No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Automatic Sorting for Multi-Neuronal Activity Recorded With Tetrodes in the Presence of Overlapping Spikes
Abstract
Multi-neuronal recording is a powerful electrophysiological technique that has revealed much of what is known about the neuronal interactions in the brain. However, it is difficult to detect precise spike timings, especially synchronized simultaneous firings, among closely neighboring neurons recorded by one common electrode because spike waveforms overlap on the electrode when two or more neurons fire simultaneously. In addition, the non-Gaussian variability (nonstationarity) of spike waveforms, typically seen in the presence of so-called complex spikes, limits the ability to sort multi-neuronal activities into their single-neuron components. Because of these problems, the ordinary spike-sorting techniques often give inaccurate results. Our previous study has shown that independent component analysis (ICA) can solve these problems and separate single-neuron components from multi-neuronal recordings. The ICA has, however, one serious limitation that the number of separated neurons must be less than the number of electrodes. The present study combines the ICA and the efficiency of the ordinary spike-sorting technique (k-means clustering) to solve the spike-overlapping and the nonstationarity problems with no limitation on the number of single neurons to be separated. First, multi-neuronal activities are sorted into an overly large number of clusters by k-means clustering. Second, the sorted clusters are decomposed by ICA. Third, the decomposed clusters are progressively aggregated into a minimal set of putative single neurons based on similarities of basis vectors estimated by ICA. We applied the present procedure to multi-neuronal waveforms recorded with tetrodes composed of four microwires in the prefrontal cortex of awake behaving monkeys. The results demonstrate that there are functional connections among neighboring pyramidal neurons, some of which fire in a precise simultaneous manner and that precisely time-locked monosynaptic connections are working between neighboring pyramidal neurons and interneurons. Detection of these phenomena suggests that the present procedure can sort multi-neuronal activities, which include overlapping spikes and realistic non-Gaussian variability of spike waveforms, into their single-neuron components. We processed several types of synthesized data sets in this procedure and confirmed that the procedure was highly reliable and stable. The present method provides insights into the local circuit bases of excitatory and inhibitory interactions among neighboring neurons.
... A central problem in this decoding task is identifying action potentials overlapping in time (a.k.a., superimposition of different spikes) [1]. Superimposed action potentials must be identified to extract the complete neural code and identify physiological mechanisms, such as neural synchronization [1,4,11]. ...
... Various spike sorting algorithms have been proposed in the literature [3,5,[11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26], and some include the analysis of overlapping spikes [5,11,[15][16][17][21][22][23][24][25][26]. Zhang et al. [24] combined different templates to minimize the residual variance using the 2 test. ...
... Various spike sorting algorithms have been proposed in the literature [3,5,[11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26], and some include the analysis of overlapping spikes [5,11,[15][16][17][21][22][23][24][25][26]. Zhang et al. [24] combined different templates to minimize the residual variance using the 2 test. ...
Objective:
Spike sorting of muscular and neural recordings requires separating action potentials that overlap in time (superimposed action potentials (APs)). We propose a new algorithm for resolving superimposed action potentials, and we test it on intramuscular EMG (iEMG) and intracortical recordings.
Methods:
Discrete-time shifts of the involved APs are first selected based on a heuristic extension of the peel-off algorithm. Then, the time shifts that provide the minimal residual Euclidean norm are identified (Discrete Brute force Correlation (DBC)). The optimal continuous-time shifts are then estimated (High-Resolution BC (HRBC)). In Fusion HRBC (FHRBC), two other cost functions are used. A parallel implementation of the DBC and HRBC algorithms was developed. The performance of the algorithms was assessed on 11,000 simulated iEMG and 14,000 neural recording superpositions, including two to eight APs, and eight experimental iEMG signals containing four to eleven active motor units. The performance of the proposed algorithms was compared with that of the Branch-and-Bound (BB) algorithm using the Rank-Product (RP) method in terms of accuracy and efficiency.
Results:
The average accuracy of the DBC, HRBC and FHRBC methods on the entire simulated datasets was 92.16±17.70, 93.65±16.89, and 94.90±15.15 (%). The DBC algorithm outperformed the other algorithms based on the RP method. The average accuracy and running time of the DBC algorithm on 10.5 ms superimposed spikes of the experimental signals were 92.1±21.7 (%) and 2.3±15.3 (ms).
Conclusion and significance:
The proposed algorithm is promising for real-time neural decoding, a central problem in neural and muscular decoding and interfacing.
... In neuroscience, improved sensors (4) have permitted large increases in the size and dimensionality of recorded datasets. An essential problem remains that the brain contains millions of simultaneously active neurons, emitting action potentials (spikes) with varying frequencies and patterns related to ongoing behavior and brain state (3,5). The identification of signals from individual neurons, in a sea of brain action potential, is critical. ...
... Also, there are often no clear boundaries between the signals of the different neurons recorded, and the density of neurons varies widely across different regions of the brain and across different recording methods (10). Overlapping clusters and strong background activity, produced by neighboring neurons, as well as the similarity of spike waveforms in given classes of neurons, present different problems for algorithms that rely on matching spike waveforms to templates, principal component analysis (PCA), density, and distance metrics (5). Compounding the complexity of the spike-sorting problem, recordings can involve 10-20 "useful" dimensions, especially those that use tetrodes or multisensor probes (8). ...
... Templatematching algorithms also present the task of preparing a template library of each of these neuron shapes. Independent component analysis (ICA) can successfully identify independent sources by decomposing a signal into multiple independent signals (5,21). However, ICA techniques operate under the assumption that the number of neurons is equal to or less than the number of electrodes, so ICA is not a good fit for spike sorting in the brain, where multiunit activity recorded by four channels of a tetrode often contains 5-10 neurons. ...
... Clustering of the detected spike shapes is employed in the later stages to reduce the signal dimensionality. One of the conventional clustering methods for autonomous unified spike sorting methods is to employ k-means clustering [14,15]. However, the number of clusters (i.e., k) has to be pre-determined. ...
... Some of the mentionable attempts addressing the above mentioned challenges include ICA (Independent component analysis) which is employed in [14] but, performance of ICA deteriorates when the number of electrodes is fewer than the number of neurons under observation. Method described in [1] uses OPTICS; a clustering algorithm to establish prior spike firing knowledge, and spike-fitting based on a modified bayesian technique is used to account for overlaps. ...
... However, more effective approaches have been adopted to minimize the loss of spike detection. As an example, the root-mean-square error of amplitude for the respective channel voltage has been used as a threshold in [14], while the standard deviation of raw voltage has been deployed in [1]. A systematic method has been described in [27], as illustrated in Figure ...
Extracellular recording from living neurons employing microelectrode arrays has attracted paramount attention in recent years as a way to investigate the functionality and disorders of the brain. To decipher useful information from the recorded signals, accurate and efficient neural spike activity detection and sorting becomes an essential prerequisite. Traditional approaches rely on thresholding to detect individual spikes and clustering to identify subset groups; however, these methods fail to identify temporally synchronous spikes due to neuronal synchrony. To address this challenge, we introduce a novel spike sorting algorithm incorporating both quantitative and probabilistic techniques to better approximate the ground truth information of the spike activity. A novel pre-clustering method for identifying key features that can form natural clusters and a dimension reduction technique for identifying the spiking activity are introduced. To address the temporal neuronal synchrony phenomenon leading to detection of multineural overlapped spikes, a procedure for template spike shape estimation and iterative recognition is developed employing the cross correlation methodology tailored to individual neuron's spike rate. A performance comparison between the proposed method and existing techniques in terms of the number of spikes identified and efficiency of sorting the spikes is presented. The outcome shows the effectiveness of the proposed method in identifying temporally synchronous spikes.
... The transformation from raw data of extracellular recordings into spike train estimates is the operation of spike sorting. There is currently no 'standard' method to perform spike sorting, despite an extensive literature (e.g., Brown et al., 2004;Fee et al., 1996;Franke et al., 2010;Kim and Kim, 2003;reviews in Lewicki, 1998;Pouzat et al., 2004;Shoham et al., 2003;Takahashi et al., 2003;Vargas-Irwin and Donoghue, 2007), and a number of methods ...
... The transformation from raw data of extracellular recordings into spike train estimates is the operation of spike sorting. There is currently no 'standard' method to perform spike sorting, despite an extensive literature (e.g. Brown et al., 2004;Fee et al., 1996;Franke et al., 2010;Kim and Kim, 2003;reviews in Lewicki, 1998;Pouzat et al., 2004;Shoham et al., 2003;Takahashi et al., 2003;Vargas-Irwin and Donoghue, 2007), and a number of methods supplemented in commercial and free softwares: e.g. klusters (http://klusters.sourceforge.net/, ...
... Alternatively, wavelet filters may be used instead in order to improve the preservation of the extracellular waveforms (Wiltschko et al., 2008). Several algorithms already exist to handle overlapping spikes (Takahashi et al., 2003;Vargas-Irwin and Donoghue, 2007) but to our knowledge none have yet been applied to cerebellar recordings. ...
The cerebellum is a brain structure involved in coordination complex motor actions such as voluntary movements. To achieve this function, the precise temporal control of a large population of neurons is required. While a large number of patterned population activity has been characterized in many major brain structures (thalamo-cortical system, basal ganglia, hippocampal formation, etc...), very little is currently known in the cerebellum. Therefore, I investigated the presence and characteristics of such an organization in freely-moving rats, especially when they perform a reach-and-grasp task. The cerebellar cortex has a strong topographical organization, such that neighboring cells share similar input sources and output targets. Therefore, studying the local network properties in the cerebellar cortex allows to access to functionally-relevant population activity. First, I demonstrated that multi-wire electrodes, tetrodes, may be used to record multiple neighboring cells in chronic recordings of freely behaving animals using a custom-made microdrive. Second, I examined in the area of the cerebellar cortex controlling limb movements how the principle cells (the Purkinje cells) coordinate their firing during rest and fast forelimb motor action. Using simultaneous electrophysiological recordings of multiple single cells, I found that neighboring Purkinje cells exhibit consistently a co-modulation of their firing rate at time scale of a few milliseconds. This correlated firing is observed during sleep and active exploration, and increases during motor execution. Our results thus indicate that during a fast and complex movement, local assemblies of Purkinje cells form dynamically at short time scales and will produce very transient episodes of inhibition in the deep cerebellar nuclei. Third, in a collaboration with the group of Richard Courtemanche, we studied the link between neuronal firing and slow local field oscillations that are observed in the cerebellum at rest. We found that a large proportion of Golgi cells and Purkinje cells are modulated during the oscillations. These results indicate that these slow oscillations, that may be also observed in the motor cortex, are propagated in the cerebellar cortex. Overall, my work has identified and characterized a number of state-dependent population activity patterns in the cerebellar cortex. How these patterns impact on the motor system largely remains to be understood and should be examined in future studies.
... The problem of temporally overlapping spikes was identified early on (Lewicki 1998), and considerable effort has been made to find solutions. Most of the approaches rely on brute force template matching of all the combinations of participating neurons and temporal shifts (Atiya 1992;Lewicki 1994;Prentice et al. 2011;Segev et al. 2004;Vargas-Irwin and Donoghue 2007;Zhang et al. 2004), independent component analysis (ICA) to demix multielectrode recordings Takahashi et al. 2003), filter-based methods (Franke et al. 2010(Franke et al. , 2015Roberts and Hartline 1975;Stein et al. 1979;Vollgraf et al. 2005) that could resolve overlapping spikes by deconvolution, and methods based on probabilistic frameworks that integrate the possibility of spike overlap (Ekanadham et al. 2014;Franke et al. 2015;Pillow et al. 2013 proaches are successful in estimating spike synchrony in tetrode recordings, and their ability to solve the overlap problem could not be evaluated on experimental data because of the absence of reliable ground truth information in real recordings. ...
... For comparison, we show the result also for the method presented in Franke et al. (2010), which is similar to the BOTM method in that it uses the known templates to construct matched filters. It then tries to resolve the cross talk between the filter outputs in a second processing step that is conceptually close to ICA Takahashi et al. 2003). Its performance was slightly lower than the performance of the BOTM method but also independent of ⌬. ...
... Alternative approaches to the overlapping problem have been suggested, e.g., ICA. In principle, it enables investigators to demix multielectrode recordings (Takahashi et al. 2003). However, a major disadvantage of this method is that the number of electrodes has to be at least as large as the number of signal sources (neurons). ...
Synchronous spike discharge of cortical neurons is thought to be a fingerprint of neuronal cooperativity. Because neighboring neurons are more densely connected to one another than neurons that are located further apart, near-synchronous spike discharge can be expected to be prevalent and it might provide an important basis for cortical computations. Using microelectrodes to record local groups of neurons does not allow for the reliable separation of synchronous spikes from different cells, because available spike sorting algorithms cannot correctly resolve the temporally overlapping waveforms. We show that high spike sorting performance of in vivo recordings, including overlapping spikes, can be achieved using a recently developed filter-based template matching procedure. Using tetrodes with a 3-dimensional structure, we demonstrate with simulated data and ground truth in vitro data, obtained by dual intracellular recording of two neurons located next to a tetrode, that the spike sorting of synchronous spikes can be as successful as the spike sorting of non-overlapping spikes, and that the spatial information provided by multielectrodes greatly reduces the error rates. We apply the method to tetrode recordings from the prefrontal cortex of behaving primates and we show that overlapping spikes can be identified and assigned to individual neurons to study synchronous activity in local groups of neurons.
Copyright © 2014, Journal of Neurophysiology.
... In neuroscience, improved sensors (4) have permitted large increases in the size and dimensionality of recorded datasets. An essential problem remains that the brain contains millions of simultaneously active neurons, emitting action potentials (spikes) with varying frequencies and patterns related to ongoing behavior and brain state (3,5). The identification of signals from individual neurons, in a sea of brain action potential, is critical. ...
