Adaptive Sensory Filtering in the Cerebellar-like

WESLEYAN UNIVERSITY
ADAPTIVE SENSORY FILTERING
IN THE CEREBELLAR-LIKE MECHANOSENSORY NUCLEUS
OF THE HINDBRAIN IN RAJA ERINACEA
by
Krista Eva Perks
A thesis submitted in partial fulfillment of the
Requirements for the degree of
Master of Arts
Wesleyan University
Middletown, Connecticut
May 2007
Abstract
In all cerebellar-like structures, a multipolar principle neuron has spiny apical
dendrites synapsing a superficial parallel fiber layer but also receives a second
functionally and anatomically distinct input. The cerebellar-like first-order nuclei in the
electrosensory system of teleost and non-teleost fishes filter reafference and other
predictable stimuli out of the system. The mechanism by which this occurs can be
described by the adaptive filter model proposed by Montgomery and Bodznick (1994).
A parallel fiber information array and primary sensory afferents converge onto the
principal neuron, with coincident activation inducing anti-Hebbian plasticity at the
molecular layer parallel fiber synapses. This present study addresses whether the same
adaptive filter model can predict the operation of the cerebellar-like medial nucleus in
the lateral line system of the little skate, Raja erinacea. Single-unit, extracellular recording
techniques were used in vivo to study the principal neurons and measure learned
compensations for an external mechanosensory stimulus made predictable by timelocking it to the natural ventilation cycle. That changes in AEN activity patterns were
predicted by the adaptive filter model suggest that the mechanism is not just a
specialization of the electrosense, and it may be underlying stimulus response
conditioning in the mechanosensory system as well. By similar logic, the adaptive filter
model may generalize to other cerebellar-like structures, and even to the cerebellum
itself.
ii
Table of Contents
Introduction………………………………………………………………………..1
Cerebellar-like Structures
2
Functional Plasticity
4
Adaptive Sensory Filtering in the Electrosense
5
The Mechanosensory Lateral Line
7
Structure and Function
11
Materials and Methods……………………………………………………………..13
Animals and Surgery
13
Stimulation
14
Recording Methods
14
Physiological Localization of Medial Nucleus
15
Single-Unit Identification
16
Coupling Experiments
16
Data Collection and Analysis
18
Morphology
21
Results………………………………………………………………………….. …22
Spontaneous Activity
22
Coupling to Ventilation
26
Control
33
Fictive Ventilation: Motor Command
35
Discussion…………………………………………………………………………..39
Generalizing the Adaptive Filter Model to a Cerebellar-like Organization 39
Limitations of the Methods Used in the Current Study
41
Functional Applications for Adaptive Filtering in the Mechanosense
43
Mode of Lateral Line Stimulus
47
Self-Organizing Systems
50
The Cerebellum
52
Literature Cited……………………………………………………………………..60
Acknowledgements…………………………………………………………………65
iii
Table of Figures
Introduction
1. Characteristics of Cerebellar-like Structures
2. Pathway of Information Through the Medial Nucleus
3. Spatial Distribution of the Canal System
Materials and Methods
4. Experimental Set-up
5. Recording Methods and Analysis Terminology
Results
6. Spontaneous Firing Patterns
7. Mixed Inter-Stimulus Interval Patterns
8. Coupling of a Mechanosensory Stimulus to Ventilation
9. Apparent Negative Image Effect
10. Possible Negative Image to an Inhibitory Stimulus Interval
11. Comparison to a Free-Running Control
12. Sufficiency of Fictive Ventilation to Induce a Cancellation Signal
13. Necessity of Coupling to the Motor Command
Discussion
14. Modulation of Output from Cerebellar-like Structures and the Cerebellum
iv
Introduction
A molecular-layer parallel fiber system is the major defining feature of a group of
primary sensory nuclei termed “cerebellar-like” structures. All cerebellar-like structures
exhibit plasticity at their molecular layer synapses, which is induced by activation of the
post-synaptic principal neuron by its other afferent input. In the cerebellar-like
electrosensory lateral-line lobe (ELL) of mormyrid electric fish, the plasticity at the
molecular layer parallel fiber synapses is the basis for a “modifiable efference copy” of
the electric organ motor command, which cancels out predictable sensory reafference
(self-stimulation) in the principal neuron from its other input, the electroreceptors (Bell,
1982). In another electrosensory cerebellar-like structure, the dorsal nucleus of
elasmobranchs, the function of plasticity in the parallel fiber system has been modeled as
an adaptive filter mechanism, which also serves to eliminate sensory reafference and
other predictable electrosensory inputs from the periphery (Montgomery and Bodznick,
1994). The primary electrosensory nuclei have thus been shown to specialize in the
feedforward predictive elimination of sensory self-stimulation. Preliminary observations
in scorpion fish have suggested that the mechanosensory lateral line system may also
have a comparable mechanism for filtering out predictable sensory afference at the level
of the hindbrain (Montgomery and Bodznick, 1994). The mechanosensory lateral line of
fishes and some amphibians is a distinct, yet very similar system to the electrosense, and
the primary sensory nucleus, the medial nucleus in the hindbrain, is another cerebellarlike structure. My hypothesis is that, based on structural similarities, the adaptive filter
model defined for the dorsal nucleus predicts an equivalent mechanism for the
1
elimination of predictable mechanosensory stimuli in the medial nucleus. Results
presented here indicate that the adaptive sensory filter is not just a specialization of the
electrosense, thus supporting the hypothesis that the adaptive filter model may describe a
fundamental dynamic operation of all cerebellar-like structures.
Cerebellar-like Structures
Cerebellar-like structures, the dorsal and medial nuclei, the ELL (electrosensory
lateral-line lobe), and the DCN (dorsal cochlear nucleus) of many mammals, are all
primary sensory nuclei, receiving information ascending via the lateral line or the octaval
nerves (Northcutt, 1980). The medial nucleus processes mechanosensory stimuli in the
hindbrain for the lateral line system of aquatic anamniotes, the dorsal nucleus processes
electrosensory stimuli in the hindbrain of electroreceptive non-teleost fishes and some
amphibians, the ELL is used for the independently-evolved electrosensory systems in
teleost fish (Bell et al., 1997), and the DCN processes spectral cues for the auditory
system in mammals (May, 2000). All of these structures are closely related by parallels in
sensory processing and by common developmental and evolutionary origins (Devor,
2000; Johnston, 1902; Larsell, 1967; Montgomery et al., 1995; Paul et al., 1977; Roberts
and Ryan, 1971). They are all classified as cerebellar-like based on the most striking
similarity among them, a unique laminar organization in which large, multipolar principal
neurons are lined up beneath a superficial parallel fiber layer from which they receive a
very large number of synaptic inputs.
The principal neurons, which in most cases are the sole output of the nucleus,
receive two functionally distinct inputs (Figure 1). In the deeper layers, primary sensory
2
Figure 1: Characteristics of Cerebellar-Like Structures.
A laminar organization featuring at least three major components is shared by all
cerebellar-like structures. The heterogeneous parallel fiber system in the superficial
molecular layer makes massive plastic synapses onto the principal neuron. Afferent
input provides information from the sensory endorgans and synapses in a deeper layer of
the nucleus. Only a few afferent axons synapse on each principal neuron to drive its
activity, so these synapses are likely stronger than the parallel fiber synapses, which seem
to serve a more modulatory function. The adaptive filter model predicts that, by antiHebbian plasticity, a cancellation signal at the parallel fiber inputs is generated in
response to sensory afference coincident with molecular layer activity.
An
environmental stimulus creates a sensory image in the afferent input (time X). Since
there is a subset of parallel fibers active at this same time, the excitatory afferent input
induces the parallel fiber synapse to be depressed. But at times Y and Z parallel fiber
activity is not coincident with AEN excitation. So those synapses are not depressed.
According to the adaptive filter model, they are actually potentiated.
3
Ascending Axon
Superficial
Molecular
Layer
Parallel
Fibers
Granule
Cells
Principal Cell
Layer
Deeper
Layers
X YZ
Primary
Afferent
afferents synapse directly (in most cases) on the smooth basal dendrites. Spiny apical
dendrites branch up into the molecular layer, which contains a parallel fiber system
arising from a population of granule cells that is continuous with the cerebellar granular
region. The granule cells receive a variety of sensory and non-sensory signals from
multiple sources. There are orders of magnitude more parallel fiber inputs than there are
afferent inputs.
Functional Synaptic Plasticity
All cerebellar-like structures feature characteristic synaptic plasticity at the
molecular layer synapses (Figure 1). Plasticity is ubiquitous in the central nervous
system, and in many instances is thought to be correlated with behavioral measures of
learning and memory. However, the exact function of plasticity within most major
structures is difficult to unravel because the identity and functional significance of the
information encoded in the inputs and outputs of the circuit are either completely
unknown or ill-defined. The cerebellar-like primary sensory nuclei are a useful model
system for studying the functional role of plasticity in behavior, first, because the
physiological properties of the neurons and the information encoded in their inputs are
discernable, and second, because the responses and interactions of the individual
neurons can be directly related to their well-defined functional roles in conditioning
sensory information to enhance the detection and identification of biologically relevant
signals from the environment.
4
Adaptive Sensory Filtering in the Electrosense
Reafference is a sensory response to stimuli generated by an animal’s behaviors.
To the sensory receptors, reafferent stimuli are physiologically indistinguishable from
other biologically relevant environmental signals. Sensory reafference is generally
stronger than other signal sources and can mask their detection (Montgomery and
Bodznick, 1999), so it must be filtered out at a low level of processing. In the
electrosensory system of the skate, active processing occurs in the first-order cerebellarlike dorsal nucleus of the hindbrain to filter out reafferent noise. The principal neuron
in the dorsal nucleus in skates and other elasmobranchs is called an AEN (Ascending
Efferent Neuron). These are the sole output neurons of the dorsal nucleus, so they are
central to the information processing of the system. AENs must selectively eliminate
reafference while staying near firing threshold and responsive to novel external stimuli.
