and allowed for rapid adaptation.
A controller that uses an adaptive Spiking Neural Network (SNN), inspired by the medullary respiratory network, has been designed and simulated. Breath volume input is used to synchronize stimulation with native breathing. The breathing frequency controller also dynamically evolves with a metabolic demand parameter.
A real-time processing hardware platform was developed to produce a digital implementation of the SNN and a custom IC-based stimulation chip which can supply the adapting current pulses required by the controller. The system has been validated and tested in vivo using both open-loop and closed-loop experiments. A closed-loop Pattern Shaper (PS) adaptive controller was developed to control breath volume by modulating charge delivery to control diaphragmatic contraction. Computational studies determined several sets of parameters which the controller could use to reduce cycle error below 5% by 20 cycles and maintain stability for at least 100 cycles. Studies on uninjured animals maintained an average of less than 10% error after an initial adaptation phase.
Authors: Ranu Jung, Sylvie Renaud, J. Abbas, Y. Bornat, A. Zbreski, J. Castelli, B. Hillen, R. Siu
Poster session II
What aspects of previous experiences guide decisions? Much research concerns how the brain computes the average, over many experiences, of rewards received for an option. But such a summary – produced by prominent models of dopaminergic incremental learning– is chiefly useful for repetitive tasks. Much less is understood about how the brain can flexibly evaluate new or changing options in more realistic tasks, which must rely on less aggregated information.
We describe studies which examine how decisions for reward can be guided by the most individuated memories, those for individual experiences or episodes. We combine novel behavioral methods with brain imaging and computational modeling to examine the relationship between hippocampal episodic memory and striatal incremental learning in decisions for reward.
Our findings suggest that both forms of learning guide decisions but that their relative contribution varies across trials as a function of the certainty of reward in the environment. Neurally, we identify both common and dissociable brain regions for how these two forms of learning guide choice. Notably, choices based on both forms of learning are related to BOLD activity in the hippocampus, raising the possibility that episodic retrieval may guide both forms of decisions. Computationally, episodic memories can support a family of learning algorithms that draw on sparse, individual experiences, such as Monte Carlo and kernel methods. These suggest novel, plausible hypotheses for how the brain solves more realistic decision problems based on rare events, and in particular how it implements a sort of reasoning known as "goal-directed" or "model-based' choice.
Authors: Raphael T. Gerraty, Daphna Shohamy, Nathaniel Daw, K. Duncan
Poster session II
Integration of sensory information across multiple senses is most likely to occur when signals are spatiotemporally coupled. Yet, recent research on audiovisual rate discrimination indicates that random sequences of light flashes and auditory clicks are integrated optimally regardless of temporal correlation. This may be due to 1) temporal averaging rendering temporal cues less effective; 2) task demands prompting decision-stage integration without low-level integration; or 3) difficulty extracting causal-inference cues from rapidly presented stimuli. We conducted a rate-discrimination task (Exp. 1) and a separate causal-judgement task (Exp. 2). We test Hypothesis 3 by using slower, more random sequences than previous studies. Exp. 1 measured unisensory and multisensory rate-discrimination thresholds to assess the effects of temporal correlation and spatial congruence on integration. Most subjects were indistinguishable from optimal for spatiotemporally coupled stimuli, and generally sub-optimal in other conditions, inconsistent with Hypotheses 1 and 2. In Exp. 2, subjects reported whether temporally uncorrelated (but spatially co-located) sequences were perceived as sharing a common source. A unified percept was affected by click-flash pattern similarity and the maximum temporal offset between individual clicks and flashes, but not on the proportion of synchronous click-flash pairs. Thus, stimulus-generation algorithms of previous studies may have prevented subjects from using the temporal mismatch of uncorrelated sequences for causal inference, as predicted in Hypothesis 3. Our results support the principle that multisensory stimuli are optimally integrated when spatiotemporally coupled, and provide insight into the forms of temporal coupling used for causal inference.
Authors: Shannon Locke, M. S. Landy
Poster session II
Advancing methods to non-invasively image and interpret neural activity in humans on fine temporal and spatial scales is essential to understanding how the brain works. Magneto-EIectroencephalography (MEG/EEG) combined with structural imaging provides reliable recordings of cortical activity with millisecond precision. Simultaneous recordings from subcortical structures, such as thalamus, have been limited due to low signal amplitudes and difficulty in source localization. Further, our understanding of the generation of the macroscopic electrical currents producing these signals from cellular events is lacking. We address these issues by integrating MEG/EEG, computational neural modeling, and invasive electrophysiology in animals and human patients to optimize MEG/EEG inverse methods to localize distributed thalamocortical sources and to interpret the underlying cellular events.
We present recent results applying our multi-modal approach to study commonly measured somatosensory evoked responses and low frequency rhythms. We show how our new inverse methods enable improved localization of distributed current sources evoked across thalamus and cortex. We also describe development of a biophysically principled neural model uniquely designed to study the circuit mechanisms underlying MEG/EEG measured source activity. This model has led to novel hypothesis on the thalamocortical mechanisms underlying spontaneous low frequency rhythms and the impact of these rhythms on sensory processing. Invasive laminar recordings in animal models, and ECoG recordings in human patients, support the model derived predictions and suggest the mechanisms underlying these rhythms are preserved across recording modalities and species. Lastly, we describe ongoing experiments and tools that are being developed to share freely with the broader neuroscience community.
