Institutions
University of Manchester, UK.
Universita di Modena e Reggio, Italia.
Istituto di Bioimmagini e Fisiologia, Italia.
Latency Coding of Whisker Deflection Direction in the Rat Lemniscal Pathway
Many studies have shown that the precise timing of spikes can convey considerable more stimulus information than a rate code. However, whether there is a role for spike timing in the coding of whisker deflection direction is not known. We recorded extracellularly the single unit responses from neurons in the trigeminal ganglion and the ventro-posterior medial (VPM) nucleus in urethane anaesthetised rats. Each principal whisker was stimulated in over 400 randomly selected directions using a 2D piezo-electric actuator. We then quantified how well each neuron encoded direction using information theoretic methods. We found that, in both ganglion and VPM, a substantial amount of mutual information about direction was conveyed by the first post-stimulus spike alone. Furthermore, in many neurons (39% in ganglion; 57% in VPM), the timing (latency) of the first spike conveyed additional information. An important issue concerning latency coding is that rat experimental subjects, unlike human experimenters, do not know the time of stimulus onsets. However, we found that a simple, reafference mechanism might address this problem. Since deflection direction was uniformly represented by ganglion neurons, we found that it was possible to decode the stimulus time by summing the population activity using a leaky integrator. This estimated stimulus time was highly accurate (1.1 SD 1.3 µs) and sufficient to preserve 98% of the information that ganglion cells convey about whisker deflection direction. This was also true of VPM cells (92%) suggesting the mechanism is preserved along pathway. In sum, our results show that considerable information about the direction of whisker deflection is available in the spike timing responses of a subpopulation of subcortical neurons and that this information is potentially decodable using reafference signals alone.
Fisher Information in Asynchronous Networks
In the cortex, neurons often fire asynchronously in response to external stimuli. The dynamics of these asynchronous states is well understood. However, it is not understood how much information such asynchronous responses may contain. For a simple, yet realistic network model receiving tuned input, we analytically calculate the information contained in the network about the external tuned input. This allows us to identify the network properties that increases information. We find that structured connectivity which matches the form of the tuned input increases information content. However, changes to the connectivity responsible for maintaining the asynchronous state make little difference to information.
Population activity in a spinal motor network monitored with extra-cellular multi-electrode recordings during local drug delivery
The mechanisms of central pattern generation in the spinal cord remain a puzzle. One major challenge is the large number of neurons involved in the generation of these motor patterns. In an attempt to address the issue of the population activity during motor activity we perform extra-cellular recordings with multi-electrode arrays in the semi-intact spinal cord of the adult turtle during fictive scratching (Keifer and Stein 1983). We find that all of the isolated units are rhythmically participating in the cyclic motor pattern (n=24). The recorded units have different but constant phase lag with respect to the hip flexor nerve in agreement with previous reports (see e.g. Berkowitz 2002).
Furthermore, in order to verify the role of inhibitory synaptic input during the scratch motor activity, we test the effects of local application of a glycinergic antagonistic agent (strychnine hydrochloride) delivered via a micro-fluidic probe in close proximity (~ 100 microns) to a multichannel sililcon array (Neuronexus multichannel array). By testing if there is an increase in spike frequency in the active period, we evaluate the hypothesis that spike activity is driven by concurrent inhibition and excitation.
Quantifying neurotransmission through an entropy measureembedding spike train metrics
We set forth a novel information theoretical measure [1] to quantifyneurotransmission reliability while taking into full account themetrical properties of the event space. This parametric informationanalysis relies on a similarity measure to estimate, as spikes flow in,the metrical relations between neural responses. In order to assess theconditional entropy and the overall information transfer, this methoddoes not require any a priori decoding algorithm to partition the spaceinto equivalence classes (e.g. clusters of neural responses based onconfusion matrices). To validate the proposed information theoreticalapproach, we study precise temporal decoding of human somatosensorysignals (i.e. the responses of fingertip mechanoreceptors to tactilestimulation) recorded via microneurography experiments [3]. For thisanalysis, a similarity measure based on the Victor-Purpura spike trainmetrics [2] is employed. It is shown that the relative spike timing ofthe mechanoreceptors responses can convey enough information to performoptimal discrimination (defined as maximum metrical information andzero conditional entropy) of 81 distinct stimuli within 60 ms of thefirst afferent spike. The proposed information theoretical measureproves to be a suitable generalisation of Shannon Mutual Information inorder to consider the metrics of temporal codes explicitly. It permitsan assessment of neurotransmission reliability in the presence of largespike train spaces (e.g. neural population codes) with high temporalprecision (e.g. 1 ms).