An adaptive filter model has been proposed as the mechanism for this function
(Montgomery and Bodznick, 1994), and it relies on anti-Hebbian plasticity at the parallel
fiber—AEN synapses (Bodznick et al., 1999). When the AEN is depolarized to spiking,
parallel fiber synapses active at the same time are depressed (Bertetto, 2007); the
depression is reversed when the parallel fiber is active in the absence of AEN
depolarization. Primary electroreceptor afferents are the normal physiological source of
depolarization/hyperpolarization to the AEN, and transmit temporally specific sensory
patterns from the environment. The molecular layer parallel fiber inputs to the AEN
comprise an information array of a variety of signals marking the skate’s own behaviors
(proprioceptive information, corollary motor command discharges, and other
descending sensory information).
5
When a response pattern in the AEN induced by the primary afferents is
predicted by activity in the parallel fiber system, a composite is molded from the
information array available at the molecular layer synapses to nullify the predicted
response pattern. The parallel fiber composite is called a cancellation signal and, by
simple summation, the cancellation signal specifically suppresses the effect on the AEN
of that predicted electrosensory afference. Only signals not predicted by parallel fiber
activity remain un-mirrored (un-opposed) and are transmitted beyond the primary
electrosensory nucleus. In summary, plasticity at this one synapse allows the rest of the
circuit to use the adaptation of a cancellation signal as a feed-forward control device for
the system’s output.
That the same basic adaptive filter mechanism is present in the evolutionarily
distinct electrosensory systems of apteronotids (Bastian, 1999), mormyrids (Bell et al.,
1999), and elasmobranchs (Bodznick et al., 1999), suggests it could be a specialization of
the electrosense. However, the cerebellar-like organization of the dorsal nucleus seems
to be the underlying key to its function. The existence of this shared and highly
specialized anatomy may be an inert consequence of evolutionary relatedness or it may
underlie a shared functional role in all of the cerebellar-like systems. These explanations
are not mutually exclusive. Can the functional role of the parallel fiber system in the
adaptive filter mechanism of the electrosensory dorsal nucleus and in the generation of a
cancellation signal to the EOD of the ELL be generalized to other cerebellar-like
structures serving different sensory systems?
6
The Mechanosensory Lateral Line
The little skate, Raja erinacea, is a primarily benthic sea creature that has both a
mechanosensory lateral line and an electrosensory system. These senses are structurally
very similar yet are functionally independent and segregated in the periphery and in the
central nervous system (Bodznick and Northcutt, 1980). Unlike electroreceptors, lateral
line mechanoreceptors receive inhibitory efferent innervation (Roberts and Russell, 1972;
Russell, 1971). This could reduce reafference by turning off the receptors during
behavior, but the efferents appear only to be activated during very vigorous behaviors
that threaten to overdrive receptors (Bodznick, 1989; Roberts and Russell, 1972). It
appears that a more dynamic filter would be useful under most circumstances.
The medial nucleus is located below the dorsal nucleus in the lateral medulla in skates,
and is the cerebellar-like primary sensory nucleus for the mechanosensory lateral line
system. As in all cerebellar-like structures, the mechanosensory principal cell (also an
AEN, as in the dorsal nucleus) is multipolar, receiving two inputs and then transmitting
the information to the second-order medial mesencephalic nucleus (MMN) in the
midbrain (Figure 2). Mechanically-evoked responses of the peripheral receptor cells are
transmitted via the anterior and posterior lateral line nerves and make monosynaptic
connections to the AENs. The medial nucleus is bound by a molecular layer parallel
fiber system, which originates from multimodal granule cells of the lateral granular (LG)
region (Schmidt and Bodznick, 1987). These three major neural components of the
circuit (primary afferents, AENs, and parallel fibers) are segregated into the three main
layers of the nucleus. An adaptive filter in the cerebellar-like medial nucleus may be used
7
Figure 2: Pathway of Information Through the Medial Nucleus
The mechanosensory AENs recorded in this dissertation are located in the
medial nucleus (MON). Afferents from the lateral line receptors enter through the
anterior and posterior lateral line nerves (ALLN and PLLN) and terminate on the basal
dendrites of the AEN (ascending efferent neuron). The lateral granular region (LG)
supplies the parallel fiber system for the molecular layer of the medial nucleus. These
fibers synapse on apical dendrites of the AENs throughout the rostro-caudal length of
the nucleus. The AEN then projects out of the medial nucleus, crosses the midline, and
terminates in the Medial Mesencephalic Nucleus (MMN), which is the second order
sensory nucleus for the lateral line system. (CC: corpus cerebellum, DON: dorsal
nucleus, DGR: dorsal granular ridge, ML: molecular layer) (bottom right, location within
the skate brain of the transverse sections diagrammed above)
8
B
MMN
A
CC
C
CC
DGR
LG
DON
ML
ML
MON
ALLN
Primary
Afferent
Hair Cell
CAUDAL
A
C
B
ROSTRAL
in the mechanosensory system to eliminate predictable sensory information from the
output of the nucleus.
Like the electrosense, the lateral line system has its own set of sub-dermal canals
distributed throughout the body; these are filled with fluid and are open to the sea water
at pores in the skin surface so that the receptors are coupled to the environment (Figure
3). The neuromast is the sensory organ of the mechanosense and is located either
subdermally in the canal system or superficially as a free-standing receptor organ on the
skin. The afferent lateral line nerves supplying the AENs in the medial nucleus innervate
the neuromasts (Boord and Northcutt, 1982; Puzdrowski RL, 1993). Each neuromast
contains hair cell receptors covered by a cupula so that shearing motion between the
fluid layer and the skin surface into which the hair cell is embedded
depolarizes/hyperpolarizes the membrane depending on the direction of fluid flow.
Near-field hydrodynamic stimuli (particle motions and pressure waves) are produced by
water currents or moving organisms (Bleckmann et al., 2003; Dusenbery, 1992; Sand,
1984), or by distortions in the hydrodynamic environment surrounding the moving
animal itself due to presence of obstacles in the near-field (Coombs S, 2003). Superficial
neuromasts encode velocity information while canal neuromasts encode acceleration of
the fluid surrounding the organism.
9
Figure 3: Spatial Distribution of the Canal System.
Illustration of the parallel spatial distribution of the electrosensory and
mechanosensory periphery. Both diagrams depict the left dorsal half of a bilaterally
symmetrical canal system in the lateral line (left) and the electrosensory (right) periphery
of Raja erinacea (After Bodznick and Schmidt 1984). The diagram of the lateral line
canals was obtained by methylene blue dye injection into the canals such that they could
be visualized and the specific location of the pores (marked by dots) could be identified.
Dotted lines indicating continuation of the canal is inferred by the presence of identified
pores and the assumption that the canals are bilaterally continuous.
10
Structure and Function
A principal tenant of biology is that structure underlies function. Although the
adaptive filter model was proposed to illustrate the mechanism for elimination of
electrosensory reafference in the first-order dorsal nucleus, my hypothesis is that the
model can be generalized to the function of all cerebellar-like structures and even the
cerebellum. The major prediction of this hypothesis tested and presented in this
dissertation is that activity of AENs in the medial nucleus will change specifically in
response to a mechanosensory lateral line stimulus coupled to ventilation in a freely
breathing skate.
According to the adaptive filter model, the parallel fiber system in the medial
nucleus provides an information array comprising neural markers for the animal’s own
behaviors, such as ventilation, to the apical AEN dendrites and those synapses follow
the rules of anti-Hebbian plasticity. So when a mechanosensory stimulus induces AEN
activity that is coincident with ventilation, a cancellation signal to the specific afferent
sensory pattern is generated. When the stimulus is then turned off, the cancellation
signal will cause a negative image of the sensory pattern in the spontaneous activity of
the AEN.
My results support this prediction. Coupling a mechanosensory stimulus to
either ventilation or fictive ventilation for 10 minutes was sufficient to induce a negative
image to the afferent sensory image. This was observed both as a decrease in response
to the stimulus over the 10-minute stimulation period and as a negative (mirror) image of
the induced activity pattern in the spontaneous activity of the AEN when the stimulus
11
was turned off. Further experiments are in progress to examine the probable functional
significance of this adaptive sensory filter in the mechanosensory system of the skate.
12
Materials and Methods
Animals and Surgery
Female skates, Raja erinacea, were collected in Long Island and Vineyard Sound.
In captivity, they were maintained at 10-14 degrees Celsius in seawater aquaria and fed
frozen whiting fillets.
Prior to experimentation, individual specimens were anesthetized by immersion
in 0.4% Benzocaine (Sigma, St. Louis, MO) and placed on ice during surgery. In
experiments using freely breathing skates, the surgical procedure began with removal of
the cranial cap to expose the brain, followed by de-cerebration by severing connections
to the forebrain. most of the spinal cord was destroyed to paralyze movement of the
fins and tail, while allowing the movements of ventilation. Often, it was also necessary
to cut the upper thoracic spinal nerves to prevent movement of the anterior part of the
expanded pectoral fins, which form the body disk in skates.
In some experiments, the skates were completely immobilized with intravenous
injection of 3mL (1.0 mg/kg) Pancuronium Bromide (Sigma, St. Loius, MO, stock
number P1918) in skate Ringer (1μg/1mL). The spinal cord was kept intact since the
drug was able to prevent all movement. In order to keep track of fictive ventilation
cycles with this preparation, an incision was made in the skin behind the spiracle to
expose the seventh cranial nerve. The motor root of this nerve was dissected from the
sensory root and transected as far laterally as possible so that a recording could be
obtained with a suction electrode. The massed discharge from this nerve is the motor
command for onset of exhalation, or fictive exhalation in a paralyzed skate.
13
After surgery, the skates were held in a seawater experimental tank on a
submerged plexiglass platform with a headstage immobilizing the cranium above the
water line. The skates were allowed to recover from anesthesia for at least one hour
before experimentation. The brain was kept flushed with skate Ringer, and the skates
were maintained by perfusion of seawater across the gills at a rate of 0.4L/min to ensure
an adequate supply of oxygen. This is especially critical for experiments in which the
skate was not freely breathing. The skates held in this way remained in good condition
and responsive to sensory stimuli for at least 2 days.
