Basic mechanisms for graded persistent activity: discrete attractors,continuous attractors,and dynamic representations

Persistent neural activity is observed in many systems, and is thought to be a neural substrate for holding memories over time delays of a few seconds. Recent work has addressed two issues. First, how can networks of neurons robustly hold such an active memory? Computer systems obtain significant robustness to noise by approximating analogue quantities with discrete digital representations. In a similar manner, theoretical models of persistent activity in spiking neurons have shown that the most robust and stable way to store the short-term memory of a continuous parameter is to approximate it with a discrete representation. This general idea applies very broadly to mechanisms that range from biochemical networks to single cells and to large circuits of neurons. Second, why is it commonly observed that persistent activity in the cortex can be strongly time-varying? This observation is almost ubiquitous, and therefore must be taken into account in our models and our understanding of how short-term memories are held in the cortex.     

Optical dissection of neural circuits responsible for Drosophila larval locomotion with halorhodopsin

Halorhodopsin (NpHR), a light-driven microbial chloride pump, enables silencing of neuronal function with superb temporal and spatial resolution. Here, we generated a transgenic line of Drosophila that drives expression of NpHR under control of the Gal4/UAS system. Then, we used it to dissect the functional properties of neural circuits that regulate larval peristalsis, a continuous wave of muscular contraction from posterior to anterior segments. We first demonstrate the effectiveness of NpHR by showing that global and continuous NpHR-mediated optical inhibition of motor neurons or sensory feedback neurons induce the same behavioral responses in crawling larvae to those elicited when the function of these neurons are inhibited by Shibire(ts), namely complete paralyses or slowed locomotion, respectively. We then applied transient and/or focused light stimuli to inhibit the activity of motor neurons in a more temporally and spatially restricted manner and studied the effects of the optical inhibition on peristalsis. When a brief light stimulus (1-10 sec) was applied to a crawling larva, the wave of muscular contraction stopped transiently but resumed from the halted position when the light was turned off. Similarly, when a focused light stimulus was applied to inhibit motor neurons in one or a few segments which were about to be activated in a dissected larva undergoing fictive locomotion, the propagation of muscular constriction paused during the light stimulus but resumed from the halted position when the inhibition (>5 sec) was removed. These results suggest that (1) Firing of motor neurons at the forefront of the wave is required for the wave to proceed to more anterior segments, and (2) The information about the phase of the wave, namely which segment is active at a given time, can be memorized in the neural circuits for several seconds.     

Transformation of the dermal collagen framework under laser heating

The aim of this study was to compare between the changes undergone by the dermal collagen framework when heated by IR laser radiation and by traditional means and to reveal the specific features of the dermal matrix modification under moderate IR laser irradiation. Rabbit skin specimens were heated to 50°C, 55°C, 60°C and 65°C in a calorimeter furnace and with a 1.68‐μm fiber Raman laser. The proportion of the degraded collagen macromolecules was determined by differential scanning calorimetry. Changes in the architectonics of the collagen framework were revealed by using standard, phase‐contrast, polarization optical and scanning electron microscopy techniques. The collagen denaturation and dermal matrix amorphization temperature in the case of laser heating proved to be lower by 10°C than that for heating in the calorimeter furnace. The IR laser treatment of the skin was found to cause a specific low‐temperature (45°C‐50°C) transformation of its collagen framework, with some collagen macromolecules remaining intact. The transformation reduces to the splitting of collagen bundles and distortion of the course of collagen fibers. The denaturation of collagen macromolecules in the case of traditional heating takes its course in a threshold manner, so that their pre‐denaturation morphological changes are insignificant.     

Interaction between neurons of the frontal cortex and hippocampus during the realization of choice of food reinforcement quality in cats

Six cats were subjected to the procedure of appetitive instrumental conditioning (with light as a conditioned stimuls) by the method of the "active choice" of reinforcement quality. Short-delay conditioned bar-press responses were rewarded with bread-meat mixture, and the delayed responses were reinforced by meat. The animals differed in behavior strategy: four animals preferred the bar-pressing with a long delay (the so-called "self-control" group), and two cats preferred the bar-pressing with a short delay (the so-called "impulsive" group). Multiunit activity in the frontal cortex and hippocampus (CA3) was recorded via chronically implanted nichrome wire semimicroelectrodes. An interaction between the neighboring neurons in the frontal cortex and hippocampus (within local neural networks) and between the neurons of the frontal cortex and hippocampus (distributed neural networks in frontal-hippocampal and hippocampal-frontal directions) was evaluated by means of statistical crosscorrelation analysis of spike trains. Crosscorrelations between neuronal spike trains in the delay range of 0-100 ms were explored. It was shown that the number of crosscorrelations between the neuronal discharges both in the local and distributed networks was significantly higher in the "self-control" cats. It was suggested that the local and distributed neural networks of the frontal cortex and hippocampus are involved in the system of brain structures which determine the behavioral strategy of animals in the "self-control" group.     