Keywords: Information theory, neurotransmission, temporal neuralcode, spike-train metrics.
References:
Neuronal dynamics in somatosensory cortex of rats after stroke
Stroke in the brain triggers massive plasticity. This remodeling of neuronal circuits occurs not only in tissue neighboring the stroke area but also in remote areas of the brain, including the contralateral hemisphere. Here, we investigate changes in neuronal dynamics occurring during stroke recovery. In particular, we focus on neuronal population activity in cortical layer V, as cells in this layer show most pronounced morphological changes after stroke and its functional significance is poorly understood. We recorded from anesthetized rats with stroke in motor cortex. Two silicon probes (32 electrodes per probe) were inserted in layer V in hind limb sensory areas in both hemispheres. We analyzed neuronal population dynamics in both spontaneous and stimulus evoked conditions (tactile stimuli were applied to multiple points on hind limbs). Neural population recordings were correlated with detailed analysis of behavioural measures of stroke recovery (skilled reaching task). The preliminary results suggest that global interactions between neuronal populations are indicative of functional recovery from stroke.
Limitations of time-scale parametric spike train distances to study precision and reliability
We consider several spike train distances that have in common a parameter that determines the time-scale of the spike times up to which the distance is sensitive to [1, 2, 3]. This parametric dependence is essential for the application of these distances to study the precision of the neural code by finding the optimal time-scale at which spike trains in response to different stimuli are discriminated the best.
We here show that this optimal time-scale is in general not informative about the neural code. By contrast it results from a non-trivial interplay of different sources of temporal structure to which the distances are sensitive to. This problem of the interpretability of the optimal time-scale obtained in the precision analysis is common to all these time-scale parametric spike train distances regardless of their particular definition. For the Schreiber et al. distance [3] we furthermore indicate a potential lack of specificity quantifying the spike train similarity caused by the way it is defined.
Hippocampal population code for discrete locations during a matching-to-multiple-places task
The hippocampus is a key brain region for cognition and notably for spatial memory. Such a role has been related to the fact that hippocampal principal cells fire in relation to space. It is thought that these "place cells" collectively provide a map-like representation of the environment, which can be used for solving spatial problems. In order to examine how place cell representations might enable solving spatial memory task, we recorded place cell activity in behaving rats during a matching-to-multiple-places task that requires frequent updating of memories for changing goal locations. The task took place on a cheeseboard maze where rats were required to find a set of rewardedwells whose locations varied from day to day but stayed constant within successive trials on a given day. On each day, rats covered longer distances before finding the rewards during the first trial than in subsequent trials. The reduction of path length after the first trial suggests that rats searched for the baited wells on the first trial, while they rapidly encoded and remembered locations for the remaining trials. We report that: (i) place-related firing patterns in CA1 regionreorganised to over-represent the learned goal locations so that learning-related CA1 population firing patterns representing learned locations correlated with memory performances in subsequent probe trials; (ii) the firing patterns of place cells that encode goal locations became more tuned together during the course of learning and exhibited a higher sharp-wave/ripple associated firing response compared to those place cells that did not encode goal locations; (iii) during the post-learning sleep period, the firing patterns of goal-centric cells exhibited stronger reactivation than other place cells and their reactivation predicted subsequent memory performances. Altogether, theseresults suggest that reorganisation and reactivation of goal-related population firing patterns sustain learning and memory retention abilities.
title and abstract follow
Time and category information in pattern-based codes
Sensory stimuli are usually composed of different features (the what) appearing at irregular times (the when). Neural responses often use spike patterns to represent sensory information. Intuitively, the what is assumed to be encoded in the identity of the elicited patterns, and the when, in the timing of patterns. However, the what and the when might not be separable concepts, for example, when they are correlated in the stimulus. In addition, there are situations where the timings of the patterns may sometimes encode the what, as for example, in latency codes. Conversely, the pattern identities may sometimes encode the when. Hence, the intuitive view turns out to be more complex. In this study, we assess the role of spike patterns by quantifying the information about the stimulus conveyed by different representations of the neural response. We establish the relation between the what and the when in the stimulus with the information conveyed by the timing (time information) and the categories (category information) of spike patterns. A formal framework is developed to assess the conditions under which the information about the what and the when coincide with the time and category information, as well as the departures from this behaviour. Finally, we study the capabilities of different neural codes to represent the what and the when in the neural response.