Stimulation
For almost all of the experiments, the external lateral line stimulus was a
depression of the skin surface within the receptive field of the AEN. Only units that had
dorsal receptive fields were used for experiments. A function generator was triggered at
the onset of each trial and caused a mechanical arm to press down once and then release
at a repetition rate of 0.3-1.0 per second.
Recording Methods
Multiunit recordings in the medial nucleus were obtained with blunted glass
micropipettes (approximately 8 μm tip diameter) filled with 4M NaCl. Extracellular
single unit recordings within the medial nucleus were obtained with indium-filled micro
pipettes (1-2μm tip diameter with a long shank), which are tipped with a 2-4 m
diameter gold ball and plated with platinum to drop the resistance between 3-7 M .
14
Extracellular single unit recordings in the anterior lateral line nerve were obtained with a
capillary glass micropipette filled with 4M NaCl and a resistance of 20-25 M .
The output of the electrode was amplified with a 3-3,000Hz bandwidth. This
raw analog signal was first viewed AC-coupled in real-time on an electron-beam
oscilloscope. It was then fed into a computer interface (Cambridge Electronic Design)
with a 500kHz 16-bit analog-digital converter under control of Spike2 software. Single
spikes were also distinguished by a voltage-level discriminator that converted them to
trigger pulses, which were also recorded in Spike2 as events that could be counted over
time.
To couple to the motor command for ventilation, the seventh cranial nerve root
was captured with a suction electrode. The raw, multiunit recording was amplified by
10,000 with a 3-3,000Hz bandwidth, and a trigger pulse was established through a
second voltage-level discriminator.
Physiological localization of medial nucleus
A concentric bipolar stimulating electrode was placed at the surface at the
midline of the contralateral superior colliculus and plunged approximately 2 mm to enter
the second order mechanosensory lateral line nucleus, the medial mesencephalic nucleus
(MMN), which is the projection site for all lateral line AEN axons. Current pulses (4
Volt, 0.5 mS), delivered via a stimulus isolation unit elicited a local field potential in the
medial nucleus. No (or minimal) field potential was elicited in the region of the dorsal
nucleus when the stimulating electrode was placed correctly in the MMN. The range in
depth through which a field potential was normally recorded was between 2,500-4,000
15
micrometers deep. The strongest field potential (with the shortest onset latency) was
usually recorded through a tract descending through 500 micrometers in the middle of
the total range over which the field potential was observed.
Single-unit Identification
Once, the range of the medial nucleus was established in each skate by recording
the induced field potential, single units were isolated in an extracellular recording with
indium electrodes. Within this range, mechanosensory ascending efferent neurons
(AENs) were identified by antidromic stimulation from the MMN. Antidromic spikes
occur with short latency (3-4 mS) after the shock to the midbrain and have a clearly
defined threshold (between 1V-5V). Mechanosensory AENs were unresponsive to an
electric field in the tank. This second criteria ensures that I have gone deep enough into
the brain and am not accidentally recording from electrosensory AENs in the dorsal
nucleus. If the cell’s receptive field is below the water line, then the cell must also
respond to natural hydrodynamic stimuli, even if that stimulus is not then used to test
the cell. This is done to ensure that the recording electrode has not gone into the octaval
nuclei that lie beneath the medial nucleus.
Coupling Experiments
In all experiments, the trigger pulse is delivered both to the function generator to
trigger a mechanical arm providing the mechanosensory stimulus and through an
interface program to be recorded with Spike2 software as a digital event. This event
marks time zero for each trial over which AEN activity is analyzed.
16
Figure 4: Experimental Set-up
Top: Cranial cap was removed to permit in vivo recordings of AENs in the medial
nucleus. Ventilatory movements were recorded via a force transducer placed over the
gill chamber, from which a voltage level discriminator generated a trigger pulse at the
onset of each exhalation. This trigger was then used to initiate the mechanosensory
stimulus, which depressed the skin surface within the receptive field of a recorded
mechanosensory AEN.
Bottom: The shaded region over the lateral lobes of the medulla marks the area
over which the medial nucleus was located (approximately 2-4mm deep). The recording
electrode was advanced from the surface down to the appropriate depth. (Diagram gratis
RG Northcutt)
17
A
AEN recording
from
medial nucleus
Raw
Ventilation
Force Transducer
4
Local
Mechanosensory
Stimulus
Trigger
Recording
in medial
nucleus
B
Rostral
Caudal
Ventilatory movements of the freely breathing skates were detected with an
isometric force transducer placed on the dorsal surface of the gills, and a trigger pulse at
the onset of inhalation was obtained via a level discriminator (Figure 4). In paralyzed
skates, a trigger was obtained from the fictive ventilation signal as described above.
Multiple triggers off of one cycle of fictive ventilation were prevented by a built-in
lockout period for the level discriminator and logic pulse generator. For the control
runs, a free-running stimulus was generated by a clock trigger that was not synchronized
with any endogenous activities or external cues.
Data Collection and Analysis (Refer to Figure 5)
Events corresponding to each action potential were recorded and first viewed as
a triggered raster plot. As illustrated in Figure 5, one trial is the time between two
triggers in a run. One run of the experiment comprises three periods: 1)the pre-stimulation
period is the baseline activity of the AEN, 2)in the coupling period the AEN is externally
stimulated by the local mechanosensory stimulus, 3)the post-stimulation period begins when
the mechanosensory stimulus is turned off. However, the trigger source to which the
stimulus had been coupled continued throughout. This trigger was also used to bin the
data in a 60-trial peri-stimulus histogram to visualize the activity profile of a single unit
during the period of single trials before, during, and after the coupling period.
18
Figure 5: Recording Methods and Analysis Terminology.
The top trace (A) depicts the spiking activity of a single AEN unit during the
period of the experiment in which activity is induced by a local mechanosensory
stimulus. That stimulus is triggered by the onset of exhalation (B: the raw signal from a
force transducer placed on the gill chamber). All spikes are recorded as events (dots)
and plotted in a raster (C). Each trigger marks the beginning (t = 0) of a trial and each
trial is the duration of one ventilation cycle.
An experiment is one run, which comprises three periods. The pre-stimulation
period is the baseline activity of the AEN.
In the 10 minute coupling period, a
mechanosensory stimulus is triggered by ventilation such that activity is induced in the
AEN for a consistent duration during each trial. This is defined as the stimulus interval
(bottom) and is applied to every trial throughout the run and used in the analysis of a
subtracted spike count for every trial. The post-stimulation period begins when the
stimulus is turned off.
19
Events = spikes
One Trial
During Coupling
Period
A
B
In.
One Ventilation
Cycle
Ex.
1s
t=0
1s
C
Pre-Stimulus
Period
600 s
Coupling
Period
Post-Stimulus
Period
Stimulus
Interval
One
Run
The stimulus interval is the time during each trial over which the sensory stimulus
induces uniform activation of the AEN. Hydrodynamic and mechanosensory stimuli are
often spatially and temporally complex. There was considerable variability in response
latency and pattern among AENs, presumably because of variations in the location of
the tactile stimulus in relation to the location of the neuromast in the canal. The units
that were used for analyses had a receptive field that was accessible (mostly dorsal) and
gave a relatively sustained response to the stimulus. Units with complex or transient
responses to the stimulus could not be used for testing because the resolution of my
statistical methods is too low.
The activity pattern induced in the AEN during coupling corresponds to a sensory
image. The negative image, then refers to an activity pattern that mirrors the sensory image.
A script was used to measure the spike rate in the stimulus and the extra-stimulus
intervals for every trial. A subtracted spike count, which is a calculation based on spike rates
in each interval and the length of each interval, was used as the final quantitative measure
of stimulus-specific changes in AEN activity. This number was obtained by normalizing
the number of spikes in the outside interval to the time inside the stimulus interval, and
then subtracting that normalized spike count from the actual spike count inside the
stimulus interval.
subtracted
spike
count
events in
=
stimulus
interval
events
-
outside
stimulus interval
x
duration
stimulus
extra- stimulus
interval
duration
20
The negative image is quantified by a change in the subtracted spike count after
coupling that opposes the change that occurred as a result of the applied stimulus during
coupling. A change in the sensory response was also quantified by a change in the
subtracted spike count at the end of the coupling period compared to the beginning.
Comparisons of statistically significant change in the subtracted spike count were made
using the Mann Whitney non-parametric one-tailed t-test between data sets of 60 trials
each (http://faculty.vassar.edu/lowry/utest.html). Since the data sets failed the
Kolmogorov-Smirnov test for normal distribution the parametric t-test was invalid
(http://www.physics.csbsju.edu/stats/KS-test.n.plot_form.html).
Morphology
Morphology of the dorsal lateral line canals was visualized by injecting methylene
blue dye into the canal and drawing the canals as viewed from the skin surface, marking
the individual pores accurately.
21
Results
Spontaneous Activity
I have encountered over 200 AENs in the medial nucleus. Of these, less than
100 could be detected by the level discriminator as events above the 60-cycle noise.
Only 57 units were cleanly recorded for at least 5 minutes before stimulation began such
that the average spike rate could be calculated for that period. The group mean average
spike rate was 2.16 spikes/s (+/- 2.24 SD). The range of rates was 0-12.1 spikes/s.
A wide range of baseline activities were observed, but with rare and minimal
modulation by ventilation. The inter-spike interval histograms (ISI) of these 57 AEN
units in the medial nucleus were analyzed (summated over 300 s for ISI <4 s, bin size
50ms) before any regular external mechanosensory stimulation was applied. These
revealed three major patterns of spontaneous firing: bursty, non-bursty, and minimallyspontaneous (Figure 6).
The 38 units with bursty spontaneous firing patterns had two features in
common (Figure 6A). First, the full 4 second range of inter-stimulus intervals was
represented. Second, there was a single largest peak, which was usually in the first 50ms
bin but in some units it was in the 50-100ms bin (Type A, Figure 6A). About 11 units
with bursty firing patterns had an additional characteristic feature of a second broad peak
much smaller than the 50ms peak (Type B, Figure 6A). This second peak was usually
around 400-1000ms and was preceded by a trough in the pattern from about100-500ms.
