|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
IDENTIFICATION OF NEURAL ACTIVITY INTAPLYSIA ABDOMINAL GANGLIA USING
A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
UNIVERSITY OF FLORIDA
O 2007 Hany Elmariah
To my parents and to Sammy and Sarina, whose love and support
has carried me throughout my life.
I thank Dr. Thomas DeMarse and Dr. Rosalind Sadleir, whose direction and guidance were
invaluable to the completion of this proj ect and my personal development as a researcher. I also
acknowledge Karl Dockendorf, Michael Furman, and II Park, all of whom made contributions to
the experiments described in my work. Finally, I thank April-Lane Derfinyak, who has worked
tirelessly to keep every student in our department on track to accomplish our goals.
TABLE OF CONTENTS
ACKNOWLEDGMENT S ........._.._ ..... ._._ ..............4.....
LIST OF TABLES ........._.._ ..... ._._ ..............7.....
LIST OF FIGURES ........._.._ ......___ ..............8....
AB STRAC T ........._.._ ..... ._._ .............._ 10...
1 INTRODUCTION ................. ..............12.......... ......
Methods for Studying Electrophysiology of Multiple Neurons ................ ................ .. 13
Impedance Tomography ................ ........... ..............18. .....
Ion Channels and Neuronal Impedance.............. ................18
Experimental Goal s.............. .............. 20..
2 MATERIALS AND METHODS.............. ..............25.
Aplysia Preparation .............. .............. 25....
Microelectrode Array Setup .............. ..............26.....
Data Acquisition .............. .............. 28....
Data Analysis .............. ..... .. ......... ........... ............3
Least-Mean-Square Adaptive Filter ................ ..............31. .......... ....
Spike Detection ................. ..............33. ...............
Moving Average Algorithm .............. ..............33.....
3 RE SULT S .............. .............. 40
Patterns of Activity ................. ..............40................
Applied Currents ................ ..............42. ...............
Total Spikes Over Time ................. ..............43.......... .....
Moving Average Filter ................ ..............44. .......... ....
Correlating Activity and Resistance.............. ..............45
Frequency Analysis.............. ..............47.
4 DI SCU S SION ................. ..............64................
Patterns of Activity and Raster Plots ................. ......... ..............64. ...
Applied Currents .................. ........... ..............65. .....
Resistance Changes and Activity ................ ..............66. ..............
Microelectrode Arrays and Future Work .............. ... .. ..............71.
Functional Magnetic Resonance Electrical Impedance Tomography. ................ ...............73
5 CONCLUSION .........._.._ ......... ..............79......
LIST OF REFERENCES .........._.._ ......... ..............80.....
BIOGRAPHICAL SKET CH .........._.._ ......... ............... 86....
LIST OF TABLES
1-1 Summary of firing patterns observed for three primary cell groups on the dorsal
surface of the abdominal ganglion ................. ......... ..............24......
LIST OF FIGURES
1-1 Image of a standard microelectrode array from Multichannel Systems ................... ........23
1-2 Diagram of the dorsal surface of the abdominal ganglion. ................ ............ ........24
2-1 Multichannel Systems 3-D microelectrode array ................ .............................35
2-2 Aplysia ganglion centered on the dish of a microelectrode array .............. ..................36
2-3 Oscilliscope function of MEABench on Mac OS X ................. .......... ...............37
2-4 Spontaneous action potential recorded from abdominal ganglion cell.............. ...............37
2-5 Schematic of the MEA setup.. .........._._._ ..............38..._..__...
2-6 Flow diagram of the experimental apparatus ......_._.__ ... .... .___ ... ..._.._........3
2-7 Example of LMS filtration. ........._.. ........_. ..............39....
3-1 Raster plots for spontaneous activity of Subj ect 1 ................ ............................49
3-2 Raster plots for Subj ect 2. ................ ..............50. .......... ..
3 -3 Raster pl ot for S ubj ect 3 ................ ................. 5......... 1..
3-4 Spike frequency over time for Subj ects 1 and 2. ................ ..............52. ..........
3-5 Spike frequency over time for Subj ect 3 ................ .......... ....... ...........5
3-6 Resistance traces from Subj ect 1 during 10 ELA/10 Hz current. ........._._.... ......_._.....54
3-7 Resistance traces from Subj ect 2 during 10 ELA/10 Hz current. ........._._.... ......_._.....55
3-8 Resistance traces from Subject 3 during 5 ELA/10 Hz current.............. .................5
3-9 Resistance traces from Subj ect 3 during 10 ELA/10 Hz current. ........._.._.._ ......_.... ....57
3-10 Comparative analy si s for Subj ect 1 during appli cati on of 10 ELA/10 Hz current .............5 8
3-11 Comparative analy si s for Subj ect 2 during appli cati on of 10 ELA/10 Hz current .............5 9
3-12 Fourier power spectrum of the moving average trace for Subject 1 during application
of a 10 ELA/10 Hz current ........._.._.. ...._... ..............60...
3-13 Fourier power spectrum of the moving average trace for Subj ect 2 during application
of a 10 ELA/10 Hz current ........._.._.. ...._... ..............61...
3-14 Fourier power spectrum of the moving average trace for Subj ect 3 during application
of a 5 ELA 5 Hz current ........._.._.. ...._... ..............62...
3-15 Fourier power spectrum of the moving average trace for Subj ect 3 during application
of a 10 ELA 10 Hz current ........._.._.. ...._... ..............63...
4-1 Moving average trace for Channel 36 of Subj ect 2 during application of 10 ELA10 Hz
current ........._.._.. ...._..._ ..............74.....
4-2 Raster plot for Subj ect 2 during application of 10 ELA10 Hz current highlighting
Channel 36 ........._.._.. ...._..._ ..............75.....
4-3 Image of four channels from the oscilliscope during Subj ect 1 recordings with
current ............ _...... ..............76....
4-4 Alternate resistance traces for Subj ect 1 during application of current. .........................77
4-5 Alternate resistance traces for Subj ect 2 during application of current. .........................78
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science
IDENTIFICATION OF NEURAL ACTIVITY INTAPLYSIA ABDOMINAL GANGLIA USING
Chair: Thomas DeMarse
Major: Biomedical Engineering
The development of a technique to effectively measure neural activity in vivo with high
spatial and temporal resolution is among the maj or challenges currently facing
neurophysiologists. A promising solution to this problem is the use of impedance tomography to
indirectly measure neural activity through inherent changes in resistivity of active neural tissue.
In order to demonstrate the viability of this solution, a consistent correlation between neural
activity and neuronal impedance must be verified. I investigated this relationship using neural
activity recorded from Aplysia abdominal ganglion with a 60-channel microelectrode arrays
(MEAs). In each trial, changes in the amplitude of a sinusoidal carrier wave, the result of
injecting sub-threshold AC currents at amplitudes ranging from 10-40 ELA, were analyzed to
determine the spatiotemporal characteristics of impedance changes in the tissue. Changes in
impedance were compared with the neural activity independently measured using a MEA. The
impedance results and the action potentials from the MEA were then analyzed in search of
spatiotemporal correlations between each measure. Spontaneous neural activity with patterns
similar to those reported in the literature were successfully recorded from the varied cell bodies
within the abdominal ganglion. I then compared impedance with activity from the MEA in both
the frequency and time domain finding only a modest change in impedance during the
production of action potentials in the ganglion. Future directions are suggested which may
improve the resolution of this method.
From the primitive bottom dwellers of the ocean to the most advanced organisms on
earth, virtually every member of the animal kingdom is linked by a common bond: the nervous
system. In all such organisms, the nervous system is crucial in controlling basic processes such
as breathing and hunger, as well as more advanced processes such as coordination and cognitive
thought. One of the key issues, however, is the understanding of how the structural foundations
of the nervous system produce their characteristic function (i.e., structure-function relationship).
This problem can be approached from a number of varied strategies arising from a reductionist
approach that begins at the basic building blocks of the nervous system, the neuron, studying the
chemical communication between neurons, such as neurotransmitters and growth signaling [1-3].
Meanwhile, others have focused on the anatomical structures of neural tissue [4-6], or on the
electrophysiology of individual neurons. However, in order to understand how the structure of
these individual building blocks leads to their functional outcomes (behavior) it is clear that more
information is needed about the activity across much of the network rather than the individual
neurons themselves. Technology that can provide information from both the network and
individual neurons with high spatial and temporal resolution would be invaluable.
The basic building block of the nervous system is the neuron. A neuron consists primarily
of dendrites, the soma (cell body), axons, synapses, and ion channels. The soma is the main body
of the neuron. From the soma extend a number of branching protrusions called dendrites which
serve to receive electrical signals and transmit them to the soma. Similarly, axons also extend
from the soma. These axons function to transmit electrical currents from the soma to another
neuron. The inter-neuronal junction which connects an axon of one neuron to the dendrites of
another neuron is known as the synapse. Ion channels, through which ions are selectively
allowed to flow, are dispersed throughout the cell membrane of the neuron . This collection of
fundamental units forms a dense network of interconnectivity that provides the foundation for
the structure-function relationship.
At rest, the environment of the soma of a single neuron is not electrically neutral. The
channels of the cell selectively block and permit the flow of charged ions through the cell
membrane. Most importantly, these channels allow the flow of sodium and potassium ions. An
energetic mechanism known as the "sodium potassium pump" pushes potassium into the cell and
sodium out of the cell in order to maintain a negative resting potential. When a neuron is
stimulated, such as by an electrical current, the sodium channels are opened, which allows
sodium ions to flow down their electrochemical gradient into the cell. This depolarizes the cell,
increasing the potential to values greater than 0 volts. This voltage increase triggers the
potassium channels to open which allows potassium to flow out of the cell. Meanwhile, the
sodium channels begin closing. As such, ions are now only flowing out of the cell, returning the
soma to resting potential. This process of depolarization and repolarization as the result of
stimuli is called an "action potential." From the soma, action potentials travel down the axon to
the synapse. This triggers the release of chemical neurotransmitters and ions which stimulate the
dendrites of another neuron, and the process begins again .
Methods for Studying Electrophysiology of Multiple Neurons
The work of Bernstein, Curtis, Cole, Hodgkin and Huxley resulted in an understanding of
neuronal electrophysiology and the means by which neurons directly communicate with one
another. This has set the foundation for today's researchers to focus on how multiple neurons
work together as a network in complex neural systems. Current research requires measuring
techniques which can record field potentials and activity patterns from large numbers of neurons.
The ideal measuring technique would encompass a wide spatial area with high temporal and
detailed spatial resolution, starting from a single neuron up to the entire brain. Unfortunately, no
single technique exists which fulfills this ideal.
Three of the most popular imaging techniques are electroencephalography (EEG),
functional magnetic resonance imaging (fMRI), and microelectrode array (1VEA) recordings.
Each of these methods is useful for certain tasks, yet none of them represent a complete solution
to the difficulties of studying electrophysiology, especially across a network. Each represents
tradeoffs between the area that can be imaged, the spatial resolution within that area, and the
temporal resolution they provide. Despite their utility, every technique used to study
electrophysiology is hindered by inherent limitations of the techniques used to indirectly detect
One of the most commonly used methods to measure activity from large areas of neural
tissue is EEG. First developed by Hans Berger , this technique is noninvasive, using
electrodes positioned on the outer surface of the scalp. The electrodes measure the extra-cellular
potentials of the neural tissue, electrical fields , creating a spatiotemporal map of the activity
of the brain. Using EEG, researchers can monitor global fluctuations in extra-cellular potentials,
particularly EPSPs (excitatory post-synaptic potentials) and IPSPs (inhibitory post-synaptic
potentials), of different regions of the brain during various activities or events. EEG has been
used to evaluate the electrophysiological causes of neural disorders, such as epileptic seizures,
including where the seizure begins and the pattern by which it propagates [9, 10]. Another
application of EEG is to correlate neural activity with certain behaviors, such as voluntary
movement, in order to develop brain-machine interfaces [11-13]. EEG can also used to elucidate
neural activity during sensory input [14, 15] and sleep studies [16, 17].
