Search

Intrinsic neuronal variability significantly limits information encoding in the primary visual cortex (V1). However, under certain conditions, neurons can respond reliably with highly precise responses to the same visual stimuli from trial to trial. This suggests that there exists intrinsic neural circuit mechanisms that dynamically modulate the intertrial variability of visual cortical neurons. Here, we sought to elucidate the role of different inhibitory interneurons (INs) in reliable coding in mouse V1. To study the interactions between somatostatin-expressing interneurons (SST-INs) and parvalbumin-expressing interneurons (PV-INs), we used a dual-color calcium imaging technique that allowed us to simultaneously monitor these two neural ensembles while awake mice, of both sexes, passively viewed natural movies. SST neurons were more active during epochs of reliable pyramidal neuron firing, whereas PV neurons were more active during epochs of unreliable firing. SST neuron activity lagged that of PV neurons, consistent with a feedback inhibitory SST→PV circuit. To dissect the role of this circuit in pyramidal neuron activity, we used temporally limited optogenetic activation and inactivation of SST and PV interneurons during periods of reliable and unreliable pyramidal cell firing. Transient firing of SST neurons increased pyramidal neuron reliability by actively suppressing PV neurons, a proposal that was supported by a rate-based model of V1 neurons. These results identify a cooperative functional role for the SST→PV circuit in modulating the reliability of pyramidal neuron activity.SIGNIFICANCE STATEMENTCortical neurons often respond to identical sensory stimuli with large variability. However, under certain conditions, the same neurons can also respond highly reliably. The circuit mechanisms that contribute to this modulation remain unknown. Here, we used novel dual-wavelength calcium imaging and temporally selective optical perturbation to identify an inhibitory neural circuit in visual cortex that can modulate the reliability of pyramidal neurons to naturalistic visual stimuli. Our results, supported by computational models, suggest that somatostatin interneurons increase pyramidal neuron reliability by suppressing parvalbumin interneurons via the inhibitory SST→PV circuit. These findings reveal a novel role of the SST→PV circuit in modulating the fidelity of neural coding critical for visual perception.

The classical concepts of reinforcement learning in the mammalian brain focus on dopamine release in the basal ganglia as the neural substrate of reward prediction errors, which drive plasticity in striatal and cortico-striatal synapses to maximize the expected aggregate future reward. This temporal difference framework, however, even when augmented with deep credit assignment, does not fully capture higher-order processes such as the influence of goal representations, planning based on learned internal models, and hierarchical decision-making implemented by diverse neocortical areas. Candidate functions for such neocortical contributions to reinforcement learning are increasingly being considered in artificial intelligence algorithms. Here, we review recent experimental neurophysiological findings focusing on the orbitofrontal cortex, a key higher-order association cortex, and highlight emerging concepts that emphasize the role of the neocortex in reward-driven computation, in addition to its role as an input to striatal structures. In this framework, reward drives plasticity in various neocortical regions, implementing multiple distinct reinforcement learning algorithms.

The uptake of glutamate by astrocytes actively shapes synaptic transmission, however its role in the development and plasticity of neuronal circuits remains poorly understood. The astrocytic glutamate transporter, GLT1 is the predominant source of glutamate clearance in the adult mouse cortex. Here, we examined the structural and functional development of the visual cortex in GLT1 heterozygous (HET) mice using two-photon microscopy, immunohistochemistry and slice electrophysiology. We find that though eye-specific thalamic axonal segregation is intact, binocular refinement in the primary visual cortex is disrupted. Eye-specific responses to visual stimuli in GLT1 HET mice show altered binocular matching, with abnormally high responses to ipsilateral compared to contralateral eye stimulation and a greater mismatch between preferred orientation selectivity of ipsilateral and contralateral eye responses. Furthermore, the balance of excitation and inhibition in cortical circuits is dysregulated with an increase in somatostatin positive interneurons, decrease in parvalbumin positive interneurons, and increase in dendritic spine density in the basal dendrites of layer 2/3 excitatory neurons. Monocular deprivation induces atypical ocular dominance plasticity in GLT1 HET mice, with an unusual depression of ipsilateral open eye responses; however, this change in ipsilateral responses correlates well with an upregulation of GLT1 protein following monocular deprivation. These results demonstrate that a key function of astrocytic GLT1 function during development is the experience-dependent refinement of ipsilateral eye inputs relative to contralateral eye inputs in visual cortex.SIGNIFICANCEWe show that astrocytic glutamate uptake via the transporter GLT1 is necessary for activity-dependent regulation of cortical inputs. Dysregulation of GLT1 expression and function leads to a disruption of binocular refinement and matching in visual cortex. Inputs from the ipsilateral eye are stronger, and monocular deprivation, which upregulates GLT1 expression in a homeostatic fashion, causes a paradoxical reduction of ipsilateral, non-deprived eye, responses. These results provide new evidence for the importance of glutamate transport in cortical development, function, and plasticity.