... Also, there are often no clear boundaries between the signals of the different neurons recorded, and the density of neurons varies widely across different regions of the brain and across different recording methods (10). Overlapping clusters and strong background activity, produced by neighboring neurons, as well as the similarity of spike waveforms in given classes of neurons, present different problems for algorithms that rely on matching spike waveforms to templates, principal component analysis (PCA), density, and distance metrics (5). Compounding the complexity of the spike-sorting problem, recordings can involve 10-20 "useful" dimensions, especially those that use tetrodes or multisensor probes (8). ...
... Templatematching algorithms also present the task of preparing a template library of each of these neuron shapes. Independent component analysis (ICA) can successfully identify independent sources by decomposing a signal into multiple independent signals (5,21). However, ICA techniques operate under the assumption that the number of neurons is equal to or less than the number of electrodes, so ICA is not a good fit for spike sorting in the brain, where multiunit activity recorded by four channels of a tetrode often contains 5-10 neurons. ...
Significance
Organizing large, multidimensional datasets by subgrouping data as clusters is a major challenge in many fields, including neuroscience, in which the spike activity of large numbers of neurons is recorded simultaneously. We present a mathematical approach for clustering such multidimensional datasets in a relatively high-dimensional space using as a prototype datasets characterized by high background spike activity. Our method incorporates features allowing reliable clustering in the presence of such strong background activity and, to deal with large size of datasets, incorporates automated implementation of clustering. Our approach effectively identifies individual neurons in spike data recorded with multiple tetrodes, and opens the way to use this method in other domains in which clustering of complex datasets is needed.
... For our experiments we selected peak to valley (PV), and energy (E), hereby called PV-E. We remark that the PV-E features contain essentially the same information as the min peak-max peak features used by [25]. Another method based on wavelet transforms that enables visualizing the data in the wavelet coefficient subspace has also been implemented in clustering packages such as Waveclus and Combinato [7,26]. ...
... Such activity may sometimes overlap in time and make multi-unit spike waveforms. The problem of overlapping spikes has been addressed by methods such as independent component analysis, ICA, (if number of electrodes is equal to or more than the number of neurons) [25] or template matching [60]. It is worth noting that in pre-clustering visualization of the data, overlapping spikes (multi-unit spikes) construct a cluster or block. ...
Sorting spikes from extracellular recording into clusters associated with distinct single units (putative neurons) is a fundamental step in analyzing neuronal populations. Such spike sorting is intrinsically unsupervised, as the number of neurons are not known a priori. Therefor, any spike sorting is an unsupervised learning problem that requires either of the two approaches: specification of a fixed value k for the number of clusters to seek, or generation of candidate partitions for several possible values of c, followed by selection of a best candidate based on various post-clustering validation criteria. In this paper, we investigate the first approach and evaluate the utility of several methods for providing lower dimensional visualization of the cluster structure and on subsequent spike clustering. We also introduce a visualization technique called improved visual assessment of cluster tendency (iVAT) to estimate possible cluster structures in data without the need for dimensionality reduction. Experimental results are conducted on two datasets with ground truth labels. In data with a relatively small number of clusters, iVAT is beneficial in estimating the number of clusters to inform the initialization of clustering algorithms. With larger numbers of clusters, iVAT gives a useful estimate of the coarse cluster structure but sometimes fails to indicate the presumptive number of clusters. We show that noise associated with recording extracellular neuronal potentials can disrupt computational clustering schemes, highlighting the benefit of probabilistic clustering models. Our results show that t-Distributed Stochastic Neighbor Embedding (t-SNE) provides representations of the data that yield more accurate visualization of potential cluster structure to inform the clustering stage. Moreover, The clusters obtained using t-SNE features were more reliable than the clusters obtained using the other methods, which indicates that t-SNE can potentially be used for both visualization and to extract features to be used by any clustering algorithm.
... The most common spike detection method is the threshold detector which is widely used in different implementations. In this method, a threshold is applied to the signal and the segments of the signal that cross the threshold are recognized as spikes [2,3]. The threshold detector has a simple structure, but it does not perform well in low signal to noise ratios (SNR). ...
... Thresholding is the most common method for spike detection. A threshold is set for the extracellular signal and the segments of the signal that cross the threshold are considered as spikes [2,3]. In low SNR, the performance of threshold detector is not convenient. ...
Objective:
Many algorithms have been suggested for detection and sorting of spikes in extracellular recording. Nevertheless, it is still challenging to detect spikes in low signal-to-noise ratios (SNR). We propose a spike detection algorithm that is based on the fractal properties of extracellular signals and can detect spikes in low SNR regimes. Semi-intact spikes are low-amplitude spikes whose shapes are almost preserved. The detection of these spikes can significantly enhance the performance of multi-electrode recording systems.
Approach:
Semi-intact spikes are simulated by adding three noise components to a spike train: thermal noise, inter-spike noise, and spike-level noise. We show that simulated signals have fractal properties which make them proper candidates for fractal analysis. Then we use fractal dimension as the main core of our spike detection algorithm and call it fractal detector. The performance of the fractal detector is compared with three frequently used spike detectors.
Main results:
We demonstrate that in low SNR, the fractal detector has the best performance and results in the highest detection probability. It is shown that, in contrast to the other three detectors, the performance of the fractal detector is independent of inter-spike noise power and that variations in spike shape do not alter its performance. Finally, we use the fractal detector for spike detection in experimental data and similar to simulations, it is shown that the fractal detector has the best performance in low SNR regimes.
Significance:
The detection of low-amplitude spikes provides more information about the neural activity in the vicinity of the recording electrodes. Our results suggest using the fractal detector as a reliable and robust method for detecting semi-intact spikes in low SNR extracellular signals.
... Correlation and cluster analysis further confirmed two cell types with the best separation (Figure 5g,h). As previously reported (Bartho et al., 2004;Takahashi et al., 2003), the spike duration of interneurons (0.35-0.40 ms, Figure 5i) was much narrower than that of the pyramidal neurons (0.80-0.85 ms, Figure 5j), and the firing rate of the interneurons was faster than that of the pyramidal neurons (Table S2) GluN2D-containing NMDARs on interneurons, leading to disinhibition for pyramidal neurons ( Figure S7). ...
Background and Purpose
Sevoflurane, a commonly used inhaled anaesthetic known for its favourable safety profile and rapid onset and offset, has not been thoroughly investigated as a potential treatment for depression. In this study, we reveal the mechanism through which sevoflurane delivers enduring antidepressant effects.
Experimental Approach
To assess the antidepressant effects of sevoflurane, behavioural tests were conducted, along with in vitro and ex vivo whole‐cell patch‐clamp recordings, to examine the effects on GluN1–GluN2 incorporated N‐methyl‐d‐aspartate (NMDA) receptors (NMDARs) and neuronal circuitry in the medial prefrontal cortex (mPFC). Multiple‐channel electrophysiology in freely moving mice was performed to evaluate sevoflurane's effects on neuronal activity, and GluN2D knockout (grin2d−/−) mice were used to confirm the requirement of GluN2D for the antidepressant effects.
Key Results
Repeated exposure to subanaesthetic doses of sevoflurane produced sustained antidepressant effects lasting up to 2 weeks. Sevoflurane preferentially inhibited GluN2C‐ and GluN2D‐containing NMDARs, causing a reduction in interneuron activity. In contrast, sevoflurane increased action potentials (AP) firing and decreased spontaneous inhibitory postsynaptic current (sIPSC) in mPFC pyramidal neurons, demonstrating a disinhibitory effect. These effects were absent in grin2d−/− mice, and both pharmacological blockade and genetic knockout of GluN2D abolished sevoflurane's antidepressant actions, suggesting that GluN2D is essential for its antidepressant effect.
Conclusion and Implications
Sevoflurane directly targets GluN2D, leading to a specific decrease in interneuron activity and subsequent disinhibition of pyramidal neurons, which may underpin its antidepressant effects. Targeting the GluN2D subunit could hold promise as a potential therapeutic strategy for treating depression.
... Spike sorting BSS algorithms can be broadly divided by whether they use lower-order or higher-order statistics to separate sources [11]. Lower-order methods find correlations between segmented AP waveforms and follow a pipeline of detection and sorting [3]. ...
p>The decomposition of neurophysiological recordings into their constituent neural sources is of major importance to a diverse range of neuroscientific fields and neuroengineering applications. The advent of high density electrode probes and arrays has driven a major need for novel semi-automated and automated blind source separation methodologies that take advantage of the increased spatial resolution and coverage these new devices offer. Independent component analysis (ICA) offers a principled theoretical framework for such algorithms, but implementation inefficiencies often drive poor performance in practice, particularly for sparse sources. Here we observe that the use of a single non-linear optimization function to identify spiking sources with ICA often has a detrimental effect that precludes the recovery and correct separation of all spiking sources in the signal. We go on to propose a projection-pursuit ICA algorithm designed specifically for spiking sources, which uses a particle swarm methodology to adaptively traverse a polynomial family of non-linearities approximating the asymmetric cumulants of the sources. We robustly prove state-of-the-art decomposition performance on recordings from high density intramuscular probes and demonstrate how the particle swarm quickly finds optimal contrast non-linearities across a range of neurophysiological datasets.</p
... Implantable neural interfacing electrodes serve as the basic research tools for neuroscience [1][2][3][4] as well as the clinical application tools for the diagnosis and treatment of neurological disorders [5][6][7][8][9] , thanks to the functions of electrophysiological recording and/or neural stimulation or modulation. Electrocorticography (ECoG) records the electrical activity in the brain from the sum of the local field potentials of the population of neurons by flexible electrodes implanted on the epidural or subdural surface of the brain 10 . ...
Bacterial cellulose (BC), a natural biomaterial synthesized by bacteria, has a unique structure of a cellulose nanofiber-weaved three-dimensional reticulated network. BC films can be ultrasoft with sufficient mechanical strength, strong water absorption and moisture retention and have been widely used in facial masks. These films have the potential to be applied to implantable neural interfaces due to their conformality and moisture, which are two critical issues for traditional polymer or silicone electrodes. In this work, we propose a micro-electrocorticography (micro-ECoG) electrode named “Brainmask”, which comprises a BC film as the substrate and separated multichannel parylene-C microelectrodes bonded on the top surface. Brainmask can not only guarantee the precise position of microelectrode sites attached to any nonplanar epidural surface but also improve the long-lasting signal quality during acute implantation with an exposed cranial window for at least one hour, as well as the in vivo recording validated for one week. This novel ultrasoft and moist device stands as a next-generation neural interface regardless of complex surface or time of duration.
... They commonly work by finding a linear projection or filter that maximizes a static non-linearity in the data, such as kurtosis, through nonlinear optimization, thus maximizing the impulse-like character of the output. Both ICA [11], [12] and blind deconvolution [13], [14] have been applied to problem of spike detection, although they have not been adopted as standard steps within current approaches to spike sorting, due in part to computational complexity and reliability. ...
Recent advances in the automated analysis of neural spike train data have been driven largely by improved techniques for clustering and the rejection of noise after candidate spikes are identified.
The effectiveness of these procedures depends crucially on the prior step of candidate spike detection, which is still most commonly handled through relatively crude ad hoc methods of filtering and thresholding.
The ideal detector of a spike waveform in noise is a template filter matched to the waveform, yet such a filter presupposes information about the target waveform that is typically unavailable at the outset. Here we exploit a well-known property of higher-order spectra (HOS)---preservation of signal phase up to time shift---to estimate matched filters for spike waveforms before any candidate spikes have been identified. We overcome limitations of previous applications of HOS to this problem, which have generally performed poorly for mixed time series and in the presence of non-Gaussian noise, by employing a recently described decomposition of HOS (HOSD) (Kovach and Howard 2019). For both univariate and multivariate time series, HOSD returns a sequence of spatiotemporal filters matched to distinct waveforms in the signal, simultaneously optimizing the suppression of noise and providing a multidimensional feature space for subsequent detection and sorting.
Using a HOS-based spike sorting toolbox under development, we demonstrate favorable performance against state-of-the-art spike sorters.
... This method is similar to template matching approaches that we will describe later. An alternative approach is Independent Component Analysis (ICA) where the first step demixes blindly the data and extract the individual source signals from which spikes are identified (Takahashi et al., 2003;G. D. Brown et al., 2001;Jäckel et al., 2012). ...
Neurons are the fundamental computing units of the central nervous system. Recent technological advances have made it possible to simultaneously record the activity of thousands of cells. A typical example is the development of microelectrode arrays with thousands of electrodes packed with a high density. A renewed challenge is to spike sort their recorded signals, by extracting the spiking activity of each neuron. I first review the issues associated with spike sorting methods, and compare the algorithms that have been proposed. I then present a new algorithm to sort spikes online from large-scale recordings. Online density-based clustering and template matching are key to reach good performances. The software has been validated on both synthetic and real ground-truth recordings. Finally, I present a specific application on the retina where online spike sorting might be useful. Classically, ganglion cells, the retinal output, are supposed to extract specific features from the visual scene such as increases or decreases of luminance (ON vs OFF cells). However, retinal processing depends on the visual context. Using a novel perturbative approach, I show that the same cell can turn ON or OFF depending on the natural context. I found that a convolutional neural network model fitted to the data can recapitulate context-dependence. Online perturbations are thus a promising tool to probe computations in sensory systems.
... Wave-cluster algorithm well liberated the artificial operation and greatly improved the classification efficiency, but it is not ideal in dealing with the noise interference (Maneesh 1999; Bar-Gad et al. 2001;Takahashi et al. 2002) and the waveform superposition (Segev et al. 2004;Lewicki 1994;Pouzat et al. 2002). In particular, in the case of the waveform superposition, the undetected rate of the wave-cluster method is very high. ...