EEG measurements are characterized by good temporal resolution, but are also hindered
by weak spatial resolution. With relatively large electrodes placed on the scalp, the resolution is
limited by the physical proximity of the electrodes to one another. This limitation is furthered by
the fact that EEG electrodes detect Hield potentials from a wide surface area (about 10 cm2)
Placing two electrodes within this area will result in the simultaneous detection of field potentials
by both electrodes. The spatial resolution of this technique is further limited by the multiple
layers of tissue between the electrodes and the brain (blood vessels, dura, skull, skin, etc.). These
layers attenuate the signals, attenuating and distorting the observed surface potentials and
completely eliminating potentials from deeper brain layers. As a result, the activity recorded
using EEG is reduced to a spatial average of activity, rather than a detailed, spatially localized
observation of specific neurons .
To improve spatial resolution, other methods have been developed which also can
indirectly image neural activity, such as fMRI, a method closely related to traditional magnetic
resonance imaging (MRI). The MRI was developed as a noninvasive means to image the
structural information of tissue. The technique is based on changes in the dipolar relationship of
water molecules in almost all living tissue and their interaction with magnetic Hields. The
magnetic Hield applied during the MRI causes these dipoles to align, and they are then perturbed
by application of radiofrequency energy. As the dipoles recoil to their resting position, they emit
energy which is then imaged. One of the advantages of this technique is that it can provide
relatively high spatial resolution (< 1 mm). While MRI is useful in mapping neural anatomy, it is
inadequate to measure any structural-functional relationships because this technique does not
provide any information about the neural activity within the areas that are imaged .
Ogawa, et al.  noted that cortical blood vessels are more clearly visible as blood
oxygen decreases due to the resulting changes in the magnetic Hield. Moreover, neurons require a
great deal of energy during activity, a process which utilizes significant amounts of oxygen. This
creates notable discrepancies in the oxygen concentrations in the arteries near the active neurons
as compared to the nearby veins. Putting these facts together, Ogawa concluded that directly
imaging changes in oxygen concentration can indirectly produce an image of neural activity.
This new method for using MRI to study neural function is appropriately titled fMRI [8, 19, 20].
Like EEG, fMRI is useful for creating spatiotemporal maps during various neurological states.
For example, fMRI has been used to study neural physiology of various emotional states [21,
22], body movement , and verbal comprehension .
Neural imaging with fMLRI provides some notable advantages. Most importantly, like
MRI, it provides high spatial resolution. Like EEG, the technique is also non-invasive. There are
some significant limitations. The specific physiology behind the altered levels of blood oxygen
in the vessels, and the ensuing ability to indirectly image activity, is not known. While fMRI can
determine the existence of activity at a spatiotemporal location, it cannot determine the type of
activity (excitatory, inhibitory, action potential propagation, etc.). The indirect nature of fMRI
imaging creates the potential that unique processes and states of neuronal activity may
nonetheless result in one common state of the blood oxygen levels in the vessels. In such
situations, the fMRI is unable to distinguish between the various neuronal activity states, a
problem that is also shared with the Hield potentials using EEG. Finally, there is a delay between
the occurrence of neural activity and the ensuing use of oxygen by the neuronal tissue. Thus,
multiple frames of images must be integrated over time to detect any changes which decreases
the temporal resolution of this technique .
Another method which provides much higher spatial and temporal resolution is the MEA.
An MEA consists of a grid of multiple electrodes, each of which has a diameter on the order of
50 microns or more. These electrodes are able to record extracellular potentials from multiple
spatial locations simultaneously. A wide variety of electrode arrays have been developed
following this basic idea including arrays such as microwire bundles, shaft electrode arrays,
flexible grid electrodes for in vivo applications and planar arrays for in vitro studies. Figure 1-1
shows an example of an array from Multichannel Systems designed for in vitro work. The MEA
is square in shape and fitted with a small, cylindrical dish centered on its face which contains the
tissue and medium used to support it. The electrode grid is embedded in the center of this dish.
The electrodes are made of either titanium nitride or platinum, depending on the application. A
series of 60 contact pads, made of either titanium or indium tin oxide, are embedded along the
perimeter of the MEA. These are connected to the electrodes via conductive tracks such that the
electrical signals may be sent from the tissue to the electrode and then to the contact pads
through the tracks. From the contact pads, signals are amplified with a filter amplifier and then,
Einally, sent to the data acquisition computer . This style of MEAs has been used for studying
pharmacology [26, 27] neural regeneration , synaptic plasticity [29, 30], visual perception
[31, 32], and robot control [33, 34]. Additionally, MEAs have been used extensively for studies
in the Hield of cardiology [35, 36].
Because MEAs directly record extracellular potentials at a reasonably high sampling
frequency, this technique boasts very little time delay and good temporal resolution. Spatially,
MEAs can record Hield potentials or even record from individual neurons. Spatial resolution is
limited by the number of electrodes that can fit on the surface of the MEA. The smallest
electrode size available on MEAs is currently 10 Epm, so the spatial resolution cannot be more
precise than 1 electrode in a 10 Epm area. Resolution is further limited by the fact that each
electrode records spike activity from an area of tissue, not an infinitesimally small point. Each
electrode detects spikes from a radius of 100 Epm. Even placing the electrodes as close as
physically possible will simply result in repetitive detection of potentials. The most obvious
disadvantage of 1VEAs is their inability to record from large-scale tissue samples, such as an
entire human brain. At most, a planar IVEA such as that shown in Figure 1-1 can be applied to
recordings from small slices of neural tissue. For larger tissue samples, clinical methods such as
EEG, fMRI, etc. are required .
Impedance tomography is a relatively new modality which is used to image the electrical
impedance of a tissue sample. Electrodes are placed in a ring around the tissue sample. One pair
of electrodes is used to propagate a test current, while the rest of the electrodes are used to record
the resulting voltages. This technique is repeated in varying orientations through the tissue. A
reconstruction algorithm is then used to create an image of the impedance throughout the tissue
sample. Technically, only four electrodes are necessary to obtain enough information to create an
impedance map. However, spatial resolution is dependent upon the number of electrodes. The
final images created are generally two-dimensional slices . This technique has a number of
applications including monitoring lung function [38, 39] and detecting various types of cancer
Ion Channels and Neuronal Impedance
Early evidence of the role of ion channels for propagating electrical signals in neurons
came from studies in the first half of the twentieth century of impedance within the neural tissue.
In 1902, with knowledge of the electrical potential gradients across the neuronal membrane,
Bernstein hypothesized that neural activity occurred when this gradient disappeared as a result of
increased permeability of the membrane to ions. This hypothesis came as the result of
mathematical analysis using the Nernst equation, by which Bernstein came to the belief that,
during activity, the membrane became completely permeable to all ions and resulted in a
depolarization of the cell to 0 volts .
Experimental support for Bernstein' s hypothesis was later established by Curtis and Cole
in 1939 . In those experiments, Curtis and Cole studied impedance within the squid giant
axon. The pair extended the axon and stimulated from one end, allowing propagation of a current
to the other end. Meanwhile, they measured the transverse impedance with electrodes positioned
on each side of the axon and an oscillograph. The results showed that the impedance within the
axon did, indeed, change during activity. In fact, Curtis and Cole found a 30-40 fold increase in
the conductance of the axon during action potentials. They also found that there was a slight
delay in the impedance change. Rather than an immediate change in conductivity occurring in
unison with the depolarization phase of the action potential, the impedance decrease was found
to occur in conjunction with the repolarization phase and the decrease in the depolarizing
Since the studies of Curtis and Cole, other researchers have further elucidated the
relationship between impedance and activity in neural tissue. Most importantly, in 1952,
Hodgkin and Huxley were finally able to determine the permeability of channels to various ions
throughout the course of action potentials [44-48]. The key to their success was their application
of a new space clamp technique. For this method, a conductive wire was threaded into the axon.
This brought the entire length of the axon to the same electrical potential. Thus, the position of
the measurement on the axon became irrelevant, and Hodgkin and Huxley were able to study the
temporal changes of current flow and conductivity across the membrane . Through these
studies, Hodgkin and Huxley created a set of differential equations which modeled the behavior
of neurons during action potentials, the basis of our current understanding of ion channel
behavior and permeability during activity.
Since the Nobel Prize-winning breakthroughs of Hodgkin and Huxley, researchers have
continued to focus on current flow and impedance in neurons. It has been shown that, at low
frequencies, neuronal membranes are essentially non-conducting. Any applied, external current
will flow exclusively through the extra-cellular space, and intracellular conductivity will
ostensibly be zero [49-53]. During activity, as the voltage gated channels in the membrane are
opened, causing depolarization and then repolarization, the impedance of the neuron decreases
and allows current to move intracellularly. Such changes have been observed in cats [54, 55],
rabbits [56, 57], rats [58, 59], and even human neural tissue [60, 61].
The direct correlation between impedance shifts and neural activity has been challenged, at
least in part . In studies using implanted cortical electrodes in cats, Klivington and Galambos
demonstrated that shifts in impedance within neurons occur only in conjunction with action
potentials. Moreover, these same studies demonstrated that the shape of the impedance shift is
constant when it does occur and, thus, independent of the shape and magnitude of the associated
action potential. They also found that action potentials may occur in the absence of any detected
shift in neuronal impedance. It is not clear whether impedance may be a viable alternative to
other imaging techniques for neural activity.
The purpose of my study was to determine whether resistance changes can be correlated
with neural activity. In other words, can resistance information be used as a unique method to
measure and identify neural activity across a small population of neurons? To accomplish this,
neural activity was independently recorded using 1VEA technology and compared with the
activity measured using impedance tomography. For preliminary studies, I studied the abdominal
ganglion of Aplysia californica, a small sea slug. Figure 1-2 shows a diagram outlining the
maj or cell bodies found in the abdominal ganglion. The ganglion consists of two hemispheres
(labeled as "L" and "R") of cells which are held together by connective tissue and covered by an
insulating sheath. These hemispheres are further subdivided into the left rostral quarter-ganglion
(LRQG), right rostral quarter-ganglion (RRQG), left caudal quarter-ganglion (LCQG), and right
caudal quarter-ganglion (RCQG), in which the ganglion neurons are located . One of the
primary advantages of using this ganglion is that distinct patterns of activity are produced by
each area of cell bodies. Hence, the activity recorded from both the IVEA and impedance
tomography can be localized anatomically based on the activity pattern that is detected.
Studying the electrophysiology of Aplysia abdominal ganglion using 1VEAs was first
accomplished by Novak and Wheeler in 1986 . In those studies, action potentials were
recorded from the ganglion cells using a 32 electrode, planar IVEA. The ganglion is surrounded
by an insulating sheath; for this reason, Novak and Wheeler measured only from the dorsal
surface of the ganglion, over which the sheath is thinner compared to the ventral side. They
reported three patterns of activity, which they used as a criterion to divide the right dorsal
ganglion into three groups: cells around R15, the RB/RC region, and the white rostral cells
(Figure 1-2). R15 can be observed in the RCQG (blue region). Lateral and rostral (above and to
the right on the Eigure) to R15 are two clusters of cells, identified as RB and RC. The white
rostral cells can be observed even further rostral, in the RRQG (yellow region) . Results
showed that neurons near the R15 neuron fire with rhythmical bursts of activity once every 10
seconds (0.1 Hz). In the regions near RB and RC, rapid firing, near 2 Hz, was observed. Finally,
Novak and Wheeler observed firing rates of 0.5 to 1 Hz in the rostral white cells. These results
are summarized in Table 1-1 .
In light of those studies, we have used a 60 channel 1VEA in order to observe ganglion
activity with unprecedented resolution. Further, we have applied this technique for the novel
purpose of observing resistance shifts during action potentials. Specifically, we inj ected the
ganglion cells with a constant AC current during each trial while recording voltages with the
IVEA. Using a constant current source, the resistance can be calculated by simply dividing the
measured voltages by the amplitude of the current, as defined by Ohm' s Law. This value was
calculated over the duration of each trial using a method combining synchronous demodulation
and a moving average algorithm. Meanwhile, a spike detection algorithm was used to determine
the spatiotemporal location of action potentials in the neural tissue. Spatiotemporal patterns of
resistance changes were then compared to spatiotemporal patterns of neural activity in hopes of
showing that action potentials are generally accompanied by changes in neuronal resistance, and
In the following experiment, the neural activity from the abdominal ganglion was assessed
in three animals using planar IVEA technology. A sinusoidal current source was then inj ected
into the IVEA using external electrodes to image activity using changes in resistance that could
be correlated with the activity detected via the IVEA. This experiment tested four different
current amplitudes ranging from 5 to 40 ELA to determine which current level provides the best
resolution. Consistent and reliable changes in resistance that correspond with changes in activity
would provide support for the viability of this technique for future studies.
