Adaptive spike-artifact removal from local field potentials uncovers prominent beta and gamma band neuronal synchronization

https://doi.org/10.1016/j.jneumeth.2019.108485Get rights and content

Abstract

Background

Many neurons synchronize their action potentials to the phase of local field potential (LFP) fluctuations in one or more frequency bands. Analyzing this spike-to-LFP synchronization is challenging, however, when neural spikes and LFP are generated in the same local circuit, because the spike’s action potential waveform leak into the LFP and distort phase synchrony estimates. Existing approaches to address this spike bleed-through artifact relied on removing the average action potential waveforms of neurons, but this leaves artifacts in the LFP and distorts synchrony estimates.

New method

We describe a spike-removal method that surpasses these limitations by decomposing individual action potentials into their frequency components before their removal from the LFP. The adaptively estimated frequency components allow for variable spread, strength and temporal variation of the spike artifact.

Results

This adaptive approach effectively removes spike bleed-through artifacts in synthetic data with known ground truth, and in single neuron and LFP recordings in nonhuman primate striatum. For a large population of neurons with both narrow and broad action potential waveforms, the use of adaptive artifact removal uncovered 20−35 Hz beta and 35−45 Hz gamma band spike-LFP synchronization that would have remained contaminated otherwise.

Comparison with existing methods

We demonstrate that adaptive spike-artifact removal cleans LFP data that remained contaminated when applying existing Bayesian and non-Bayesian methods of average spike-artifact removal.

Conclusions

Applying adaptive spike-removal from field potentials allows to estimate the phase at which neurons synchronize and the consistency of their phase-locked firing for both beta and low gamma frequencies. These metrics may prove essential to understand cell-to-circuit neuronal interactions in multiple brain systems.

Introduction

Spiking activity of many different neuron types synchronize to the local field potential (LFP) (Klausberger and Somogyi, 2008; Roux and Buzsaki, 2015). The strength of this spike-LFP phase synchronization can predict how a neuron contributes to the functioning of the circuit (Pesaran et al., 2018; Womelsdorf and Everling, 2015), how strong it gates afferent inputs through the circuit (Cardin et al., 2009; Tiesinga et al., 2008; Womelsdorf et al., 2014b), and how strong it communicates with upstream brain areas (Bastos et al., 2015; Fries, 2015; Womelsdorf et al., 2007). The importance of spike-LFP relationships is highlighted by reports that the likelihood of spiking responses can sometimes be better predicted by the phase of the local field than by spiking of neurons at nearby electrodes (Besserve et al., 2010). One reason for the informativeness of the LFP is that it is generated by currents flowing along dendritic and axonal membranes (Einevoll et al., 2013; Mitzdorf, 1985; Reimann et al., 2013; Schomburg et al., 2012). When these transmembrane currents reflect synaptic inputs or subthreshold depolarization levels they can effectively set a gain control for the spike output of neurons and thereby provide crucial insights into information processing (Azouz and Gray, 2003; Fries, 2005; Womelsdorf et al., 2014b).

However, the close relationship of spiking activity to the LFP can be artifactual when the local field potential activity is contaminated by the action potentials of the spiking neurons themselves as opposed to be based on transmembrane currents from the local circuit (Buzsaki et al., 2012; Einevoll et al., 2013; Ray, 2015). Such contamination of the LFP can occur for spiking activity of neurons in the immediate surroundings (within ∼200 μm) of a recorded neuron (Watson et al., 2018) and becomes evident as spike-bleed through artifacts in spike-triggered LFP averages or as artifactual spike-LFP synchrony, because the LFP phase of the spiking activity is contaminated by the spike itself and thus is not informative beyond the spike time itself. This spike bleed through effect is not only evident for the high frequencies at which the fast action potentials main power reside (around ∼0.8−5 kHz), but it can dominate spike-LFP measurements down to ∼100 Hz, with lower but discernible contributions for frequencies as low as 25 Hz (Ardid et al., 2015; Ray, 2015; Schomburg et al., 2012; Zanos et al., 2011).