Temperature-dependent Activation of Neurons by Continuous Near-infrared Laser

Optical control of neuronal activity has a number of advantages over electrical methods and can be conveniently applied to intact individual neurons in vivo. In this study, we demonstrated an experimental approach in which a focused continuous near-infrared (CNI) laser beam was used to activate single rat hippocampal neurons by transiently elevating the local temperature. Reversible changes in the amplitude and kinetics of neuronal voltage-gated Na and K channel currents were recorded following irradiation with a single-mode 980 nm CNI-laser. Using single-channel recordings under controlled temperatures as a means of calibration, it was estimated that temperature at the neuron rose by 14°C in 500 ms. Computer simulation confirmed that small temperature changes of about 5°C were sufficient to produce significant changes in neuronal excitability. The method should be broadly applicable to studies of neuronal activity under physiological conditions, in particular studies of temperature-sensing neurons expressing thermoTRP channels.     

Recurrent neural networks of integrate-and-fire cells simulating short-term memory and wrist movement tasks derived from continuous dynamic networks

Dynamic recurrent neural networks composed of units with continuous activation functions provide a powerful tool for simulating a wide range of behaviors, since the requisite interconnections can be readily derived by gradient descent methods. However, it is not clear whether more realistic integrate-and-fire cells with comparable connection weights would perform the same functions. We therefore investigated methods to convert dynamic recurrent neural networks of continuous units into networks with integrate-and-fire cells. The transforms were tested on two recurrent networks derived by backpropagation. The first simulates a short-term memory task with units that mimic neural activity observed in cortex of monkeys performing instructed delay tasks. The network utilizes recurrent connections to generate sustained activity that codes the remembered value of a transient cue. The second network simulates patterns of neural activity observed in monkeys performing a step-tracking task with flexion/extension wrist movements. This more complicated network provides a working model of the interactions between multiple spinal and supraspinal centers controlling motoneurons. Our conversion algorithm replaced each continuous unit with multiple integrate-and-fire cells that interact through delayed "synaptic potentials". Successful transformation depends on obtaining an appropriate fit between the activation function of the continuous units and the input-output relation of the spiking cells. This fit can be achieved by adapting the parameters of the synaptic potentials to replicate the input-output behavior of a standard sigmoidal activation function (shown for the short-term memory network). Alternatively, a customized activation function can be derived from the input-output relation of the spiking cells for a chosen set of parameters (demonstrated for the wrist flexion/extension network). In both cases the resulting networks of spiking cells exhibited activity that replicated the activity of corresponding continuous units. This confirms that the network solutions obtained through backpropagation apply to spiking networks and provides a useful method for deriving recurrent spiking networks performing a wide range of functions.     

A self-regulating feed-forward circuit controlling C. elegans egg-laying behavior

BACKGROUND: Egg laying in Caenorhabditis elegans has been well studied at the genetic and behavioral levels. However, the neural basis of egg-laying behavior is still not well understood; in particular, the roles of specific neurons and the functional nature of the synaptic connections in the egg-laying circuit remain uncharacterized. RESULTS: We have used in vivo neuroimaging and laser surgery to address these questions in intact, behaving animals. We have found that the HSN neurons play a central role in driving egg-laying behavior through direct excitation of the vulval muscles and VC motor neurons. The VC neurons play a dual role in the egg-laying circuit, exciting the vulval muscles while feedback-inhibiting the HSNs. Interestingly, the HSNs are active in the absence of synaptic input, suggesting that egg laying may be controlled through modulation of autonomous HSN activity. Indeed, body touch appears to inhibit egg laying, in part by interfering with HSN calcium oscillations. CONCLUSIONS: The egg-laying motor circuit comprises a simple three-component system combining feed-forward excitation and feedback inhibition. This microcircuit motif is common in the C. elegans nervous system, as well as in the mammalian cortex; thus, understanding its functional properties in C. elegans may provide insight into its computational role in more complex brains.     