Non-bursty firing patterns (10 units) could also be classified by a comparable set
of features (Figure 6B). First, there was usually no inter-stimulus interval larger than 2
22
seconds. Second, there was a single broad peak with a smooth decline that usually had
its maximum around the 200-250ms bin. Variation in the breadth of this peak and its
maximum did occur. A few patterns (8 units) seemed to be a combination of both
bursty and non-bursty patterns (Figure 7).
Minimally-spontaneous units (1) were generally silent when first encountered but
after a period of sensory stimulation inter-stimulus intervals from 0.05s to at least 15 s
were sparsely represented, but with only one or two spikes per bin (Figure 6C).
23
Figure 6: Spontaneous Firing Patterns.
Representative inter-stimulus interval histograms of the three major spontaneous firing
pattern types encountered: Bursty (A), Non-Bursty (B), and Minimally-spontaneous (C).
Histogram width 0-4s (except where noted as 8s), bin size 0.05s. Numbers to the left of
each denote the maximum of the y-axis.
24
A
Bursty
Type A
90
SPIKES
160
25
INTERVAL
4 seconds
B
Non-Bursty
70
200
C
Minimally-Spontaneous
25
8 seconds
Type B
Figure 7: Mixed Inter-stimulus Interval Patterns.
Representative histograms for activity patterns that were a mix between bursty
and non-bursty. All inter-stimulus intervals (0-4s, bin size 0.05s) were represented as in
the bursty units, but the single 0-50ms peak was replaced by a broader peak usually with
a maximum around the 200ms bin. Numbers to the left of each denote the maximum of
the y-axis.
25
Mixed ISI Patterns
SPIKES
14
INTERVAL
30
35
2s
The units identified as AENs in the medial nucleus were responsive to a variety
of hydrodynamic stimulus sources (such as surface disturbances, turbulence, and water
jets), which are by nature spatially and temporally ill defined. A tactile lateral line
stimulus that either depressed the skin or displaced the fin near the receptive field was
also effective at activating the AENs, presumably by compressing the fluid in the canals
to stimulate neuromasts in that region. All receptive fields for this localized
mechanosensory stimulus were ipsilateral to the recording site in the hindbrain. The
response patterns from a tactile lateral line stimulus were still often complex and the lift
of the mechanical stimulator back off of the skin and out of the water often caused
surface disturbances, which induced a choppy response pattern with multiple stimulus
intervals. In addition, there were variations in the duration and reliability of the response
in different units.
Coupling to Ventilation
By applying the predictions of the adaptive filter model to the medial nucleus in
the mechanosensory system, it is expected that the effect of the molecular layer inputs
on the AEN will change when afferent activity is predicted by signals conveyed through
the parallel fiber system. An AEN sensory response was induced by coupling a localized
mechanosensory lateral line stimulus to the ventilation cycle in a freely breathing skate.
A temporally specific cancellation signal to the predictable pattern of afferent stimulation
should manifest in two ways, 1) by a reduction in the response to the regularly presented
stimulus, and 2) by a negative image of the induced sensory pattern imposed in the
26
spontaneous activity in the post-stimulation period. The more striking of these effects is
the absence of activity specific to the stimulus interval at stimulus offset.
In only 17 of the 57 units used for analysis of spontaneous activity was a dorsal
receptive field localized that had a singular near-tonic response to the tactile lateral line
stimulus from which a subtracted spike count could be analyzed in at least one run.
Only skates with moderate breathing rates (ventilation periods of 1.5-5 s) were tested.
Most units were lost before the three full periods of the run had been completed. Only
AENs isolated for 5 minutes of baseline, 10 minutes of stimulus coupling, and at least 5
minutes post-stimulus were of use for a full analysis of any changes in activity induced by
the repeated mechanosensory stimulus. If a unit was lost, it was usually because the cell
died or the skate moved. When possible, multiple runs were performed on the same cell
in order to test with different stimulus interval onset times and duration.
The negative image effect for an excitatory stimulus is quantified by a decreased
subtracted spike count at stimulus-offset compared to baseline. This decrease occurred
for all 17 units recorded in which 23 total runs were performed (while the inverse was
true for an inhibitory stimulus in the two cases where it was tested). However, the
decrease was not always statistically significant. In 14/23 runs in 12 units a statistically
significant negative image was generated (Figure 8).
27
Figure 8: Coupling of Mechanosensory Stimulus to Ventilation
A cancellation signal developed against the applied sensory stimulus pattern is
revealed by both a decrease in the subtracted spike count in the post-coupling period (D)
compared to baseline (A) and a decrease in the subtracted spike count by the end of the
coupling period (C) compared to the beginning (B), which were statistically significant.
From post-coupling (D) to 40 minutes (F), a recovery from the negative image to the
baseline activity pattern seems evident. (E: 10 minutes after the stimulus offset). The
bar under each histograms denotes the duration and timing of the stimulus interval. To
the left of the raster plot, vertical bars represent the time of the 60 trials over which the
histograms were obtained.
28
1s
1000 s
A
A
B
C
B
D
C
D
E
40 spikes
E
F
F
1s
Both a negative image and a decrease in the subtracted spike count during
coupling are predicted by the adaptive filter model. This was true for 8 of the 12 units
that generated a significant negative image (Figure 8). In the other 4, expected response
decline was not observed. However, the response to the stimulus during coupling
changed in a total of 13/17 units (16/23 runs). The change was a statistically significant
decrease in 12 of these runs. In 4 runs (2 of which were in the same unit) the change
was measured as an unpredicted increase.
Two of the units that did not generate a statistically significant negative image
according to my quantitative methods nevertheless appear, by visual inspection, to have
a very clear negative image of the applied sensory pattern after coupling (Figure 9).
Furthermore, for both of these units, two runs were performed, each with a different
stimulus onset time, and the most significant aspect of the qualitative analysis for these
two units is that the apparent negative image shifted in onset latency according to the
stimulus interval shift. This temporal specificity of the apparent negative image is
significant for the predictions of the adaptive filter model.
29
Figure 9: Apparent Negative Image Effect
A comparison of two runs performed in the same unit, but with different
stimulus interval onset. Raster plot (center) and histogram analyses (right: representing
60-trial periods from the run in the first half of raster, left: representing the run recorded
in the second half of raster). A negative image in the post-coupling period of each run
seemed clear, but was not statistically significant. A: pre-coupling baseline activity. B:
beginning of the coupling period. The black bar under each histogram represents the
stimulus interval. C: end of the coupling period. The reduced response to the stimulus
was statistically significant in both runs.
D: Post-stimulation period in which the
apparent negative image in the post coupling period of the second run has shifted its
onset in concordance with the stimulus interval shift during coupling. Note the very
short duration of 60 trials (black bars lateral to the raster) relative to the duration over
which the negative image appears.
30
1000 s
1s
A
A
B
B
C
A
C
D
E
B
D
F
C
A
B
20 spikes
E
F
1s
25 spikes
D
E
C
D
F
E
F
1s
The responses of at least half of the AENs tested in the medial nucleus revealed
a negative image or a stimulus decline (or both), which indicated that the predicted
cancellation signal was generated for a mechanosensory stimulus coupled to ventilation
in a freely breathing skate. In some units, the latency to the first spike in the stimulus
interval appeared to drift during the coupling period (Figure 8). This was not quantified,
and is not necessarily predicted by the adaptive filter model. However it signifies a
modulation of the effect that the stimulus has on the AEN sensory response pattern.
Another prediction of the model is that, in the post-stimulation period, there
should be an active process of recovery to baseline because the parallel fiber composite
generated to oppose the coupled stimulus is, in the post-stimulation period, active in the
absence of AEN activity. In some cases, a unit was held long enough to see a full
recovery of the baseline activity pattern after about 20-40 minutes, although not enough
samples were obtained to complete a comprehensive quantitative comparison.
Two of the 17 AENs tested had an inhibitory receptive field (Figure 10). At least
a weak negative image seems apparent by qualitative assessment, but in neither of these
cases was there a statistically significant change
31
Figure 10: Possible Negative Image to an Inhibitory Stimulus Interval
The response to the mechanosensory stimulus was a distinct inhibition followed
by a weaker excitation.
Both an inhibitory stimulus interval (left bar under each
histogram) and an excitatory stimulus interval (right bar under each histogram) were used
to analyze changes in the subtracted spike count.
In the post-stimulation period
histogram at left (D) it appears that the baseline (A) activity profile has been smoothed
to mirror the pattern induced during coupling (B and C). All histograms represent 60
trials
32
50 spikes
A
D
1s
200s
1s
A
B
B
C
C
D
Control
In a free-running stimulus condition, the uncoupled sensory response should not
be predictable by the parallel fiber system and thus no change in the activity patterns of
the AEN should be induced. In four of the 12 units that generated a cancellation signal
to a stimulus coupled to ventilation, an external stimulus was presented every 3 seconds
for 10 minutes, equivalent to the coupling period length in the previous experiment.
The regular presentation of the stimulus alone was not sufficient to produce the negative
image effect (Figure 11). There was in two units, however, a significant decline in the
response to the stimulus over the coupling period.
33
Figure 11: Comparison to a Free-Running Control
For the same unit depicted in Figure 8 (histograms at left), a mechanosensory
stimulus was presented free-running for 10 minutes (histograms at right). The negative
image in the post-stimulation period (D) for the coupling condition was not mimicked in
the free-running control, even though the sensory pattern induced during coupling (B
and C) were comparable and the baseline condition (A) was similar in each run. All
histograms represent 60 trials.
34
Coupled
to Ventilation
Free-Running
Control
Baseline
(pre-stimulation
period)
Begin
Stimulation
t=0
40 spikes
End
Stimulation
t = 10 min
Post-Stimulation
Period
1s
Fictive Ventilation: Motor Command
The adaptive filter model predicts that the molecular layer parallel fiber system
provides the information array that is molded into a cancellation signal in the AEN
induced by coincident sensory afference. Coupling a sensory response to ventilation in a
freely breathing skate provided sufficient conditions to cause changes in AEN activity
that imply the induction of an underlying cancellation signal. It is known that, in the
electrosensory system, some granule cells in the DGR (dorsal granular ridge) are
modulated independently by the ventilatory motor command discharge (Hjelmstad et al.,
1996). The second hypothesis of the adaptive filter model tested here is that coupling a
mechanosensory stimulus to fictive ventilation (recorded from the motor root of the
seventh cranial nerve) in a completely paralyzed skate would provide a sufficient
condition for inducing plasticity in the medial nucleus to cancel out the applied sensory
stimulus.
