Classic computational theories of the hippocampus posit that the dentate gyrus region is specialized for pattern separation and the CA3 region is specialized for pattern completion. These theories were inspired by anatomical data. The dentate gyrus contains a larger number of cells than its input region (the entorhinal cortex), suggesting that the dentate gyrus performs an "expansion recoding" computation as originally proposed for cerebellar granule cells by Marr. The CA3 region contains a system of recurrent collaterals that suggest the existence of attractor circuitry and dynamics that might underlie pattern completion, generalization, and error correction. In support of these theories, I will present single unit recording data showing (a) sparse encoding of dentate gyrus granule cells predicted by the expansion recoding model of dentate gyrus and (b) a gradient of pattern completion (error correction) along the transverse axis of CA3 that correlates with the density of recurrent collaterals along this axis.
Monday, February 12th, 2018
The standard framework of hippocampal memory function states that CA3 serves as a Hopfield type auto-associative memory that stores patterns in its recurrent connections. Dentate gyrus orthogonalizes patterns and prepares them for storage in CA3. CA1 helps decoding patterns from CA3.
The CRISP theory recently proposed by Sen Cheng states that the main site of storage are not the recurrent connections in CA3 but the feed-forward connections to and from CA1. CA3 serves as a sequence generator to which memory sequences are associated.
I present here some simulation results that investigate the implications of the CRISP theory and support the idea that storage in feed-forward connections might have computational advantages over the standard view of storage in recurrent connections in CA3.
No abstract available.
Hippocampal place cell ensembles form a cognitive map of space during exposure to novel environments. However, surprisingly little evidence exists to support the idea that synaptic plasticity in place cells is involved in forming new place fields. Here we used high-resolution functional imaging to determine the signaling patterns in CA1 soma, dendrites, and axons associated with place field formation when mice are exposed to novel virtual environments. We found that putative local dendritic spikes often occur prior to somatic place field firing. Subsequently, the first occurrence of somatic place field firing was associated with widespread regenerative dendritic events, which decreased in prevalence with increased novel environment experience. This transient increase in regenerative events was likely facilitated by a reduction in dendritic inhibition. Since regenerative dendritic events can provide the depolarization necessary for Hebbian potentiation, these results suggest that activity-dependent synaptic plasticity underlies the formation of many CA1 place fields.
The ability to represent multiple possible futures, or prospection, is essential to cognition. In human studies, prospective thinking activates and requires the hippocampus, suggesting a critical role for hippocampal neural activity. Recent research focusing on rodent spatial navigation has found that hippocampal neural activity in ~100 ms periods represents spatial paths that project from a subject’s current location, thus establishing neural representation of single prospective scenarios at the sub-second timescale. However, it has remained unclear whether the brain can regularly switch between representations of different possibilities at this timescale. Here I will discuss results demonstrating that the hippocampus is capable of continuously switching between divergent representations at 8 Hz. We found that hippocampal neurons representing differing spatial experiences (bifurcating paths and paths in different travel directions) routinely fire in alternation at 8 Hz. Notably, firing consistent with 8 Hz alternation was strongest when a navigational choice requiring prospection was imminent. Further, we found that hippocampal neural firing at the population level commonly represented divergent spatial experiences in continuous alternation at 8 Hz, and that this activity pattern occurred in the absence of overt deliberative behavior. These findings establish that prospective representation in the hippocampus is rapidly flexible, and moreover outline a temporally- and anatomically- specific neural mechanism capable of supporting prospection. These findings also suggest that cognitive capacities dependent on prospection (e.g. decision-making, planning, and imagination) should be evaluated at the sub-second timescale.
Erratum: Following this talk, we discovered an error in our analysis code related to the predictions of choices in the new W-track. The corrected analyses revealed only very weak predictions, suggesting that the alternating representations of future possibilities is not biased toward actual choices. The corrected results can be seen on slide 17 of the new pdf.