Authors: Stephanie Jones, Alexandre Gramfort, Y. Bekhti, J. Guerin, M. Jas, R. Law, S. Lee, P. Sundaram, M. Hamalainen, W. Asaad
Poster session II
Long-term memories are thought to be maintained in part by persistent increases in synaptic strength. A key molecule in this maintenance mechanism has been proposed to be the PKC isoform, PKMzeta. Unlike most PKCs that are transiently activated by second messengers, PKMzeta is constitutively and therefore persistently active. Evidence indicates that the persistent activity of PKMzeta maintains LTP and long-term memory. Recently, PKMzeta’s role was challenged by PKMzeta-knockout mice showing normal appearing LTP and memory. Moreover, like in wild-type mice, the PKMzeta-inhibitor ZIP disrupts LTP and memory in PKMzeta-null mice. Two hypotheses can account for these findings. First, PKMzeta is unimportant for LTP or memory. Second, PKMzeta is essential for late-LTP and long-term memory in wild-type mice, and PKMzeta-null mice recruit compensatory mechanisms, which are also inhibited by ZIP. We used a pharmacogenetic approach to distinguish between these hypotheses. PKMzeta-antisense in hippocampus blocks late-LTP and spatial long-term memory in wild-type mice, but not in PKMzeta-null mice without the target mRNA. Whereas PKMzeta persistently increases in LTP in wild-type mice, another ZIP-sensitive isoform, PKCiota/lambda, persistently increases in LTP in PKMzeta-null mice, and a PKCiota/lambda-antagonist disrupts late-LTP and memory in PKMzeta-null mice, not wild-type mice. Thus, PKMzeta is essential for wild-type LTP and long-term memory. In PKMzeta-null mice, PKCiota/lambda takes on properties of PKMzeta, allowing it to be persistently active and compensate for PKMzeta in LTP and long-term memory. Computational modeling indicates PKMzeta and the PKCiota/lambda ‘back-up’ mechanism for LTP and memory provide a robust, bistable molecular mechanism for information storage.
Authors: Todd Sacktor, H. Shouval, P. Tsokas, A. Fenton
Poster session II
Our project studies how rats use their whiskers (vibrissae) to sense air currents and the implications this sensory ability has for their behavior and neural processing. We have addressed the broader question of whisker-air interactions using three approaches: behavioral studies of rats, bio-inspired whiskered robotics, and complementary experiments and numerical simulations of the fluid dynamics of whisker-air interactions.
Nearly all eutherian mammals have facial whiskers arranged on the cheek in an ordered array of rows and columns. To date, most studies of sensing in the rodent vibrissal-trigeminal system have examined tactile stimulation in which the whisker comes into direct contact with an object. However, even when not in contact with a solid object, the vibrissae are immersed in a fluid medium - air - whose velocity fluctuations are continually transmitting momentum, and thus force, to the whisker. Thus, whisker-air interactions are likely to be an important component of vibrissal sensing that has helped shape both rodent behavior and brain evolution.
Recent publications from our project have quantified the mechanical response of isolated whiskers to airflow (Yu et al., 2016a) and also shown that rats can use their whiskers to track airflow (Yu et al., 2016b). In a parallel exploration of whisker functionality, we constructed a robot that uses four whiskers to steer towards the source of an air current using control signals derived from the fluctuations of the whiskers in the airstream. When the whisker signals were used as inputs to a neural network, the system could accurately recognize “flowscapes,” or local flow environments. Our poster presents our latest behavioral, robotic, experimental and simulation results and discusses the advances we have made in how whiskers can sense flow.
Authors: V. Gopal, A. Beverage, M. Genovese, I. Ghouse, Y. Yu, M. Graff, C. Bresee, Y. Man, W. Kou, S. Park, N. Patankar, M. Hartmann
Poster session II
Deep brain stimulation (DBS) of subcallosal cingulate white matter (SCCwm) is effective in alleviating symptoms of treatment resistant depression (TRD). Mechanistic models of SCCwm-DBS therapeutic action in TRD require quantitative readouts of the brain’s activity and the activity’s relation to TRD severity. Recent DBS hardware advances allow for chronic intracranial electrophysiology in patients implanted with DBS stimulators, enabling long-term study of brain activity changes at the time scales of TRD remission. Local field potentials (LFPs) collected from the SCC may reflect depression state and clarify the effects of therapeutic DBS on SCC activity.
Here, we outline a multiscale approach to modeling the SCC-LFP in patients with DBS for TRD in order to address key needs in SCC-DBS research. Specifically (a) biophysical modeling of the brain tissue generating the SCC-LFP signal is needed to better understand how brain activity gives rise to disease, and (b) signal modeling of the relationship between SCC-LFP and behavior/disease state is needed to develop a clinically actionable metric of treatment efficacy. Our collaborative approach between engineers, neuroscientists, and clinicians is beginning to identify small- and large-scale changes in the SCC region through acute and chronic experimental designs. The modeling approaches being used to understand the origins of these signals and the implications of these signals are presented with some initial findings. This work will be a critical first step in understanding the therapeutic mechanism of SCCwm-DBS and standardizing the implementation of SCCwm-DBS for TRD neuromodulation.
Author: Vineet Tiruvadi
Poster session II
Spike-timing dependent plasticity (STDP) is a universal rule of synaptic plasticity. According to this rule, positively correlated pre- and postsynaptic activity leads to long-term synaptic potentiation (LTP) whereas negative correlation induces long-term synaptic depression (LTD). As most plasticity involves signaling by intracellular calcium, the extracellular calcium concentration is likely to play a critical role. Yet, most if not all in vitro studies devoted so far to STDP used non-physiological Ca2+ concentration (2-3 mM) whereas the physiological Ca2+ concentration is approx. 1.5 mM. Theoretical models of STDP based on calcium dynamics indicate that different types of STDP curves can be obtained as a function of calcium (Graupner & Brunel, 2012). In particular, with lower calcium concentration, only LTD occurs. CA1 pyramidal neurons were recorded in whole-cell patch-clamp in acute slices of young rat hippocampus and STDP was examined at the Schaffer to CA1 pyramidal cell-synapse in physiological extracellular Ca2+ concentration (1.3/1.8 mM). Confirming theoretical predictions, no change was observed with 1.3 mM Ca2+ and only LTD could be induced by positive or negative correlation in 1.8 mM Ca2+. LTP could be restored if the number of post-synaptic spikes was increased from 1 to 3 or 4. Furthermore, LTP could be induced with a single post-synaptic spike when the pairing was performed in the presence of the beta-adrenergic agonist isoprenaline. These data suggest that the STDP rule is altered in physiological Ca2+ but that a normal STDP profile can be restored under specific activity regimes or by application of a neuromodulator.