The prefrontal cortex is vital for a range of cognitive processes, including working memory, attention, and decision-making. Notably, its absence impairs the performance of tasks requiring the maintenance of information through a delay period. In this paper, we formulate a rodent task—which requires maintenance of delay-period activity—as a Markov decision process and treat optimal task performance as an (active) inference problem. We simulate the behavior of a Bayes optimal mouse presented with 1 of 2 cues that instructs the selection of concurrent visual and auditory targets on a trial-by-trial basis. Formulating inference as message passing, we reproduce features of neuronal coupling within and between prefrontal regions engaged by this task. We focus on the micro-circuitry that underwrites delay-period activity and relate it to functional specialization within the prefrontal cortex in primates. Finally, we simulate the electrophysiological correlates of inference and demonstrate the consequences of lesions to each part of our in silico prefrontal cortex. In brief, this formulation suggests that recurrent excitatory connections—which support persistent neuronal activity—encode beliefs about transition probabilities over time. We argue that attentional modulation can be understood as the contextualization of sensory input by these persistent beliefs.

Cognitive maps are mental representations of spatial and conceptual relationships in an environment, and are critical for flexible behavior. To form these abstract maps, the hippocampus has to learn to separate or merge aliased observations appropriately in different contexts in a manner that enables generalization and efficient planning. Here we propose a specific higher-order graph structure, clone-structured cognitive graph (CSCG), which forms clones of an observation for different contexts as a representation that addresses these problems. CSCGs can be learned efficiently using a probabilistic sequence model that is inherently robust to uncertainty. We show that CSCGs can explain a variety of cognitive map phenomena such as discovering spatial relations from aliased sensations, transitive inference between disjoint episodes, and formation of transferable schemas. Learning different clones for different contexts explains the emergence of splitter cells observed in maze navigation and event-specific responses in lap-running experiments. Moreover, learning and inference dynamics of CSCGs offer a coherent explanation for disparate place cell remapping phenomena. By lifting aliased observations into a hidden space, CSCGs reveal latent modularity useful for hierarchical abstraction and planning. Altogether, CSCG provides a simple unifying framework for understanding hippocampal function, and could be a pathway for forming relational abstractions in artificial intelligence., Higher-order sequence learning using a structured graph representation - clone-structured cognitive graphs (CSCG) – can explain how the hippocampus learns cognitive maps. CSCG provides novel explanations for transferable schemas and transitive inference in the hippocampus, and for how place cells, splitter cells, lap-cells and a variety of phenomena emerge from the same set of fundamental principles.

Human visual systems can parse a scene composed of novel objects and infer their surfaces and occlusion relationships without relying on object-specific shapes or textures. Perceptual grouping can bind together spatially disjoint entities to unite them as one object even when the object is entirely novel, and bind other perceptual properties like color and texture to that object using object-based attention. Border-ownership assignment, the assignment of perceived occlusion boundaries to specific perceived surfaces, is an intermediate representation in the mammalian visual system that facilitates this perceptual grouping. Since objects in a scene can be entirely novel, inferring border ownership requires integrating global figural information, while dynamically postulating what the figure is, a chicken-and egg process that is complicated further by missing or conflicting local evidence regarding the presence of boundaries. Based on neuroscience observations, we introduce a model -- the cloned Markov random field (CMRF)-- that can learn attention-controllable representations for border-ownership. Higher-order contour representations that distinguish border-ownerships emerge as part of learning in this model. When tested with a cluttered scene of novel 2D objects with noisy contour-only evidence, the CMRF model is able to perceptually group them, despite clutter and missing edges. Moreover, the CMRF is able to use occlusion cues to bind disconnected surface elements of novel objects into coherent objects, and able to use top-down attention to assign border ownership to overlapping objects. Our work is a step towards dynamic binding of surface elements into objects, a capability that is crucial for intelligent agents to interact with the world and to form entity-based abstractions.

