NEURAL INSTRUCTIVE SIGNALS IN THE CEREBELLUM
DISSERTATION SUBMITTED TO THE PROGRAM IN
NEUROSCIENCES AND THE COMMITTEE ON GRADUATE
STUDIES OF STANFORD UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
Michael C. Ke May 2010
http://creativecommons.org/licenses/by-nc/3.0/us/
This dissertation is online at: http://purl.stanford.edu/wy417ny4640
© 2010 by Michael Chinwen Ke. All Rights Reserved.
Re-distributed by Stanford University under license with the author.
This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.
ii
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Jennifer Raymond, Primary Adviser
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
John Huguenard
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
Eric Knudsen
I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.
William Newsome
Approved for the Stanford University Committee on Graduate Studies.
Patricia J. Gumport, Vice Provost Graduate Education
This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.
iii
iv
Abstract
An understanding of the neural patterns available to guide plasticity in vivo is needed
to bridge our knowledge of synaptic plasticity to its function in learning. I
investigated the patterns of neural activity that trigger plasticity in vivo in a simple
cerebellum-dependent motor learning task, adaptation of the vestibulo-ocular reflex
(VOR), with the specific goal of determining which neurons carry the instructive
signals that trigger plasticity in the circuit for the VOR. The VOR stabilizes images
on the retina during head turns by using vestibular signals to generate compensatory
smooth eye movements in the opposite direction of head motion. Motor learning
maintains the accuracy of the VOR by modifying the gain and timing of the reflex
whenever retinal image motion is persistently associated with head movements. In the
laboratory, motor learning in the VOR can be acutely induced by pairing head
movements with motion of a visual stimulus.
Two specific hypotheses have been proposed regarding the neural signals that guide
motor learning in the VOR. One suggests that learning is guided by the activity of
Purkinje cells, the output neurons of the cerebellum[1]. The other hypothesis suggests
that learning is guided by climbing fiber input to the Purkinje cells[2-4]. Previous
experiments addressing which neurons carry instructive signals have typically used a
single training condition for increasing VOR gain and a single training condition for
decreasing VOR gain[5, 6]. These two training conditions each elicited Purkinje cell
and climbing fiber signals that carried information about the required direction of
learning, and since the patterns of neural activity were consistent with both
hypotheses, data are needed to provide constraints that could discriminate between the
hypotheses.
The goal of my research is to provide such constraints by recording the patterns of
neural activity present in Purkinje cells and climbing fibers during a broader range of
visual-vestibular stimuli that induce motor learning in the VOR. I induced motor
learning in the VOR by pairing head movements with complex visual stimuli. These
v
novel behavioral manipulations elicited many different combinations of Purkinje cell
and climbing fiber signals, allowing us to evaluate how each of these neural signals
contributes to learning.
My data demonstrated that neither instructive signals in the climbing fibers nor
Purkinje cells are necessary for learning, although either signals appear to be sufficient
to support learning. Additionally, the largest changes in VOR gain occurred when
both signals were present, suggesting that the changes mediated by Purkinje cell-
triggered mechanisms and climbing-fiber triggered mechanisms are additive in their
effects at the behavioral level. These findings are evidence that motor learning in the
VOR is accomplished by parallel and independent operation of climbing fiber-
triggered and Purkinje cell-triggered plasticity mechanisms. If cerebellum dependent
motor learning is supported by the parallel and independent operation of plasticity
mechanisms, similar motor learning need not be accomplished in a stereotyped
fashion, but rather similar motor learning can be achieved by engaging distinct subsets
of plasticity mechanisms each under the control of a unique instructive signal.
vi
PREFACE
Publications
The body of this thesis is in 3 chapters (Chapters 2-4) and will correspond to
publications in which I will be co-first author. Chapter 3 discuss results that have
been published[7]. The second publication is in preparation, and will focus on the
effects of target and background motion on motor learning in the VOR and the
putative instructive signals hypothesized to guide VOR adaptation (Chapter 2). The
data from the generalization of learning studies discussed in Chapter 4 have been
published along with other work in mice from the J. Raymond lab[8].
Structure of collaborations
The behavioral and recording experiments in the thesis involving Monkey L and C
were carried out by myself in Jennifer Raymond’s lab. The experiments conducted on
Monkey E were carried out by Christine Guo, a graduate student in the lab. Based on
her studies in Monkey E, she was the first to recognize the complex interaction
between target slip and the effect of conflicting background motion on climbing fiber
responses, as described in Chapter 2. The data analyses for this project were
conducted in collaboration with Christine, and the writing of manuscripts has been
split between Christine, Jennifer, and me.
Acknowledgments
Many people have played an integral role during my studies at Stanford University.
First, I would like to thank my labmates in the Raymond lab, for their academic
support, and continual affirmation that the end is indeed in sight: Rhea Kimpo, Grace
Zhou, Akira Katoh, Adam Bristol, Edward Boyden, Christine Guo, Barbara Nguyen,
and Soon-lim Shin. Without the technical and laboratory support provided by these
lab managers, the experiments would not have been possible: Kate Wesley, Rachel
Levine, and in particular, Pam Louderback, who’s motherly patience for my constant
clutter in lab was much appreciated. I would like to thank my committee members,
Richard Reimer, John Huguenard, William Newsome, and Eric Knudsen for their
time, brilliant insights, and frank, but diplomatic assessments of my progress. Of
course, I am thankful for the financial support, particularly MSTP at Stanford under
vii
the guidance of Greg Barsh and Seung Kim. As an East Coast transplant, I was
fortunate to make many new friends outside Stanford, and I am grateful for their
companionship and words of encouragement. In particular, I would like to thank these
friends who were a constant presence in my life throughout my entire tenure at
Stanford—Hugh Huynh, Adam Schetky, Phuong Quach, Rick Quach, Rich Stein, and
Euell Tochip. Of course, without the support of my family, I would not be able to
reach any of my personal goals, including graduating from Stanford. I would like to
thank my Mom and Dad for not asking precisely when I am graduating, but instead
allowing me to find my own path, no matter how tortuous it might have been. And I
would like to thank my sister, Sandra, for somehow convincing my parents that
whatever my decisions were, they were the right ones. The partnership, however, that
I am most thankful for, was the one I have with my partner, David Chiu. Thank you
for the amazing memories, and I look forward to many more. And finally, this would
not be at all possible, without the mentorship of my advisor, Jennifer Raymond, who
provided guidance not only on scientific issues, but life issues as well. After every
office meeting with Jennifer, regardless of the topic discussed, I always left a little
more inspired, a little more excited about the scientific endeavors we were
undertaking.
viii
Table of Contents
Chapter 1 Background
1.1 Motor learning . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . 1
1.2 Cerebellum-dependent motor learning . . . . . . . . . . . . . 2
1.3 Motor learning in the vestibule-ocular reflex (VOR) . . . . 3
1.4 Two influential but competing VOR learning models . . . . 5
1.5 Challenges to the Marr-Albus-Ito and Miles-Lisberger
Model for motor learning in the VOR . . . . . . . . . . . . . . . . . 9
1.6 Multiple mechanisms underlie motor learning in the
VOR . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 The next step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Chapter 2 Effects of visual target and background on motor learning in the VOR
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2.1 Training conditions with different combinations of
target and background . . . . . .. . . . . . . . . . . . . . . . . 17
2.2.2 Retinal slip of the target and backgro. . . . . . . . . . . 18
2.2.3 Effects of target and background on tracking eye
movements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.4 Effects of target and background on motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.5 Phase changes induced by consistent and conflicting
training conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2.6 Motor learning during training conditions at 5 Hz . 27
2.2.7 Effects of target and background on floccular
Purkinje cell simple spikes .. . . . . . . . . . . . . . . . . . . 29
ix
2.2.8 Effects of target and background on floccular
climbing fiber activity . . . . . . . . . . . . . . . . . . . . . . 31
2.2.9 Purkinje cell simple spikes and climbing fiber
signals modulate independently. . . . . . . . . . . . . . . 33
2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.1 Learning cues during the induction of motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.2 Retinal slip velocity and tracking eye movements . 36
2.3.3 Behaviorally-relevant and foveal/parafoveal stimuli 37
2.3.4 Effects of target and background motion on
putative neural instructive signals for motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.5 Climbing fibers do not encode net image motion
on the retina . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.6 Modulation of Purkinje cell simple spikes is
independent of climbing fiber signals . . . . . . . . . . . 42
2.4 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Chapter 3 Elimination of climbing fiber instructive signals during motor learning
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.1 Standard training stimuli elicit instructive signals
In climbing fibers . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.2 Novel training stimuli eliminate instructive signals
in climbing fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.2.3 Instructive signals carried by Purkinje cell simple
Spikes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.4 Learning in the absence of instructive signals in the
climbing fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
x
3.2.5 Learning in the absence of instructive signals in the
simple spikes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.6 Purkinje cell simple spike and climbing fiber signals
together predict behavioral changes . . . . . . . . . . . . 66
3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.1 VOR circuit physiology and Marr-Albus-Ito Model 69
3.3.2 Could another population of climbing fibers induce
changes in VOR gain? . . . . . . . . . . . . . . . . . . . . . 70
3.3.3 Other instructive signals for cerebellum-dependent
motor learning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.3.4 Multiple instructive signals during cerebellum-
dependent motor learning. . . . . . . . . . . . . . . . . . . . . 75
3.4 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Chapter 4 Specificity and generalization of motor learning in the VOR
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.2.1 Generalization of learned changes in VOR gain . . . 82
4.2.2 Can VOR gain changes at the non-training
frequencies be explained by a combination of
climbing fiber and Purkinje cell signals? . . . . . . . . 84
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3.1 Neuronal population coding schemes to achieve
generalization and specificity . . . . . . . . . . . . . . . . . 88
4.3.2 Generalization of VOR gain increases versus
decreases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3.3 Patterns of generalization for high versus low
frequency training . . . . . . . . . . . . . . . . . . . . . . . . 89
xi
4.3.4 Generalization and specificity during motor
learning in the VOR. . . . . . . . . . . . . . . . . . . . . . . . 89
4.4 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Chapter 5 Conclusions and future directions . . . . . . . . . . . . 93
Appendix
Individual climbing fiber and Purkinje cell simple spike responses
for each training stimulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Bibilography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
xii
List of tables
Table 2.1 Visual vestibular stimuli used to induce motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . 18
Table 3.1 Statistical evaluation of climbing fiber and
Purkinje cell simple spike responses to the training
stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Table 3.2 Climbing fibers carrying “gain decrease” versus
“gain increase” signals during x0T/x2BG training
did not differ in any other property . . . . . . . . . . . 73
xiii
List of figures
Figure 1.1 General motor learning scheme. . . . . . . . . . . . . . 1
Figure 1.2 VOR circuitry. . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Figure 1.3 Purkinje cell and climbing fiber response during x0
and x2 stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Figure 1.4 Marr-Albus-Ito hypothesis. . . . . . . . . . . . . . . . . . . 6
Figure 2.1 Two independently moving visual stimuli can
create multiple image motions on the retina . . . . . 17
Figure 2.2 Retinal slip of the target and background during
visual-vestibular stimuli used to induce motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 2.3 Target motion is correlated with gaze velocity. . . . 20
Figure 2.4 Target slip velocity is correlated to gaze velocity . . 21
Figure 2.5 Motor learning in the VOR induced by coherent
motion of target and background . . . . . . . . . . . . . . 22
Figure 2.6 Motor learning in the VOR is correlated to target
slip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Figure 2.7 Conflicting background motion impairs motor
learning in the VOR . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 2.8 Target slip mediates effects of conflicting
Background motion on motor learning in the VOR . 25
Figure 2.9 Changes in VOR phase induced by consistent and
conflicting training conditions . . . . . . . . . . . . . . . . 26
Figure 2.10 Motor learning in the VOR is correlated to target
slip velocity but not gaze velocity at 5 Hz . . . . . . . 28
Figure 2.11 Target and background slip velocities during
training stimuli at 5 Hz . . . . . . . . . . . . . . . . . . . . 29
xiv
Figure 2.12 Purkinje cell simple spike responses are well-
correlated to gaze . . . . . . . . . . . . . . . . . . . . . . . 30
Figure 2.13 Changes in Purkinje cell simple spike responses
during conflicting conditions are correlated to eye
movement responses . . . . . . . . . . . . . . . . . . . . . 31
Figure 2.14 Climbing fiber responses to consistent and
conflicting training conditions . . . . . . . . . . . . . . 32
Figure 2.15 Target slip mediates effects of conflicting
background motion on climbing fiber responses . . 33
Figure 2.16 Training conditions using target and background
motion elicit different combinations of Purkinje
cell and climbing fiber signals . . . . . . . . . . . . . . . 34
Figure 3.1 VOR circuit and Marr-Albus-Ito hypothesis for
VOR learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure 3.2 Climbing fiber responses to standard and novel
training stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Figure 3.3 Neural responses and changes in VOR gain elicited
by the x2T/x0BG training stimulus . . . . . . . . . . . . 59
Figure 3.4 Average firing rate of climbing fibers did not vary
across training stimuli . . . . . . . . . . . . . . . . . . . . . . 60
Figure 3.5 Purkinje cell simple spike responses during
training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Figure 3.6 In the absence of instructive signals in the
climbing fibers, learning was correlated with
simple spike responses during training . . . . . . . . . . 64
Figure 3.7 Purkinje cell simple spike and climbing fiber
signals predict learned behavioral changes . . . . . . 67
Figure 3.8 Climbing fiber responses in non-HGVPs during
x0T/x0BG and x0T/x2BG training . . . . . . . . . . . . . 71
xv
Figure 3.9 Climbing fibers with “gain decrease” and “gain
increase” responses during x0T/x2BG training
had similar responses during standard training
stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Figure 3.10 Most of the learning occurs early in the training
session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Figure 4.1 Motor learning in the VOR at non-training
frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Figure 4.2 Generalization indices for consistent and
conflicting training conditions . . . . . . . . . . . . . 83
Figure 4.3 Climbing fiber signals are correlated to learning
at non-training frequencies in Monkey L . . . . . . 85
Figure 4.4 Purkinje cell signals are correlated with learning
at non-training frequencies . . . . . . . . . . . . . . . . . 86
Figure 4.5 Motor learning in the VOR at non-training
frequencies can be predicted by a linear
combination of Purkinje cell and climbing fiber
responses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Figure 4.6 Purkinje cell signals contribute more to VOR gain
changes at non-training frequencies . . . . . . . . . . 87
1
Chapter 1 Background
1.1 Motor learning
Motor learning is the process of improving the smoothness and accuracy of
movements. While motor learning is apparent for complicated movements such as
riding a bicycle, or swinging a bat to hit a ball, it is also necessary for calibrating
simple movements such as reflexes, as parameters of the body and environment
change over time.
One approach to understanding motor learning is to trace the effects of stimuli that
induce motor learning through different levels of analysis (Fig. 1.1). At the behavioral
level, non-optimal movements
result in sensory and motor cues
that can be used as “error”
signals to induce learning.
These behavioral error signals
are translated into specific
patterns of neural activity in the
brain. These patterns of activity
act as neural instructive signals
to selectively induce cellular and
synaptic plasticity events, which
in turn alter signal processing in
the learning circuit so that the
behavior is adaptively modified.
In order to link system-level
analyses of learning with cellular analyses of plasticity, a circuit-level understanding
of which patterns of neural activity induce synaptic plasticity in the awake, behaving
animal is needed. This knowledge has been difficult to attain since in most learning
tasks, the precise neural circuits underlying learning have not been delineated, and
therefore the neural instructive signals impinging on a site of plasticity involved in
2
learning are mostly not known. Thus, a first step in understanding what patterns of
neural activity trigger plasticity in vivo, is to identify which population of neurons in a
learning circuit carries the instructive signal used to guide learning. The specific aim
of this thesis is to determine which neurons carry the instructive signals that trigger
plasticity during one specific form of motor learning, adaptation of the vestibulo-
ocular reflex (VOR).
1.2 Cerebellum-dependent motor learning
The cerebellum is a structure critical for motor learning in vertebrates. One of the
most striking features of the cerebellum is its basic architecture, described as being
crystalline due its numerous, repeated modules, each containing the same few cell
types connected in the same manner[9]. While different regions within the cerebellum
receive different inputs, and project to different outputs, the overall uniformity of the
cerebellum suggests that it processes these signals in similar ways.
One of the first and most influential models of cerebellar function was proposed by
Marr and Albus[2, 4]. In their model, mossy fibers via granule cells and their
projections, the parallel fibers, provide sensory and motor input to the cerebellar
Purkinje cells. The Purkinje cells provide the lone output signal of the cerebellum, and
its activity drives specific movements. Over a hundred thousand parallel fibers
synapse onto a single Purkinje cell. The Marr-Albus model proposed that changes in
the strength at specific parallel fiber-Purkinje cell synapses stored stimulus response
associations by linking the inputs with appropriate motor outputs[2, 4].
Another striking feature of the cerebellum is the climbing fiber input —each Purkinje
cell receives input from just one climbing fiber, and the climbing fiber forms a very
strong synapse onto the Purkinje cell in that each presynaptic spike triggers a
postsynaptic spike [10]. However, despite this incredibly powerful connection, the
climbing fiber input makes a small contribution (1-3 Hz) to the overall Purkinje cell
3
firing rate (>60 Hz), suggesting that the function of the climbing fiber is for something
other than ordinary signal transmission.
Marr and Albus proposed that the function of the climbing fiber is to provide an
instructive signal to guide changes at the parallel fiber to Purkinje cell synapse, but
they disagreed on the precise role of the climbing fiber. Marr believed that it was a
positive reinforcer, strengthening parallel fiber to Purkinje cell synapses when the
Purkinje cell output was correct, whereas Albus believed the climbing fibers to carry
an error signal and weakened synapses when the output was incorrect[2, 4].
Consistent with Albus’ prediction, it was later discovered that electrical stimulation of
climbing fibers induced a decrease in synaptic strength, called cerebellar long-term
depression (LTD), at parallel fibers that are active simultaneously[3].
1.3 Motor learning in the vestibulo-ocular reflex (VOR)
One model system to study cerebellum-dependent motor learning is the vestibulo-
ocular reflex (VOR). The VOR system offers a key advantage over other types of
learning: much of the neural pathway underlying the VOR and VOR learning is well
characterized, allowing us to link events occurring in particular neurons and synapses
to circuit-level events during the induction and expression of learning[11].
The VOR is a reflexive eye movement that stabilizes images on the retina by using
vestibular signals to generate compensatory smooth eye movements in the opposite
direction from head motion. The VOR functions to improve vision, but the
performance of the VOR is measured in the dark to isolate eye movements driven by
the vestibular stimuli from eye movements driven by visual stimuli. The performance
of the VOR is characterized by its gain, which is defined as the ratio between eye and
head velocities, and its phase, which reflects the relative timing of eye and head
movements when tested using sinusoidal vestibular stimuli.
4
When the VOR is not well calibrated, head movements result in image motion on the
retina, which impairs visual acuity. When image motion is persistently present during
head movements, a form of motor learning, known as VOR adaptation, gradually
corrects the reflex
and restores proper
function. In the
laboratory, this form
of motor learning is
typically induced by
pairing passive head
movements (e.g.
head-fixed rotation
on a turn table) with
movement of a
single, large visual stimulus. If the visual stimulus moves exactly with the head, a
decrease in the gain of the VOR is induced. Such training stimuli are called “x0”,
because the VOR gain required to stabilize images is zero. If, instead, the visual
stimulus moves exactly opposite to head motion, an increase in VOR gain is induced.
Such training stimuli are called “x2” because the ideal VOR gain would be two (eye
velocity equal to twice head velocity).
The VOR circuitry has been well described (Fig. 1.2)[11]. Neural signals for head
movements are encoded in the vestibular nerve and are processed in parallel pathways.