Authors: Yanis Inglebert, J. Aljadeff, N. Brunel, D. Debanne
Poster session II
The principle of efficient coding suggests that visual processing in early sensory systems should be adapted to the statistical properties of the stimulus. By comparing intracellular responses of V1 neurons to stimulus statistics of different complexity, we have shown that the temporal reliability of the neural code is optimized for natural statistics and that the stimulus-locked trial-to-trial variability of the subthreshold membrane potential waveforms is modulated by the statistics of the full field stimulus. Using the exact same stimulus seed but also additional stimuli as motion clouds and natural movies, we present here a multiscale analysis based on mesoscopic measures. We recorded, with high-density silicon probes (up to 64 channels), the unit activity (single and multi-unit activity) and the local field potential, on the anesthetized and paralyzed cat primary visual cortex. Our aim is to explore if the single-cell observations can be related (or not) to specific behavior and stimulus dependency shared by local ensemble of neurons. Also, if a laminar dependency of the observed effects can be detected by these mesoscopic methods and what is the global impact of input statistics on the correlation between the mesoscopic signals.
Authors: Yannick Passarelli, J. Vacher, M. Larroche, L. Foubert, Y. Frégnac, C. Monier
Poster session II
Source imaging based on magnetoencephalography (MEG) and electroencephalography (EEG) allows for the non-invasive analysis of brain activity with high temporal and good spatial resolution. As the bioelectromagnetic inverse problem is ill-posed, constraints are required. For the analysis of evoked brain activity, spatial sparsity of the neuronal activation is a common assumption. It is often taken into account using convex constraints based on the l1-norm. The resulting source estimates are however biased in amplitude and often suboptimal in terms of source selection due to high correlations in the forward model. We demonstrate that an inverse solver based on a block-separable penalty with a Frobenius norm per block and a l0.5-quasinorm over blocks addresses both of these issues. For solving the resulting non-convex optimization problem, we propose the iterative reweighted Mixed Norm Estimate in the standard domain (irMxNE) or in the Time-Frequency domain (irTF-MxNE). Source localization in the time-frequency domain has already been investigated using Gabor dictionary, however the choice of an optimal dictionary remains unsolved. Due to a mixture of signals, i.e. short transient signals (right after the stimulus onset) and slower brain waves, the choice of a single dictionary explaining simultaneously both signals types in a sparse way is difficult. We then introduce a method to improve the source estimation relying on a multi-scale dictionary, i.e. multiple dictionaries with different scales concatenated to fit short transients and slow waves at the same time. We demonstrate the benefits of the approach in terms of reduced leakage, temporal smoothness and detection of both signals types.
Authors: Yousra Bekhti, D. Strohmeier, M. Jas, R. Badeau, A. Gramfort
Poster session II
Mammalian grid cells discharge when an animal crosses the points of an imaginary hexagonal grid tessellating the environment. In this presentation, I will show how animals can navigate by reading out the population activity of grid cells across multiple spatial scales. The theory explains key experimental results, makes testable predictions for future physiological and behavioral experiments, and provides a mathematical foundation for the concept of a "neural metric" for space. For goal-directed navigation, the proposed allocentric grid cell representation can be readily transformed into the egocentric goal coordinates needed for planning movements. These results show that the grid-cell code provides a powerful and highly flexible neural substrate to solve various cognitive tasks.
The presentation is based on and extends the paper "Connecting multiple spatial scales to decode the population activity of grid cells" by M. Stemmler, A. Mathis and A. V. M. Herz.
Authors: Andreas V. M. Herz, M. Stemmler, A. Mathis
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - A. V. M. Herz (T).pdf | 2.47 MB |
The signal attenuation in magnetic resonance (MR) diffusion experiments is sensitive not only to the random motion of water molecules in tissues, diffusion, but also to the pseudo-random flow through the vascular network, Intravoxel Incoherent Motion (IVIM). Traditionally the IVIM signal decay has been modeled as a mono-exponential function from which microcirculatory perfusion parameters can be extracted. In this work we propose a bi-exponential model to analyze the IVIM imaging data. Through experiments and numerical simulations we demonstrate the validity of this model and show that it provides a more complete description of the cerebral vasculature reflecting the presence of two separate vascular pools, corresponding to two different flow regimes. By evaluating the influence of the experimental parameters on the model outputs we aim to establish the optimum protocols for IVIM cerebral perfusion measurements. Specifically, we show that the fast flowing pool is harder to detect at long diffusion encoding times for which the bi- and mono-exponential models converge. On the contrary, employing spin echo acquisitions with short repetition times enhances the contribution to the IVIM signal of the fast flowing spins facilitating the differentiation and characterization of the two vascular pools. While introduced and validated in a preclinical setting the bi-exponential model we propose has potential clinical applications, as lesions or responses to therapy are likely to be characterized by a differential balance between the fast and slow vascular pools.