Episodic memory is defined as memory for events occurring at a specific time and place. Neurophysiological recordings from brain regions in behaving rodents demonstrate neuronal response properties that may code space and time for episodic memory. This includes the coding in entorhinal cortex and hippocampus of spatial location by place cells (O'Keefe and Burgess, 1996) and grid cells (Hafting et al., 2005; Stensola et al., 2012). This talk will focus on data on the coding of temporal intervals by time cells (Kraus et al., 2013; 2015) of running speed by speed cells (Kropff et al., 2015; Hinman et al., 2016) and of environmental barriers by allocentric boundary cells (Solstad et al., 2008; Lever et al., 2009) as well as egocentric boundary cells (Hinman et al., 2017). Experimental data and computational modeling have explored potential cortical and subcortical mechanisms for the neural coding of time and space. Inactivation of input from the medial septum impairs the responses of neurons coding space (Brandon et al., 2011) and time (Wang et al., 2014) indicating subcortical contributions, but grid cells and boundary cells alter their coding in response to sensory input about environmental barriers conveyed by cortical inputs.
Mammalian grid cells discharge when an animal crosses the points of an imaginary hexagonal grid tessellating the environment. In this talk, I will show how animals can use this code by reading out the grid-cell population activity across multiple spatial scales. The theory explains key experimental results (grid-field alignment, modularity, grid-scale ratio), makes testable predictions for future physiological and behavioral studies (specific changes in foraging when particular parts of the system are silenced), 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 for spatial cognition.
Tuesday, February 13th, 2018
The basic architecture of basal ganglia circuits has been conserved for at least half-a-billion years, helping animals learn through trial-and-error to obtain rewards. I will briefly outline this architecture and how it is thought to process information. I will also briefly describe the fruitful dialog between psychology/neuroscience and computer science that has played an important role in the development of Reinforcement Learning algorithms for contemporary artificial intelligence. As time permits I will describe some key current debates about the functions of the basal ganglia and how they are modulated by dopamine.
Neuroscientists have generally viewed learning as being implemented by a few, generic synaptic plasticity rules, with the specialization for specific behavioral tasks arising from the circuit architecture. In contrast, I will describe evidence that the synaptic plasticity rules themselves can be precisely tuned to functional requirements. More specifically, my laboratory discovered heterogeneity in the learning rules implemented by different subsets of parallel fiber-to-Purkinje cell synapses in the cerebellum. Our results suggest that the rules governing synaptic plasticity at these synapses are tuned to compensate for circuit-level feedback delays, thereby solving the temporal credit assignment problem.
The thalamus and cerebral cortex are intimately interconnected by a dense network of feedforward and feedback circuits. In the visual system, the response properties of neurons in the lateral geniculate nucleus (LGN) of the thalamus and primary visual cortex (V1) are governed by the anatomical organization of connections and the temporal patterns of impulse arrival. Recent advances in multielectrode recording technology have made it possible to study the role of both features in the same experiment. Results will be presented from experiments that examined the specificity of neuronal connections and the role of spike timing and behavioral modulation in the reciprocal exchange of information between the LGN and V1 in the alert macaque monkey. These results reveal a striking relationship between the parallel feedforward and feedback processing streams, as well as the biophysical properties that govern spike transfer and the encoding of visual information in neuronal spike trains.
Corticogeniculate circuits connect the primary visual cortex with the visual thalamus in the feedback direction and are a subset of corticothalamic circuits that are ubiquitous across mammals and sensory systems. Corticogeniculate feedback is anatomically robust in that the number of synapses onto thalamic relay neurons originating from the cortex far outnumber those originating from the retina. However, the functional role of corticogeniculate feedback in vision has remained a stubborn puzzle in visual neuroscience. We employed a novel combination of virus-mediated gene delivery and optogenetics to selectively and reversibly manipulate the activity of corticogeniculate neurons in vivo. We recorded the responses of thalamic neurons in response to visual stimulation with additional optogenetic activation of corticogeniculate feedback. We discovered that corticogeniculate feedback dramatically sharpens the response precision of visual thalamic neurons. Specifically, stimulation of corticogeniculate feedback reduced onset response latencies, increased spike timing precision, and reduced the size of the classical receptive field among thalamic neurons. We propose a simple circuit-based model to explain these effects. Together, our findings suggest that corticogeniculate feedback controls the timing of thalamic responses to visual stimuli and broadly support the notion that the thalamus is much more than a simple relay of sensory information.