Acupuncturing the Zusanli (ST 36) point with different types of manual acupuncture manipulations (MAs) and different frequencies can evoke a lot of neural response activities in spinal dorsal root neurons. The action potential is the basic unit of communication in the neural response process. With the rapid development of the electrode acquisition technology, we can simultaneously obtain neural electrical signals of multiple neurons in the target area. So it is crucial to extract spike trains of each neuron from raw recorded data. To solve the problem of variability of the spike waveform, this paper adopts a optimization algorithm based on model to improve the wave-cluster algorithm, which can provide higher accuracy and reliability. Further, through this optimization algorithm, we make a statistical analysis on spike events evoked by different MAs. Results suggest that numbers of response spikes under reinforcing manipulations are far more than reducing manipulations, which mainly embody in synchronous spike activities.
... A proper sorting of spikes with respect to their origin ameliorate considerably the decoding operation. Then, before representing the position of the animal in terms of firing patterns, sorting spikes must be accomplished and the spike's rate of each neuron is calculated [11,12]. Nevertheless, signal instabilities could arise from physical factors like the implants movement, reaction of the tissue or material degradation. ...
Brain-machines capture brain signals in order to restore communication and movement to disabled people who suffer from brain palsy or motor disorders. In brain regions, the ensemble firing of populations of neurons represents spatio-temporal patterns that are transformed into outgoing spatio-temporal patterns which encode complex cognitive task. This transformation is dynamic, non-stationary (time-varying) and highly nonlinear. Hence, modeling such complex biological patterns requires specific model structures to uncover the underlying physiological mechanisms and their influences on system behavior. In this study, a recent multi-electrode technology allows the record of the simultaneous neuron activities in behaving animals. Intra-cortical data are processed according to these steps: spike detection and sorting, than desired action extraction from the rate of the obtained signal. We focus on the following important questions about (i) the possibility of linking the brain signal time events with some time-delayed mapping tools; (ii) the use of some suitable inputs than others for the decoder; (iii) a consideration of separated data or a special representation founded on multi-dimensional statistics. This paper concentrates mostly on the analysis of parallel spike train when certain critical hypotheses are ignored by the data for the working method. We have made efforts to define explicitly whether the underlying hypotheses are actually achieved. In this paper, we propose an algorithm to define the embedded memory order of NARX recurrent neural networks to the hand trajectory tracking process. We also demonstrate that this algorithm can improve performance on inference tasks.
... At the same time, as rates increase, under population methods, the lower fractional yield of well identified spikes still represented many more spikes than at lower rates, albeit flattening the frequency increase somewhat. The complete separation of spatial sites into the 9 activations, means that the local collisions may be mitigated and available for further analysis as tetrode data even at higher rates (Gray et al., 1995;Takahashi et al., 2003;Franke et al., 2010), although here we did not explore this directly. This feature of spatial site separations eliminated significant fractions of collisions directly. ...
The braided multielectrode probe (BMEP) is an ultrafine microwire bundle interwoven into a precise tubular braided structure, which is designed to be used as an invasive neural probe consisting of multiple microelectrodes for electrophysiological neural recording and stimulation. Significant advantages of BMEPs include highly flexible mechanical properties leading to decreased immune responses after chronic implantation in neural tissue and dense recording/stimulation sites (24 channels) within the 100–200 μm diameter. In addition, because BMEPs can be manufactured using various materials in any size and shape without length limitations, they could be expanded to applications in deep central nervous system (CNS) regions as well as peripheral nervous system (PNS) in larger animals and humans. Finally, the 3D topology of wires supports combinatoric rearrangements of wires within braids, and potential neural yield increases. With the newly developed next generation micro braiding machine, we can manufacture more precise and complex microbraid structures. In this article, we describe the new machine and methods, and tests of simulated combinatoric separation methods. We propose various promising BMEP designs and the potential modifications to these designs to create probes suitable for various applications for future neuroprostheses.
... This is because it was almost impossible to accurately separate the firing of neighboring multiple neurons and, therefore, to detect synchronous firing among the neurons by the extracellular recording method, which records neuronal activity for a long time in free-moving animals. However, it became possible by using the spike-sorting method utilizing independent component analysis (ICA; Takahashi et al., 2003). It was demonstrated that approximately 80% of the neuron pairs in the prefrontal cortex in monkeys showed firing correlation and jitters of spike times of 1-5 ms. ...
In this review article we focus on research methodologies for detecting the actual activity of cell assemblies, which are populations of functionally connected neurons that encode information in the brain. We introduce and discuss traditional and novel experimental methods and those currently in development and briefly discuss their advantages and disadvantages for the detection of cell-assembly activity. First, we introduce the electrophysiological method, i.e., multineuronal recording, and review former and recent examples of studies showing models of dynamic coding by cell assemblies in behaving rodents and monkeys. We also discuss how the firing correlation of two neurons reflects the firing synchrony among the numerous surrounding neurons that constitute cell assemblies. Second, we review the recent outstanding studies that used the novel method of optogenetics to show causal relationships between cell-assembly activity and behavioral change. Third, we review the most recently developed method of live-cell imaging, which facilitates the simultaneous observation of firings of a large number of neurons in behaving rodents. Currently, all these available methods have both advantages and disadvantages, and no single measurement method can directly and precisely detect the actual activity of cell assemblies. The best strategy is to combine the available methods and utilize each of their advantages with the technique of operant conditioning of multiple-task behaviors in animals and, if necessary, with brain–machine interface technology to verify the accuracy of neural information detected as cell-assembly activity.
... To sample more members of a circuit, multiple electrodes can be recorded simultaneously, and the shape of the recorded action potential at each different electrode can be used to putatively assign that action potential to a distinct neuron generator (Abeles & Goldstein, 1977), though each added electrode risks traumatizing the neurons of interest. The most frequent electrode configuration employed for extracellular recording is a tetrode, to allow sorting of action potentials into different putative units on the basis of spike morphology at each of the different electrodes (Harris, Henze, Csicsvari, Hirase, & Buzsáki, 2000;Takahashi, Anzai, & Sakurai, 2003), without causing a great deal of damage to the neurons of interest. Yet spike variability over time, especially with different behavioral conditions, can seriously compromise the ability of spike sorting to properly attribute spike activity when compared with intracellular recordings (Stratton et al., 2012). ...
The majority of 20th century investigations into anesthetic effects on the nervous system have used electrophysiology. Yet some fundamental limitations to electrophysiologic recordings, including the invasiveness of the technique, the need to place (potentially several) electrodes in every site of interest, and the difficulty of selectively recording from individual cell types, have driven the development of alternative methods for detecting neuronal activation. Two such alternative methods with cellular scale resolution have matured in the last few decades and will be reviewed here: the transcription of immediate early genes, foremost c-fos, and the influx of calcium into neurons as reported by genetically encoded calcium indicators, foremost GCaMP6. Reporters of c-fos allow detection of transcriptional activation even in deep or distant nuclei, without requiring the accurate targeting of multiple electrodes at long distances. The temporal resolution of c-fos is limited due to its dependence upon the detection of transcriptional activation through immunohistochemical assays, though the development of RT-PCR probes has shifted the temporal resolution of the assay when tissues of interest can be isolated. GCaMP6 has several isoforms that trade-off temporal resolution for signal to noise, but the fastest are capable of resolving individual action potential events, provided the microscope used scans quickly enough. GCaMP6 expression can be selectively targeted to neuronal populations of interest, and potentially thousands of neurons can be captured within a single frame, allowing the neuron-by-neuron reporting of circuit dynamics on a scale that is difficult to capture with electrophysiology, as long as the populations are optically accessible.
... In the literature, this step is simply known as splitting and merging, either manually 15,24,25 or systematically. [26][27][28] Apart from this error analysis, an ideal clustering technique should be robust against modes of firing, namely, regular spiking versus bursting, 29-31 multispike trains 7,21,32,33 and local field potentials. 34,35 Classical cluster analysis techniques do not easily generalize to other unsupervised data partitioning tasks. ...
An important goal in visual neuroscience is to understand how neuronal population coding in vertebrate retina mediates the broad range of visual functions. Microelectrode arrays interface on isolated retina registers a collective measure of the spiking dynamics of retinal ganglion cells (RGCs) by probing them simultaneously and in large numbers. The recorded data stream is then processed to identify spike trains of individual RGCs by efficient and scalable spike detection and sorting routines. Most spike sorting software packages, available either commercially or as freeware, combine automated steps with judgment calls by the investigator to verify the quality of sorted spikes. This work focused on sorting spikes of RGCs into clusters using an integrated analytical platform for the data recorded during visual stimulation of wild-type mice retinas with whole field stimuli. After spike train detection, we projected each spike onto two feature spaces: a parametric space and a principal components space. We then applied clustering algorithms to sort spikes into separate clusters. To eliminate the need for human intervention, the initial clustering results were submitted to diagnostic tests that evaluated the results to detect the sources of failure in cluster assignment. This failure diagnosis formed a decision logic for diagnosable electrodes to enhance the clustering quality iteratively through rerunning the clustering algorithms. The new clustering results showed that the spike sorting accuracy was improved. Subsequently, the number of active RGCs during each whole field stimulation was found, and the light responsiveness of each RGC was identified. Our approach led to error-resilient spike sorting in both feature extraction methods; however, using parametric features led to less erroneous spike sorting compared to principal components, particularly for low signal-to-noise ratios. As our approach is reliable for retinal signal processing in response to simple visual stimuli, it could be applied to the evaluation of disrupted physiological signaling in retinal neurodegenerative diseases.
... Tetrodes, with wires brought down to a low impedance, can also be used to record stable LFPs [11] and units as well, enclosed in small microdrives [12,13]. Tetrode recordings, by quadrangulating in a few tens of μm the various sources of voltage, along with the use of powerful software algorithms, contribute to advantageous unit detection, along with LFP recordings [14][15][16][17]. This can be done in vitro, in a rodent brain slice of approximately 450 μm [18]. ...
This chapter introduces basic and some advanced methods for recording, analyzing, and comparing local network electrophysiological activity in rodents and primates. Attention will be paid to the acquisition of network signals that consider the local field potentials (LFPs) and single-unit and multiunit activity. Analysis methods for extracting main features from LFP signals, their frequency, power, and coherence will be discussed, as well as the relation of unit activity with the LFPs. The relationship with movement and behavior will be developed, and so will the relation of these measures with network activity in specific pathologies of the local networks.
... This method is similar to template matching approaches that we will describe later. An alternative approach is independent component analysis (ICA) where the first step demix blindly the data and extract the individual source signals from which spikes are identified (Takahashi et al., 2003;Brown et al., 2001;Jäckel et al., 2012). Note that variants, such as the convolutional independent component analysis (cICA) of Leibig et al. (2016), has been developed. ...
In recent years, arrays of extracellular electrodes have been developed and manufactured to record simultaneously from hundreds of electrodes packed with a high density. These recordings should allow neuroscientists to reconstruct the individual activity of the neurons spiking in the vicinity of these electrodes, with the help of signal processing algorithms. Algorithms need to solve a source separation problem, also known as spike sorting. However, these new devices challenge the classical way to do spike sorting. Here we review different methods that have been developed to sort spikes from these large-scale recordings. We describe the common properties of these algorithms, as well as their main differences. Finally, we outline the issues that remain to be solved by future spike sorting algorithms.
... Details of spike sorting were reported previously (Takahashi et al., 2003). Recorded spike trains were sorted to isolate individual neuronal activities using ICSort, a spike sorting method with independent component analysis (Fast ICA) and KlustaKwik (Takahashi, 2013). ...
Planning of multi-step actions based on the retrieval of acquired information is essential for efficient foraging. The hippocampus (HPC) and prefrontal cortex (PFC) may play critical roles in this process. However, in rodents, many studies investigating such roles utilized T-maze tasks that only require one-step actions (i.e., selection of one of two alternatives), in which memory retrieval and selection of an action based on the retrieval cannot be clearly differentiated. In monkeys, PFC has been suggested to be involved in planning of multi-step actions; however, the synchrony between HPC and PFC has not been evaluated. To address the combined role of the regions in planning of multi-step actions, we introduced a task in rats that required three successive nose-poke responses to three sequentially illuminated nose-poke holes. During the task, local field potentials (LFP) and spikes from hippocampal CA1 and medial PFC (mPFC) were simultaneously recorded. The position of the first hole indicated whether the following two holes would be presented in a predictable sequence or not. During the first nose-poke period, phase synchrony of LFPs in the theta range (4-10 Hz) between the regions was not different between predictable and unpredictable trials. However, only in trials of predictable sequences, the magnitude of theta phase synchrony during the first nose-poke period was negatively correlated with latency of the two-step ahead nose-poke response. Our findings point to the HPC-mPFC theta phase synchrony as a key mechanism underlying planning of multi-step actions based on memory retrieval rather than the retrieval itself. This article is protected by copyright. All rights reserved.
... Measurement noise may come from direct physical sources like, for instance, instabilities and movement in the tissue surrounding the recording electrodes, noise properties of the recording devices themselves, the mere fact that only a fraction of all system variables is experimentally accessed ('sampling noise'), or may result from preprocessing steps like spike sorting (e.g. [62,63]). This is therefore a quite different scenario from the comparatively lowdimensional and low-noise situations in, e.g., laser physics [64], and delay-embedding-based approaches to the reconstruction of neural dynamics may have to be augmented by machine learning techniques to retrieve at least some of its most salient features [7,8]. ...