Author: Luisa Ciobanu
Talk
The principle of efficient coding suggests that visual processing in early sensory systems should be adapted to the statistical properties of the stimulus. By comparing intracellular responses of V1 neurons to stimulus statistics of different complexity, we have shown that the temporal reliability of the neural code is optimized for natural statistics and that the stimulus-locked trial-to-trial variability of the subthreshold membrane potential waveforms is modulated by the statistics of the full field stimulus. Using the exact same stimulus seed but also additional stimuli as motion clouds and natural movies, we present here a multiscale analysis based on mesoscopic measures. We recorded, with high-density silicon probes (up to 64 channels), the unit activity (single and multi-unit activity) and the local field potential, on the anesthetized and paralyzed cat primary visual cortex. Our aim is to explore if the single-cell observations can be related (or not) to specific behavior and stimulus dependency shared by local ensemble of neurons. Also, if a laminar dependency of the observed effects can be detected by these mesoscopic methods and what is the global impact of input statistics on the correlation between the mesoscopic signals.
Authors: Cyril Monier, Y. Passarelli, J. Vacher, M. Larroche, L. Foubert, Y. Frégnac
Talk
Widening economic inequity is a key concern for our society, and previous cohort-based studies have also suggested a link between economic inequity and depression. However, little is known about the underlying neural mechanism of the link, due to substantial individual differences. Here, we demonstrate that functional magnetic resonance imaging activity patterns in the amygdala/hippocampus induced by the inequity during an economic game can predict both present and future (measured one year later) depression index (BDI-II). Such predictions were not possible by behavioral measures of the participants. These findings reveal the connection between the sensitivity of the amygdala/hippocampus to economic inequity, and the present and future depression index, and suggest an important involvement of the amygdala and hippocampus into the effect of economic inequity on human mental states.
Author: Masahiko Haruno
Talk
In addition to the yearly CRCNS call for research/data proposals, the funders of the CRCNS programmes run other programmes of interest for the community of the attending researchers. In this information session such funding opportunities will be presented. The speakers are:
- Kenneth Whang (NSF) and Andrew Rossi (NIH) for the US
- Sheyla Mejia-Gervacio (ANR) for France
- Yair Rotstein (BSF) for Israel
Info session
Attachment | Size |
---|---|
CRCNS Conference 2016 - Funding opportunities.pdf | 2.35 MB |
Hughes BERSINI is Professor in Université Libre de Bruxelles and Co-Director of the IRIDIA laboratory. He teaches Artificial Intelligence, Object-Oriented technologies (UML, Design Patterns, Java, .Net, ...) and Web Programming (Django/Python) both for academics and for enterprises. He is a member of the Belgium Royal Academy of Science.
Hughes BERSINI's main research interests are modelling and control of complex systems, neural networks and fuzzy control, data/graph/text mining, autonomous agents and dynamics of biological networks, computational chemistry, immune engineering, cognitive sciences, bioinformatics and object-oriented technology.
He has written several French books essentially dedicated to computer science. Some go along his teaching activity and have become with the years quite popular in the academic world. Others are essays to help readers to better understand complex systems. And finally two books at the crossroad between science and fictions.
I will first discuss how since the birth of AI two traditions have always been very active, labelled for simplicity “conscious and unconscious”, then how and why the second one, based on Big Data and Machine Learning, is dangerously taking the lead today. Thus I will discuss what more recent researches in neural networks such as deep learning, self-adapting nets and chaotic Hopfield networks can bring to neurosciences.
Keynote lecture
Understanding how neural networks process information based on high-dimensional noisy measurements requires a breakthrough both in large scale data collection and in the development of analysis and modeling. We address these key issues in the context of sensory-motor control. We focus on the characterization of the relations between time-dependent neural activities in different cortical regions (motor and sensory) and associate them to behavior through novel manifold learning techniques.
To deal with these questions, we develop novel tools for recording neuronal data in behaving animals, using a newly developed motor reach task in awake head-fixed mice, and genetically encoded calcium indicators to chronically record the activity over extended periods from specific identified neuronal populations, using two photon microscopy. This motor task is complex, versatile, yet natural to mice. To deal with the high dimensionality, complexity and richness of this data, we develop a nonlinear, data-driven and model-free approach, which leads to the construction of intrinsic dynamic representations of the data that provide a coherent explanation of the observed data and testable experimental predictions.
Specifically, our approach gives rise to the joint organization of neurons and dynamic patterns in data-driven hierarchical structures. We show that this methodology enables us to extract hidden biological features and accurate indications of pathological dysfunction. We demonstrate its capability to identify, solely from observations, patterns of neuronal activity and variability related to external triggers (e.g., a tone) and behavioral events (e.g., the sequence of motor actions), at different time scales, and among specific neuronal sub-groups.
Authors: Ron Meir, G. Mishne, R. Talmon, J. Schiller, M. Lavzin, U. Dubin, R. R. Coifman
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - R. Meir.pdf | 2.07 MB |
In this talk we report on the following abstracts:
Abstract 1:
Throughout adulthood the mouse olfactory bulb is repopulated with new granule cells (GC) neurons through a process termed adult neurogenesis. It is unknown whether this persistent remodeling can potentially drive synaptic plasticity, as the pre-existing circuitry needs to adapt to allow new neuron integration. In this study, we tracked the dendritic and spine development of adult-born GCs using 2-photon in vivo imaging. Overall, adult born GC dendritic structure stabilized at one month. In contrast, adult born GC distal apical dendritic spine dynamics plateaued at two months, but remained highly dynamic. We compared the spine dynamics of the adult born GCs with pre-existing, early-postnatal born GCs and found matching spine turnover. To test whether the GC spine dynamics correlated with synapse turnover of their synaptic partners, mitral/tufted cells (MC), we labeled mitral/tufted cells with fluorescently labeled gephryn, a GABA receptor structural protein. In vivo imaging of the gephryn puncta demonstrated dynamics matching GC spines. Using a computational model of GC-MC structural plasticity, we found that these dynamics enable the network to rapidly optimize its output in response to changes in the odor environment. Surprisingly, odor representations quickly settled into a steady state despite continued rapid remodeling of the circuit. GC to mitral/tufted cell synapses have so far been demonstrated to lack synaptic strength plasticity, therefore the olfactory bulb appears to be unique in the adult brain, where persistent structural remodeling plasticity may be its dominant form of plasticity.