So far, 5 AENs have been tested. The group mean firing rate is 2.53 spikes/s
(+/- 1.08 SD). Three inter-stimulus interval patterns were represented in this sample (3
bursty, 1 non-bursty, and 1 mixed). Four of these AENs had dorsal receptive fields and
gave a sustained response to tactile lateral line stimulus. One unit generated a negative
image, but without a change in the subtracted spike count during the coupling period
(Figure 12). A free-running control stimulus was then presented to the same unit, with
no resulting negative image (Figure 13). In 2/4 units the subtracted spike count
decreased significantly by the end of coupling compared to the beginning, but no
negative image was measured in the post-coupling period. One unit out of the 4 did not
show any change as a result of the coupling.
35
An interesting effect occurred in one unit that may have been significant, but was
not quantified. This AEN’s baseline condition seemed to be modulated with the fictive
ventilation cycle. Since the skate was completely paralyzed, this apparent response could
not be a result of sensory reafference. Instead, the adaptive filter model would predict
that there was an outstanding cancellation signal for ventilatory reafference to which the
AEN was responding in the baseline condition.
36
Figure 12: Sufficiency of Fictive Ventilation to Induce a Cancellation Signal
Preliminary evidence suggests that coupling a mechanosensory stimulus to fictive
ventilation alone is sufficient for an appropriate cancellation signal to be generated.
Raster plot (right) depicts the one run, from which the histograms (left) of 60 trials each
were obtained at the denoted periods of the experiment. The effect of the cancellation
signal can be observed as a negative image at stimulus offset (D) of the sensory pattern
induced during the coupling period (B and C, in which the stimulus interval is denoted
by the bar under the histograms). This unit also seems to recover to baseline condition
(A) after 10 minutes (E) in the post-stimulus period (in which the stimulus interval is
denoted by a dotted line, indicating that the stimulus was not actually being presented
during this period of the run).
37
1s
500s
A
A
B
B
C
C
D
10 spikes
D
E
1s
E
Figure 13: Necessity of Coupling to the Motor Command
For the same unit depicted in Figure 12 (histograms at left), a mechanosensory
stimulus was presented free-running for 10 minutes (histograms at right). The negative
image in the post-stimulation period (D) for the coupling condition was not mimicked in
the free-running control, even though the sensory pattern induced during coupling (B
and C) were comparable and the baseline condition (A) was similar in each run. All
histograms represent 60 trials.
38
Free-Running
Control
Coupled to
Ventilation
Baseline
(pre-stimulation
period)
Begin
Stimulation
t=0
10 spieks
End
Stimulation
t = 10 min
Post-Stimulation
Period
1s
Discussion
Generalizing the Adaptive Filter Model to a Cerebellar-like Organization
In cerebellar-like structures, two inputs are compared at the level of the principal
neuron. An adaptive filter model was proposed to describe the mechanism by which the
elimination of predictable electrosensory stimuli such as reafference could be
accomplished in the first-order dorsal nucleus without turning down the gain of the
sensory system in general. The granule cells supplying the molecular layer of the dorsal
nucleus are modulated by various sources providing information about the skate’s own
behavior (Hjelmstad et al., 1996). Anti-Hebbian plasticity occurs specifically at these
molecular layer synapses (Bodznick et al., 1999) such that any activity in the afferent
input that is coincident with (predicted by) activity in the parallel fiber system will be
opposed and filtered from the output of the nucleus. This is observed as a change in
activity profile of the AEN both during the coupling period as a decrease in the sensory
response and at stimulus offset as a negative image of the stimulus pattern evoked during
coupling.
The parallel fiber system seems to provide a unique mechanism for conditioning
the output of cerebellar-like systems by experience-dependent plasticity. Prompted by
the hypothesis that the adaptive filter model applies to the operation of all cerebellar-like
structures based on their fundamental organization, I tested the prediction that the
medial nucleus in the mechanosensory system of the skate would filter the effect of a
sensory stimulus from its output if that stimulus was coupled to the skate’s ventilation.
Results from extracellular in vivo recordings of the principle neurons (AENs) in the
39
medial nucleus of the lateral line have been presented here and provide evidence that
inputs to the AEN were modified as a result of the presentation of the predictable
stimulus. That not all AENs generated a negative image for the predictable stimulus was
not a surprising result, since there are AENs in the dorsal nucleus that are not plastic
with this paradigm either (Montgomery and Bodznick, 1994). There could be distinct
populations of AENs in the medial nucleus that differ in the plasticity of their molecular
layer synapses.
A recurring stimulus did not induce the generation of a negative image in the
medial nucleus unless it was specifically timed with the ventilation cycle. Since the
parallel fiber system is the likely source of behavioral markers for ventilation, the parallel
fiber—AEN synapse seems essential to the operational capability of the circuit to predict
afferent activity and filter it out. Plasticity at the parallel fiber-AEN synapse has been
verified in the dorsal nucleus (Bodznick et al., 1999), but further experiments are needed
to test whether the same is true in the medial nucleus.
The generation of a cancellation signal to a predictable stimulus as reflected by
changes in the response pattern over the 10-minute coupling period were often difficult
to detect, which either means the stimulus was driving the cell above its normal
physiological range, or that 10 minutes is not enough time for a cancellation of a
stimulus to be effective. In the dorsal nucleus, the development of a cancellation signal
during coupling is similar in its time course to the decay in the post-stimulation recovery
period. In the medial nucleus, recovery seems to have a longer time-scale (20-40
minutes) than in the dorsal nucleus (5-10 min, Bodznick prs com). For a stronger
cancellation signal to be developed that has a greater effect on reducing the stimulus
40
reponse, the 10 minute coupling period may have to be extended closer to 40 minutes
for the mechanosensory system. Rates of onset and recovery need to be more
quantitatively measured in both electrosensory and mechanosensory AENs.
In the dorsal nucleus, a behavior is decomposed into many component signals
available through the parallel fiber system that can serve as markers for an external
stimulus, and induce changes in molecular layer synaptic inputs to the AEN. In the
medial nucleus, the motor command alone during fictive ventilation provided sufficient
conditions in for generation of a negative image to the sensory response induced by a
coincident mechanosensory stimulus in the medial nucleus.
The medial and dorsal nuclei are part of independent systems, but what is
fundamental to both is a cerebellar-like organization. This dissertation presents evidence
that stimulation of an AEN in the medial nucleus at a fixed point in the ventilation cycle
is sufficient to induce changes in AEN activity similar to those observed in the dorsal
nucleus. Sensory filtering in the medial nucleus demonstrated both temporal specificity
and a dependence on coupling the stimulus to either ventilation or the ventilatory motor
command alone. These are similar to the properties reported for the dorsal nucleus.
The results indicate that the same mechanism is operating in these two distinct sensory
systems. And it is the unique cerebellar-like organization that is common to both.
Limitations of the Methods Used in the Current Study
Neither comparing the subtracted spike count in the pre- to the post-stimulation
period nor within the coupling period seemed to capture the observed effects of a
cancellation signal in all units tested. Two mechanosensory AENs in particular appeared
41
to develop a clear negative image in the histogram analysis and a statistically significant
decrease in its response to the stimulus over the coupling period, but there was no
significant change in the subtracted spike count after coupling in either of the two runs
performed (Figure 9). I would not disagree with the statistical evaluation of these cells if
it were not for the fact that multiple runs were performed for both of these that
qualitatively appear to refute the result of the quantitative analysis. Two coupling runs
with different stimulus onset generated negative images with different timing that was
equivalent to the change in timing of the stimulus interval. Also, for one of these
examples a third run was completed in which a negative image was statistically
significant, but for which the effect did not appear to be as strong.
These seemingly contradictory results may reflect the inadequacy of our statistical
methods at accurately quantifying the effect of a cancellation signal. A comparison of 60
trials seemed a fair compromise for the duration of the negative image effect in most
cells. However, as seem in the example given in Figure 9, the standard 60 trials was
sometimes a poor temporal reflection of the time course of the apparent stimulusspecific change in activity for a few units. In general, our quantitative methods seem
conservative, with at least these two probable false negatives, but no apparent false
positives.
Several factors complicated the detection of a cancellation signal. First, the
cancellation signal is the phenomena that underlies the generation of a negative image to
the induced stimulus pattern at stimulus offset. So it cannot actually be measured with
extracellular techniques. Measuring changes in spike rate that correspond to the negative
image only implies the formation of a cancellation signal in the AEN. Second, the
42
subtracted spike count measure is essentially based on a comparison between the average
rate inside and outside of the stimulus interval. It is possible, however, for the average
rate over an interval within a trial to remain the same while the temporal pattern of the
activity over that interval changes. The subtracted spike count would not capture the
details of complex activity patterns that are often induced by lateral line stimuli.
Third, sporadic changes in spike rate were often observed that appeared to be
independent from any effects of our stimulation. The subtracted spike count was
developed as an analytical method to account for the random fluctuations in
spontaneous activity by normalizing the spike rate inside and outside the stimulus
interval for the AEN throughout an experiment. But there were still several cases where
low and sporadic activity in the baseline condition may have been the factor preventing
changes in spike rate specific to the actual stimulus interval from being detected.
Experiments in which the graded changes in synaptic strength are recorded would help
increase the resolution with which induction of plasticity is measured.
Functional Applications of Adaptive Filtering in the Mechanosense
Animals unintentionally detect signals produced by their own behaviors. This
sensory reafference is usually strong and can mask useful signals from the environment.