Detecting the direction of an object’s motion is essential for our representation of the visual environment. Visual cortex is one of the main stages in the mammalian nervous system where motion direction may be computed de novo. Experiments and theories indicate that cortical neurons respond selectively to motion direction by combining inputs that provide information about distinct spatial locations with distinct time-delays. Despite the importance of this spatiotemporal offset for direction selectivity its origin and cellular mechanisms are not fully understood. I will showthat ~80+/-10 thalamic neurons responding with distinct time-courses to stimuli in distinct locations contribute to the excitation of mouse visual cortical neurons during visual stimulation. Integration of thalamic inputs with the appropriate spatiotemporal offset provides cortical neurons with the primordial bias for direction selectivity. These data show how cortical neurons selectively combine the spatiotemporal response diversity of thalamic neurons to extract fundamental features of the visual world.
Wednesday, February 14th, 2018
My team's research explores how neurons integrate information and how cooperative activity gives rise to high-level brain function and dysfunction. In my presentation I'll discuss our approach which is best described by the triptych “signals, systems and psyche”: by detailed comprehension of signals elicited by the brain, we seek to understand computation during a behavioral task (systems) and how such processing gives rise to a high-level, behavioral readout (psyche). To this end we use a combination of methods such as large-scale biophysically detailed simulations as well as slice and in vivo electrophysiology.
Neural dynamics represent the hard-to-interpret substrate of circuit computations. Advances in large-scale recordings have highlighted the sheer spatiotemporal complexity of circuit dynamics, portraying in detail the difficulty of interpreting such dynamics and relating it to computation. Considering preparatory activity in a delayed response task we utilized neural recordings performed simultaneously with optogenetic perturbations to probe circuit dynamics. First, we revealed a surprising robustness in the detailed evolution of certain dimensions of the population activity, beyond what was thought to be the case experimentally and theoretically. Second, the robust dimension in activity space carries nearly all of the decodable behavioral information whereas other non-robust dimensions contained nearly no decodable information, consistent with the circuit being setup to make informative dimensions stiff, i.e., resistive to perturbations, leaving uninformative dimensions sloppy, i.e., sensitive to perturbations. This constitutes important evidence for earlier theoretical proposals of dimension specific computation since the circuit is built to selectively stabilize the computationally significant dimensions. Finally, we show that this robustness can be achieved by a modular organization of circuitry, whereby modules whose dynamics normally evolve independently can correct each other’s dynamics when an individual module is perturbed, a common design feature in robust systems engineering.
I will present a new method to control the activity of neural ensembles in the intact brain with holographic multiphoton optogenetics. We combined 3D light sculpting with newly engineering microbial opsins, to permit the cellular resolution, millisecond timing control of user-defined subsets of neurons in the awake mouse brain. This experimental approach should allow researchers to test fundamental theories of neural computation and coding in the brain.
Thursday, February 15th, 2018
A hallmark of sensory cortical circuit function is the reliable expression of selective neuronal responses that enable the encoding of specific stimulus features. In principle, a variety of factors may contribute to a cortical neuron’s response selectivity including the tuning and strength of excitatory and inhibitory synaptic inputs, dendritic nonlinearities, and spike threshold. But how these factors act in combination to produce response selectivity remains poorly understood. Here we employ a combination of techniques including in vivo whole-cell recording, synaptic and cellular resolution in vivo 2-photon calcium imaging, and GABAergic-selective optogenetic manipulation to dissect the factors contributing to direction selective responses of layer 2/3 neurons in ferret visual cortex. Two-photon calcium imaging of dendritic spines revealed that each neuron receives a mixture of excitatory synaptic inputs selective for the somatic preferred or null direction of motion. In contrast, in vivo whole-cell patch clamp recordings revealed a striking degree of direction selectivity in subthreshold membrane potential responses that was significantly correlated with the degree of direction selectivity evident in the neuron’s spiking behavior. Several lines of evidence including conductance measurements suggest that inhibition preferentially suppresses responses to the null direction of motion. Consistent with this idea, optogenetic inactivation of GABAergic neurons in layer 2/3 preferentially enhanced responses to the null direction of motion, causing a reduction in membrane potential direction selectivity. Furthermore, using a new technique to optogenetically map connectivity from inhibitory onto excitatory neurons in layer 2/3 in vivo, we find that layer 2/3 inhibitory neurons make long-range, intercolumnar projections to excitatory neurons that prefer the opposite direction of motion. We conclude that intracortical inhibition exerts a major influence on the degree of direction selectivity in layer 2/3 of ferret visual cortex by preferentially suppressing responses to the null direction of motion.