Abstract 2:
Centrifugal feedback projections from higher to lower brain areas are pervasive in the mammalian brain. They can provide the receiving brain area with information that is not available in the feedforward stream and thereby modify the information processing by the lower brain area in a task-dependent manner. These functions depend essentially on the network connectivity, but it is only poorly understood what controls its formation and specificity. The rodent olfactory system is well suited to investigate the mechanisms controlling the evolution of network connectivity and functionalities arising from it. Even in adult animals the olfactory bulb, which directly receives the sensory information from the nose, is persistently rewired through the addition and controlled removal of granule cells (GCs), which comprise the bulb's dominant interneuron population, and through highly dynamic fluctuations of the dendritic spines connecting the inhibitory GCs with the bulb's principal neurons, mitral/tufted cells (MC/TC). We developed computational models for the network evolution via spine fluctuations and adult neurogenesis of the GCs. Our modeling shows that Hebbian-type spine stability can explain the experimentally observed increase in spine stability with spine age. Our neurogenesis model focuses on the role of centrifugal projections from olfactory cortex onto GCs. The experimentally supported activity-dependence of GC survival leads to the formation of subnetworks that provide bidirectional connections between the bulbar and the cortical representations of learned stimuli. The resulting specificity of cortical inhibition of MC/TC via GCs allows, e.g., context-enhanced discrimination of occluded stimuli and cortical switching of bulbar processing.
Authors: Kurt A. Sailor, Hermann Riecke, Pierre-Marie Lledo, M. T. Valley, M. T. Wiechert, G. J. Sun, H. Song
Talk
Haim Sompolinsky is the William N. Skirball Professor of Neuroscience at the Edmond and Lily Safra Center for Brain Sciences, and a Professor of Physics at the Racah Institute of Physics at The Hebrew University of Jerusalem, Israel. He is also a visiting professor in the Center of Brain Science at Harvard University and the director of Harvard’s Swartz Program in Theoretical Neuroscience.
Haim Sompolinsky received his Ph.D. in physics from Bar-Ilan University in Tel Aviv. He then worked as postdoctoral fellow in the physics department at Harvard University. He was appointed associate professor of physics at Bar-Ilan University, when he moved in 1986 to the Hebrew University of Jerusalem as professor of physics.
Since the mid-1980s, he has pioneered the new field of computational neuroscience, introducing methods and concepts of theoretical physics to the study of neuronal circuits, memory, learning and neuronal information processing. Sompolinsky’s research includes spike-based neural learning and computation, neuronal population codes, sensory representations, dynamics and function of sensory and motor cortical circuits, and large-scale structure and dynamics of human brain. He also studies the relation between physics, neuroscience, and human volition, freedom and agency.
Neural circuits in the cerebral cortex appear to operate in a regime in which strong excitatory synaptic currents are roughly balanced by inhibition, resulting in a net input that is considerably smaller than either of these components individually. Theoretical and computational studies of the past twenty years have shown that simple neuronal circuits with strong synapses settle into a balanced state, characterized by fluctuation-driven, irregular, asynchronous firing patterns. These studies also elucidated the computational advantages and limitations of balanced networks.
In my talk, I will describe recent work that examines the balanced state in biologically realistic models of cortical circuits. I will show that realistic heterogeneous input connectivity of these circuits threatens to prevent the dynamic balance between excitation and inhibition, yielding, instead, unrealistically sparse and temporally regular firing. I will then show how homeostatic synaptic plasticity as well as cellular adaptation can restore the fluctuation-driven, asynchronous dynamics.
Previous work on balanced states focused largely on sensory processing circuits. I will describe a novel state of excitation-inhibition balance in recurrent circuits of associative memory. This state emerges from synaptic learning that incorporates the realistic requirement of noise-robust memory retrieval.
Taken together, these studies provide new insights into the mechanisms underlying excitation-inhibition balance in realistic cortical circuits and into their surprising functional benefits.
Keynote lecture
Midbrain dopaminergic (DA) neurons display two distinct patterns of spiking: low frequency tonic spiking and short burst-like episodes of high-frequency spiking. Synaptic inputs are suggested to be responsible for the bursts, but the mechanisms are not clear. In order to explore this, we model the circuitry of the ventral tegmental area (VTA) and analyzed the influence of synaptic inputs on the model. We find that the burstiness (the number of spikes in bursts) is drastically reduced when NMDA, but not AMPA receptors are blocked. When DA and GABAergic neurons are combined to simulate the local circuitry of the VTA, we find that high-frequency firing can be facilitated by the interaction between these two types of neurons. In particular, when GABA inputs are synchronized, the GABAa receptor currents contribute to high-frequency firing of the DA neuron by augmenting the fast AHP. When the GABA inputs are desynchronized, they only inhibit the DA neuron, and a burst emerges at the offset of these inputs, according to the disinhibition mechanism. Thus, the study suggests a combination of NMDA, AMPA and GABA receptor activation that contributes most significantly to the bursting pattern of the VTA DA neurons. These mechanisms are explored in the context of the action of alcohol and salient environmental stimuli.