The computational properties of neural systems are often thought to be implemented in terms of their network dynamics. Hence, recovering the system dynamics from experimentally observed neuronal time series, like multiple single-unit (MSU) recordings or neuroimaging data, is an important step toward understanding its computations. Ideally, one would not only seek a state space representation of the dynamics, but would wish to have access to its governing equations for in-depth analysis. Recurrent neural networks (RNNs) are a computationally powerful and dynamically universal formal framework which has been extensively studied from both the computational and the dynamical systems perspective. Here we develop a semi-analytical maximum-likelihood estimation scheme for piecewise-linear RNNs (PLRNNs) within the statistical framework of state space models, which accounts for noise in both the underlying latent dynamics and the observation process. The Expectation-Maximization algorithm is used to infer the latent state distribution, through a global Laplace approximation, and the PLRNN parameters iteratively. After validating the procedure on toy examples, the approach is applied to MSU recordings from the rodent anterior cingulate cortex obtained during performance of a classical working memory task, delayed alternation. A model with 5 states turned out to be sufficient to capture the essential computational dynamics underlying task performance, including stimulus-selective delay activity. The estimated models were rarely multi-stable, but rather were tuned to exhibit slow dynamics in the vicinity of a bifurcation point. In summary, the present work advances a semi-analytical (thus reasonably fast) maximum-likelihood estimation framework for PLRNNs that may enable to recover the relevant dynamics underlying observed neuronal time series, and directly link them to computational properties.
... This method is similar to template matching approaches that we will describe later. An alternative approach is independent component analysis (ICA) where the first step demix blindly the data and extract the individual source signals from which spikes are identified (Takahashi et al., 2003;Brown et al., 2001;Jäckel et al., 2012). Note that variants, such as the convolutional independent component analysis (cICA) of Leibig et al. (2016), has been developed. ...
In recent years, arrays of extracellular electrodes have been developed and manufactured to record simultaneously from hundreds of electrodes packed with a high density. These recordings should allow neuroscientists to reconstruct the individual activity of the neurons spiking in the vicinity of these electrodes, with the help of signal processing algorithms. Algorithms need to solve a source separation problem, also known as spike sorting. However, these new devices challenge the classical way to do spike sorting. Here we review different methods that have been developed to sort spikes from these large-scale recordings. We describe the common properties of these algorithms, as well as their main differences. Finally, we outline the issues that remain to be solved by future spike sorting algorithms.
... ICA was shown to be useful in modalities such as EEG and MEG. It has also been successfully applied to multi-neuronal recordings [68]. Mostly, it is used for artifact reduction [61], [152] or for real-time control in BCI applications [65]. ...
In the past decade, numerous advances in the study of the human brain were fostered by successful applications of blind source separation (BSS) methods to a wide range of imaging modalities. The main focus has been on extracting “networks” represented as the underlying latent sources. While the broad success in learning latent representations from multiple datasets has promoted the wide presence of BSS in modern neuroscience, it also introduced a wide variety of objective functions, underlying graphical structures, and parameter constraints for each method. Such diversity, combined with a host of datatype-specific knowhow, can cause a sense of disorder and confusion, hampering a practitioner’s judgment and impeding further development. We organize the diverse landscape of BSS models by exposing its key features and combining them to establish a novel unifying view of the area. In the process, we unveil important connections among models according to their properties and subspace structures. Consequently, a high-level descriptive structure is exposed, ultimately helping practitioners select the right model for their applications. Equipped with that knowledge, we review the current state of BSS applications to neuroimaging. The gained insight into model connections elicits a broader sense of generalization, highlighting several directions for model development. In light of that, we discuss emerging multi-dataset multidimensional (MDM) models and summarize their benefits for the study of the healthy brain and disease-related changes.
... The results of spikes detected by selective sorting is compared with a number of popularly adopted techniques including noise standard deviation, root mean square of channel voltage and standard deviation of the channel 4,8,[12][13][14] . Figure 8(a) demonstrates the superiority of selective sorting to identify a larger number of spikes. ...
Extracellular data analysis has become a quintessential method for understanding the neurophysiological responses to stimuli. This demands stringent techniques owing to the complicated nature of the recording environment. In this paper, we highlight the challenges in extracellular multi-electrode recording and data analysis as well as the limitations pertaining to some of the currently employed methodologies. To address some of the challenges, we present a unified algorithm in the form of selective sorting. Selective sorting is modelled around hypothesized generative model, which addresses the natural phenomena of spikes triggered by an intricate neuronal population. The algorithm incorporates Cepstrum of Bispectrum, ad hoc clustering algorithms, wavelet transforms, least square and correlation concepts which strategically tailors a sequence to characterize and form distinctive clusters. Additionally, we demonstrate the influence of noise modelled wavelets to sort overlapping spikes. The algorithm is evaluated using both raw and synthesized data sets with different levels of complexity and the performances are tabulated for comparison using widely accepted qualitative and quantitative indicators.
... k-means (Chan et al. (2008)), superparamagnetic (Quiroga et al. (2004)) and hierachical adaptive means (HAM) (Paraskevopoulou et al. (2014)), have been proposed for better accuracy. Overlapping spikes can be handled by either iteration and subtraction (Lewicki (1994); Zhang et al. (2004); Prentice et al. (2011); Pillow et al. (2013); Ekanadham et al. (2014)) or independent component analysis (Takahashi et al. (2003); Takahashi and Sakurai (2005); Jäckel et al. (2012)). Most of these methods are off-line, requiring the collection of all spikes before running the analysis, unsuitable for real-time closed-loop applications. ...
Background:
Computationally efficient spike recognition methods are required for real-time analysis of extracellular neural recordings. The enteric nervous system (ENS) is important to human health but less well-understood with few appropriate spike recognition algorithms due to large waveform variability.
New method:
Here we present a method based on dynamic time warping (DTW) with high tolerance to variability in time and magnitude. Adaptive temporal gridding for "fastDTW" in similarity calculation significantly reduces the computational cost. The automated threshold selection allows for real-time classification for extracellular recordings.
Results:
Our method is first evaluated on synthesized data at different noise levels, improving both classification accuracy and computational complexity over the conventional cross-correlation based template-matching method (CCTM) and PCA + k-means clustering without time warping. Our method is then applied to analyze the mouse enteric neural recording with mechanical and chemical stimuli. Successful classification of biphasic and monophasic spikes is achieved even when the spike variability is larger than millisecond in width and millivolt in magnitude.
Comparison with existing method(s):
In comparison with conventional template matching and clustering methods, the fastDTW method is computationally efficient with high tolerance to waveform variability.
Conclusions:
We have developed an adaptive fastDTW algorithm for real-time spike classification of ENS recording with large waveform variability against colony motility, ambient changes and cellular heterogeneity.
... In fact in real recordings these waveforms are likely to signicantly deviate from linear dependence. Nonetheless, linear dependence between recording channels is assumed especially by spike sorting procedures based on independent component analysis (ICA) (Takahashi et al. (2002(Takahashi et al. ( , 2003b; Madanymamlouk et al. (2005); Takahashi and Sakurai (2005)). ...
For the understanding of how neural networks grow, learn and are able to fulfill their impressive functions, a reliable way to monitor their activity is crucial. The neuroscientist has a steadily growing toolbox for this purpose with one tool being of particular importance even though it is also one of the oldest techniques applied, namely extracellular recordings. The main reasons for its broad usage are probably the low cost and relative ease of application: an electrode needs to be placed in the vicinity of a neuron. Since the most important way of neurons to communicate is by means of electrical events called action potentials, extracellular recordings provide the possibility to "listen" to the communication in a neural network. A major drawback of that technique, however, is the complexity of the analysis of the recordings that is caused by low signal quality as well as the measurement inherent simultaneous recording of not only one but many neurons whose signals need to be separated. Despite the long lasting history of extracellular recordings a satisfactory solution to its associated problem of detection and correct classification of single neuronal action potentials - called spike sorting - is still not readily available. Especially with the massive increase of its usage and the development of more sophisticated electrode arrays that are in principle able to record over long time periods from hundreds of neurons simultaneously, extracellular recordings will also play a major role for future neuroscience, brain machine interfaces and medical applications. But these applications make great demands on the algorithms used: because of the huge amount of extracellular recordings humans will not be able to supervise the spike sorting process any longer so the algorithms need to be fully automatic. For so called closed-loop experiments, where the spike sorting result will be e.g. used to stimulate other neurons in real-time the spike sorting procedures also have to provide the results in real-time. The quickly increasing number of recording electrodes will force the methods to optimally combine information from all available electrodes and deal with a higher computational burden. And finally, in long lasting experiments slow changes in the recording setup like changes in electrode position will cause non-adaptive algorithms to fail. This work is the investigation of the applicability of linear filters for the purpose of fast automatic and adaptive spike detection and sorting. Linear filters provide the advantage of easy implementation - also in hardware - are computationally fast and embedded in a well developed theory. The calculations in this work are done in the discrete-time signal space which simplifies the derivations and shows interesting connections of the spike sorting problem to optimal matched filters, beamforming, space-time adaptive processing, time series forecasting and Wiener filtering. The role of the noise covariance matrix that arises naturally in these domains and is also crucial for many - especially linear-filter based - spike sorting algorithms is investigated. Based on those filters two spike sorting algorithms are proposed that are suited for real-time implementation, can adapt to non-stationary data and make optimal use of multielectrode recordings. The first algorithm uses optimal filters and successive source separation to maximize the signal to noise ratio and demix the single neuronal signals. The second approach shows that the filter output of the optimal matched filters can be interpreted in a Bayesian sense and can this way be used to derive a linear discriminant function based spike sorter. The performance of the methods is compared to that of others on simulated as well as unique real experimental data that is especially suited for that purpose and the real-time ability of the algorithms is discussed. However, the principal problem of benchmarking the quality of a spike sorting procedure on real data and the burdensome lack of a widely accepted benchmark pose a serious challenge. A way to overcome that obstacle with a community approach is proposed. A platform to host this approach is implemented in form of a website for the automatic and blind evaluation of spike sorting algorithms.
... Details of spike sorting were previously reported (Takahashi et al., 2003a,b; Sakurai and Takahashi, 2006). Recorded spike trains were sorted to isolate the individual neuronal activities by a method of independent component analysis (ICA) and k-means clustering called ICsort (Takahashi et al., 2003a,b). ...
The discovery of time cells revealed that the rodent hippocampus has information regarding time. Previous studies have suggested that the role of hippocampal time cells is to integrate temporally segregated events into a sequence using working memory with time perception. However, it is unclear whether hippocampal cells contribute to time perception itself because most previous studies employed delayed matching-to-sample tasks that did not separately evaluate time perception from working memory processes. Here, we investigated the function of the rat hippocampus in time perception using a temporal discrimination task. In the task, rats had to discriminate between durations of 1 and 3 s to get a reward, and maintaining task-related information as working memory was not required. We found that some hippocampal neurons showed firing rate modulation similar to that of time cells. Moreover, theta oscillation of local field potentials (LFPs) showed a transient enhancement of power during time discrimination periods. However, there were little relationships between the neuronal activities and theta oscillations. These results suggest that both the individual neuronal activities and theta oscillations of LFPs in the hippocampus have a possibility to be engaged in seconds order time perception; however, they participate in different ways.
... However, if the spikes are too close they can create complex waveforms, which in turn will difficult the cluster isolation and may even give rise to new clusters. The problem of overlapping spikes is one of the most challenging issues in spike sorting and although several methods have been proposed (Hulata et al., A c c e p t e d M a n u s c r i p t 9 2002; Lewicki 1994;Pillow et al., 2013;Prentice et al., 2011;Takahashi et al., 2003), there is no optimal approach to deal with it. In this regard, the use of tetrodes or multielectrode probes can be helpful to deal with this issue since what appears as an overlap on one channel might be an isolated unit on another (Lewicki 1998). ...
Spike sorting is a crucial step to extract information from extracellular recordings. With new recording opportunities provided by the development of new electrodes that allow monitoring hundreds of neurons simultaneously, the scenario for the new generation of algorithms is both exciting and challenging. However, this will require a new approach to the problem and the development of a common reference framework to quickly assess the performance of new algorithms. In this work, we review the basic concepts of spike sorting, including the requirements for different applications, together with the problems faced by presently available algorithms. We conclude by proposing a roadmap stressing the crucial points to be addressed to support the neuroscientific research of the near future.
Copyright © 2015 The Authors. Published by Elsevier Inc. All rights reserved.
... In [6] , Principal Component Analysis (PCA) is introduced to group spikes through analyzing spikes to get several principal components and projecting spikes into each component. Takahashi et al. apply Independent Component Analysis (ICA) [7, 8] to separate spikes. As both PCA-and ICA-based methods require strong spike correlation and variances, they will not work well in case of low signal to noise ratio (SNR). ...
In this paper, hidden Markov models (HMM) is studied for spike sorting. We notice that HMM state sequences have capability to represent spikes precisely and concisely. We build a HMM for spikes, where HMM states respect spike significant shape variations. Four shape variations are introduced: silence, going up, going down and peak. They constitute every spike with an underlying probabilistic dependence that is modelled by HMM. Based on this representation, spikes sorting becomes a classification problem of compact HMM state sequences. In addition, we enhance the method by defining HMM on extracted Cepstrum features, which improves the accuracy of spike sorting. Simulation results demonstrate the effectiveness of the proposed method as well as the efficiency.
... In this study, more automated clustering techniques must be attempted. Combination of Kmeans with another clustering technique is one example, as shown by Takahashi et al [34], where each class obtained from the K-means is split and later (similar ones) are agglomerated to overcome the problem of overlapped classes. In all assessments, a meticulous visual assessment of the results is essential, ...
Background
Fasciculation potentials (FPs) are important in supporting the electrodiagnosis of Amyotrophic Lateral Sclerosis (ALS). If classified by shape, FPs can also be very informative for laboratory-based neurophysiological investigations of the motor units.
