This content has been downloaded from IOPscience. Please scroll down to see the full text.
Download details:
IP Address: 193.205.81.18
This content was downloaded on 24/11/2016 at 09:59
Please note that terms and conditions apply.
You may also be interested in:
Quantifying long-term microelectrode array functionality using chronic in vivo impedance testing
Abhishek Prasad and Justin C Sanchez
Evaluation of PEDOT films for improving neural recordings
Kip A Ludwig, Jeffrey D Uram, Junyan Yang et al.
Progress towards biocompatible intracortical microelectrodes for neural interfacing applications
Mehdi Jorfi, John L Skousen, Christoph Weder et al.
Conducting polymer coated neural recording electrodes
Alexander R Harris, Simeon J Morgan, Jun Chen et al.
Integrated device for optical stimulation and spatiotemporal electrical recording of neural
activity
Jiayi Zhang, Farah Laiwalla, Jennifer A Kim et al.
A MEMS-based flexible multichannel ECoG-electrode array
Birthe Rubehn, Conrado Bosman, Robert Oostenveld et al.
A new EC--PC threshold estimation method for in vivo neural spike detection
Zhi Yang, Wentai Liu, Mohammad Reza Keshtkaran et al.
Carbon nanotube composite coating of neural microelectrodes preferentially improves the
multiunit signal-to-noise ratio
View the table of contents for this issue, or go to the journal homepage for more
2011 J. Neural Eng. 8 066013
(http://iopscience.iop.org/1741-2552/8/6/066013)
Home Search Collections Journals About Contact us My IOPscience
IOP PUBLISHING JOURNAL OF NEURAL ENGINEERING
J. Neural Eng. 8 (2011) 066013 (12pp) doi:10.1088/1741-2560/8/6/066013
Carbon nanotube composite coating of
neural microelectrodes preferentially
improves the multiunit signal-to-noise
ratio
Gytis Baranauskas
1,3
, Emma Maggiolini
1,3
, Elisa Castagnola
1,3
,
Alberto Ansaldo
1
, Alberto Mazzoni
1
, Gian Nicola Angotzi
1
,
Alessandro Vato
1
, Davide Ricci
1
, Stefano Panzeri
1
and
Luciano Fadiga
1,2,4
1
Robotics, Brain and Cognitive Sciences Department, Italian Institute of Technology, Genoa, Italy
2
Section of Human Physiology, University of Ferrara, Ferrara, Italy
E-mail: luciano.fadiga@iit.it
Received 22 July 2011
Accepted for publication 13 October 2011
Published 8 November 2011
Online at stacks.iop.org/JNE/8/066013
Abstract
Extracellular metal microelectrodes are widely used to record single neuron activity in vivo.
However, their signal-to-noise ratio (SNR) is often far from optimal due to their high
impedance value. It has been recently reported that carbon nanotube (CNT) coatings may
decrease microelectrode impedance, thus improving their performance. To tease out the
different contributions to SNR of CNT-coated microelectrodes we carried out impedance and
noise spectroscopy measurements of platinum/tungsten microelectrodes coated with a
polypyrrole–CNT composite. Neuronal signals were recorded in vivo from rat cortex by
employing tetrodes with two recording sites coated with polypyrrole–CNT and the remaining
two left untreated. We found that polypyrrole–CNT coating significantly reduced the
microelectrode impedance at all neuronal signal frequencies (from 1 to 10 000 Hz) and
induced a significant improvement of the SNR, up to fourfold on average, in the 150–1500 Hz
frequency range, largely corresponding to the multiunit frequency band. An equivalent circuit,
previously proposed for porous conducting polymer coatings, reproduced the impedance
spectra of our coated electrodes but could not explain the frequency dependence of SNR
improvement following polypyrrole–CNT coating. This implies that neither the neural signal
amplitude, as recorded by a CNT-coated metal microelectrode, nor noise can be fully
described by the equivalent circuit model we used here and suggests that a more detailed
approach may be needed to better understand the signal propagation at the electrode–solution
interface. Finally, the presence of significant noise components that are neither thermal nor
electronic makes it difficult to establish a direct relationship between the actual electrode noise
and the impedance spectra.
S Online supplementary data available from stacks.iop.org/JNE/8/066013/mmedia
(Some figures in this article are in colour only in the electronic version)
3
These authors contributed equally to this work.
4
Author to whom any correspondence should be addressed.
