Towards reliable spike-train recordings from thousands of neurons with multielectrodes

https://doi.org/10.1016/j.conb.2011.10.001Get rights and content

The new generation of silicon-based multielectrodes comprising hundreds or more electrode contacts offers unprecedented possibilities for simultaneous recordings of spike trains from thousands of neurons. Such data will not only be invaluable for finding out how neural networks in the brain work, but will likely be important also for neural prosthesis applications. This opportunity can only be realized if efficient, accurate and validated methods for automatic spike sorting are provided. In this review we describe some of the challenges that must be met to achieve this goal, and in particular argue for the critical need of realistic model data to be used as ground truth in the validation of spike-sorting algorithms.

Highlights

► New silicon multielectrodes may reliably record spikes from thousands of neurons. ► Validated and automated spike-sorting methods needed to fulfill potential. ► Realistic ground-truth data required both for algorithm development and validation.

Introduction

The electrical recording of spikes  the extracellular signatures of neuronal action potentials  is among the oldest [1] and most commonly used recording techniques in neuroscience. Accurate and reliable recording of the sequence of spikes from individual neurons is crucial as such spike trains appear to be the main carrier of information in mammalian neural systems. Despite its importance, the problem of reliably deciphering the recorded signal from an electrode in terms of spike contributions from the various neurons located in its vicinity is still largely unresolved. This spike-sorting problem is akin to the well-known ‘cocktail-party problem’ of tuning into one of the conversations at a busy party [2], and the development of automatic computerized spike-sorting methods has been an active research topic for several decades [3, 4, 5]. However, spike sorting is still an art as much as a science, and in most neuroscience labs spike sorting still involves a large manual component. This is not only very labor intensive, the sorting will by its nature be idiosyncratic, and the estimated spike trains will depend on who is doing the analysis [6, 7].

The new generation of silicon-based multielectrodes [8, 9] with hundreds or more electrode contacts [10, 11•, 12•] offers exciting opportunities for dramatically increasing the yield and allowing for simultaneous spike-train recordings from thousands of neurons. However, this potential can only be realized if these advances in electrode hardware are accompanied by the development of validated automatic methods for spike-train extraction and quality estimation. Such algorithms would ideally operate in real-time to allow for continuous monitoring of recordings and use in closed-loop therapeutic [13] and brain–machine interface applications [14].

As illustrated in Figure 1, spike-train estimation from raw potential traces involves many steps. In addition to the obvious prerequisite of suitable recording electrodes and measurement electronics, all these must be performed adequately for successful spike-train estimation. In this short review, some key challenges in this processing chain will be discussed. We argue that the design and use of detailed mock experiments in a computer-model setting will be crucial for the development and validation of the necessary signal-analysis methods at each processing step. In such models the ground-truth, that is, the true underlying spike trains, will be known so that the estimation error can be readily assessed. The forward modeling of realistic extracellular signatures of spikes is an essential part of this. This was a central question discussed at the recent workshop ‘Validation of Automatic Spike-Sorting Methods’ held in Ski, Norway in May 2011 (see www.g-node.org/spike) and will also be a key point addressed in this review.

Section snippets

Challenges of automatic spike sorting

Spike sorting is an umbrella term lumping together the several stages involved in extracting single-neuron spike trains from raw extracellular recordings [5]. This is illustrated in Figure 1 and described in more detail in its caption for a recording with a ‘tetrode’, a commonly used multielectrode with four closely spaced electrode contacts [15, 16]. The analysis steps of the recorded potential traces can be organized as follows: (1) preprocessing of raw data (b  c in Figure 1), (2) spike

Validation of automatic spike-sorting algorithms

Spike-sorting methods should ideally provide not only a best estimation of spike trains, but also a measure of the sorting reliability. A thorough recent review of methods that can be used in the absence of ground-truth data can be found in [30••], and some easily implemented tests are illustrated in Figure 2. A thorough evaluation of spike-sorting algorithms is difficult to do, however, without proper ground-truth test data, that is, spike-sorting data for which the true underlying spike