Methods
This study describes a Matlab program for classification of FPs recorded by multi-channel surface electromyogram (EMG) electrodes. The program applies Principal Component Analysis on a set of features recorded from all channels. Then, it registers unsupervised and supervised classification algorithms to sort the FP samples. Qualitative and quantitative evaluation of the results is provided for the operator to assess the outcome. The algorithm facilitates manual interactive modification of the results. Classification accuracy can be improved progressively until the user is satisfied. The program makes no assumptions regarding the occurrence times of the action potentials, in keeping with the rather sporadic and irregular nature of FP firings.
Results
Ten sets of experimental data recorded from subjects with ALS using a 20-channel surface electrode array were tested. A total of 11891 FPs were detected and classified into a total of 235 prototype template waveforms. Evaluation and correction of classification outcome of such a dataset with over 6000 FPs can be achieved within 1–2 days. Facilitated interactive evaluation and modification could expedite the process of gaining accurate final results.
Conclusion
The developed Matlab program is an efficient toolbox for classification of FPs.
... However, these methods generally rely on brute-force examination of all combinations of spike waveforms at all time separations (impractical for simultaneous recordings of many cells), or use of "greedy algorithms that iteratively subtract the waveform of the best-fitting cell until the residual amplitude is within the range expected for noise. A notable exception is the family of ICA-based spike sorting methods [138,139,49], which bear some resemblance to our approach, but have not been developed or implemented in the context of a unified probabilistic model for the voltage measurements, and have not been extensively tested and compared to traditional clustering methods. ...
Understanding neural circuit functions requires tools that can interrogate the activity of a large number of neurons over extended periods of time. Implantable neural probes can record and modulate neuronal activities at high spatiotemporal resolution, representing one of the most applied tools in basic and applied neuroscience. Over the past decade, implantable neural probe technologies have evolved considerably in terms of capability and reliability. This article focuses on the latest developments in implantable neural probe technologies that are enabled by advanced materials and processing strategies. We highlight implantable neural probes that can allow for large-scale and long-lasting neural activity recordings. In addition, we describe recent developments in multifunctional neural probes for combined electrophysiological recording and modulation functionalities. The wide dissemination and clinical translation of these technologies will rapidly advance our understanding of the brain and provide new opportunities for the treatment of neurological diseases.Graphical abstract
Sorting spikes from extracellular recording into clusters associated with distinct single units (putative neurons) is a fundamental step in analyzing neuronal populations. Such spike sorting is intrinsically unsupervised, as the number of neurons are not known a priori. Therefor, any spike sorting is an unsupervised learning problem that requires either of the two approaches: specification of a fixed value c for the number of clusters to seek, or generation of candidate partitions for several possible values of c, followed by selection of a best candidate based on various post-clustering validation criteria. In this paper, we investigate the first approach and evaluate the utility of several methods for providing lower dimensional visualization of the cluster structure and on subsequent spike clustering. We also introduce a visualization technique called improved visual assessment of cluster tendency (iVAT) to estimate possible cluster structures in data without the need for dimensionality reduction. Experimental results are conducted on two datasets with ground truth labels. In data with a relatively small number of clusters, iVAT is beneficial in estimating the number of clusters to inform the initialization of clustering algorithms. With larger numbers of clusters, iVAT gives a useful estimate of the coarse cluster structure but sometimes fails to indicate the presumptive number of clusters. We show that noise associated with recording extracellular neuronal potentials can disrupt computational clustering schemes, highlighting the benefit of probabilistic clustering models. Our results show that t-Distributed Stochastic Neighbor Embedding (t-SNE) provides representations of the data that yield more accurate visualization of potential cluster structure to inform the clustering stage. Moreover, The clusters obtained using t-SNE features were more reliable than the clusters obtained using the other methods, which indicates that t-SNE can potentially be used for both visualization and to extract features to be used by any clustering algorithm.
Spike sorting for neuron recordings is one of the core tasks in brain function studies. Spike sorting always consists of spike detection, feature extraction and clustering. Most of the clustering algorithms adopted in spike sorting schemes are subject to the shapes and structures of the signal except the spectral clustering algorithm. To improve the performance of spectral clustering algorithm for spike sorting, in this paper, a locally weighted co-association matrix is employed as the similarity matrix and the Shannon entropy is also introduced to measure the dependability of clustering. Experimental results show that the performance of spike sorting with the improved spectral clustering algorithm is superior to that of spike sorting with other classic clustering algorithms.
Hippocampal oscillations, particularly theta (6-12 Hz) and gamma (30-90 Hz) frequency bands, play an important role in several cognitive functions. Theta and gamma oscillations show cross-frequency coupling (CFC), wherein the phase of theta rhythm modulates the amplitude of the gamma oscillation, and this CFC is believed to reflect cell assembly dynamics in cognitive processes. Previous studies have reported that CFC strength correlates with the learning process. However, details on these dynamic correlations have not been elucidated. In the present study, we analyzed local field potentials recorded from the rat hippocampus during the learning of a rule-switching task. The modulation index, an index of the CFC strength, became higher in rule-guided behavior than in the no rule condition. The enhanced coupling between theta and high-gamma oscillations (60-90 Hz) changed during the late stage of learning. In contrast, the coupling between theta and low-gamma oscillations (30-60 Hz) did not show any changes during learning. These results suggest that the coupling between theta and gamma bands occurs during rule learning and that high- and low-gamma bands play different roles in rule switching.
In our minds we can vividly re-experience a series of past events or episodes that occurred along space and time. In 1957, a highly influential clinical investigation published by Scoville and Milner suggested that episodic memory retrieval is severely impaired after physical damage occurs to the hippocampus. In fact, loss of the hippocampus can lead to a profound disturbance to spatial navigation. Within the hippocampus reside place cells: excitatory pyramidal cells that maximally fire at a particular location. The presence of hippocampal place cells across several mammalian species, including mice, rats, bats, and primates, has led to a theory that the hippocampus is a locus of the ‘where’ element of episodic memory. However, growing evidence suggests that place cell’s firings do not merely encode spatial information, but also represent a series of experienced events that occur along the spatiotemporal continuum. This suggestion inspires the following questions: is the ensemble activity of place cells a neuronal substrate for forming or recalling episodic memory? And, why does the hippocampus enable flexible spatial navigation?
For the purpose of visualization and for the ease of interpretation, to remove redundancies from the data or to combat the curse of dimensionality (Sect. 4.4), it may be useful to reduce the dimensionality of the original p-dimensional feature space. This, of course, should be done in a way that minimizes the potential loss of information, where the precise definition of “loss of information” may depend on the statistical and scientific questions asked. There are both linear and nonlinear methods for dimensionality reduction. This chapter will start with the by far most popular procedure, principal component analysis.
The computational and cognitive properties of neural systems are often thought to be implemented in terms of their (stochastic) network dynamics. Hence, recovering the system dynamics from experimentally observed neuronal time series, like multiple single-unit recordings or neuroimaging data, is an important step toward understanding its computations. Ideally, one would not only seek a (lower-dimensional) state space representation of the dynamics, but would wish to have access to its statistical properties and their generative equations for in-depth analysis. Recurrent neural networks (RNNs) are a computationally powerful and dynamically universal formal framework which has been extensively studied from both the computational and the dynamical systems perspective. Here we develop a semi-analytical maximum-likelihood estimation scheme for piecewise-linear RNNs (PLRNNs) within the statistical framework of state space models, which accounts for noise in both the underlying latent dynamics and the observation process. The Expectation-Maximization algorithm is used to infer the latent state distribution, through a global Laplace approximation, and the PLRNN parameters iteratively. After validating the procedure on toy examples, and using inference through particle filters for comparison, the approach is applied to multiple single-unit recordings from the rodent anterior cingulate cortex (ACC) obtained during performance of a classical working memory task, delayed alternation. Models estimated from kernel-smoothed spike time data were able to capture the essential computational dynamics underlying task performance, including stimulus-selective delay activity. The estimated models were rarely multi-stable, however, but rather were tuned to exhibit slow dynamics in the vicinity of a bifurcation point. In summary, the present work advances a semi-analytical (thus reasonably fast) maximum-likelihood estimation framework for PLRNNs that may enable to recover relevant aspects of the nonlinear dynamics underlying observed neuronal time series, and directly link these to computational properties.
Hippocampal oscillations, particularly theta (5–10 Hz) and gamma (30–90 Hz) bands, play important roles in several cognitive functions. Theta and gamma oscillations show cross-frequency coupling (CFC), and the theta rhythm phase can modulate the amplitude of gamma oscillations. This phase–amplitude modulation is believed to have several cognitive roles. Previous studies found that the strength of CFC is correlated with task performance during learning. To explore this correlation in more detail, we recorded local field potentials from the hippocampus of rats while they were performing a rule-switching task. Our preliminary data suggested that the coupling between theta and high-gamma oscillations changed through the learning stages, but the coupling between theta and low-gamma oscillations did not show these changes. These results suggest that high-gamma and low-gamma oscillations play different roles in rule switching.
Deep brain stimulation (DBS) has been used as a treatment of brain diseases such as Parkinson's disease and is a promising therapy for epilepsy. But the mechanisms of high frequency stimulation (HFS) used by DBS are still uncertain. In order to investigate the changes of action penitential firings of individual neurons (single unit activity, SUA) during the period of HFS, a new algorithm based on window detection was designed to detect spikes in broadband-frequency recording signals. The results show that orthodromic-HFS (O-HFS) could excite the neurons in CA1 regions, and the firing rate of interneurons and pyramidal neurons increased significantly. In particular, a decrease in spike amplitude for both interneurons and pyramidal neurons was observed during the period of O-HFS. The amplitude decrease of unit spikes was most remarkable with the presence of HFS-induced population spike (PS). These results suggest that the stimulation pulses of O-HFS could activate the downstream neurons continuously, leading to the downstream neurons being unable to repolarize completely. The results are important for tracking individual neuron activity during HFS and for further understanding of DBS mechanisms.
Unsupervised identification of action potentials in multi-channel extracellular recordings, in particular from high-density microelectrode arrays with thousands of sensors, is an unresolved problem. While independent component analysis (ICA) achieves rapid unsupervised sorting, it ignores the convolutive structure of extracellular data, thus limiting the unmixing to a subset of neurons.
Here we present a spike sorting algorithm based on convolutive ICA (cICA) to retrieve a larger number of accurately sorted neurons than with instantaneous ICA while accounting for signal overlaps. Spike sorting was applied to datasets with varying signal-to-noise ratios (SNR: 3 to 12) and 27% spike overlaps, sampled at either 11.5 or 23_kHz_ on 4365 electrodes.
We demonstrate how the instantaneity assumption in ICA-based algorithms has to be relaxed in order to improve the spike sorting performance for high-density microelectrode array recordings. Reformulating the convolutive mixture as an instantaneous mixture by modeling several delayed samples jointly is necessary to increase signal-to-noise ratio. Our results emphasize that different cICA algorithms are not equivalent.
Spike sorting performance was assessed with ground-truth data generated from experimentally derived templates. The presented spike sorter was able to extract ≈90% of the true spike trains with an error rate below 2%. It was superior to two alternative (c)ICA methods (≈80% accurately sorted neurons) and comparable to a supervised sorting.
Our new algorithm represents a fast solution to overcome the current bottleneck in spike sorting of large datasets generated by simultaneous recording with thousands of electrodes.
In awaking and working brains, it is still not clear how dynamic the neuronal activity and their functional connectivity are and how high potentials for learning they have. Our research project will detect actual features of the dynamics and learning potentials of the working brains by constructing and applying a high-performance brain-machine interface (BMI). With the BMI system, neuronal activities of animals learning behavioral tasks directly control machine-outputs instead of animal’s body-outputs and learn to perform behavioral tasks correctly. The keys to construct such a successful BMI are training animals in appropriate behavioral tasks, recording multi-neuronal activities from the behaving animals for long periods, and decoding neuronal code representing valid information in the working brain. The neuronal code might be expressed in synchronized activity among neighboring neurons which could consist a cell assembly. We, therefore, have developed a method to detect precise sub-millisecond activity interactions among closely neighboring neurons in the brains of behaving animals. The system uses a combination of independent component analysis (ICA) and newly developed multi-electrodes.
The present chapter discusses why cell-assembly coding, i.e., ensemble coding by functionally connected neurons, is an appropriate view of the brain's neuronal code and how it operates in the working brain. The cell-assembly coding has two major properties, i.e., partial overlapping of neurons among assemblies and connection dynamics within and among the assemblies. The former is the ability of one neuron to participate in different types of information processing. The latter is the capability for functional synaptic connections, detected by synchrony of firing of the neurons, to change among different types of information processing. Examples of experiments which detected these two major properties are then given. Several relevant points concerning the detection of cell assemblies and dual-coding by cell assemblies and single neurons are also enumerated. Finally, technical and theoretical improvements necessary for future researches of cell-assembly coding are discussed. They include an unique technique of spike-sorting with independent component analysis and theories of sparse coding by distributed overlapped assemblies and coincidence detection as a role of individual neurons to bind distributed neurons into cell assemblies.
The present review discusses why multi-neuronal recording is requisite to detect real features of information processing in the working brain. We introduce several technical key points and to-be-solved problems in the recording experiments. Then we refer to “cell-assembly”, ensemble activity of functionally connected neurons, as a functional unit representing neural information in the brain. Several relevant points concerning detection of the actual dynamics of cell assemblies are enumerated. Finally we introduce a rapidly growing research subject, brain-machine interface (BMI). It tries to detect and utilize neural information by cell assemblies in the working brain to operate artificial devices instead of the animal or human bodies. Multi-neuronal recording and analyzing systems detecting actual cell assemblies are keys to construct successful BMI. We finally show a newly developed and high performance system for our BMI research, which can detect precise submillisecond activity interactions among closely neighboring neurons in the behaving animal. (Japanese Journal of Physiological Psychology and Psychophysiology, 24 (1) : 57-67, 2006.)