In the direct pathway, the vestibular afferents project to the vestibular nucleus (VN)
that sends projections to premotor and motor neurons (MN) that control eye
movements. A second pathway is an inhibitory side loop into the flocculus and
ventral paraflocculus of the cerebellum. This pathway conveys vestibular and eye
movement signals via mossy fibers (MF) and granule cells (GrC). The axons of the
granule cells form parallel fibers (PF) that make direct excitatory connections with
Purkinje cells (PC). In addition, Purkinje cells receive excitatory inputs from climbing
5
fibers (CF) that arise from neurons in the inferior olive (IO) and which carry visual
signals related to image motion on the retina. The Purkinje cells are the sole output
neuron of the cerebellum and inhibit VOR interneurons called flocculus target neurons
(FTN) in the vestibular nucleus.
1.4 Two influential but competing VOR learning models
There are two
predominate
hypotheses regarding
the neural learning
rules implemented in
the VOR circuit
during the induction
of learning. Each
proposes that a
particular population of neurons carries the key instructive signal that guides changes
in the circuit. One hypothesis, proposed by Masao Ito (1972) based on the theoretical
works of Marr (1969) and Albus (1971), suggests a learning rule whereby climbing
fibers carry the instructive signals[2, 4, 12]. The other hypothesis, proposed by Miles
and Lisberger (1981), suggests a learning rule whereby signals in Purkinje cells
provide the instructions controlling changes in the VOR circuit[1].
Previous recordings from climbing fibers and Purkinje cells during the induction of
learning provide some support for each of these hypotheses. Climbing fibers and
Purkinje cells each exhibit different responses when sinusoidal head motion is paired
with a visual stimulus moving in a way that increased (x2) versus decreased (x0) VOR
gain[5, 6, 13](Fig. 1.3).
Climbing fibers in the floccular complex are driven by contraversive image motion
and inhibited by ipsiversive image motion. During the x0 stimulus, the climbing
6
fibers increased their firing rate during contraversive head motion (Fig. 1.3), and
decreased their firing rate during ipsiversive head motion. During the x2 stimulus, the
climbing fibers increased their firing rate during ipsiversive head motion, and
decreased their firing rate during contraversive head motion. This difference in the
timing of the climbing fiber firing relative to head movement, (and therefore activity
of vestibular afferents) during x0 versus x2 training reflects the difference in the
timing of contraversive image motion. Cellular studies have indicated that parallel
fibers active around the time of increased climbing fiber activity should be
depressed[3, 14],
whereas parallel fibers
active during decreased
climbing fiber activity
should be potentiated.
Therefore, according to
the Marr-Albus-Ito
model, during x0
training, the climbing
fiber response would
induce selective LTD
of vestibular parallel
fibers that fire during
contraversive head
movements and
selective long-term
potentiation (LTP) of
vestibular parallel
fibers that fire during ipsiversive head movements (Fig. 1.4). On the other hand, the
x2 training should induce selective LTD of parallel fibers that fire during ipsiversive
head movement and LTP of parallel fibers that fire during contraversive movement.
For x0 training, this selective induction of LTD at a particular set of parallel fibers
7
would alter the Purkinje cell simple spike output, causing the Purkinje cells to fire
more in-phase with VOR interneurons in the vestibular nucleus after training as
compared with before training (Fig. 1.4). Since Purkinje cells inhibit VOR
interneurons, the altered Purkinje cell output should tend to cancel the response in the
vestibular interneurons and result in a decrease in VOR gain. In contrast, during x2
training, the selective induction of LTD would alter the Purkinje cell simple spike
output, causing the Purkinje cell to fire more out-of-phase with VOR interneurons
after training, causing an increase in VOR gain (Fig. 1.4).
In summary, the Marr-Albus-Ito hypothesis predicts that climbing fiber activity during
the induction of learning lead to selective LTD and LTP of vestibular parallel fiber
synapses onto Purkinje cells. The engagement of these plasticity events, in turn, alter
Purkinje cell responses during the performance of the VOR in such a way that can
account for the changes in VOR gain.
In contrast to the Marr-Albus-Ito hypothesis, Miles and Lisberger proposed that
activity in the Purkinje cells serve as the neural instructive signal guiding motor
learning in the VOR. Previous recordings of Purkinje cells have provided some
support for the Miles and Lisberger model[15]. Floccular Purkinje cells carry signals
that reflect the eye movements the animals makes to track the visual target[16]. Since
the tracking eye movements were different for x0 versus x2 training, the Purkinje cell
signals were also different. Specifically, during conditions that decrease VOR gain
(x0), there was elevated Purkinje cell activity during ipsiversive head motion, whereas
Purkinje cell activity was reduced during ipsiversive head motion during conditions
that increased (x2) VOR gain (Fig. 1.3). These observations suggest an alternative
learning rule for VOR learning: if increased Purkinje cell activity coincided with
activity in the vestibular afferents during ipsiversive head movement, then synaptic
changes is induced to decrease the gain of the VOR, whereas decreased Purkinje cell
activity coinciding with activity in the vestibular afferents during ipsiversive head
movements would lead to synaptic changes that would increase VOR gain. Since
8
floccular Purkinje cells project to the vestibular nucleus, these synaptic changes are
predicted to occur at the level of the synapses between vestibular afferents and
neurons in the vestibular nucleus (see lightening bolt, Fig. 1.2).
In the Miles and Lisberger hypothesis, Purkinje cell activity during the induction of
learning guide plasticity in the vestibular nucleus. The engagement of these plasticity
mechanisms, in turn, alter the responses of vestibular interneurons during performance
of VOR in such a way that can account for the changes in VOR gain.
Several studies have reported learning-related changes in Purkinje cell activity during
VOR performance [13, 15, 17]. According to Marr-Albus-Ito, these altered signals
result directly from LTD or LTP of synapses onto Purkinje cells from parallel fibers
carrying vestibular input[12]. Miles and Lisberger, on the other hand, suggested that
the altered signals in the Purkinje cells during the VOR reflect altered input to the
cerebellum from an efference copy pathway[1]. Purkinje cell output can influence eye
movements, but the Purkinje cells also receive input from mossy fiber pathways
carrying signals related to eye movement[18, 19]. Thus, there is a feedback loop.
Given this feedback loop, changes almost anywhere in the circuit could be expressed
at any given site. In particular, Miles and Lisberger suggested that learning caused
changes in the vestibular nucleus, which resulted in an altered eye movement
command during the VOR, which in turn was conveyed to the Purkinje cells as an
efference copy signal.
In the past, testing the hypotheses regarding whether climbing fibers or Purkinje cells
carry the instructive signals that guide VOR leaning was conducted by comparing
signals in climbing fibers and Purkinje cells during one stimulus that increased VOR
gain (x2) and one stimulus that decreased VOR gain (x0). Since the patterns of neural
activity present during these stimuli were consistent with both hypotheses, more data
are needed to test each hypothesis individually, and this dissertation provides such
data.
9
In summary, these hypotheses make distinctly different predictions regarding the role
of the cerebellum in motor learning. First, each makes a different prediction on which
neurons carry the instructive signal guiding motor learning. Marr-Albus-Ito contends
it is the climbing fibers, whereas Miles and Lisberger argue it is the simple spike
output of the Purkinje cells. Second, each hypothesis makes a different prediction
regarding the site of plasticity. Marr-Albus-Ito proposes that the site of memory is at
the parallel fiber to Purkinje cell synapse in the cerebellar cortex. Miles and Lisberger
assume the changes to be in the vestibular nucleus. Last, each hypothesis makes
different predictions on the source of learning-related changes in Purkinje cell simple
spike activity during VOR performance after learning has occurred. Marr-Albus-Ito
believed that the changes in Purkinje cell are the direct result of plasticity in the
cerebellar cortex. Miles and Lisberger, on the other hand, believed that the changes in
the Purkinje cell simple spike during VOR performance are reflective of the efference
copy of the altered eye movement command created in the vestibular nucleus.
1.5 Challenges to the Marr-Albus-Ito and Miles-Lisberger model for motor
learning in the VOR
There are several challenges to each of these influential models as the model for motor
learning in the VOR. The biggest challenge to the Marr-Albus-Ito hypothesis is the
observation that after motor learning is induced in the VOR, changes in signal
processing by the Purkinje cells cannot be predicted by the neural learning rules
governing cerebellar LTD[16]. The Marr-Albus-Ito hypothesis predicts that the
learning-related changes observed in Purkinje cell signals during VOR after learning
result directly from LTD or LTP of synapses onto Purkinje cells from parallel fibers
carrying vestibular input[12]. However, the altered responses may be attributed to
either a change in the vestibular or efference copy input to the Purkinje cells. To
assess how much of the altered response was attributed to just the vestibular inputs, an
additional test stimulus besides the VOR was used. The test was to record Purkinje
cells while the animal cancelled its VOR (VORC) by tracking a target that moved
10
exactly with him during sinusoidal head turns. During VORC, the eye velocity was
close to zero throughout the head movement, so there should be little to no firing rate
modulation in the efference copy pathway, and therefore the response in the Purkinje
cell during VORC was taken as a measure of the strength of its vestibular inputs.
Several studies using this method revealed that learning-related changes in the
vestibular sensitivity were in the wrong direction to account for the change in VOR
gain [13, 15, 20], and in the wrong direction from what would be predicted if the
climbing fiber responses observed during induction of learning produced LTD in the
vestibular parallel fibers active simultaneously.
The major challenge to the Miles and Lisberger hypothesis is that adaptive gain
changes can occur under some conditions where Purkinje cell signals cannot
discriminate between stimuli that increase and decrease VOR gain, and therefore
provide no useful instructive signals to guide the appropriate plasticity[21]. Typically,
VOR adaptation is induced with 0.5 Hz sinusoidal x0 and x2 visual-vestibular stimuli,
and during these stimuli, both climbing fibers and Purkinje cells carry potentially
useful instructive signals (Fig. 1.3). VOR adaptation can also be successfully induced
with 5 Hz sinusoidal x0 and x2 visual-vestibular stimuli[22]. However, recordings of
Purkinje cells and climbing fiber signals during the induction of learning at 5 Hz,
revealed that climbing fibers carried information about the direction of learning, but
Purkinje cells did not[6]. This indicated that Purkinje cells cannot provide the
instructive signals that guide learning induced with high frequency stimuli.
1.6 Multiple mechanisms underlie motor learning in the VOR
In the last several years, evidence using in vitro preparations have demonstrated many
plasticity mechanisms in the cerebellum and related circuitry, suggesting a wide
variety of plasticity mechanisms are available to be implemented within cerebellum-
dependent motor learning tasks[23]. One possible resolution to the challenges faced
by the Marr-Albus-Ito and Miles-Lisberger model is both these models as well as
other mechanisms contributes to motor learning in the VOR.
11
A model that incorporates distributed learning and multiple plasticity mechanisms was
first proposed to explain why the vestibular sensitivity of Purkinje cells changes in the
wrong direction to account for the change in VOR gain. Lisberger and Sejnowski
proposed that having changes only in the vestibular nucleus would cause unstable eye
movements due to a positive feedback loop within the VOR circuit[24]. Since
Purkinje cells can drive eye movements, and signals related to these eye movement
commands then feed back to drive more Purkinje cell activity (Fig. 1.2), unstable eye
movements would result after learning-related changes in the vestibular nucleus have
been induced. However, additional changes in the cerebellar cortex, namely the
observed changes in vestibular sensitivity are in the appropriate direction to maintain
proper movement dynamics.
The gain of the VOR can be adaptively increased or decreased, and the original Marr-
Albus-Ito hypothesis attributes a single plasticity mechanism (cerebellar LTD) at
different sets of parallel fiber-Purkinje cell synapses to account for VOR gain
increases and decreases. If the model was accurate, then increases and decreases in
VOR gain should have similar properties stemming from their shared mechanism.
However, several studies have demonstrated asymmetries in the behavioral properties
of VOR gain increases and decreases, suggesting that they depend on different
plasticity mechanisms. For example, increases in VOR gain passively decayed
quicker than decreases in VOR gain[25], increases in VOR gain can be actively
reversed more readily than decreases in gain[26], and increases in gain generalize less
than decreases[8], when learning is measured in a context different from one used for
training. In addition, several pharmacological and genetic studies have also
demonstrated that VOR gain increases and decreases are dependent on different
cellular mechanisms—inhibiting nitric oxide receptors, a component in the induction
of cerebellar LTD, impairs gain increases but not decreases in goldfish[27]. And mice
lacking CAMKIV, a required molecule for long-lasting LTD significantly impairs
long-term retention of VOR gain increases, but not decreases[28].
12
1.7 The next step
During the last several decades, an attempt to discriminate between the Marr-Albus-
Ito and Miles-Lisberger hypothesis has dominated research on motor learning in the
VOR. However, with recent in vitro and in vivo evidence strongly suggesting a role of
multiple mechanisms, the question now shifts from, “Which plasticity mechanism is
correct?” to “How may a single plasticity mechanism contribute to motor learning in
the VOR?”
The goal of the this thesis project is to help elucidate how plasticity mechanisms, as
proposed by the Marr-Albus-Ito and Miles-Lisberger models, may contribute to motor
learning in the VOR. The overall experimental goal is to correlate the presence and
absence of motor learning in the VOR with the presence and absence of neural
instructive signals carried by the climbing fibers and Purkinje cells during the
induction of learning. By correlating learning with the available neural instructive
signals, I hope to address fundamental questions regarding the necessity, and possibly
sufficiency of each neural instructive signal to learning. Answering these fundamental
questions will yield powerful insights into the learning algorithms utilized by the
cerebellum, and its learning capacity.
The following thesis is divided into three major chapters. The first chapter will focus
on the effects of pairing complex visual stimuli with head movements to induce motor
learning in the VOR. These experiments were designed to study how systematic
changes in the sensory and motor stimuli known to guide motor learning in the VOR
affect the putative neural instructive signals and motor learning. The second chapter
will focus on a specific subset of the training conditions that allowed us to test the
necessity of the climbing fiber instructive signals for motor learning. In addition, this
chapter examines how Purkinje cell and climbing fiber instructive signals may
together account for learning. Finally, the last chapter examines how training with
complex visual stimuli can affect the generalization of motor learning in the VOR,
13
specifically examining whether learning-related changes at the training frequency may
be expressed at non-training frequencies. If learning can be expressed at non-training
frequencies, this chapter examines whether Purkinje cell and climbing fiber instructive
signals may account for these changes.
14
Chapter 2 Effects of visual target and background on motor learning in the VOR
2.1 Introduction
The cerebellum plays a critical role in motor learning, the process of improving the
smoothness and accuracy of movements. The dominant theory over the past several
decades postulated that the climbing fiber input to the cerebellum encodes an error in
the motor output, and that this neural signal guides plasticity in the cerebellum in such
a way that the error in motor output is subsequently reduced[2, 4, 12]. Several studies
have provided evidence for this hypothesis, namely that under training conditions that
induce motor learning, climbing fiber activity is well-correlated to errors in motor
performance[6, 29]. These studies, however, were mostly carried out in well-
controlled laboratory settings, where learning was induced with robust, well-defined
errors. In contrast, in the real-world, motor learning typically occurs under a more
complex learning environment, where multiple, and at times conflicting errors in
motor performance, may be present. The effects of multiple errors on learning and
neural instructive signals are largely unknown.
The oculomotor system is an ideal motor system for laboratory studies because eye
movements can be easily quantified, and the neural pathways underlying the behavior
are well understood[11]. The main goal of any oculomotor system is to stabilize
images on the retina, and if eye movements are not well-calibrated, retinal slip or
image motion occurs, and cerebellum-dependent motor learning is induced to adjust
the eye movement response so the retinal slip is reduced. Climbing fibers in the
floccular complex, the part of the cerebellum thought to mediate this form of motor
learning, carry information related to retinal slip[5]. Previous experiments induced
motor learning with coherent motion of a visual stimulus on the retina[13, 16, 22].
However, this simplified condition is not reflective of the real-world, where animals
are likely to encounter multiple image motions. For example, tracking a small visual
target would result in opposing image motions of the target and background on the
retina. Or during motion parallax, the differential displacement of near and far objects
15
may result in opposing image motions on the retina. Little is known how complex,
opposing image motions may affect oculomotor learning and putative neural
instructive signals.
We examined the effects of a complex learning environment on motor learning and
putative neural instructive signals, using a cerebellum-dependent motor learning
paradigm, motor learning in the vestibulo-ocular reflex (VOR). The VOR stabilizes
images on the retina during head movements by generating compensatory eye
movements in the opposite direction of the head motion. When VOR is not
appropriately calibrated, retinal slip is persistently present during head movements,
and motor learning is induced to adjust the amplitude and timing of the eye movement
responses. Motor learning in the VOR can be induced in the laboratory by pairing
head movements with motion of single coherent visual stimulus. Neural recordings
during these experiments have demonstrated that climbing fiber activities carry
sensory signals related to retinal slip, and Purkinje cell activities carry motor signals
related to eye movements[18, 30]. Both these signals have been suggested to be
neural instructive signals guiding plasticity during the induction of motor learning in
the VOR.
The training condition with a coherent visual stimulus has been used successfully in
understanding the basic characteristic of the neural circuit underlying VOR learning.
Here we studied motor learning in the VOR with paradigms more complex than
traditional training paradigms to determine the effects of multiple, conflicting image
motions on the neural instructive signals and motor learning in the VOR. In
particular, we induced learning with the simultaneous presentation of two visual
stimuli moving independently-- a small visual target, and a large-field visual
background. The use of two visual stimuli moving independently permitted us to
design training conditions where the image motions of the target and background are
in the same direction on the retina, producing "consistent" image
motions. Alternatively, we can design training conditions where the image motion of
16
the target and background are in opposite direction on the retina, therefore producing
"conflicting" image motions. We recorded climbing fiber and Purkinje cell simple
spikes during the induction of learning with training conditions eliciting consistent and
conflicting image motions. An understanding of the effects of multiple image motions
on the putative neural instructive signals may provide insights on how learning is
selectively induced under more complex environments.
17
2.2 Results
2.2.1 Training conditions with different combinations of target and background
In our experiments, we paired head motion with motion of two visual stimuli—a
visual target (0.5˚ diameter) that the animal was trained to track, and a large visual
background (20˚ x 30˚). The target and background moved independently from each
other. Figure 2.1 illustrates
two different combinations of
target and background. During
these training conditions, the
head was moving with a
sinusoidal velocity profile at
0.5 Hz ± 10 ˚/s. In both the
x0T/x0BG and x0T/x2BG
training condition, the visual
target and head were moving
together (brown and purple
lines). Since the target
motions were the same during
both training conditions, the
eye movements (blue lines)
were also comparable. However, the monkeys were not perfectly tracking the target,
resulting in small, but significant retinal slip of the target (typically less than 3 ˚/s,
purple lines). In the illustrated pair of training conditions, the background was moving
differently. In the x0T/x0BG training stimulus, the background was moving exactly
the same as the target; whereas during the x0T/x2BG training stimulus, the
background was moving with the same speed as the target, but in the opposite
direction (compare gray and purple lines labeled target and background). Thus, the
resulting background slip during each condition was remarkably different.
18
We tested 22 different
target and background
combinations in
Monkey L, 16
different combinations
in Monkey C, and 13
different combinations
in Monkey E (Table
2.1, Methods). In
each case, the
vestibular stimulus
was sinusoidal head
movement at 0.5 Hz,
± 10 ˚/s, with the
exception of 6 training
stimuli in monkey L
that were tested at 5
Hz, ± 10 ˚/s. (Table 2.1, Methods).
2.2.2 Retinal slip of the target and background
In our experiments, the animals were trained to track the visual target while ignoring
motion of the background. The presence of two visual stimuli moving independently,
created two different types of retinal image motion—image motion of the visual target
and image motion of the background. The target slip velocity was calculated by
determining the difference between gaze velocity and target velocity. Gaze velocity
was the linear addition of the eye and head velocities, and reflected where the animal
was looking in earth-fixed coordinates. Target slip velocity was a direct measure of
how well the animal tracked the target. The better an animal tracked, the smaller the
target slip velocity on the retina. In Figure 2.2, the target slip and background slip
during peak head velocity were plotted for each 0.5 Hz training stimulus.