To prevent the spike bleed through effects, a clean LFP has to be estimated. In many studies this is achieved by measuring the LFP from electrode tips that are far (>200 μm) away from the electrode recording the spiking neuron so that it does not affect the potential. This approach might be successful when the LFP is homogeneous over several hundreds of micrometers, However, there are major limitations to this approach in many settings. Firstly, the LFP might not be homogeneous between two electrodes that are >200 μm apart, but rather changes its phase systematically as quantified e.g. in traveling waves. Secondly, it is unclear how much distance between spike- and LFP electrode should be considered safe. For example, with silicon shank electrodes in somatosensory cortex there are apparent spike-bleed through artifacts evident at ∼≥100 Hz spike-LFP wavelet spectra with channels 200 μm away (e.g. see Fig. 6, Suppl. 3 and 11 in Watson et al. (2018)). This is a problem when considering that local cell-to-circuit interactions between neurons can be limited to a <150 μm diameter with spike triggered averages being essentially flat at larger distances of the neuron to the site of LFP recording (Fujisawa et al., 2008).

These considerations suggest that a versatile approach to estimate the LFP around the spiking neuron is to clean the LFP activity from influences of the spike of a neuron itself. Existing approaches for removing spike artifacts have either focused on the average waveform of a neuron to subtract its average contribution over an arbitrarily defined time window around the spike time (Pesaran et al., 2002; Zanos et al., 2012, 2011), estimated the spike through a dictionary based marching pursuit algorithm (Ray et al., 2008), or designed an on-average optimal linear filter predicting spiking influences on the LFP (David et al., 2010), or removed data and interpolated across the time around the spike occurrence to remove the spikes influence on spectral estimates of spike-LFP synchrony (Ardid et al., 2015; Galindo-Leon and Liu, 2010; Okun et al., 2010; Womelsdorf et al., 2010). However, all of these approaches have in common that they do leave residual spike-artifacts in the data in those lower 25−100 Hz frequencies that contain physiologically interesting information (Pesaran et al., 2018).

Here, we report of a novel approach that removes spike-related transients in low frequencies of the LFP by addressing several limitations of prior approaches. First, in contrast to existing methods the novel approach does not require an a priori determination of a short duration that should be taken into account to include in the artifact estimation. This is important to be able to remove slower spike accompanying events (e.g. slow after-hyperpolarization, plateau potentials or preceding EPSC barrages of activated nearby synapses) as sources of the spike-artifact. Secondly, the proposed method estimates the peak time and duration for individual spikes and is not based on a grand average spike waveform and thus allows for variable width and height (i.e. shape) of action potential waveforms. This feature is critically important to allow for changes in spike amplitude and shape that occur e.g. for burst firing neurons. By incorporating these features, we show in ground truth simulations that our approach effectively removes the action potential contributions to the LFP without introducing distortions of the phase that prior methods could not avoid. We then show in electrophysiological recordings in nonhuman primate striatum that removal of spike-transients is essential to detect narrow band spike-LFP phase synchrony in the beta (20−30 Hz) and gamma (35−45 Hz) bands. Notably, phase synchrony was reliably estimated for neurons with narrow as well as broad waveforms, suggesting that the adaptive spike-artifact removal approach allows distinguishing how different cell classes synchronize to the local LFP.

Section snippets

Electrophysiological recording

Data was collected from two male rhesus macaques (Macaca mulatta) from the head of the caudate and the ventral striatum as described in full in Oemisch et al. (2019). All animal care and experimental protocols were approved by the York University Council on Animal Care and were in accordance with the Canadian Council on Animal Care guidelines. Extra-cellular recordings were made with tungsten electrodes (impedance 1.2–2.2 MOhm, FHC, Bowdoinham, ME) through rectangular recording chambers

Adaptive Spike Removal (ASR) on simulated spike-LFP data

We first tested the Adaptive Spike Removal (ASR) method on synthetic data with known ground-truth spike waveform shape and known frequency and noise components of the LFP. In simulations we ensured that spikes injected to the simulated LFP introduced the typical spike aligned bleed through into the lower frequency band of the LFP (Fig. 3A,B). Applying ASR to the spike-triggered LFP of 100 simulated neurons with varying spike amplitude and firing rates efficiently removed the spike artifact,

Discussion

We showed how a method that adaptively estimates the frequency components of individual action potentials of a neuron is capable to effectively remove the spike artifact that would otherwise bleed-through into the lower frequency range (<100 Hz) of LFPs. The described adaptive spike removal approach effectively extracts ground-truth spike action potential waveforms injected into synthetic wideband LFP irrespective of the amplitudes of the spike artefact. We then showed that the novel approach

Acknowledgements

This research was supported by a grant from the Canadian Institutes of Health Research (T.W.) CIHR Grant MOP_102482 and by the National Institute Of Biomedical Imaging And Bioengineering of the National Institutes of Health under Award Number R01EB028161 (T.W.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the Canadian Institutes of Health Research or the National Institutes of Health. We thank Mariann Oemisch and Seyed Ali

References (64)

  • S. Ray

    Challenges in the quantification and interpretation of spike-LFP relationships

    Curr. Opin. Neurobiol.