Extracellular recordings of multi-unit neural activity have become indispensable in neuroscience research. The analysis of the recordings begins with the detection of the action potentials (APs), followed by a classification step where each AP is associated with a given neural source. A feature extraction step is required prior to classification in order to reduce the dimensionality of the data and the impact of noise, allowing source clustering algorithms to work more efficiently. New Method: In this paper, we propose a novel framework for multi-sensor AP feature extraction based on the so-called Matched Subspace Detector (MSD), which is shown to be a natural generalization of standard single-sensor algorithms.
Clustering using both simulated data and real AP recordings taken in the locust antennal lobe demonstrates that the proposed approach yields features that are discriminatory and lead to promising results. Comparison with Existing Method(s): Unlike existing methods, the proposed algorithm finds joint spatio-temporal feature vectors that match the dominant subspace observed in the two-dimensional data without needs for a forward propagation model and AP templates.
The proposed MSD approach provides more discriminatory features for unsupervised AP sorting applications.
Copyright © 2015. Published by Elsevier B.V.
Neuronal noise sources and systematic variability in the shape of a spike limit the ability to sort multiple unit waveforms recorded from nervous tissue into their single neuron constituents. Here we present a procedure to efficiently sort spikes in the presence of noise that is anisotropic, i.e., dominated by particular frequencies, and whose amplitude distribution may be non-Gaussian, such as occurs when spike waveforms are a function of interspike interval. Our algorithm uses a hierarchical clustering scheme. First, multiple unit records are sorted into an overly large number of clusters by recursive bisection. Second, these clusters are progressively aggregated into a minimal set of putative single units based on both similarities of spike shape as well as the statistics of spike arrival times, such as imposed by the refractory period. We apply the algorithm to waveforms recorded with chronically implanted micro-wire stereotrodes from neocortex of behaving rat. Natural extensions of the algorithm may be used to cluster spike waveforms from records with many input channels, such as those obtained with tetrodes and multiple site optical techniques.
In an overcomplete basis, the number of basis vectors is greater than the dimensionality of the input, and the representation of an input is not a unique combination of basis vectors. Overcomplete representations have been advocated because they have greater robustness in the presence of noise, can be sparser, and can have greater flexibility in matching structure in the data. Overcomplete codes have also been proposed as a model of some of the response properties of neurons in primary visual cortex. Previous work has focused on finding the best representation of a signal using a fixed overcomplete basis (or dictionary). We present an algorithm for learning an overcomplete basis by viewing it as probabilistic model of the observed data. We show that overcomplete bases can yield a better approximation of the underlying statistical distribution of the data and can thus lead to greater coding efficiency. This can be viewed as a generalization of the technique of independent component analysis and provides a method for Bayesian reconstruction of signals in the presence of noise and for blind source separation when there are more sources than mixtures.
Signal processing and classification algorithms often have limited applicability resulting from an inaccurate model of the signal's underlying structure. We present here an efficient, Bayesian algorithm for modeling a signal composed of the superposition of brief,
Poisson-distributed functions. This methodology is applied to the specific problem of modeling and classifying extracellular neural waveforms which are composed of a superposition of an unknown number of action potentials CAPs). Previous approaches have had
limited success due largely to the problems of determining the spike shapes, deciding how many are shapes distinct, and decomposing overlapping APs. A Bayesian solution to each of these problems is obtained by inferring a probabilistic model of the waveform. This
approach quantifies the uncertainty of the form and number of the inferred AP shapes and is used to obtain an efficient method for decomposing complex overlaps. This algorithm can extract many times more information than previous methods and facilitates the
extracellular investigation of neuronal classes and of interactions within neuronal circuits.
In the present study, nearly two-thirds of all hippocampal pyramidal units isolated under barbiturate anesthesia, which maximizes these cell's activity, were behaviorally silent. These "silent cells" showed no spontaneous firing activity in the awake, freely-behaving rat. Both reanesthetization and antidromic stimulation, however, activated these silent cells. More than 92% of the remaining spontaneously active hippocampal pyramidal cells recorded from freely-behaving rats were place cells; i.e., they exhibited spatially specific changes in firing activity in at least one environment. The firing rates of these place cells varied depending on the animal's location in this environment. Interestingly many of these place cells displayed low or no spontaneous activity and no spatial specificity in other, dissimilar environments; i.e., their lack of firing in some spatial environments mirrored the behavioral silence of the more numerous silent cells reported here. In complex information processing, such as the processing of spatial information by the hippocampus demonstrated here, neural silence may be as important a signal as neural activity.
Neuronal interactions were studied in the cat's cortex (area 17) by analyzing the cross-correlation of impulse discharges in two simultaneously recorded cells. In order to obtain a reliable correlogram in a short period of time, impulse discharges in cortical cells were enhanced by electrophoretic ejection of glutamate ions or by stimulating their receptive fields with a stationary or moving light bar. The cross-correlation study of 208 neuronal pairs sampled from 17 cats revealed three types of interneuronal interactions. The first type of interaction was a conjoint excitation of two cortical cells. A positive correlation occurred with practically no delay: the maximum positivity was located ± 0.3 ms around zero time and the positivity extended only for ±0.6 ms. The second type of interaction was a delayed excitation of one cell following the excitation of the other. The third type of interaction was a delayed inhibition of one cell following the excitation of the other. A negative correlation started with a monosynaptic delay and declined relatively slowly (total duration more than 80 ms). Conjoint excitation occurred in about half of the neuronal pairs studied (105/208) and delayed excitation and inhibition in about 1/10 of them (24 and 26/208). Conjoint excitation of cortical cells may be ascribed to common excitatory inputs from the lateral geniculate cells (common excitation). Delayed excitation and inhibition may represent, respectively, excitatory and inhibitory interactions through intracortical connections (intracortical excitation and intracortical inhibition).
1. Activity of up to 10 single units was recorded in parallel from frontal areas of behaving monkeys. 2. Spatiotemporal firing patterns were revealed by a method that detects all excessively repeating patterns regardless of their complexity or single-unit composition. 3. Excess of repeating patterns was found in 30-60% of the cases examined when timing jitter of 1-3 ms was allowed. 4. An independent test refuted the hypothesis that these patterns represented chance events. 5. In a given behavioral condition there were usually many different patterns, each repeating several times, and not one (or a few) pattern repeating many times. 6. In 13 out of 20 cases, when a single unit elevated its firing rate in association with an external event beyond 40/s, most of the spikes within that period were associated with excessively repeating spatiotemporal patterns. 7. Of 157 types of patterns whose excess was most marked, 107 were composed of spikes from one single unit, 45 of the patterns contained spikes from two single units, and only one was composed of spikes from three different single units. 8. These properties suggest that the patterns were generated by reverberations in a synfire mode within self-exciting cell assemblies.
A paradox that exists in auditory and electrosensory neural systems is that they encode behaviorally relevant signals in the range of a few microseconds with neurons that are at least one order of magnitude slower. The importance of temporal coding in neural information processing is not clear yet. A central question is whether neuronal firing can be more precise than the time constants of the neuronal processes involved. Here we address this problem using the auditory system of the barn owl as an example. We present a modelling study based on computer simulations of a neuron in the laminar nucleus. Three observations explain the paradox. First, spiking of an 'integrate-and-fire' neuron driven by excitatory postsynaptic potentials with a width at half-maximum height of 250 micros, has an accuracy of 25 micros if the presynaptic signals arrive coherently. Second, the necessary degree of coherence in the signal arrival times can be attained during ontogenetic development by virtue of an unsupervised hebbian learning rule. Learning selects connections with matching delays from a broad distribution of axons with random delays. Third, the learning rule also selects the correct delays from two independent groups of inputs, for example, from the left and right ear.
The objective of this study was to determine whether each of several different memory processes is encoded exclusively by specific single neurons (single-neuron coding) or by overlapped groups of neurons (population coding by cell assembly). Single neuronal activity was recorded from the rat hippocampal formation (CA1, CA3, dentate gyrus) and temporal cortex during the performance of simple auditory, simple visual, and configural auditory-visual discrimination tasks. All the tasks employed the identical apparatus and time parameters and differed only in the type of stimuli to be processed for correct performance. Single neurons showing significantly differential activity among the discriminative stimuli in each task were judged to be task-related and involved in the memory process of the task. Of the total number of neurons recorded from the hippocampal formation and temporal cortex, 21-26% of the neurons showed task-related activity in only one task, in two tasks, or in all three tasks. This result indicates some overlapping among the neurons involved in each of teh different memory processes. A cross-correlation analysis tested activity correlations among the neurons recorded simultaneously. Most pairs of the hippocampal neurons related to the same tasks (same memory processes) showed correlations during performance of the related tasks. This result showing coactivation of the same types of task-related neurons, together with the result showing the overlapping of task-related neurons, supports the concept of population coding by cell assemblies specifically in the hippocampal formation during memory processing.
The invasion of sodium spikes from the soma into dendrites was studied in hippocampal pyramidal cells by simultaneous extracellular and intracellular recordings in anesthetized rats and by simultaneous extracellular recordings of the somatic and dendritic potentials in freely behaving animals. During complex-spike patterns, recorded in the immobile or sleeping animal, dendritic invasion of successive spikes was substantially attenuated. Complex-spike bursts occurred in association with population discharge of CA3-CA1 pyramidal cells (sharp wave field events). Synaptic inhibition reduced the amplitude of sodium spikes in the dendrites and prevented the occurrence of calcium spikes. These findings indicate that (i) the voltage-dependent calcium influx into the dendrites is under the control of inhibitory neurons and (ii) the temporal coincidence of synaptic depolarization and activation of voltage-dependent calcium conductances by the backpropagating spikes during sharp wave bursts may be critical for synaptic plasticity in the intact hippocampus.
It is now commonly accepted that planning and execution of movements are based on distributed processing by neuronal populations
in motor cortical areas. It is less clear, though, how these populations organize dynamically to cope with the momentary computational
demands. Simultaneously recorded activities of neurons in the primary motor cortex of monkeys during performance of a delayed-pointing
task exhibited context-dependent, rapid changes in the patterns of coincident action potentials. Accurate spike synchronization
occurred in relation to external events (stimuli, movements) and was commonly accompanied by discharge rate modulations but
without precise time locking of the spikes to these external events. Spike synchronization also occurred in relation to purely
internal events (stimulus expectancy), where firing rate modulations were distinctly absent. These findings indicate that
internally generated synchronization of individual spike discharges may subserve the cortical organization of cognitive motor
processes.
In vitro experiments suggest that dendritic fast action potentials may influence the efficacy of concurrently active synapses by enhancing Ca2+ influx into the dendrites. However, the exact circumstances leading to these effects in the intact brain are not known. We have addressed these issues by performing intracellular sharp electrode recordings from morphologically identified sites in the apical dendrites of CA1 pyramidal neurons in vivo while simultaneously monitoring extracellular population activity. The amplitude of spontaneous fast action potentials in dendrites decreased as a function of distance from the soma, suggesting that dendritic propagation of fast action potentials is strongly attenuated in vivo. Whereas the amplitude variability of somatic action potentials was very small, the amplitude of fast spikes varied substantially in distal dendrites. Large-amplitude fast spikes in dendrites occurred during population discharges of CA3-CA1 neurons concurrent with field sharp waves. The large-amplitude fast spikes were associated with bursts of smaller-amplitude action potentials and putative Ca2+ spikes. Both current pulse-evoked and spontaneously occurring Ca2+ spikes were always preceded by large-amplitude fast spikes. More spikes were observed in the dendrites during sharp waves than in the soma, suggesting that local dendritic spikes may be generated during this behaviorally relevant population pattern. Because not all dendritic spikes produce somatic action potentials, they may be functionally distinct from action potentials that signal via the axon.
Spike transmission probability between pyramidal cells and interneurons in the CA1 pyramidal layer was investigated in the behaving rat by the simultaneous recording of neuronal ensembles. Population synchrony was strongest during sharp wave (SPW) bursts. However, the increase was three times larger for pyramidal cells than for interneurons. The contribution of single pyramidal cells to the discharge of interneurons was often large (up to 0.6 probability), as assessed by the presence of significant (<3 ms) peaks in the cross-correlogram. Complex-spike bursts were more effective than single spikes. Single cell contribution was higher between SPW bursts than during SPWs or theta activity. Hence, single pyramidal cells can reliably discharge interneurons, and the probability of spike transmission is behavior dependent.
In cultures of dissociated rat hippocampal neurons, persistent potentiation and depression of glutamatergic synapses were induced by correlated spiking of presynaptic and postsynaptic neurons. The relative timing between the presynaptic and postsynaptic spiking determined the direction and the extent of synaptic changes. Repetitive postsynaptic spiking within a time window of 20 msec after presynaptic activation resulted in long-term potentiation (LTP), whereas postsynaptic spiking within a window of 20 msec before the repetitive presynaptic activation led to long-term depression (LTD). Significant LTP occurred only at synapses with relatively low initial strength, whereas the extent of LTD did not show obvious dependence on the initial synaptic strength. Both LTP and LTD depended on the activation of NMDA receptors and were absent in cases in which the postsynaptic neurons were GABAergic in nature. Blockade of L-type calcium channels with nimodipine abolished the induction of LTD and reduced the extent of LTP. These results underscore the importance of precise spike timing, synaptic strength, and postsynaptic cell type in the activity-induced modification of central synapses and suggest that Hebb's rule may need to incorporate a quantitative consideration of spike timing that reflects the narrow and asymmetric window for the induction of synaptic modification.