19
The slip velocities reported in this chapter are the amplitude of target and background
slip that occurred during peak head velocity (see Methods). If peak retinal slip
occurred in the same direction as head motion, then the retinal slip velocities were
assigned positive values. If peak retinal slip occurred opposite direction of head
motion, then the retinal slip velocities were assigned negative values. If both the
target and background slip occurred in the same direction on the retina relative to head
motion, the conditions were called “consistent” conditions (white quadrants, Fig. 2.2).
If a training condition elicited target slip on the retina opposite in direction of
background slip during head motion, then the stimulus was called a “conflicting”
condition (blue quadrants, Fig. 2.2).
2.2.3 Effects of target and background on tracking eye movements
In our many training conditions, we paired head movements with target motion to give
us a range of gaze velocity responses. Recall that gaze velocity is the linear addition
of head and eye velocities, and reflects where the animals are looking in earth fixed
coordinates. Since animals were trained to track a visual target, target motion was
correlated to gaze velocity. Figure 2.3 plots the amplitude of gaze velocity during
peak head velocity (see Methods) during training conditions with different target
20
motion. Positive values were
assigned if gaze was in the same
direction as head movements;
whereas negative values were
assigned to gaze in the opposite
direction of head movements. For
example, the x0T/x0BG training
stimulus paired a target and
background moving in the same
speed and direction as the head.
Therefore, the gaze of a monkey
tracking the target would be in the
same direction as the head and
was assigned a positive value. On the other hand, the x2T/x2BG training condition
paired a target and background both moving in the opposite direction but at the same
speed as the head. In this training condition, the gaze of a monkey tracking a target
would be in the opposite direction of head motion, and therefore assigned a negative
value. During x1T/x1BG, the visual stimulus was stationary relative to the world, so a
monkey tracking the target maintained central fixation throughout head rotation, and
no gaze velocity was elicited.
The target slip elicited by each training condition was correlated to gaze velocity.
Figure 2.4 plots the relationship between gaze velocity and target slip velocity induced
by each training condition. In general, as gaze velocity increased, target slip velocity
increased. Since gaze velocity was determined by target motion (Fig. 2.3), and the
target slip reflected how well the animal was tracking the target, the relationship
between gaze velocity and target slip velocity demonstrate that as the target moved
faster, animals became less effective at tracking the target.
21
In a subset of our training conditions, the background slip was in the opposite
direction of the target slip on the retina. During these “conflicting” conditions,
tracking eye movements were affected (gray diamonds, Fig. 2.4). In Monkey L, the
presence of a conflicting background decreased target slip velocity, revealing that
Monkey L’s tracking eye movements became more effective in the presence of
conflicting background motion. In Monkey C, the presence of a conflicting
background had an opposite effect to that on Monkey L, in that target slip velocity
increased. In Monkey C, the presence of conflicting background motion made its
tracking eye movements less effective. In Monkey E, the presence of conflicting
background motion had no effect on tracking eye movements. Although the presence
of a conflicting background affected tracking eye movements in Monkey L and
Monkey C, the absolute effects were small, in that the gaze velocity during conflicting
training conditions were similar to gaze velocity elicited by consistent training
conditions.
2.2.4 Effects of target and background on motor learning in the VOR
We tested the effectiveness of training conditions at inducing motor learning in the
VOR. Motor learning in the VOR was induced by presenting a training stimulus for
either 1 h (Monkey E) or 2 h (Monkey C and Monkey L). VOR learning was assessed
22
comparing eye movement responses to head movements in complete darkness before
and after training (see Methods for more details).
For training conditions that elicited consistent retinal slip of the target and
background, learned changes in VOR gain (∆ VOR gain ) were well-correlated with
target motion, as shown in previous experiments [6, 31, 32]. Training that paired
target and background slip in the same direction as head movements (x0T, x0.5T
conditions) induced a decrease in VOR gain, whereas training conditions that paired
target and background slip in the opposite direction as head movements (x2T, x1.5T
conditions) induced an increase in VOR gain (Fig. 2.5)
The critical difference between consistent training conditions that induced a gain
increase versus gain decrease is the direction of target slip on the retina. If the target
slip was in the same direction as head motion, then the VOR gain was too small to
keep the target on the retina, and a gain increase was induced, whereas if the target
slip was in the opposite direction as head motion, then the VOR gain was too large,
and a gain decrease was induced. To evaluate if learning was affected by the
amplitude of target slip, we plotted the amount of learning according to the target slip
velocities elicited by each consistent training condition (Fig. 2.6).
23
A linear regression analysis demonstrated that learning was well-correlated to target
slip velocities (R2 = 0.94 for Monkey L, R2 = 0.97 for Monkey E, R2 = 0.50 for
Monkey C). Monkey C had variable data, particularly for x2T/x2BG training
condition, that either elicited moderate gain increases or no change in VOR gain. This
consistent training condition has been previously used in many studies and typically
induced a robust increase in VOR gain as demonstrated by Monkey L and Monkey E
[22, 31]. Monkey C’s inability to effectively learn may be due to its ability to track
the visual stimulus throughout the 2 h of training. If the animal looks away from the
moving visual stimulus during training, it will be receiving at x1 stimulus, which can
trigger changes within the circuit to offset changes triggered by either x0 or x2
training stimulus. Interestingly, Monkey C had a harder time tracking a x2 stimulus
than a x0 stimulus (note larger error bars for Monkey C for target slip during negative
gaze velocities that correspond to gain up stimuli in Figure 2.4), and it is the x2
consistent training conditions that had the most variable amounts of learning.
For each animal, we fitted the data points with a linear regression and derived the
regression coefficient which represented the rate of learning for each animal (Fig. 2.6,
black lines). The regression coefficients were -30.3 %/°/s for Monkey L, -16.2 %/°/s
for Monkey E, and -7.6 %/°/s for Monkey C. The different rates of learning most
likely reflect intrinsic differences in learning capacity, as well as differences in
training conditions (e.g. 1 h training for Monkey E, 2 h training for Monkey L) which
were optimized to each animal’s work abilities.
24
To combine the animals’ datasets for further analyses, we normalized each animal’s
learning data by dividing the ∆ VOR gain by the learning rate as calculated in Fig. 2.6.
As expected a linear regression analysis applied to the normalized ∆ VOR gain
yielded a coefficient of “1.00” (Fig. 2.7, left). It is important to keep in mind that the
training conditions used for this analysis are the consistent training conditions
comprised of a coherently moving target and background.
We next evaluated how conflicting target and background may affect motor learning
in the VOR. The ∆ VOR gain induced by each conflicting training condition was
normalized by the learning rates for each animal (Fig. 2.7, right). If the conflicting
background had no effect on learning, we would expect the learning to be similar to
learning induced by consistent training conditions, and a regression analysis on the
normalized ∆ VOR gain induced by conflicting training conditions should yield a
coefficient of 1.00. In this situation, the amount learning would be predicted by target
slip alone. However, a regression analysis on the normalized ∆ VOR gain for
25
conflicting training conditions was 0.29, demonstrating that conflicting background
slip impaired learning by approximately 70% relative to the learning induced by
consistent training conditions.
The correlation between target slip and learning induced by conflicting training
conditions (R2 = 0.42) was not as robust as the correlation between target slip and
learning induced by consistent training conditions (R2 = 0.73), suggesting that the
learning impairment is not a global 70% reduction at all target slip velocities. To take
a closer look at the effects of the conflicting background on learning, we grouped our
training conditions according to
their target motion (x0, x2 for
Monkey L and Monkey C; x0, x0.5,
x1, x1.5, x2 for Monkey E). For
each target motion, we calculated
the difference in learning between
the consistent and conflicting
training conditions, and divided the
difference by the learning induced
with consistent conditions. These
values represent the effect of the
conflicting background motion on
learning for each target motion. We
plotted the effect of conflicting
background motion and the average target slip associated with each target motion (Fig.
2.8). In general, when the target slip was small (when the animal was tracking well),
the effect of the conflicting background was large. However, when the target slip was
larger (when the animal was not tracking well), the effect of the conflicting
background diminished.
26
2.2.5 Phase changes induced by consistent and conflicting training conditions
Changes in VOR gain were accompanied by changes in the phase of the VOR. The
VOR phase is a measure of when peak eye velocity occurs relative to peak head
velocity during the VOR. If the peak eye velocity response to head motion occurred
before peak head velocity, then there was a phase lead. If peak eye velocity response
to head motion occurred after peak head velocity, then there was a phase lag. Visual-
vestibular stimuli that induced VOR gain changes also induced VOR phase changes.
A change in VOR phase was calculated as the phase of the VOR after learning minus
the VOR phase before learning. Typically, VOR gain changes induced at the training
frequency were not accompanied by a change in VOR phase [22]. However, changes
in VOR gain at non-training frequencies were accompanied by a change in phase.
Figure 2.9 plots the phase profiles for consistent and conflicting training conditions.
In general, increases in gain (x2 training conditions, blue) induced an increase in phase
lag at higher test frequencies, whereas decreases in gain (x0 training conditions, red)
induced an increase in phase lead at higher test frequencies. In Monkey C, the x0
training conditions did not elicit consistent phase leads, so those results were
replicated in Monkey E. x0 and x2 training conditions did not induce different
changes in the phase of the VOR measured at 0.2 Hz. Phase changes during
conflicting training conditions were similar to consistent training conditions for x2
(compare light blue and dark blue), however, x0 conflicting training conditions (light
pink) induced little phase changes at the higher testing frequencies.
27
2.2.6 Motor learning during training conditions at 5 Hz
During the induction of motor learning in the VOR with 0.5 Hz visual-vestibular
stimuli, the target slip velocity was correlated with learning (Fig. 2.6). Since gaze
velocity and target slip velocity were correlated (Fig. 2.4), learning can also be
correlated to gaze velocity. The correlations between VOR learning and target slip
and gaze velocities suggest that the effects of target motion on VOR learning may be
mediated by either the retinal slip it creates or by the eye movements it elicits.
To study whether motor learning in the VOR can be guided by target slip velocity, but
not gaze velocity, we tested in Monkey L a select subset of the target and background
combinations moving at 5 Hz, with a peak velocity of ± 10 deg/s. At 5 Hz, the
monkey was unable to track the visual stimuli. The position excursion of the target
moving at 5 Hz was small (± 0.3 deg), so in order to get the juice reward, the monkey
simply attempted to maintain fixation on the target. At 5 Hz, the eye movements were
similar for x0 and x2 training conditions, and therefore, the gaze velocity were the
same during these training conditions. Figure 2.10 plots the background velocity
(red), target velocity (blue), and gaze velocity (green) against the learning induced by
six different training conditions at 5 Hz for Monkey L. The median amount of
learning induced by a training condition was plotted on the vertical axis, and the error
bars represented the range of gain changes that were induced.
28
Motor learning induced by 5 Hz training conditions was not well correlated to
background slip velocity or gaze velocity. However, target slip velocity was
correlated to the induced changes in VOR gain. If target slip occurred in the same
direction as head motion, a decrease in VOR gain was induced. If target slip occurred
in the opposite direction of head motion, an increase in VOR gain was induced.
Therefore, target slip was capable of influencing learning, and the effects of the target
on learning may not necessarily be mediated by its effects on eye movements.
During 5 Hz stimuli, the monkeys were unable to track the visual target. The eye
movement responses during x0 and x2 training conditions at 5 Hz resulted in nearly
equal target and background slip velocities. Figure 2.11 plots target and background
slip velocities, and the learning induced by each of the 6 different visual-vestibular
stimuli tested at 5 Hz.
29
Similar to 0.5 Hz stimuli, pairing of head movements with a target alone was
sufficient to induce robust changes in VOR gain. VOR gain during all 6 stimuli went
in the direction predicted by the
target. The x2T/x0BG training
condition elicited greater target slip
than background slip. This was the
only conflicting condition at either 0.5
or 5 Hz where the target slip velocity
was bigger than background slip
velocity. The x2T/x0BG training
condition at 5 Hz induced increases in
gain, suggesting that motor learning in
the VOR need not be in the direction
predicted by the visual stimulus
eliciting the smallest retinal slip
velocity.
2.2.7 Effects of target and background on floccular Purkinje cell simple spikes
To analyze the effects of consistent and conflicting training conditions on the putative
neural instructive signals during motor learning in the VOR, we recorded Purkinje
cells in the floccular complex of the same three rhesus monkeys used for the
behavioral experiments. We focused our analysis on a subclass of Purkinje cell
neurons, known as Horizontal Gaze Velocity Purkinje cells (HGVPs), which have
been implicated in motor learning in the VOR [11, 19]. Spikes in the climbing fiber
reliably trigger calcium spikes, called complex spikes, in its Purkinje cell targets in a
one-to-one manner, therefore we used complex spike activity as a measure of activity
in its climbing fiber input [10, 33].
Individual neural responses to all training conditions for each monkey are supplied in
the appendix. The peak amplitude of the neural response elicited by each visual-
30
vestibular stimulus, and its phase relative to head velocity were calculated. To
systematically compare neural responses to many different training conditions, we
compared responses at a particular phase of head movement, and we report the neural
response aligned with peak head velocity (see Methods). Responses with positive
values correspond to increased firing during peak ipsiversive head movement, and
responses with negative values correspond to increased firing during peak
contraversive head movement.
The simple spike output of the Purkinje cells has been proposed to be a neural
instructive signal for motor learning in the VOR [1]. As shown in previous studies,
the simple spike activity was well-correlated to gaze velocity (Fig. 2.12). Gaze in the
same direction as head motion (positive gaze) elicited increases in Purkinje cell simple
spike firing rate during ipsiversive head motion, whereas gaze in the opposite direction
as head motion (negative gaze) elicited decreases in simple spike firing rate during
ipsiversive head motion.
The presence of a conflicting background had small effects on the tracking eye
movements in Monkey L and Monkey C. These effects were opposite in nature—in
Monkey L, presence of a conflicting background improved tracking eye movements,
whereas in Monkey C, presence of a conflicting background impaired tracking eye
movements. These effects can be reflected in the gaze velocity elicited by the
31
conflicting training conditions (Fig. 2.4). In Figure 2.13, we compared gaze velocities
elicited by conflicting and consistent training conditions. For each target motion, the
averaged gaze velocity elicited by
the conflicting training conditions
was normalized by dividing the
value by the averaged gaze
velocity elicited by consistent
training conditions. For Monkey
L, values for normalized gaze
velocity for each target motion
were greater than 1.0, whereas for
Monkey C, the normalized gaze
velocities were less than 1.0.
Thus, for Monkey L, conflicting training conditions increased gaze velocities relative
to consistent training conditions, whereas for Monkey C, conflicting training
conditions decreased gaze velocities. When we examined the change in Purkinje cell
simple spike firing rate modulations during conflicting training conditions relative to
the firing rate modulations elicited by the consistent training conditions (abscissa, Fig.
2.13), we observed opposite effects of the conflicting background on Purkinje cell
responses between Monkey L and Monkey C. For Monkey L, conflicting conditions
increased Purkinje cell responses relative to consistent conditions, whereas for
Monkey C, conflicting training conditions decreased Purkinje cell responses. Thus, the
changes in tracking eye movements as measured by gaze velocity during conflicting
training conditions parallel the observed changes in Purkinje cell simple spike activity.
2.2.8 Effects of target and background on floccular climbing fiber activity
Floccular climbing fiber activity is thought to carry information regarding retinal slip
[5]. During consistent training conditions, climbing fiber activity was well-correlated
to target slip—target slip in the same direction as the head (positive slip) elicited
increases in climbing fiber activity during contraversive head movements, whereas
32
target slip in the opposite direction as the head (negative slip) elicited increases in
climbing fiber activity during ipsiversive head movements (Fig. 2.14, gray diamonds).
Climbing fibers have an average spontaneous firing rate of approximately 1 Hz, and
the maximal firing rate modulation was approximately ± 1 sp/s.
Presence of a conflicting background reduced the firing rate modulation of climbing
fibers. With few exceptions, the climbing fiber responses remained in the same
direction as the responses elicited by consistent training conditions (Fig. 2.14, white
diamonds). Floccular climbing fibers are typically thought to be driven by
contraversive image motion[5], and the phase of the climbing fiber response relative to
head movement correlated to when contraversive image motion occurred during the
visual-vestibular stimuli. During conflicting training conditions, the retinal slip of the
target and background were moving in opposite directions. If the climbing fibers were
more sensitive to the background slip than target slip, then we would expect during
conflicting conditions, the climbing fiber response to be out-of-phase relative to the
consistent conditions. We observed this effect for only a small subset of the
conflicting training conditions (x0.5T/x1.5BG and x1T/x0BG in Monkey E),
suggesting for the most part target slip was more effective at driving climbing fiber
responses than background slip.
33
For Monkey E, the x0.5T/x1.5BG and
x1T/x0BG conflicting training
conditions elicited relatively smaller
amounts of target slip (0.58 and 0.08
deg/s, respectively) compared to
x0T/x2BG and x2T/x0BG conflicting
training conditions (0.92 and 1.06
deg/s, respectively). We suspected
that the amplitude of target slip may
gate the effect of the conflicting
background on climbing fiber
modulation. Indeed, when we
examined the effects of the conflicting background on climbing fiber responses
according to the elicited target slip, we find that the conflicting background has a
larger effect on climbing fiber activity when the target slip velocity was small (Fig.
2.15).
Interestingly, the effects of the conflicting background on learning were paralleled by
its effect on climbing fiber activity (Fig. 2.8 and Fig. 2.15). In both cases, the amount
of target slip mediated the effect of the conflicting background, with the largest effect
of the conflicting background when the target slip was small.
2.2.9 Purkinje cell simple spikes and climbing fiber signals modulate
independently
Cerebellar researchers have proposed that the climbing fiber signals directly modulate
the simple spike signals[34], resulting in antiphasic simple spike and complex spike
activity. If climbing fiber activity directly affected simple spike modulation, then we
would expect the background effect on climbing fiber signals should also affect the
Purkinje cell simple spike response.
34
Our experiments can directly test this hypothesis, since our training conditions elicited
different combinations of climbing fiber and Purkinje cell responses. We can test
whether Purkinje cell simple spike and climbing fiber responses covary, as predicted if
climbing fiber activity modulates Purkinje cell simple spike activity. In our set of
training conditions, we created pairs of training conditions where climbing fiber
responses were the same, but Purkinje cell simple spike responses were different, or
vice versa. For example, in Monkey L, the x2T/x2BG training condition (Fig. 2.16,
green datapoint) elicited statistically similar Purkinje cell responses as the x2T/x0BG
training condition (Fig. 2.16, blue datapoint, p = 0.51, paired t-test), but statistically
different climbing fiber responses (p < 0.001, paired t-test). Likewise, in Monkey L,
the x2T/x0BG (Fig. 2.16, red datapoint) elicited similar climbing fiber responses to the
x0T/x2BG training condition (Fig. 2.16, blue datapoint, p = 0.50, paired t-test), but
different Purkinje cell responses (p < 0.001, paired t-test). Therefore, the covariation
between Purkinje cell simple spike responses and climbing fiber responses can be
broken using our behavioral manipulations.
Further evidence that climbing fiber and Purkinje cell signals modulate independently
comes from the x0T/x2BG training conditions in Monkey L and Monkey E. The
x0T/x2BG training condition elicits no climbing fiber modulation, yet Purkinje cell
simple spikes remain robustly modulated (p = 0.39, Monkey L; p = 0.84, Monkey E;
one sample t-test, Fig. 2.16). During the x0T/x2BG training condition, the climbing
35
fiber activity is similar to its spontaneous activity (see Chapter 3). Thus, something
other than climbing fiber activity must be driving Purkinje cell simple spike
modulation during this training condition. Since the simple spike modulation was
well-correlated to gaze (Fig. 2.12), the modulation was most likely driven by eye
movement inputs.