Dynamics of population code for working memory in the prefrontal cortex

Some neurons (delay cells) in the prefrontal cortex elevate their activities throughout the time period during which the animal is required to remember past events and prepare future behavior, suggesting that working memory is mediated by continuous neural activity. It is unknown, however, how working memory is represented within a population of prefrontal cortical neurons. We recorded from neuronal ensembles in the prefrontal cortex as rats learned a new delayed alternation task. Ensemble activities changed in parallel with behavioral learning so that they increasingly allowed correct decoding of previous and future goal choices. In well-trained rats, considerable decoding was possible based on only a few neurons and after removing continuously active delay cells. These results show that neural activity in the prefrontal cortex changes dynamically during new task learning so that working memory is robustly represented and that working memory can be mediated by sequential activation of different neural populations.     

The individual organization of frontal-hippocampal networks during realization of different behavioral tasks

Four cats were subjected to appetitive instrumental conditioning with light as a conditioned stimulus by the method of "active choice" of the reinforcement quality: short-delay conditioned bar-press responses were followed by bread-meat mixture and the delayed responses--by meat. The animals differed in behavior strategy: four animals preferred bar-pressing with long delay (so called "self-control" group); two animal preferred bar-pressing with short-delay (so called "impulsive" group). Then all the animals were learned to short-delay (1 s) instrumental conditioned reflex to light (CS+) reinforced by meat. The multiunit activity in the frontal cortex and the hippocampus (CA3) was recorded through chronically implanted nichrome-wire semimicroelectrodes. The interactions among the neighboring neurons in the frontal cortex and hippocampus (within the local neuronal networks) and between the neurons of the frontal cortex and hippocampus (distributed neuronal networks of frontal-hippocampal and hippocampal-frontal directions) were evaluated by means of statistical crosscorrelation analysis of the spike trains. Crosscorrelation interneuronal connections in the delay range 0-100 ms were explored. It was shown that the functional organization of the frontal and hippocampal neuronal networks differed in choice behavior and was similar during realization of short-delayed conditioned reflex. We suggest that the local and distributed neural networks of the frontal cortex and hippocampus take part in the realization of cognitive behavior, in particularly in the processes of the decision making.     

Interrelations between neurons of the frontal cortex and hippocampus under cholinergic deficit in choice behavior of cats

Appetitive instrumental conditioned reflexes on light (CS+) were formed in six cats by the method of "active choice" of quality of reinforcement; bread-meat mixture was given after short-delay conditioned bar-press responses, and the delayed responses were rewarded by meat. The animals differed in choice behavior strategy: "self-control", "ambivalent", "impulsive". The multiunit activity in the frontal cortex and hippocampus (CA3) was recorded. Cross-correlation analysis was used for estimation of correlation of activities in neuronal pairs in the frontal cortex and hippocampus (distributed frontal-hippocampal networks) and pairs within the same structure (frontal and hippocampal local neuronal networks). It was shown that the number of cross-correlations between the discharges of neurons both in the local and distributed networks was significantly higher in "self-control" cats. Under conditions of systemic administration of antagonists of muscarinic central cholinoreceptors (trihexyphenidyl and scopolamine), the bar-press conditioning impaired, the number of direct interneuronal connections decreased, and the number of externally synchronized correlations ("common input") significantly increased. The results suggest that the local and distributed neural networks of the frontal cortex and hippocampus are involved in the system of brain structures that determine the behavioral strategy of "self control".     

Rhythmically firing (20–50 Hz) neurons in monkey primary somatosensory cortex: Activity patterns during initiation of vibratory-cued hand movements

The activity patterns of rhythmically firing neurons in monkey primary somatosensory cortex (SI) were studied during trained wrist movements that were performed in response to palmar vibration. Of 1,222 neurons extracellularly recorded in SI, 129 cells (11%) discharged rhythmically (at 30 Hz) during maintained wrist position. During the initiation of vibratory-cued movements, neuronal activity usually decreased at 25 ms after vibration onset followed by an additional decrease in activity at 60 ms prior to movement onset. Rhythmically firing neurons are not likely to be integrate-and-fire neurons because, during activity changes, their rhythmic firing pattern was disrupted rather than modulated. The activity pattern of rhythmically firing neurons was complimentary to that of quickly adapting SI neurons recorded during the performance of this task (Nelson et al., 1991). Moreover, disruptions of rhythmic activity of individual SI neurons were similar to those reported previously for local field potential (LFP) oscillations in sensorimotor cortex during trained movements (Sanes and Donoghue, 1993). However, rhythmic activity of SI neurons did not wax and wane like LFP oscillations (Murthy and Fetz, 1992; Sanes and Donoghue, 1993). It has been suggested that fast (20–50 Hz) cortical oscillations may be initiated by inhibitory interneurons (Cowan and Wilson, 1994; Llinas et al., 1991; Stern and Wilson, 1994). We suggest that rhythmically firing neurons may tonically inhibit quickly adapting neurons and release them from the inhibition at go-cue onsets and prior to voluntary movements. It is possible that rhythmically active neurons may evoke intermittent oscillations in other cortical neurons and thus regulate cortical population oscillations.     