The detection of neural spike activity is a technical challenge that is a prerequisite for studying many types of brain function. Measuring the activity of individual neurons accurately can be difficult due to large amounts of background noise and the difficulty in distinguishing the action potentials of one neuron from those of others in the local area. This article reviews algorithms and methods for detecting and classifying action potentials, a problem commonly referred to as spike sorting. The article first discusses the challenges of measuring neural activity and the basic issues of signal detection and classification. It reviews and illustrates algorithms and techniques that have been applied to many of the problems in spike sorting and discusses the advantages and limitations of each and the applicability of these methods for different types of experimental demands. The article is written both for the physiologist wanting to use simple methods that will improve experimental yield and minimize the selection biases of traditional techniques and for those who want to apply or extend more sophisticated algorithms to meet new experimental challenges.
It is not clear how information related to cognitive or psychological processes is carried by or represented in the responses of single neurons. One provocative proposal is that precisely timed spike patterns play a role in carrying such information. This would require that these spike patterns have the potential for carrying information that would not be available from other measures such as spike count or latency. We examined exactly timed (1-ms precision) triplets and quadruplets of spikes in the stimulus-elicited responses of lateral geniculate nucleus (LGN) and primary visual cortex (V1) neurons of the awake fixating rhesus monkey. Large numbers of these precisely timed spike patterns were found. Information theoretical analysis showed that the precisely timed spike patterns carried only information already available from spike count, suggesting that the number of precisely timed spike patterns was related to firing rate. We therefore examined statistical models relating precisely timed spike patterns to response strength. Previous statistical models use observed properties of neuronal responses such as the peristimulus time histogram, interspike interval, and/or spike count distributions to constrain the parameters of the model. We examined a new stochastic model, which unlike previous models included all three of these constraints and unlike previous models predicted the numbers and types of observed precisely timed spike patterns. This shows that the precise temporal structures of stimulus-elicited responses in LGN and V1 can occur by chance. We show that any deviation of the spike count distribution, no matter how small, from a Poisson distribution necessarily changes the number of precisely timed spike patterns expected in neural responses. Overall the results indicate that the fine temporal structure of responses can only be interpreted once all the coarse temporal statistics of neural responses have been taken into account.
Independent component analysis or blind source separation is a new technique of extracting independent signals from mixtures. It is applicable even when the number of independent sources is unknown and is larger or smaller than the number of observed mixture signals. This article extends the natural gradient learning algorithm to be applicable to these overcomplete and undercomplete cases. Here, the observed signals are assumed to be whitened by preprocessing, so that we use the natural Riemannian gradient in Stiefel manifolds.
Multichannel tetrode array recording in awake behaving animals provides a powerful method to record the activity of large numbers of neurons. The power of this method could be extended if further information concerning the intracellular state of the neurons could be extracted from the extracellularly recorded signals. Toward this end, we have simultaneously recorded intracellular and extracellular signals from hippocampal CA1 pyramidal cells and interneurons in the anesthetized rat. We found that several intracellular parameters can be deduced from extracellular spike waveforms. The width of the intracellular action potential is defined precisely by distinct points on the extracellular spike. Amplitude changes of the intracellular action potential are reflected by changes in the amplitude of the initial negative phase of the extracellular spike, and these amplitude changes are dependent on the state of the network. In addition, intracellular recordings from dendrites with simultaneous extracellular recordings from the soma indicate that, on average, action potentials are initiated in the perisomatic region and propagate to the dendrites at 1.68 m/s. Finally we determined that a tetrode in hippocampal area CA1 theoretically should be able to record electrical signals from approximately 1, 000 neurons. Of these, 60-100 neurons should generate spikes of sufficient amplitude to be detectable from the noise and to allow for their separation using current spatial clustering methods. This theoretical maximum is in contrast to the approximately six units that are usually detected per tetrode. From this, we conclude that a large percentage of hippocampal CA1 pyramidal cells are silent in any given behavioral condition.
Precise spatiotemporal patterns in neural discharge are a possible mechanism for information encoding in the brain. Previous studies have found that such patterns repeat and appear to relate to key behavioral events. Whether these patterns occur above chance levels remains controversial. To address this question, we have made simultaneous recordings from between two and nine neurons in the primary motor cortex and supplementary motor area of three monkeys while they performed a precision grip task. Out of a total of 67 neurons, 46 were antidromically identified as pyramidal tract neurons. Sections of recordings 60 s long were searched for patterns involving three or more spikes that repeated at least twice. The allowed jitter for pattern repetition was 3 ms, and the pattern length was limited to 192 ms. In all 11 recordings analyzed, large numbers of repeating patterns were found. To assess the expected chance level of patterns, "surrogate" datasets were generated. These had the same moment-by-moment modulation in firing rate as the experimental spike trains, and matched their interspike interval distribution, but did not preserve the precise timing of individual spikes. The number of repeating patterns in 10 randomly generated surrogates was used to form 99% confidence limits on the repeating pattern count expected by chance. There was close agreement between these confidence limits and the number of patterns seen in the experimental data. Analysis of high complexity patterns was carried out in four long recordings (mean duration 23.2 min, mean number of neurons simultaneously recorded 7.5). This analysis logged only patterns composed of a larger number (7-11) of spikes. The number of patterns seen in the surrogate datasets showed a small but significant excess over those seen in the original experimental data; this is discussed in the context of surrogate generation. The occurrence of repeating patterns in the experimental data were strongly associated with particular phases of the precision grip task; however, a similar task dependence was seen for the surrogate data. When a repeating pattern was used as a template to find inexact matches, in which up to half of the component spikes could be missing, similar numbers of matches were found in experimental and surrogate data, and the time of occurrence of such matches showed the same task dependence. We conclude that the existence of precise repeating patterns in our data are not due to cortical mechanisms that favor this form of coding, since as many, if not more, patterns are produced by spike trains constructed only to modulate their firing rate in the same way as the experimental data, and to match the interspike interval histograms. The task dependence of pattern occurrence is explicable as an artifact of the modulation of neural firing rate. The consequences for theories of temporal coding in the cortex are discussed.
To elucidate cortical mechanisms involved in higher cortical functions such as working memory, we have examined feedforward excitation transmitted by identified pyramidal cells to interneurons with predominantly horizontal axonal arbors, using dual somatic recordings in prefrontal cortical slices. Interneurons with local (narrow) axonal arbors, especially chandelier interneurons, exhibited extremely narrow action potentials and high evoked firing rates, whereas neurons identified with wide arbor axons generated wider spikes and lower evoked firing rates with considerable spike adaptation, resembling that of pyramidal cells. Full reconstruction of differentially labeled neuronal pairs revealed that local arbor cells generally received a single but functionally reliable putative synaptic input from the identified pyramidal neuron member of the pair. In contrast, more synapses (two to five) were necessary to depolarize medium and wide arbor neurons reliably. The number of putative synapses and the amplitude of the postsynaptic response were remarkably highly correlated within each class of local, medium, and wide arbor interneurons (r = 0.88, 0.95, and 0.99, respectively). Similarly strong correlations within these subgroups were also present between the number of putative synapses and variance in the EPSP amplitudes, supporting the validity of our morphological analysis. We conclude that interneurons varying in the span of their axonal arbors and hence in the potential regulation of different numbers of cortical modules differ also in their excitatory synaptic input and physiological properties. These findings provide insight into the circuit basis of lateral inhibition and functional interactions within and between cortical columns in the cerebral cortex.
This is the first report that introduces appropriate behavioral tasks for monkeys for investigations of working memory for temporal and nontemporal events. Using several behavioral tests, the study also shows how temporal information is coded during retention intervals in the tasks. Each of three monkeys was trained with two working memory tasks: delayed matching-to-sample of stimulus duration (DMS-D) and delayed matching-to-sample of stimulus color (DMS-C). The two tasks employed an identical apparatus and responses and differed only in the temporal and nontemporal attribute of the stimuli to be retained for correct performance. When a retention interval between the sample and comparison stimuli was prolonged, the monkeys made more incorrect responses to short samples in the DMS-C task, suggesting "trace decay" of memory for short stimuli. However, the same monkeys showed no such increase in incorrect responses to short samples in the DMS-D task, suggesting active coding of temporal information, that is, the length of stimulus duration, during the retention interval. When variable lengths of samples were presented with a fixed retention interval, the monkeys made more incorrect responses when length differences between short and long samples were small in the DMS-D task, but not in the DMS-C task. This suggests that the codes of working memory retained in the DMS-D task were not absolute (analogical) but rather were relative (categorical) and related to differences in the duration of the samples.
The detection of neural spike activity is a technical challenge that is a prerequisite for studying many types of brain function. Measuring the activity of individual neurons accurately can be difficult due to large amounts of background noise and the difficulty in distinguishing the action potentials of one neuron from those of others in the local area. This article reviews algorithms and methods for detecting and classifying action potentials, a problem commonly referred to as spike sorting. The article first discusses the challenges of measuring neural activity and the basic issues of signal detection and classification. It reviews and illustrates algorithms and techniques that have been applied to many of the problems in spike sorting and discusses the advantages and limitations of each and the applicability of these methods for different types of experimental demands. The article is written both for the physiologist wanting to use simple methods that will improve experimental yield and minimize the selection biases of traditional techniques and for those who want to apply or extend more sophisticated algorithms to meet new experimental challenges.
The present review discusses why cell-assembly coding, i.e. ensemble coding by functionally connected neurons, is a tenable view of the brain's neuronal code and how it operates in the working brain. The cell-assembly coding has two major properties, i.e., partial overlapping of neurons among assemblies and connection dynamics within and among the assemblies. The former is the ability of one neuron to participate in different types of information processing. The latter is the capability for functional synaptic connections, detected by activity correlations of the neurons, to change among different types of information processing. An example of a series of experiments which detected these two major properties is then given. Several relevant points concerning the detection of the actual dynamics of cell-assembly coding are also enumerated. They include the dependence of the type of cell-assembly coding on types of information-processing in different structures of the brain, sparse coding by distributed overlapped assemblies, and coincidence detection as a role of individual neurons to bind distributed neurons into cell assemblies.
Classifying spike shapes in multi-unit recordings has been important to extract single neuronal activities from nervous tissue. Although several methods for this purpose have been developed, most of them have had limitations in their ability to decompose single unit activities. When more than two neurons generate action potentials simultaneously, it is difficult to identify the spikes because of the overlap of the spike waveforms. In this paper, we suggest a procedure that solves this problem using independent component analysis. By testing for the refractory period of spikes in each independent component, the proposed procedure is efficient for the decomposition of neuronal activities.
Different types of memory are thought to be processed by different neuronal networks. The present study evaluated the functional synaptic connections among neurons by means of cross-correlation analysis when two types of memory, working and reference memory, were being processed in behaving rats. Some of the synaptic connections of neurons in the hippocampal CA1, CA3, dentate gyrus and auditory cortex functioned when the rat was performing only one of the memory tasks. Working memory needed more of these memory-type dependent synapses in the hippocampal regions than reference memory.
The independent component analysis (ICA) of a random vector consists of searching for a linear transformation that minimizes the statistical dependence between its components. In order to define suitable search criteria, the expansion of mutual information is utilized as a function of cumulants of increasing orders. An efficient algorithm is proposed, which allows the computation of the ICA of a data matrix within a polynomial time. The concept of ICA may actually be seen as an extension of the principal component analysis (PCA), which can only impose independence up to the second order and, consequently, defines directions that are orthogonal. Potential applications of ICA include data analysis and compression, Bayesian detection, localization of sources, and blind identification and deconvolution.
Objective evaluation of spike sorting algorithms such as those used to decompose tetrode recordings into distinct spike trains requires a priori knowledge of the correct classification for a given recording. Intracellular recording can unambiguously assign spikes to a single neuron, and thus provide correct classification if signals from that neuron concurrently appear in a tetrode recording. Simultaneous single or paired intracellular and tetrode recordings are used here to evaluate a contemporary spike sorting algorithm for isolated as well as overlapped events. These data are also used to demonstrate that overlapping extracellular spikes combine additively, and to introduce a means for quantifying variability in action potential shape.
Independent component analysis (ICA) is a statistical method for transforming an observed multidimensional random vector into components that are statistically as independent from each other as possible. In this paper, we use a combination of two different approaches for linear ICA: Comon's information-theoretic approach and the projection pursuit approach. Using maximum entropy approximations of differential entropy, we introduce a family of new contrast (objective) functions for ICA. These contrast functions enable both the estimation of the whole decomposition by minimizing mutual information, and estimation of individual independent components as projection pursuit directions. The statistical properties of the estimators based on such contrast functions are analyzed under the assumption of the linear mixture model, and it is shown how to choose contrast functions that are robust and/or of minimum variance. Finally, we introduce simple fixed-point algorithms for practical optimization of the contrast functions. These algorithms optimize the contrast functions very fast and reliably.
The statistical analysis of two simultaneously observed trains of neuronal spikes is described, using as a conceptual framework the theory of stochastic point processes.The first statistical question that arises is whether the observed trains are independent; statistical techniques for testing independence are developed around the notion that, under the null hypothesis, the times of spike occurrence in one train represent random instants in time with respect to the other. If the null hypothesis is rejected-if dependence is attributed to the trains-the problem then becomes that of characterizing the nature and source of the observed dependencies. Statistical signs of various classes of dependencies, including direct interaction and shared input, are discussed and illustrated through computer simulations of interacting neurons. The effects of nonstationarities on the statistical measures for simultaneous spike trains are also discussed. For two-train comparisons of irregularly discharging nerve cells, moderate nonstationarities are shown to have little effect on the detection of interactions.Combining repetitive stimulation and simultaneous recording of spike trains from two (or more) neurons yields additional clues as to possible modes of interaction among the monitored neurons; the theory presented is illustrated by an application to experimentally obtained data from auditory neurons.A companion paper covers the analysis of single spike trains.