Neuroanalysis.org and STAToolkit: Open-Source Information-Theoretic Analyses.
To advance understanding of neural coding, and complement widely-available conventional methods, we have developed the STAToolkit, a set of Open Source information-theoretic algorithms available at neuroanalysis.org (Goldberg et al. Neuroinformatics 7, 165-178, 2009). Analyses of neural coding require such a suite of methods, because different neural systems use specialized representations, and many analytic methods require specific types or amounts of data.
New STAToolkit v1.5 (2/12/2010) now:
Toward aiding formulation and test of hypotheses about neural coding, sensory discrimination, firing pattern variability, synchrony and other relations characterizing multineuronal recordings, we are expanding capabilities of neuroanalysis.org. Utilizing algorithmic equations, code, and/or insight from European,UK, and US collaborators, our set of complementary analytic methods will include additional information measures, dimensional reduction, methods to distinguish information from purely biophysical variation, and generation of surrogate multineuronal data sets.
Supported by Human Brain Project/Neuroinformatics MH068012 from US NIMH, NINDS, NIA, NIBIB and NSF.
Cortical spike trains encode active grasp behaviors
To analyze neural correlates of prehension quantitatively, we trained macaque monkeys to perform a task in which they grasp, lift, hold, lower, and release round or rectangular objects in response to visual cues. Digital video simultaneously recorded hand movements and spike trains of neurons in posterior parietal cortex (PPC). To compare the timing of firing of different cortical populations from serial single unit recording sessions, we evaluated spike trains with both standard statistical tests and STAToolkit information-theoretic algorithms. We found that task factors controlled by the animal -- how and when he used his hand -- played the principal roles in modulating firing rates of neurons recorded in rostral area 5 and area AIP. Forward and lateral approach trials --; those that included reaching, wrist rotation, and hand preshaping prior to contact -- evoked the most vigorous responses in 80% of the population studied. High firing rates began during reach as the hand was directed to the target, and carried over into subsequent contact, grasp, and lift stages; strong responses were associated with fluid movement of the hand as the object was acquired and lifted. Moreover, spike trains evoked during these reach trials were longer in duration than in the no-reach condition in which the animal grasped an object in the immediate environment (local approach) or lifted objects repeatedly (regrasps). Actions of the hand and arm before contact thus influence neural responses in subsequent stages. Firing rates during reach trials were only weakly modulated by direction of reach or location or features of the object.
We also used the metric space technique to measure how much information about the task kinematics and the properties of the grasped object was conveyed by the internal temporal structure of spike trains. The greatest information about task kinematics was conveyed in the 200 ms interval prior to contact, the period that distinguished reach and no-reach trials. Significant information was conveyed by spike timing about the approach style, but not the object size or shape.
Our spike train analyses suggest that complex visuomotor and somatosensory actions preceding contact reinforce subsequent neural activity accompanying grasp and lift, allowing subjects to acquire and manipulate objects in a continuous, smooth sequence.
Support provided by National Institute of Neurological Diseases and Stroke (NINDS) Research Grant R01 NS-011862.
A metric space approach to the information capacity of spike trains.
Classical information theory can be either discrete or continuous, corresponding to discrete or continuous random variables. However, although spike times in a spike train are described by continuous variables, the information content is usually calculated using discrete information theory. This is because the number of spikes, and hence, the number of variables, varies from spike train to spike train, making the continuous theory difficult to apply. It is possible to avoid this difficulty by using a metric space approach to spike trains. A metric can be prescribed, giving a distance between different spike trains. The continuous version of information theory is then rephrased in terms of metric quantities, to give a method for calculating the information capacity of spike trains. This method works by matching the distribution of distances between responses to the same stimulus to a chi-distribution: the chi-distribution is the length distribution for a vector of Gaussian variables. This defines an effective dimension for the spike train and gives a bound on the channel capacity. As an example, this capacity is calculated for a dataset recorded from auditory neurons in zebra finch.
