Elsevier

Neurocomputing

Volume 70, Issues 10–12, June 2007, Pages 2064-2068
Neurocomputing

Validation of task-related excess of spike coincidences based on NeuroXidence

https://doi.org/10.1016/j.neucom.2006.10.142Get rights and content

Abstract

One of the key findings supporting the assembly hypothesis was found in recordings from the primary motor cortex of behaving monkeys involved in a delayed pointing task [A. Riehle, S. Grün, M. Diesmann, A. Aertsen, Spike synchronization and rate modulation differentially involved in motor cortical function, Science 278 (1997) 1950–1953]. Based on the unitary event (‘UE’) method, the authors have shown that excess coincidences between simultaneously recorded neurons occur dynamically at behaviorally relevant points in time. However, sensitivity of the UE method for non-stationarities and regularity of spike trains caused fear that the results presented might be, at least in part, false positives. We reanalyzed the same data with the new non-parametric method NeuroXidence, which is robust against firing-rate modulations, rate changes across trials, regularity or burstiness, as well as low rates. Our results based on NeuroXidence confirm the results presented in Riehle et al. [Spike synchronization and rate modulation differentially involved in motor cortical function, Science 278 (1997) 1950–1953] and demonstrate behaviorally modulated excesses of coincidences on a time scale of 3–5 ms.

Introduction

In recent years, there has been a great deal of controversial discussion about the assembly hypothesis. On the one hand, evidence for fine-temporal structure in spiking activity is accumulating. On the other hand, there are still concerns about the appropriate analysis techniques to demonstrate the existence of precise temporal coordination of spiking activities of neurons and to diminish the fear of false positives due to violations of inherent assumptions of the analysis method [2], [5], [13]. One of the key findings that supports the assembly hypothesis was found in recordings from the primary motor cortex of behaving monkeys involved in a delayed pointing task [12]. It was shown that excess joint-spike activity of simultaneously recorded neurons occurs dynamically at behaviorally relevant points in time. The analysis of the data was based on the unitary event analysis method (‘UE’ ; [3], [4]).

Section snippets

UE method

The core idea of the UE method is based on a statistical hypothesis test that compares the amount of empirically measured coincidences within simultaneously recorded neurons with the expected number. The latter is estimated based on the null-hypothesis (H0) that assumes that the recorded spike trains within the analysis window can be described by independent and stationary Bernoulli processes [3]. To compare the expected and the empirical numbers by a statistical significance test, the UE

NeuroXidence

NeuroXidence is, likewise the UE method, based on a hypothesis test that compares the expected and the empirical numbers of coincidences occurring in simultaneously recorded spike trains [11]. It detects fine-temporal cross-structure on a time scale τc (e.g. τc=5ms) that is not explained by either the auto-structure of individual spike trains or the cross-structure, which may exist on a time scale slower than τr (e.g. τr=15ms). NeuroXidence utilizes surrogate data to derive the statistical

Results

We re-analyzed the data presented in [12] with both methods discussed above (UE, [3], [4] and NeuroXidence, parameters: S=20, τc=5ms, τr=15ms). We used the same parameters as those used in the original study (sliding window of 100 ms duration, shifted in 5 ms offset along the data, test level 5%, allowed coincidence width 5 ms). To describe the statistical properties of the recorded spike trains, we computed also the Inter-Spike-Interval (‘ISI’) distributions and the corresponding coefficient of

Conclusion

Taken together, we conclude that the results shown in [12] are not due to an effect of false positives. Indeed, our results based on NeuroXidence [11] confirm excess synchronized activity on a time scale of 3–5 ms at 0.9, 1.2, and 1.5 s after PS, at moments in time at which the monkey expected the occurrence of the GO signal. NeuroXidence does not assume specific properties of neuronal data and thus, is robust against features that may lead to false positives when using other methods. However, it

Acknowledgments

We thank Wolf Singer for stimulating discussions and his support. This work was in part funded by the Hertie Foundation (GP), the Stifterverband für die Deutsche Wissenschaft (SG), and the Volkswagen Foundation (SG, GP).

Gordon Pipa was born in 1974. He holds several patents (Patents: US6231185, WO99/03066, PCT01454, EP0993657, DE9701454) the honorary medal of the Friedrich-Bläse-Foundation and a federal scholarship. He studied Physics and Electrical Engineering at the Technical University of North Rhine-Westphalia in Aachen and received his M.Sc. in Physics in 2002 from the J.W. Goethe University (Frankfurt a.M., Germany). His Masters thesis was carried out in the Department of Neurophysiology at the

References (13)

There are more references available in the full text version of this article.