Figure 1-1. Image of a standard microelectrode array from Multichannel Systems. This 60
channel array consists of 30 Epm electrodes spaced 200 Epm apart. Each electrode can
measure the electrical changes produced during action potentials of neuronal tissue.
The right panel shows and example of an MEA with dissociated rat cortical neurons.
The lower left panel shows and example of an 8x8 grid of extracellular potentials
recorded from neurons (200 ms window for each electrode).
Table 1-1. Summary of firing patterns observed for three primary cell groups on the dorsal
surface of the abdominal ganglion .
Cell Group Pattern of Firing Frequency (Hz)
R15/surrounding cells Bursting 0.1
RB/RC Irregular 2
White Rostral Cells Regular 0.5-1.0
DORSAL SURFACE OF GANGLION
Lefi Rosiral Qularlr~-Gangllon
-`- C~ Cells Lef icudal Quadr-GaJngllon
SiphR Nrve _JRighi Rosiral Gua~rtGanglion
Genital-pericardial Nerve Branchjal Nerve
Figure 1-2. Diagram of the dorsal surface of the abdominal ganglion. The neuropils, which lie
dorsal to each hemisphere, have not been included in order to show the neurons.
MATERIALS AND METHODS
Aplysia californica for all experiments were obtained from the NIH Aplysia facility at the
University of Miami. For each set of experiments, the animals were delivered the week
preceding experimentation and then stored in a 10-gallon saltwater tank. The tank was filled with
regular tap water and then a commercial aquarium salt mix was added. The salinity was
measured with a hydrometer and adjusted to a level of 33.5 ppt with a specific gravity of 1.024 at
room temperature (75 degrees Fahrenheit). To neutralize chemicals in the water which could
potentially harm the Aplysia, 5 mL of a commercial water conditioner was added to the tank. A
water pump was then attached to the side of the aquarium to circulate the water. While kept in
the tank, the animals were fed a piece of iceberg lettuce every 1-2 days.
Preceding each set of experiments, the abdominal ganglion of an Aplysia was surgically
removed. First, a solution of 77.05 g/L of MgCl2 and 3.5745 g/L of HEPES buffer was made for
use as an anesthetic. The animals were placed on a foam plate with the animal's lateral side
facing up. The anesthetic was inj ected into the foot process, middle, and head. As a defense
mechanism, Aplysia may release ink during the injections. In order to avoid this, the animal was
first inj ected with 5 cc of anesthetic with a 1 cc syringe through a 2861/2 needle. Five inj sections
were made into the foot process, followed by the middle, and Einally the head. This technique
was repeated such that a total of 20-25 cc of anesthetic was inj ected into the organism through
the small gauge needle. After these first inj sections, it is safe to use a large, 16G1/2 gauge needle
and inj ect 10 mL of anesthetic with each inj section. Again, the anesthetic was inj ected at the foot,
middle, and head. This technique was repeated until the animal was non-responsive to physical
stimuli and body movement had ceased, generally a total of approximately 80 mL of anesthetic.
All inj sections were made perpendicular to the longitudinal axis (with the needle pointed down
into the foam plate). This prevented the anesthetic from being inj ected into the abdomen and
damaging the ganglion.
Once the specimen was anesthetized, the surgery proceeded with a mid-dorsal,
longitudinal incision from the foot process to the head. A scalpel was used to gradually dissect
the connective internal connective tissue. As necessary, the skin was pinned back in order to
expose the abomen of the organism, allowing access to the ganglion. The ganglion can be
identified as a small, orange bundle in the center of the abdomen. It is held in place by nervous
tissue, which was detached with a razor blade or a scalpel. Care was taken to preserve the axons
protruding from the ganglion. With the naked eye, there appear to be four such axons,
approximately evenly spaced around the circumference of the circular surface of the ganglion.
However, one of these axons is thicker than the other three and is, in fact, two closely positioned
axons. This "double" axon can be used as a marker for positioning the ganglion .
The unused Aplysia tissue was placed in a designated biohazard bag and sealed for
disposal. The bag was placed in a biohazard waste box for proper removal by the University of
Florida Office of Environmental Health and Safety. All gloves, syringes, razors, etc. used in the
procedure were also disposed as biohazard waste in compliance with the NIH/CDC guidelines,
the State of Florida Administrative Code 64E-6, and restrictions of the Alachua County landfill.
Microelectrode Array Setup
Recording from the Aplysia abdominal ganglion employed a "3-D" microelectrode array
from Ayanda Biosystems in Lausanne, Switzerland. These IVEAs are ideal for recording from
thick pieces of neural tissue, such as cortical slices, hippocampal slices, and ganglion cells for a
multitude of applications such as neuroplasticity, visual perception, and impedance studies. The
electrodes in the dish are shaped with a three-dimensional orientation to better penetrate the deep
layers of tissue. In Figure 2-1, the electrodes are magnified to show their raised, conic shape
which penetrates into deep tissue. The electrodes are aligned in an 8x8 grid (with the corners
removed) and are separated by 200 Epm. Each electrode is 50 Epm high with a diameter of 40 Epm
at the base which comes to a Eine point at the tip. The electrodes, tracks, and contact pads are all
composed of platinum, and the IVEA is insulated with SU-8.
As the spatial activity patterns of Aplysia abdominal ganglia are well documented, it is
useful to deliberately orient the neurons on the surface of the IVEA. While looking down at the
IVEA, the "Ayanda Biosystems" logo was oriented such that it was at the top of the face of the
IVEA and readable from left to right (the same orientation was used when resting the IVEA on
the amplifier base, as discussed below). Next, the ganglion was positioned in the center of the
array. Because the ganglion cells are surrounded by a nonsymmetrical, insulating sheath, the
ganglion was placed such that the thinner side of the sheath was facing the IVEA surface, as the
thicker side may decrease the quality of data acquisition. The thinner side can be identified
because it has a bright orange color, as opposed to the thicker side which has a brown
appearance. The ganglion was rotated such that the "double" axon is pointing towards the
bottom-right side of the IVEA surface (with the "Ayanda Biosystems" logo designated as the
top). In this orientation, the R15 neuron is positioned on the lower, left side of the IVEA. Once
the ganglion was positioned on the IVEA, it was viewed under a microscope and images were
taken using a Macintosh iSight digital camera (Figure 2-2). As seen in the Eigure, variations in
the orange color (lower right) are the different cell bodies contained within the abdominal
ganglion. One of the axons can be observed looping around the left hand side of the image
outside the electrode grid.
In order to improve the adhesion of the ganglion cells to the MEA, the MEA was treated
with polyethyleneimine (PEI). The treatment was performed under a sterilized fume hood. Using
a micropipette, 100 ELL of PEI was deposited on the surface of the MEA. This small quantity of
PEI was left on the MEA for approximately 30 minutes and then removed from the surface using
a Eine tipped vacuum. Care must be taken in this step not to damage the leads of the electrodes on
the bottom of the MEA. After treatment with PEI, the dish was further rinsed with ultra-purified
water, from a Millipore Simplicity 185 purifier, using the same procedure. However, the
procedure is repeated 10 times, such that 1 mL of water is used in the process.
For the maintenance of activity in the extracted neurons, the ganglion was placed in an
artificial seawater (ASW) bath in the MEA. The bath closely resembles the seawater conditions
in which the Aplysia usually live, allowing the neurons to remain physiologically active for up to
8 hours. The bath consisted of 26.8824 g/L NaCL, 0.7753 g/L KC1, 1 1.1815 g/L MgCl2, 1.221
g/L CaCl2, and 3.5745 g/L of HEPES to maintain a pH of 7.8.
During experiments, a fitted Teflon lid is placed around the MEA dish. The lid is
equipped with 4 electrodes, which are equally spaced around the circumference of the ring. Two
opposing electrodes were used to propagate the measuring sine wave from an electrical stimulus
generator. In order for these electrodes to successfully send and receive current, the ASW
solution was filled to the top of the lid, creating a conductive path for electricity. In order to
prevent leakage of the ASW between the dish and the lid, a fitted rubber O-ring was placed in
the intervening crevice.
An MEA1060-BC amplifier from Multi Channel Systems was used to record neural
activity. This amplifier was designed specifically for MEAs possessing a special base on which
to rest the MEA and a series of pins which make direct connect with the contact pads of the
MEA. The amplifier is connected to a data acquisition A/D card which collects the data and then
outputs it to a computer for processing. The sampling rate used was 25 k
0. 1-10 k
as a single 200 second data fie requires approximately 800 MB of hard drive space.
Experimental data was collected from the data acquisition card on a PowerMac dual G5
desktop. A software bundle called "MEA Bench," developed by Dr. Thomas DeMarse for Mac
OS X, was used to perform a number of powerful functions for data acquisition and analysis.
Most notably, for these experiments, the "scope" and "spike detector" functions proved to be the
most useful. The scope program serves as an on-screen oscilliscope. The data can be observed in
real time as it is collected, saved, and then replayed. All of the MEA channels are displayed
simultaneously, with a function to zoom in on any specific channels) or adjust the voltage and
time scales as appropriate. As seen in Figure 2-3, aside from the four corners, each window on
the oscilliscope represents one of the digital MEA channels. The four corners display analog
channels; most notably the top left corner outputs a square wave in phase with the inj ected
sinusoidal current. The icons at the top of the window allow the user to control time scale,
voltage scale, and to freeze and rewind the data. Figure 2-4 shows a magnified image of one of
the channels during an action potential (highlighted in red). The window shows 200 ms of data
with a voltage scale of 50 ELV on the vertical axis. The spike detector will be described in more
A sinusoidal measuring current was used to measure changes in the impedance of
neurons in the ganglion. This current was produced by an STG 1008 stimulator from Multi
Channel Systems. This stimulus generator is capable of producing sinusoidal, square,
monophasic, biphasic, and ramp waveforms with a voltage range of -8 to 8 V and a current range
of -0.8 to 0.8 mA. The output signal is controlled using MC_Stimulus, and included a software
package with which the user can set the stimulus parameters and then upload them to the
hardware. The stimulator was connected to the electrodes which were implanted in the Teflon
ring using standard alligator clips. Hence, the measuring waveform was propagated through the
ASW and the tissue. Figure 2-5 shows a schematic of this setup. The MEA is at the bottom of the
image, with the Teflon lid fitted around the dish. The electrodes are embedded within the ring
(left and right sides of the Eigure) such that the current can propagate from one side to the other.
The lid is filled with the ASW solution as shown, with the tissue resting at the bottom on the
A flow diagram of the setup of the recording apparatus in Figure 2-6 shows the stimulator
connected to the electrodes on the MEA lid to send a current into the dish, while the MEA
electrodes record voltages and send the information to the amplifier. When positioned on the
MEA in the ASW solution, the ganglion cells spontaneously fire. For this reason, none of the
recordings discussed here employed a stimulating current to elicit action potentials. Instead, all
inj ected currents used were sub-threshold. Each current can be described by its amplitude and
frequency, parameters which were controlled to obtain the most useful data.
A key factor of the measuring current is that it must propagate at a low frequency. At
high frequencies, the current is able to penetrate the intracellular space through the cell
membrane of an inactive neuron. Conversely, low frequency currents are trapped in the extra-
cellular space, unable to cross the membrane . Yet, during cellular activity, the conductivity
increases by 30-40 times , allowing either current to move into the cell. Hence, the low
frequency current demonstrates a higher measurable impedance change in the neuronal cell
membranes. However, at low frequencies, the current amplitude necessary to stimulate action
potentials is lower than for high frequency currents. Thus, 10 Hz currents were used in these
experiments to maximize the measurable impedance change, at amplitudes of 5, 10, 20, and 40
ELA to avoid stimulation.