Comparative strategies in the investigation of neural networks.

Comparative studies on nervous systems, though infrequently undertaken for the purpose of comparison, have yielded some important generalities about the formats of nervous networks, and about the cell biology of certain neural types. In the first category, it is clear that convergent evolutionary processes arrived at very similar networks to accomplish reciprocal and lateral inhibition, and load-compensation in "resistance reflexes." A newer general network format is described, command-derived inhibition, in which the central nervous elements controlling a rapid movement deliver presynaptic inhibition to the terminals of sensory neurons that carry reafferent excitation from the movement. It is argued that such circuits occur in several groups of animals, and that they include as a special class the efferent inhibitory neurons innervating acoustico-lateralis receptors in vertebrates. The properties of circuit elements that now seem to constitute useful generalizations include size principle (the inverse relationship between size and excitability in a variety of neurons), and the late differentiation of sensory neurons, failure to decussate, and their inability to mediate inhibition. Many other generalities have emerged, only to fall; one conclusion from such searches is that many supposedly "basic" properties of cell types or neural circuits are in fact not phylogenetically conservative, however much the physiologist may expect them to be.     

Dynamical coding of sensory information with competitive networks.

Based on experiments with the locust olfactory system, we demonstrate that model sensory neural networks with lateral inhibition can generate stimulus specific identity-temporal patterns in the form of stimulus-dependent switching among small and dynamically changing neural ensembles (each ensemble being a group of synchronized projection neurons). Networks produce this switching mode of dynamical activity when lateral inhibitory connections are strongly non-symmetric. Such coding uses 'winner-less competitive' (WLC) dynamics. In contrast to the well known winner-take-all competitive (WTA) networks and Hopfield nets, winner-less competition represents sensory information dynamically. Such dynamics are reproducible, robust against intrinsic noise and sensitive to changes in the sensory input. We demonstrate the validity of sensory coding with WLC networks using two different formulations of the dynamics, namely the average and spiking dynamics of projection neurons (PN).     

Local lateral inhibition: a key to spike synchronization?

 Starting from the idea that neural group activity as such is unlikely to be immediately relevant for neural synchronization, we investigate mechanisms that act at the level of individual nerve impulses (spikes). Hence, we consider populations of formal spike-emitting ‘leaky integrate and fire’ neurons instead of networks built from non-spiking oscillators. After outlining the principle of synchronization for basic forms of recurrent impulse coupling by using a pair of simplified formal neurons, we show that local lateral inhibition results in robust impulse synchronization in networks with non-vanishing transmission delays. Received: 12 January 1994/Accepted in revised form: 25 April 1995     

Modeling of laser coagulation of tissue with MRI temperature monitoring

Light energy from a laser source that is delivered into body tissue via a fiber-optic probe with minimal invasiveness has been used to ablate solid tumors. This thermal coagulation process can be guided and monitored accurately by continuous magnetic resonance imaging (MRI) since the laser energy delivery system does not interfere with MRI. This report deals with mathematical modeling and analysis of laser coagulation of tissue. This model is intended for "real-time" analysis of magnetic resonance images obtained during the coagulation process to guide clinical treatment. A mathematical model is developed to simulate the thermal response of tissue to a laser light heating source. For fast simulation, an approximate solution of the thermal model is used to predict the dynamics of temperature distribution and tissue damage induced by a laser energy line source. The validity of these simulations is tested by comparison with MRI-based temperature data acquired from in vivo experiments in rabbits. The model-simulated temperature distribution and predicted lesion dynamics correspond closely with MRI-based data. These results demonstrate the potential for using this combination of fast modeling and MRI technologies during laser heating of tissue for online prediction of tumor lesion size during laser heating.     

Predictive Coding of Dynamical Variables in Balanced Spiking Networks

Two observations about the cortex have puzzled neuroscientists for a long time. First, neural responses are highly variable. Second, the level of excitation and inhibition received by each neuron is tightly balanced at all times. Here, we demonstrate that both properties are necessary consequences of neural networks that represent information efficiently in their spikes. We illustrate this insight with spiking networks that represent dynamical variables. Our approach is based on two assumptions: We assume that information about dynamical variables can be read out linearly from neural spike trains, and we assume that neurons only fire a spike if that improves the representation of the dynamical variables. Based on these assumptions, we derive a network of leaky integrate-and-fire neurons that is able to implement arbitrary linear dynamical systems. We show that the membrane voltage of the neurons is equivalent to a prediction error about a common population-level signal. Among other things, our approach allows us to construct an integrator network of spiking neurons that is robust against many perturbations. Most importantly, neural variability in our networks cannot be equated to noise. Despite exhibiting the same single unit properties as widely used population code models (e.g. tuning curves, Poisson distributed spike trains), balanced networks are orders of magnitudes more reliable. Our approach suggests that spikes do matter when considering how the brain computes, and that the reliability of cortical representations could have been strongly underestimated.     