1741-2560/11/066013+12$33.00
1 © 2011 IOP Publishing Ltd Printed in the UK
J. Neural Eng. 8 (2011) 066013 G Baranauskas et al
1. Introduction
Extracellular microelectrodes are one of the most important
tools to study the function of the brain in vivo [1]. The
neural signals recorded by extracellular microelectrodes have
provided a significant amount of knowledge about brain
function in behaving animals [2] and are central to a number
of clinical applications [3], including the development of
brain machine interfaces (BMIs), a technology that translates
cortical brain activities into commands for operating robotic
arms or other external devices [4]. Such neural signals are
distributed over a wide range of frequencies and each part
of the frequency spectrum reflects different neural processes.
The low frequency band (approximately 1–250 Hz) contains
the so-called local field potentials (LFPs) and reflects synaptic
activity and other subthreshold integrative processes [5].
Spikes of individual neurons can be detected in the high
frequency range (500–3000 Hz) while the power variations
in the same frequency band represent the overall spiking
activity of multiple neurons close to the electrode tip (multiunit
activity signal, MUA) [5–7]. Because of their different neural
origins, different frequency bands of extracellularly recorded
neural signals often carry complementary information about
external functional correlates such as sensory activity or motor
commands [8–11] and their simultaneous measure increases
the amount of information available, e.g. to control prosthetic
devices [12].
An ideal extracellular microelectrode should provide
the highest signal quality, expressed as signal-to-noise
ratio (SNR), for the whole frequency band containing any
useful neural information. Currently, the vast majority
of neurophysiology laboratories record extracellular neural
signals using microelectrodes with metal tips [1, 2]. However,
their SNR is often far from ideal because of relatively
large noise levels, mainly thought to arise from thermal
noise, directly related to the microelectrode impedance values
[1, 13, 14]. Thermal noise amplitude becomes larger
when the electrode impedance real part increases. The
electrode impedance real part is inversely proportional to the
electrode tip surface area and large area electrodes should
generate thermal noise of small amplitude. Theoretical and
experimental considerations, however, indicate that to obtain
high selectivity in action potential recordings from individual
neurons, the microelectrode tip size in any direction should
not exceed 20 μm, corresponding to the maximal unmodified
surface area of ∼1000 μm
2
[1, 15]. Due to such a small
exposed metal area, the microelectrode impedance is often
very large (∼0.1–1 M) leading to high thermal noise levels
and, consequently, low SNR.
To improve the performance of modern extracellular
microelectrodes, various surface modifications that lower the
tip impedance without significantly increasing the tip size have
been studied [16–24]. A significant advance was recently
made by Keefer and colleagues [21], who proposed using
carbon nanotube (CNT) coatings to increase the tip surface
area, thus decreasing the microelectrode impedance without
significantly affecting the geometrical tip size. This study
reported that coating microelectrodes with CNT composites
improves neural signal quality for all three major types of
extracellular microelectrode signals—LFPs, MUA and spikes
[21]. The reported data demonstrate, however, that there is no
straightforward relationship between the electrode impedance
and its properties. Theory shows that the thermal noise root
mean square (rms) should be proportional to the square root
of the impedance real part [20, 21, 23]. Thus, for the observed
drop in the impedance values of 30–100 times following
CNT coating, the expected thermal rms noise reduction is
more than fivefold or >80%. However, in practice CNT
coating reduces the microelectrode rms noise only by 40–
60%, i.e. approximately twofold, much less than predicted.
This discrepancy between the decrease in the microelectrode
impedance real part and the rms noise contradicts the view
that the impedance determines most of the microelectrode
performance [1, 14, 16, 20, 25]. Similarly, theory predicts
that the neural signal amplitude is attenuated only when
the amplifier input impedance is comparable to the
microelectrode impedance in a given frequency range
[1, 14, 15]. Since most modern amplifiers have input
impedance of 30 M or more, no significant changes in neural
signal amplitude should occur for electrodes, the impedance
of which is well below 10 M. However, the experimental
results show that even in the high frequency range (>500 Hz),
in which non-modified electrode impedance is below 2–3 M,
coating with an electrochemical co-deposition of CNTs and
conductive polymers, typically polypyrrole (PPy), or gold,
increases the neural signal amplitude [20, 21].