In addition to inj ected current recordings, baseline recordings of spontaneous activity
were obtained for each specimen. These recordings are useful in observing the natural,
spontaneous activity of the neurons. Specifically, these recordings were used as a reference point
with which to determine if the inj ected current recordings were, indeed, sub-threshold.
Three specimen were used for separate experimental trials. Four separate trials were
performed for each specimen. First, 200 seconds of baseline activity was recorded. Then, 200
seconds of data was recorded for each of three current values. For Specimens 1 and 2, currents
of 10, 20, and 40 ELA were used, all at a frequency of 10 Hz. The same procedure was used for
Specimen 3, though an added trial was included using a current of 5 ELA at 10 Hz.
Least-Mean-Square Adaptive Filter
Before an analysis of impedance could be conducted it was necessary to Birst identify any
neural activity embedded in the 10 Hz sinusoidal wave so that the timing of these events could
be compared with any changes in impedance. Applying the typical spike detection method would
be useless given the large amplitude fluctuations produced by the 10 Hz sinusoidal current.
Hence, the waveform from the current source must be separated from those produced by the
action potentials of the abdominal ganglion. To accomplish this, a least-mean-square (LMS)
adaptive fi1ter was used to remove the 10 Hz sinusoidal currents  before spike detection
occurred. The LMS fi1ter is a simple tool which can compare an input waveform to a desired
waveform and then output the error between the two. For example, in these experiments, the
input waveform was a sinusoidal wave recorded during propagation through the neural tissue
(which included neural data and noise). The desired wave used, then, was the 10 Hz sinusoid that
was initially inj ected into the neural tissue (without the neural activity). The LMS algorithm then
outputted the error between these two waveforms which, ideally, is equivalent to the neural
activity and the noise which would have been observed during a baseline recording without a
sinusoidal carrier current. The top panel of Figure 2-7 shows a baseline recording with a
prominent action potential at 0.975 seconds. The middle panel shows an example of a recording
during application of the current. Potentials recorded from the tissue (including action potentials)
are visibly embedded on the wave, though the spike detector software is unable to detect this.
After filtration (bottom panel), the raw data is isolated from the sinusoidal wave, allowing for
detection of action potentials with the spike detector.
The mechanism of the LMS filter is characterized by two primary functions: filtration
and adaptation. During filtration, the input wave is compared to a desired waveform and then the
algorithm outputs the error between the two, as previously noted. In the adaptive process, the
algorithm determines the proper weights for the tap weight vector, wn. This algorithm is
described by the following equations:
p = filter order, EL= stepsize, n=1, e=error (1)
x(n) = [x(n).. .x(n-p)] (2)
e(n)= d(n)-wnx(n) (3)
wn+1= Wn + pE*E(n)x(n) (4)
where wn is initialized to zero, p = 10 taps, and the step size EL was set to 0.000000001 for slow
adaptation of the filter.
The process of LMS filtration was necessary in order to isolate the raw neural activity.
From this data, then, the spatio-temporal locations of actions potentials were determined using a
spike detection algorithm. The algorithm is built into the MEA Bench software which was also used
to record the data. The spike detector first calculated the standard deviation of 5 seconds of random
noise at the beginning of the data set. The value obtained was then used as a threshold by which to
determine the location of spikes. Any data value in the data set which exceeded five times the
standard deviation from the mean was deemed a spike. Information about each spike including
channel, time, height, and width was then stored to a file for further processing.
Moving Average Algorithm
In addition to action potentials, it was necessary to determine the location of impedance
shifts in the neural data. This was determined from the sinusoidal carrier wave with the embedded
neural data. Each injected sinusoidal current was characterized by a constant amplitude (either 5, 10,
20, or 40 ELA). As this current moved through the conductive tissue, the tissue acted as a resistor. By
Ohm's Law, the voltages recorded with the MEA are directly proportional to the impedance of
the tissue. Hence, the impedance values were obtained by a simple division of the measured
voltages by the current amplitude.
For analysis of the impedances, the sinusoidal signal was demodulated by multiplication
with a square wave synchronous with the inj ected current waveform (recorded on one of the
MEABench amplifier's analog auxilliary channels), and then estimating the resistance or
reactance amplitude using a moving average algorithm. The moving average algorithm calculated
the average resistance of the 10 Hz voltage data using a sliding window that was fitted to one full
period of the sinusoid, approximately 2500 samples. After each calculation, the window slides by
one sample and then calculates another average over one period of the wave. Hence, the final
moving average vector was the same length as the original voltage data vector and displayed the
average impedance of the neural tissue for each sample window.
Final analysis consisted of comparisons between the moving average data and the spike
detection data. The time and location of impedance fluctuations were evaluated against spike
detection data. Correlated occurrences of impedance changes and neural activity in the same
location are indicative of a relationship between the two.
Figure 2-1. Multichannel Systems 3-D microelectrode array A) Entire Array B) Magnified
image of the electrodes C) Further magnified image. This array consists of 60
electrodes separated by 200 Epm. Unlike the standard planar MEA, the electrodes on
these arrays consist of 50 Epm high cones which are designed to penetrate the tissue
from which data is recorded and, hence, improve the signal-to-noise.
Figure 2-2. Aplysia ganglion centered on the dish of a microelectrode array. Although it is
difficult to see, variations in the orange color (e.g., lower right) are the different cell
bodies contained within the abdominal ganglion. One of the axons can be seen
looping around the left hand side of the picture outside the electrode grid. (Image is
from a Nikon TS-100 confocal microscope at 200x magnification.)
Figure 2-3. Oscilliscope function of MEABench on Mac OS X. The image shows data collected
during injection of a 10 EtA/10 Hz sinusoidal current. Electrical activity from the 8x8
grid of electrodes is shown. Each window represents 200 ms of the output from one
channel. The vertical scale for each window is 50 ELV. The data from the four corners
are from auxiliary channels used to record the timing of the sinusoidal current source.
T-ime (m ls)
Figure 2-4. Spontaneous action potential (red box) recorded from abdominal ganglion cells and
displayed on the MEABench scope tool. The window shows 200 ms of data with a
voltage scale of 50 EtV vertically.
\ I ~lectrode
Tissue Samiple MEA
Figure 2-5. Schematic of the MEA setup. Stimulator is attached to the electrodes to propagate a
current from one electrode, through the tissue/ASW, to the other electrode.
~1~ 1I~r H7.
n 2n tallk
Figure 2-6. Flow diagram of the experimental apparatus. The stimulator inj ects currents across
the tissue in the MEA. Potentials recorded by the MEA electrodes are sent to the
Raw Sqland Deerd 10 Mr Carat
13 435 441 445 4.5 4.55 41
LMS Fnterd Skyal
13 4 35 44 445 45 4 55 41
Figure 2-7. Example of LMS filtration. A) Spontaneous neural activity from Aplysia abdominal
ganglion. B) Spontaneous neural activity during application of a 10 ELA/10 Hz
sinusoidal current. C) LMS filtration to remove the sinusoid.
In this experiment, the neural activity using MEA technology from the abdominal
ganglion of three subj ects was compared to changes in the resistance under four different current
levels, a 10 ELA, a 20 ELA, and a 40 ELA injected current. The third subject also included a 5 ELA
current recording. All propagated currents were characterized by a frequency of 10 Hz. For each
subj ect, several minutes of spontaneous activity were first recorded to characterize the activity
patterns before the application of any current source. Each of the currents was then separately
applied and recorded for several minutes.
Patterns of Activity
Measurements using a 60-channel MEA were successful for the purposes of displaying and
detecting action potentials from the abdominal ganglion neurons. As indicated in Table 1-1, the
expected results could be divided into three categories: regular activity, irregular activity, and
bursting. Moreoever, the frequencies of firing expected for these three categories were 0.5-1 Hz,
2 Hz, and 0.1 Hz, respectively.
In order to visually display patterns of activity, raster plots were created to show activity
on all channels over time. These are scatter plots which display time across the horizontal axis
and MEA channel number on the vertical axis. Each marker on the plot, then, represents the
occurrence of an action potential at a specific time on a specific channel. Hence, raster plots
display a spatiotemporal map of all activity measured by the MEA. Figures 3-1A, 3-2A, and 3-
3A show the spontaneous activity of the abdominal ganglion before current application from
subj ects one through three, respectively.
The most noticeable feature of the spontaneous activity for Subject 1 (cf. Figure 3-1) is
dense activity on channels 27-32. By locating these channels on the vertical axis and then
following to the right, one can observe a cluster of action potentials occurring on many of these
channels beginning at 2 seconds. Vertically aligned markers such as these on the raster plot
indicate simultaneous activity of multiple neurons, known as bursts. This pattern continues at an
average rate of every 12 seconds. Hence, the first Subj ect did demonstrate periodic bursting, with
a period of approximately 12 seconds (0.08 Hz) on channels 27-32. On these channels, few
spikes can be observed outside of the bursts, indicating very little activity occurred outside of the
bursts. Weaker and shorter bursts were also occasionally observed on channels 33-40, with a
frequency below 0.05 Hz. Another notable feature of Figure 3-1 was the activity of channel 22.
Spikes can be observed during almost every 2 seconds for much of the recording period. This
denotes regular activity observed on channel 22, with spiking at about 0.5 Hz. The rest of
channels exhibited sporadic activity.
The spontaneous activity recorded from Subject 2 (cf. Figure 3-2), shows strong activity on
channels 49-59 just before 10 seconds, 30 seconds, 50 seconds, etc., approximately every 20
seconds. From this, it is evident that Subj ect 2 neurons also fired in bursts approximately every
20 seconds (or at a frequency of 0.05 Hz) with little or no activity between bursts. Like channel
22 from Subject 1, channel 36 demonstrated higher frequency activity, around 0.5-1 Hz,
evidenced by spikes nearly every 1-2 seconds. Channel 53 behaved uniquely, bursting every 20
seconds in unison with channels 54-59, but also, remaining active between bursts at a frequency
of about 0.5-1 Hz. During the spontaneous recording there was only one burst seen on channels
0-35 that occurred during the first 20 seconds of recording.
The data from recording the spontaneous activity of Subj ect 3 (Figure 3-3) was similar
than Subjects 1 and 2. However, in this subject, large bands of high frequency bursting was
observed on channels 13-22. Hence, for this data set, the synchronized bursting occurred at a
much higher frequency (1 Hz) than Subjects 1 and 2 (0.05 Hz). Subject 3 also displayed the
typical, unsynchronized high-frequency (0.5-1 Hz) active channels (0, 2, 3, 5, 7, 8, 10, 12, 57,
58). The tissue of Subj ect 3 was significantly more active than all other Subj ects. Although there
was some evidence of slower bursting (see channels 3 5-40), activity from Subj ect 3 could be
characterized by more active channels, higher spike frequencies, and even larger amplitude
action potentials than other Subj ects.
Data collected with an applied current was filtered and used to create raster plots as well.
Patterns of activity during application of each current were in general consistent with those
observed during the preceding spontaneous recordings. Raster plots of neural activity during
current applications are shown in Figures 3-1B, 3-2B, 3-3B, and 3-3C.
For Subject 1, the pattern of activity during the 10 ELA current (Figure 3-1A) was similar to
that observed spontaneously (Figure 3-1B). Unlike the baseline activity, the rate of bursting
during the current was slower, decreasing from 0.08 Hz to 0.05 Hz (every 20 seconds). Bursts
can be observed on channels 27-32 at 0 seconds, 20 seconds, 40 seconds, and continue with a
period of 20 seconds.
Similarly, activity for Subj ect 2 was comparable in frequency and regularity to baseline
recordings (Figure 3-2B). The bursts of activity on channels 47 and above showed periodic
bursting during both spontaneous (Figure 3-2A) and current source recordings. During the
current source recording several bursts of activity on channels 0-35 were also recorded, though
this was observed only once during the initial spontaneous measure (comparison of Figure 3-2A
vs. Figure 3-2B for channels 0-35).