36
2.3 Discussion
2.3.1 Learning cues during the induction of motor learning in the VOR
To acutely induce motor learning in the VOR in previous experiments, head
movements were paired with motion of a single visual stimulus, providing image
motion of a single stimulus in one direction on the retina. However, in a more natural
environment with multiple visual cues, multiple and at times conflicting image
motions on the retina may be present. The preceding set of experiments induced
motor learning in the VOR with multiple visual stimuli, in the hope of understanding
the effects of multiple image motions on putative neural instructive signals and motor
learning in the VOR. We find that motor learning in the VOR is affected by both
target and visual background motion. Several factors may influence the potency of the
target and background at inducing learning.
2.3.2 Retinal slip velocity and tracking eye movements
At low frequency, imperfect tracking of the visual target result in small-velocity
retinal slip of the target. During presentation of the visual background, there is
simultaneous retinal slip of the background that can be either in the same or opposite
direction as target slip During conditions of conflicting, or opposing slip, learning is
strongly influenced by the background, although in most conditions we tested, the
target predicted the direction of learning. Since at these lower frequencies, the
animals have the ability to track the target, the slip velocities of the target were smaller
than the slip velocities of the background. Several lines of earlier evidence suggest
how the target may exert its influence: First, the visual target controls tracking eye
movements that drives the simple spikes of the Purkinje cells, another putative
instructive signal shown to be well correlated to learning when present[1]. Second,
climbing fiber signals, a putative instructive signal, can be driven my image motion of
a target. Third, climbing fiber signals may be driven by the eye movements used to
track the target[30]. And last, the climbing fiber tend to be tuned to image motion of
low slip velocities, therefore, the low image slip of the target may be more effective at
driving climbing fiber signals relative to the background[5].
37
Our results demonstrate that the effects of the target on VOR motor learning is not due
to only tracking eye movements. First, during training conditions with similar eye
movements (e.g. compare x0T/x0BG versus x0T/x2BG), but very different
combinations of target and background slip, different amounts of learning was
induced, suggesting that eye movements are not the lone behavioral learning cue for
VOR learning. While our experiments exclude the possibility that eye movements
during 0.5 Hz stimuli is the lone behavioral cue guiding motor learning in the VOR, it
does not exclude an eye movement contribution to learning.
The experiments at 5 Hz also demonstrate that motor learning in the VOR are not
guided by images with the lowest slip velocities. At 5 Hz, the target and background
slip are equal, but learning went in the direction predicted by the target, suggesting
that something other than tuning to lower image motion velocity is influencing the
target’s effect on putative instructive signals and learning.
2.3.3 Behaviorally-relevant and foveal/parafoveal stimuli
The animals in the experiments were trained to track the visual target, and rewarded
for such behavior. Therefore, the visual target was the behaviorally-relevant stimulus,
and probably the focus of the animal’s attention. Recent evidence demonstrate that
climbing fiber responses in afoveate rabbits during the presence of transparent slip are
associated with the retinal slip of the visual stimulus that triggers an eye
movement[35]. Therefore, the effect of the target on learning may be mediated
through the attentional gating of instructive signals carried by the climbing fibers.
In animals with a fovea, the visual stimulus that is the object of the animal’s attention
is often placed on the fovea. Therefore, it is difficult to dissociate whether attention
gates the climbing fiber response to the target, or whether the target exerts its
influence on the climbing fiber response because floccular climbing fibers place
greater weight on foveal image motion. There is evidence that suggest image motion
38
on the fovea plays critical importance in driving climbing fibers. A major input to the
dorsal cap of the inferior olive, the origin of floccular climbing fibers, is the nucleus of
the optic tract (NOT)[36]. Neurons in NOT in monkeys have been divided into two
classes based on their response properties[37, 38]. One class of neurons responds
during ipsiversive tracking of a visual target when presented alone or in the presence
of a stationary background. Another class of neurons responds during contraversive
tracking of a visual target but only when the target motion occurs in the presence of a
background. It appears that these neurons, respond to either retinal slip associated
with the target or background, respectively. Thus, the influence of the target on
climbing fiber signals may be reflected in the response properties of the neurons that
innervate the source of climbing fibers, the inferior olive.
The function of the VOR, based on studies in phylogenetics and VOR kinematics, is to
stabilize full-field images, paying no specific attention to stabilizing the visual
stimulus to one specific area of the retina. Based on this understanding of VOR
function, it may be predicted that full-field image motion would play a more important
role in guiding VOR learning[39, 40]. However, our results demonstrate that the
brain does not just care about net motion energy on the retina as what might be
predicted. Rather factors related to the target must influence the putative neural
instructive signals that guide learning.
2.3.4 Effects of target and background motion on putative neural instructive
signals for motor learning in the VOR
The combinations of target and background motion during visual-vestibular stimuli
used to induce VOR adaptation elicited many different combinations of Purkinje cell
and climbing fiber instructive signals. Our data demonstrated two main results: first,
climbing fibers in the floccular complex encode information about the target and
background, with the climbing fiber response reflecting a complex, non-linear
interaction between the two visual stimuli. Second, the ability to break the previously
observed covariation between climbing fiber and Purkinje cell signals[34] demonstrate
39
that to some degree, inputs to Purkinje cells in the floccular complex, can
independently drive simple and complex spike activity.
2.3.5 Climbing fibers do not encode net image motion on the retina
The earliest studies of floccular climbing fiber responses used motion of a single,
large-field visual stimulus to induce coherent image motion on the retina[5, 41].
These studies found that floccular climbing fibers increased their firing during
contraversive retinal image motion, and were tuned to low retinal slip velocities.
More recently, studies in primates have demonstrated that motion of a small visual
target was sufficient to drive floccular climbing fibers, although the climbing fiber
response were typically smaller than when a large-field stimulus was nused[30].
Classically, floccular climbing fibers have been thought to carry only visual
information. However, more recent recordings of floccular climbing fibers
demonstrated a population of neurons that modulate in the light during performance of
the optokinetic reflex (OKR) and during VOR in complete darkness, suggesting that
these climbing fibers carry a non-visual, vestibular signal[42]. In the study, during
performance of the OKR, the climbing fibers increased their firing during
contraversive motion of the optokinetic drum, and decreased their firing during
ipsiversive motion of the drum, consistent with the earlier findings studies that
floccular climbing fibers are driven by contraversive image motion, and inhibited by
ipsiversive image motion on the retina. The same climbing fibers during performance
of the VOR in complete darkness increased their firing during contraversive head
movements, and decreased their firing during ipsiversive head movements. The
climbing fiber response was also tested during performance of the VOR in the light
(x1 stimulus). In the light, contraversive head rotations (which drive the climbing
fiber response) result in ipsilateral image motion (which inhibits the climbing fiber
response), so if the climbing fibers response were to reflect both its visual and
vestibular inputs, then the climbing fibers should not respond during VOR in the light.
However, the same climbing fibers that carry non-visual signals are strongly
modulated during VOR in the light, in the direction predicted by image motion. These
40
results demonstrated two important qualities of floccular climbing fibers. First, there
is a non-visual signal in floccular climbing fibers. And second, there is a complex,
non-linear interaction between the climbing fiber inputs. For example, it may be
possible that the presence of a visual signal gates the effectiveness of the vestibular
input at driving climbing fiber activity. If climbing fiber signals were an error signal
guiding motor learning in the VOR, one may expect that the climbing fiber
modulation during VOR in the dark to induce VOR learning. However, a vestibular
stimulus in the absence of any visual stimuli did not induce any consistent changes in
VOR gain [43].
Another challenge to earlier findings that floccular climbing fibers carry only visual
signals related to image motion is a study by Frens, et al. In their study, they recorded
floccular climbing fibers during transparent motion[35]. Transparent motion occurs
when there are multiple motions in the same part of the visual space, and can be
created by two overlapping visual stimuli moving in opposite direction. When rabbits
are exposed to transparent motion stimuli, the rabbits follow one of the two visual
stimuli. The induced eye movements would create multiple image motions on the
retina. Frens, et al. found that responses in the floccular climbing fibers were best
correlated to the retinal image slip of the visual stimulus that was being followed,
suggesting that climbing fibers do not encode net image motion on the retina, but
rather the sensory consequence of a movement, or a “performance error.” However,
there were several confounds that may dispute this conclusion. First, since climbing
fibers were correlated to eye movements in the study, it may be possible that climbing
fiber signals contained a motor signal. Indeed follow up studies using optokinetic
white noise stimuli demonstrated that climbing fiber responses were better correlated
to the eye movements, than to retinal slip[44]. Also, since the visual stimulus that
elicited an eye movement is most likely to be the behaviorally-relevant or attended
stimulus, it may be possible that attention gates the climbing fiber response. And last,
since the eye movements elicited by a visual stimulus would decrease the retinal slip
velocity of that stimulus relative to visual stimuli that were stationary or moving in the
41
opposite direction, it may be possible that climbing fibers respond best to the image
motion moving the slowest on the retina.
Our current study also addressed how floccular climbing fibers respond during
simultaneous presentation of multiple, and at times, oppositely directed image motion
on the retina. In our study, the visual target was the behaviorally-relevant stimulus, in
that the animals were trained to track the target for a juice reward. A key difference
between our findings and those observed in the Frens’ study is that in the Fren’s study,
the climbing fiber response only reflected the image motion resulting from the visual
stimulus eliciting the eye movement. In our study, climbing fiber responses were
affected not only by the stimulus driving eye movements—the target, but also by the
large background, especially when background elicited retinal slip in the opposite
direction of target slip.
Another key difference in our study was that we used rhesus monkeys instead of
rabbits. Rhesus monkeys, unlike rabbits, are foveated animals. It may be possible that
visual stimuli placed on the fovea are given greater weight at driving climbing fiber
responses. Indeed, recordings from neurons in the nucleus of the optic tract (NOT),
the main input neurons to the dorsal cap of the inferior olive which give rise to
floccular climbing fibers, demonstrated two classes of neurons in near equal numbers:
neurons that were sensitive to motion of a foveal target, and those sensitive to motion
of a large visual background [37, 38]. It may be possible that oppositely directed
motion of a target and background would tend to cancel the response at the level of the
inferior olive. However, in our current results, there was a non-linear interaction
between the target and background at driving climbing fiber responses. During
training stimuli that elicited smaller target slip velocities (due to more effective
tracking eye movements), the effectiveness of an oppositely directed background at
reducing the climbing fiber responses was greater. One possible explanation may be
that the NOT neurons encoding target motion may be tuned to higher target slip
velocities, so as target slip velocities increase, the target played a greater role relative
42
to the background at driving climbing fiber responses. The precise velocity tuning of
these two distinct populations of NOT neurons remains unknown.
2.3.6 Modulation of Purkinje cell simple spikes is independent of climbing fiber
signals
In a previous study, despite blocking vestibular primary afferent signals to the
ipsilateral uvula-nodulus by unilateral labryinthectomy, simple spike modulation in
ipsilateral Purkinje cells was observed during head rotations, suggesting that climbing
fiber signals originating from the contralateral inferior olive modulated the simple
spike response[34]. In our study, we were able to remove the climbing fiber response
while maintaining the simple spike modulation, suggesting that at least in the
flocculus, a component of the simple spike modulation is independent of climbing
fiber signals. In our study, we recorded the simple and complex spike response of a
single neuron to many different visual-vestibular stimuli, and found that within a
single neuron, climbing fiber response may be altered while simple spike responses
remained constant. Thus, our data suggest that the modulation of simple spikes in a
particular Purkinje cell may be independent of modulation of the climbing fiber that
projects to that same Purkinje cell. It may be possible that climbing fibers activating
adjacent Purkinje cells or cerebellar interneurons drive the simple spike modulation in
the recorded Purkinje cell. However, our data across animals do not support this
argument. Across the three animals used in our experiments, conflicting background
motion uniformly reduced the climbing fiber response relative to consistent
background motion, so it is reasonable to assume this manipulation would have a same
effect on neighboring climbing fiber responses in each of the animals. If these
neighboring climbing fibers modulate Purkinje cell simple spikes, we would expect a
conflicting background to have the same effect on simple spike modulation in each of
the three animals. However, there were different effects of the conflicting background
on the Purkinje cell simple spike responses. In one animal, the simple spike response
was the same as during consistent background motion (Monkey E), in another monkey
the simple spike response was reduced (Monkey C), and in the last animal the simple
43
spike response was enhanced (Monkey L). Therefore, the different effects of the
conflicting background on simple spike modulation are unlikely to be the result of
neighboring climbing fiber responses. Rather, the different Purkinje cell simple spike
responses were correlated to the different effects of the conflicting background on
tracking eye movements. Thus, the eye movement inputs to the Purkinje cells and the
retinal slip inputs to the climbing fibers can effectively drive signals in each
population of neurons independently of each other.
44
2.4 Materials and methods
General Procedures
Experiments were conducted on three male rhesus monkeys. Animals were trained to
perform a visual fixation task to obtain liquid reinforcement. Previously described
surgical procedures were used to implant orthopedic plates in the skull for restraining
the head[22, 45], and to implant a coil of wire in one eye for measuring vertical and
horizontal eye position[46]. A second surgical procedure was conducted to
stereotaxically place a recording cylinder for single-unit recording. During
experiments, each monkey sat in a specially designed primate chair to which his
implanted head holder was secured. Vestibular stimuli were delivered using a servo-
controlled turntable (Ideal Aerosmith) that rotated the animal, the primate chair, and a
set of magnetic coils (CNC Engineering) together about an earth-vertical axis. Visual
motion stimuli were provided by a visual “target” subtending 0.5˚ of visual angle,
which the animal was rewarded for tracking, and a visual “background” consisting of
a high contrast, black and white checkerboard pattern (20˚ x 30˚). The visual stimuli
were reflected off mirror galvanometers onto the back of a tangent screen 114 cm in
front of the eyes.
Behavioral experiments
Motor learning in the VOR was induced by presenting combined visual-vestibular
stimuli for one hour (Monkey E) or two hours (Monkey L). VOR performance was
tested before, after and at 15 minute intervals (Monkey E) or 30 minute intervals
(Monkey L) during training, by delivering the vestibular stimulus in total darkness.
The vestibular stimulus used to measure the VOR and induce learning had a
sinusoidal velocity profile (0.5 Hz, peak velocity ±10˚/s or in a few cases, where
noted, ±20˚/s). The visual-vestibular training stimuli are described by the eye velocity
gain (relative to head movement) required to stabilize the image of the target (T) and
45
the background (BG) on the retina. During training stimuli with a x0T, the target
moved exactly with the head, so an eye movement gain of zero would be required to
stabilize the target on the retina. During training stimuli with a x2T, the visual target
moved at the same speed as the head but 180˚ out of phase with the head, thus, eye
movements would need to be twice as large as the head movements (gain of 2) to
stabilize the target on the retina. During training stimuli with a x0.5T or x1.5T, the
target moved in phase or 180˚ out of phase with the head, respectively at one-half the
head speed. During training stimuli with x1T, the visual target was earth-stationary.
With the exception of two experiments conducted in Monkey L (x0T only and x2T
only), the visual stimulus was comprised of visual background and target motions. In
some experiments, the motion of the target and background were the same
(x0T/x0BG, x0.5T/x0.5BG, x1T/x1BG, x1.5T/x1.5BG, x2T/x2BG). In the remaining
experiments, there was non-coherent motion of the target and background. During
x0T/x2BG, x0.5T/x1.5BG, x1.5T/x0.5BG and x2T/x0BG stimuli, the visual
background moved at the same speed but 180˚ out of phase with target motion.
During the x0T/x1BG, x0.5T/x1BG, x1.5T/x1BG and x2T/x1BG stimuli, the visual
background was earth-stationary during target motion. During x0T/x0.5BG and
x2T/x1.5 BG stimuli, the background moved at one-half the speed and in-phase with
target motion. During the x0.5T/x0BG and x1.5T/x2BG stimuli, the background
moved at twice the speed and in phase with target motion. During the x1T/x0BG and
x1T/x2BG stimuli, the visual target was earth-stationary, while the background mo-
ved at the same speed and in phase or 180˚ out of phase with the head, respectively.
Experiments were separated by at least 24 hours to allow the gain of the VOR to
readapt to its normal value before the next experiment. There were a minimum of
three replications of the behavioral experiments for each training stimulus.
46
Electrophysiology
Tungsten electrodes (FHC, Microprobe) were used to make extracellular recordings
from Purkinje cells in the flocculus/ventral paraflocculus of the cerebellum. After a
Purkinje cell was isolated, its sensitivity to eye velocity and head velocity were first
measured by recording its responses during (1) smooth pursuit eye movements evoked
by horizontal motion of the visual target with a sinusoidal velocity profile at a
frequency of 0.5 Hz and a peak velocity of 20˚/s or greater and (2) as the monkey
canceled his VOR by tracking a visual target that moved exactly with sinusoidal head
rotation about an earth vertical axis at 0.5 Hz and at a peak velocity of 20˚/s or greater.
The present work focuses on horizontal-gaze velocity Purkinje cells (HGVPs), a well
studied class of Purkinje cells in the floccular complex. Purkinje cells were classified
as HGVPs if (1) during horizontal smooth pursuit eye movements, simple spike firing
rate was modulated by at least ±0.3 sp/s per ˚/s, and there was a phase difference of
less than 45 degrees between peak firing rate and peak ipsiversive eye velocity; (2)
during cancellation of the VOR, simple spike firing rate was modulated by at least
±0.3 sp/s per ˚/s and the phase difference between peak firing rate and peak ipsiversive
head velocity was less than 45 degrees [6, 16]. [6, 16]. A total of 76 HGVPs were
recorded. Complex spikes were well isolated in 58/76 HGVPs. In the 18 recorded
neurons without complex spike activity, the neurons were deemed Purkinje cells
because of their simple spike wave form, their irregular simple spike firing rate, the
ability to record the cells through several hundred micrometers of cerebellum, and in
some cases, their characteristic injury discharge at the end of recording
We compared the instructive signals carried by the same neuron during several
different training stimuli used to induce motor learning in the VOR. Each training
stimulus was presented for 60-90 seconds. Recordings were made when the gain of the
VOR was at baseline, and the training stimuli were not presented long enough to
induce measurable changes in VOR performance as measured in the dark.
47
Data analysis
Voltages related to the position and velocity of eye, head and visual stimulus were
recorded during the experiment at 500 Hz/channel. Eye velocity records were edited
to remove the rapid deflections caused by saccades. The data were then analyzed by
aligning stimulus cycles on head or target velocity, and averaging. Most averages
contained 10 or more cycles, and analyses were limited to cycles for which gaze
position was within 15 degrees of straight-ahead gaze. Average eye and head velocity
traces were subjected to a sines fit. The gain of the VOR was calculated as the ratio of
peak eye to head velocity derived from the fitted sinusoidal functions.
To calculate horizontal gaze velocity, the raw eye and head velocity traces were
summed for each stimulus cycle. In these calculations, the rapid deflections caused
saccadic eye movements were not removed from the records. The gaze velocity
records were aligned to peak head velocity and subjected to a sines fit. The amplitude
of the gaze velocity was the amplitude of the fitted sinusoidal function. Gaze
velocities were assigned positive and negative values according to its relative
relationship to head movements. The phase of gaze velocity was defined as the
timing of the occurrence of peak gaze velocity relative to head velocity. Positive
phase values for gaze velocity indicated a phase lead, where the occurrence of peak
gaze velocity preceded peak head velocity.
To calculate retinal slip of the target, target velocity was calculated from an offline
differentiation of the recorded target position trace. Retinal slip of the target was
calculated as the difference in horizontal gaze velocity and target velocity for each
individual stimulus cycle. The retinal slip records were aligned to peak head velocity
and subjected to a sines fit. The amplitude of the target slip velocity was the
amplitude of the fitted sinusoidal function. The phase of target slip velocity was
defined as the timing of the occurrence of peak target slip velocity relative to head
48
velocity. Positive phase values for retinal slip indicated a phase lead, where the
occurrence of peak target slip preceded peak head velocity.
