Task-induced neural covariability as a signature of approximate Bayesian learning and inference

Perception is often characterized computationally as an inference process in which uncertain or ambiguous sensory inputs are combined with prior expectations. Although behavioral studies have shown that observers can change their prior expectations in the context of a task, robust neural signatures of task-specific priors have been elusive. Here, we analytically derive such signatures under the general assumption that the responses of sensory neurons encode posterior beliefs that combine sensory inputs with task-specific expectations. Specifically, we derive predictions for the task-dependence of correlated neural variability and decision-related signals in sensory neurons. The qualitative aspects of our results are parameter-free and specific to the statistics of each task. The predictions for correlated variability also differ from predictions of classic feedforward models of sensory processing and are therefore a strong test of theories of hierarchical Bayesian inference in the brain. Importantly, we find that Bayesian learning predicts an increase in so-called “differential correlations” as the observer’s internal model learns the stimulus distribution, and the observer’s behavioral performance improves. This stands in contrast to classic feedforward encoding/decoding models of sensory processing, since such correlations are fundamentally information-limiting. We find support for our predictions in data from existing neurophysiological studies across a variety of tasks and brain areas. Finally, we show in simulation how measurements of sensory neural responses can reveal information about a subject’s internal beliefs about the task. Taken together, our results reinterpret task-dependent sources of neural covariability as signatures of Bayesian inference and provide new insights into their cause and their function.     

Optimal Prediction of Moving Sound Source Direction in the Owl

Capturing nature’s statistical structure in behavioral responses is at the core of the ability to function adaptively in the environment. Bayesian statistical inference describes how sensory and prior information can be combined optimally to guide behavior. An outstanding open question of how neural coding supports Bayesian inference includes how sensory cues are optimally integrated over time. Here we address what neural response properties allow a neural system to perform Bayesian prediction, i.e., predicting where a source will be in the near future given sensory information and prior assumptions. The work here shows that the population vector decoder will perform Bayesian prediction when the receptive fields of the neurons encode the target dynamics with shifting receptive fields. We test the model using the system that underlies sound localization in barn owls. Neurons in the owl’s midbrain show shifting receptive fields for moving sources that are consistent with the predictions of the model. We predict that neural populations can be specialized to represent the statistics of dynamic stimuli to allow for a vector read-out of Bayes-optimal predictions.     

Ensembles of Spiking Neurons with Noise Support Optimal Probabilistic Inference in a Dynamically Changing Environment

It has recently been shown that networks of spiking neurons with noise can emulate simple forms of probabilistic inference through “neural sampling”, i.e., by treating spikes as samples from a probability distribution of network states that is encoded in the network. Deficiencies of the existing model are its reliance on single neurons for sampling from each random variable, and the resulting limitation in representing quickly varying probabilistic information. We show that both deficiencies can be overcome by moving to a biologically more realistic encoding of each salient random variable through the stochastic firing activity of an ensemble of neurons. The resulting model demonstrates that networks of spiking neurons with noise can easily track and carry out basic computational operations on rapidly varying probability distributions, such as the odds of getting rewarded for a specific behavior. We demonstrate the viability of this new approach towards neural coding and computation, which makes use of the inherent parallelism of generic neural circuits, by showing that this model can explain experimentally observed firing activity of cortical neurons for a variety of tasks that require rapid temporal integration of sensory information.     