The data for Subj ect 3, shown in Figure 3-3, was unique among the three subj ects in that
activity occurred at much higher frequencies across more channels than the other two subj ects.
During the 5 and 10 ELA current source there are a few instances where bursting had occurred, for
example, at 75 seconds (Figure 3-3B) and 125 seconds (Figure 3-3C), which is similar to
patterns recorded during baseline activity. However, it appears that application of either the 5 or
10 ELA current source seemed to slow the rate of activity in terms of the number of channels that
were active and the rate of activity on those channels. This slow down is similar to that observed
for Subject 1, as described earlier.
Total Spikes Over Time
In order to quantitatively display the amount of activity over the entire dish for each trial,
the number of spikes during one-second time bins was collected and plotted. Figure 3-4 shows
the rate of activity over time as a histogram obtained from recordings using a 10 ELA/10 Hz
applied current on Subject 1 (Figure 3-4A) and Subject 2 (Figure 3-4B). In this histogram, the
number of spikes detected on the entire IVEA is plotted on the vertical axis and time on the
horizontal axis. The time axis is divided into interval bins of 1 second. Hence, each bin
represents the total number of spikes detected in a one second interval, essentially smoothing the
variability of the rate of activity over time.
For most of the bins for Subj ect 1, 20-30 total spikes were observed. However, beginning
at 1 second and repeating every 20 seconds, the number of total spikes increases sharply to 40
spikes or more. This means that the sum total of the spikes detected with the IVEA suddenly
increased at a frequency of 0.5 Hz. Similarly, Subject 2 also showed periodicity. For this trial,
most of the bins contained 30-40 spikes. Around 10 seconds, a large increase in the number of
spikes per bin can be observed, the largest of which contains 75 spikes. Like Subj ect 1, this
sudden increase repeats at intervals of approximately 20 seconds. These histograms reflect the
underlying burst events shown in the raster plots described earlier.
Figure 3-5 shows a histogram of spike rate over time for Subj ect 3 during the 5 ELA/5 Hz
(Figure 3-5A) and 10 ELA/10 Hz current (Figure 3-5B). In this subject, bursts were exhibited at
much high frequencies with or without the current compared to the other two subjects (near 1
Hz). It follows that histograms created for Subj ect 3 should display sudden increases in the total
number of spikes every second. However, since the bin size is only 1 second, this means that
every bar on the histogram corresponds to the location of an expected burst of activity. Hence, all
of the bars should be roughly equal in magnitude, without noticeable increases in activity. As
expected, the magnitudes of the bars remain relatively constant throughout the trial, with two
exceptions at 135 seconds and 170 seconds (Figure 3-5A) and 75 seconds (Figure 3-5B).
Moving Average Filter
The results of the moving average filter were used to display changes in resistance of the
ganglion over time. The algorithm was used to calculate moving average results for individual
channels, and then to average these results over multiple channels. The moving average trace is
simply a graph of resistance or reactance values (vertical axis) shown over time (horizontal axis).
A value of zero ohms on the vertical axis of a moving average trace is equivalent to the baseline
resistance of the neurons during inactivity. Thus extrema represent temporal locations of sudden
shifts in the resistance from baseline. Since such resistance shifts are the main focus of our
experiments, these extrema are similarly the main focus of analysis of the moving average traces.
Specifically, consistent spatiotemporal correlations of these extrema and spike activity would
show evidence of a relationship between these two phenomena.
The resistance using the moving average method is shown in Figures 3-6 through 3-7
during the 10 ELA current for Subjects 1 and 2, respectively. The data for Subject 3 under the 5
and 10 ELA currents is shown in Figures 3-8 and 3-9, respectively. In each figure, three moving
average traces were produced. Panel A of each Eigure shows the resistance values calculated by
averaging together the moving average traces of all 60 channels of the IVEA. Panel B shows
resistance values averaged over only a band of active channels (generally channels on which
bursting was observed). Panel C of each Eigure shows values calculated by averaging together
the resistance traces of a cluster of inactive channels for comparison. Specifically, the active and
inactive traces were qualitatively compared for features such as the number and locations of
extrema. A direct subtraction of the two traces proved impractical due to the significantly
varying magnitudes of the resistance of the tissue at different spatial locations.
For Subject 1 in Figure 3-6, averaging over all channels during the 10 ELA current resulted
in a large resistance peak just before 1 10 seconds. Twenty seconds later, a series of significant
maxima and minima were observed. Large peaks were also observed for Subj ect 2 and Subj ect 3
in panel A of Figures 3-7 through 3-9. Analysis over the active channels or inactive channels
showed peaks for only one of the three subjects (Subject 3). For subject 1 and Subject 2, there
were occasional peaks in the data which were not apparent across each panel. In contrast, Subj ect
3 shows a clear peak at 95 seconds during the 5 ELA current source and 180 seconds during the 10
ELA current source which appears in each of the three panels. Of the three subj ects only the
subj ect which showed the highest rate of activity across the maj ority of channels showed any
Correlating Activity and Resistance
A maj or function of these experiments was to identify correlations between resistance
changes and action potentials in the neural tissue. In order to accomplish this, data from each
raster plot and histogram was compared to the corresponding moving average trace. For Subj ects
1 and 2, spatiotemporal locations of bursts were compared directly to the resistance traces. First
the bursts were located using the raster plot. These locations were confirmed with comparisons
to the histogram, which showed peaks at the temporal location of each burst. Then, the resistance
trace at the same instant of time was investigated for maxima and minima. The presence of
extrema on the resistance trace of bursting channels occurring simultaneously with the bursts on
those channels would indicate a correlation between resistance and activity.
This analysis is shown in Figure 3-10 from Subject 1 and Figure 3-11 for Subject 2. In
the Eigure, vertical line segments are used to designate the location of bursts. Following the
segment to the moving average trace from the active channels (Figures 3-10C and 3-11C) allows
for observation of the existence of extrema. In the Eigure, green segments are used to mark bursts
in which a simultaneous moving average extremum is present, while red segments are used to
designate bursts in which no such correlation could be observed. As both Eigures show, most
bursts did not correspond to an extremum, though a few bursts did show a correlation. The
moving average trace was further scrutinized for the presence of observable extrema which did
not correspond with a burst. Two such maxima were observed in from Subj ect 1 (Figure 3-12) at
125 and 175 seconds, and from Subject 2 at 5 and 40 seconds. Each is indicated in the Eigure by a
red circle. For Subj ect 3, this analysis was also performed, though it is not shown in the Eigure.
Since activity was persistent with a frequency near 1 Hz, such a visual representation would not
In general, for all trials, strong activity was sometimes accompanied by extrema in the
resistance trace, though usually were not. Beginning instead by finding resistance shifts and then
comparing to the raster plots and histograms showed similarly mixed results. While some
correlations were observed, most moving average extrema were present without a corresponding
increase in firing rate or neural activity.
Correlations between resistance shifts and spike activity could also potentially be
observed using frequency analyses. The periodicity of neuronal firing in the ganglion cells of
each subj ect has been shown using raster plots and histograms. If a correlation exists between
activity and resistance, it may be indicated by corresponding periodic deviations in the resistance
trace. For example, bursts were observed in Subject 1 every 20 seconds (0.5 Hz) on channels 27-
32. If neural activity can be indicated by resistance changes, then the resistance traces of these
channels should also have some component (decrease, increase, minimum, maximum, etc.)
which appears every 20 seconds. The presence of such a feature can be isolated using one of
several techniques for frequency analysis.
To evaluate frequency dependent factors, a Fourier transform was performed on each
moving average trace over clusters of active channels. The Fourier transform was used to convert
the data to the frequency domain such that the multiple frequency components of the wave could
be directly observed. This resulting data was then displayed on a power spectrum with each
frequency value across the horizontal axis and the corresponding power at each frequency on the
vertical axis. Prominent frequency components were then compared to the bursting frequencies.
Only frequency values between 0 and 1.5 Hz were considered in this analysis since the observed
frequencies of bursting activity did not exceed this range.
For Subj ect 1, bursting was observed with a frequency of 0.05 Hz over channels 27-32.
Fourier analysis of the same channels (Figure 3-12) shows a moderately sized peak at this
frequency. The strongest frequency component for this trial was observed at 0.5 Hz. Subj ect 2
exhibited bursting activity at the same frequency. For this subj ect, two moderately sized peaks
are shown around 0.05 Hz (Figure 3-13). However, the graph shows a much stronger component
around 0. 16 Hz. For Subject 3, periodic firing occurred at frequencies between 0.5-1.0 Hz.
Fourier analysis for each current value (Figure 3-14 and 3-15) shows only small to moderately
sized peaks in this frequency range. The largest components observed were at smaller
frequencies near 0.2 Hz.
Loannelve, I Ime or spires
nanne VB IIn fa le
JU U IU U IlU lG IS U lb U U (III
I Irne ls)
Figure -1. Raser plot for spntaneo s atvs.ity of Subjet1shwnteCanladTieo
each~ ~ ~ ~ ~~~~~" spk o )Bsln ciiyadB ppiaino 0P/0H urn.Ec
bUlu mrker deinae a deece acinptnil ust r dni da
siulaeosfiinsofmltpe hanls hee a b bsredpeidialyo
chnes2-2wthafeunynar01H.Drn curn apition patrso
aciiywr iia osotnosatvt bevdwtottecretatog
bustn frqunc di pert eres o00 z
na n v .IIrne or .pil~es
I Im~e ISl
renannel vs. Ilme or 3pIKe
U U t iU IUU I I 9U I I liiU UI.
r-I Im is]=
Fiue32 atrposfrSbjc uigA aeieatvtyadB plcto f1 A1
Hz ~~ curn.Hrbrtn a eosre tafeunyo .5H ncanl 9
59 hne 6as xiissgiiatsotnosatvt ohwti n usd
eac bust Busigwssilosre ncanes4-9a rqec f00 z
In ithsai l thrwaaloeiecofocsoabussootecan lsd ig
th plcto ftecret ore(U. hnes02)
cnannel vs. lrime or .pines
I Ire gal
nnnei vs. Ilme or 3p re
t- _~ ----1---~_1 -I---~=~- -- -;I--= ---- -=1--=- =~-~~.~~-'~1 ~-~I- ~-
DU U IU UU I bU U~U I bU
remainedi activ near 1r Hz.
:-plKes Prequency r-er ::econa
U LU LtU OU OU IUU IdU
I Ira elsj
IltU IOU IOU LUU
:-plKes rrequency I-*r ::.cona
U EU TU DU OU IUU
I Irna el s
IEU ITU IDU IOU EUU
Figure 3-4. Spike frequency over time (spikes in each second of data) for A) Subject 1 during
application of the 10 ELA/10 Hz current and B) Subj ect 2 during the 10 ELA/10 Hz
current. The almost periodic peaks seen in these figures correspond to the
spontaneous bursts that occurred regularly in these animals.
-.pies 1-requency r-er -.econa
U JU LtU DU UU l
I Imf e [Sj
IUU l~U IRU IOU
-.plK8 1-requency I-er --econa
Figure 3-5. Spike frequency over time (spikes in each second of data) for Subject 3 during A)
application of the 5 ELA/10 Hz current and B) the 10 ELA/10 Hz current For Subj ect 3,
peaks in spike rate are less apparent since activity in this animal consisted almost
entirely of rapid spike activity (cf. Figure 3-3 and 3-7).
Ui u U 9U DU U;U IUU ICU 19U IBU IOIU
I trne 18}
Figure 3-6. Resistance traces from Subject 1 during 10 ELA/10 Hz current calculated for A) all
channels, B) active channels, and C) inactive channels. Deviations in the traces
indicate shifts in the measured resistance.
Moving A~vera ge, rrl Lnannels
LU 4u bu iu mu ILu 14u Ibu mIu
I Ime (s]
Moving Hvera~ge, Acarve Lnannels
U LU SU DU UU IUU ILU 191
I Ire ls)
v ra~n e nacovell I n~nese
U IOU IUU
M OVHn g AVera g e, AII L anin OIs
U LU 9U bU UU IUU ILU ISU IbU IUU
x iu'' Moving Average, Acarve ULnannels
I Ime Isl
X IU MVn erg 10CI LianS
Figure 3-7. Resistance traces from Subject 2 during 10 ELA/10 Hz current calculated for A) all
channels, B) active channels, and C) inactive channels.