"This talk explores the possibility that horizontal intra-cortical connections contribute to the dynamic emergence of facilitation and predictions in the primary visual cortex (V1) of cat and non-human primates (NHP)). Contextual long-range interactions, studied at the spiking level, are generally thought to depend on cortico-cortical feedback (Li et al., 2006; Gilbert and Li, 2013) and attention (Lamme & Roelfsema 2000). In contrast, the contribution of intra-V1 mechanisms such as recurrent amplification and lateral propagation (review in Frégnac, 2002 and Frégnac and Bathellier, 2015) is less known.
Experimental work in my lab (UNIC) has been focusing on the dynamic processes based on lateral propagation in V1, through which spatio-temporal inferences (continuous movement or apparent motion sequences) may be generated and facilitate predictive responses along a trajectory (""filling-in""). The stimulation paradigms that we have designed interleave a variety of apparent or continuous animations of local oriented stimuli along trajectories, forming predictable global patterns according to their local to global configuration, and their specific spatial and temporal properties. We have measured the cortical dynamics evoked at two scales of neuronal integration, from micro- (intracellular, SUA) to meso-scopic levels (dense laminar electrode (MEA) and voltage sensitive dye imaging (VSDI)) in the anesthetized cat.
The picture drawn from our intracellular observations is that synaptic activity in cat V1 reveals a built-in bias in the structural organization in the absence of attention-related feedback from higher cortical areas (Gerard-Mercier et al, 2016). This neural process could be related to perceptual biases expressed in NHPs and humans. Taken together with findings obtained in collaboration with the group of Frederic Chavane (INT, Marseille), and based on LFP recordings (Benvenuti et al, submitted), our results suggest, surprisingly, that distinct neural mechanisms - implicating horizontal connectivity - may operate along orthogonal configurations of orientation and direction features and activate differently V1 receptive fields (RF): on one side, surround facilitatory modulation along the orientation preference axis; on the other, anticipatory responses for direction perpendicular to orientation, i.e. along the RF width axis.
Each set of data (still in progress) points to two specific roles of horizontal propagation in V1, with possibly different spatial anisotropy profiles, different spatial scales and different speed selectivities. The first one concerns collinear-biased propagation of iso-orientation preference at high speed (0.1-0.3 m/s; 100°/sec (monkey) -250°/sec (cat)) and fits the general concept of the ""perceptual"" association field” (Field et al, 1993; Li et al, 2006; Gerard-Mercier et al, 2016). The second one concerns isotropic horizontal propagation (10-30°/sec (cat and monkey)) and the directional build-up of anticipatory activity facilitating integration of a moving object along a trajectory (Benevenuti et al, submitted). This propagation process could operate at lower speed than the orientation selective component, and most likely imply a cascade of shorter-range horizontal interactions triggered by a sequence of feedforward inputs. Its synaptic correlates are still unknown.
At a more conceptual level, our working hypothesis posits that horizontal connectivity participates to the propagation of a network-based belief, resulting in some kind of ""prediction"" (filling-in) or ""anticipatory"" build-up process travelling through the V1 network. This view may be compared - and why not opposed - to the classical ""predictive coding"" schema (Rao and Ballard, 1999) where what is transmitted laterally or through feedback onto V1 is the error signal itself (error from the expectation generated in higher cortical areas) rather than the prediction inferred from the global perceptual information. When the contrast is high, the lateral broadcast of the peripheral context leads to a suppression of redundancy of activity in V1 (Martin and von der Heydt, 2015). In contrast when the feedforward signal is weak (low contrast or stimulus absence in ""illusory contour"") the contextual information boosts the gain of feedforward-related activity (Fregnac et al, 1996). Future project development focuses on the possible dependency of the balance between horizontal connectivity and feedback from higher cortical areas on species-specific V1 architecture features and cognitive behavior.