Automatic classification of single- and multi-unit activity from extracellular recordings
Abstract follows
Affiliations
Quantification of the contribution of temporal integration and coincidence detection to the irregularity of cortical neurons at high rates
In this paper, we address the question of quantifying the relative contribution of temporal integration and coincidence detection to the high firing variability observed in cortical neurons [1]. Such a quantification of the two neuronal operational modes would be a significant step in shedding light to the underlying coding scheme, as substantially more temporal integration would suggest that information is encoded in the rate of the neuron's output spike train, while substantially more coincidence detection would indicate the importance of the timing of individual spikes in the encoding of information, pointing towards a temporal code.
In a previous attempt to identify the operational mode of the cortical neuron, Bugmann et al. [2] used reverse correlation (RC) graphs of the stimulus of a leaky integrate-and-fire (LIF) neuron model with the partial somatic reset mechanism (which has been shown to be the most likely candidate to reflect the mechanism used in the brain for reproducing the irregular firing of cortical neurons at high rates [2,3]). They were unable to provide a conclusive answer however (despite that RC graphs were recommended for such a purpose by Mainen and Sejnowski [4]), and they suggested that RC of the membrane potential may be more informative.
Based on this suggestion, in our current attempt to determine the operational mode of the cortical neuron, we introduce a novel, heuristic measure which uses the RC of the membrane potential of a LIF neuron with partial reset. Our method relies on the similarity between the shape of the curve of the membrane potential time course preceding each spike and the shape of the RC of the membrane potential of a pure temporal integrator and a pure coincidence detector neuron. The outcome is a measure of the number of spikes caused by each operational mode, which indicates the relative contribution of each mode to the production of spikes in our model. Our preliminary results show that at firing frequencies from 70Hz to 170Hz the model operates mainly as a temporal integrator, while at firing rates between 170Hz and 280Hz, coincidence detection is the dominant mode. These results indicate that our proposed heuristic method can provide a form of quantification between the two modes and confirm that the RC of the membrane potential is indeed more informative. An alternative, analytical approach is also presented, which relies on analytical derivations of the RC graphs.
References:
Affiliations:
Time-resolved and time-scale adaptive measures of spike train synchrony
Estimating the degree of synchrony between two or more spike trains is a frequent task in both experimental and computational neuroscience. One of the most recent approaches is the ISI-distance, a simple method that extracts information from the interspike intervals (ISIs) by evaluating the ratio of the instantaneous firing rates. In contrast to most other measures that had been proposed previously it is parameter free and time scale independent, however, it is not well suited to track changes in synchrony that are based on spike coincidences. Here we propose the SPIKE-distance, a measure complementary to the ISI-distance in the sense that it is sensitive to spike coincidences. The new method shares the fundamental advantages of the ISI-distance, in particular, it is easy to visualize in a time-resolved manner. Like the ISI-distance it can easily be extended to a method that is also applicable to larger sets of spike trains.
EMG Prediction from Motor Cortical Recordings
in the State-Space
A constrained Kalman filtering (KF)-optimization mechanism for prediction of electromyogram (EMG) signals from multi-channel neural spike recordings in the state-space is proposed here. Sub-optimality of filters from the Kalman family in dealing with non-Gaussian observations or when the state evolution violates the linear-Gaussian Markov process assumption has inspired this work. To rectify the problem of non-Gaussian observations (here spikes), this approach can encapsulate several covariates of neural activity, e.g. the neurons' own spiking history, concurrent ensemble activity, and extrinsic covariates such as stimuli or behavior in the neurons' spiking probability by using a combined generalized linear model (GLM) - point process structure. This approach provides reasonable characterizations between neural activity and motor behavior. Moreover, as the envelopes of EMGs are strictly non-negative, the predictor is penalized for negative EMG estimates in a constrained optimization interpretation of traditional Kalman filters, termed Constrained Kalman filter (CKF). The EMG signals were predicted for twelve forearm and hand muscles of a behaving monkey during a squash ball grip task. Fit and generalization [over time] rates - in terms of the coefficient of determination R^2 - computed using such method showed improvement of on average 5% over that of the linear filtering when the in GLM model no history or coupling elements was included. Including these extra elements improved the predictions scores by a further 5% on average. Moreover, the approach was tested for different bin sizes and delays [between spikes and motor output]. Results confirmed that bin sizes and delay values of respectively 20 and 40 ms provide highest prediction rates.