These types of coatings were also tested by us [26] and we
found that, compared to the electrodes coated with the gold–
CNT composite, those coated with PPy and CNT composite
(PPy–CNT) have better electrochemical performances, in
agreement with Keefer and colleagues [21]. Although the
PPy–CNT coating can degrade within hours during constant
polarization [27], we found that the electrode properties
are essentially unchanged during several hour long acute
recordings when no polarization is applied and the subsequent
tests showed that the impedance is changed by less than 25%
after such recording sessions.
Based on these results, to understand to what extent
the impedance measurements can predict the microelectrode
performance, we investigated in detail impedance, noise
and neural signals on PPy and chemically-functionalized
multi-walled CNT (MWCNT–PPy) composite-coated metal
microelectrodes.
Our working hypothesis is that the main electrode noise
component derives from sources other than the thermal noise.
In contrast to thermal noise, other noise components are
frequency dependent. Thus, to tease out the contribution
of thermal noise and other noise sources to the overall
microelectrode noise, we compared the impedance and
noise spectra of MWCNT–PPy modified microelectrodes
and non-coated, pristine, platinum/tungsten microelectrodes.
Since both the power of neural signals and the increase in
neural signal power produced by MWCNT–PPy coating are
frequency dependent [21, 28], we also performed neural signal
spectral analysis in order to compare noise and neural signal
levels and, thus, to obtain SNR frequency dependence for the
2
J. Neural Eng. 8 (2011) 066013 G Baranauskas et al
(A) (B) (C)
Frequency (kHz)
impedance phaseelectrode surface
Frequency (kHz)
MWCNT-Ppy
coated metal tip
MWCNT-Ppy surface
non-coated
metal tip
50 μm2μm50 μm
10
0
10
1
10
2
10
3
10
4
10
5
10
1
10
2
10
3
10
4
10
0
10
1
10
2
10
3
10
4
10
5
0
30
60
90
(
p
hase
o
)
pristine
MWCNT-Ppy
pristine
MWCNT-Ppy
k
Z
Ω
II
)
(
Figure 1. MWCNT–PPy coating strongly reduced the impedance of extracellular metal microelectrodes. (A) SEM images at low resolution
show the tips of non-coated and coated microelectrodes. High resolution imaging reveals extensive porosity at sub-micron scale of
MWCNT–PPy-coated microelectrodes. (B), (C) Impedance spectroscopy shows a significant decrease in the impedance modulus (B) and a
complete change in the phase frequency dependence (C). The light orange shaded area corresponds to the frequency range in which the
background noise levels were reduced significantly after MWCNT–PPy coating.
whole neural signal frequency band. We found that, though
the impedance spectroscopy can be used to predict some
electrode properties, there is no direct relationship between
the microelectrode performance and its impedance. This
observation is explained in part by the presence of significant
noise components with no direct relationship to the impedance
values.
2. Materials and methods
2.1. Metal microelectrodes
Single core quartz insulated metal electrodes were prepared in-
house by mechanically grinding 95% platinum/5% tungsten
microwires of 20 μm diameter coated with 30 μm of quartz
(ThomasRecording, Giessen, Germany). The impedance of
these electrodes, measured in a saline solution at 1 kHz,
was typically between 500 and 700 k and the estimated
exposed metal tip area was ∼900 μm
2
[26]. For neural
activity measurements, two platinum/tungsten tetrodes
(ThomasRecording, Giessen, Germany, 400–1200 k at
1 kHz) were used. These specially designed microelectrodes
have four independent recording sites of diameter between 14
and 26 μm, separated by less than 40 μm on the tip of the
same shaft (figure 6(B)). We have chosen such electrodes as
they make it possible to record signals of the same neuron from
more than one site [29]. In our experiments, two recording
sites of the tetrode were coated with MWCNT–PPy while the
other two remained untreated.
2.2. PPy–CNT deposition
We made a CNT electrochemical deposition with pyrrole
in an aqueous solution using procedures recently described
[21, 26, 30]. Briefly, PPy and chemically-functionalized
multi-walled CNTs were co-electrodeposited from an aqueous
solution of 0.5 M pyrrole (Sigma-Aldrich), 1 mg ml
−1
COOH–
MWCNTs (Nanocyl 3151, <4% of –COOH functional
groups) and 0.4% poly-sodium 4-styrene-sulfonate (PSS)
(Sigma-Aldrich). As purchased, COOH–MWCNTs were
suspended in ultrapure water (Milli-Q, Millipore) by horn
sonication (6 s at 66% duty cycle pulses, 4 W ml
−1
)for
minimum 15 min and up to 60 min while keeping the solution
cooled with an ice bath. PSS and pyrrole were added to the
suspension immediately afterwards and the solution was kept
deoxygenated by bubbling with nitrogen. The electrochemical
deposition was carried out under an inert atmosphere in
potentiostatic mode. The polymerization potential was set
to 550 mV versus Ag/AgCl reference electrode while the
deposition time varied from 5 to 50 s. The amount of charge
passed during the deposition was between 100 and 400 mC
cm
−2
.