Flowers and weeds: cell-type specific pruning in the developing visual thalamus

In the first weeks of vertebrate postnatal life, neural networks in the visual thalamus undergo activity-dependent refinement thought to be important for the development of functional vision. This process involves pruning of synaptic connections between retinal ganglion cells and excitatory thalamic neurons that relay signals on to visual areas of the cortex. A recent report in Neural Development shows that this does not occur in inhibitory neurons, questioning our current understanding of the development of mature neural circuits. See research article: http://www.neuraldevelopment.com/content/8/1/24     

Stochastic resonance modulates neural synchronization within and between cortical sources

Neural synchronization is a mechanism whereby functionally specific brain regions establish transient networks for perception, cognition, and action. Direct addition of weak noise (fast random fluctuations) to various neural systems enhances synchronization through the mechanism of stochastic resonance (SR). Moreover, SR also occurs in human perception, cognition, and action. Perception, cognition, and action are closely correlated with, and may depend upon, synchronized oscillations within specialized brain networks. We tested the hypothesis that SR-mediated neural synchronization occurs within and between functionally relevant brain areas and thus could be responsible for behavioral SR. We measured the 40-Hz transient response of the human auditory cortex to brief pure tones. This response arises when the ongoing, random-phase, 40-Hz activity of a group of tuned neurons in the auditory cortex becomes synchronized in response to the onset of an above-threshold sound at its "preferred" frequency. We presented a stream of near-threshold standard sounds in various levels of added broadband noise and measured subjects' 40-Hz response to the standards in a deviant-detection paradigm using high-density EEG. We used independent component analysis and dipole fitting to locate neural sources of the 40-Hz response in bilateral auditory cortex, left posterior cingulate cortex and left superior frontal gyrus. We found that added noise enhanced the 40-Hz response in all these areas. Moreover, added noise also increased the synchronization between these regions in alpha and gamma frequency bands both during and after the 40-Hz response. Our results demonstrate neural SR in several functionally specific brain regions, including areas not traditionally thought to contribute to the auditory 40-Hz transient response. In addition, we demonstrated SR in the synchronization between these brain regions. Thus, both intra- and inter-regional synchronization of neural activity are facilitated by the addition of moderate amounts of random noise. Because the noise levels in the brain fluctuate with arousal system activity, particularly across sleep-wake cycles, optimal neural noise levels, and thus SR, could be involved in optimizing the formation of task-relevant brain networks at several scales under normal conditions.     

Variability of the EEG autocorrelation structure in adolescents with schizophrenia spectrum disorders

Quantitative analysis of changes in the autocorrelation structure of EEG short segments (in the range of several seconds) was performed in healthy adolescents (n = 39) and adolescents with schizophrenia spectrum disorders (n = 39). The variability of the EEG autocorrelation structure was shown to be higher in patients with the greatest trend in the frontal leads. It is suggested that psychopathology of the schizophrenia spectrum is accompanied by a break in the mutual determination of cortical neural networks with predominant localization of this process in the frontal areas of the brain cortex.     

Network Adaptation Improves Temporal Representation of Naturalistic Stimuli in Drosophila Eye: I Dynamics

Because of the limited processing capacity of eyes, retinal networks must adapt constantly to best present the ever changing visual world to the brain. However, we still know little about how adaptation in retinal networks shapes neural encoding of changing information. To study this question, we recorded voltage responses from photoreceptors (R1–R6) and their output neurons (LMCs) in the Drosophila eye to repeated patterns of contrast values, collected from natural scenes. By analyzing the continuous photoreceptor-to-LMC transformations of these graded-potential neurons, we show that the efficiency of coding is dynamically improved by adaptation. In particular, adaptation enhances both the frequency and amplitude distribution of LMC output by improving sensitivity to under-represented signals within seconds. Moreover, the signal-to-noise ratio of LMC output increases in the same time scale. We suggest that these coding properties can be used to study network adaptation using the genetic tools in Drosophila, as shown in a companion paper (Part II).