Authors: Alexey Kuznetsov, Boris Gutkin, C. Lapish
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - A. Kuznetsov.pdf | 1.99 MB |
In this talk we report on two abstracts:
Abstract 1:
Risky choice is subject to a number of violations of normative decision-making, which are accounted by a number of descriptive models of choice, such as the Prospect Theory. These models, however, are not process models. In our study we examine the time-course of risky choices (lotteries), while we also track the eye-trajectories of the participants, who chose between the lotteries. The behavioral results indicate risk-aversion, which appears to increase with the lottery-amount. We fit the data via a number of computational models and we compute Bayesian Information Criteria. We also use the eye-trajectories to better understand the development of preference within each trial. Towards this aim we developed a neurocomputational model for risky choice, in which units that encode log-probability and log-value are summed to generate a measure of preference, and the weight of the units is modulated by the eye-location. The model is able to account for the choices to a degree roughly equal to what can be obtained within the Prospect Theory, under a more parsimonious framework. The model also allows us understand the role of attention in generating preference and in making decisions.
Abstract 2:
Multi-attribute decision making requires integration of attributes to generate a value estimate. This process could occur in parallel by integrating all attributes simultaneously, or sequentially by switching attentional focus between different attributes. To distinguish between these alternatives, we developed a novel gamble task to determine where attention is deployed during the decision process. The task will be used for both monkeys and humans. Subjects choose between gamble options, each characterized by two attributes: reward amount and probability. The amount and probability of a given option are each represented by the length of a line segment on a screen, and the two segments corresponding to the same option are visually connected on the screen. Using eyetracking, a line segment is only revealed when the subject directly looks at it, allowing us to observe which attribute the subject is attending at a given time. Subjects are free to make eye movements and to inspect each cue as long and as often as necessary. They indicate their final choice by a hand movement. Preliminary experiments in humans show that on most trials more than four fixations are made before the final choice. The number of fixations rises with the difficulty of the choice. In the first four fixations, the subjects showed strong idiosyncratic spatial biases that were absent in later fixations. Subjects showed two different search strategies, with one group prioritizing inspection of the different attributes of the same gamble option while the other group compared primarily the same attribute across gamble options.
Authors: Marius Ushe, Moshe Glickman, Ernst Niebur, Veit Stuphorn, Dino Levy
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - M. Usher.pdf | 1.1 MB |
Can a sensory stimulus be learned from an animal’s behavioral response? How are different behaviors organized and controlled by an animal? We study this question in the context of thermal escape response in the worm, C. elegans, a model for nociception. Here we develop a model of the stimulus-behavior relation to infer the perceived level of thermal nociception from the worm's stereotyped escape response. We analyze the behavior of ibuprofen-treated worms and a TRPV (transient receptor potential) mutant, and show that the perception of the thermal signal for the ibuprofen treated worms is lower than the wild-type, demonstrating a method for quantification of pharmacological effects. At the same time, our model shows that the mutant changes the worm's behavior beyond affecting sensory transduction. Thus the model also provides a method of assigning behavioral changes to the sensory or the motor system. Further, we use the recently developed algorithm called Sir Isaac for automated inference of the dynamical system underlying the escape behavior. The algorithm discovers that the behavior is approximated well by having one hidden degree of freedom beyond the observed worm velocity. This degree of freedom integrates the thermal signal and uses it through intertwined negative and positive feedback loops to initiate and stabilize the escape. According to the inferred model, the phase space of the worm’s behavior has two attractors, corresponding to forward and escape motions. The sensory stimulus affects the stability of these attractors, destroying the forward attractor and forcing the escape for large stimuli levels.
Authors: William Ryu, K. Leung, A. Mohammadi, B. C. Daniels, I. Nemenman
Talk
People’s concern for others is a long-celebrated feature of human sociality, yet people do not appear to extend this concern uniformly. How people treat others depends on who those others are—for example on their race, gender, or nationality—violating prevailing models of social preferences. We present a parsimonious, generalizable computational model model that explains a variety of group-contingent effects on behavior, such as differences in altruism shown by subjects toward people of various ethnic and nationality backgrounds, occupation, and behavioral tendencies. We show that our model provides highly accurate predictions of how subjects behave toward counterpart players that belong to a variety of different social groups (e.g., Nurse, Irish, Japanese, Elderly, Drug Addict). In contrast, standard other-regarding preferences models cannot explain these differences. In addition, the model generalizes across counterpart group membership, in that once calibrated, the model makes accurate predictions about subjects’ behavior toward novel social groups. For example, how subjects behave toward Nurses and Greeks can be used to predict how they behave toward elderly and homeless. It is generalizable across games, including well-known ones such as Dictator, Ultimatum, Trust, and Deception games. Finally, it is generalizable across populations, in that the model calibrated on perception and choices of a sample from a particular population make highly accurate predictions of those of another sample from a different population. Together, these results elucidate how social perception influences well-studied components of social valuation, including preferences for fairness and reciprocity, with implications for scientific understanding of discrimination.
Author: Ming Hsu
Talk
Odors activate large populations of olfactory bulb mitral cells. These cells project to piriform cortex where they activate sparse and distributed cortical odor ensembles. We obtained simultaneous recordings from large populations of mitral cells and piriform cortex cells in awake, head-fixed mice to understand how the dense odor responses in bulb are transformed in cortex. Individual piriform cells could be activated or suppressed by different odors and a linear classifier could accurately decode odors from the population using single sniff-spike counts, indicating that these representations are reliable and robust. At the population level, odors evoked a sustained increase in net bulb output. By contrast, there was almost no net change in total cortical output, with a small, brief increase in spiking immediately followed by sustained suppression. We developed a spiking network olfactory bulb-piriform cortex model to reveal the origins of this transformation. In the model, odors activated distinct sets of glomeruli and associated mitral cells at specific phases throughout the sniff cycle. In piriform cortex, a small subset of pyramidal cells were activated by the earliest bulb inputs, which helped recruit other pyramidal cells that only received subthreshold bulb input through their recurrent collateral connections. However, this ramping cascade of cortical activity then recruits strong feedback inhibition that suppresses cortical spiking and discounts the impact of later bulb input. Thus, using a combination of experimental and computational approaches, we provide a mechanistic explanation for how the sparse cortical odor responses are largely defined by the earliest activated bulb inputs.