Spike-based population coding and working memory

Compelling behavioral evidence suggests that humans can make optimal decisions despite the uncertainty inherent in perceptual or motor tasks. A key question in neuroscience is how populations of spiking neurons can implement such probabilistic computations. In this article, we develop a comprehensive framework for optimal, spike-based sensory integration and working memory in a dynamic environment. We propose that probability distributions are inferred spike-per-spike in recurrently connected networks of integrate-and-fire neurons. As a result, these networks can combine sensory cues optimally, track the state of a time-varying stimulus and memorize accumulated evidence over periods much longer than the time constant of single neurons. Importantly, we propose that population responses and persistent working memory states represent entire probability distributions and not only single stimulus values. These memories are reflected by sustained, asynchronous patterns of activity which make relevant information available to downstream neurons within their short time window of integration. Model neurons act as predictive encoders, only firing spikes which account for new information that has not yet been signaled. Thus, spike times signal deterministically a prediction error, contrary to rate codes in which spike times are considered to be random samples of an underlying firing rate. As a consequence of this coding scheme, a multitude of spike patterns can reliably encode the same information. This results in weakly correlated, Poisson-like spike trains that are sensitive to initial conditions but robust to even high levels of external neural noise. This spike train variability reproduces the one observed in cortical sensory spike trains, but cannot be equated to noise. On the contrary, it is a consequence of optimal spike-based inference. In contrast, we show that rate-based models perform poorly when implemented with stochastically spiking neurons.     

The boundary cap: a source of neural crest stem cells that generate multiple sensory neuron subtypes

The boundary cap (BC) is a transient neural crest-derived group of cells located at the dorsal root entry zone (DREZ) that have been shown to differentiate into sensory neurons and glia in vivo. We find that when placed in culture, BC cells self-renew, show multipotency in clonal cultures and express neural crest stem cell (NCSCs) markers. Unlike sciatic nerve NCSCs, the BC-NCSC (bNCSCs) generates sensory neurons upon differentiation. The bNCSCs constitute a common source of cells for functionally diverse types of neurons, as a single bNCSC can give rise to several types of nociceptive and thermoreceptive sensory neurons. Our data suggests that BC cells comprise a source of multipotent sensory specified stem cells that persist throughout embryogenesis.     

Sensation in a single neuron pair represses male behavior in hermaphrodites

Pheromones elicit innate sex-specific mating behaviors in many species. We demonstrate that in C.?elegans, male-specific sexual attraction behavior is programmed in both sexes but repressed in hermaphrodites. Repression requires a single sensory neuron pair, the ASIs. To repress attraction in adults, the ASIs must be present, active, and capable of sensing the environment during development. The ASIs release TGF-β, and ASI function can be bypassed by experimental activation of TGF-β signaling. Sexual attraction in derepressed hermaphrodites requires the same sensory neurons as in males. The sexual identity of both these sensory neurons and a distinct subset of interneurons must be male to relieve repression and release attraction. TGF-β may therefore act to change connections between sensory neurons and interneurons during development to engage repression. Thus, sensation in a single sensory neuron pair during development reprograms a common neural circuit from male to female behavior.     

Zebrafish narrowminded suggests a genetic link between formation of neural crest and primary sensory neurons.

In the developing vertebrate nervous system, both neural crest and sensory neurons form at the boundary between non-neural ectoderm and the neural plate. From an in situ hybridization based expression analysis screen, we have identified a novel zebrafish mutation, narrowminded (nrd), which reduces the number of early neural crest cells and eliminates Rohon-Beard (RB) sensory neurons. Mosaic analysis has shown that the mutation acts cell autonomously suggesting that nrd is involved in either the reception or interpretation of signals at the lateral neural plate boundary. Characterization of the mutant phenotype indicates that nrd is required for a primary wave of neural crest cell formation during which progenitors generate both RB sensory neurons and neural crest cells. Moreover, the early deficit in neural crest cells in nrd homozygotes is compensated later in development. Thus, we propose that a later wave can compensate for the loss of early neural crest cells but, interestingly, not the RB sensory neurons. We discuss the implications of these findings for the possibility that RB sensory neurons and neural crest cells share a common evolutionary origin.     