U LU 90 bU ;CIU IUU 120 19U 100 IBIU
MOMn g AVerag AicrIVe L.nannel3
U LU 9U DU UU IUU ILU IMU IDU IOIU
M oung wverage, mnacavle Lnannels
M Oung A~Vert ge, AI Liannels
"~'' '' ~I11'1'R~'1'11'''''~ '"''1 ~''~~ r'''''" r'1"'1' 'il ''j~'l
L L I 1
9U bU UtU IUU ILU 19U IbU IBiU
Figure 3-8. Resistance traces from Subject 3 during 5 ELA/10 Hz current calculated for A) all
channels, B) active channels, and C) inactive channels. In this animal, a large
deviation can be seen at approximately 95 seconds which occurs on all channels, even
the channels which had not shown any spontaneous activity in the absence of current.
Moung A~Vera.ge, p-91 nannels
Zu 4u ou oiu luu I~u I4u lou lou
I lme Is]
Movin g A~vera g e, cmo~ve L nann els
2U 4U bU UU IUU 12U 19U IbU IBU
I lme Is]
Moving A~vera~ge, Inacave Lnannels
2U 4U bU UU IUU 12U 19U IbU IBU
I lme Isl
Figure 3-9. Resistance traces from Subject 3 during 10 ELA/10 Hz current calculated for A) all
channels, B) active channels, and C) inactive channels.
PFIN U I 'T ] 1- *n
| | |
Lnbnnel VS. Ilme ~r spiKes
f ~ ;I~
|m In Sl
JU 4U bU UU IUU
I Im e I.s.
I U 14U IbU IBU ZUU
Figure 3-10. Comparative analysis for Subject 1 during application of 10 ELA/10 Hz current
between A) raster plot, B) histogram, and C) moving average trace for active
channels. Overlaid lines indicate areas of bursting. Green lines designate a possible
correlation while red indicate no apparent correlation. Red circles on the panel C
indicate two very large maxima which were not simultaneous with bursting.
UI dU 91 br J t U U 19 IU t U
~U 4U JU iU IU I IU 1.I'U ItiU ItiU rUU
plK$ rrequen: y r-er::.e~na
LI ZU 4U tIU iUl IU J IZL I' U ItiU li-iU rUU
I I lffl [SJ
~$ .rf* *
++ i *++ i
ZU 4U tiU tU IUU
I U 19U ItiU ItiU JUU
Figure 3-11. Comparative analysis for Subject 2 during application of 10 EtA/10 Hz current
between A) raster plot, B) histogram, and C) moving average trace for active
channels. Overlaid lines indicate areas of bursting. Green lines designate a possible
correlation while red indicate no apparent correlation. Red circles on the bottom panel
indicate two very large maxima which were not simultaneous with bursting.
Ln6~nnelVS. Ilme ~r spiKes
~~ ~~ r k* d ~J r
~m*+ +* *
rr + + +* ~
+ + ++ ++ ++
f f~ +t*l* r .~ ~ ~: ~ ~+
1-011118 l f8118100
U U1 U.2 U..; U.9t U.b U. U. U. U~~ U.Y
Figure 3-12. Fourier power spectrum of the moving average trace for Subject 1 during
application of a 10 EtA/10 Hz current
I OUnef I ran810ffA
U U.1 U.2 U.J U.9t U.b U.b U.? U.0 U.Y
Figure 3-13. Fourier power spectrum of the moving average trace for Subject 2 during
application of a 10 EPA/10 Hz current
r Ouflef l8I asF Off
UU U.b I I.b
Figure 3-14. Fourier power spectrum of the moving average trace for Subject 3 during
application of a 5 PcA/5 Hz current
rourier I ranerorrn
I cbI .
Fiue3-5 orirpwr pcru fte oigavrg taefr ujc 3drn
aplcto fa10P/0H urn
Patterns of Activity and Raster Plots
Using our 64-channel 1VEA, I was able to successfully record action potentials from the
Aplysia abdominal ganglion. Moreoever, results showed firing patterns which were reasonably
consistent with those reported by Novak and Wheeler . However, every type of activity was
not observed for every tissue sample used. Instead, slow periodic bursting (0.05-0.08 Hz) was
observed in Subj ects 1 and 2, while faster bursts (on the order of 1 Hz) were observed in Subj ect
3. For all trials, activity was generally observed in a cluster of 5-20 adj acent channels for every
data set. A few of the remaining channels (fewer than 10) also exhibited activity, each of which
was generally unsynchronized with the rest of the channels, and high frequency (~1 Hz). The
remaining channels were characterized by sporadic, irregular activity.
The varying patterns of activity observed were likely due to differences in the spatial
orientation of the ganglion on the IVEA. For each trial, the intention was to position the ganglion
such that measurements from the region near R15 would be recorded directly. This proved to be
a difficult task as the external features of the ganglion are sometimes not easy to distinguish after
dissection, even under a microscope. Based on the previous studies and our results, it is apparent
that this placement was successful in Subj ects 1 and 2, but not Subj ect 3. The areas near R15
were those which periodically burst every 10-20 seconds. Though this firing rate is slower than
that reported by Novak and Wheeler, this is probably due to variations between animals, each of
which demonstrates slightly unique activity patterns. For Subj ect 3, the rapid firing activity was
indicative of poor placement of the ganglion. The 0. 5-1 Hz frequency of firing was typical of the
white rostral cells, as reported by Novak and Wheeler. However, the bursting observed is in
contradiction with Novak and Wheeler' s classification of regular firing. This discrepancy may be
explained by a unique step in our methodology. For this subj ect only, extra care was taken to
remove the insulating sheath which surrounds the ganglion cells in order to increase spike
amplitude. Since action potentials were so large, they may have been measured effectively
enough by distant channels for detection by the spike detector algorhythm. Hence, one action
potential may have appeared on multiple channels, creating the appearance of bursting activity.
This is also significant in explaining the high number of active channels for this Subj ect. Hence,
it is likely that a smaller fraction of the ganglion was firing regularly at 0.5-1 Hz than the results
from Subject 3 would indicate.
Application of a 10 ELA current consistently provided the most manageable and useful
results as this current value provided efficient and accurate filtration of the sinusoidal current.
Even so, changes were noted in the firing patterns of the tissue as compared to spontaneous
activity. Specifically, the frequency of bursting in Subject I was slower (0.05 Hz) during
application of the current versus that observed with no current (0.08 Hz). This decrease in
frequency could indicate that the current values used were large enough to affect the behavior of
the tissue. Conversely, the changed behavior could be a byproduct of lost data during filtration
and spike detection. This latter explanation is also applicable to the sparseness of the raster plots
obtained from applied current trials as compared to baseline recordings.
As the 20 and 40 ELA currents provided a superior signal to noise ratio in the moving
average traces, the relative success of the 10 ELA data was unexpected. However, the STG 1008
stimulator which was used to provide the current often created problems in the signal beyond 10
ELA. On these trials, artifacts from the stimulator were observed at every transition between one
period of the wave to the next. Every data set using a current higher than 10 ELA was marred by
this effect. To resolve the problem, these artifacts were later removed from the data set before
filtration. However, this process would also remove any action potentials which occurred
simultaneously with the artifacts. The artifacts sometimes spanned up to 200 samples, or 0.008
seconds and were removed from every period of the wave, every 0. 1 seconds at 10 Hz. Hence,
up to 8% of the data was removed in order to eliminate these artifacts. While useful information
was still available in the remaining portion of each data set, these recordings were not considered
in the Einal results and analyses here. The 5 and 10 ELA data, however, consistently proved useful,
could be efficiently filtered, and was devoid of artifacts.
Resistance Changes and Activity
The purpose of the moving average filtration was to quantitatively and visually represent
changes in the resistance of the neural tissue over time. Ideally, these deviations would show
some consistent relationship to the results of spike detection as represented visually using raster
plots. For a single neuron, deviations from this baseline value were expected to show temporal
correlation with spiking activity.
Results from the moving average fi1ter from all trials provided modest evidence that
resistance changes occur in conjunction with neural activity. Correlations between activity and
deviations in the moving average trace were successfully found, though not consistently.
Qualitatively, changes in the shape of the moving average trace appeared similarly random, with
little dependence on activity. Traces from active channels and inactive channels were hardly
distinguishable in Subject 1, both remaining reasonably flat with only a few small extrema. In
contrast, the active and inactive traces calculated for Subj ect 2 showed little correlation to one
another. For Subject 3 at both current amplitudes, active and inactive channels showed
deviations away from baseline, sometimes correlating together and sometimes not. Hence, it
could not be concluded that the presence of maxima and minima in each trace was dependent on
the amount of activity present on the channel from which the trace was calculated.
Another method used to isolate relationships between activity and resistance was the
comparison of histogram data to the resistance traces. The histograms were useful as a second
means by which to visualize instances of high neural activity in addition to raster plots. The
resistance traces obtained over all of the active channels of a trial were compared to the sum total
activity of these channels over time. Instances at which the amount of spikes in a second over the
whole dish drastically increased should create a change in the resistance of all of the active
channels, which would be apparent in the resistance trace. Lining up the time axes of the
histograms with their corresponding resistance traces and comparing showed the same results as
comparisons to the raster plots. Correlations existed, though not on a consistent basis. This is
especially obvious using data from Subjects 1 and 2. In these data sets, the histograms show
sharp increases in activity which correspond to the bursting of the R15 region of the ganglion
once every 20 seconds. In contrast, the active channel resistance traces for these trials remained
rather flat. Further, what extrema were present did not visibly show any correlating periodicity.
Large peaks present on the moving average trace of Subject 1 at 125 and 175 seconds were not
accompanied by a burst. Large peaks observed in Subj ect 2 at 5 and 40 seconds come too early
and too late, respectively, to correlate with observed bursts. For Subject 3 at both current levels,
the result is reversed; the histogram is relatively flat due to the constant firing of the cells, though
extrema were present in the resistance traces. Hence, a consistent relationship between histogram
data and the resistance traces also could not be confirmed.
Frequency and periodicity were also an indicator as to whether correlations existed
between resistance changes and neural activity. If neural activity is truly characterized by
changes in neuronal resistance, then the resistance traces should show periodic character with the
frequency of the associated periodic neuronal activity. For Subjects 1 and 2, active channels
burst with a frequency of 0.05 Hz, or every 20 seconds during application of current. Hence, if a
correlation does exist between activity and resistance changes, moving average traces on these
channels should display periodic deviations with a period of 20 seconds. Similarly, for Subj ect 3,
the spike frequency of 1 Hz should correspond with deviations in the resistance trace.
Fourier transforms were used to help affirm the absence of correlations between
frequency and activity. For Subject 1 and 2, a peak was expected at or near 0.05 Hz on the
Fourier power spectrum. For Subj ect 3, a prominent peak was expected at 1 Hz. For each trial,
peaks expected on the Fourier power spectrum sometimes existed, but were never prominent.
The very presence of peaks at the expected frequencies may provide some possible indication of
the correct periodicity, but the results remain inconclusive due to the weakness of the amplitude.
It should be noted that any high frequency components were lost during the moving
average process. Because the moving average window is 0.1 seconds wide, equivalent to one
period of the 10 Hz sinusoidal current, the average magnitude of the current within one window
is always zero. Moreover, any waveform with a frequency that is a multiple of 10 Hz, such as 60
Hz, will similarly have an average magnitude of zero. Hence, these components were cancelled
out of the frequency band. Furthermore, calculation of the response characteristics of the moving
average filter showed that the gain of the filter decreases with increasing frequency for values
which are not multiples of 10 Hz. At high frequencies, on the order of k
drops to less than 0.003. Thus, components at these frequencies, such as action potentials (which
have a duration of approximately 1 ms, equivalent to 1 k
Because of this, action potentials themselves do not influence the values of the moving average
trace. Instead, only the indirect effect of the action potentials on the resistance should be
manifested in the trace.