The adaptive filter model was originally proposed to explain how in the electrosensory
system, at the level of the secondary sensory neuron (the AEN) in the dorsal nucleus,
sensory responses to self stimulation in the primary afferents are eliminated. The
molecular layer parallel fiber system is essentially an information array comprising signals
marking the animal’s own behavior. Both biologically relevant signals and reafferent
43
noise are physiologically indistinguishable to the peripheral receptors, but only
reafference would reliably be coincident with activity in a subset of the parallel fibers. By
anti-Hebbian plasticity, the template of activity comprising the parallel fiber information
array is molded into a composite that actively compensates for any coincident
electrosensory input to the dorsal nucleus. The medial nucleus was capable of generating
a negative image to an applied mechanosensory stimulus predicted by ventilation. Thus,
the mechanosensory system may use this mechanism for the subtraction of sensory
reafference just like the electrosensory system.
Unlike the electrosense, the mechanosensory system is already protected in some
circumstances from intense sensory reafference because the neuromast endorgans are
innervated by an inhibitory efferent system that is active during vigorous motor activity
(Roberts and Russell, 1972; Russell, 1971). But it is not active during normal behavior,
in which case reafference would still be a noise issue. The adaptive filter mechanism has
an additional advantage over the efferent system in that it does not work by changing the
gain of the system and so it is probably more functional under most circumstances in
which the animal must remain sensitive to environmental stimuli.
The mechanosensory efferent system is also selectively activated by corollary
motor discharges from certain behaviors. In the medial nucleus, coupling a stimulus to
the motor command during fictive ventilation was sufficient to induce plasticity in
accordance with the adaptive filter model. Thus, these results provide evidence that
corollary discharge information is also being used in the dynamic context of the adaptive
filter mechanism in the medial nucleus. There is now a need to determine the precise
relationship between the two mechanisms to which the corollary discharge contributes to
44
better understand the functional applications for the adaptive filter to sensory processing
in the mechanosensory system during normal behavior.
Ventilation was used in the initial experiments in the medial nucleus because it is
the behavior most amenable to in vivo electrophysiological recordings and sitmulus
coupling to this behavior provided sufficient conditions for a cancellation signal to
develop to the mechanosensory response in the medial nucleus AEN. The adaptive
filter mechanism in the dorsal nucleus is behaviorally significant to the animal because
otherwise, intense electrosensory reafferent stimulation from behaviors such as
ventilation would mask the detection of biological signals in the environment. However,
it is not clear that the mechanosensory system incurs the same intense reafference from
ventilation. It may be more relevant to test the probable behavioral function of the
adaptive filter mechanism in the medial nucleus by coupling a stimulus to a behavior that
would more strongly modulate the mechanosensory system such as swimming and other
fin movements.
Every behavior must be made up of a different information array of signals
associated with it, including motor commands, descending sensory information, or
proprioceptive information. Essential to determining the probable function of a sensory
filter in the mechanosensory system is to determine the specific conditions under which
plasticity can actually occur. In the dorsal nucleus, each of these separate kinds of
information are individually sufficient to induce the generation of a cancellation signal
(Montgomery and Bodznick, 1999). Signals from other stimulus modalities should be
available in the parallel fiber system of the medial nucleus. Preliminary results from
coupling a mechanosensory stimulus to a uniform electric field alone has not yielded
45
statistically significant evidence of induction of a cancellation signal. To establish the
multimodal capabilities of the medial nucleus adaptive filter, stimuli could be coupled to
any eighth-nerve stimulus (possibly auditory), since they are known to terminate in the
lateral granular (LG) region supplying the medial nucleus parallel fiber system (Schmidt,
1985).
Since the response to a free-running stimulus declined in two of the control runs,
future experiments should be conducted to determine whether any habituation is
contributing to filtering mechanisms within the hindbrain medial nucleus. A habituation
mechanism could decrease the response to a regular stimulus whether it was predictable
by the parallel fiber system or not. However, habituation would not account for the
temporally specific negative image of the response pattern at stimulus offset, which is a
product of the adaptive sensory filter.
The responses of AENs in the medial nucleus to mechanosensory stimuli seem
more complex and variable than the responses of AENs in the dorsal nucleus to
electrosensory stimuli. The complex and ill-defined nature of hydrodynamic stimulus
patterns could contribute to the variability in the capability of our methods to accurately
document physiological changes induced by mechanical stimulation. Since my statistical
analysis of changes in activity induced by complex response patterns seemed inadequate,
it raises the question of whether the organism itself can filter these complex sensory
patterns. It is not clear what resolution for an adaptive filter should be expected for the
mechanism to be functionally significant to the animal. Therefore a future goal should
be to classify the temporal capabilities of the filter in the medial nucleus, and to know
more about real patterns of self-stimulation.
46
Mode of Lateral Line Stimulus
A mechanosensory stimulus can come from a variety of sources. Neuromast hair
cells of the lateral line encode shearing motion between the fluid and the skin surface. In
the water, pressure differences between canal pores cause canal fluid motion
corresponding to the accelerations of the external water flow and activate the neuromast
hair cells accordingly (Sand, 1984). Investigators have mostly used a high-frequency
vibrating sphere in studies to classify the physiological properties of neurons in the
lateral line system (Bleckmann et al., 2003; Coombs and Braun, 2003). But
hydrodynamic stimuli, effective at activating the medial nucleus AENs, often have illdefined temporal and spatial distributions.
For our purposes in initially testing the possibility of adaptive filter properties in
the medial nucleus, crude measures of plasticity by analysis of average spike rates were
used, and having the stimulus interval be as long in duration as possible was critical to an
accurate measure of subtracted spike count. The tactile stimulation method used in most
experiments was a depression of the skin made by the lowering of a mechanical arm,
which provided a more near-DC mechanical stimulus to elicit near-tonic responses from
most AENs. This method was developed to lengthen the duration of the stimulation
interval so that our quantification methods would be better at representing the activity
patterns of the AEN.
Although to some this tactile stimulus may seem to be contrived or unnatural for
the lateral line, there is reason to believe that the lateral line system may actually process
exactly such a stimulus in a functional context. Specifically, the mechanotactile
hypothesis was proposed by Maruska and Tricas (2004) as a function for the seemingly
47
unadaptive ventral non-pored canals of some batoids and elasmobranchs. Coupling of
the skin and canal fluid should result in lateral line afferent responses to the velocity of
the skin movement.
Our model species, Raja erinacea, does have such a ventral non-pored canal
distribution. Although I have mostly examined units with dorsal receptive fields, the
mechanical arm used in the experiments presented in this dissertation generated tactile
depression of the skin to varying degrees at a frequency near that used to test the
mechanotactile hypothesis. If depression of the skin is an effective stimulus for the
AENs, then it seems to be a valid method to test the filtering capabilities of the medial
nucleus for the lateral line.
More tests of a natural lateral line stimulus will need to be conducted, but with
the current experimental set-up it is rare to find an AEN with a receptive field in a
location that is maximally stimulated by just touching the water with the rod. If our
statistical methods are not high enough resolution to analyze some stimulation
paradigms such as those in which a complex response is induced by purely
hydrodynamic stimuli, does that mean the organism itself cannot filter those types of
stimuli? The results obtained thus far at least encourage a search with more natural
stimuli for analysis of a greater variety of effects to determine if the adaptive filter
mechanism will have functional significance for the mechanosensory system during
normal behavior.
48
Self-Organizing Systems
“Comparative studies can provide important clues toward an
understanding of the more basic operations of a neural structure,
encouraging the formation of more generalized theories of function.”
(Nixon, 2003)
In this dissertation I support the proposal to consider all cerebellar-like
structures and the cerebellum in the same conceptual framework based on the principles
of the adaptive filter model. It was in the behavioral context of the elasmobranch
sensory systems that a function for the adaptive filter model in eliminating sensory
reafference was proposed (Montgomery and Bodznick, 1999). The parallel fibers convey
signals from heterogeneous sources and provide an information array that is a template
of neural markers for the fish’s behavior. Plasticity at these molecular layer synapses
follows anti-Hebbian rules of association such that the information array is continuously
molded to provide a temporal excitation/inhibition pattern that opposes and attenuates
any predictable patterns of afferent-induced activity in the AEN. Only signals from
sources that are not predictable by the parallel fiber activity are transmitted to higher
brain regions, thus removing reafference from the system.
Now it is in the context of the adaptive filter model that I am addressing a
fundamental relationship among cerebellar-like structures that permits them to condition
the output of a principal neuron by continuously changing the contribution of the
molecular layer inputs to its activity. The key functional characteristic of cerebellar-like
structures is that the molecular layer parallel fiber array makes many synapses on the
principal cell, which also receives a second functionally and anatomically segregated type
49
of input (Figure 1). Sensory and non-sensory signals converge on the principal neuron
from these two sources and are summated, but the output of the system always depends
on the state of the molecular layer synapses, which are molded by the recent history of
activity in the circuit.
A self-organizing system generates complex patterns of behavior with no topdown control. Instead, the behavior emerges when the interacting components follow
simple rules of association (Johnson, 2001). The adaptive filter model represents the
rules by which the principal neuron of a cerebellar-like structure responds dynamically to
the association of its inputs: 1) molecular layer synapses that are consistently active
coincident with afferent excitation of the principal neuron are depressed, and 2)
molecular layer synapses active in the absence of afferent activation are potentiated.
From these rules, the principal neuron selects a composite of the molecular layer parallel
fiber information array to which it responds. Access to an array of information in the
parallel fiber system and plasticity at those molecular layer synapses seems to have a
fundamental role in the self-organizing behavior of the principal neuron to modulate its
output from a cerebellar-like circuit. Defining the behavior of a system from such a
bottom-up approach gives the principles translational value among systems that, on the
macro level may be functionally distinct, but at the circuit level comprise the same
principles of self-organization.
50
The Cerebellum
Theories of mind are often based on a modular hypothesis and one-region onefunction view (Fodor, 1983). The function of a complex structure such as the
cerebellum is mostly defined by lesion and neuroimaging studies. Until recently, the
function of motor control has been ascribed to the cerebellum. Now the cerebellum
appears to be involved in a great variety of motor, sensory, and cognitive tasks, which
has puzzled investigators in trying to define the seemingly heterogeneous function.