Outlook

Methods for spike sorting of arrays of tetrodes are well developed and, if not optimal and validated, good enough to be used routinely to record from up to ∼200 neurons simultaneously in vivo. The new generation of high-density MEAs presents the potential to record from thousands of neurons in vivo, but new algorithms must be developed to spike-sort this data. The problem of developing optimal and validated methods for extraction of spike trains from this data, is not only difficult but also

References and recommended reading

Papers of particular interest, published within the period of review, have been highlighted as:

  • • of special interest

  • •• of outstanding interest

Acknowledgements

Useful discussions with the participants at the workshop on ‘Validation of Automatic Spike-Sorting Methods’ at Ski, Norway in May 2011 and ‘Challenges in Extracellular Electrophysiology Data Analysis’ at Janelia Farm, USA in May 2010 (arranged by Gyuri Buzsaki and Dima Rimberg) are gratefully acknowledged. Work was supported by the Research Council of Norway (NevroNor, eScience, NOTUR) and the International Neuroscience Coordinating Facility (INCF).

References (50)

  • G.R. Holt et al.

    Electrical interactions via the extracellular potential near cell bodies

    J Comp Neurosci

    (1999)
  • C. Gold et al.

    On the origin of the extracellular action potential waveform: a modeling study

    J Neurophysiol

    (2006)
  • C. Gold et al.

    Using extracellular action potential recordings to constrain compartmental models

    J Comp Neurosci

    (2007)
  • J.C. Letelier et al.

    Spike sorting based on discrete wavelet transform coefficients

    J Neurosci Meth

    (2000)
  • E.D. Adrian et al.

    Impulses in the pyramidal tract

    J Physiol

    (1939)
  • S. Haykin et al.

    The cocktail party problem

    Neural Comp

    (2005)
  • M. Abeles et al.

    Multispike train analysis

    Proc IEEE

    (1977)
  • M.S. Lewicki

    A review of methods for spike sorting: the detection and classification of neural action potentials

    Network: Comp Neur Syst

    (1998)
  • R.Q. Quiroga

    Spike sorting

    Scholarpedia

    (2007)
  • F. Wood et al.

    On the variability of manual spike sorting

    IEEE Trans Biomed Engin

    (2004)
  • K.D. Harris et al.

    Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements

    J Neurophysiol

    (2000)
  • G. Buzsaki

    Large-scale recording of neuronal ensembles

    Nature Neurosci

    (2004)
  • T.J. Blanche et al.

    Polytrodes: high-density silicon electrode arrays for large-scale multiunit recording

    J Neurophysiol

    (2005)
  • R. Segev et al.

    Recording spikes from a large fraction of the ganglion cell's in a retinal patch

    Nature Neurosci

    (2004)
  • A. Lambacher et al.

    Identifying firing mammalian neurons in networks with high-resolution multi-transistor array (MTA)

    Appl Phys A

    (2011)
  • Cited by (144)

    • Recent Advances in Electrical Neural Interface Engineering: Minimal Invasiveness, Longevity, and Scalability

      2020, Neuron
      Citation Excerpt :

      The automated clustering process in these algorithms, while working well at low dimensions, often performs poorly as the dimension of the analyzed data increases, an effect termed the “curse of dimensionality” (Rossant et al., 2016). Second, previous algorithms require substantial manual processing, such as curating the results by selecting which clusters to split, merge, or reject (Einevoll et al., 2012). The manual process, while manageable with low-count probes, cannot scale up to hundreds and thousands or more channels.

    • Implanted multichannel microelectrode array for simultaneous electrophysiological signal detection of hippocampal CA1 and DG neurons of simulated microgravity rats

      2020, Biochemical and Biophysical Research Communications
      Citation Excerpt :

      Neural cells in the rat hippocampus can be divided into two classes: putative pyramidal neurons from DG and putative interneurons from CA1, distinguishable by the width of their waveform, firing rates, and the inter-spike interval [16]. In this study, cluster analysis was used to distinguish the two types of nerve cells [17,18]. In this study the Spike duration was calculated, and the symmetry index was set to (a-b)/(a+b), where the values of a and b were taken from Fig. S4(a) and subjected to cluster analysis.

    View all citing articles on Scopus
    View full text