Interactions between the prefrontal cortex (PFC) and mediodorsal thalamus are critical for cognitive flexibility, yet the underlying computations are unknown. To investigate frontothalamic substrates of cognitive flexibility, we developed a behavioral task in which mice switched between different sets of learned cues that guided attention toward either visual or auditory targets. We found that PFC responses reflected both the individual cues and their meaning as task rules, indicating a hierarchical cue-to-rule transformation. Conversely, mediodorsal thalamus responses reflected the statistical regularity of cue presentation and were required for switching between such experimentally specified cueing contexts. A subset of these thalamic responses sustained context-relevant PFC representations, while another suppressed the context-irrelevant ones. Through modeling and experimental validation, we find that thalamic-mediated suppression may not only reduce PFC representational interference but could also preserve unused cortical traces for future use. Overall, our study provides a computational foundation for thalamic engagement in cognitive flexibility.

Cognitive maps are mental representations of spatial and conceptual relationships in an environment. These maps are critical for flexible behavior as they permit us to navigate vicariously, but their underlying representation learning mechanisms are still unknown. To form these abstract maps, hippocampus has to learn to separate or merge aliased observations appropriately in different contexts in a manner that enables generalization, efficient planning, and handling of uncertainty. Here we introduce a specific higher-order graph structure – clone-structured cognitive graph (CSCG) – which forms different clones of an observation for different contexts as a representation that addresses these problems. CSCGs can be learned efficiently using a novel probabilistic sequence model that is inherently robust to uncertainty. We show that CSCGs can explain a variety cognitive map phenomena such as discovering spatial relations from an aliased sensory stream, transitive inference between disjoint episodes of experiences, formation of transferable structural knowledge, and shortcut-finding in novel environments. By learning different clones for different contexts, CSCGs explain the emergence of splitter cells and route-specific encoding of place cells observed in maze navigation, and event-specific graded representations observed in lap-running experiments. Moreover, learning and inference dynamics of CSCGs offer a coherent explanation for a variety of place cell remapping phenomena. By lifting the aliased observations into a hidden space, CSCGs reveal latent modularity that is then used for hierarchical abstraction and planning. Altogether, learning and inference using a CSCG provides a simple unifying framework for understanding hippocampal function, and could be a pathway for forming relational abstractions in artificial intelligence.

Sequence learning is a vital cognitive function and has been observed in numerous brain areas. Discovering the algorithms underlying sequence learning has been a major endeavour in both neuroscience and machine learning. In earlier work we showed that by constraining the sparsity of the emission matrix of a Hidden Markov Model (HMM) in a biologically-plausible manner we are able to efficiently learn higher-order temporal dependencies and recognize contexts in noisy signals. The central basis of our model, referred to as the Cloned HMM (CHMM), is the observation that cortical neurons sharing the same receptive field properties can learn to represent unique incidences of bottom-up information within different temporal contexts. CHMMs can efficiently learn higher-order temporal dependencies, recognize long-range contexts and, unlike recurrent neural networks, are able to natively handle uncertainty. In this paper we introduce a biologically plausible CHMM learning algorithm, memorize-generalize, that can rapidly memorize sequences as they are encountered, and gradually generalize as more data is accumulated. We demonstrate that CHMMs trained with the memorize-generalize algorithm can model long-range structure in bird songs with only a slight degradation in performance compared to expectation-maximization, while still outperforming other representations.

Hippocampus encodes cognitive maps that support episodic memories, navigation, and planning. Under-standing the commonality among those maps as well as how those maps are structured, learned from experience, and used for inference and planning is an interesting but unsolved problem. We propose higher-order graphs as the general principle and present, as a plausible model, a cloned hidden Markov model (HMM) that can learn these graphs efficiently from experienced sequences. In our experiments, we use the cloned HMM for learning spatial and abstract representations. We show that inference and planning in the learned CHMM encapsulates many of the key properties of hippocampal cells observed in rodents and humans. Cloned HMM thus provides a new frame-work for understanding hippocampal function.

The thalamus has long been suspected to have an important role in cognition, yet recent theories have favored a more corticocentric view. According to this view, the thalamus is an excitatory feedforward relay to or between cortical regions, and cognitively relevant computations are exclusively cortical. Here, we review anatomical, physiological, and behavioral studies along evolutionary and theoretical dimensions, arguing for essential and unique thalamic computations in cognition. Considering their architectural features as well as their ability to initiate, sustain, and switch cortical activity, thalamic circuits appear uniquely suited for computing contextual signals that rapidly reconfigure task-relevant cortical representations. We introduce a framework that formalizes this notion, show its consistency with several findings, and discuss its prediction of thalamic roles in perceptual inference and behavioral flexibility. Overall, our framework emphasizes an expanded view of the thalamus in cognitive computations and provides a roadmap to test several of its theoretical and experimental predictions.