However, in the complexity of the cerebellum is regularity. A striking uniform
organization has been noted by all who study it. The simplest description of the
organization is that of a laminar architecture in which a principal neuron (the Purkinje
cell) receives two distinct inputs: one from a population of parallel fibers in the
molecular layer carrying mixed-source information, and one from climbing fibers
originating in the Inferior Olive (IO). Plasticity at the parallel fiber synapses is
associative with induction of LTD upon climbing fiber (afferent) activation. This basic
organization (coupled with plasticity) is that which underlies the adaptive filter
mechanism now attributable to cerebellar-like structures in general. Studies on the
synaptic level are not sufficient for elucidating the function of the cerebellar circuit
because the significance of inputs and outputs to the structure are ill-defined. It may be
useful to generalize the principles of the fundamental computational role for a cerebellarlike parallel fiber system suggested by studies of the primary nuclei of the lateral line and
octaval senses to the cerebellum in an attempt to define the function of such a complex
structure from a bottom-up approach.
51
Structures studied in the human brain tend to acquire operational definitions
from a top-down approach. In the early 1900s, behavior of the cerebellar patients were
observed for abnormalities in behavior and ability to complete various tasks involving
motor coordination and fine movement control. Rudolfo Llinas notes that although the
cerebellum is innervated by a multitude of heterogeneous sources, the only major deficits
seem to be the inability to coordinate movement (Llinas, 1992). The cerebellar system
has thus been viewed as a neuronal machine for the control of these motor functions that
are lacking when the cerebellum is lesioned or ablated.
Not all information currently available supports the conventional view of the
cerebellum as a control point for the organization of movements. Behavioral and
cognitive studies demonstrate that cerebellar patients are also slower at sensory
processing and discrimination (such as distinguishing between two sounds), have trouble
with language (such as evoking specific verbs quickly), and have maladjusted emotional
reactions (see Bower and Parsons (2003) for review).
Although sensory and motor functions are inextricably intertwined in nearly
every behavior, Parsons and coworkers have attempted to parse the degree of cerebellar
involvement in each. He found that the dentate, a deep cerebellar nucleus, was more
strongly activated with the purposeful acquisition of sensory information than with fine
movement of the fingers alone in absence of such intent (Parsons et al., 1997). Thus the
results of their neuroimaging study suggest that the degree of activation was correlated
with the intent to use motor behaviors in the context of sensory discrimination.
The motor control hypothesis has been modified by some investigators to
encompass these results. Apps (2005) describes the cerebellum as a sensorimotor control
52
system. And Bower’s new hypothesis attributes cerebellar control to being “motor for
sensory’s sake” (Bower, 1997). By mapping tactile-evoked responses of cerebellar
regions, Bower (1997) describes a pattern of sensory inputs to the cerebellum that seems
to represent discrete regions of the animal somatotopically in multiple regions of the
cerebellar cortex topography. The spatial relationship between areas of the body surface
appear to Bower to be mapped according to use of that body part. The body part
featured varies between species, but it always seems to be the one prominently used by
the animal in active tactile exploration (whiskers are featured in the rat cerebellum,
forepaws in the cat, and fingers in primates) (Bower 1997).
Apparent involvement of the cerebellum in motor control may be a product of
the fact that motor coordination requires sensory information about the environment.
Parsons (1997) believes that the “cerebellum is specifically involved in monitoring and
adjusting the acquisition of most of the sensory data on which the rest of the nervous
system depends. Thus it increases the efficiency of function of other sensory systems.” By
this hypothesis, it is predicted that cerebellar patients may demonstrate ataxia and slowed
reaction times due to “poorly controlled sensory data” (Parsons 1997) as opposed to
deficiencies in “motor control.”
Blakemore, Wolpert and Frith (2000) concluded that the cerebellum is part of a
system that underlies our inability to tickle ourselves. Their hypothesis is that the
cerebellum provides predictions about the sensory consequences of actions and can be
used to cancel the perception of tactile sensations from self -generated movements.
Similar to their hypothesis, and to that of Parsons (1997), Devor (2000) notes that the
cerebellum now seems more closely related to the need for distinguishing movement of
53
one’s self from movement of objects in the environment during active sensory
exploration. She predicts that the cerebellum is involved most intimately with tasks
involving comparisons between sensory stimuli.
Organization of the cerebellar cortex has been conserved throughout vertebrate
evolution (Bower and Parsons, 2003; Roberts and Ryan, 1971) and is mirrored in the
cerebellar-like sensory structures of teleost and non-teleost fish (Larsell, 1967).
Evolution and development often provide clues to a structure’s function. All cerebellarlike structures (including the cerebellum itself) develop from the same somatosensory
plate of the neural tube (Devor, 2000), which may be a clue for its apparent involvement
in sensory processes. Or it may indicate that structures requiring similar architecture for
their function, such as the parallel fiber system of cerebellar-like structures, all develop
from the same type of neural tube tissue.
The question that still emerges is how the cerebellum could control all of the
disparate functions attributed to it. Disruption of motor behavior is known to result
from ablation of the cerebellum, but this is likely because motor acts are the only directly
observable effect. Neuroimaging studies lack comprehensiveness because activation of a
brain region during a task demonstrates only that it is involved, and does not even
necessitate that a structure is directly contributing to the task at all. Although Bower’s
hypothesis is supported and diverges from the classic description of the cerebellum’s
function as purely motor, these studies still contribute nothing to how the cerebellum
performs any function. Due to these limitations, the field still lacks consensus about
what the cerebellum does.
54
Authors DeSchutter and Meax (1996) began to re-examine the longstanding
dogmas of cerebellar control theory and address the need for a more bottom-up
approach to defining the role of the cerebellum in behavior (DeSchutter and Meax,
1996). Ohyama and Mauk (2003) view cerebellar function in terms of the information
processing it accomplishes and use Pavlovian eyelid conditioning as an example that
defines the computational capacity of the cerebellum as a type of feedforward control.
Ivry and Spencer (2004) study the cerebellum with computational models and view it as a
structure specialized for a timing mechanism. However, the most direct way of
addressing the function of a neuronal circuit is by studying the organization and
physiological properties of its components.
At the neuronal level, the cerebellar cortex has unique structural characteristics
and is recognized for its uniform, yet unusual, organization in which the sole output of
the cortex (the Purkinje cell) receives synapses from two distinct sources. Granule cells
are innervated by mossy fibers and give rise to a parallel fiber system in the molecular
layer, which makes a large number of synapses on the spiny apical dendrites of the
Purkinje cell. From the Inferior Olive (IO), one climbing fiber synapses on each
Purkinje cell.
One of the main problems with studying the cerebellum at a synaptic level is that
the neural circuits are too many synapses deep into the brain. The connections of the
cerebellum to other cortical structures are complex and its role in behavioral modulation
is not direct. The inputs are already filtered at the level of the climbing fiber, and the
output of the cerebellum is to the deep nuclei whose contribution to modulating motor
output is loose and ill-defined.
55
Structures in which data processing relies on knowledge of self seem to share the
common features of a laminar, cerebellar-like organization. The cerebellar-like medullary
structures of the elasmobranch, which have been the focus of this dissertation, are
sensory structures with a seemingly well understood functional significance. The
adaptive filter model has provided insight into sensory processing in general and the
elimination of reafference based on the adaptive filter model. Perhaps the adaptive filter
model can also contribute more broadly to understanding the mechanism involved in
conditioning the output of other cerebellar-like structures.
The organization of these primary sensory nuclei is called cerebellar-like because
the most striking feature of the common architecture is, like the cerebellum, the
molecular layer parallel fiber system, which (along with stellate cell inhibitory
interneurons) makes a multitude of synapses on the principal neuron. The principal
neurons are Purkinje-like because they have a widely-branched array of spiny apical
dendrites that extend into the molecular layer, while concurrently receiving a second,
functionally distinct, afferent input. The cerebellum epitomizes the dichotomy between
the number of parallel fiber inputs (many) to afferent inputs (only one climbing fiber) to
the principal (Purkinje) cell. The parallel fibers carry a variety of signals because the
granule cells have a wide variety of sources, while all of the afferent inputs have a single
common source.
Synaptic plasticity is a salient feature of both the cerebellar-like structures and
the cerebellum and depends on associated principal cell activation such that, in
accordance with the information driving the granule cells, the outputs of the system is
adjusted (Figure 14). However there are some differences between the two systems.
56
Figure 14: Modulation of output from Cerebellar-like Structures and the Cerebellum.
Schematic comparison of the two streams of input at the principal neuron of
cerebellar-like structures and the cerebellum (created according to the review in Devor
2000). The major structural commonality is the parallel fiber system originating from a
heterogeneous population of granule cells and making en-passante synapses on the spiny
apical dendrites in the molecular layer.
Although the structures differ in the
characteristics of their afferent input, the key feature is that the parallel fiber system
common to a cerebellar-like organization seems to provides a unique mechanism for
conditioning the output of the system based on characteristic associative plasticity that
molds the parallel fiber information array, which then modulates the activity of the
principal neuron. Plasticity is continuously induced in the cerebellar-like structures by
ongoing afferent activity (or the lack thereof, according to the adaptive filter model).
However in the cerebellar cortex, a continuous stream of parallel fiber activity is
interrupted by the induction of plasticity upon non-continuous climbing fiber activation.
The output of the principle cell depend both on the continuous activity in the parallel
fiber system and the state of the synaptic weights in the molecular layer.
57
Cerebellum
Cerebellar-like Structures
Midbrain
or
Efferent
Sensory
Apparatus
Deep
Nuclei
Sensory
Apparatus
IO
Plasticity in the electrosensory and mechanosensory systems seems to be continuous.
The adaptive filter model predicts that the molecular layer synapses are constantly
constructing negative images of any ongoing afferent activity that is correlated to activity
in the parallel fiber information array. However, in the cerebellum, climbing fiber
activity, which carries only information about unexpected sensory inputs, induces
plasticity. And inhibitory inputs suppress olivary excitation when the plasticity is not
needed (Devor 2000 for review).
The fundamental operation of cerebellar-like structures is self-organization to
effect emergent adaptive behavior for the systems to which they belong. Changing
environmental conditions effect changes in the parallel fiber system, which modulates
the activity of the principal neuron. Despite some physiological differences, the
cerebellum itself may also be viewed as a self-organizing system, whose function at the
circuit level may be to adaptively contribute to behaviors in a changing and complex
environment.