Exploring Memory Representations with Activity-Based Genetics

The brain is thought to represent specific memories through the activity of sparse and distributed neural ensembles. In this review, we examine the use of immediate early genes (IEGs), genes that are induced by neural activity, to specifically identify and genetically modify neurons activated naturally by environmental experience. Recent studies using this approach have identified cellular and molecular changes specific to neurons activated during learning relative to their inactive neighbors. By using opto- and chemogenetic regulators of neural activity, the neurons naturally recruited during learning can be artificially reactivated to directly test their role in coding external information. In contextual fear conditioning, artificial reactivation of learning-induced neural ensembles in the hippocampus or neocortex can substitute for the context itself. That is, artificial stimulation of these neurons can apparently cause the animals to “think” they are in the context. This represents a powerful approach to testing the principles by which the brain codes for the external world and how these circuits are modified with learning.A central feature of nervous systems is that, to function properly, specific neurons must become active in response to specific stimuli. The nature of this selective activation and its modification with experience is the focus of much neuroscience research, ranging from studies of sensory processing in experimental animals to disorders of thought such as schizophrenia in humans. The central dogma of neuroscience is that perceptions, memories, thoughts, and higher mental functions arise from the pattern and timing of the activity in neural ensembles in specific parts of the brain at specific points in time. Until quite recently, the investigation of these “circuit”-based questions has primarily been limited to observational techniques, such as single unit recording, functional magnetic resonance imagery (fMRI), and calcium imaging, to document the patterns of neural activity evoked by sensory experience or even complex psychological contingencies in human fMRI studies. These techniques have been enormously successful and created a framework for understanding information processing in the brain. For example, recordings in the visual system have indicated that, in the primary visual cortex, neurons are tuned to the orientation of linear stimuli (Hubel and Wiesel 1962). In contrast, neurons in higher brain areas can respond to discrete items. The most striking example of this specificity comes from in vivo recording in the human medial temporal lobe in which single units have been identified that respond to photos of the actress Halle Berry as well as her written name (Quiroga et al. 2005). This highly selective tuning of neural activity is suggestive of function, but how can this be directly tested? What would be the effect of stimulating just this rare population of neurons, a memory of the actress, a sensory illusion of her image? How does this type of specific firing arise? Do these neurons differ from their nonresponsive neighbors in terms of biochemistry, cell biology, or connectivity? Do they undergo molecular alterations when new information is learned about this individual and are these changes required for the learning? These types of questions have recently become accessible to study in mice through the use of activity-based genetic manipulation, in which neurons that are activated by a specific sensory stimulus can be altered to express any gene of experimental interest. These studies and approaches will be the focus of this work.     

Genes, lineages and the neural crest: a speculative review

Sensory and sympathetic neurons are generated from the trunk neural crest. The prevailing view has been that these two classes of neurons are derived from a common neural crest-derived progenitor that chooses between neuronal fates only after migrating to sites of peripheral ganglion formation. Here I reconsider this view in the light of new molecular and genetic data on the differentiation of sensory and autonomic neurons. These data raise several paradoxes when taken in the context of classical studies of the timing and spatial patterning of sensory and autonomic ganglion formation. These paradoxes can be most easily resolved by assuming that the restriction of neural crest cells to either sensory or autonomic lineages occurs at a very early stage, either before and/or shortly after they exit the neural tube.     

Sensory Adaptation and Short Term Plasticity as Bayesian Correction for a Changing Brain

Neurons in the sensory system exhibit changes in excitability that unfold over many time scales. These fluctuations produce noise and could potentially lead to perceptual errors. However, to prevent such errors, postsynaptic neurons and synapses can adapt and counteract changes in the excitability of presynaptic neurons. Here we ask how neurons could optimally adapt to minimize the influence of changing presynaptic neural properties on their outputs. The resulting model, based on Bayesian inference, explains a range of physiological results from experiments which have measured the overall properties and detailed time-course of sensory tuning curve adaptation in the early visual cortex. We show how several experimentally measured short term plasticity phenomena can be understood as near-optimal solutions to this adaptation problem. This framework provides a link between high level computational problems, the properties of cortical neurons, and synaptic physiology.     