To corroborate all of this evidence, a moving average trace was created for an individual
channel, 36 from Subject 2 using a 10 ELA current (Figure 4-1). This channel showed unique
behavior as evidenced by the associated raster plot for this data set (Figure 4-2). The
distinguishing feature of this channel was that it was active for the first 150 seconds of recording,
but then becomes noticeably less active in the last 50 seconds. Hence, if a correlation exists, the
moving average trace should show some type of change in its behavior for 150 seconds, and then
transition to a different type of behavior in the final portion of the recording. Results from the
moving average trace in the figure showed no such transition. Channel 4 of the 5 ELA recording
of Subject 3 was also inactive and then active for long, separate stretches of the trial. Like
channel 36 in Subj ect 2, no distinct transition in the resistance trace corresponded to the activity
measured on the channel.
While it is apparent that deviations in the moving average trace showed only little
correlation to activity in the neurons, the question remains as to where these deviations originate.
The answer is likely due to random conditions which alter the initial 1VEA recording. While
watching the data move across the oscilliscope software during recordings, the signal can be
observed deviating away from its properly centered alignment. The most obvious example of this
is random noise. However, noise would doubtfully create deviations on the scale of those seen in
some of the moving average traces.
Poor contact between the ganglion tissue and the IVEA could lead to erroneous data. One
of the most difficult aspects of the setup of these experiments was the placement of the neural
tissue on the IVEA dish. Even with the use of PEI, the tissue often adheres poorly to the dish.
Hence, during recordings, the tissue may have been partially floating in the ASW, allowing the
tissue to move in the dish during recordings. In this situation, the movement of the tissue may
not have significantly affected action potential recordings, but may have had a drastic effect on
the moving average calculations.
Wherever they originate from, there is some evidence that deviations occurring during
recordings corrupt the moving average data. Referring back to Subj ect 1, the resistance trace
across all channels displayed a very large peak just before 1 10 seconds, and a number of other
extrema thereafter. However, the traces from the active channels and the inactive channels do not
show these extrema and, in fact, are relatively flat. Review of the original raw data showed that
channels 58 and 59 began to visibly drift during the initial peak before 110 seconds, and
continued to drift through the remainder of the data set (Figure 4-3). Figure 4-4 shows resistance
traces of channels 58-59, and the average of channels 0-57, and the overall average across all
channels. Panel A (all channels) and panel C (channels 58-59) are quite similar, both showing
deviations which temporally correspond to the drifting. Once the drifting channels (58-59) were
removed and the average was recalculated across the rest of the channels (0-57), the large peaks
disappeared (Figure 4-4C).
This finding leads to two conclusions. First, deviations in the initial recordings caused by
external factors can have significant effects on the end results and analysis. Second, though
multiple channels were averaged together, one or two corrupt channels can have drastic effects
on the overall averages. The large problem, however, is not these drifting channels. Channels
like 58 and 59 in Subject 1 are obviously corrupt and can be removed from the moving average
calculation. Instead, lower magnitude drifting is not so easy to identify and remove. This could
be the cause of some of the observed extrema in the resistance traces, and a difficult obstacle to
avoid in future work.
Another example of this behavior occurred in the results from Subj ect 2. In Figure 3-6,
moving average traces were shown for this data set, including averages over the active channels
(54-59). However, activity was actually observed on channels 51-59, though channels 51-53
were removed from the collection due to significant drifting. Figure 4-5 shows the original trace
from Channels 54-59, and the new trace, calculated over channels 51-59. The new trace is much
different, now exhibiting a peak at 115 seconds. Hence, while the specific set of channels used to
calculate resistance traces exerts an impact on the shape and characteristics of the trace, the
amount of activity on the channels chosen is not the causal variable. Moreover, the new trace
proves that channels 51-53 are large contributors to the overall shape of the trace calculated over
all channels, showing helping to further affirm that drifting channels cause large deviations in the
Microelectrode Arrays and Future Work
Data collection using a 64-channel 3-D IVEA proved successful in obtaining
spatiotemporal information about neuron spiking activity. Action potentials were clearly visible
to the human observer, and easily detected using 1VEABench OSX software. Furthermore,
patterns of activity observed using these IVEAs were consistent with those reported in past
literature. However, some obstacles existed in using this modality for identifying changes in
resistance. There appear to be problems within the experimental setup that would have to be
corrected in order to use IVEAs as a viable means for measuring such resistance changes. Care
must be taken to prevent any perturbations in the system which could affect data acquisition.
Most importantly, it is essential that strong contact is made and maintained between the tissue
and the IVEA for the duration of all recordings.
Aside from externally driven shifts in the moving average trace, even those shifts that were
apparently real were difficult to analyze. Local extrema were successfully observed in each trial.
However, the small magnitudes of some maxima and minima left doubt as to whether these were
true extrema or merely the results of noise and random fluctuations. This could be remedied with
an improved signal to noise ratio. Unfortunately, the stimulator produced artifacts during
application of currents larger than 10 ELA. If this problem can be solved, the signal to noise ratio
can be improved, which would increase the size of observable changes in resistance. The
resistance is calculated by dividing the voltage by the current as defined by Ohm's Law.
Specifically, the moving average trace is calculated for the voltages recorded by the IVEA and
then divided by the magnitude of the inj ected current to obtain the resistance. Hence, increasing
the magnitude of the current will cause a proportional increase in the recorded voltages for a
given value of the resistance. Thus, small deviations caused by noise, perturbation, movement,
etc. may prove insignificant relative to the large magnitude of the voltage changes (and thus, the
resistance changes). Further, though large voltages would overwhelm the action potentials, spike
detection will be unaffected since the sinusoidal current is filtered our of the signal first.
Significant improvement may require currents even larger than 40 ELA, which would stimulate
the tissue. Though I have studied only spontaneous activity here, stimulation may prove more
useful. Indeed, considering the randomness of the results observed here, the predictability and
control gained with evoked activity may prove superior.
Functional Magnetic Resonance Electrical Impedance Tomography
A new technique, called functional magnetic resonance electrical resistance tomography
(fMLREIT), would be a promising new method for imaging neural activity in vivo. This method
uses an MRI scanner to collect maps of the magnetic flux density within the tissue, a result of the
flow of electric current. Hence, this data reflects changes in resistance. During neuronal activity,
the increase in the flow of current into the intracellular space will be apparent using an fMLREIT
scan due to the accompanying change in conductivity of the cell membrane. Thus, by imaging
the changes in resistance, the scan is indirectly imaging neural activity. Most importantly, if
successful, this technique would image neural activity with the spatial resolution of an MRI, but
with improved temporal resolution .
MoMn g A~vera.g e
U JU 'tU DU iiU IUU IZU IRU IOU ItiU
I lil e Isl
Figure 4-1. Moving average trace for Channel 36 of Subj ect 2 during application of 10 LA/1 0
Hz current. This channel was active for the first 150 seconds of the recording, and
ZU LtU DU aU IUU IdU 19~U lbU IBU eUU
I Ina i (3
Figure 4-2. Raster plot for Subject 2 during application of 10 EtA/10 Hz current. Channel 36
(indicated by a red box) is active for the first 150 seconds and then becomes relatively
LILHanOS VS. 1100 Of spin, te
1 er u ** ** *r r
* r ec ++ + ~ + e *r *
r + C *+ <* r r
es' * e *+
*c 4 *b *i e4 kl es r+
e + +
so 4 *
II *I *
* ** M *
~ I ~ ,
r rr r
r r+r rr +, r *
r r r r r *
r +1 1
++4- 1 ++
t *r*+ r r~ +r*
1 1111 1 1C11 1 1111 I
.1. ~ r~II ~~CI+ I~r~rtll 1. f I
Fiue43 mg ffu hnnl rmteoclicp urn ujc eodnswt
Fiue -.Img f channels ono th etsie 8an 9 ega tsisop durift while the otecrchnnes remined
U 2U 4U bU UU IUU 12U 19U IbU IBU
I lme Is]
x lu' Moving Avera~ge, Lnannels sties
u ~U 4u bu ou luu I u 14u Ibu ~Uu
I lme Is]
S x lu
Moung A~vera~ge, c411unannels
2U 4U bU UU IUU 12U 19U IbU IBU
I lme Isl
M ovn g A~vera~ge, L nann els u-5,
Figure 4-4. Resistance traces for Subject 1 during application of current for A) all channels, B)
channels 0-57, and C) only channels 58-59.
Movin g A~vera g e, Mit L nann els
A j u
d U EU 4U bU UU IUU IZU 19U IbU IBU
I Ime [sl
SX U' MOuing A~verfge, McIcVe Unannels DR-D'-
ZU 4U bU UU IUU IZU 19U
I Ime [s]
MOnng AVera~ge, HcnVe Unannel3 b I-bs
JU 4U bU UU IUU IZU 19U IbU IBU
I Ime [sJ
Figure 4-5. Resistance traces for Subj ect 2 during application of current for A) all
active channels 54-59, and C) active channels 51-59.
The results obtained in the experiments described have shown that MEAs are well suited
for recording the spatiotemporal characteristics of neural activity from Aplysia abdominal
ganglion. Unfortunately, the results could not be used to confirm a consistent relationship
between resistance changes and activity in neural tissue, whether that neural activity is always
accompanied by resistance changes, or vice versa. This relationship has been observed and
reported in previous literature. Trouble in the experimental setup and analysis techniques caused
the inability to prove the desired relationship. Poor signal to noise ratios led to difficulty in
identifying resistance changes. Under these conditions, the results may have been susceptible to
external factors which created corrupted results. The results here also have not led to a
conclusion that a relationship between resistance changes and activity in neural tissue absolutely
do not exist. With the proper improvements to the experimental setup, the topic still merits
LIST OF REFERENCES
 M. Pascual, E. Pozas, M. J. Barallobre, M. Tessier-Lavigne, and E. Soriano,
"Coordinated functions of Netrin-1 and Class 3 Semaphorins in the guidance of
reciprocal septohippocampal connections," M~olecular and Cellular Neuroscience, vol.
26, pp. 24-33, 2003.
 E. Stein and M. Tessier-Lavigne, "Hierarchichal organization of guidance receptors:
silencing of Netrin attraction by Slit through a Robo/DCC receptor complex," Science,
vol. 291, pp. 1928-1939, 2001.
 A. C. Yu, E. Hertz, and L. Hertz, "Alterations in uptake and release rates for GABA,
glutamate,and glutamine during biochemical maturation of highly purified cultures of
cerebral cortical neurons, a GABAergic preparation," J.Neurochem, vol. 42, pp. 951-960,
 R. Albin, A. Young, and J. Penney, "The functional anatomy of basal ganglia disorders,"
Trends Neurosci, vol. 12, pp. 366-75, 1989.
 A. Nobin and A. Bj orklund, "Topography of the monoamine neuron systems in the
human brain as revealed in fetuses," ActaPhysiool Scand Suppl., vol. 3 88, pp. 1-40,
 P. Somogyi, A. J. Hodgson, and A. D. Smith, "An approach to tracing neuron networks in
the cerebral cortex and basal ganglia. Combination of Golgi staining, retrograde transport
of horseradish peroxidase and anterograde degeneration of synaptic boutons in the same
material," Neuroscience, vol. 4, pp. 1805-1852, 1979.
 A. C. Guyton and J. E. Hall, Textbook of2\~edicalPhysiology, vol. 11. Atlanta, Georgia:
W.B. Saunders Co, 2005.
 G. Buzsaki, Rhythms of the Brain. USA: Oxford University Press, 2006.
 C. Croona, M. Kihlgren, S. Lundberg, O. Eeg-Olofsson, and K. E. Eeg-Olofsson,
"Neuropsychological findings in children with benign childhood epilepsy with
centrotemporal spikes," Developmental Medicine and Child Neurology, vol. 41, pp. 8 13 -
 M. Modarreszadeh and R. N. Schmidt, "Wireless, 32-channel, EEG and epilepsy
monitoring system," presented at 19th Annual International Conference of the IEEE,
Chicago, IL, USA, 1997.