Cortical neurons often respond to identical sensory stimuli with large variability. However, under certain conditions, the same neurons can also respond highly reliably. The circuit mechanisms that contribute to this modulation, and their influence on behavior remains unknown. Here we used novel double transgenic mice, dual-wavelength calcium imaging and temporally selective optical perturbation to identify an inhibitory neural circuit in visual cortex that can modulate the reliability of pyramidal neurons to naturalistic visual stimuli. Our results, supported by computational models, suggest that somatostatin interneurons (SST-INs) increase pyramidal neuron reliability by suppressing parvalbumin interneurons (PV-INs) via the inhibitory SST→PV circuit. Using a novel movie classification task, we further show that, by reducing variability, activating SST-INs can improve the ability of mice to discriminate between ambiguous stimuli. Together, these findings reveal a novel role of the SST→PV circuit in modulating the fidelity of neural coding critical for visual perception.

Intrinsic neuronal variability significantly limits information encoding in the primary visual cortex (V1). Certain stimuli can suppress this intertrial variability to increase the reliability of neuronal responses. In particular, responses to natural scenes, which have broadband spatiotemporal statistics, are more reliable than responses to stimuli such as gratings. However, very little is known about which stimulus statistics modulate reliable coding and how this occurs at the neural ensemble level. Here, we sought to elucidate the role that spatial correlations in natural scenes play in reliable coding. We developed a novel noise-masking method to systematically alter spatial correlations in natural movies, without altering their edge structure. Using high-speed two-photon calcium imaging in vivo, we found that responses in mouse V1 were much less reliable at both the single neuron and population level when spatial correlations were removed from the image. This change in reliability was due to a reorganization of between-neuron correlations. Strongly correlated neurons formed ensembles that reliably and accurately encoded visual stimuli, whereas reducing spatial correlations reduced the activation of these ensembles, leading to an unreliable code. Together with an ensemble-specific normalization model, these results suggest that the coordinated activation of specific subsets of neurons underlies the reliable coding of natural scenes., National Institutes of Health (U.S.) (Grant EY007023), National Institutes of Health (U.S.) (Grant NS090473), National Science Foundation (U.S.) (Grant EF1451125), Howard Hughes Medical Institute (International Student Research Fellowship)

Spike timing-dependent plasticity (STDP) is a temporally asymmetric form of Hebbian plasticity in which the sign and magnitude of synaptic strength is determined by the precise timing between pre- and postsynaptic spikes. Postsynaptic NMDA receptors (NMDARs) play an important role in STDP due to their characteristic magnesium ion (Mg2+) gating process [1]. The Mg2+ gate is opened by near coincident presynaptic glutamate release and postsynaptic spiking. Contrary to previous belief, this gating process is not instantaneous but has distinct fast and slow time constants, which might influence the time window over which spike-EPSP coincidence occurs [2,3]. The aim of this study is to investigate how local activity in dendrites alters NMDAR kinetics and in turn controls the time window over which STDP occurs. Using a detailed biophysical model of an NR2A subunit, we find the NMDAR current to be temporally asymmetric with a faster block than unblock process. This is due to the voltage-dependent fast and slow time constants of Mg2+ gating. These time constants were on the order of 0.8 ms and 5 ms respectively and changed nonlinearly with membrane potential. By varying both amplitude and duration of a model action potential (AP), we find that dendritic Ca2+ spikes cause Mg2+ unblock to be dominated by its fast time constant. Conversely, somatic Na+ spikes elicit a slower unblocking. Since backpropagating APs (BAPs) attenuate along the dendritic arbor, our results suggest that NMDAR kinetics vary depending on synapse location. Next, we incorporated this biophysical model into a compartmentalized model of a neocortical pyramidal cell in order to investigate the effect that dendritic inhibition and synapse location has on STDP. NMDAR kinetics is significantly different in the basal and apical compartments, which gives rise to a heterogeneous set of STDP curves along the length of the dendrite. Distal synapses had a broader potentiation window than apical synapses, suggesting that distal branches are less selective to correlated inputs. This synapse-location dependent STDP learning rule agrees with previous findings [4]. Dendritic inhibition plays an important role in shaping the computations performed by pyramidal neurons in the neocortex. However, the role that inhibition plays in controlling synaptic plasticity is less clear. By incorporating inhibitory synapses in our model, we find that varying the timing of inhibition relative to the EPSP shifted the peak of the STDP curve, resulting in a reversal of the temporal order of depression and potentiation. Inhibition altered NMDAR kinetics by increasing its blocking rate. As a result, changing the balance between excitation and inhibition in the dendritic branch results in a sliding threshold between LTP and LTD. In summary, we show that NMDAR kinetics can be altered in an activity-dependent manner resulting in different STDP learning rules. When taken in context of the BCM theory, our finding implies that dendritic inhibition can dynamically alter the threshold between LTP and LTD by altering the time constants of the NMDAR. Our model provides a novel framework for understanding how STDP in different dendritic compartments selects inputs alters information processing in individual neurons.