Temporal encoding in a nervous system

We examined the extent to which temporal encoding may be implemented by single neurons in the cercal sensory system of the house cricket Acheta domesticus. We found that these neurons exhibit a greater-than-expected coding capacity, due in part to an increased precision in brief patterns of action potentials. We developed linear and non-linear models for decoding the activity of these neurons. We found that the stimuli associated with short-interval patterns of spikes (ISIs of 8 ms or less) could be predicted better by second-order models as compared to linear models. Finally, we characterized the difference between these linear and second-order models in a low-dimensional subspace, and showed that modification of the linear models along only a few dimensions improved their predictive power to parity with the second order models. Together these results show that single neurons are capable of using temporal patterns of spikes as fundamental symbols in their neural code, and that they communicate specific stimulus distributions to subsequent neural structures.     

Delta/Notch signaling promotes formation of zebrafish neural crest by repressing Neurogenin 1 function

In zebrafish, cells at the lateral edge of the neural plate become Rohon-Beard primary sensory neurons or neural crest. Delta/Notch signaling is required for neural crest formation. ngn1 is expressed in primary neurons; inhibiting Ngn1 activity prevents Rohon-Beard cell formation but not formation of other primary neurons. Reducing Ngn1 activity in embryos lacking Delta/Notch signaling restores neural crest formation, indicating Delta/Notch signaling inhibits neurogenesis without actively promoting neural crest. Ngn1 activity is also required for later development of dorsal root ganglion sensory neurons; however, Rohon-Beard neurons and dorsal root ganglion neurons are not necessarily derived from the same precursor cell. We propose that temporally distinct episodes of Ngn1 activity in the same precursor population specify these two different types of sensory neurons.     

Pax3-expressing trigeminal placode cells can localize to trunk neural crest sites but are committed to a cutaneous sensory neuron fate

The cutaneous sensory neurons of the ophthalmic lobe of the trigeminal ganglion are derived from two embryonic cell populations, the neural crest and the paired ophthalmic trigeminal (opV) placodes. Pax3 is the earliest known marker of opV placode ectoderm in the chick. Pax3 is also expressed transiently by neural crest cells as they emigrate from the neural tube, and it is reexpressed in neural crest cells as they condense to form dorsal root ganglia and certain cranial ganglia, including the trigeminal ganglion. Here, we examined whether Pax3+ opV placode-derived cells behave like Pax3+ neural crest cells when they are grafted into the trunk. Pax3+ quail opV ectoderm cells associate with host neural crest migratory streams and form Pax3+ neurons that populate the dorsal root and sympathetic ganglia and several ectopic sites, including the ventral root. Pax3 expression is subsequently downregulated, and at E8, all opV ectoderm-derived neurons in all locations are large in diameter, and virtually all express TrkB. At least some of these neurons project to the lateral region of the dorsal horn, and peripheral quail neurites are seen in the dermis, suggesting that they are cutaneous sensory neurons. Hence, although they are able to incorporate into neural crest-derived ganglia in the trunk, Pax3+ opV ectoderm cells are committed to forming cutaneous sensory neurons, their normal fate in the trigeminal ganglion. In contrast, Pax3 is not expressed in neural crest-derived neurons in the dorsal root and trigeminal ganglia at any stage, suggesting either that Pax3 is expressed in glial cells or that it is completely downregulated before neuronal differentiation. Since Pax3 is maintained in opV placode-derived neurons for some considerable time after neuronal differentiation, these data suggest that Pax3 may play different roles in opV placode cells and neural crest cells.     