In this article the relationships among firing rate, probability of firing and counts per bin are examined. It is suggested that PSTHs, autocorrelations and crosscorrelations of neuronal activity should all be expressed in units of firing rates (spikes/s), since the values obtained by such scaling are independent of bin size and of total time of measurement. A simple method for these histograms is described. Methods to compute confidence limits for PSTHs, autocorrelations and crosscorrelations are suggested. The computations are based on the null hypothesis that the spike train(s) is (are) the realization of (independent) Poisson-point process(es). The validity and the limitations of these computations methods, when applied to spike trains, are discussed. Methods to smooth out random fluctuation with little distortion of the histogram's shape are described. It is suggested that one can minimize the distortion of the histogram in the time-domain and in the frequency-domain by using a bell-shaped bin whose center point slides continuously along the histogram. The article aims at giving the potential user of the methods some insight for the meaning of the formulae. It describes in detail how the methods are applied in practice and illustrates each method by using real data from single-unit recordings.
Ensemble recordings of 73 to 148 rat hippocampal neurons were used to predict accurately the animals' movement through their environment, which confirms that the hippocampus transmits an ensemble code for location. In a novel space, the ensemble code was initially less robust but improved rapidly with exploration. During this period, the activity of many inhibitory cells was suppressed, which suggests that new spatial information creates conditions in the hippocampal circuitry that are conducive to the synaptic modification presumed to be involved in learning. Development of a new population code for a novel environment did not substantially alter the code for a familiar one, which suggests that the interference between the two spatial representations was very small. The parallel recording methods outlined here make possible the study of the dynamics of neuronal interactions during unique behavioral events.
The majority of techniques for separating multiple single-unit spike trains from a multi-unit recording rely on the assumption that different cells exhibit action potentials having unique amplitudes and waveforms. When this assumption fails, due to the similarity of spike shape among different cells or to the presence of complex spikes with declining intra-burst amplitude, these methods lead to errors in classification. In an effort to avoid these errors, the stereotrode (McNaughton et al., 1983) and later the tetrode (O'Keefe and Reece, 1993; Wilson and McNaughton, 1993) recording techniques were developed. Because the latter technique has been applied primarily to the hippocampus, we sought to evaluate its performance in the neocortex. Multi-unit recordings, using single tetrodes, were made at 28 sites in area 17 of 3 anesthetized cats. Neurons were activated with moving bars and square wave gratings. Single units were separated by identification of clusters in 2-D projections of either peak-to-peak amplitude, spike width, spike area, or the 1st versus 2nd principal components of the waveforms recorded on each channel. Using tetrodes, we recorded a total of 154 single cells (mean = 5.4, max = 9). By cross-checking the performance of the tetrode with the stereotrode and electrode, we found that the best of the 6 possible stereotrode pairs and the best of 4 possible electrodes from each tetrode yielded 102 (mean = 3.6, max = 7) and 95 (mean = 3.4, max = 6) cells, respectively. Moreover, we found that the number of cells isolated at each site by the tetrode was greater than the stereotrode or electrode in 16/28 and 28/28 cases, respectively. Thus, both stereotrodes, and particularly electrodes, often lumped 2 or more cells in a single cluster that could be easily separated by the tetrode. We conclude that tetrode recording currently provides the best and most reliable method for the isolation of multiple single units in the neocortex using a single probe.
1. Here we study the variability in extracellular records of action potentials. Our work is motivated, in part, by the need to construct effective algorithms to classify single-unit waveforms from multiunit recordings. 2. We used microwire electrode pairs (stereotrodes) to record from primary somatosensory cortex of awake, behaving rat. Our data consist of continuous records of extracellular activity and segmented records of extracellular spikes. Spectral and principal component techniques are used to analyze mean single-unit wave-forms, the variability between different instances of a single-unit waveform, and the underlying background activity. 3. The spectrum of the variability between different instances of a single-unit waveforms is not white, and falls off above 1 kHz with a frequency dependence of roughly f-2. This spectrum is different from that of the mean spike waveforms, which falls off roughly as f-4, but is essentially identical with the spectrum of background activity. The spatial coherence of the variability on the 10-micron scale also falls off at high frequencies. 4. The variability between different instances of a single-unit waveform is dominated by a relatively small number of principal components. As a consequence, there is a large anisotropy in the cluster of the spike waveforms. 5. The background noise cannot be represented as a stationary Gaussian random process. In particular, we observed that the spectrum changes significantly between successive 20-ms intervals. Furthermore, the total power in the background activity exhibits larger fluctuations than is consistent with a stationary Gaussian random process. 6. Roughly half of the single-unit spike waveforms exhibit systematic changes as a function of the interspike interval. Although this results in a non-Gaussian distribution in the space of waveforms, the distribution can be modeled by a scalar function of the interspike interval. 7. We use a set of 44 mean single-unit waveforms to define the space of differences between spike waveforms. This characterization, together with that of the background activity, is used to construct a filter that optimizes the detection of differences between single-unit waveforms. Further, an information theoretic measure is defined that characterizes the detectability.
Several lines of evidence indicate that brief (< 25 ms) bursts of high-frequency firing have special importance in brain function. Recent work shows that many central synapses are surprisingly unreliable at signaling the arrival of single presynaptic action potentials to the postsynaptic neuron. However, bursts are reliably signaled because transmitter release is facilitated. Thus, these synapses can be viewed as filters that transmit bursts, but filter out single spikes. Bursts appear to have a special role in synaptic plasticity and information processing. In the hippocampus, a single burst can produce long-term synaptic modifications. In brain structures whose computational role is known, action potentials that arrive in bursts provide more-precise information than action potentials that arrive singly. These results, and the requirement for multiple inputs to fire a cell suggest that the best stimulus for exciting a cell (that is, a neural code) is coincident bursts.
Most neurons in the mammalian CNS encode and transmit information via action potentials. Knowledge of where these electrical events are initiated and how they propagate within neurons is therefore fundamental to an understanding of neuronal function. While work from the 1950s suggested that action potentials are initiated in the axon, many subsequent investigations have suggested that action potentials can also be initiated in the dendrites. Recently, experiments using simultaneous patch-pipette recordings from different locations on the same neuron have been used to address this issue directly. These studies show that the site of action potential initiation is in the axon, even when synaptic activation is powerful enough to elicit dendritic electrogenesis. Furthermore, these and other studies also show that following initiation, action potentials actively backpropagate into the dendrites of many neuronal types, providing a retrograde signal of neuronal output to the dendritic tree.
Several early studies suggested that spikes can be generated in the dendrites of CA1 pyramidal neurons, but their functional significance and the conditions under which they occur remain poorly understood. Here, we provide direct evidence from simultaneous dendritic and somatic patch-pipette recordings that excitatory synaptic inputs can elicit dendritic sodium spikes prior to axonal action potential initiation in hippocampal CA1 pyramidal neurons. Both the probability and amplitude of dendritic spikes depended on the previous synaptic and firing history of the cell. Moreover, some dendritic spikes occurred in the absence of somatic action potentials, indicating that their propagation to the soma and axon is unreliable. We show that dendritic spikes contribute a variable depolarization that summates with the synaptic potential and can act as a trigger for action potential initiation in the axon.
Peaks in spike train correlograms are usually taken as indicative of spike timing synchronization between neurons. Strictly speaking, however, a peak merely indicates that the two spike trains were not independent. Two biologically plausible ways of departing from independence that are capable of generating peaks very similar to spike timing peaks are described here: covariations over trials in response latency and covariations over trials in neuronal excitability. Since peaks due to these interactions can be similar to spike timing peaks, interpreting a correlogram may be a problem with ambiguous solutions. What peak shapes do latency or excitability interactions generate? When are they similar to spike timing peaks? When can they be ruled out from having caused an observed correlogram peak? These are the questions addressed here. The previous article in this issue proposes quantitative methods to tell cases apart when latency or excitability covariations cannot be ruled out.
Covariations in neuronal latency or excitability can lead to peaks in spike train covariograms that may be very similar to those caused by spike timing synchronization (see companion article). Two quantitative methods are described here. The first is a method to estimate the excitability component of a covariogram, based on trial-by-trial estimates of excitability. Once estimated, this component may be subtracted from the covariogram, leaving only other types of contributions. The other is a method to determine whether the covariogram could potentially have been caused by latency covariations.
1. Extracellular and intracellular records were made from guinea-pig hippocampal slices to examine the contributions of intrinsic cellular properties and synaptic events to the generation of neuronal activity. Extracellular signals were filtered to pass action potentials, which could be detected within a distance of about 80 microm from a discharging cell. 2. Spontaneous action potentials were invariably detected in records from the stratum pyramidale of CA3 region. Blocking excitatory synaptic transmission with NBQX and APV reduced their frequency by 23 +/- 35 %. Suppressing synaptic inhibition, while excitation was already blocked, increased the rate of spike discharge to 177 +/- 71 % of its control value. 3. Most action potentials recorded intracellularly from CA3 pyramidal cells were initiated in the absence of a detectable synaptic event. In contrast, most action potentials generated by inhibitory cells located close to stratum pyramidale were preceded by an EPSP. 4. In 31 simultaneous recordings, intracellular pyramidal cell action potentials appeared consistently to initiate extracellular spikes with a mean latency of 2.2 +/- 1.0 ms. Single inhibitory cell action potentials could initiate a reduction in the frequency of extracellular spikes of duration 10-30 ms. 5. Some identified extracellular spikes (n = 9) consistently preceded intracellularly recorded IPSPs. IPSPs were initiated monosynaptically with latencies of 0.9-1.5 ms. In reciprocal interactions, single pyramidal cell action potentials could trigger the discharge of an identified unit that in turn appeared to initiate an IPSP in the same pyramidal cell. 6. These data suggest that intrinsic cellular mechanisms underly much of the spontaneous activity of pyramidal cells of the CA3 region of the hippocampus in vitro. Both synaptic inhibition and a strong excitation of inhibitory cells by pyramidal cells act to reduce population activity.
Simultaneous recording from large numbers of neurons is a prerequisite for understanding their cooperative behavior. Various recording techniques and spike separation methods are being used toward this goal. However, the error rates involved in spike separation have not yet been quantified. We studied the separation reliability of "tetrode" (4-wire electrode)-recorded spikes by monitoring simultaneously from the same cell intracellularly with a glass pipette and extracellularly with a tetrode. With manual spike sorting, we found a trade-off between Type I and Type II errors, with errors typically ranging from 0 to 30% depending on the amplitude and firing pattern of the cell, the similarity of the waveshapes of neighboring neurons, and the experience of the operator. Performance using only a single wire was markedly lower, indicating the advantages of multiple-site monitoring techniques over single-wire recordings. For tetrode recordings, error rates were increased by burst activity and during periods of cellular synchrony. The lowest possible separation error rates were estimated by a search for the best ellipsoidal cluster shape. Human operator performance was significantly below the estimated optimum. Investigation of error distributions indicated that suboptimal performance was caused by inability of the operators to mark cluster boundaries accurately in a high-dimensional feature space. We therefore hypothesized that automatic spike-sorting algorithms have the potential to significantly lower error rates. Implementation of a semi-automatic classification system confirms this suggestion, reducing errors close to the estimated optimum, in the range 0-8%.
The temporal pattern and relative timing of action potentials among neocortical neurons may carry important information. However,
how cortical circuits detect or generate coherent activity remains unclear. Using paired recordings in rat neocortical slices,
we found that the firing of fast-spiking cells can reflect the spiking pattern of single-axon pyramidal inputs. Moreover,
this property allowed groups of fast-spiking cells interconnected by electrical and γ-aminobutyric acid (GABA)–releasing (GABAergic)
synapses to detect the relative timing of their excitatory inputs. These results indicate that networks of fast-spiking cells
may play a role in the detection and promotion of synchronous activity within the neocortex.
This study reports how hippocampal individual cells and cell assemblies cooperate for neural coding of pitch and temporal information in memory processes for auditory stimuli. Each rat performed two tasks, one requiring discrimination of auditory pitch (high or low) and the other requiring discrimination of their duration (long or short). Some CA1 and CA3 complex-spike neurons showed task-related differential activity between the high and low tones in only the pitch-discrimination task. However, without exception, neurons which showed task-related differential activity between the long and short tones in the duration-discrimination task were always task-related neurons in the pitch-discrimination task. These results suggest that temporal information (long or short), in contrast to pitch information (high or low), cannot be coded independently by specific neurons. The results also indicate that the two different behavioral tasks cannot be fully differentiated by the task-related single neurons alone and suggest a model of cell-assembly coding of the tasks. Cross-correlation analysis among activities of simultaneously recorded multiple neurons supported the suggested cell-assembly model.Considering those results, this study concludes that dual coding by hippocampal single neurons and cell assemblies is working in memory processing of pitch and temporal information of auditory stimuli. The single neurons encode both auditory pitches and their temporal lengths and the cell assemblies encode types of tasks (contexts or situations) in which the pitch and the temporal information are processed.
Multispike trains are encountered often, either purposely or inadvertently, when one records from neural populations. This paper focuses on techniques for detecting and identifying the spikes in multispike trains. Relatively simple methods are briefly reviewed. Most of these require a high signal-to-noise ratio. A method based on signal detection by template matching, which works well with relativeiy small spikes, is described in detail. Use of this technique is illustrated by an investigation of the biophysical aspects of extracellular recording in sensory cortex. A further application is the analysis of multi-unit records to display relationships between two or three neurons recorded simultaneously.