Authors: Kevin M. Franks, Alexander Fleischmann
Talk
Spiking records from the sensory neocortex have revealed a remarkable degree of functional selectivity. For example, Hubel and Wiesel (1962) demonstrated exquisite selectivity in cat and primate visual cortical neurons for the orientation, motion and depth of objects in the world. These visual selectivities have been shown across mammals, though the functional organization within visual cortex varies: neurons in carnivores and primates are organized into functional maps where nearby neurons share retinotopy, orientation and motion preference whereas functional maps beyond retinotopy are largely absent in rodents. This difference in the functional organization across species for orientation raises the question of whether the emergence of orientation selectivity depends on a feature specific connectivity, or whether random connectivity between neurons is sufficient to account for orientation selectivity within visual cortex. We will present theoretical and physiological evidence showing that a circuit composed of random connectivity can account for mouse orientation selectivity. The random connectivity model makes critical predictions about the dependence of orientation preference on the specific form of the stimulus. We tested and confirmed these predictions using both intracellular recordings and calcium imaging in mouse visual cortex. Our theoretical and experimental results thus indicate that the selectivity observed in mouse visual cortex could emerge from random connectivity alone instead of specific connectivity between neurons.
Authors: Nicholas Priebe, C. van Vreeswijk, D. Hansel, J. Pattadkal, G. Mato
Talk
In this collaborative US-German project with Leibniz Institute for Neurobiology (LIN) experiments have been conducted using a well-established rodent learning model, the two-way active avoidance paradigm in the shuttle box, to analyze neuronal signatures of auditory discrimination learning in Mongolian gerbils implanted with an array of 5x4 electrodes over the auditory cortex. Our experiments confirm previous results on sudden behavioral change from the initial naive state to the avoidance state as learning progresses using CS+ and CS- stimuli.
- We defined various causality metrics, including the “inverse causality interaction” (ICI) index, to quantify the relationship between channel pairs before and after the animal acquired successful discrimination.
- We found that the number of channel pairs collected based on the ICI criterion is significantly increased after the animals transitioned to the avoidance stage.
- Spatial patterns between some electrode pairs emerged after the transition to the avoidance stage, indicating the formation of localized connectivity patterns between cortical areas as the result of the learning process.
- We have evaluated for classification performance of electrode pairs based on ICI index and demonstrated statistically significant discrimination between CS+ and CS- conditions after the transition to avoidance behavior. To investigate learning effects in the cortex, we used a simple graph theory model with coupled excitatory-inhibitory units and having rewiring effect (small-world properties).
- We studied the effect of the following parameters: proportion of inhibitory nodes, probability of long-range (non-local) connections, and intrinsic noise level.
- Our modeling study shows learning-induced changes in the discrimination properties can be reproduced by the graph theory model suggesting that connectivity changes may be responsible the observed behavioral transitions from naïve to avoidance state in gerbils.
Our findings underscore the importance of functional reorganization in early sensory areas and provide ISI as a possible neural correlate to behavioral changes.
Talk
Attachment | Size |
---|---|
CRCNS Conference 2016 - R. Kozma.pdf | 2.89 MB |
Beta oscillations (15 – 30 Hz) are a ubiquitous finding in the cortico-basal ganglia-thalamic (CBT) motor loop. They are elevated in Parkinson’s disease and correlate with movement disability, yet mechanisms controlling their expression are not understood. Though a series of computational and laboratory experiments, we have established that elevated striatal cholinergic tone can enhance beta oscillations in CBT circuits. Additionally, we show that optogenetic stimulation of striatal cholinergic interneurons produces parkinsonian-like motor deficits in normal mice. Modeling based on the experimental results provides further insight into a potential therapeutic mechanism of action of deep brain stimulation (DBS). Our models propose that DBS may act through the indirect pathway of the basal ganglia, suppressing the overactivity of striatal medium spiny neurons. Overall, these results suggest striatal cholinergic tone is a key modulator of CBT beta oscillations associated with parkinsonian-like motor symptoms and give insight into possible mechanisms mediating their production and transmission in CBT circuits.
Authors: Michelle M. McCarthy, K. Kondabolu, E. A. Roberts, B. R. Pittman-Polletta, A. Quach, A. I. Mohammed, M. Romano, M. Bucklin, N. Kopell, X. Han
Talk
Alain Destexhe is Research Director at CNRS. He is Editor in Chief of the Journal of Computational Neuroscience, and has obtained grants from several institutions, such as French National Research Agency, European Commission, Medical Research Council of Canada, NIH (National Institutes of Health, US), and Human Frontier Science Program.
His main research interests in computational neuroscience stand at the interface between physics (dynamical systems) and neurosciences (electrophysiology). The current research directions include the study of stochastic activity in cerebral cortex and its consequence on integrative properties and coding, the study of collective dynamics in networks (oscillations, chaotic states) and local field potentials, as well as the conception of methods at the interface with electrophysiology (conductance extraction, analysis of multiple electrode recordings, Active Electrode Compensation method, etc.).
We illustrate that in integrative neuroscience, computational modeling is tightly integrated with experimental data, and that this integration is very powerful to answer a number of problems in neuroscience. A first area where there is such a tight model-experiment association is about the study of single-neuron dynamics, and more generally, the integrative properties of single neurons. This subject requires three steps:
- to measure from in vivo experiments the characteristics of neuronal activity such as conductance and sub-threshold dynamics, as well as spiking activity;
- to design computational models of these dynamics, and predict how single neurons function in such states;
- to integrate the models into in vitro experiments in order to test predictions and better understand the integrative properties of the neurons.