Supported by The Wellcome Trust
Collaborators: Liam Paninski (Columbia), Christian Ethier, James M. Rebesco, and Lee E. Miller (Northwestern) and R. Chris Miall (Birmingham)
Spike train metric based on Cox
method
CurrentMulti-Electrode Array (MEA) technique allows to simultaneouslyrecording many spike trains and analyzing of them is a challengingproblem. We describe a statistical technique called Cox method forestimating of influences between spike trains. There are severalimportant advantages of this method: binless; efficient (works for alimited amount of spikes); sensitive to weak connections; allowssimultaneously analysing influences from many spike trains to thetarget spike train and estimating the strength of each influence. Wehave used the estimates of the Cox method to derive a metric(distance) between two spike trains. This distance has been used fork-means cluster analysis to identify the scheme of functionalconnections between spike trains. We have tested this technique usingboth generated and experimental data (up to 100 spike trains) andfound a good performance of the method.
Hippocampal population representation of different environments
The hippocampal formation is a brain area known to be involved in the encoding and consolidation of spatial and episodic memories. During waking activity, hippocampal principle cells fire in relation to space, such that their activity is restricted to discrete regions of the environment, known as “place fields”. In different environments, these “place cells” remap by firing in different locations. In addition, such cells can change their average firing rate, from environment to environment. This ‘rate remapping’ can occur in conjunction with, or independently of, place field remapping. It has been hypothesised that spatial selectivity and firing rate may separately underlie coding for spatial and episodic memory, respectively, however it is not clear the degree to which firing rate is independent in different environments. Moreover, the degree to which interneuron firing patterns change from one environment to another, as well as their contribution to pyramidal cell ensemble code, remains to be determined.
We recorded ensembles of neurons in the hippocampal formation of behaving rats during exploration of different environments. CA1 principle neurons showed changes in spatial selectivity and firing rate across different environments, as previously reported. However, the change in population firing rates was not random: mean and peak firing rates established in different recording arenas were correlated. Moreover, this correlation was significantly higher than when firing rates were randomly shuffled, indicating that at the population level both the place related firing rates and overall firing probability were not completely reorganised in different arenas. In contrast, the firing relationships between cells were completely reorganised from environment to environment. This suggests that different cell assemblies are active in each environment. Finally, while the average firing rates of interneurons were similar in different environments, the relationship between the fine structure of interneuron firing and population pyramidal cell activity differed across recording arenas. These data collectively indicate that spatial environments are represented in the hippocampus by a unique combination of associations between cells.
Avoiding correlations in neural activity on neuromorphic hardware
Because of inherent physical constraints, communication bandwidth on hardware is usually limited. Especially in the case of the FACETS neuromorphic hardware, where the emphasis is placed on configurability and interneuron communication, most of the available physical space is taken by the corresponding circuits. When a large number of on-chip neurons require independent stimulation, a limited number of inputs is obviously problematic. We have devised a method which allows us to predict correlations in neural activity only from knowledge of the number of common and independent input channels. Also, we have developed an algorithm which searches for input configurations with minimal pairwise overlap within a limited pool ov available inputs. These tools have allowed us to emulate networks on our hardware systems which would otherwise greatly exceed the available input bandwidth.