This work, in its initial phase, has been supported by the ramp-up phase of the Human Brain Project (HBP-SP3), and the Marie-Curie Fellowship program. It is currently funded by the CNRS and the French Agence Nationale de la Recherche (ANR: Horizontal-V1).
Benvenuti, G., Chemla, S., Boonman, A., Masson, GS and Chavane F (submitted). Anticipatory responses along motion trajectories in awake monkey V1.
Chavane, F., Sharon, D., Jancke, D., Marre, O., Frégnac, Y., & Grinvald, A. (2011). Lateral Spread of Orientation Selectivity in V1 is Controlled by Intracortical Cooperativity. Frontiers in Systems Neuroscience, 5(4): 1-26.
Field, D.J., Hayes, A., and Hess, R.F. (1993). Contour integration by the human visual system: evidence for a local “association field.” Vision Research 33, 173–193.
Frégnac, Y. (2012). Reading out the synaptic echoes of low level perception in V1. Lecture Notes in Computer Science. 7583: 486-495
Frégnac, Y. and Bathellier, B. (2015). Cortical correlates of low-level perception : from neural circuits to percepts. Neuron. 88(1): 110-126.Fregnac et al 1996
Frégnac, Y., Bringuier, V. & Chavane, F. (1996) Synaptic integration fields and associative plasticity of visual cortical cells in vivo. Journal of Physiology (Paris) 90:367–372
Gerard-Mercier, F., Pananceau, M., Carelli, P., Troncoso, X. and Frégnac, Y. (2016). Synaptic correlates of low-level perception in V1. The Journal of Neuroscience. 36(14) : 3925-3942.
Gilbert, C. D., & Li, W. (2013). Top-down influences on visual processing. Nature Reviews Neuroscience, 14(5), 350–363. Lamme and Roeflsema
Lamme, V. A., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences, 23(11), 571–579.
Li, W., Piëch, V. & Gilbert, C.D. (2006). Contour saliency in primary visual cortex. Neuron 50:951–962.
Martin, A.B. & von der Heydt, R. (2015). Spike Synchrony Reveals Emergence of Proto-Objects in Visual Cortex. The Journal of Neuroscience 35: 6860-6870.
Muller, L., Reynaud, A., Chavane, F., & Destexhe, A. (2014). The stimulus-evoked population response in visual cortex of awake monkey is a propagating wave. Nature Communications, 5(3675).
Rao, R.P. and Ballard, D.H. (1999). Predictive coding in the visual cortex : a functional interpretation of some extra-classical receptive-field effects. Nature Neuroscience 2(1): 79-87."
The foveal region of V1 vastly over-represents not just the visual field, but also retinal ganglion cells, in comparison to parafoveal or peripheral V1. This implies that there are some fairly unique visual computations happening in the fovea that don't happen in other parts of V1, but what these are is still largely unknown. Here I shall present one hypothesis based on the idea that foveal V1 neurons are super-resolving the information provided by retinal ganglion cells by exploiting the eye's own drift motions. I shall describe the neural circuitry needed to allow V1 to build a more complete and detailed representation of visual information than is provided statically by the cones.
With Alex Anderson, Kavitha Ratnam and Austin Roorda.
No abstract available.
I will discuss our recent work on developing large-scale modeling of mammalian neocortex, based on mesoscopic directed- and weighted inter-areal connectivity data (Wang and Kennedy 2016). By taking into account a gradient of circuit property across cortical areas, the model naturally gives rise to a hierarchy of timescales (Chaudhuri et al. 2015), and captures interactions of bottom-up and top-down processes mediated by layer-dependent connections (Mejias et al. 2016). Moreover, in a complex brain system, routing of information between areas must be flexibly gated according to behavioral demands. We propose such a gating mechanism with a disinhibitory circuit motif implemented by three subtypes of (PV+, SOM+ and VIP+) inhibitory neurons, the latter’s relative distribution varies markedly across areas in the mammalian cortex (Kim et al. 2017). This model is now being used to investigate distributed working memory representation in a large-scale brain system. Circuit modeling across levels represents a promising approach to elucidate high-dimensional neural dynamics and cognitive functions of the global brain, as well as their deficits associated with psychiatric disorders.