Cited by (28)

  • The principle of coherence in multi-level brain information processing

    2013, Progress in Biophysics and Molecular Biology
    Citation Excerpt :

    The higher spatio-temporal precision relating neuronal oscillatory correlates to specific brain functions, and the additional minimisation of possible physical artifacts or statistical biases provided further evidence that synchronisation indeed regulates various cognitive processes and is not merely an epiphenomenon or a reflection. Some of these newer methods are: instantaneous coherence (Schack and Krause, 1995) and event-related coherence (ERCoh) (Andrew and Pfurtscheller, 1996), which allow a high temporal and frequency resolution; dynamic topographic analysis and other methods that employ Hilbert transform to visualise rapid bursts of desynchronisation and phase reset (e.g., Breakspear et al., 2004; Freeman and Rogers, 2003; Thatcher et al., 2009); methods that enable assessment of direction of information transmission, like partial directed coherence (Astolfi et al., 2006) and directed transfer function (Babiloni et al., 2005); the unitary event analysis, which employs complex statistics (Pipa et al., 2007, 2008) and can detect individual events of coincidence firing; spike-field coherence (Fries et al., 2002), which estimates consistent phase relations between the discharges of individual neurons and LFP oscillations; and pairwise phase consistency (Vinck et al., 2010b), which computes how similar the relative phase observed in one trial is to the relative phase observed in another trial, suitable for measuring rhythmic synchronisation for both EEG–EEG, MEG–MEG, spike–LFP, and spike–spike pairs. In recent years it has become increasingly clear that neuronal oscillatory coherence correlates with all basic cognitive functions.

  • Multiple firing coherence resonances in excitatory and inhibitory coupled neurons

    2012, Communications in Nonlinear Science and Numerical Simulation
    Citation Excerpt :

    Neurophysiological studies have revealed the existence of accurately timed patterns of spikes by a variety of cognitive and motoric tasks [1–6].

View all citing articles on Scopus

Gordon Pipa was born in 1974. He holds several patents (Patents: US6231185, WO99/03066, PCT01454, EP0993657, DE9701454) the honorary medal of the Friedrich-Bläse-Foundation and a federal scholarship. He studied Physics and Electrical Engineering at the Technical University of North Rhine-Westphalia in Aachen and received his M.Sc. in Physics in 2002 from the J.W. Goethe University (Frankfurt a.M., Germany). His Masters thesis was carried out in the Department of Neurophysiology at the Max-Planck-Institute for Brain Research (Frankfurt a.M., Germany). In 2006 he received his Ph.D. degree (main topic: the neuronal code) from the Technical University Berlin, Germany.

Alexa Riehle received the B.Sc. degree in Biology (main topic: deciphering microcircuitries in the frog retina) from the Free University, Berlin, Germany, in 1976, and the Ph.D. degree in Neurophysiology (main topic: neuronal mechanisms of temporal aspects of color vision in the honey bee) from the Biology Department of the Free University, Berlin, Germany, in 1980.

From 1980 to 1984, she was a Postdoctoral Fellow at the CNRS in Marseille, France (main topic: neuronal mechanisms of elementary motion detectors in the fly visual system). In 1984, she moved to the Cognitive Neuroscience Department at the CNRS, Marseille, France, and is since then mainly interested in the study of cortical information processing and neural coding in cortical ensembles during movement preparation and execution in non-human primates.

Sonja Grüun was born in 1960 in Germany, where she obtained her Diploma in physics (Eberhard-Karls University Tübingen). She did her Ph.D. work in the field of computational neuroscience at the Ruhr-University Bochum, Germany, and at the Weizmann Inst. of Science, Rehovot, Israel, and obtained her Ph.D. in physics (Ruhr-University Bochum). After her post-doctoral work at the Hebrew University in Jerusalem, Israel, she worked as a senior fellow at the Max-Planck Institute for Brain Research in Frankfurt/M, Germany. From 2002 to 2006 she was an Assistant Professor for Neuroinformatics/Theoretical Neuroscience at the Free University in Berlin, Germany and was a founding member of the Bernstein Center for Computational Neuroscience in Berlin. Since September 2006 she is the head of a research unit at the RIKEN Brain Science Institute in Wako, Japan. Her main interests are in statistical neuroscience which includes modeling of stochastic processes and the development of data analysis techniques for multiple parallel neuronal time series.

View full text