Unsupervised wavelet optimization for detection of intra-cortical action potentials with low signal-to-noise ratio
Neural recordings are analyzed as a source of information on the activity in different parts of the brain. Neural signals recorded from the motor area can be used for the prediction of movement to build brain-computer interfaces (BCIs) for neurorehabilitation purposes. Improving the detection of neural activity against the background noise is an important aspect for building a robust framework for BCI applications. We propose a method for the detection of neural spikes in the wavelet domain. We optimized the method by introducing a new signal based criterion for selecting the best wavelet which improves the detection performance. The method was tested in 3 sets of noisy simulated signals. Each set of signals was analyzed with 3 levels of added white Gaussian noise and the performance was measured by the detection error rate, as summation of false positives and false negatives divided by total number of spikes. The proposed method was compared with previously proposed detection methods.
A new method to determine functional connectivity: IncrementalMutual Information
As multi-electrode and imaging technology begin to provide us withsimultaneous recordings of large neuronal populations, new tools forthe analysis and of modeling of these data must also be developed. Wedevelop a new method for characterizing the strength and dynamics ofthe functional connectivity between neurons: incremental mutualinformation (IMI). IMI measures how informative the activity of oneneuron is about another a particular delay, after the past activity ofboth neurons has already been considered. IMI improves on thecorrelation-based measures that are typically used to study functionalconnectivity in two important ways: 1) IMI does not assume linearityand 2) IMI measures only the dependencies due to connectivity,ignoring those due to external stimuli or shared inputs, without theneed for repeated trials. I will demonstrate the utility of IMI incharacterizing the functional connectivity between cells in themammalian visual pathway.
Code Analysis Repository and Modelling for e-Neuroscience: theCARMENproject
The CARMEN project aims to deliver a virtual laboratory forneurophysiology, enabling sharing and collaborative exploitation ofdataand analysis code for neural activity recordings (time series). This isachieved through portal technology (so all that the user needs is a webbrowser), and users can upload and download data, and apply services toprocess this data. The datasets are primarily electrophysiology data,including spike train data. The portal is freely available for use.
Spike trains recorded extracellularly from cortical cultures on multi-electrode arrays (MEAs) reflect activity of nearby neurones. Due to interactions between such areas the spike trains are not independent. Moreover, the structure of the dependencies changes as a function of the nature of the interactions, including changes in synaptic strength. The evolution of these dependencies was studied by creating a model of complex evolving networks. Preliminary results indicate that although the evolution of interactions captures gross changes of network activities some variability is not accounted for indicating a more complex information processing flow.
+these two authors contributed equally to the work
Predictable thalamocortical firing patterns in chronic pain models
Most studies on sensory mechanisms in the brain have targeted system responses to stimuli or task-evoked activations. However, brain activity can be elevated even in the absence of explicit inputs. Spontaneous activity is usually considered to be a kind of noisy background in studies on stimulus processing. It was even suggested that stimulus evoked neural activities act as a trigger to release latent circuit dynamics rather than directly reflecting the structure of the input signal itself. While in the spinal cord sustained noxious stimulation modifies (e.g. by sensitization or wind up) the dynamics of ongoing neural firing patterns, at higher level in the sensory pathways a proper characterization of ongoing activity in still uncertain. We analysed the firing patterns in rat VPL thalamus and S1 cortex at single and multi-unit level and compared our results in control animals with those obtained by recording from diverse models of chronic pain.
First we investigated the possibility that chronic pain conditions modulated the similarity among spike patterns. As a measure of similarity we used the Normalized Compression Distance (NCD) that allowed us to jointly detect low and high-order correlations. We computed the NCD between all the recorded neuron pairs.
Then we asked whether the overall level of activity represented the only feature that constrains Multi-Unit Spike Patterns (MUSPs) during ongoing activity. The alternative hypothesis was that more complex features are necessary to provide more accurate predictions of incoming MUSPs. To address this point we combined clustering algorithms and information theoretic analyses.
From the first set of analyses we found that neurons recorded from neuropathic animals express more stereotyped patterns. This result was particularly strong in VPL thalamic nuclei although also significant in S1.
The second set of analyses showed that both in VPL and S1 the overall level of activity alone is not sufficient to describe MUSP dynamics and that the activation of specific unit subsets must be taken into account. From these analyses we observed that in S1 the dynamical constraints, as above, were massively distributed in the neuropathic models in comparison with controls.
Information Visualization of multi-dimensional
spike traindatasets
This work focuses on the analysis of multi-dimensional spike train datasets. The software developed is part of the ongoing ViSA (Visualization of Inter-Spike Associations) project which aims to provide software support for the visual exploration of large datasets [1].