Since the voltage signal conveying the target command was used to control motion of
the visual background, the retinal slip of the background was calculated using the
recorded target position trace. For example, if the background moved 180˚ out-of-
phase but at the same speed as the target (e.g. during x0T/x2BG), the values
pertaining to background position were equal to values associated with target position
multiplied by (-1). If the background moved in-phase but at one-half the speed of the
target, the values pertaining to background position were equal to values associated
with target position multiplied by (0.5). Once the background position was
calculated, the amplitude and phase of background slip were derived using the same
method as calculating amplitude and phase of target slip.
The component of the gaze velocity, background slip velocity, and target slip velocity
aligned with peak head velocity was calculated by multiplying the amplitude of the
response with cosine of the phase, and these values were used for statistical analysis.
The simple-spike activity of Purkinje cells was detected with a hardware window
discriminator, and the times of the resulting pulses were recorded to the nearest
10 µsec. In addition, unit activity was sampled at 50 kHz, and complex spikes were
discriminated using off-line spike sorting with time and amplitude windows or
template matching algorithms (Spike2, Cambridge Electronic Design, Cambridge,
UK). In addition, the occurrence of each complex spike was confirmed by visual
inspection of the raw traces.
Data analysis was performed in Matlab and Excel. The simple spike data were
analyzed after the experiment by aligning the records on head velocity or visual
stimulus position. The amplitude of firing rate modulation and phase of the simple
spike responses relative to peak contraversive head velocity were estimated as the
49
amplitude and phase of the fundamental components provided by Fourier analysis of
the averages.
Because some stimuli drive climbing fiber activity into firing rate cutoff, the climbing
fiber responses were not always well described by a sinusoid. Therefore, to quantify
climbing fiber responses during training stimuli, complex spike data were analyzed
using a vector analysis. Stimulus cycles were aligned on head velocity and averaged.
The stimulus cycle was divided into 1000 equal bins. Each time bin was represented
as a vector with the magnitude of the vector equal to the average firing rate in that bin,
and the phase determined by the phase of the bin relative to peak ipsiversive head
velocity. The phase and amplitude of the climbing fiber response were calculated as
the phase and one-half the amplitude of the vector sum.
The component of the climbing fiber response or simple spike response aligned with
peak head velocity was calculated by multiplying the amplitude of the response with
cosine of the phase, and these values were used for statistical analysis. The
significance of each neural response was determined by performing the vector analysis
on each cycle of head movement during a given training stimulus. Significance was
tested using a one-sample t-test.
50
Chapter 3 Elimination of climbing fiber instructive signals during motor learning
3.1 Introduction
To understand the algorithm a neural circuit uses to learn, one must determine how its
patterns of activity during the induction of learning are translated into the cellular
changes that encode memory. The cerebellum, which supports motor learning, is one
brain region for which there is a well-developed theory about the neural events that
induce plasticity during learning. The dominant theory over the last several decades
has postulated that the climbing fibers provide the neural instructive signals that guide
cerebellum-dependent learning[2, 4, 12]. In support of the climbing fiber hypothesis,
in vivo recordings have shown that climbing fiber activity signals errors during a
number of different motor learning paradigms[6, 29, 47, 48]. Moreover, previous
studies have documented the sufficiency of climbing fiber signals to induce learning—
training with paired electrical stimulation of the climbing fibers and a conditioned
stimulus (CS) can induce a conditioned response (CR) when the CS is subsequently
presented alone[49].
The classic Marr-Albus-Ito theory attributed cerebellum-dependent learning to
climbing fiber-triggered plasticity at the parallel fiber-to-Purkinje cell synapses in the
cerebellar cortex[3, 12]. More recent evidence suggests that multiple, distributed
mechanisms contribute to cerebellum-dependent learning[26, 50-54]. Nevertheless,
climbing fiber-triggered plasticity is still considered central to cerebellum-dependent
learning[50, 54-56], since the climbing fibers are positioned to control multiple
plasticity mechanisms in the cerebellum and related circuitry. At the parallel fiber-to-
Purkinje cell synapses, an increase or decrease in climbing fiber activity relative to its
baseline activity could provide an instructive signal that triggers long-term depression
(LTD) or long-term potentiation (LTP), respectively[14, 57]. At other classes of
synapses in the cerebellar cortex, climbing fiber activation controls several additional
plasticity mechanisms[23, 33, 58-60]. Moreover, changes in the deep cerebellar nuclei
(DCN) or vestibular nuclei (VN) during cerebellum-dependent learning are widely
51
viewed as secondary to or dependent upon climbing fiber-triggered changes in the
cerebellar cortex[24, 50, 53, 54]. Thus, neural instructive signals in the climbing
fibers could, either directly or indirectly, orchestrate the induction of multiple forms of
plasticity in the cerebellum and related circuitry. The central question addressed in
this chapter is whether these climbing fiber-controlled plasticity mechanisms are
necessary for motor learning, or whether plasticity mechanisms controlled by other
neural instructive signals could independently support learning.
Previous tests of the necessity of the climbing fibers for motor learning have been
inconclusive. Lesion or pharmacological inactivation of the source of the climbing
fibers, the inferior olive, abolishes cerebellum-dependent learning[61, 62]. However,
such manipulations abolish spontaneous neural activity in the climbing fibers as well
as the task-related signals carried by changes in firing rate above and below the
spontaneous level of activity. The elimination of spontaneous activity has the effect of
producing abnormal neural activity at multiple sites in the cerebellar circuit[63, 64].
Thus, the inability to learn after such manipulations cannot be directly attributed to the
loss of the instructive signals in the climbing fibers, but could simply reflect the gross
cerebellar dysfunction associated with disrupted basal activity. To avoid this
confound, we developed a behavioral approach to selectively eliminate the instructive
signals in the climbing fibers without affecting their baseline level of activity.
The paradigm we used was the adaptive modification of the vestibulo-ocular reflex
(VOR) by motor learning. Signaling in the VOR circuit is understood at a level that
enabled us to design training stimuli that would abolish instructive signals in the
climbing fibers, while leaving intact most other aspects of the training conditions
known to induce VOR learning. The VOR is a reflexive eye movement that functions
to stabilize images on the retina by generating eye movements in the opposite
direction from head motion. If the VOR fails to stabilize images during head
movements, motor learning can adjust the amplitude, or gain, of the VOR (i.e., the
ratio of eye velocity to head velocity) to restore image stability. This form of motor
52
learning requires the floccular complex of the cerebellum, comprising the cerebellar
flocculus and ventral paraflocculus. Lesions of the floccular complex abolish the
adaptive modification of the VOR through motor learning without affecting its
baseline performance[65, 66], whereas lesions of other regions of the cerebellum
carrying vestibular, visual, or oculomotor signals do not affect motor learning in the
VOR[67-69].
In the laboratory, motor learning in the VOR is typically induced by pairing head
movements with motion of a single, large, coherently moving visual stimulus (Fig.
3.1b). If the visual stimulus moves exactly with the head, a decrease in the gain of the
VOR is induced. Such training stimuli are called “x0”, because the VOR gain
required to stabilize images on the retina is zero. If, instead, the visual stimulus moves
exactly opposite to head motion, an increase in VOR gain is induced. Such training
stimuli are called “x2” because the ideal VOR gain would be 2 (eye velocity equal to
twice head velocity).
One line of evidence that instructive signals in the climbing fibers control motor
learning in the VOR has been that the climbing fibers carry information about the
53
required direction of learning during x0 and x2 training. Climbing fibers in the
floccular complex are driven by contraversive image motion and inhibited by
ipsiversive image motion[5, 41]. During x0 training, contraversive image motion
occurs during contraversive head motion, whereas during x2 training, contraversive
image motion occurs during ipsiversive head motion. Thus, the timing of the peak
climbing fiber activity relative to head motion carries information about whether the
VOR gain needs to increase or decrease[5, 6, 13]. Based on the known physiology of
the circuit, it has been proposed that these differently timed climbing fiber responses
induce LTD and/or LTP in the appropriate vestibular parallel fiber-to-Purkinje cell
synapses to support the observed changes in VOR gain[2, 4, 12, 26, 70] (Fig. 3.1).
During head rotations in the dark, when no visual cues are available to guide motor
learning in the VOR, and no consistent changes in VOR gain are induced, the same
climbing fibers have little or no response, but simply fire at their spontaneous rate[43,
71].
Thus, with standard training conditions, the climbing fibers carry information about
whether the VOR gain needs to be altered, and they appear to be capable of
implementing appropriate changes in the VOR circuit to accomplish the required
behavioral changes. In the present study, we compared the climbing fiber responses
during standard x0 and x2 training stimuli, with those present during variations of
these stimuli designed to eliminate the instructive signals in the climbing fibers. Our
results suggest that learning can be induced in the absence of instructive signals in the
climbing fibers and that plasticity mechanisms controlled by other neural instructive
signals make a significant and independent contribution to motor learning.
54
3.2 Results
We analyzed the neural instructive signals available in the VOR circuit during a range
of training paradigms that induce motor learning in the VOR.
In the floccular complex of two rhesus monkeys, 102 Purkinje cells with task-related
activity (head and/or eye movement sensitivity) were recorded. Our analysis focused
on 58 of these Purkinje cells, which were identified as horizontal gaze velocity
Purkinje cells (HGVPs), a subclass of neurons that have been implicated in VOR
learning[15, 16]. Since spikes in a climbing fiber reliably trigger complex spikes in its
Purkinje cell targets in a ‘one-to-one’ manner, complex spike activity in a Purkinje
cell was used as a measure of activity in its climbing fiber input[10], and is referred to
as a “climbing fiber response”. Complex spikes were well isolated in 48/58 HGVPs
and 20/44 non-HGVPs. We also analyzed the simple spike responses of the Purkinje
cells to each training stimulus. In each individual neuron recorded, the responses to a
number of different visual-vestibular training stimuli were compared.
A similar set of behavioral and electrophysiology experiments were conducted on a
third animal, Monkey C, whose behavior and electrophysiology data is presented in
Chapter 2. However, the x0T/x2BG training condition which eliminated climbing
fiber responses in Monkey L and Monkey E (see Results below), elicited a statistically
significant climbing fiber response in Monkey C (see Appendix for individual
responses to x0T/x2BG; p < 0.001, one sample t-test), and therefore, we were unable
to test the necessity of climbing fiber instructive signals to motor learning in this
animal. Therefore, we excluded Monkey C’s dataset from this chapter.
3.2.1 Standard training stimuli elicit instructive signals in climbing fibers
Consistent with previous studies from a number of labs, the population of climbing
fibers we recorded discriminated between a x0 versus a x2 training stimulus[5, 6, 13],
and thus carried information about whether the VOR gain needed to increase or
decrease. During standard, x0 training, with coherent motion of a visual target (T;
55
subtending 0.5° of visual angle) and a visual background (BG; 20° x 30°), both
moving exactly with the head (0.5 Hz, ±10°/s), climbing fiber firing peaked during
contraversive head movement (x0T/x0BG training, Fig. 3.2); whereas during standard,
x2 training, climbing fiber firing peaked during ipsiversive head movement
(x2T/x2BG, Fig. 3.2).
The climbing fiber responses during the x0T/x0BG and x2T/x2BG training stimuli
reflected both an increased probability of firing during the “preferred” half-cycle of
the stimulus, and a decreased probability of firing during the “non-preferred” half-
cycle of the stimulus, relative to the baseline, spontaneous activity measured in the
absence of head movement or visual stimulus movement (Fig. 3.2b). Across the
population of climbing fibers, the responses were remarkably uniform (gray traces,
Fig. 3.2b). During the “preferred” half-cycle, a large fraction of the climbing fibers
fired with a probability two standard deviations (SD) above baseline, and almost no
climbing fibers fired with a probability 2 SD below baseline (Fig. 3.2c). Likewise,
during the “non-preferred” half-cycle, a large fraction of the climbing fibers fired with
a probability 2 SD below baseline, and almost no cells fired with a probability 2 SD
above baseline (Fig. 3.2c).
56
The stimulus-driven changes in the probability of climbing fiber firing were not
restricted to a particular time during the stimulus. Rather, throughout the stimulus
cycle, the probability of firing was 2 SD above or below baseline in the majority of
cells, and the probability of firing smoothly transitioned from low to high during the
stimulus cycle, reflecting the sinusoidal velocity profile of the training stimulus.
Therefore, each climbing fiber response was summarized by the amplitude of the
overall firing rate modulation during the stimulus cycle, and the phase of the peak
57
firing relative to head movement, calculated using a vector analysis (Fig. 3.2d; see
Methods).
The significance of each climbing fiber response was calculated by analyzing the
cycle-by-cycle variability. During the standard training stimuli, x0T/x0BG and
x2T/x2BG, almost all climbing fibers had responses significantly different from zero
(42/42 climbing fibers for x2T/x2BG, 45/48 climbing fibers for x0T/x0BG, Fig. 3.2d,
bold symbols), and, for each stimulus, the timing (phase) of peak firing relative to the
head movement was similar in all neurons with significant responses (within 70°).
Thus, the population response of the climbing fibers was significantly different from
zero for both x0T/x0BG and x2T/x2BG training stimuli (p<0.05, one sample t-test,
Table 3.1). Moreover, the population responses to the x0T/x0BG and x2T/x2BG
stimuli were significantly different from each other, since the phase of peak firing
relative to head movement was approximately 180° opposite for the two stimuli (Fig.
3.2d, Table 3.1). These climbing fiber responses were similar to those reported
previously in primates and other species[6, 13, 72].
58
3.2.2 Novel training stimuli eliminate instructive signals in climbing fibers
To eliminate climbing fiber responses during VOR training, and thereby test their
necessity for motor learning, we paired head movements with oppositely directed
motion of two visual stimuli. The origin of floccular climbing fibers is the
contralateral dorsal cap of the inferior olive, which in turn receives its major input
from the ipsilateral nucleus of the optic tract (NOT)[73]. Neurons in the NOT
increase their firing rate during ipsiversive image motion, and decrease their firing rate
during contraversive image motion. Within this population are two distinct subtypes
of neurons, found in approximately equal numbers: those sensitive to motion of a
foveal target, and those sensitive to motion of a large visual background[37, 38].
Therefore, we expected that the effects of oppositely directed motion of the target and
background might tend to cancel at the level of the inferior olive, and this was
confirmed by our recordings from the climbing fibers. When head movements were
paired with target motion plus oppositely directed background motion (x0T/x2BG,
Fig. 3.2; x2T/x0BG, Fig. 3.3), the responses of the climbing fibers were greatly
reduced compared with the responses in the same climbing fibers when the target and
background moved together (x0T/x0BG, x2T/x2BG, Fig. 3.2).
The reduction in climbing fiber response was greater for the x0T/x2BG training
stimulus than the x2T/x0BG stimulus. At the population level, the climbing fiber
response to the x2T/x0BG stimulus was significantly different from zero (one sample
t-test, Fig. 3.3, Table 3.1), therefore this stimulus did not provide a good test of the
necessity of climbing fiber instructive signals for motor learning. In contrast, the
population response to the x0T/x2BG stimulus was not significantly different from
zero (P > 0.05, one sample t-test; Table 3.1). Therefore, we conducted additional
analyses to evaluate whether the climbing fiber response was truly eliminated during
the x0T/x2BG stimulus.
59
When the cycle-by-cycle variability of individual climbing fiber responses was
considered, only 4/15 (Monkey L) and 1/30 (Monkey E) climbing fibers had
significant responses to the x0T/x2BG stimulus (Fig. 3.2d, x0T/x2BG, bold symbols).
Moreover, the timing of peak firing was not consistent in these five climbing fibers
with significant responses. Three climbing fibers increased their firing during
ipsiversive head movement, as observed during x2T/x2BG training, and two climbing
fibers increased their firing during contraversive head movement, as observed during
x0T/x0BG training.
We considered the possibility that climbing fiber responses may be restricted to a very
specific time in the stimulus cycle, but found no evidence for a temporally specific
climbing fiber response to the x0T/x2BG training stimulus. We compared the
probability of climbing fiber firing during each 200-ms segments of the x0T/x2BG
60
stimulus cycle with spontaneous activity and with spike trains obtained by randomly
shuffling the interspike intervals measured during the x0T/x2BG training stimulus to
remove any potential signal (shuffled, spontaneous, Fig. 3.2). Because of natural
variability in the climbing fiber interspike intervals, when spontaneous or shuffled
activity was averaged across 2000-ms “stimulus cycles” (see Methods), there was, in
each 200-ms time bin, a small percentage of the climbing fibers firing with a
probability 2 SD above or below the mean (controls, Fig. 3.2). However, in each time
bin, a similar percentage of climbing fiber had increased versus decreased firing,
indicating there was no signal carried by the population. In contrast, at any given time
point during the x0T/x0BG and x2T/x2BG training stimuli, a larger percentage of
climbing fibers fired 2 SD above or below baseline, and the percentage of climbing
fibers with increased versus decreased firing was highly asymmetric, reflecting the
signals carried by the population of climbing fibers during these standard stimuli.
During the x0T/x2BG training stimulus, climbing fiber activity was indistinguishable
from spontaneous climbing fiber activity and shuffled spike trains. Furthermore,
inspection of the raw climbing fiber spike trains revealed no evidence for a
temporally-specific response to the x0T/x2BG training stimulus on a finer timescale or
within a subset of trials.
Although the conflicting background motion during the x0T/x2BG stimulus
eliminated the climbing fiber responses, it did not affect the overall average firing rate.
On average, the overall firing rate
in the climbing fibers during the
x0T/x2BG training stimulus was
the same as the average firing
rate during spontaneous activity,
and during x0T/x0BG and
x2T/x2BG training stimuli (P >
0.05, ANOVA, Fig. 3.4). Hence,
the x0T/x2BG stimulus achieved
61
selective elimination of instructive signals in the climbing fibers without affecting the
baseline firing rate, as required to test the necessity of the instructive signals in the
climbing fibers for learning.
3.2.3 Instructive signals carried by Purkinje cell simple spikes
One candidate instructive signal that was previously proposed to guide motor learning
in the VOR is the simple spike output of the Purkinje cells, and during the novel
x0T/x2BG stimulus, the Purkinje cell simple spikes carried robust signals that could
potentially guide the induction of learning. As with the climbing fibers, the timing of
peak simple spike activity relative to the head movement discriminated between the
standard, x0 and x2 training stimuli when the target and background moved together
(x0T/x0BG, x2T/x2BG, Fig. 3.5). Unlike the climbing fibers, the Purkinje cell simple
spike activity also carried large, potentially useful instructive signals during
x0T/x2BG training (Fig. 3.5). The simple spike activity of the Purkinje cells encodes
both the vestibular input and the eye movements the monkey makes to track the visual
target[16]. The vestibular stimulus was the same across the experiments. The
tracking eye movements were also similar during x0T/x0BG and x0T/x2BG training,
because motion of the visual target was the same. Therefore, the Purkinje cell simple
spike responses during x0T/x2BG training were indistinguishable from those during
x0T/x0BG training (P > 0.19, paired t-test; Fig. 3.5, Table 3.1), thus providing a way
to assess the potential contribution of Purkinje cell simple spikes to the induction of
learning in the absence of instructive signals in the climbing fiber.
62
3.2.4 Learning in the absence of instructive signals in the climbing fiber
Despite the absence of instructive signals in the climbing fibers, the x0T/x2BG
training stimulus induced consistent motor learning. In the two animals used for
neural recordings, motor learning in the VOR was induced by presenting one of the
training stimuli for one hour (Monkey E) or two hours (Monkey L). VOR learning
was assessed by comparing the eye movement response to head movements in
complete darkness before and after training. The standard, x0T/x0BG and x2T/x2BG
training stimuli induced decreases and increases in VOR gain, respectively. The
x0T/x2BG training stimulus also induced a significant decrease in VOR gain (P <
63
0.001, one sample t-test; 4 of 4 training sessions in Monkey L; 5 of 5 training sessions
in Monkey E; Fig. 3.6).