[1l] J. D. R. Millan, F. Renkens, J. Mourino, and W. Gerstner, "Noninvasive brain-actuated
control of a mobile robot by human EEG, IEEE Trans BiomedEng, vol. 51, pp. 1026-
 M. A. L. Nicolelis, "Actions from thoughts," in Nature, vol. 409, 2001, pp. 403-407.
 M. A. L. Nicolelis and J. K. Chapin, "Controlling robots with the mind," Scientific
American, vol. 287, 2002, pp. 46-53.
 C. G. Gross, C. E. Rocha-Miranda, and D. B. Bender, "Visual properties of neurons in
inferotemporal cortex of the Macaque," JNeurophysiol, vol. 35, pp. 96-111, 1972.
 C. S. Herrmann, "Human EEG responses to 1-100 Hz flicker: resonance phenomena in
visual cortex and their potential correlation to cognitive phenomena, Exp Brain Res, vol.
137, pp. 345-353, 2001.
 D. J. Dijk, D. G. Beersma, and R. H. van den Hoofdakker, "All night spectral analysis of
EEG sleep in young adult and middle-aged male subjects," NeurobiolAging, vol. 10, pp.
 E. V. Evarts, "Activity of neurons in visual cortex of the cat during sleep with low
voltage fast eeg activity," J. Neurophys., vol. 25, pp. 812-816, 1962.
 S. Ogawa, T. Lee, A. Kay, and D. Tank, "Brain magnetic resonance imaging with
contrast dependent on blood oxygenation," Proc NatlAcadSci USA, vol. 87, pp. 9868-
 N. K. Logothetis, "The underpinnings of the BOLD functional magnetic resonance
imaging signal," JNeurosci, vol. 23, pp. 3963-3971, 2003.
 M. E. Raichle, "Functional brain imaging and human brain function," JNeurosci, vol. 23,
pp. 3959-3962, 2003.
 J. D. Greene, R. B. Sommerville, L. E. Nystrom, J. M. Darley, and J. D. Cohen, "An
fMRI investigation of emotional engagement in moral judgment, Science, vol. 293, pp.
 K. L. Phan, T. Wager, S. F. Taylor, and I. Liberzon, "Functional neuroanatomy- of
emotion: A meta-analysis of emotion activation studies in PET and FMRI," Neuroimage,
vol. 16, pp. 331-348, 2002.
 R. M. Birn, P. A. Bandettini, R. W. Cox, and R. Shaker, "Event-related fMRI of tasks
involving brief motion," Interscience, vol. 7, pp. 106-1 14, 1999.
 A. D. Frieiderici, M. Meyer, and D. Y. von Cramon, "Auditory language comprehension:
An event-related fMRI study on the processing of syntactic and lexical information, "
Brain and Language, vol. 74, pp. 289-300, 2000.
 M~icroelectrode Array (M~EA) User Manual. Reutlingen, Germany: Multichannel Systems
MCS GmbH, 2005.
 S. Chawla, P. Vanhoutte, F. J. Arnold, C. L. Huang, and H. Bading, "Neuronal activity-
dependent nucleocytoplasmic shuttling of HDAC4 and HDAC5," JNeurochem, vol. 85,
pp. 151-159, 2003.
 M. A. Corner, J. van Pelt, P. S. Wolters, R. E. Baker, and R. H. Nuytinck, "Physiological
effects of sustained blockade of excitatory synaptic transmission on spontaneously active
developing neuronal networks--an inquiry into the reciprocal linkage between intrinsic
biorhythms and neuroplasticity in early ontogeny," Neurosci Biobehav Rev, vol. 26, pp.
 F. Hofmann, "Functional re-establishment of the perforant pathway in organotypic co-
cultures on microelectrode arrays," Brain Res, vol. 1017, pp. 184-196, 2004.
 F. H. Arnold, F. Hofmann, P. Bengtson, M. Wittmann, P. Vanhoutte, and H. Bading,
"Microelectrode array recordings of cultured hippocampal networks reveal a
simple model for transcription and protein synthesis-dependent plasticity," JPhysiol, vol.
564, pp. 3-19, 2005.
 F. Lante, M. C. de Jesus Ferreira, J. Guiramand, M. Recasens, and M. Vignes, "Low-
frequency stimulation induces a new form of LTP, metabotropic glutamate (mGlu(5))
receptor- and PKA-dependent, in the CAl area of the rat hippocampus," Hippocampus,
 M. J. Berry, J. L. Puchalla, E. Schneidman, and R. A. Harris, "Redundancy in the
population code of the retina," Neuron, vol. 46, pp. 493-504, 2005.
 A. H. Chen, Y. Zhou, H. Q. Gong, and P. J. Liang, "Luminance adaptation increased the
contrast sensitivity of retinal ganglion cells," Neuroreport, vol. 16, pp. 371-375, 2005.
 D. J. Bakkum, Shkolnik, A. C., Ben-Ary, G., Gamblen, P., DeMarse, T. B. and Potter, S.
M., "Removing some 'A' from AI: embodied cultured networks. Embodied artificial
intelligence," EmbodiedArtificiallIntelligence, F. lida, Steels, L. and Pfeifer, R., Ed.:
Springer-Verlag, In Press.
 T. B. DeMarse, D. A. Wagenaar, A. W. Blau, and S. M. Potter, "The neurally controlled
animat: Biological brains acting with simulated bodies," Autonom Robots, vol. 11, pp.
 K. Banach, M. D. Halbach, P. Hu, J. Hescheler, and U. Egert, "Development of electrical
activity in cardiac myocyte aggregates derived from mouse embryonic stem cells," Am J
Physiol Heart Circ Physiol, vol. 284, pp. H12114-23, 2003.
 J. Hescheler, Y. Duan, M. Tang, and H. Linag, "Investigation on spontaneous electrical
activity of murine embryonic heart using microelectrode arrays," Acta Physiologica
Sinica, vol. 58, pp. 65-70, 2006.
 J. Malmivuo and R. Plonsey, Bioelectromagnetism Principles and Applications of
Bioelectric andBiomagnetic Fields. New York, NY: Oxford University Press, 1995.
 A. Adler, R. Amyot, R. Guardo, J. H. T. Bates, and Y. Berthiaume, "Monitoring changes
in lung air and liquid volumes with electrical impedance tomography," JAppl Physiol,
vol. 83, pp. 1762-1767, 1997.
 I. Frerichs, J. Hinz, P. Herrmann, G. Weisser, G. Hahn, T. Dudykevych, M. Quintel, and
G. Hellige, "Detection of local lung air content by electrical impedance tomography
compared with electron beam CT," JApplPhysiol, vol. 93, pp. 660-666, 2002.
 V. Cherepenin, "A 3D electrical impedance tomography (EIT) system for breast cancer
detection," Physiol2~ea~s, vol. 22, pp. 9-18, 2001.
 Y. Zou and Z. Guo, "A review of electrical impedance techniques for breast cancer
detection," M~edEng Phys., vol. 25, pp. 79-90, 2003.
 J. Keener and J. Sneyd, Mathematical Physiology. New York, NY: Springer-Verlag,
 K. S. Cole and H. J. Curtis, "Electric impedance of the squid giant axon during activity,"
The Journal of General Physiology, vol. 22, pp. 649-670, 1939.
 A. L. Hodgkin and A. F. Huxley, "Currents carried by sodium and potassium ions
through the membrane of the giant axon of Loligo," JPhysiol, vol. 1 16, pp. 449-472,
 A. L. Hodgkin and A. F. Huxley, "Measurement of current-voltage relations in the
membrane of the giant axon of Loligo," JPhysiol, vol. 1 16, pp. 424-448, 1952.
 A. L. Hodgkin and A. F. Huxley, "A quantitative description of ion currents and its
applications to conduction and excitation in nerve," JPhysiol, vol. 116, pp. 500-544,
 A. L. Hodgkin and A. F. Huxley, "The components of membrane conductance in the
giant axon of Loligo," JPhysiol, vol. 116, pp. 473-496, 1952.
 A. L. Hodgkin and A. F. Huxley, "The dual effect of membrane potential on sodium
conductance in the giant axon of Loligo," JPhysiol, vol. 116, pp. 497-506, 1952.
 K. Boone, D. Barber, and B. Brown, "Imaging with electricity: report of the European
Concerted Action on Impedance Tomography,"1 M edEng Technol, vol. 21, pp. 201-32,
[ 50] G. Grimnes and O. G. Martinsen, Bioimped anced~~~~~dddd~~~~ an2dBioelectricity Basics. London, UK:
Academic Press, 2000.
[ 51] D. Holder, ed., Clinical and Physiological Applications of Electrical Impedanced~~~~dddd~~~~ddd
Tomoguraphy. London, UK: UCL Press, 1993.
[ 52] D. Holder, Electrical Impedanced~~~~dddd~~~~ddd Tomoguraphy ofBrain Function. London, UK: IOP
 J. G. Webster, ed., Electrical Impedanced~~~~dddd~~~~ddd Tomoguraphy. Bristol, UK: Adam Hilgar, 1990.
 W. R. Adey, R. T. Kado, and J. Didio, "Impedance measurements in brain tissue of
animals using microvolt signals," Experimental Neurology, vol. 5, pp. 47-66, 1962.
 A. Van Harreveld and J. P. Schade, "Changes in the electrical conductivity of cerebral
cortex during seizure activity," Experimental Neurology, vol. 5, pp. 383-400, 1962.
 N. A. Aladj olova, Slow electrical processes in the brain," Progress In Brain Research,
vol. 7, pp. 155-237, 1964.
 J. B. Ranck, Jr., Specific impedance of rabbit cerebral cortex," Experimental Neurology,
vol. 7, pp. 144-52, 1963.
 S. E. Fox, "Membrane potential and impedance changes in hippocampal pyramidal cells
during theta rhythm," Experimental Brain Research, vol. 77, pp. 283-294, 1989.
 S. E. Fox and C. Y. Chan, "Location of membrane conductance changes by analysis of
the input impedance of neurons. II. Implementation," Journal ofNeurophysiology, vol.
54, pp. 1594-1606, 1985.
 L. Geddes and L. E. Baker, "The specific resistance of biological material-a compendium
of data for the biomedical engineer and physiologist, M~edBiol Eng, vol. 5, pp. 271-93,
 A. T. Tidswell, A. Gibson, R. H. Bayford, and D. S. Holder, "Electrical impedance
tomography of human brain activity with a two-dimensional ring of scalp electrodes,"
Physiol2~ea~s, vol. 22, pp. 167-75, 2001.
 K. A. Klivington and R. Galambos, "Rapid resistance shifts in cat cortex during click-
evoked responses," Journal ofNeurophysiology, vol. 31, pp. 565-573, 1968.
 W. T. Frazier, E. R. Kandel, I. Kupfermann, R. Waziri, and R. E. Coggeshall,
"Morphological and functional properties of identified neurons in the abdominal ganglion
of Aplysia californica, Journal ofNeurophysiology, vol. 30, 1967.
 J. L. Novak and B. C. Wheeler, "Recording from the Aplysia abdominal ganglion with a
planar microelectrode array," IEEE Transactions on Biomedical Engineering, vol. BME-
33, pp. 196-202, 1986.
 S. Haykin, Adaptive Filter Theory (4th Edition). Upper Saddle River, NJ: Prentice Hall,
 R. J. Sadleir, S. C. Grant, X. Silver, S. U. Zhang, E. J. Woo, S. Y. Lee, T. S. Kim, S. H.
Oh, B. I. Lee, and J. K. Seo, "Magnetic resonance electrical impedance tomography
(MREIT) at 11 tesla field strength: preliminary experimental study," presented at Proc.
5th BEM and 5th NFSI, Minneapolis, Minnesota, 2005.
Hany Elmariah was born on August 10, 1982 in St. Petersburg, Florida. Two years later,
his family relocated to Panama City, Florida where he spent the rest of his childhood. Hany
attended Bay High School, where he fostered strong interests in Biology and Physics. As an
undergraduate at Duke University, Hany earned a Bachelor of Science in Physics, while also
completing the prerequisites to pursue a graduate education in the health sciences. This
background proved useful for his next academic pursuit, a Master of Science in Biomedical
Engineering at the University of Florida. Hany plans to attend medical school in the fall of 2007.