Gene expression profiling of cranial sensory ganglia that transmit food intake stimuli

Peripheral cranial sensory nerves projecting into the oral cavity receive food intake stimuli and transmit sensory signals to the central nervous system. They are derived from four cranial sensory ganglia, trigeminal, geniculate, petrosal, and nodose ganglia, each of which contains multiple kinds of sensory neurons with different cell morphologies and neuronal properties. We investigated the complex properties of these neurons from the viewpoint of gene expression using DNA microarrays. The 498 genes were selected from a total of 8,740 genes as showing tissue-dependent expression on the microarray by hierarchical cluster analysis, in which several genes known to be differentially expressed in cranial sensory ganglia are included. This suggests that DNA microarray cluster analysis revealed a number of characteristic genes for sensory neurons in these ganglia. Among the selected 498 genes, 44 genes are associated with neurotransmission, such as neuropeptides, their receptors, and vesicle transport, and 26 are ion channels regulating membrane potentials. The identification of a number of genes related directly to neural properties indicates that these sensory ganglia contain heterogeneous types of neurons with different neural properties.     

Variability of perceptual multistability: from brain state to individual trait

Few phenomena are as suitable as perceptual multistability to demonstrate that the brain constructively interprets sensory input. Several studies have outlined the neural circuitry involved in generating perceptual inference but only more recently has the individual variability of this inferential process been appreciated. Studies of the interaction of evoked and ongoing neural activity show that inference itself is not merely a stimulus-triggered process but is related to the context of the current brain state into which the processing of external stimulation is embedded. As brain states fluctuate, so does perception of a given sensory input. In multistability, perceptual fluctuation rates are consistent for a given individual but vary considerably between individuals. There has been some evidence for a genetic basis for these individual differences and recent morphometric studies of parietal lobe regions have identified neuroanatomical substrates for individual variability in spontaneous switching behaviour. Moreover, disrupting the function of these latter regions by transcranial magnetic stimulation yields systematic interference effects on switching behaviour, further arguing for a causal role of these regions in perceptual inference. Together, these studies have advanced our understanding of the biological mechanisms by which the brain constructs the contents of consciousness from sensory input.     

Escape behavior elicited by single, channelrhodopsin-2-evoked spikes in zebrafish somatosensory neurons

Somatosensory neurons in teleosts and amphibians are sensitive to thermal, mechanical, or nociceptive stimuli [1, 2]. The two main types of such cells in zebrafish--Rohon-Beard and trigeminal neurons--have served as models for neural development [3-6], but little is known about how they encode tactile stimuli. The hindbrain networks that transduce somatosensory stimuli into a motor output encode information by using very few spikes in a small number of cells [7], but it is unclear whether activity in the primary receptor neurons is similarly efficient. To address this question, we manipulated the activity of zebrafish neurons with the light-activated cation channel, Channelrhodopsin-2 (ChR2) [8, 9]. We found that photoactivation of ChR2 in genetically defined populations of somatosensory neurons triggered escape behaviors in 24-hr-old zebrafish. Electrophysiological recordings from ChR2-positive trigeminal neurons in intact fish revealed that these cells have extremely low rates of spontaneous activity and can be induced to fire by brief pulses of blue light. Using this technique, we find that even a single action potential in a single sensory neuron was at times sufficient to evoke an escape behavior. These results establish ChR2 as a powerful tool for the manipulation of neural activity in zebrafish and reveal a degree of efficiency in coding that has not been found in primary sensory neurons.     

Forward and Backward Inference in Spatial Cognition

This paper shows that the various computations underlying spatial cognition can be implemented using statistical inference in a single probabilistic model. Inference is implemented using a common set of ‘lower-level’ computations involving forward and backward inference over time. For example, to estimate where you are in a known environment, forward inference is used to optimally combine location estimates from path integration with those from sensory input. To decide which way to turn to reach a goal, forward inference is used to compute the likelihood of reaching that goal under each option. To work out which environment you are in, forward inference is used to compute the likelihood of sensory observations under the different hypotheses. For reaching sensory goals that require a chaining together of decisions, forward inference can be used to compute a state trajectory that will lead to that goal, and backward inference to refine the route and estimate control signals that produce the required trajectory. We propose that these computations are reflected in recent findings of pattern replay in the mammalian brain. Specifically, that theta sequences reflect decision making, theta flickering reflects model selection, and remote replay reflects route and motor planning. We also propose a mapping of the above computational processes onto lateral and medial entorhinal cortex and hippocampus.     