The result of electrochemical deposition was the
formation of a nanostructured coating of PPy and MWCNTs
on the microelectrode exposed tip (figure 1(A)). Reproducible
coatings could be attained as demonstrated by the impedance
modulus and phase measurements shown in figures 1(B)
and (C). Following preliminary tests, we selected deposition
parameters in such a way that the microelectrodes used for this
study had impedance at 1 kHz no more than 25 k and a total
charge transfer capability no more than 200 mC cm
−2
.
2.3. Electrochemical and morphological characterizations
All measurements were carried out using a potentiostat/
galvanostat (Parstat 2273, Princeton Applied Research,
Oak Ridge, TN, USA) and a standard three-electrode
electrochemical cell configuration with a Pt wire as a counter
electrode and the Ag/AgCl reference electrode.
Galvano electrochemical impedance spectroscopy was
performed in a saline solution (0.9% NaCl) by applying a
sine wave of 300 rms nA current amplitude. The impedance
values were determined at ten frequencies per decade over the
range 1–10
5
Hz.
For cyclic voltammetry tests the potential on the working
electrode was swept between 0.6 and −1.0 V versus Ag/AgCl
in saline solution at a scan rate of 100 mV s
−1
, starting at
open-circuit potential and sweeping in the positive direction
first. The total charge transfer capability was calculated as the
time integral of a whole cyclic voltammetry cycle.
The morphology of CNT nano-composites was examined
using a cold field emission gun high resolution scanning
electron microscope (HR-SEM, Jeol JSM-7500 FA)
(figure 1(A)).
3
J. Neural Eng. 8 (2011) 066013 G Baranauskas et al
Figure 2. Equivalent circuit model of electrode–nanocoating–
electrolyte interface. WE stands for the working electrode and RE
for the reference electrode. Role and equation of the single elements
are reported in table 1.
2.4. Fitting an equivalent circuit model to the impedance data
To describe in detail the MWCNT–PPy-coated electrode–
electrolyte interface impedance, we used the equivalent circuit
model (figure 2), proposed by Abidian and colleagues [31],
that takes into account the conducting polymer coating
interface, and the corresponding equations:
Z = R
s
+ Z
elect
= R
s
+
1
jωC
c
+
1
R
p
+
1
1
Z
CPE
+
1
R
t
+Z
T
, (1)
where R
s
is the solution resistance, Z
elect
stands for the
electrode impedance without the solution impedance, j is
the imaginary number and ω is the angular frequency, C
c
is the coating capacitance, R
p
the pore resistance, Z
CPE
the
interface capacitance, R
t
the charge transfer resistance and Z
T
the finite diffusion element.
To verify if the model satisfied the impedance response
of our modified neural microelectrodes we estimated
the parameter set of equation (1) by a best-fit procedure of
the experimental impedance (both modulus and phase).
The best fit was achieved by minimizing (over the space
of parameter values) the sum of square errors of model
predictions of both modulus and phase. The minimization
was performed using the Levenberg–Marquardt down-hill
algorithm while constraining the parameters to remain in the
real domain. We used as initial guess for the parameters
the values for PPy nanotubes reported by [31], rescaled for
the geometrical area of our electrode (900 μm
2
). The statistical
test of goodness of fit was assessed using the left χ
2
test [32]
which reports the p value of the null hypothesis that data
are generated by the model. We set the model acceptance
threshold to p > 0.95. We assessed the tolerance range for each
parameter by varying its value around the best-fit value and
computing the goodness-of-fit χ
2
p value after each parameter
variation. The tolerance range for each parameter was given
by the range of variations for which the χ
2
test significance
remained above p = 0.95. The narrower the tolerance range,
the more accurate the estimation of the parameter value is.
The wider the tolerance range, the less sensitive to the precise
value of the parameter the fit between model and data is
(figure 3).