Comparison of the x0T/x2BG stimulus with other novel stimuli revealed that training
stimuli that elicited similar instructive signals in the climbing fibers could induce
opposite changes in the behavior, which were correlated with the simple spike
responses. In Monkey L, the climbing fiber response to the x2T/x0BG stimulus was
not statistically different from the response during the x0T/x2BG stimulus (P > 0.05,
paired t-test; Table 3.1). Nevertheless, in 3 of 3 training sessions, x2T/x0BG training
induced a learned increase in VOR gain in Monkey L (Fig. 3.3), in contrast to the
decrease in VOR gain induced by x0T/x2BG training in 4 of 4 training sessions in the
same monkey (Fig. 3.6). The simple spike response elicited by the x2T/x0BG
stimulus was similar to that elicited by the x2T/x2BG stimulus (P > 0.05, paired t-test;
Table 3.1, Fig. 3.3 and Fig. 3.6), and therefore may have contributed to the increase in
VOR gain observed in each case.
64
In Monkey E, the climbing fiber response to the x2T/x0BG training stimulus was
significantly different from the response to the x0T/x2BG stimulus (P = 0.002, one
sample t-test; Table 3.1). Therefore, we designed additional training stimuli for this
monkey. Three training stimuli— x0T/x2BG, plus x0.5T/x1BG and x1.5T/x0.5BG
65
(see Methods for detailed description of training paradigms), each elicited no
significant climbing fiber response (Fig. 3.6a, see Appendix for individual neural
responses). However, the target motion was different during each of these three
training stimuli, which elicited different eye movements and, therefore, different
Purkinje cell simple spike responses (Fig. 3.6b). This provided an opportunity to
assess the ability of different simple spike responses to induce learning. Although the
climbing fiber responses to these three stimuli were the same, the learned changes in
VOR gain they induced were different (Fig. 3.6c, Table 3.1). Moreover, the changes
in VOR gain were correlated with the simple spike responses during training; for these
three stimuli, the biggest decreases in VOR gain occurred when the Purkinje cell
simple spike responses were most similar to those during the standard, x0T/x0BG
training stimulus.
3.2.5 Learning in the absence of instructive signals in the simple spikes
The observation of motor learning in the absence of instructive signals in the climbing
fibers does not exclude a contribution of climbing fiber-triggered plasticity
mechanisms to VOR learning. An additional training stimulus, x1T/x0BG, was used
to isolate the climbing fiber contribution by eliminating the putative instructive signals
in the Purkinje cell simple spikes. The x1T/x0BG stimulus consisted of head
movement paired with an earth-stationary target plus a background moving exactly
with the head. This stimulus drove the climbing fibers to respond in a manner similar
to their response during x0T/x0BG training, which induces a decrease in VOR gain
(compare x0T/x0BG and x1T/x0BG in Fig. 3.6, see Appendix for individual neural
responses). In contrast, the simple spike responses of the Purkinje cells were
eliminated in Monkey L, and reversed in Monkey E (Fig. 3.6), so that it was more
similar to the response elicited by the x2T/x2BG stimulus, which induces an increase
in VOR gain. At the behavioral level, training with the x1T/x0BG stimulus induced a
decrease in VOR gain in both monkeys (Fig. 3.6c), as one might predict from the
climbing fiber response.
66
3.2.6 Purkinje cell simple spike and climbing fiber signals together predict
behavioral changes
Our results indicate that motor learning can occur when information about the required
direction of learning is carried by both the climbing fibers and Purkinje cell simple
spikes, or when only one of these two instructive signals is available. To evaluate how
instructive signals carried by the simple spikes and climbing fibers may interact during
the induction of motor learning in the VOR, we recorded climbing fiber and Purkinje
cell simple spike responses to many novel training stimuli, and measured the
effectiveness of each training stimulus at inducing VOR learning. Each training
stimulus paired head movements with a different combination of target and
background motion. In total, 16 training stimuli were compared in Monkey L, and 17
training stimuli were compared in Monkey E. These stimuli elicited different
combinations of instructive signals in the climbing fibers and Purkinje cell simple
spikes (Fig. 3.7a), which allowed us to analyze their individual contributions to
learning. Some pairs of training stimuli elicited similar Purkinje cell simple spike
responses, but different climbing fiber responses (e.g. compare stimuli b and e in
Monkey L; Fig. 3.7a). Other pairs of training stimuli elicited similar climbing fiber
responses, but different Purkinje cell simple spike responses (e.g. compare stimuli e
and i in Monkey L; Fig. 3.7a).
67
We conducted a pairwise analysis to evaluate how each neural signal may contribute
to learning. For each pair of training stimuli, we calculated the difference in climbing
fiber response (∆climbing fiber), the difference in Purkinje cell simple spike response
(∆simple spike), and the difference in learning (∆learning). The 16 training stimuli
used in Monkey L yielded 120 pairwise comparisons, and the 17 training stimuli used
in Monkey E yielded 136 comparisons.
If climbing fibers provide instructive signals guiding learning, then difference in
learning induced by any pair of training stimuli should be related to the difference in
the climbing fiber signals that they elicit during training. Indeed, ∆climbing fiber was
linearly correlated with ∆learning (Fig. 3.7b). We then divided all the stimulus pairs
into three groups on the basis of the ∆simple spike value associated with each pair.
68
For a given ∆climbing fiber, ∆learning systematically varied with ∆simple spike.
Moreover, ∆learning was linearly correlated with ∆simple spike, particularly when the
data were grouped according to ∆climbing fiber (Fig. 3.7c). This influence of both
∆simple spike and ∆climbing fiber on ∆learning is consistent with an independent
contribution of both the climbing fibers and simple spikes to learning.
The contributions of the two putative instructive signals to learning were estimated by
using the average correlation coefficients obtained from the pairwise analyses to
predict the behavioral changes induced by each training stimulus (see Methods).
When only the signals in the climbing fibers were considered, the predicted learning
was well correlated with the observed learning (R2 = 0.85, Monkey L; 0.78, Monkey
E), however, the amount of learning was generally underestimated by 30-40% (Fig.
3.7d, correlation coefficient = 0.72 in Monkey L, 0.61 in Monkey E). The learning
predicted from the two neural signals together was well correlated with the observed
learning (R2 = 0.89, Monkey L; 0.81, Monkey E). Moreover, the correlation
coefficients were close to the ideal value of 1.0, which would be attained if the
predicted learning was equal to the observed learning (black lines, Fig. 3.7d,
correlation coefficient = 1.02 in Monkey L, 0.86 in Monkey E). Thus, a linear
combination of climbing fiber and Purkinje cell instructive signals better accounted for
motor learning in the VOR, compared to climbing fiber signals alone.
69
3.3 Discussion
We designed visual-vestibular stimuli that did not affect baseline firing, but eliminated
instructive signals carried by neurons that have been implicated in VOR learning. The
x0T/x2BG training stimulus eliminated putative instructive signals carried by the
climbing fibers, and the x1T/x0BG training stimulus eliminated putative instructive
signals carried by the Purkinje cell simple spikes, yet each of these training stimuli
induced robust motor learning in the VOR. Thus, our data indicate that neither
instructive signals carried by climbing fibers nor instructive signals carried by
Purkinje cell simple spikes are necessary for learning. Rather, each of these neural
instructive signals may operate in parallel, with each capable of inducing learning in
the other’s absence.
3.3.1 VOR circuit physiology and Marr-Albus-Ito model
Models of motor learning in the VOR are constrained by a great deal of information
about how activity at each site in the circuit should affect the gain of the VOR.
Purkinje cells are the sole output of the cerebellar cortex, therefore plasticity
mechanisms in the cerebellar cortex can influence the gain of the VOR only via their
effects on Purkinje cell output during VOR performance, which is tested during head
turns in total darkness. For a change in Purkinje cell output to cause a decrease in
VOR gain, the Purkinje cells should fire less during contraversive head movement
and/or more during ipsiversive head movement, so that inhibitory input from the
Purkinje cells to VOR interneurons in the vestibular nucleus counteracts some of the
excitatory drive to the interneurons from the vestibular afferents, which increase their
firing during ipsiversive head movement. On the other hand, a decrease in inhibition
from the Purkinje cells during ipsiversive head movement would create a bigger
response in the VOR interneurons, and hence, an increase in VOR gain (Fig. 3.1).
The Marr-Albus-Ito model attributed motor learning to climbing fiber-triggered
changes in the weights of parallel fiber synapses onto Purkinje cells. For the
70
appropriate changes in Purkinje cell output to be accomplished by LTD at the parallel
fiber-to-Purkinje cell synapses, there should be selective depression of those parallel
fibers that fire during one specific phase of the head movement. A decrease in VOR
gain would require selective depression of those parallel fibers that provide excitatory
input to the Purkinje cells during contraversive head movement, whereas an increase
in VOR gain would require selective depression of the parallel fibers that fire during
ipsiversive head movement (Fig. 3.1). Since climbing fiber activation can trigger LTD
in parallel fibers active simultaneously[74], the responses present during the
x0T/x0BG and x2T/x2BG training stimuli would be expected to trigger LTD in the
appropriate vestibular parallel fibers to account for the observed changes in VOR gain
(Fig. 3.1). In contrast, there is no reason to believe that the x0T/x2BG training
stimulus could induce selective LTD of the appropriate parallel fibers to produce the
observed decrease in VOR gain. During x0T/x2BG training, the climbing fibers fired
with the same probability during ipsiversive and contraversive head movement,
making it equally probable that parallel fibers active during contraversive or
ipsiversive head movements would undergo LTD. LTD of both groups of parallel
fibers may decrease the average firing rate of the Purkinje cells, but should not cause
the change in firing rate modulation during head movements required to decrease the
VOR gain. Moreover, if climbing fiber activity at the spontaneous rate is not effective
at inducing LTD, then there may be no LTD at the parallel fiber-to-Purkinje cell
synapses during x0T/x2BG training.
3.3.2 Could another population of climbing fibers induce changes in VOR gain?
During standard, x0T/x0BG and x2T/x2BG, training stimuli, the population of
climbing fibers we recorded carried robust signals, which were previously
hypothesized to guide VOR learning. We compared the responses in the same
individual neurons to many different training stimuli. The very same climbing fibers
that carried no instructive signals during our novel training stimuli (x0T/x2BG,
x0.5T/x1BG, and x1.5T/x0.5BG) carried robust and uniform instructive signals during
standard training (x0T/x0BG or x2T/x2BG, Fig. 3.2), and therefore our sample is
71
drawn from the same population that has been previously implicated in VOR learning.
It is not likely that there is another population of unrecorded climbing fibers that carry
signals during the training stimuli that elicited no responses in the recorded climbing
fibers. Within the floccular complex, we tested all of the Purkinje cells we isolated
with any task-related
activity, and found no
significant response of their
climbing fiber inputs during
the x0T/x2BG training
stimulus (P > 0.05, one-
sample t-test; Fig. 3.8). It is
unlikely that climbing fibers
in other parts of the
cerebellum carry instructive
signals to guide motor
learning in the VOR. There
is compelling evidence from several labs that VOR learning requires the floccular
complex, and not other regions of the cerebellum. Floccular Purkinje cells undergo
learned changes in their output during VOR performance, which should support the
observed changes in VOR gain (Fig. 3.1)[13, 15-17]. Lesions of the floccular
complex completely abolish VOR gain learning[19, 67]. In contrast, lesions of other
regions of the cerebellum that receive vestibular input, including the uvula/nodulus,
lobule VI, and lobule VII[75, 76], do not affect VOR learning[67-69].
72
It is also unlikely that a
subpopulation of the climbing
fibers we recorded during the
x0T/x2BG stimulus can
account for the decreases in
VOR gain induced by this
stimulus. Although the
population response during
the x0T/x2BG stimulus was
not significantly different
from zero, there was variation
around the mean response of
zero (Fig. 3.2). Therefore, a
subset of the individual
climbing fibers (21/39) had small responses in the correct, “gain decrease” direction
during the x0T/x2BG stimulus to account for the observed decrease in VOR gain.
However, if one considers the cycle-by-cycle variability of the responses, only 2/21
climbing fibers had significant “gain decrease” responses during the x0T/x2BG
stimulus. Moreover, 18 climbing fibers had small responses in the incorrect, “gain
increase” direction, three of which were significant. Thus, any effects of the climbing
fibers with small “gain decrease” responses should be cancelled by the effects of the
equal number of climbing fibers with small “gain increase” responses, unless the
Purkinje cells receiving the “gain decrease” climbing fibers have unique properties
that endow them with privileged control over the behavior, and there was no evidence
for that. The neurons receiving small “gain increase” versus “gain decrease”
instructive signals during x0T/x2BG training were indistinguishable in all other
respects (Fig. 3.9, Table 3.2). Finally, the responses of the full population of climbing
fibers during x0T/x2BG training were normally distributed, consistent with the
responses being drawn from a single, uniform population of neurons rather than two
distinct populations (P = 0.26; D’Agostino-Pearson test). Thus, the variation around
73
the mean climbing fiber response of zero during the x0T/x2BG stimulus appears to be
biological noise, rather than any kind of signal that could guide the induction of the
observed behavioral changes.
3.3.3 Other instructive signals for cerebellum-dependent motor learning
In the absence of climbing fiber instructive signals, learning was correlated with the
Purkinje cell simple spike responses present during training, suggesting that the
signals carried by the simple spikes may contribute to the induction of learning.
Purkinje cell simple spikes have been previously proposed as an instructive signal for
motor learning[1]. Purkinje cells are well positioned to control the induction of
plasticity in their main target, the vestibular nucleus (VN) or deep cerebellar nuclei
(DCN), and there are several lines of evidence for changes within the VN/DCN during
learning[24, 45, 77]. The cellular mechanisms for such changes are not fully
understood, but it is plausible that the correlation of activity between the vestibular
afferent and Purkinje cell inputs to the vestibular nucleus could induce synaptic
changes that decrease or increase VOR gain.
Although some models of cerebellum-dependent learning include sites of plasticity
outside the cerebellar cortex, plasticity at these sites has been viewed as secondary to
74
LTD at the parallel fiber-to-Purkinje cell synapses (pf-Pk LTD), or dependent on pf-
Pk LTD for their appropriate expression. For example, a model of eyeblink
conditioning proposes that pf-Pk LTD causes changes in the Purkinje cell simple spike
output, which in turn induces plasticity in the DCN[54]. According to this model,
learned changes in the DCN are not expressed if the pf-Pk LTD is reversed[78]. Other
recent work has suggested that the consolidation of cerebellum-dependent learning
involves a transition from an initial dependence on pf-Pk LTD to a dependence on
other plasticity mechanisms [50, 77]. Thus, despite the accumulating evidence that
multiple plasticity mechanisms contribute to learning, climbing fiber-triggered
plasticity has been viewed as central to and necessary for cerebellum-dependent
learning. In contrast, our current results suggest that instructive signals carried by
Purkinje cell simple spikes may induce learning in the absence of any climbing fiber-
triggered plasticity.
Our experiments measured the neural
instructive signals available to guide
learning at the beginning of training,
when the gain of the VOR was at
baseline. Each training stimulus was
presented for just 1-2 min so that the
responses of a single neuron to many
stimuli could be compared. It is
possible that climbing fiber responses
to the training stimuli could emerge as
learning progresses, even if no
response was present at the beginning
of training. However, most of the
behavioral changes occurred early in
the training session (Fig. 3.10) and
therefore cannot depend on any late-
75
developing climbing fiber responses. Thus, even if climbing fibers did make a small
contribution to learning late in x0T/x2BG training, it would have to be secondary to a
climbing fiber-independent plasticity mechanism that drove the initial changes in the
circuit.
3.3.4 Multiple instructive signals during cerebellum-dependent motor learning
Our results suggest that neither instructive signals carried by the climbing fibers nor
the Purkinje cell simple spikes are necessary for motor learning. However, either
neural instructive signal appears to be sufficient to support learning, thus imparting the
cerebellum with distinct, independent ways to control the induction of motor learning.
We cannot rule out the possibility that other neural instructive signals, besides those
carried by climbing fibers and Purkinje cell simple spikes, could contribute to the
induction of learning. However, for the range of stimuli tested, the learned changes in
behavior were well predicted from the signals present in the climbing fibers and
simple spikes during training. The pairwise analysis in Fig. 3.7 suggests that changes
in VOR gain are a linear function of the climbing fiber response amplitude during
training and a linear function of the Purkinje cell simple spike response amplitude
during training. When both neural instructive signals are present during training, their
effects appear to sum linearly, suggesting that climbing fiber and Purkinje cell
instructive signals operate independently and in parallel during the induction of
learning.
The independent operation of multiple instructive signals offers three advantages.
First, it provides more than one way to achieve a similar behavioral outcome. The
learning induced by x0T/x2BG training and x1T/x0BG training is similar at the
behavioral level, in each case a decrease in VOR gain. However, different neural
instructive signals are available during each training stimulus. Therefore, the
underlying memory traces may be quite different, potentially involving distinct
locations within the circuit. Second, the use of multiple instructive signals could allow
76
learning to occur under a broader range of conditions, since each neural instructive
signal encodes different aspects of the learning environment, with the climbing fibers
more sensitive to motion of the visual background and the Purkinje cells more
sensitive to the visually-driven eye movements made during training. Third, when
multiple instructive signals are recruited, more learning is induced. Thus, to optimize
motor learning in clinical or other settings, one should design the training environment
to recruit each of the available instructive signals.
77
3. 4 Materials and methods
Surgical preparation, calibration, behavioral experimentation, and data analysis were
performed as described in Chapter 2.
To test for temporally-restricted climbing fiber responses, each cycle was divided into
ten 200-ms epochs and the average probability of a complex spike was calculated for
each epoch. When spontaneous activity was used as a control condition, climbing
fiber responses were divided into 2000-ms “trials” and aligned to the onset of
recording, and the same analysis was performed as on the training stimuli.
The baseline probability of climbing fiber firing was estimated from all recordings of
spontaneous activity in climbing fibers. Spike trains of spontaneous activity were
divided into 200-ms bins, and the probability of complex spike firing was calculated
from a random selection, without replacement, of 35 200-ms bins (the typical number
of bins used to calculate the firing probability during a training stimulus). This
measure was repeated to derive the mean and standard deviation used as the baseline
probability.
Pairwise comparison analysis
Pairwise comparison analysis in Fig. 3.7 was based on the neural and behavioral
responses to 16 training stimuli for Monkey L and 17 training stimuli for Monkey E.
For each training stimulus, the average response in the climbing fibers and Purkinje
cell simple spike populations were calculated by averaging the components of the
individual neural response aligned with peak head velocity. For learning, the median
value of all behavioral replications was used. For each pair of training stimuli, the
differences in Purkinje cell simple spike responses, climbing fiber responses, and
78
learning were calculated. For Monkey L, the 16 training stimuli yielded 120 total
comparisons, and the 17 training stimuli yielded 136 total comparisons for Monkey E.
To assess the contribution of climbing fiber signals to learning, the range of
differences in Purkinje cell simple spike responses was divided into five bins of equal
size, and each stimulus pair was sorted into a bin according to its difference in
Purkinje cell simple spike response. A linear regression was performed between the
differences in learning and the differences in climbing fiber signals for the stimulus
pairs in a given bin. The correlation coefficients obtained from the five bins of
stimulus pairs were averaged, and this value (CCF) was used to estimate the climbing
fiber contribution to learning.
To assess the contribution of simple spike signals to learning, the range of differences
in climbing fiber responses was divided into five bins of equal size, and each stimulus
pair was sorted into a bin according to its difference in climbing fiber response. A
linear regression was performed between the differences in learning and the
differences in simple spike signals for the stimulus pairs in a given bin. The
correlation coefficients obtained from the five bins of stimulus pairs were averaged,
and this value (CSS) was used to estimate the Purkinje cell simple spike contribution to
learning.
The coefficients, CCF and CSS, derived from the pairwise analysis were used to predict
the amount of learning induced by each training stimulus. To predict the amount of
learning based only on the climbing fiber instructive signals present during each
training stimulus, the following equation was used:
LCF(stimulus) = CCF * CFmeasured(stimulus)
where LCF(stimulus) is the predicted learning for a training stimulus and
CFmeasured(stimulus) is the measured climbing fiber response during that stimulus.