Version II of the software provides the user with two key visual representations (i) the iGridTM[2] representation which provides an overview of multiple cross-correlograms for a number of spike trains and (ii) the iRasterTM[3] representation which is a highly interactive raster plot with extensive functionality. These representations have been tested and proved useful using both generated and real data.
This poster presents the prototype of Version III of the software. This is a substantial upgrade to previous versions. The software has been entirely re-engineered such that (i) the interface of the software is now based on a visual programming paradigm, and, (ii) the functionality of the software is modular. In addition, a completely new underlying framework has been developed.
References
A Novel Algorithm for Automated Cell Finding
Simultaneous multi-channel single unit recordings are a key tool toinvestigate neural function. Whilst some success has been achievedusing chronic arrays of stationary electrodes, the yield andsignal-to-noise ratio of recordings are usually inferior to thoseobtained from moveable electrodes acutely implanted on the day of therecording. However, a key limitation to this approach is the need tomove many electrodes to locate and keep clean unit activity. In ourexperience, a single experimenter can work effectively with only 3-5electrodes in this way.
We have developed a simple algorithm to automate the process offinding extracellular single units in real time using a multi-channelmicroelectrode drive. The algorithm uses brief (1s) periods ofrecording to assess recording quality. The microelectrode is advanceduntil well isolated units are detected.
The autonomous algorithm consists of four steps:
The algorithm differs from more usual approaches to unit clustering,in that it does not attempt to assign cluster identities to individualspikes. Rather, the analysis is focussed on providing only measures ofrecording quality. Consequently, the algorithm is effective on onlyshort records, and fast enough to implement in real time. We arecurrently incorporating this algorithm into our experimental system toallow automated location of units across a large electrode array.
Funded by: EPSRC and The Wellcome Trust
Parallel processing of visual information in the superior colliculus: difference in response variability to visual stimulation
Cells in the early visual system, belonging to different visual information processing channels, differ in morphological and physiological properties. Visually responding neurons in the superficial layers of the cat superior colliculus (SC) receive retinal inputs from two channels, Y and W, what is reflected in their spatio-temporal frequency properties and velocity response profiles. We quantified the time-dependent variability of responses of SC neurons to stimuli moving with different velocities by Fano factor (FF). Changes of activity of cells responding to fast moving stimuli, processed by Y-pathway, correlated negatively with FF whether the response was excitatory or suppressive, while the FF for cells responding to low-velocity stimuli, thus receiving W-inputs, increased with the increase in the firing rate. In order to explain these differences in variability of responses we used four stochastic models of experimental data: inhomogeneous Poisson process, parametric and non-parametric versions of inhomogeneous Markov interval processes and inhomogeneous Gamma process. Neither model could fully recover dependencies between firing rate and response variability of SC cells, even though all models reproduced quite well average changes in neuronal activity evoked by visual stimulation. The inclusion of the time-from-last-spike dependence to some models recovered most of the observed features of responses to fast moving stimuli, but failed to reproduce the variability of low-velocity responses. In this last case probably more complex time dependencies or network influences need to be considered. Our results indicate that Y- and W-channels may differ in reliability of responses to visual stimulation. This suggests that, apart from morphological and physiological differences of cells belonging to different visual information processing channels, response variability may be a new feature distinguishing parallel visual pathways.
Supported by MSHE/Poland grant N N303 07234
Information Theoretic Analysis of Proprioceptive Encoding DuringFinger Flexion in Monkey Sensorimotor Cortex
We recorded single units from motor cortex, somatosensory cortex,parietal cortex and cerebellum in two monkeys performing a fingerflexion task. The task consisted of making a finger flexion toone of four different positions against either high or low torque(giving eight task conditions). Each trial consisted of a flexionmovement followed by a stationary hold period. We calculated the firingrate and firing irregularity during a one second section of the holdperiod and calculated the task information based on these two codingparameters. Here we report the results from the different areas, thecoding of position information vs. torque information and also thedegree of redundancy in the combined firing rate/firing irregularitycode.
Supported by The Wellcome Trust.
© 2010 by -thowe-