2.5. Animal surgery
Experiments were carried out in acute sessions on eight
anaesthetized Long-Evans male rats, weighing 300–400 g. Six
animals were used for single electrodes and two for tetrodes.
The experimental plan was designed in compliance with the
Italian law regarding the care and use of experimental animals
(DL116/92) and approved by the institutional review board of
the University of Ferrara and by the Italian Ministry of Health.
Rats were anaesthetized with a mixture of Zoletil (30 mg kg
−1
)
and Xylazine (5 mg kg
−1
) delivered intraperitoneally. For
the duration of the experiment, the depth of anaesthesia was
monitored by testing for the absence of hindlimb withdrawal
reflex and was maintained by additional doses of anaesthetic
(i.p. or i.m.). Under anaesthesia, the body temperature was
maintained at 36–38
◦
C with a thermostatically controlled
heating pad. In each recording session the anaesthetized
animal was placed in a stereotaxic apparatus (Myneurolab,
St Louis, MO) and a small craniotomy (2 × 2mm
2
)was
made in the parietal bone to expose the vibrissa region of the
somatosensory cortex according to vascular landmarks and
stereotaxic coordinates [33–35]. Dura mater was left intact
and quartz–platinum/tungsten microelectrodes and tetrodes
were lowered perpendicular through the cortical surface using
a hydraulic microdrive (Kopf, 2650) to a depth of >900 μm
(infragranular layer) [36–38]. After testing and confirming
the placement of the electrodes by recording the extracellular
neuronal discharges to manual whisker stimulation, the
spontaneous activity of infragranular layer, consisting of
bursts’ firing (action potential clusters), was recorded [39].
Seven MWCNT–PPy-coated and five control, non-coated
single electrodes were used in this study. Both modified
and non-modified electrodes were tested on the same day,
i.e. for each rat we tested at least one MWCNT–PPy-coated
and one non-coated electrode. For each electrode one to three
penetrations/tracks in a rat brain were made, in total seven
tracks for non-coated electrodes and ten tracks for MWCNT–
PPy-coated electrodes. Three tracks were made with each of
the two tetrodes. Each tetrode had two recording sites coated
with MWCNT–PPy while the remaining two recording sites
were left uncoated.
2.6. Neural recordings and spike sorting
The neural activity was recorded using a Plexon Multichannel
Acquisition Processor (Plexon Inc. Dallas, TX, USA) 1000×
preamplifier connected to a high impedance headstage
(40 M,HST/16o25-18P-GR) with the amplifier ground
connected to the ground electrode. Most analyses were
performed with custom written routines employing Igor Pro
program (Wavemetrics, Lake Oswego, OR, USA) while
spike sorting was performed with the Plexon offline sorter
software.
For single unit separation (spike sorting), we employed
the Plexon Expectation Maximization algorithm followed by
manual verification of spike cluster quality. Our goal was
to compare the quality of unit separation for different types
of electrodes and not to achieve the best sorting results. We
were more concerned to have a reproducible procedure and
to have confirmation of the presence of a single unit in a
cluster.
4
![Figure 3. Measured impedance spectra and best model fit. (A) Impedance modulus. The grey area represents the mean ± standard deviation range of the recorded data. The black line represents the best fit of the dataset obtained using the equivalent circuit model (figure 1, equation [1] in section 2). The grey line indicates the average impedance module across data. (B) Impedance phase. The fit was performed simultaneously for impedance phase and modulus and the best obtained fit with the parameter values of table 1 is shown (left χ 2 test of goodness of fit; p = 0.985 for the range 1 Hz–10 kHz). (C) Tolerance range for parameters estimated in the model. Parameters are normalized by dividing them by the corresponding best-fit values of table 1. Each parameter was modulated in the range indicated on the x axis while all the others were fixed to their best-fit value. The grey level bar indicates the level of significance of the χ2 goodness-of-fit test for the model for the given parameter value (up to a rejection threshold of p < 0.95 indicated in white). The tolerance range for each parameter is indicated by the extent of the area around the reference value in which the p value remains above 0.95. Rs , Cc, Rp and Rt definitions are provided in table 1. Q and n determine the behaviour of the interface capacitance ZCPE and Y0 and B are the parameters in the equation for the diffusion element ZT (see the corresponding expressions in table 1).](/figures/figure-3-measured-impedance-spectra-and-best-model-fit-a-jbrr7mz2.webp)