None of the ideas presented here contradict the fact that the cerebellum is
involved in motor coordination and control. More significant than defining the extent
of cerebellar involvement is elucidating the nature of its function. Redefining the
function of the cerebellum from a bottom-up approach in the context of the adaptive
filter model would encompass the seemingly contradictory involvement in sensory
discrimination, cognition, emotion, and motor control. The cerebellum does not control
motor output directly, but the operational capacity of a cerebellar-like circuit may more
generally underlie the ability for the rest of the central nervous system to carry out
proper processing in all behavioral and cognitive functions.
58
LITERATURE CITED
Apps, R. and Garwicz, M. (2005). Anatomical and physiological foundations of
cerebellar information processing. Nat Rev Neurosci 6, 297-311.
Bastian, J. (1999). Plasticity of feedback inputs in the apteronotid electrosensory
system. J Exp Biol 202, 1327-1337.
Bell, C. (1982). Properties of a Modifiable Efference Copy in an Electric Fish.
Journal of Neurophysiology 47, 1043-1056.
Bell, C., Bodznick, D., Montgomery, J. and Bastian, J. (1997). The
generation and subtraction of sensory expectations within cerebellum-like structures.
Brain Behav Evol 50 Suppl 1, 17-31.
Bell, C., Han, V., Sugawara, Y. and Grant, K. (1999). Synaptic plasticity in the
mormyrid electrosensory lobe. J Exp Biol 202, 1339-1347.
Bertetto, L. (2007). Functional Synaptic Plasticity in the Electrosensory System
of the little skate, Raja erinacea. In Biology, vol. PhD. Middletown: Wesleyan Universiry.
Blakemore, S. J., Wolpert, D. and Frith, C. (2000). Why can't you tickle
yourself? Neuroreport 11, R11-6.
Bleckmann, H., Mogdans, J. and Dehnhardt, G. (2003). Processing of
Dipole and More Complex Hydrodynamic Stimuli Under Still- and Running-Water
Conditions. In Sensory Processing in Aquatic Environments, (ed. M. N. Collin SP). New York:
Springer.
60
Bodznick, D. (1989). Comparisons Between Electrosensory and
Mechanosensory Lateral Line Systems. In The Mechanosensory Lateral Line: Neurobiology and
Evolution, eds. C. S. G. P. and M. H.). New York: Springer-Verlag.
Bodznick, D., Montgomery, J. C. and Carey, M. (1999). Adaptive
mechanisms in the elasmobranch hindbrain. J Exp Biol 202, 1357-64.
Bodznick, D. and Northcutt, R. G. (1980). Segregation of electro- and
mechanoreceptive inputs to the elasmobranch medulla. Brain Res 195, 313-21.
Boord, R. L. and Northcutt, R. G. (1982). Ascending lateral line pathways to
the midbrain of the clearnose skate. J Comp Neurol 207, 274-282.
Bower, J. (1997). Is the cerebellum sensory for motor's sake, or motor for
sensory's sake; the view from the whiskers of a rat? Progress in brain research 114, 463-496.
Bower, J. and Parsons, L. (2003). Rethinking the "lesser brain". Sci Am 289, 5057.
Coombs S, B. C. (2003). Information Processing by the Lateral Line System. In
Sensory Processing in Aquatic Environments, (ed. M. N. Collin SP). New York: Springer.
Coombs, S. and Braun, C. B. (2003). Information Processing by the Lateral
Line System. In Sensory Processing in Aquatic Environments, (ed. M. N. Collin SP). New
York: Springer.
DeSchutter, E. and Meax, R. (1996). The cerebellem: cortical processing and
theory. Curr Opin Neurobiol. 6, 759-764.
Devor, A. (2000). Is the cerebellum like cerebellar-like structures? Brain Res Brian
Res Rev 34, 149-156.
61
Dusenbery, D. B. (1992). Sensory Ecology: how organisms acquire and respond
to information. New York: WH Freeman and Company.
Fodor, J. (1983). The modularity of mind. Cambridge, Mass: MIT Press.
Hjelmstad, G., Parks, G. and Bodznick, D. (1996). Motor corollary discharge
activity and sensory responses related to ventilation in the skate vestibulolateral
cerebellum: implications for electrosensory processing. J Exp Biol 199, 673-81.
Ivry, R. B. and Spencer, R. M. (2004). The neural representation of time. Curr
Opin Neurobiol. 14, 225-232.
Johnson, S. (2001). Emergence: The Connected Lives of Ants, Brains, Cities,
and Software. New York: Touchstone.
Johnston, J. B. (1902). The brain of Petromyzon. J Comp Neurol 7, 2-82.
Larsell, O. (1967). The Comparative Anatomy and Histology of the Cerebellum
from Myxinoids through Birds. Minneapolis: University of Minnesota Press.
Llinas, R. (1992). Cerebellum Revisited. New York: Springer-Verlag.
Maruska, K. P. and Tricas, T. C. (2004). Test of the mechanotactile
hypothesis: neuromast morphology and response dynamics of mechanosensory lateral
line primary afferent in the stingray. J Exp Biol 207, 3463-3476.
May, B. (2000). Role of the dorsal cochlear nucleus in the sound localization
behavior of cats. Hear. Res. 148, 74-87.
Montgomery, J. C. and Bodznick, D. (1994). An adaptive filter that cancels
self-induced noise in the electrosensory and lateral line mechanosensory systems of fish.
Neurosci Lett 174, 145-8.
62
Montgomery, J. C. and Bodznick, D. (1999). Signals and noise in the
elasmobranch electrosensory system. J Exp Biol 202, 1349-55.
Montgomery, J. C., Coombs, S., Conley, R. A. and Bodznick, D. (1995).
Hindbrain sensory processing in lateral line, electrosensory, and auditory systems: a
comparative overview of anatomical and functional similarities. Auditory Neuroscience 1,
207-231.
Nixon, P. (2003). The role of the cerebellum in preparing responses to
predictable sensory events. Cerebellum 2, 114-122.
Northcutt, R. G. (1980). Central Auditory Pathways in Anamniotic Vertebrates.
In Comparative Studies of Hearing in Vertebrates, (ed. A. N. P. a. R. R. Fay), pp. 79-118. New
York: Springer-Verlag.
Ohyama, T., Nores, W. L., Murphy, M. and Mauk, M. D. (2003). What the
cerebellum computes. Trends Neurosci 26, 222-227.
Parsons, L. M., Bower, J. M., Gao, J., Xiong, J., Li, J. and Fox, P. (1997).
Lateral Cerebellar Hemispheres Actively Support Sensory Acquisition and
Discrimination Rather Than Motor Control. Learning & Memory 4, 49-62.
Paul, D. H., Roberts, B. L. and Ryan, K. P. (1977). Comparisons between the
lateral-line lobes of the dogfish and the cerebellum: An ultrastructural study. J. Hirnforsch
18, 335-343.
Puzdrowski RL, L. R. (1993). The octavolateral systems in the stingray,
Dasyatis sabina. I. Primary Projections of the octaval and lateral line nerves. J Comp
Neurol 332, 21-37.
63
Roberts, B. L. and Russell, I. J. (1972). The activity of lateral-line efferent
neurons in stationary and swimming dogfish. J Exp Biol 57, 435-448.
Roberts, B. L. and Ryan, K. P. (1971). The fine structure of the lateral-line
sense organs of dogfish. Proc. R. Soc. Lond. B 179, 157-169.
Russell, I. (1971). The role of the lateral-line efferent system in Xenopus Laevis.
J Exp Biol 54, 621-641.
Sand, O. (1984). Lateral-line systems. In Comparative Physiology of sensory systems,
(ed. K. R. Bolis L, Maddrell SHP). Cambridge: Cambridge University Press.
Schmidt, A. W. (1985). The Afferent and Efferent Connections of pars lateralis.
In Biology, vol. M.A., pp. 47. Middletown, CT: Wesleyan University.
Schmidt, A. W. and Bodznick, D. (1987). Afferent and efferent connections of
the vestibulateral cerebellum of the little skate Raja erinacea. Brain Behav Evol 30, 282302.
64
Acknowledgements
David Bodznick
Proper thanks to an immeasurably good teacher and quality human being.
Colleagues
Lisa: guidance and support both inside and outside the lab, Zhi: you provide everyone in
the lab with unconditional support, Janet: I have worked closely with you in many
contexts, in all of which I have appreciated your friendship and responsibility. You will
succeed in everything you do, Lianne Morris-Smith: for helping me initially find my way
around the lab, the equipment, and the medial nucleus.
The Thesis Committee: John Kirn and Gloster Aaron
I hope it was not too much of a bear to tackle. I appreciate your patience and
confidence.
I have gotten to know many of the Wesleyan faculty and it makes me proud to have
gone to a school in which such relationships are withheld and valued. Learning can’t just
happen inside the lecture room.
There are many graduate students who have known me since my years as a work-study in
the biology office. Thank you for never looking down on me and for welcoming me
into the graduate life this year. Your support has grounded me.
Blanche, Marjorie, and Susan
Your unlimited help and kindness towards everyone in the department is heartening.
The bio office feels like a safe space for me and I have appreciated your support
throughout the years. Our community would not stand together without your care and
interest in both our individual and our common experiences.
65
Mom: You are a much larger part of my science stuff that I think that you realize. I
would not appreciate what I do without an underlying interest in artistic expression and
creativity. Many of the things I do at school remind me of you and make me appreciate
all you have taught me. Thank you for your unconditional (but not blind) support in
every decision that I make, even if you have differing opinions.
Dad: I still remember driving to elementary school and drawing molecular structures. I
was probably the first 5th grader to have a sense of what an empirical formula was. But
you’ve never pushed me in academics, which makes me even more pleased about where
I am now and how everything has turned out.
…From both of you I have learned above all else to appreciate all the little things in life,
nature, and people. And although I can be sad, I have often smiled at times when others
may not.
Friends and Family
Thanks to those of you who have extended my immediate family. All best friends, aunts,
uncles, grandparents, first and second cousins, and those once-removed…I am lucky to
be part of the close network that makes me want to continue growing up close to our
collective homes.
66