79
To predict the amount of learning, Ltotal(stimulus), based on both Purkinje cell simple
spike and climbing fiber responses, the amount of learning predicted from each signal
was calculated and then summed:
Ltotal(stimulus) = CCF * CFmeasured(stimulus) + Css * SSmeasured(stimulus)
where SSmeasured(stimulus) is the measured simple spike response during the training
stimulus.
80
Chapter 4 Specificity and generalization of motor learning in the VOR
4. 1 Introduction
Once motor learning has occurred, it can only be useful if it is expressed in
appropriate contexts. To some extent, learning must be able to generalize to contexts
slightly different from those used for training, otherwise natural situational variations
may restrict learning from ever being expressed. On the other hand, learning can be
maladaptive if expressed in an inappropriate context.
One example of context dependency of motor learning in the VOR is the specificity of
adapted gain to head tilt present during training. In a previous study, gain up training
of rotational VOR with the head tilted at 90 degrees to the right induced a gain
increase that was not at all expressed when the head was tilted 90 degrees to the left
[79]. This finding suggests an interaction between neural signals encoding
information from the semicircular canals, which measures head rotation, and the
otolith organs, which measure head tilt. The interactions of these signals during motor
learning in the VOR with head tilt to the left and to the right must activate non-
overlapping populations of neuronal synapses undergoing plasticity since no learning
was expressed when the head was tilted in the direction opposite that present during
training.
Another example of stimulus specificity for motor learning in the VOR is the
expression of learning when measured at head-rotation frequencies different from the
training frequency. If training is induced with visual-vestibular stimuli at a single
frequency, the observed change in VOR gain is much smaller when measured at head
movement frequencies different from the one used for training [22, 80]. This finding
suggests that the circuit for the VOR contains parallel, frequency-tuned signal-
processing channels that can each be modifiable during training. Interestingly, the
amount of training observed at the non-training frequencies is dependent on the
training frequency. For example, if training is done with 0.5 Hz stimuli, there is little
81
expression of learning when the VOR gain is measured with 5 Hz head rotation[80].
When training is done with 5 Hz stimuli, however, learning is expressed when VOR
gain is measured at 0.5 Hz [22, 80]. One possible explanation for this observed
difference in generalization with high versus low frequency training is their
dependence on different plasticity mechanisms. Previous studies (and studies from
Chapter 2 and 3) have shown that instructive signals carried by both Purkinje cell
simple spikes and climbing fiber signals could guide learning at low-frequency;
whereas climbing fibers are the best candidate instructive signals to guide plasticity
during high-frequency training [6].
As described in previous chapters, the behavioral manipulations using a target and
visual background paired with head movements to induce motor learning in the VOR
allowed us to titrate the amount of neural instructive signals present in Purkinje cells
and climbing fibers. Examining the generalization profile, or the amount of learning
at the non-training frequencies during training with complex visual-vestibular stimuli,
may provide insights to how climbing fiber-triggered plasticity mechanisms versus
Purkinje cell-triggered plasticity mechanisms contribute to the generalization of motor
learning in the VOR.
82
4.2 Results
4.2.1 Generalization of learned changes in VOR gain
Motor learning in the VOR was typically assessed by measuring changes in VOR gain
at the frequency used for training. We also measured the effectiveness of the training
stimulus to induce gain changes at the non-training frequencies, to assess if learning
induced at one frequency generalized to other frequencies. In our experiments, we
induced motor learning in the VOR with visual-vestibular stimuli at 0.5 Hz frequency,
and tested the generalization of learning at 4 other non-training frequencies (0.2, 1, 2
and 5 Hz) in Monkeys L and C. Figure 4.1 plots the learned gain changes at the
training (0.5 Hz) and non-training frequencies (0.2, 1, 2, and 5 Hz), elicited by x0 and
x2 consistent and conflicting training conditions.
In general, the learned changes in VOR gain generalized to non-training frequencies.
Conflicting conditions induced smaller gain changes than consistent conditions at the
training frequency, however, the generalization profiles between consistent and
conflicting conditions were similar, in that the learning induced at the training
frequency generalized to other frequencies.
83
To compare the overall pattern of generalization, a generalization index was derived
from the data[8]. It was calculated as the fraction of learning at the training frequency
that was expressed, on average, at test frequencies other than the training frequency. It
was calculated as:
∆Gain(i) represented the percent change in VOR gain measured at one of the j head
rotation frequencies other than the training frequency, and ∆Gain (train) was the
percentage change in gain at the training frequency. Increasing values of the
generalization index indicated increasing extents of generalization. If the average
change in the non-training frequencies was equal to ∆Gain(train), then the index was
1. If the average change in the non-training frequencies was 0, then the index was 0,
indicating specificity. The index is greater than 1 if the changes in the non-training
frequencies were larger than the change at the training frequency. Lastly, the index
was less than 0 if the changes in the non-training frequencies were maladaptive.
The calculated generalization indices for Monkey L and Monkey C during x0T/x0BG
were 0.67 and 0.96, respectively. For
x2T/x2BG in Monkey L and C, the
values were 0.39 and -0.05,
respectively. These results
demonstrated that increases in gain
generalized less than decreases in VOR
gain, and these findings were also
replicated in mice studies[8]. The
smaller changes in VOR gain induced
by conflicting stimuli at training and
non-training frequencies produced a noisy range of generalization indices, which
showed no systematic trend of the effect of conflicting stimuli on generalization (Fig.
84
4.2). During these conditions, small absolute changes in gain have a big impact on the
index value.
4.2.2 Can VOR gain changes at the non-training frequencies be explained by a
combination of climbing fiber and Purkinje cell signals?
We analyzed if gain changes at the non-training frequencies can be explained by a
combination of climbing fiber and Purkinje cell signals that were recorded at the
training frequency. If both signals contribute to VOR gain changes at the non-training
frequencies, then we may expect both signals to account for learning, similar to how
both signals account for learning at the training frequency (see Chapter 3). On the
other hand, only one of the two signals may selectively contribute to learning
expressed at the non-training frequencies.
VOR gain changes at the non-training frequencies were measured for the 16
conditions in Monkey L and C. We conducted a pairwise comparison analysis as
explained in Chapter 3 to analyze the contributions of each instructive signal to
learning at non-training frequencies. For each pair of stimuli, we calculated the
difference in Purkinje cell responses, and ranked ordered the pairs according to their
difference. We took pairs of stimuli that were within 10 percent (± ~4 spikes/s) of the
maximal difference in Purkinje cell response (± ~40 spikes/s). These stimuli pairs
represent pairs of training conditions that elicited similar Purkinje cell signals. Next,
we calculated the difference in climbing fiber responses and learning between the
pairs, and plotted these differences. If climbing fiber signals contribute to VOR gain
changes, we would expect a correlation between the difference in learning and the
difference in climbing fiber signals when Purkinje cell responses were held nearly
constant. We observed a correlation in climbing fibers signals elicited by the training
stimulus at 0.5 Hz to VOR gain changes induced at non-training frequencies (0.2, 1, 2,
and 5 Hz) for Monkey L (Fig. 4.3). In Monkey C, there was no correlation at the non-
training frequencies. In Monkey C, the gain changes induced at the training
frequencies were smaller (Fig. 4.1), compared to the changes induced in Monkey L.
85
Since the change at the training frequency was small, the VOR gain changes at the
non-training frequencies were also comparably smaller, or no changes were induced
despite robust climbing fiber signals being elicited at the training frequency. These
small changes or lack of changes at the non-training frequencies could account for the
lack of correlation observed in Monkey C.
For each test frequency, we determined the learning rate attributed to climbing fiber
instructive signals by calculating the regression coefficients for the data in Fig. 4.3 for
Monkey L. The values were 16.0% ∆VOR gain/sp/s for 0.2 Hz, 18.8% ∆VOR
gain/sp/s for 1Hz, 16.8% ∆VOR gain/sp/s for 2 Hz, and 11.9% ∆VOR gain/sp/s for 5
Hz. Using these coefficients, we calculated the amount of learning attributed to the
climbing fiber signals by multiplying the learning rate with the measured climbing
fiber response for each training condition. We subtracted the amount of learning
attributed to the climbing fibers from the observed learning. This unaccounted,
residual amount of learning was then compared to the Purkinje cell responses elicited
by the training condition (Fig. 4.4).
86
The Purkinje cell responses were correlated to the residual learning during 0.2 and 1
Hz stimuli, moderately correlated during 2 Hz stimuli, and not correlated during 5 Hz
stimuli (Fig. 4.4). These results suggest that Purkinje cell signals can account for
learning-related changes not attributed to climbing fiber signals during 0.2, 1, and 2
Hz test frequencies. Regression coefficients were calculated for each frequency in
Figure 4.4 to determine the learning rate associated with Purkinje cell signals. The
Purkinje cell and climbing fiber learning rates were then used to predict the individual
contributions of each signal to learning. With the exception of learning-related
changes at 5 Hz, a linear combination of Purkinje cell and climbing fiber signals can
account for learning-related changes at non-training frequencies (Fig. 4.5).
87
Our visual-vestibular stimuli elicited many different combinations of Purkinje cell and
climbing fiber responses, and some visual-vestibular stimuli elicited no response in the
climbing fiber response (e.g. x0T/x2BG), while other stimuli elicited no response in
the Purkinje cells (e.g. x1T/x0BG).
According to the linear model that predicts learning using both signals, Purkinje cells
would account for nearly 100 percent of the learning when climbing fiber signals were
absent, and account for very little learning when the Purkinje cell responses were
absent. We asked if the contributions of Purkinje cell and climbing fiber signals to
VOR gain change were similar at training versus non-training frequencies. We took
our 16 conditions in Monkey L and at
each test frequency (with the exception
of 5 Hz where there was no correlation
between learning and Purkinje cell
signals), calculated the VOR gain change
predicted by the Purkinje cell signals,
and divided it by the total gain change
predicted by the Purkinje cell and
climbing fiber signals. These
calculations reflect the percentage of the
total learning attributed to the Purkinje
cell instructive signal. We then compared the percentage of the total learning
attributed to the Purkinje cell signals at the training frequency to the percentages at the
non-training frequencies (Fig 4.6). At the non-training frequencies, Purkinje cells
consistently accounted for 5-15 percent more of the total learning. Both Purkinje cells
and climbing fiber signals were needed to account for the total change in VOR gain at
non-training frequencies. However, a greater Purkinje cell contribution at non-training
frequencies suggest that Purkinje-cell triggered plasticity mechanisms contributed
relatively more to the generalization of learned changes in VOR gain.
88
4.3 Discussion
4.3.1 Neuronal population coding schemes to achieve generalization and
specificity
Generalization and specificity of motor learning in the VOR can each be achieved by a
population coding scheme that features either sparse encoding or overlapping
representations. In sparse encoding, different contexts are represented in non-
overlapping population of neurons. Therefore, during a specific training context a
distinct set of neurons are activated and modified. For learning to be expressed, the
modified pathway is only triggered if the specific context is present, since non-training
condition cannot activate the modified pathway. On the other hand, in a network
using overlapping representations, different contexts are represented by overlapping
population of neurons. In this model, a specific training context will activate and
modify a distinct set of neurons, but since these neurons can also be activated by other
contexts, learning can be expressed in a non-training condition.
Neurons in the circuit for the VOR have been found to have different tuning
properties. In the vestibular nucleus, neurons exhibit a wide range of tuning to head-
rotation frequency, with some neurons encoding a specific frequency, while others
encode a wide range of frequencies [81, 82]. The tuning properties of granule cells in
the cerebellar cortex have not been studies, but the large number of granule cells may
permit each neuron to be tightly tuned to a particular context when a movement is
made [2, 4]. Therefore, one possible hypothesis has been that learning expressed in a
narrow range of stimuli would be mediated by plasticity in the cerebellar cortex,
whereas learning that generalized broadly would be mediated by plasticity in the deep
cerebellar nuclei (DCN) or vestibular nucleus (VN).
4.3.2 Generalization of VOR gain increases versus decreases
Since increases in VOR gain generalize less than decreases in VOR gain, this “cortex-
nucleus” model suggests that increases in VOR gain are more dependent on plasticity
in the cerebellar cortex, whereas decreases in VOR gain are dependent on plasticity in
89
the vestibular nucleus. This hypothesis, however, is at odds with several studies of
cerebellum-dependent learning. Post-training lesions of the cerebellar cortex resulted
in similar effects on both learned increases and decreases in VOR gain [83, 84]. In
another form of cerebellum-dependent motor learning, eyeblink conditioning,
disconnecting the cerebellar cortical input to the DCN did not degrade the specificity
of learned eye blinks for the tone used in conditioning, suggesting that plasticity in the
DCN supported stimulus-specific learning. Additionally, our current study
demonstrated that VOR gain changes at non-training frequencies can be best
accounted for by both Purkinje cell simple spike and climbing fiber instructive signals.
Since the simple spikes are more likely to guide plasticity in the vestibular nucleus,
whereas climbing fibers are more likely to act on neurons in the cerebellar cortex, our
results in combination with previous studies suggest that it is unlikely that increases
and decreases in VOR gain rely on entirely separate sites in the VOR circuit.
4.3.3 Patterns of generalization for high versus low frequency training
Previous studies have shown that when VOR learning is induced with low frequency
stimuli (0.5 Hz), learned changes in VOR gain do not generalize as well as when VOR
learning is induced with high frequency stimuli (5 Hz) [22]. The “nucleus-cortex”
model would predict that changes induced by low frequency stimuli would be
localized to the cerebellar cortex, whereas changes induced by high frequency stimuli
would be localized to the brainstem.
However, studies of the available neural instructive signals to guide VOR learning
predict opposite outcomes. During high frequency training (5 Hz), Purkinje cell
simple spikes do not provide instructive signals that can discriminate between training
conditions that induce a gain increase versus gain decrease, and therefore, it is unlikely
any changes in the brainstem could account for the observed generalization of
learning. Rather, the best candidate instructive signal is the climbing fiber [6].
However, climbing fibers do project to the vestibular nucleus [33] and could mediate
plasticity events that can be expressed in non-training contexts.
90
To successfully stabilize images on the retina across varying conditions, the circuit for
the VOR must coordinate the specificity or generalization of learning across many
dimensions of sensory input, including both vestibular stimulus parameters and
context cues. For example, the vertical VOR can be adapted to different gains during
upward, versus downward, head turns[85]. Subjects can be trained to have different
horizontal VOR gains when their eyes are angled upward versus downward[86].
Learning is also specific for the vergence angle held by the eyes during training[87].
Motor learning in the translational VOR is expressed only in one eye, if the other is
not required to move in response to the training stimulus[88]. In the real world,
learning is regulated by all of these dimensions simultaneously, and so the underlying
coding and plasticity mechanisms in the VOR must be of sufficient richness to enable
regulation of learning by multiple dimensions of sensory input.
91
4.4 Materials and Methods
Surgical preparation, calibration, behavioral experimentation, and data analysis were
performed as described in Chapter 3, except that changes in VOR gain were assayed at
five frequencies (0.2, 0.5, 1, 2, and 5 Hz). All frequencies tested were used in the
calculation of the generalization index.
Pairwise comparison analysis in Fig. 4.3 was based on the neural and behavioral
responses to 16 training stimuli for Monkey L and Monkey C.
For each training stimulus, the average response in the climbing fibers and Purkinje
cell simple spike populations were calculated by averaging the components of the
individual neural response aligned with peak head velocity. For learning, the median
value of all behavioral replications was used. The difference in learning was
calculated for all test frequencies.
For each pair of training stimuli, the differences in Purkinje cell simple spike
responses and learning were calculated. To assess the contribution of climbing fiber
signals to learning, the range of differences in Purkinje cell simple spike responses
among the stimulus pairs was determined, and pairs within 10 percent (approx. 4
sps/s) of the maximal difference were extracted. For each test frequency, a linear
regression was performed between the differences in learning and the differences in
climbing fiber signals for the extracted stimulus pairs. In total, five correlation
coefficients (CCF) were calculated, each coefficient specific to a test frequency (0.2, 1,
2, 5 Hz).
To assess the contribution of simple spike signals to learning, the amount of learning
attributed to climbing fiber signals was first calculated for each test frequency by the
following equation:
92
LCF(stimulus, frequency) = CCF(frequency) * CFmeasured(stimulus)
where LCF(stimulus, frequency) is the predicted learning for a training stimulus at a
specific test frequency and CFmeasured(stimulus) is the measured climbing fiber
response elicited by a training stimulus. Next, the amount of learning unaccounted by
climbing fiber signals (Lnon-CF) was calculated:
Lnon-CF(stimulus, frequency) = Lmeasured(stimulus, frequency) - LCF(stimulus,
frequency)
where Lmeasured(stimulus, frequency) is the amount of measured VOR gain change
elicited by a training condition at a specific test frequency. For each test frequency, a
linear regression was performed between Lnon-CF(stimulus, frequency) and the
Purkinje cell simple spike signals elicited by each training condition. In total, five
correlation coefficients (Css) were calculated, each coefficient specific to a test
frequency (0.2, 1, 2, 5 Hz).
For each test frequency, the coefficients, CCF and CSS, were used to predict the amount
of learning induced by each training stimulus. To predict the amount of learning,
Ltotal(stimulus, frequency), based on both Purkinje cell simple spike and climbing
fiber responses, the amount of learning predicted from each signal was calculated and
then summed:
Ltotal(stimulus, frequency) = CCF(frequency) * CFmeasured(stimulus) +
Css(frequency) * SSmeasured(stimulus)
where SSmeasured(stimulus) is the measured simple spike response during the training
stimulus.
93
Chapter 5 Conclusions and future directions
The dominant theory for the last several decades has postulated that the climbing
fibers provide the neural instructive signals that control the induction of cerebellum-
dependent learning. However, studies in this thesis showed that climbing fiber-
independent plasticity can also support cerebellum-dependent learning, because motor
learning in the vestibulo-ocular reflex (VOR) can be induced with training stimuli that
do not elicit responses in the climbing fibers. We also generated evidence that the
simple spike output of cerebellar Purkinje cells provides the neural instructive signals
guiding the climbing fiber-independent component of motor learning; in the absence
of instructive signals in the climbing fibers, learned changes in VOR gain were
correlated with the simple spike responses during training.
The results from our current study suggest that the simple spike output of Purkinje
cells plays a role in providing the instructive signals guiding motor learning, but a
more limited role than the climbing fibers (Fig. 3.7). This may be because the set of
training stimuli used in the current study elicited only modest simple spike responses
(up to about ± 20 sp/s), compared to the physiological range (at least ± 50 sp/s),
whereas the climbing fiber responses elicited by the training stimuli (up to about ± 1
sp/s) represents a greater portion of their physiological range. Therefore it is possible
that the climbing fiber-triggered component of learning was near maximal whereas the
simple spike-triggered component was not maximal, but could actually make a
substantially larger contribution than these initial results revealed.
A future direction is to design visual-vestibular training stimuli to elicit different
simple spike responses, spanning much of the physiological range. Two general
approaches may be used—first, “stronger” training stimuli such as x3, x4, etc.
conditions should elicit bigger eye movements and therefore larger simple spike
responses during training. Climbing fiber instructive signals tended to saturate during
x0 and x2 consistent training conditions, so climbing fiber signals should remain
94
unaffected by these stronger stimuli. Second, it is possible to stabilize the target on
the retina during head movements to gain control over target and background slip, and
thereby eliminating climbing fiber responses[89]. Image stabilization during target
motion that elicits large eye movements could create big responses in the simple
spikes during head movements. The results of these experiments could reveal answers
to several questions: How much can the simple-spike triggered plasticity contribute to
changes in VOR gain? Does simple spike-triggered plasticity contribute equally to
increases and decreases in VOR gain? Does simple spike triggered plasticity
contribute to learned changes in VOR phase? To what extent do simple spike-
triggered changes in VOR generalize? Can climbing fiber activity regulate the simple
spike-triggered plasticity mechanism?