    (2015)
  • M.W. Reimann et al.

    A biophysically detailed model of neocortical local field potentials predicts the critical role of active membrane currents

    Neuron

    (2013)
  • L. Roux et al.

    Tasks for inhibitory interneurons in intact brain circuits

    Neuropharmacology

    (2015)
  • E.W. Schomburg et al.

    Theta phase segregation of input-specific gamma patterns in entorhinal-hippocampal networks

    Neuron

    (2014)
  • M. Vinck et al.

    The pairwise phase consistency: a bias-free measure of rhythmic neuronal synchronization

    Neuroimage

    (2010)
  • T. Womelsdorf et al.

    Burst firing synchronizes prefrontal and anterior cingulate cortex during attentional control

    Curr. Biol.

    (2014)
  • T. Womelsdorf et al.

    Long-range attention networks: circuit motifs underlying endogenously controlled stimulus selection

    Trends Neurosci.

    (2015)
  • K.I. Amemori et al.

    Striatal microstimulation induces persistent and repetitive negative decision-making predicted by striatal beta-band oscillation

    Neuron

    (2018)
  • S. Ardid et al.

    Mapping of functionally characterized cell classes onto canonical circuit operations in primate prefrontal cortex

    J. Neurosci.

    (2015)
  • D.J. Bakkum et al.

    Tracking axonal action potential propagation on a high-density microelectrode array across hundreds of sites

    Nat. Commun.

    (2013)
  • J.J. Belic et al.

    Interplay between periodic stimulation and GABAergic inhibition in striatal network oscillations

    PLoS One

    (2017)
  • P. Berens

    CircStat: a MATLAB toolbox for circular statistics

    J. Stat. Softw.

    (2009)
  • J.D. Berke

    Fast oscillations in cortical-striatal networks switch frequency following rewarding events and stimulant drugs

    Eur. J. Neurosci.

    (2009)
  • M. Besserve et al.

    Causal relationships between frequency bands of extracellular signals in visual cortex revealed by an information theoretic analysis

    J. Comput. Neurosci.

    (2010)
  • G. Buzsaki

    Large-scale recording of neuronal ensembles

    Nat. Neurosci.

    (2004)
  • G. Buzsaki et al.

    The origin of extracellular fields and currents – EEG, ECoG, LFP and spikes

    Nat. Rev. Neurosci.

    (2012)
  • J.A. Cardin et al.

    Driving fast-spiking cells induces gamma rhythm and controls sensory responses

    Nature

    (2009)
  • J.E. Carmichael et al.

    Gamma oscillations in the rat ventral striatum originate in the piriform cortex

    J. Neurosci.

    (2017)
  • J. Catanese et al.

    Low- and high-gamma oscillations deviate in opposite directions from zero-phase synchrony in the limbic corticostriatal loop

    J. Neurophysiol.

    (2016)
  • S.V. David et al.

    Decoupling action potential bias from cortical local field potentials

    Comput. Intell. Neurosci.

    (2010)
  • G.T. Einevoll et al.

    Modelling and analysis of local field potentials for studying the function of cortical circuits

    Nat. Rev. Neurosci.

    (2013)
  • F.R. Fernandez et al.

    Dendritic Na+ current inactivation can increase cell excitability by delaying a somatic depolarizing afterpotential

    J. Neurophysiol.

    (2005)
  • Cited by (22)

    • Oscillations without cortex: Working memory modulates brainwaves in the endbrain of crows

      2022, Progress in Neurobiology
      Citation Excerpt :

      Data of single-cell neuronal activity for analysis of the spiking rate of the neuronal population (Fig. S10) was obtained from our previous study (Hahn et al., 2021). Prior to extracting frequency power from our signals, we removed possible spike-related traces from the LFP signals using the algorithm of Banaie Boroujeni et al. (2020). We further processed our LFP signals using the FieldTrip open-source software package for Matlab (Oostenveld et al., 2010).

    • Schrödinger filtering: a precise EEG despiking technique for EEG-fMRI gradient artifact

      2021, NeuroImage
      Citation Excerpt :

      Additional Fourier components come from any periodicity of the spikes’ occurrence in the time series, which also overlaps with the EEG signal. Thus, frequency-domain bandpass filtering is inadequate for spike removal (Boroujeni et al., 2020). Secondly, spikes are detrimental to EEG correlation analyses commonly done for explorations of functional connectivity.

    View all citing articles on Scopus
    View full text