A second example of such tight experiments-theory interaction is to study network-level phenomena, such as the emergence of oscillations in the brain. Similar to above, the oscillations must first be characterized in vivo, then integrated into computational models which can generate predictions that can be tested in vitro.
We illustrate this interaction for the case of absence epileptic seizures in the brain. Such problems require a tight combination of in vivo and in vitro experiments, and computational models play the link between them.
Finally, we illustrate another example, where computational models are used to link between experiments and the design of dedicated "neuro-morphic" circuits. In this case, the biological experiments provide a number of principles, which are formulated by models, and later integrated by engineers into the design of electronic circuits integrating these principles. Such an approach is supported by EC-funded projects such as BrainScales and The Human Brain Project.
Keynote lecture
The reinforcement learning algorithms, developed in the 80's to endow artificial agents with stronger autonomy, have had a deep influence on the study of decision-making in neuroscience after the discovery, in the 90's, of the striking similarity between these algorithms and the interactions of the cortex and basal ganglia loops with the dopaminergic system. A new exciting current in Artificial Intelligence (AI) and robotics is the development of deep learning methods. They have been steadily gaining momentum, and these have also been taught widely in the popular press as a watershed for modelling intelligence and the brain.
This workshop working group will put together views from the AI and robotics community and from neuroscience (computational and experimental) to address the following questions:
- Is state-of-the-art AI still in position of fruitfully inseminating neuroscience? Have deep learning methods the potential to lead to new success stories? How could these and other current methods be used to analyse data, and to formalise information processing principles in neural circuits?
- Conversely, are discoveries in computational neuroscience in position of feeding AI and robotics with new ideas? In these domains, is neuro-inspiration complementary to the traditional engineering approach? What might neuroscience bring to the deep learning community?
Time |
October 26
F. Jacob Auditorium
|
10:00 |
Introduction and goals - Mehdi Khamassi, Boris Gutkin
Questions & discussion
|
10:10 |
Position statement by Samuel Gershman
Questions & discussion
|
10:35 |
Position statement by Haim Sompolinsky
Questions & discussion
|
11:00 | Coffee break |
11:30 | Position statement by Sophie Deneve |
11:55 | Position statement by Ron Meir |
12:20 | General discussion and conclusions |
13:00 | End of working group session |
Workshop working group
Attachment | Size |
---|---|
CRCNS Conference 2016 - WG2.pdf | 2.89 MB |
Neuroimaging includes a large portfolio of techniques capable of investigating the brain at different spatial scales. Combining the various measurements in order to understand how microscopic characteristics give rise to macroscopic behaviors is a major challenge of modern neuroscience. From a medical point of view, putting together results from different neuroimaging studies in order to formulate medical diagnoses is often one of the most difficult tasks faced by physicians.
This workshop working group will identify difficulties encountered when bringing together specific neuroimaging methods such as functional MRI, microscopic MR, PET and optical imaging from the point of view of signal modeling and data analysis. The discussions will be initiated by leading experts who will present advantages, disadvantages and limitations of different brain imaging technologies.
Time |
October 26
Fernbach Building, Room Aubert
|
10:00 |
Introduction and goals - Luisa Ciobanu
Questions & discussion
|
10:15 |
Neuromechanical coupling: Evidence from diffusion MRI and electrophysiology studies - Denis Le Bihan
Questions & discussion
|
10:30 |
Complementarity of PET and MR - Hans Wehrl
Questions & discussion
|
10:45 |
From medical imaging to decision support systems - Guy Courbebaisse
Questions & discussion
|
11:00 | Coffee break |
11:30 | General discussion |
13:00 | End of working group session |
Workshop working group
Attachment | Size |
---|---|
CRCNS Conference 2016 - WG1.pdf | 5.08 MB |
Particularly for more complex cognitive tasks, neural recordings (single and mutli unit electrodes and arrays, EEG, iEEG, fMRI, MEG) reveal complex high dimensional combinations of task and context factors that can be difficult to directly correlate with behavior. When observed at a population level however, this coding scheme is revealed to reflect a general high dimensional representation with universal properties. This same coding mechanism is seen in neural networks with fixed recurrent connections (Enel et al 2016, Fusi et al 2016, Rigotti et al 2013, Sussillo & Barak 2013).
This workshop working group will address how high dimensional representations in recurrent reservoir networks can be applied both to the interpretation and decoding of, and the modeling of high dimensional neurophysiological data.
Peter F. Dominey will make a 10 min introductory presentation briefly reviewing high dimensional coding in reservoir computing and primate cortical neurophysiology. The three co-moderators will then each make 10 min presentations where they identify open questions in the context of the goal of the workshop. In the second part of the workshop, the discussion will be focused on the points previously raised: In particular one of the goals of the working group is to identify limits of how far we can go in considering that the cortex displays reservoir dynamics, and that these dynamics provide a general purpose machine for higher cognitive function, and that this framework can form the basis for future developments in interpretation and decoding of neurophysiological signals.
Time |
October 26
Lwoff Building, Room 14
|
10:00 |
Introduction and goals - Peter F. Dominey
Questions & discussion
|
10:15 |
Mixed selectivity - Omri Barak
Questions & discussion
|
10:30 |
Interpreting high dimensional data - Alberto Bernacchia
Questions & discussion
|
10:45 |
Reservoir computing and rodent DMS task - Anand Subramoney
Questions & discussion
|
11:00 | Coffee break |
11:30 | General discussion |
13:00 | End of working group session |
Workshop working group
Attachment | Size |
---|---|
CRCNS Conference 2016 - WG3.pdf | 1.48 MB |