These future experiments would provide correlative evidence to support the role of
simple spike activity in the induction of motor learning in the VOR. However, such
correlations could be observed if the instructive signals come not from the Purkinje
cells, but another population of neurons whose activity covaries with the simple spike
activity of Purkinje cells. To test the sufficiency of Purkinje cell output for inducing
learning, one future direction is to test whether direct activation of the Purkinje cells
can induce learning when paired with head movements in the absence of visual
stimuli. Direct electrical stimulation of floccular Purkinje cells can be conducted in
primates or mice, while optogenetic activation using the channelrhodopsin-2 system
can be accomplished in mice [90].
What is the algorithm a neural circuit uses to learn? More specifically, for
cerebellum-dependent learning, which neurons are monitoring the accuracy of
movements, and how do they determine when the circuit controlling a movement
needs to be modified? The question of what controls the induction of plasticity in vivo
is one of the most central questions one can ask about the neural mechanisms of
learning, but it has not been studied extensively. For many decades, studies of
cerebellum-dependent learning have focused on the role of the climbing fibers in
95
carrying the neural instructive signals that control the induction of
plasticity. However, the studies in this thesis have shown that climbing fiber-
independent mechanisms also contribute to cerebellum-dependent learning. With the
insight that cerebellum-dependent learning is not a unitary process, the next step is
undertaking systematic research to dissect cerebellum-dependent learning into its
component parts, and identify the molecular, cellular, and electrophsyiological
underpinnings of each.
96
Appendix
Individual climbing fiber and Purkinje cell simple spike responses for each
training condition
The following are polar plots of individual climbing fiber and Purkinje cell simple
spike responses for each of the training conditions. Each point on the polar plots
represents the response recorded in a single Horizontal Gaze Velocity Purkinje cell
(HGVP) in Monkey L (purple), Monkey C (green) and Monkey E (orange). Distance
from the origin represents the amplitude of the response (change in firing rate above
and below baseline firing rate) and the phase represents the timing of the neural
response relative to peak head velocity. Clockwise rotation represents increased phase
lead.
97
98
-
99
100
101
102
103
104
105
106
107
108
109
110
111
112
Bibliography
1. Miles, F.A. and S.G. Lisberger, Plasticity in the vestibulo-ocular reflex: a new
hypothesis. Annu Rev Neurosci, 1981. 4: p. 273-99. 2. Marr, D., A theory of cerebellar cortex. J. Physiol., 1969. 202: p. 437-70. 3. Ito, M. and M. Kano, Long-lasting depression of parallel fiber-Purkinje cell
transmission induced by conjunctive stimulation of parallel fibers and
climbing fibers in the cerebellar cortex. Neurosci Lett, 1982. 33(3): p. 253-8. 4. Albus, J., A theory of cerebellar function. Math. Biosci., 1971. 10: p. 25-61. 5. Simpson, J.I. and K.E. Alley, Visual climbing fiber input to rabbit vestibulo-
cerebellum: a source of direction-specific information, in Brain Res. 1974. p. 302-8.
6. Raymond, J.L. and S.G. Lisberger, Neural learning rules for the vestibulo-
ocular reflex. J Neurosci, 1998. 18(21): p. 9112-29. 7. Ke, M.C., C.C. Guo, and J.L. Raymond, Elimination of climbing fiber
instructive signals during motor learning. Nat Neurosci, 2009. 12(9): p. 1171-9.
8. Kimpo, R.R., et al., Distinct patterns of stimulus generalization of increases
and decreases in VOR gain. J Neurophysiol, 2005. 94(5): p. 3092-100. 9. John C. Eccles, M.I., Janos Szenthagothai, The cerebellum as a neuronal
machine. 1967, Berlin, Heidelberg, New York: Springer-Verlag. 335. 10. Eccles, J.C., R. Llinas, and K. Sasaki, The excitatory synaptic action of
climbing fibres on the purinje cells of the cerebellum. J Physiol, 1966. 182(2): p. 268-96.
11. du Lac, S., et al., Learning and memory in the vestibulo-ocular reflex. Annu Rev Neurosci, 1995. 18: p. 409-41.
12. Ito, M., Neural design of the cerebellar motor control system. Brain Res, 1972. 40(1): p. 81-4.
13. Hirata, Y. and S.M. Highstein, Acute adaptation of the vestibuloocular reflex:
signal processing by floccular and ventral parafloccular Purkinje cells. J Neurophysiol, 2001. 85(5): p. 2267-88.
14. Coesmans, M., et al., Bidirectional parallel fiber plasticity in the cerebellum
under climbing fiber control. Neuron, 2004. 44(4): p. 691-700. 15. Miles, F.A., et al., Long-term adaptive changes in primate vestibuloocular
reflex. III. Electrophysiological observations in flocculus of normal monkeys. J Neurophysiol, 1980. 43(5): p. 1437-76.
16. Lisberger, S.G. and A.F. Fuchs, Role of primate flocculus during rapid
behavioral modification of vestibuloocular reflex. I. Purkinje cell activity
during visually guided horizontal smooth-pursuit eye movements and passive
head rotation. J Neurophysiol, 1978. 41(3): p. 733-63. 17. Watanabe, E., Neuronal events correlated with long-term adaptation of the
horizontal vestibulo-ocular reflex in the primate flocculus. Brain Res, 1984. 297(1): p. 169-74.
113
18. Stone, L.S. and S.G. Lisberger, Visual responses of Purkinje cells in the
cerebellar flocculus during smooth-pursuit eye movements in monkeys. I.
Simple spikes. J Neurophysiol, 1990. 63(5): p. 1241-61. 19. Lisberger, S.G., F.A. Miles, and D.S. Zee, Signals used to compute errors in
monkey vestibuloocular reflex: possible role of flocculus. J Neurophysiol, 1984. 52(6): p. 1140-53.
20. Lisberger, S.G., et al., Neural basis for motor learning in the vestibuloocular
reflex of primates. II. Changes in the responses of horizontal gaze velocity
Purkinje cells in the cerebellar flocculus and ventral paraflocculus. J Neurophysiol, 1994. 72(2): p. 954-73.
21. Raymond, J.L., Learning in the oculomotor system: from molecules to
behavior. Curr Opin Neurobiol, 1998. 8(6): p. 770-6. 22. Raymond, J.L. and S.G. Lisberger, Behavioral analysis of signals that guide
learned changes in the amplitude and dynamics of the vestibulo-ocular reflex. J Neurosci, 1996. 16(23): p. 7791-802.
23. Evans, G.J., Synaptic signalling in cerebellar plasticity. Biol Cell, 2007. 99(7): p. 363-78.
24. Lisberger, S.G. and T.J. Sejnowski, Motor learning in a recurrent network
model based on the vestibulo-ocular reflex. Nature, 1992. 360(6400): p. 159-61.
25. Miles, F.A. and B.B. Eighmy, Long-term adaptive changes in primate
vestibuloocular reflex. I. Behavioral observations. J Neurophysiol, 1980. 43(5): p. 1406-25.
26. Boyden, E.S. and J.L. Raymond, Active reversal of motor memories reveals
rules governing memory encoding. Neuron, 2003. 39(6): p. 1031-42. 27. Li, J., S.S. Smith, and J.G. McElligott, Cerebellar nitric oxide is necessary for
vestibulo-ocular reflex adaptation, a sensorimotor model of learning. J Neurophysiol, 1995. 74(1): p. 489-94.
28. Boyden, E.S., et al., Selective engagement of plasticity mechanisms for motor
memory storage. Neuron, 2006. 51(6): p. 823-34. 29. Gilbert, P.F. and W.T. Thach, Purkinje cell activity during motor learning.
Brain Res, 1977. 128(2): p. 309-28. 30. Stone, L.S. and S.G. Lisberger, Visual responses of Purkinje cells in the
cerebellar flocculus during smooth-pursuit eye movements in monkeys. II.
Complex spikes. J Neurophysiol, 1990. 63(5): p. 1262-75. 31. Miles, F.A. and J.H. Fuller, Adaptive plasticity in the vestibulo-ocular
responses of the rhesus monkey. Brain Res, 1974. 80(3): p. 512-6. 32. Gauthier, G.M. and D.A. Robinson, Adaptation of the human vestibuloocular
reflex to magnifying lenses. Brain Res, 1975. 92(2): p. 331-5. 33. Shinoda, Y., et al., The entire trajectory of single climbing and mossy fibers in
the cerebellar nuclei and cortex. Prog Brain Res, 2000. 124: p. 173-86. 34. Barmack, N.H. and V. Yakhnitsa, Cerebellar climbing fibers modulate simple
spikes in Purkinje cells. J Neurosci, 2003. 23(21): p. 7904-16. 35. Frens, M.A., A.L. Mathoera, and J. van der Steen, Floccular complex spike
response to transparent retinal slip. Neuron, 2001. 30(3): p. 795-801.
114
36. Mustari, M.J., et al., Anatomical connections of the primate pretectal nucleus
of the optic tract. J Comp Neurol, 1994. 349(1): p. 111-28. 37. Mustari, M.J. and A.F. Fuchs, Discharge patterns of neurons in the pretectal
nucleus of the optic tract (NOT) in the behaving primate. J Neurophysiol, 1990. 64(1): p. 77-90.
38. Ilg, U.J. and K.P. Hoffmann, Responses of neurons of the nucleus of the optic
tract and the dorsal terminal nucleus of the accessory optic tract in the awake
monkey. Eur J Neurosci, 1996. 8(1): p. 92-105. 39. Misslisch, H. and B.J. Hess, Three-dimensional vestibuloocular reflex of the
monkey: optimal retinal image stabilization versus listing's law. J Neurophysiol, 2000. 83(6): p. 3264-76.
40. Angelaki, D.E., H.H. Zhou, and M. Wei, Foveal versus full-field visual
stabilization strategies for translational and rotational head movements. J Neurosci, 2003. 23(4): p. 1104-8.
41. Alley, K., R. Baker, and J.I. Simpson, Afferents to the vestibulo-cerebellum
and the origin of the visual climbing fibers in the rabbit. Brain Res, 1975. 98(3): p. 582-9.
42. Belton, T., M. Suh, and J.I. Simpson, The nonvisual complex spike signal in
the flocculus responds to challenges to the vestibulo-ocular reflex gain. Ann N Y Acad Sci, 2002. 978: p. 503-4.
43. Cohen, M.R., et al., Reversal of motor learning in the vestibulo-ocular reflex in
the absence of visual input. Learn Mem, 2004. 11(5): p. 559-65. 44. Winkelman, B. and M. Frens, Motor coding in floccular climbing fibers. J
Neurophysiol, 2006. 95(4): p. 2342-51. 45. Lisberger, S.G., T.A. Pavelko, and D.M. Broussard, Neural basis for motor
learning in the vestibuloocular reflex of primates. I. Changes in the responses
of brain stem neurons. J Neurophysiol, 1994. 72(2): p. 928-53. 46. Robinson, D.A., A Method of Measuring Eye Movement Using a Scleral
Search Coil in a Magnetic Field. IEEE Trans Biomed Eng, 1963. 10: p. 137-45.
47. Soetedjo, R. and A.F. Fuchs, Complex spike activity of purkinje cells in the
oculomotor vermis during behavioral adaptation of monkey saccades. J Neurosci, 2006. 26(29): p. 7741-55.
48. Kitazawa, S., T. Kimura, and P.B. Yin, Cerebellar complex spikes encode both
destinations and errors in arm movements. Nature, 1998. 392(6675): p. 494-7. 49. Mauk, M.D., J.E. Steinmetz, and R.F. Thompson, Classical conditioning using
stimulation of the inferior olive as the unconditioned stimulus. Proc Natl Acad Sci U S A, 1986. 83(14): p. 5349-53.
50. Shutoh, F., et al., Memory trace of motor learning shifts transsynaptically from
cerebellar cortex to nuclei for consolidation. Neuroscience, 2006. 139(2): p. 767-77.
51. Boyden, E.S., A. Katoh, and J.L. Raymond, Cerebellum-dependent learning:
the role of multiple plasticity mechanisms. Annu Rev Neurosci, 2004. 27: p. 581-609.
115
52. Hansel, C., et al., alphaCaMKII Is essential for cerebellar LTD and motor
learning. Neuron, 2006. 51(6): p. 835-43. 53. Lisberger, S.G., Neural basis for motor learning in the vestibuloocular reflex
of primates. III. Computational and behavioral analysis of the sites of
learning. J Neurophysiol, 1994. 72(2): p. 974-98. 54. Medina, J.F., et al., Mechanisms of cerebellar learning suggested by eyelid
conditioning. Curr Opin Neurobiol, 2000. 10(6): p. 717-24. 55. Jirenhed, D.A., F. Bengtsson, and G. Hesslow, Acquisition, extinction, and
reacquisition of a cerebellar cortical memory trace. J Neurosci, 2007. 27(10): p. 2493-502.
56. Ogasawara, H., et al., Nitric oxide regulates input specificity of long-term
depression and context dependence of cerebellar learning. PLoS Comput Biol, 2007. 3(1): p. e179.
57. Lev-Ram, V., et al., A new form of cerebellar long-term potentiation is
postsynaptic and depends on nitric oxide but not cAMP. Proc Natl Acad Sci U S A, 2002. 99(12): p. 8389-93.
58. Rancillac, A. and F. Crepel, Synapses between parallel fibres and stellate cells
express long-term changes in synaptic efficacy in rat cerebellum. J Physiol, 2004. 554(Pt 3): p. 707-20.
59. Jorntell, H. and C.F. Ekerot, Reciprocal bidirectional plasticity of parallel
fiber receptive fields in cerebellar Purkinje cells and their afferent
interneurons. Neuron, 2002. 34(5): p. 797-806. 60. Mittmann, W. and M. Hausser, Linking synaptic plasticity and spike output at
excitatory and inhibitory synapses onto cerebellar Purkinje cells. J Neurosci, 2007. 27(21): p. 5559-70.
61. Haddad, G.M., J.L. Demer, and D.A. Robinson, The effect of lesions of the
dorsal cap of the inferior olive on the vestibulo-ocular and optokinetic systems
of the cat. Brain Res, 1980. 185(2): p. 265-75. 62. Mintz, M., et al., Unilateral inferior olive NMDA lesion leads to unilateral
deficit in acquisition and retention of eyelid classical conditioning. Behav Neural Biol, 1994. 61(3): p. 218-24.
63. Demer, J.L., D.A. Echelman, and D.A. Robinson, Effects of electrical
stimulation and reversible lesions of the olivocerebellar pathway on Purkinje
cell activity in the flocculus of the cat. Brain Res, 1985. 346(1): p. 22-31. 64. Zbarska, S., J.R. Bloedel, and V. Bracha, Cerebellar dysfunction explains the
extinction-like abolition of conditioned eyeblinks after NBQX injections in the
inferior olive. J Neurosci, 2008. 28(1): p. 10-20. 65. Rambold, H., et al., Partial ablations of the flocculus and ventral
paraflocculus in monkeys cause linked deficits in smooth pursuit eye
movements and adaptive modification of the VOR. J Neurophysiol, 2002. 87(2): p. 912-24.
66. Robinson, D.A., Adaptive gain control of vestibuloocular reflex by the
cerebellum. J Neurophysiol, 1976. 39(5): p. 954-69.
116
67. Torte, M.P., et al., Anatomical segregation of different adaptative processes
within the vestibulocerebellum of the cat. Exp Brain Res, 1994. 99(3): p. 441-54.
68. Ito, M., P.J. Jastreboff, and Y. Miyashita, Specific effects of unilateral lesions
in the flocculus upon eye movements in albino rabbits. Exp Brain Res, 1982. 45(1-2): p. 233-42.
69. Cohen, H., et al., Habituation and adaptation of the vestibuloocular reflex: a
model of differential control by the vestibulocerebellum. Exp Brain Res, 1992. 90(3): p. 526-38.
70. Ito, M., N. Nisimaru, and M. Yamamoto, Specific patterns of neuronal
connexions involved in the control of the rabbit's vestibulo-ocular reflexes by
the cerebellar flocculus. J Physiol, 1977. 265(3): p. 833-54. 71. Ghelarducci, B., M. Ito, and N. Yagi, Impulse discharges from flocculus
Purkinje cells of alert rabbits during visual stimulation combined with
horizontal head rotation. Brain Res, 1975. 87(1): p. 66-72. 72. Nagao, S., Behavior of floccular Purkinje cells correlated with adaptation of
vestibulo-ocular reflex in pigmented rabbits. Exp Brain Res, 1989. 77(3): p. 531-40.
73. Maekawa, K. and T. Takeda, Origin of descending afferents to the rostral part
of dorsal cap of inferior olive which transfers contralateral optic activities to
the flocculus. A horseradish peroxidase study. Brain Res, 1979. 172(3): p. 393-405.
74. Safo, P.K. and W.G. Regehr, Endocannabinoids control the induction of
cerebellar LTD. Neuron, 2005. 48(4): p. 647-59. 75. Carleton, S.C. and M.B. Carpenter, Distribution of primary vestibular fibers in
the brainstem and cerebellum of the monkey. Brain Res, 1984. 294(2): p. 281-98.
76. Epema, A.H., N.M. Gerrits, and J. Voogd, Secondary vestibulocerebellar
projections to the flocculus and uvulo-nodular lobule of the rabbit: a study
using HRP and double fluorescent tracer techniques. Exp Brain Res, 1990. 80(1): p. 72-82.
77. De Zeeuw, C.I. and C.H. Yeo, Time and tide in cerebellar memory formation. Curr Opin Neurobiol, 2005. 15(6): p. 667-74.
78. Medina, J.F., K.S. Garcia, and M.D. Mauk, A mechanism for savings in the
cerebellum. J Neurosci, 2001. 21(11): p. 4081-9. 79. Yakushin, S.B., T. Raphan, and B. Cohen, Context-specific adaptation of the
vertical vestibuloocular reflex with regard to gravity. J Neurophysiol, 2000. 84(6): p. 3067-71.
80. Lisberger, S.G., F.A. Miles, and L.M. Optican, Frequency-selective
adaptation: evidence for channels in the vestibulo-ocular reflex? J Neurosci, 1983. 3(6): p. 1234-44.
81. Broussard, D.M., A.J. Priesol, and Y.F. Tan, Asymmetric responses to rotation
at high frequencies in central vestibular neurons of the alert cat. Brain Res, 2004. 1005(1-2): p. 137-53.
117
82. Dickman, J.D. and D.E. Angelaki, Dynamics of vestibular neurons during
rotational motion in alert rhesus monkeys. Exp Brain Res, 2004. 155(1): p. 91-101.
83. Luebke, A.E. and D.A. Robinson, Gain changes of the cat's vestibulo-ocular
reflex after flocculus deactivation. Exp Brain Res, 1994. 98(3): p. 379-90. 84. McElligott, J.G., P. Beeton, and J. Polk, Effect of cerebellar inactivation by
lidocaine microdialysis on the vestibuloocular reflex in goldfish. J Neurophysiol, 1998. 79(3): p. 1286-94.
85. Hirata, Y., J.M. Lockard, and S.M. Highstein, Capacity of vertical VOR
adaptation in squirrel monkey. J Neurophysiol, 2002. 88(6): p. 3194-207. 86. Shelhamer, M., D.A. Robinson, and H.S. Tan, Context-specific adaptation of
the gain of the vestibulo-ocular reflex in humans. J Vestib Res, 1992. 2(1): p. 89-96.
87. Lewis, R.F., R.A. Clendaniel, and D.S. Zee, Vergence-dependent adaptation of
the vestibulo-ocular reflex. Exp Brain Res, 2003. 152(3): p. 335-40. 88. Zhou, W., et al., Rapid motor learning in the translational vestibulo-ocular
reflex. J Neurosci, 2003. 23(10): p. 4288-98. 89. Morris, E.J. and S.G. Lisberger, Different responses to small visual errors
during initiation and maintenance of smooth-pursuit eye movements in
monkeys. J Neurophysiol, 1987. 58(6): p. 1351-69. 90. Boyden, E.S., et al., Millisecond-timescale, genetically targeted optical control
of neural activity. Nat Neurosci, 2005. 8(9): p. 1263-8.