Expression of muscarinic acetylcholine receptors and choline acetyltransferase enzyme in cultured antennal sensory neurons and non‐neural cells of the developing moth Manduca sexta

Antennal sensory neurons of Manduca sexta emerge from epidermal cells that also give rise to sheath cells surrounding the peripheral parts of the neurons and to glial cells that enwrap the sensory axons in the antennal nerve. Reciprocal interactions between sensory neurons and glial cells are believed to aid in axon growth and guidance, but the exact nature of these interactions is not known. We investigated the possibility of cholinergic interactions in this process by locating muscarinic acetylcholine receptors (mAChRs) and choline acetyltransferase (ChAT) enzyme in cultured antennal sensory neurons and non‐neural cells. ChAT and mAChRs were present in the sensory neurons from the first day in culture. Therefore, the sensory neurons are probably cholinergic, as previously suggested, but they may also be controlled by ACh. In 7‐day‐old cultures a subgroup of small non‐neural cells with processes expressed ChAT activity, and in 14‐day‐old cultures non‐neural cells that formed lamellipodia and scaffoldlike structures on the culture substrate were labeled with ChAT antibody. mAChR activity was detected in similar non‐neural cells but only in areas surrounding the nuclei. In addition, mAChRs were found in flat lamellipodia and filopodia forming cells that were present in 1‐day‐old cultures and grew in size during the 2 week investigation period. These findings suggest muscarinic cholinergic interactions between the neural and non‐neural cells during the development of Manduca antenna. © 2004 Wiley Periodicals, Inc. J Neurobiol, 2005     

Painful channels in sensory neurons

Pain is an unpleasant sensation experienced when tissues are damaged. Thus, pain sensation in some way protects body from imminent threat or injury. Peripheral sensory nerves innervated to peripheral tissues initially respond to multiple forms of noxious or strong stimuli, such as heat, mechanical and chemical stimuli. In response to these stimuli, electrical signals for conducting the nociceptive neural signals through axons are generated. These action potentials are then conveyed to specific areas in the spinal cord and in the brain. Sensory afferent fibers are heterogeneous in many aspects. For example, sensory nerves are classified as Aa, -b, -d and C-fibers according to their diameter and degree of myelination. It is widely accepted that small sensory fibers tend to respond to vigorous or noxious stimuli and related to nociception. Thus these fibers are specifically called nociceptors. Most of nociceptors respond to noxious mechanical stimuli and heat. In addition, these sensory fibers also respond to chemical stimuli [Davis et al. (1993)] such as capsaicin. Thus, nociceptors are considered polymodal. Recent advance in research on ion channels in sensory neurons reveals molecular mechanisms underlying how various types of stimuli can be transduced to neural signals transmitted to the brain for pain perception. In particular, electrophysiological studies on ion channels characterize biophysical properties of ion channels in sensory neurons. Furthermore, molecular biology leads to identification of genetic structures as well as molecular properties of ion channels in sensory neurons. These ion channels are expressed in axon terminals as well as in cell soma. When these channels are activated, inward currents or outward currents are generated, which will lead to depolarization or hyperpolarization of the membrane causing increased or decreased excitability of sensory neurons. In order to depolarize the membrane of nerve terminals, either inward currents should be generated or outward currents should be inhibited. So far, many cationic channels that are responsible for the excitation of sensory neurons are introduced recently. Activation of these channels in sensory neurons is evidently critical to the generation of nociceptive signals. The main channels responsible for inward membrane currents in nociceptors are voltage-activated sodium and calcium channels, while outward current is carried mainly by potassium ions. In addition, activation of non-selective cation channels is also responsible for the excitation of sensory neurons. Thus, excitability of neurons can be controlled by regulating expression or by modulating activity of these channels.