Natural scenes, although complex in appearance, contain numerous statistical regularities. For instance, owing to surfaces and textures, neighboring pixels in natural images are strongly correlated in both space and time (Figure 1B). These correlations create strong dependencies between neurons that may limit neural coding efficiency. Thus, to produce an efficient representation, it has been proposed that neural circuits in the early visual system function to remove extraneous stimulus correlations. Although extensively studied in the retina and LGN, the mechanisms of efficient coding in primary visual cortex (V1) remain elusive. The goal of this study is to establish a relationship between population coding in V1 and spatial correlations in natural scenes. Mathematically, the Fourier amplitude spectrum describes second-order correlations between pixels. In contrast, the phase spectrum describes higher-order correlations, which arise due to edges and other salient features. To understand how stimulus correlations influence coding in V1, we developed a novel technique that allowed us to systematically perturb the Fourier amplitude spectrum of natural movies, while keeping the edge structure intact. Using this method, we generated movies with four different levels of spatial correlation (Figure 1A). Visually, edges in the most decorrelated movies (K0 spectrum) appear shaper, while edges in the most strongly correlated movies appear occluded (K2 spectrum). To assess how populations of neurons respond to these movies, we performed high-speed 2-photon calcium imaging in layer 2/3 of mouse V1. This method permitted us to monitor the activity of close to 100 neurons with high spatial and temporal resolution.

Networks in the neocortex serve an important purpose of extracting information from sensory inputs. One appealing theory is the efficient coding hypothesis, which postulates that information about a stimulus is maximized when redundancy between neurons is reduced [1]. This gives rise to the notion of a sparse population code where only neurons that are highly selective to certain stimulus features respond. However, sparse codes are highly sensitivity to noise. An alternative strategy is to use a synchronized code, which is robust to noise but reduces the representational capacity of the network [2,3]. Hence, it is only logical to ask what level of redundancy between neurons maximizes the amount of information contained in a population code. In this study, we investigated how the collective dynamics of a population of neurons represent stimulus-specific information. Our aim was to investigate (1) the conditions required for a network to synchronize and (2) the amount of information contained in a synchronized rate code. By applying nonlinear contraction analysis on a recurrent network with time varying inputs, we found that the degree of synchronization depended critically on the gain of the feedforward projection. Our analysis showed that changing the feedforward gain exploits different symmetries in the network, resulting in different patterns of synchrony and different population codes. We hypothesized that the coding strategy can by optimized by altering this synaptic gain. We tested this prediction computationally using a two-layered network. The receiver layer was a balanced recurrent network of 5000 integrate-and-fire neurons. Input to the network was provided by a population of "visually responsive" neurons, which was modeled using a linear-nonlinear Poisson (LNP) cascade. The LNP neurons were tuned to 16 orientations and projected nonspecifically to 20% of the neurons in the receiver layer. We assumed that the stimulus was a sequence of drifting gratings with random orientations. In response to stimuli, the network displayed transiently synchronized responses. Because similarly tuned LNP neurons projected to different subsets of neurons, the pattern of network activity was different for each stimulus sequence. By gradually modifying the feedforward gain, we were able to alter this pattern of stimulus-specific synchronization, confirming the results from our mathematical analysis. In the high gain condition, neurons became redundant and responded to all stimuli. However, when gain was low, only sparse subsets of neurons were activated by different stimuli. Finally, we used an optimal linear decoder to measure of how well the network encoded the stimuli. We found that classification accuracy varied as a function of the degree of synchronization between neurons. Classification accuracy was highest when approximately 15% of the neurons responded concurrently to the stimulus. In both densely synchronized and sparse codes, the decoder performed below chance. Our results show that stimulus-specific information is maximized only when small ensembles of neurons are synchronized. Our finding that stimulus-dependent synchronization is controlled by the gain of feedforward synapses is unique as it suggests that mechanisms which alter synaptic gain such as attention can alter the coding strategy employed by the neocortex.